Megatest

# make install CSCOPTS='-accumulate-profile -profile-name $(PWD)/profile-ww$(shell date +%V.%u)'
# rm <files>.o ; make install CSCOPTS='-profile' ; ... ;  chicken-profile | less

PREFIX=$(PWD)
CSCOPTS=
INSTALL=install
SRCFILES = common.scm items.scm launch.scm \
   ods.scm runconfig.scm server.scm configf.scm \
   db.scm keys.scm margs.scm megatest-version.scm \
   process.scm runs.scm tasks.scm tests.scm genexample.scm \
   http-transport.scm filedb.scm tdb.scm \
   client.scm synchash.scm daemon.scm mt.scm \
   client.scm daemon.scm mt.scm \
   ezsteps.scm lock-queue.scm sdb.scm \
   rmt.scm api.scm \
   portlogger.scm archive.scm env.scm diff-report.scm cgisetup/models/pgdb.scm

# Eggs to install (straightforward ones)
EGGS=matchable readline apropos base64 regex-literals format regex-case test coops trace csv \
dot-locking posix-utils posix-extras directory-utils hostinfo tcp-server rpc csv-xml fmt \

ndboard : newdashboard.scm $(OFILES) $(GOFILES)
	csc $(CSCOPTS) $(OFILES) $(GOFILES) newdashboard.scm -o ndboard

mtut: $(OFILES) mtut.scm
	csc $(CSCOPTS) $(OFILES) mtut.scm -o mtut

TCMTOBJS = \
	api.o \
	archive.o \
	cgisetup/models/pgdb.o \
	client.o \
	common.o \
	configf.o \
	daemon.o \
	db.o \
	env.o \
	http-transport.o \
	items.o \
	keys.o \
	launch.o \
	lock-queue.o \
	margs.o \
	mt.o \
	megatest-version.o \
	ods.o \
	portlogger.o \
	process.o \
	rmt.o \
	rpc-transport.o \
	runconfig.o \
	runs.o \
	server.o \
	tasks.o \
	tdb.o \
	tests.o \


tcmt : $(TCMTOBJS) tcmt.scm
	csc $(CSCOPTS) $(TCMTOBJS) tcmt.scm -o tcmt

# install documentation to $(PREFIX)/docs
# DOES NOT REBUILD DOCS
#
$(PREFIX)/share/docs/megatest_manual.html : docs/manual/megatest_manual.html
	mkdir -p $(PREFIX)/share/docs
	$(INSTALL) docs/manual/megatest_manual.html $(PREFIX)/share/docs/megatest_manual.html
	for png in $(PNGFILES);do $(INSTALL) docs/manual/$$png $(PREFIX)/share/docs/$$png;done

$(PREFIX)/bin/.$(ARCHSTR)/mtut : mtut
	$(INSTALL) mtut $(PREFIX)/bin/.$(ARCHSTR)/mtut

$(PREFIX)/bin/mtutil : $(PREFIX)/bin/.$(ARCHSTR)/mtut utils/mk_wrapper
	utils/mk_wrapper $(PREFIX) mtut $(PREFIX)/bin/mtutil
	chmod a+x $(PREFIX)/bin/mtutil

$(PREFIX)/bin/.$(ARCHSTR)/tcmt : tcmt
	$(INSTALL) tcmt $(PREFIX)/bin/.$(ARCHSTR)/tcmt

$(PREFIX)/bin/tcmt : $(PREFIX)/bin/.$(ARCHSTR)/tcmt utils/mk_wrapper
	utils/mk_wrapper $(PREFIX) tcmt $(PREFIX)/bin/tcmt
	chmod a+x $(PREFIX)/bin/tcmt

# $(PREFIX)/bin/.$(ARCHSTR)/mdboard : multi-dboard
#	$(INSTALL) multi-dboard $(PREFIX)/bin/.$(ARCHSTR)/mdboard

# $(PREFIX)/bin/mdboard : $(PREFIX)/bin/.$(ARCHSTR)/mdboard  utils/mk_wrapper
# 	utils/mk_wrapper $(PREFIX) mdboard $(PREFIX)/bin/mdboard
# 	chmod a+x $(PREFIX)/bin/mdboard

	chmod a+x $(PREFIX)/bin/dashboard
	$(INSTALL) dboard $(PREFIX)/bin/.$(ARCHSTR)/dboard

install : $(PREFIX)/bin/.$(ARCHSTR) $(PREFIX)/bin/.$(ARCHSTR)/mtest $(PREFIX)/bin/megatest \
          $(PREFIX)/bin/.$(ARCHSTR)/dboard $(PREFIX)/bin/dashboard $(HELPERS) $(PREFIX)/bin/nbfake \
	  $(PREFIX)/bin/nbfind $(PREFIX)/bin/loadrunner $(PREFIX)/bin/viewscreen $(PREFIX)/bin/mt_xterm \
	  $(PREFIX)/share/docs/megatest_manual.html $(PREFIX)/bin/remrun $(PREFIX)/bin/mtutil \
          $(PREFIX)/share/db/mt-pg.sql $(PREFIX)/bin/.$(ARCHSTR)/ndboard $(PREFIX)/bin/newdashboard
          $(PREFIX)/share/db/mt-pg.sql $(PREFIX)/bin/.$(ARCHSTR)/ndboard  $(PREFIX)/bin/tcmt

# $(PREFIX)/bin/newdashboard

$(PREFIX)/bin/.$(ARCHSTR) : 
	mkdir -p $(PREFIX)/bin/.$(ARCHSTR)
	mkdir -p $(PREFIX)/bin/.$(ARCHSTR)/lib

test: tests/tests.scm
	cd tests;csi -I .. -b -n tests.scm

	if csi -ne '(use postgresql)';then \
	   echo "(use postgresql)(hash-table-set! *available-db* 'postgresql #t)" >> altdb.scm;\
	fi

portlogger-example : portlogger-example.scm api.o archive.o client.o common.o configf.o daemon.o dashboard-tests.o db.o dcommon.o ezsteps.o filedb.o genexample.o gutils.o http-transport.o items.o keys.o launch.o lock-queue.o margs.o megatest-version.o mt.o ods.o portlogger.o process.o rmt.o rpc-transport.o runconfig.o runs.o sdb.o server.o synchash.o tasks.o tdb.o tests.o tree.o
	csc $(CSCOPTS) portlogger-example.scm api.o archive.o client.o common.o configf.o daemon.o dashboard-tests.o db.o dcommon.o ezsteps.o filedb.o genexample.o gutils.o http-transport.o items.o keys.o launch.o lock-queue.o margs.o megatest-version.o mt.o ods.o portlogger.o process.o rmt.o rpc-transport.o runconfig.o runs.o sdb.o server.o synchash.o tasks.o tdb.o tests.o tree.o

# create a pdf dot graphviz diagram from notations in rmt.scm
rmt.pdf : rmt.scm
	grep ';;DOT' rmt.scm | sed -e 's/.*;;DOT //' > rmt.dot;dot -Tpdf rmt.dot -o rmt.pdf

    get-all-run-ids
    get-prev-run-ids
    get-run-ids-matching-target
    get-runs-by-patt
    get-steps-data
    get-steps-for-test
    read-test-data
    read-test-data*
    login
    tasks-get-last
    testmeta-get-record
    have-incompletes?
    synchash-get
    ))

     (debug:print 0 *default-log-port* " message: "  ((condition-property-accessor 'exn 'message) exn))       
     (vector #f (vector exn call-chain dat))) ;; return some stuff for debug if an exception happens
   (cond
    ((not (vector? dat))                    ;; it is an error to not receive a vector
     (vector #f (vector #f "remote must be called with a vector")))
    ((> *api-process-request-count* 20) ;; 20)
     (debug:print 0 *default-log-port* "WARNING: api:execute-requests received an overloaded message.")
     (set! *server-overloaded* #t)
     (vector #f (vector #f 'overloaded))) ;; the inner vector is what gets returned. nope, don't know why. please refactor!
    (else  
     (let* ((cmd-in            (vector-ref dat 0))
            (cmd               (if (symbol? cmd-in)
				   cmd-in
				   (string->symbol cmd-in)))
            (params            (vector-ref dat 1))

                   ((get-keys-write)                        (db:get-keys dbstruct)) ;; force a dummy "write" query to force server; for debug in -repl
                   
                   ;; SERVERS
                   ((start-server)                    (apply server:kind-run params))
                   ((kill-server)                     (set! *server-run* #f))

                   ;; TESTS

                   ((test-set-state-status-by-id)     (apply mt:test-set-state-status-by-id dbstruct params))
                   ;;((test-set-state-status-by-id)     (apply mt:test-set-state-status-by-id dbstruct params))
                   ;;BB - commented out above because it was calling below, eventually, incorrectly (dbstruct passed to mt:test-set-state-status-by-id, which previosly did more, but now only passes thru to db:set-state-status-and-roll-up-items.
                   ((test-set-state-status-by-id)

                    ;; (define (db:set-state-status-and-roll-up-items dbstruct run-id test-name item-path state status comment)
                    (db:set-state-status-and-roll-up-items
                     dbstruct
                     (list-ref params 0) ; run-id
                     (list-ref params 1) ; test-name
                     #f                  ; item-path
                     (list-ref params 2) ; state
                     (list-ref params 3) ; status
                     (list-ref params 4) ; comment
                     ))
                   
                   ((delete-test-records)             (apply db:delete-test-records dbstruct params))
                   ((delete-old-deleted-test-records) (apply db:delete-old-deleted-test-records dbstruct params))
                   ((test-set-state-status)           (apply db:test-set-state-status dbstruct params))
                   ((test-set-top-process-pid)        (apply db:test-set-top-process-pid dbstruct params))
                   ((set-state-status-and-roll-up-items) (apply db:set-state-status-and-roll-up-items dbstruct params))
                   ((top-test-set-per-pf-counts)      (apply db:top-test-set-per-pf-counts dbstruct params))
                   ((test-set-archive-block-id)       (apply db:test-set-archive-block-id dbstruct params))

                   ((get-tests-tags)            (db:get-tests-tags dbstruct))

                   ;; TASKS
                   ((tasks-add)                 (apply tasks:add dbstruct params))   
                   ((tasks-set-state-given-param-key) (apply tasks:set-state-given-param-key dbstruct params))
                   ((tasks-get-last)            (apply tasks:get-last dbstruct params))

		   ;; NO SYNC DB
		   ((no-sync-set)               (apply db:no-sync-set         *no-sync-db* params))
		   ((no-sync-get/default)       (apply db:no-sync-get/default *no-sync-db* params))
		   ((no-sync-del!)              (apply db:no-sync-del!        *no-sync-db* params))
		 
                   ;; ARCHIVES
                   ;; ((archive-get-allocations)   
                   ((archive-register-disk)     (apply db:archive-register-disk dbstruct params))
                   ((archive-register-block-name)(apply db:archive-register-block-name dbstruct params))
                   ((archive-allocate-testsuite/area-to-block)(apply db:archive-allocate-testsuite/area-to-block dbstruct block-id testsuite-name areakey))

                   ;;======================================================================

                   ;; STEPS
                   ((get-steps-data)               (apply db:get-steps-data dbstruct params))
                   ((get-steps-for-test)           (apply db:get-steps-for-test dbstruct params))

                   ;; TEST DATA
                   ((read-test-data)               (apply db:read-test-data dbstruct params))
                   ((read-test-data*)              (apply db:read-test-data* dbstruct params))

                   ;; MISC
                   ((get-latest-host-load)         (apply db:get-latest-host-load dbstruct params))
                   ((have-incompletes?)            (apply db:have-incompletes? dbstruct params))
                   ((login)                        (apply db:login dbstruct params))
                   ((general-call)                 (let ((stmtname   (car params))
                                                         (run-id     (cadr params))

(define (archive:main linktree target runname testname itempath options)
  (let ((testdir (conc linktree "/" target "/" runname "/" testname "/" itempatt))
	(flavor  'plain) ;; type of machine to run jobs on
	(maxload 1.5)   ;; max allowed load for this work
	(adisks  (archive:get-archive-disks)))
    ;; get testdir size
    ;;   - hand off du to job mgr
    (if (and (file-exists? testdir)
    (if (and (common:file-exists? testdir)
	     (file-is-writable? testdir))
	(let* ((dused  (jobrunner:run-job 
			flavor  ;; machine type
			maxload ;; max allowed load
			'()     ;; prevars - environment vars to set for the job
			common:get-disk-space-used  ;; if a proc call it, if a string it is a unix command
			(list testdir)))

	      
	      (toplevel/children (and (db:test-get-is-toplevel test-dat)
				      (> (rmt:test-toplevel-num-items run-id test-name) 0)))
	      (test-partial-path (conc target "/" run-name "/" (db:test-make-full-name test-name item-path)))
	      ;; note the trailing slash to get the dir inspite of it being a link
	      (test-path         (conc linktree "/" test-partial-path))
	      (mutex-lock! rp-mutex)
	      (test-physical-path (if (file-exists? test-path) 
	      (test-physical-path (if (common:file-exists? test-path) 
				      (common:real-path test-path)
				      #f))
	      (mutex-unlock! rp-mutex)
	      (partial-path-index (if test-physical-path (substring-index test-partial-path test-physical-path) #f))
	      (test-base         (if (and partial-path-index 
					  test-physical-path )
				     (substring test-physical-path
						0
						partial-path-index)
				     #f)))
	 
 	 (cond
	  (toplevel/children
	   (debug:print 0 *default-log-port* "WARNING: cannot archive " test-name " with id " test-id " as it is a toplevel test with children"))
	  ((not (file-exists? test-path))
	  ((not (common:file-exists? test-path))
	   (debug:print 0 *default-log-port* "WARNING: Cannot archive " test-name "/" item-path " as path " test-path " does not exist"))
	  (else
	   (debug:print 0 *default-log-port*
			"From test-dat=" test-dat " derived the following:\n"
			"test-partial-path  = " test-partial-path "\n"
			"test-path          = " test-path "\n"
			"test-physical-path = " test-physical-path "\n"

	      (bup-index-params (append (list "-d" archive-dir "index") test-paths))
	      (bup-save-params  (append (list "-d" archive-dir "save" ;; (conc "--strip-path=" linktree)
					      (conc "-" compress) ;; or (conc "--compress=" compress)
					      "-n" (conc (common:get-testsuite-name) "-" run-id)
					      (conc "--strip-path=" disk-group))
					test-paths))
	      (print-prefix      #f)) ;; "Running: ")) ;; change to #f to turn off printing
	 (if (not (file-exists? archive-dir))
	 (if (not (common:file-exists? archive-dir))
	     (create-directory archive-dir #t))
	 (if (not (file-exists? (conc archive-dir "/HEAD")))
	 (if (not (common:file-exists? (conc archive-dir "/HEAD")))
	     (begin
	       ;; replace this with jobrunner stuff enventually
	       (debug:print-info 0 *default-log-port* "Init bup in " archive-dir)
	       ;; (mutex-lock! bup-mutex)
	       (run-n-wait bup-exe params: bup-init-params print-cmd: print-prefix)
	       ;; (mutex-unlock! bup-mutex)
	       ))

	      (toplevel/children (and (db:test-get-is-toplevel test-dat)
				      (> (rmt:test-toplevel-num-items run-id test-name) 0)))
	      (test-partial-path (conc target "/" run-name "/" (db:test-make-full-name test-name item-path)))
	      ;; note the trailing slash to get the dir inspite of it being a link
	      (test-path         (conc linktree "/" test-partial-path))
	      ;; if the old path was not deleted then prev-test-physical-path will end up pointing to a real directory
	      (mutex-lock! rp-mutex)
	      (prev-test-physical-path (if (file-exists? test-path)
	      (prev-test-physical-path (if (common:file-exists? test-path)
					   ;; (read-symbolic-link test-path #t)
					   (common:real-path test-path)
					   #f))
	      (mutex-unlock! rp-mutex)
	      (new-test-physical-path  (conc best-disk "/" test-partial-path))
	      (archive-block-id        (db:test-get-archived test-dat))
	      (archive-block-info      (rmt:test-get-archive-block-info archive-block-id))
	      (archive-path            (if (vector? archive-block-info)
					   (vector-ref archive-block-info 2) ;; look in db.scm for test-get-archive-block-info for the vector record info
					   #f)) ;; no archive found?
	      (archive-internal-path   (conc (common:get-testsuite-name) "-" run-id "/latest/" test-partial-path)))
	 
	 ;; some sanity checks, move an existing path out of the way - iif it is not a toplevel with children
	 ;;
	 (if (and (not toplevel/children)  ;; special handling needed for toplevel with children
		  prev-test-physical-path
		  (file-exists? prev-test-physical-path)) ;; what to do? abort or clean up or link it in?
		  (common:file-exists? prev-test-physical-path)) ;; what to do? abort or clean up or link it in?
	     (let* ((base (pathname-directory prev-test-physical-path))
		    (dirn (pathname-file      prev-test-physical-path))
		    (newn (conc base "/." dirn)))
	       (debug:print-error 0 *default-log-port* "the old directory " prev-test-physical-path ", still exists! Moving it to " newn)
	       (rename-file prev-test-physical-path newn)))

	 (if (and archive-path ;; no point in proceeding if there is no actual archive

;;======================================================================
;; Copyright 2006-2012, Matthew Welland.
;; 
;;  This program is made available under the GNU GPL version 2.0 or
;;  greater. See the accompanying file COPYING for details.
;; 
;;  This program is distributed WITHOUT ANY WARRANTY; without even the
;;  implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
;;  PURPOSE.
;;======================================================================

(use srfi-1 posix regex-case base64 format dot-locking csv-xml z3 sql-de-lite hostinfo md5 message-digest typed-records directory-utils stack
     matchable regex posix srfi-18 extras)
     matchable regex posix srfi-18 extras
     pkts (prefix dbi dbi:))

(import (prefix sqlite3 sqlite3:))
(import (prefix base64 base64:))

(declare (unit common))

(include "common_records.scm")

	      (debug:print-error 0 *default-log-port* "bad value for setenv, key=" key ", value=" val)
	    (setenv key val))
	  (debug:print-error 0 *default-log-port* "bad value for setenv, key=" key ", value=" val))))

(define home (getenv "HOME"))
(define user (getenv "USER"))

;; GLOBAL GLETCHES
;; GLOBALS

;; CONTEXTS
(defstruct cxt
  (taskdb #f)
  (cmutex (make-mutex)))
;; (define *contexts* (make-hash-table))
;; (define *context-mutex* (make-mutex))

(define *task-db*             #f) ;; (vector db path-to-db)
(define *db-access-allowed*   #t) ;; flag to allow access
(define *db-access-mutex*     (make-mutex))
(define *db-transaction-mutex* (make-mutex))
(define *db-cache-path*       #f)
(define *db-with-db-mutex*    (make-mutex))
(define *db-api-call-time*    (make-hash-table)) ;; hash of command => (list of times)
;; no sync db
(define *no-sync-db*          #f)

;; SERVER
(define *my-client-signature* #f)
(define *transport-type*    'http)             ;; override with [server] transport http|rpc|nmsg
(define *runremote*         #f)                ;; if set up for server communication this will hold <host port>
;; (define *max-cache-size*    0)
(define *logged-in-clients* (make-hash-table))
;; (define *server-id*         #f)
(define *server-info*       #f)  ;; good candidate for easily convert to non-global
(define *time-to-exit*      #f)
(define *server-run*        #t)
(define *run-id*            #f)
(define *server-kind-run*   (make-hash-table))
(define *home-host*         #f)
;; (define *total-non-write-delay* 0)
(define *heartbeat-mutex*   (make-mutex))
(define *api-process-request-count* 0)
(define *max-api-process-requests* 0)
(define *server-overloaded*  #f)

;; client
(define *rmt-mutex*         (make-mutex))     ;; remote access calls mutex 

;; RPC transport
(define *rpc:listener*      #f)

;; KEY info
(define *target*            (make-hash-table)) ;; cache the target here; target is keyval1/keyval2/.../keyvalN
(define *keys*              (make-hash-table)) ;; cache the keys here
(define *keyvals*           (make-hash-table))
(define *toptest-paths*     (make-hash-table)) ;; cache toptest path settings here
(define *test-paths*        (make-hash-table)) ;; cache test-id to test run paths here
(define *test-ids*          (make-hash-table)) ;; cache run-id, testname, and item-path => test-id
(define *test-info*         (make-hash-table)) ;; cache the test info records, update the state, status, run_duration etc. from testdat.db

(define *run-info-cache*     (make-hash-table)) ;; run info is stable, no need to reget
(define *launch-setup-mutex* (make-mutex))     ;; need to be able to call launch:setup often so mutex it and re-call the real deal only if *toppath* not set
(define *homehost-mutex*     (make-mutex))

;; Miscellaneous
(define *triggers-mutex*     (make-mutex))     ;; block overlapping processing of triggers

(defstruct remote
  (hh-dat            (common:get-homehost)) ;; homehost record ( addr . hhflag )
  (server-url        (if *toppath* (server:check-if-running *toppath*))) ;; (server:check-if-running *toppath*) #f))
  (last-server-check 0)  ;; last time we checked to see if the server was alive
  (conndat           #f)
  (transport         *transport-type*)
  (server-timeout    (server:get-timeout)) ;; default from server:get-timeout

(define (common:get-last-run-version-number)
  (string->number 
   (substring (common:get-last-run-version) 0 6)))

(define (common:set-last-run-version)
  (rmt:set-var "MEGATEST_VERSION" (common:version-signature)))

;; postive number if megatest version > db version
;; negative number if megatest version < db version
(define (common:version-db-delta)
         (- megatest-version (common:get-last-run-version-number)))

(define (common:version-changed?)
  (not (equal? (common:get-last-run-version)
	       (common:version-signature))))
               (common:version-signature))))

(define (common:api-changed?)
  (not (equal? (substring (->string megatest-version) 0 4)
               (substring (conc (common:get-last-run-version)) 0 4))))
  
;; Move me elsewhere ...
;; RADT => Why do we meed the version check here, this is called only if version misma
;;
(define (common:cleanup-db dbstruct)
  (db:multi-db-sync 
(define (common:cleanup-db dbstruct #!key (full #f))
  (apply db:multi-db-sync 
   dbstruct
   'schema
   ;; 'new2old
   'killservers
   'dejunk
   'adj-target
   ;; 'old2new
   'new2old
   (if full
       '(dejunk)
   )
  (if (common:version-changed?)
       '()))
  (if (common:api-changed?)
      (common:set-last-run-version)))

;; Rotate logs, logic: 
;;                 if > 500k and older than 1 week:
;;                     remove previous compressed log and compress this log
;; WARNING: This proc operates assuming that it is in the directory above the
;;          logs directory you wish to log-rotate.

             (file-age (- (current-seconds)(file-modification-time fullname))))
        (if (or (and (string-match "^.*.log" file)
                     (> (file-size fullname) 200000))
                (and (string-match "^server-.*.log" file)
                     (> (- (current-seconds) (file-modification-time fullname))
                        (* 8 60 60))))
            (let ((gzfile (conc fullname ".gz")))
              (if (file-exists? gzfile)
              (if (common:file-exists? gzfile)
                  (begin
                    (debug:print-info 0 *default-log-port* "removing " gzfile)
                    (delete-file gzfile)))
              (debug:print-info 0 *default-log-port* "compressing " file)
              (system (conc "gzip " fullname)))
            (if (> file-age (* (string->number (or (configf:lookup *configdat* "setup" "log-expire-days") "30")) 24 3600))
                (handle-exceptions
                 exn
                 #f
                 (delete-file fullname)))))))
   '()
   "logs"))

;; Force a megatest cleanup-db if version is changed and skip-version-check not specified
;; Do NOT check if not on homehost!
;;
(define (common:exit-on-version-changed)
  (if (common:on-homehost?)
      (if (common:version-changed?)
      (if (common:api-changed?)
	  (let* ((mtconf (conc (get-environment-variable "MT_RUN_AREA_HOME") "/megatest.config"))
                (dbfile (conc (get-environment-variable "MT_RUN_AREA_HOME") "/megatest.db"))
                (read-only (not (file-write-access? dbfile)))
                (dbstruct (db:setup #t)))
	    (debug:print 0 *default-log-port*
			 "WARNING: Version mismatch!\n"
			 "   expected: " (common:version-signature) "\n"
			 "   got:      " (common:get-last-run-version))
            (cond
             ((get-environment-variable "MT_SKIP_DB_MIGRATE") #t)
             ((and (file-exists? mtconf) (file-exists? dbfile) (not read-only)
             ((and (common:file-exists? mtconf) (common:file-exists? dbfile) (not read-only)
                   (eq? (current-user-id)(file-owner mtconf))) ;; safe to run -cleanup-db
              (debug:print 0 *default-log-port* "   I see you are the owner of megatest.config, attempting to cleanup and reset to new version")
              (handle-exceptions
               exn
               (begin
                 (debug:print 0 *default-log-port* "Failed to switch versions.")
                 (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
                 (print-call-chain (current-error-port))
                 (exit 1))
               (common:cleanup-db dbstruct)))
             ((not (file-exists? mtconf))
             ((not (common:file-exists? mtconf))
              (debug:print 0 *default-log-port* "   megatest.config does not exist in this area.  Cannot proceed with megatest version migration.")
              (exit 1))
             ((not (file-exists? dbfile))
             ((not (common:file-exists? dbfile))
              (debug:print 0 *default-log-port* "   megatest.db does not exist in this area.  Cannot proceed with megatest version migration.")
              (exit 1))
             ((not (eq? (current-user-id)(file-owner mtconf)))
              (debug:print 0 *default-log-port* "   You do not own megatest.db in this area.  Cannot proceed with megatest version migration.")
              (exit 1))
             (read-only
              (debug:print 0 *default-log-port* "   You have read-only access to this area.  Cannot proceed with megatest version migration.")
              (exit 1))
             (else
              (debug:print 0 *default-log-port* " to switch versions you can run: \"megatest -cleanup-db\"")
              (exit 1)))))
      (begin
	(debug:print 0 *default-log-port* "ERROR: cannot migrate version unless on homehost. Exiting.")
	(exit 1))))
              (exit 1)))))))
;;      (begin
;;	(debug:print 0 *default-log-port* "ERROR: cannot migrate version unless on homehost. Exiting.")
;;	(exit 1))))

;;======================================================================
;; S P A R S E   A R R A Y S
;;======================================================================

(define (make-sparse-array)
  (let ((a (make-sparse-vector)))

;; if lock is older than expire-time then remove it and try again
;; to get the lock
;;
(define (common:simple-file-lock fname #!key (expire-time 300))
  (handle-exceptions
      exn
      #f ;; don't really care what went wrong right now. NOTE: I have not seen this one actually fail.
    (if (file-exists? fname)
    (if (common:file-exists? fname)
	(if (> (- (current-seconds)(file-modification-time fname)) expire-time)
	    (begin
	      (delete-file* fname)
	      (common:simple-file-lock fname expire-time: expire-time))
	    #f)
	(let ((key-string (conc (get-host-name) "-" (current-process-id))))
	  (with-output-to-file fname
	    (lambda ()
	      (print key-string)))
	  (thread-sleep! 0.25)
	  (if (file-exists? fname)
	  (if (common:file-exists? fname)
	      (with-input-from-file fname
		(lambda ()
		  (equal? key-string (read-line))))
	      #f)))))

(define (common:simple-file-lock-and-wait fname #!key (expire-time 300))
  (let ((end-time (+ expire-time (current-seconds))))

      #f ;; I don't really care why this failed (at least for now)
    (delete-file* fname)))

;;======================================================================
;; S T A T E S   A N D   S T A T U S E S
;;======================================================================

;; BBnote: *common:std-states* - dashboard filter control and test control state buttons defined here; used in set-fields-panel and dboard:make-controls
(define *common:std-states*   
(define *common:std-states*   ;; for toggle buttons in dashboard
  '((0 "ARCHIVED")
    (1 "STUCK")
    (2 "KILLREQ")
    (3 "KILLED")
    (4 "NOT_STARTED")
    (5 "COMPLETED")
    (6 "LAUNCHED")
    (7 "REMOTEHOSTSTART")
    (8 "RUNNING")
    ))

;; BBnote: *common:std-statuses* dashboard filter control and test control status buttons defined here; used in set-fields-panel and dboard:make-controls
(define *common:std-statuses*
  '(;; (0 "DELETED")
    (1 "n/a")
    (2 "PASS")
    (3 "SKIP")
    (4 "WARN")
    (5 "WAIVED")
    (6 "CHECK")
    (7 "STUCK/DEAD")
    (8 "FAIL")
    (9 "ABORT")))

(define *common:ended-states*       ;; states which indicate the test is stopped and will not proceed
  '("COMPLETED" "ARCHIVED" "KILLED" "KILLREQ" "STUCK" "INCOMPLETE"))

(define *common:badly-ended-states* ;; these roll up as CHECK, i.e. results need to be checked
  '("KILLED" "KILLREQ" "STUCK" "INCOMPLETE" "DEAD"))

;; BBnote: *common:running-states* used from db:set-state-status-and-roll-up-items
(define *common:running-states*     ;; test is either running or can be run
  '("RUNNING" "REMOTEHOSTSTART" "LAUNCHED"))
  '("RUNNING" "REMOTEHOSTSTART" "LAUNCHED" "STARTED"))

(define *common:cant-run-states*    ;; These are stopping conditions that prevent a test from being run
  '("COMPLETED" "KILLED" "UNKNOWN" "INCOMPLETE" "ARCHIVED"))

(define *common:not-started-ok-statuses* ;; if not one of these statuses when in not_started state treat as dead
  '("n/a" "na" "PASS" "FAIL" "WARN" "CHECK" "WAIVED" "DEAD" "SKIP"))

      (getenv "MT_TESTSUITE_NAME")
      (if (string? *toppath* )
          (pathname-file *toppath*)
          #f))) ;; (pathname-file (current-directory)))))

(define common:get-area-name common:get-testsuite-name)

(define (common:get-db-tmp-area)
(define (common:get-db-tmp-area . junk)
  (if *db-cache-path*
      *db-cache-path*
      (if *toppath*
	  (let ((dbpath (create-directory (conc "/tmp/" (current-user-name)
						"/megatest_localdb/"
						(common:get-testsuite-name) "/"
						(string-translate *toppath* "/" ".")) #t)))
	    (set! *db-cache-path* dbpath)
	    dbpath)
      (if *toppath* ;; common:get-create-writeable-dir
	  (handle-exceptions
	      exn
	      (begin
		(debug:print-error 0 *default-log-port* "Couldn't create path to " dbdir)
		(exit 1))
	    (let ((dbpath (common:get-create-writeable-dir
			   (list (conc "/tmp/" (current-user-name)
				       "/megatest_localdb/"
				       (common:get-testsuite-name) "/"
				       (string-translate *toppath* "/" ".")))))) ;;  #t))))
	      (set! *db-cache-path* dbpath)
	      dbpath))
	  #f)))

(define (common:get-area-path-signature)
  (message-digest-string (md5-primitive) *toppath*))

(define (common:get-signature str)
  (message-digest-string (md5-primitive) str))

;;======================================================================
;; E X I T   H A N D L I N G
;;======================================================================

(define (common:run-sync?)
    (and (common:on-homehost?)
	 (args:get-arg "-server")))

;;   (let ((ohh (common:on-homehost?))
;; 	(srv (args:get-arg "-server")))
;;     (and ohh srv)))
    ;; (debug:print-info 0 *default-log-port* "common:run-sync? ohh=" ohh ", srv=" srv)

;;;; run-ids
;;    if #f use *db-local-sync* : or 'local-sync-flags
;;    if #t use timestamps      : or 'timestamps
(define (common:sync-to-megatest.db dbstruct) 
  (let ((start-time         (current-seconds))
	(res                (db:multi-db-sync dbstruct 'new2old)))
    (let ((sync-time (- (current-seconds) start-time)))
      (debug:print-info 3 *default-log-port* "Sync of newdb to olddb completed in " sync-time " seconds pid="(current-process-id))
      (if (common:low-noise-print 30 "sync new to old")
	  (debug:print-info 0 *default-log-port* "Sync of newdb to olddb completed in " sync-time " seconds pid="(current-process-id))))
    res))




(define *wdnum* 0)
(define *wdnum*mutex (make-mutex))
;; currently the primary job of the watchdog is to run the sync back to megatest.db from the db in /tmp
;; if we are on the homehost and we are a server (by definition we are on the homehost if we are a server)

        (debug:print-info 13 *default-log-port* "duration-since-last-sync="duration-since-last-sync)
        (if (and (not *time-to-exit*)
                 (< duration-since-last-sync sync-cool-off-duration))
            (thread-sleep! (- sync-cool-off-duration duration-since-last-sync)))
        (if (not *time-to-exit*)
            (let ((golden-mtdb-mtime (file-modification-time golden-mtpath))
                  (tmp-mtdb-mtime    (file-modification-time tmp-mtpath)))
              (if (> golden-mtdb-mtime tmp-mtdb-mtime)
                  (let ((res (db:multi-db-sync dbstruct 'old2new)))
                    (debug:print-info 13 *default-log-port* "rosync called, " res " records transferred.")))
	      (if (> golden-mtdb-mtime tmp-mtdb-mtime)
		  (if (< golden-mtdb-mtime (- (current-seconds) 3)) ;; file has NOT been touched in past three seconds, this way multiple servers won't fight to sync back
		      (let ((res (db:multi-db-sync dbstruct 'old2new)))
			(debug:print-info 13 *default-log-port* "rosync called, " res " records transferred."))))
              (loop (current-seconds)))
            #t)))
    (debug:print-info 0 *default-log-port* "Exiting readonly-watchdog timer, *time-to-exit* = " *time-to-exit*" pid="(current-process-id)" mtpath="golden-mtpath)))


        
(define (common:writable-watchdog dbstruct)
  (thread-sleep! 0.05) ;; delay for startup
  (let ((legacy-sync (common:run-sync?))
	(debug-mode  (debug:debug-mode 1))
	(last-time   (current-seconds))
        (this-wd-num     (begin (mutex-lock! *wdnum*mutex) (let ((x *wdnum*)) (set! *wdnum* (add1 *wdnum*)) (mutex-unlock! *wdnum*mutex) x))))
    (debug:print-info 2 *default-log-port* "Periodic sync thread started.")
    (debug:print-info 3 *default-log-port* "watchdog starting. legacy-sync is " legacy-sync" pid="(current-process-id)" this-wd-num="this-wd-num)
    (if (and legacy-sync (not *time-to-exit*))
	(let* (;;(dbstruct (db:setup))
	       (mtdb     (dbr:dbstruct-mtdb dbstruct))
	       (mtpath   (db:dbdat-get-path mtdb)))
	  (debug:print-info 0 *default-log-port* "Server running, periodic sync started.")
	  (let loop ()
	    ;; sync for filesystem local db writes
	    ;;
	    (mutex-lock! *db-multi-sync-mutex*)
	    (let* ((need-sync        (>= *db-last-access* *db-last-sync*)) ;; no sync since last write
		   (sync-in-progress *db-sync-in-progress*)
		   (should-sync      (and (not *time-to-exit*)
                                          (> (- (current-seconds) *db-last-sync*) 5))) ;; sync every five seconds minimum
		   (start-time       (current-seconds))
		   (mt-mod-time      (file-modification-time mtpath))
		   (recently-synced  (< (- start-time mt-mod-time) 4))
		   (will-sync        (and (or need-sync should-sync)
					  (not sync-in-progress)
					  (not recently-synced))))
              (debug:print-info 13 *default-log-port* "WD writable-watchdog top of loop.  need-sync="need-sync" sync-in-progress="sync-in-progress" should-sync="should-sync" start-time="start-time" mt-mod-time="mt-mod-time" recently-synced="recently-synced" will-sync="will-sync)
	      ;; (if recently-synced (debug:print-info 0 *default-log-port* "Skipping sync due to recently-synced flag=" recently-synced))
	      ;; (debug:print-info 0 *default-log-port* "need-sync: " need-sync " sync-in-progress: " sync-in-progress " should-sync: " should-sync " will-sync: " will-sync)
	      (if will-sync (set! *db-sync-in-progress* #t))
	      (mutex-unlock! *db-multi-sync-mutex*)
	      (if will-sync
		  (let ((res (common:sync-to-megatest.db dbstruct))) ;; did we sync any data? If so need to set the db touched flag to keep the server alive
		    (if (> res 0) ;; some records were transferred, keep the db alive
			(begin
			  (mutex-lock! *heartbeat-mutex*)
			  (set! *db-last-access* (current-seconds))
			  (mutex-unlock! *heartbeat-mutex*)
			  (debug:print-info 0 *default-log-port* "sync called, " res " records transferred."))
			(debug:print-info 2 *default-log-port* "sync called but zero records transferred"))))
	      (if will-sync
		  (begin
		    (mutex-lock! *db-multi-sync-mutex*)
		    (set! *db-sync-in-progress* #f)
		    (set! *db-last-sync* start-time)
		    (mutex-unlock! *db-multi-sync-mutex*)))
	      (if (and debug-mode
		       (> (- start-time last-time) 60))
		  (begin
		    (set! last-time start-time)
		    (debug:print-info 4 *default-log-port* "timestamp -> " (seconds->time-string (current-seconds)) ", time since start -> " (seconds->hr-min-sec (- (current-seconds) *time-zero*))))))
	    
	    ;; keep going unless time to exit
	    ;;
	    (if (not *time-to-exit*)
		(let delay-loop ((count 0))
                  ;;(debug:print-info 13 *default-log-port* "delay-loop top; count="count" pid="(current-process-id)" this-wd-num="this-wd-num" *time-to-exit*="*time-to-exit*)
                                                            
		  (if (and (not *time-to-exit*)
			   (< count 4)) ;; was 11, changing to 4. 
		      (begin
			(thread-sleep! 1)
			(delay-loop (+ count 1))))
		  (if (not *time-to-exit*) (loop))))
	    (if (common:low-noise-print 30)
		(debug:print-info 0 *default-log-port* "Exiting watchdog timer, *time-to-exit* = " *time-to-exit*" pid="(current-process-id)" this-wd-num="this-wd-num)))))))

;; TODO: for multiple areas, we will have multiple watchdogs; and multiple threads to manage
(define (common:watchdog)
  (debug:print-info 13 *default-log-port* "common:watchdog entered.")
  (if (launch:setup)
  (if (common:on-homehost?)
      (let ((dbstruct (db:setup #t)))
	(debug:print-info 13 *default-log-port* "after db:setup with dbstruct="dbstruct)
	(cond
	 ((dbr:dbstruct-read-only dbstruct)
	  (debug:print-info 13 *default-log-port* "loading read-only watchdog")
	  (common:readonly-watchdog dbstruct))
	 (else
	  (debug:print-info 13 *default-log-port* "loading writable-watchdog.")
	  (common:writable-watchdog dbstruct)))
	(debug:print-info 13 *default-log-port* "watchdog done."))
      (debug:print-info 13 *default-log-port* "no need for watchdog on non-homehost")))
      (if (common:on-homehost?)
	  (let ((dbstruct (db:setup #t)))
	    (debug:print-info 13 *default-log-port* "after db:setup with dbstruct=" dbstruct)
	    (cond
	     ((dbr:dbstruct-read-only dbstruct)
	      (debug:print-info 13 *default-log-port* "loading read-only watchdog")
	      (common:readonly-watchdog dbstruct))
	     (else
	      (debug:print-info 13 *default-log-port* "loading writable-watchdog.")
	      (server:writable-watchdog dbstruct)))
	    (debug:print-info 13 *default-log-port* "watchdog done."))
	  (debug:print-info 13 *default-log-port* "no need for watchdog on non-homehost"))))


(define (std-exit-procedure)
  (on-exit (lambda () 0))
  ;;(debug:print-info 13 *default-log-port* "std-exit-procedure called; *time-to-exit*="*time-to-exit*)
  (let ((no-hurry  (if *time-to-exit* ;; hurry up
		       #f

				  (let ((db (cdr *task-db*)))
				    (if (sqlite3:database? db)
					(begin
					  (sqlite3:interrupt! db)
					  (sqlite3:finalize! db #t)
					  ;; (vector-set! *task-db* 0 #f)
					  (set! *task-db* #f)))))
                              (http-client#close-all-connections!)
                              (if (and *runremote*
                                       (remote-conndat *runremote*))
                                  (begin
                                    (http-client#close-all-connections!))) ;; for http-client
                              ;; (if (and *runremote*
                              ;;          (remote-conndat *runremote*))
                              ;;     (begin
                              ;;       (http-client#close-all-connections!))) ;; for http-client
                              (if (not (eq? *default-log-port* (current-error-port)))
                                  (close-output-port *default-log-port*))
			      (set! *default-log-port* (current-error-port))) "Cleanup db exit thread"))
	  (th2 (make-thread (lambda ()
			      (debug:print 4 *default-log-port* "Attempting clean exit. Please be patient and wait a few seconds...")
			      (if no-hurry
                                  (begin

(define (common:which cmds)
  (if (null? cmds)
      #f
      (let loop ((hed (car cmds))
		 (tal (cdr cmds)))
	(let ((res (with-input-from-pipe (conc "which " hed) read-line)))
	  (if (and (string? res)
		   (file-exists? res))
		   (common:file-exists? res))
	      res
	      (if (null? tal)
		  #f
		  (loop (car tal)(cdr tal))))))))
  
(define (common:get-install-area)
  (let ((exe-path (car (argv))))
    (if (file-exists? exe-path)
    (if (common:file-exists? exe-path)
	(handle-exceptions
	 exn
	 #f
	 (pathname-directory
	  (pathname-directory 
	   (pathname-directory exe-path))))
	#f)))

				   (thread-sleep! delay-time)
				   (common:get-homehost trynum: (- trynum 1)))
				 (begin
				   (mutex-unlock! *homehost-mutex*)
				   (debug:print 0 *default-log-port* "ERROR: Failed to read .homehost file after trying five times. Giving up and exiting, message: "  ((condition-property-accessor 'exn 'message) exn))
				   (exit 1)))
			   (let ((hhf (conc *toppath* "/.homehost")))
			     (if (file-exists? hhf)
			     (if (common:file-exists? hhf)
				 (with-input-from-file hhf read-line)
				 (if (file-write-access? *toppath*)
				     (begin
				       (with-output-to-file hhf
					 (lambda ()
					   (print bestadrs)))
				       (begin

	(cdr hh)
	#f)))

;; do we honor the caches of the config files?
;;
(define (common:use-cache?)
  (let ((res #t)) ;; priority by order of evaluation
    (if *configdat*
    (if *configdat* ;; sillyness here. can't use setup/use-cache to know if we can use the cached files!
	(if (equal? (configf:lookup *configdat* "setup" "use-cache") "no")
	    (set! res #f)
	    (if (equal? (configf:lookup *configdat* "setup" "use-cache") "yes")
		(set! res #t))))
    (if (args:get-arg "-no-cache")(set! res #f)) ;; overrides setting in "setup"
    (if (getenv "MT_USE_CACHE")
	(if (equal? (getenv "MT_USE_CACHE") "yes")

          ((< (+ load (/ (random 250) 1000))         ;; add a random factor to keep from getting in a rut
              (+ best-load (/ (random 250) 1000))  )
           (set! best-load load)
           (set! best-host hostname)))))
     hosts)
    best-host))




(define (common:wait-for-cpuload maxload numcpus waitdelay #!key (count 1000) (msg #f)(remote-host #f))
  (let* ((loadavg (common:get-cpu-load remote-host))
	 (first   (car loadavg))
	 (next    (cadr loadavg))
	 (adjload (* maxload numcpus))
	 (loadjmp (- first next)))
    (cond
     ((and (> first adjload)
	   (> count 0))
      (debug:print-info 0 *default-log-port* "waiting " waitdelay " seconds due to load " first " exceeding max of " adjload (if msg msg ""))
      (debug:print-info 0 *default-log-port* "waiting " waitdelay " seconds due to load " first " exceeding max of " adjload " " (if msg msg ""))
      (thread-sleep! waitdelay)
      (common:wait-for-cpuload maxload numcpus waitdelay count: (- count 1)))
     ((and (> loadjmp numcpus)
	   (> count 0))
      (debug:print-info 0 *default-log-port* "waiting " waitdelay " seconds due to load jump " loadjmp " > numcpus " numcpus (if msg msg ""))
      (thread-sleep! waitdelay)
      (common:wait-for-cpuload maxload numcpus waitdelay count: (- count 1))))))

(define (common:wait-for-homehost-load maxload msg)
  (let* ((hh-dat (if (common:on-homehost?) ;; if we are on the homehost then pass in #f so the calls are local.
                     #f
                     (common:get-homehost)))
         (hh     (if hh-dat (car hh-dat) #f))
         (numcpus (common:get-num-cpus hh)))
    (common:wait-for-normalized-load maxload msg: msg remote-host: hh)))

(define (common:get-num-cpus remote-host)
  (let ((proc (lambda ()
		(let loop ((numcpu 0)
			   (inl    (read-line)))
		  (if (eof-object? inl)
		      numcpu
		      (loop (if (string-match "^processor\\s+:\\s+\\d+$" inl)
				(+ numcpu 1)
				numcpu)
			    (read-line)))))))
    (if remote-host
	(with-input-from-pipe 
	 (conc "ssh " remote-host " cat /proc/cpuinfo")
	 proc)
	(with-input-from-file "/proc/cpuinfo" proc))))

;; wait for normalized cpu load to drop below maxload
;;
(define (common:wait-for-normalized-load maxload #!key (msg #f)(remote-host #f))
  (let ((num-cpus (common:get-num-cpus remote-host)))
    (common:wait-for-cpuload maxload num-cpus 15 msg: msg)))
    (common:wait-for-cpuload maxload num-cpus 15 msg: msg remote-host: remote-host)))

(define (get-uname . params)
  (let* ((uname-res (process:cmd-run->list (conc "uname " (if (null? params) "-a" (car params)))))
	 (uname #f))
    (if (null? (car uname-res))
	"unknown"
	(caar uname-res))))

       (lambda ()
          (for-each (lambda (keyval)
		      (let* ((key   (car keyval))
			     (val   (cdr keyval))
			     (delim (if (string-search whitesp val) 
					"\""
					"")))
			(print (if (member key ignorevars)
			(print (if (or (member key ignorevars)
				       (string-search whitesp key))
				   "# setenv "
				   "setenv ")
			       key " " delim (mungeval val) delim)))
		    envvars)))
     (with-output-to-file (conc fname ".sh")
       (lambda ()
          (for-each (lambda (keyval)
		      (let* ((key (car keyval))
			     (val (cdr keyval))
			     (delim (if (string-search whitesp val) 
					"\""
					"")))
			(print (if (or (member key ignorevars)
				       (string-search whitesp key)
				       (string-search ":" key)) ;; internal only values to be skipped.
				   "# export "
				   "export ")
			       key "=" delim (mungeval val) delim)))
                    envvars)))))

;; set some env vars from an alist, return an alist with original values
;; (("VAR" "value") ...)
(define (alist->env-vars lst)
  (if (list? lst)
      (let ((res '()))
	(for-each (lambda (p)
		    (let* ((var (car  p))
			   (val (cadr p))
			   (prv (get-environment-variable var)))
		      (set! res (cons (list var prv) res))
		      (if val 
			  (setenv var (->string val))
			  (safe-setenv var (->string val))
			  (unsetenv var))))
		  lst)
	res)
      '()))

;; clear vars matching pattern, run proc, set vars back
;; if proc is a string run that string as a command with

  (string-intersperse 
   (map (lambda (x)
          (number->string x 16))
        (map string->number
             (string-split instr)))
   "/"))

(define (common:faux-lock keyname)
  (if (rmt:get-var keyname)
      #f
(define (common:faux-lock keyname #!key (wait-time 8))
  (if (rmt:no-sync-get/default keyname #f) ;; do not be tempted to compare to pid. locking is a one-shot action, if already locked for this pid it doesn't actually count
      (if (> wait-time 0)
	  (begin
	    (thread-sleep! 1)
	    (if (eq? wait-time 1) ;; only one second left, steal the lock
		(begin
		  (debug:print-info 0 *default-log-port* "stealing lock for " keyname)
		  (common:faux-unlock keyname force: #t)))
	    (common:faux-lock keyname wait-time: (- wait-time 1)))
	  #f)
      (begin
        (rmt:set-var keyname (conc (current-process-id)))
        (equal? (conc (current-process-id)) (conc (rmt:get-var keyname))))))
        (rmt:no-sync-set keyname (conc (current-process-id)))
        (equal? (conc (current-process-id)) (conc (rmt:no-sync-get/default keyname #f))))))

(define (common:faux-unlock keyname #!key (force #f))
  (if (or force (equal? (conc (current-process-id)) (conc (rmt:get-var keyname))))
  (if (or force (equal? (conc (current-process-id)) (conc (rmt:no-sync-get/default keyname #f))))
      (begin
        (if (rmt:get-var keyname) (rmt:del-var keyname))
        (if (rmt:no-sync-get/default keyname #f) (rmt:no-sync-del! keyname))
        #t)
      #f))

  
(define (common:in-running-test?)
  (and (args:get-arg "-execute") (get-environment-variable "MT_CMDINFO")))

;; first read ~/views.config if it exists, then read $MTRAH/views.config if it exists
;;
(define (common:load-views-config)
  (let* ((view-cfgdat    (make-hash-table))
	 (home-cfgfile   (conc (get-environment-variable "HOME") "/.mtviews.config"))
	 (mthome-cfgfile (conc *toppath* "/.mtviews.config")))
    (if (file-exists? mthome-cfgfile)
    (if (common:file-exists? mthome-cfgfile)
	(read-config mthome-cfgfile view-cfgdat #t))
    ;; we load the home dir file AFTER the MTRAH file so the user can clobber settings when running the dashboard in read-only areas
    (if (file-exists? home-cfgfile)
    (if (common:file-exists? home-cfgfile)
	(read-config home-cfgfile view-cfgdat #t))
    view-cfgdat))

;;======================================================================
;; Manage pkts, used in servers, tests and likely other contexts so put
;; in common
;;======================================================================

(define common:pkt-spec
  '((server . ((action    . a)
	       (pid       . d)
	       (ipaddr    . i)
	       (port      . p)))
    			  
    (test   . ((cpuuse    . c)
	       (diskuse   . d)
	       (item-path . i)
	       (runname   . r)
	       (state     . s)
	       (target    . t)
	       (status    . u)))))

(define (common:get-pkts-dirs mtconf use-lt)
  (let* ((pktsdirs-str (or (configf:lookup mtconf "setup"  "pktsdirs")
			   (and use-lt
				(conc *toppath* "/lt/.pkts"))))
	 (pktsdirs  (if pktsdirs-str
			(string-split pktsdirs-str " ")
			#f)))
    pktsdirs))

;; use-lt is use linktree "lt" link to find pkts dir
(define (common:with-queue-db mtconf proc #!key (use-lt #f)(toppath-in #f))
  (let* ((pktsdirs (common:get-pkts-dirs mtconf use-lt))
	 (pktsdir  (if pktsdirs (car pktsdirs) #f))
	 (toppath  (or (configf:lookup mtconf "scratchdat" "toppath")
		       toppath-in))
	 (pdbpath  (or (configf:lookup mtconf "setup"  "pdbpath") pktsdir)))
    (cond
     ((not (and  pktsdir toppath pdbpath))
      (debug:print 0 *default-log-port* "ERROR: settings are missing in your megatest.config for area management.")
      (debug:print  0 *default-log-port* "  you need to have pktsdir in the [setup] section."))
     ((not (common:file-exists? pktsdir))
      (debug:print 0 *default-log-port* "ERROR: pkts directory not found " pktsdir))
     ((not (equal? (file-owner pktsdir)(current-effective-user-id)))
      (debug:print 0 *default-log-port* "ERROR: directory " pktsdir " is not owned by " (current-effective-user-name)))
     (else
	(let* ((pdb  (open-queue-db pdbpath "pkts.db"
				    schema: '("CREATE TABLE groups (id INTEGER PRIMARY KEY,groupname TEXT, CONSTRAINT group_constraint UNIQUE (groupname));"))))
	  (proc pktsdirs pktsdir pdb)
	  (dbi:close pdb))))))

(define (common:load-pkts-to-db mtconf)
  (common:with-queue-db
   mtconf
   (lambda (pktsdirs pktsdir pdb)
     (for-each
      (lambda (pktsdir) ;; look at all
	(cond
	 ((not (common:file-exists? pktsdir))
	  (debug:print 0 *default-log-port* "ERROR: packets directory " pktsdir " does not exist."))
	 ((not (directory? pktsdir))
	  (debug:print 0 *default-log-port* "ERROR: packets directory path " pktsdir " is not a directory."))
	 ((not (file-read-access? pktsdir))
	  (debug:print 0 *default-log-port* "ERROR: packets directory path " pktsdir " is not readable."))
	 (else
	  (debug:print-info 0 *default-log-port* "Loading packets found in " pktsdir)
	  (let ((pkts (glob (conc pktsdir "/*.pkt"))))
	    (for-each
	     (lambda (pkt)
	       (let* ((uuid    (cadr (string-match ".*/([0-9a-f]+).pkt" pkt)))
		      (exists  (lookup-by-uuid pdb uuid #f)))
		 (if (not exists)
		     (let* ((pktdat (string-intersperse
				     (with-input-from-file pkt read-lines)
				     "\n"))
			    (apkt   (pkt->alist pktdat))
			    (ptype  (alist-ref 'T apkt)))
		       (add-to-queue pdb pktdat uuid (or ptype 'cmd) #f 0)
		       (debug:print 4 *default-log-port* "Added " uuid " of type " ptype " to queue"))
		     (debug:print 4 *default-log-port* "pkt: " uuid " exists, skipping...")
		     )))
	     pkts)))))
      pktsdirs))))

(define (common:get-pkt-alists pkts)
  (map (lambda (x)
	 (alist-ref 'apkt x)) ;; 'pkta pulls out the alist from the read pkt
       pkts))

;; given list of pkts (alist mode) return list of D cards as Unix epoch, sorted descending
;; also delete duplicates by target i.e. (car pkt)
;;
(define (common:get-pkt-times pkts)
  (delete-duplicates
   (sort 
    (map (lambda (x)
	   `(,(alist-ref 't x) . ,(string->number (alist-ref 'D x))))
	 pkts)
    (lambda (a b)(> (cdr a)(cdr b))))      ;; sort descending
   (lambda (a b)(equal? (car a)(car b))))) ;; remove duplicates by target

(include "common_records.scm")

;; return list (path fullpath configname)
(define (find-config configname #!key (toppath #f))
  (if toppath
      (let ((cfname (conc toppath "/" configname)))
	(if (file-exists? cfname)
	(if (common:file-exists? cfname)
	    (list toppath cfname configname)
	    (list #f      #f     #f)))
      (let* ((cwd (string-split (current-directory) "/")))
	(let loop ((dir cwd))
	  (let* ((path     (conc "/" (string-intersperse dir "/")))
		 (fullpath (conc path "/" configname)))
	    (if (file-exists? fullpath)
	    (if (common:file-exists? fullpath)
		(list path fullpath configname)
		(let ((remcwd (take dir (- (length dir) 1))))
		  (if (null? remcwd)
		      (list #f #f #f) ;;  #f #f) 
		  (loop remcwd)))))))))

(define (config:assoc-safe-add alist key val #!key (metadata #f))

;; apply-wildcards: #t/#f - apply vars from targets with % wildcards to all matching sections
;;
(define (read-config path ht allow-system #!key (environ-patt #f)            (curr-section #f)
		     (sections #f)              (settings (make-hash-table)) (keep-filenames #f)
		     (post-section-procs '())   (apply-wildcards #t))
  (debug:print 9 *default-log-port* "START: " path)
  (if (and (not (port? path))
	   (not (file-exists? path))) ;; for case where we are handed a port
	   (not (common:file-exists? path))) ;; for case where we are handed a port
      (begin 
	(debug:print-info 1 *default-log-port* "read-config - file not found " path " current path: " (current-directory))
	;; WARNING: This is a risky change but really, we should not return an empty hash table if no file read?
	#f) ;; (if (not ht)(make-hash-table) ht))
      (let ((inp        (if (string? path)
			    (open-input-file path)
			      path)) ;; we can be handed a port

							     (full-conf     (if (absolute-pathname? include-file)
										include-file
										(common:nice-path 
										 (conc (if curr-conf-dir
											   curr-conf-dir
											   ".")
										       "/" include-file)))))
							(if (file-exists? full-conf)
							(if (common:file-exists? full-conf)
							    (begin
							      ;; (push-directory conf-dir)
							      (debug:print 9 *default-log-port* "Including: " full-conf)
							      (read-config full-conf res allow-system environ-patt: environ-patt curr-section: curr-section-name sections: sections settings: settings keep-filenames: keep-filenames)
							      ;; (pop-directory)
							      (loop (configf:read-line inp res (calc-allow-system allow-system curr-section-name sections) settings) curr-section-name #f #f))
							    (begin
							      (debug:print '(2 9) #f "INFO: include file " include-file " not found (called from " path ")")
							      (debug:print 2 *default-log-port* "        " full-conf)
							      (loop (configf:read-line inp res (calc-allow-system allow-system curr-section-name sections) settings) curr-section-name #f #f)))))
	       (configf:script-rx ( x include-script params);; handle-exceptions
                                  ;;    exn
                                  ;;    (begin
                                  ;;      (debug:print '(0 2 9) #f "INFO: include from script " include-script " failed.")
                                  ;;      (loop (configf:read-line inp res (calc-allow-system allow-system curr-section-name sections) settings) curr-section-name #f #f))
							 (if (and (file-exists? include-script)(file-execute-access? include-script))
							 (if (and (common:file-exists? include-script)(file-execute-access? include-script))
							     (let* ((new-inp-port (open-input-pipe (conc include-script " " params))))
							       (debug:print '(2 9) *default-log-port* "Including from script output: " include-script)
							      ;;  (print "We got here, calling read-config next. Port is: " new-inp-port)
							       (read-config new-inp-port res allow-system environ-patt: environ-patt curr-section: curr-section-name sections: sections settings: settings keep-filenames: keep-filenames)
							       (close-input-port new-inp-port)
							       (loop (configf:read-line inp res (calc-allow-system allow-system curr-section-name sections) settings) curr-section-name #f #f))
							     (begin

	    (let ((match (assoc var sectdat)))
	      (if match ;; (and match (list? match)(> (length match) 1))
		  (cadr match)
		  #f))
	    ))
      #f))

;; use to have definitive setting:
;;  [foo]
;;  var yes
;;
;;  (configf:var-is? cfgdat "foo" "var" "yes") => #t
;;
(define (configf:var-is? cfgdat section var expected-val)
  (equal? (configf:lookup cfgdat section var) expected-val))

(define configf:lookup config-lookup)
(define configf:read-file read-config)

;; safely look up a value that is expected to be a number, return
;; a default (#f unless provided)
;;
(define (configf:lookup-number cfdat section varname #!key (default #f))
  (let* ((val (configf:lookup *configdat* section varname))
         (res (if val
                  (string->number (string-substitute "\\s+" "" val #t))
                  #f)))
    (cond
     (res  res)
     (val  (debug:print 0 *default-log-port* "ERROR: no number found for [" section "], " varname ", got: " val))
     (else default))))

(define (configf:section-vars cfgdat section)
  (let ((sectdat (hash-table-ref/default cfgdat section '())))
    (if (null? sectdat)
	'()
	(map car sectdat))))

		 (res '()))
	(let ((newres (append res (list (string-substitute (regexp "\n") "\n         " hed #t)))))
	  (if (null? tal)
	      newres
	      (loop (car tal)(cdr tal) newres))))))

(define (configf:file->list fname)
  (if (file-exists? fname)
  (if (common:file-exists? fname)
      (let ((inp (open-input-file fname)))
	(let loop ((inl (read-line inp))
		   (res '()))
	  (if (eof-object? inl)
	      (begin
		(close-input-port inp)
		(reverse res))

;; refdb
;;======================================================================

;; reads a refdb into an assoc array of assoc arrays
;;   returns (list dat msg)
(define (configf:read-refdb refdb-path)
  (let ((sheets-file  (conc refdb-path "/sheet-names.cfg")))
    (if (not (file-exists? sheets-file))
    (if (not (common:file-exists? sheets-file))
	(list #f (conc "ERROR: no refdb found at " refdb-path))
	(if (not (file-read-access? sheets-file))
	    (list #f (conc "ERROR: refdb file not readable at " refdb-path))
	    (let* ((sheets (with-input-from-file sheets-file
			     (lambda ()
			       (let loop ((inl (read-line))
					  (res '()))

  (handle-exceptions
      exn
      #f
    (configf:alist->config
     (with-input-from-file fname read))))

(define (configf:write-alist cdat fname)
    (if (common:faux-lock fname)
        (let* ((dat  (configf:config->alist cdat))
               (res
                (begin
                  (with-output-to-file fname ;; first write out the file
                    (lambda ()
                      (pp dat)))
                  
                  (if (common:file-exists? fname)   ;; now verify it is readable
                      (if (configf:read-alist fname)
                          #t ;; data is good.
                          (begin
                            (handle-exceptions
                             exn
                             #f
                             (debug:print 0 *default-log-port* "WARNING: content " dat " for cache " fname " is not readable. Deleting generated file.")
                             (delete-file fname))
                            #f))
                      #f))))
  (if (not (common:faux-lock fname))
      (debug:print 0 *default-log-port* "INFO: Could not get lock on " fname))
  (let* ((dat  (configf:config->alist cdat))
         (res
          (begin
            (with-output-to-file fname ;; first write out the file
              (lambda ()
                (pp dat)))
            
            (if (common:file-exists? fname)   ;; now verify it is readable
                (if (configf:read-alist fname)
                    #t ;; data is good.
                    (begin
                      (handle-exceptions
                       exn
                       #f
                       (debug:print 0 *default-log-port* "WARNING: content " dat " for cache " fname " is not readable. Deleting generated file.")
                       (delete-file fname))
                      #f))
                #f))))
          
          (common:faux-unlock fname)
          res)
    (common:faux-unlock fname)
    res))
        (begin
          (debug:print 0 *default-log-port* "WARNING: could not get faux-lock on " fname)
          #f)))

  
;; convert hierarchial list to ini format
;;
(define (configf:config->ini data)
  (map 
   (lambda (section)
     (let ((section-name (car section))
	   (section-dat  (cdr section)))

			   (db:test-get-uname testdat))) ;; )
	    )))))

;; if there is a submegatest create a button to launch dashboard in that area
;;
(define (submegatest-panel dbstruct keydat testdat runname testconfig)
  (let* ((subarea (configf:lookup testconfig "setup" "submegatest"))
	 (area-exists (and subarea (file-exists? subarea))))
	 (area-exists (and subarea (common:file-exists? subarea))))
    ;; (debug:print-info 0 *default-log-port* "Megatest subarea=" subarea ", area-exists=" area-exists)
    (if subarea
	(iup:frame 
	 #:title "Megatest Run Info" ; #:expand "YES"
	 (iup:button
	  "Launch Dashboard"
	  #:action (lambda (obj)

				   (cadr keyval))
				 keydat)
			    "/"))
	       (item-path  (db:test-get-item-path testdat))
	       ;; this next block was added to fix a bug where variables were
               ;; needed. Revisit this.
	       (runconfig  (let ((runconfigf (conc  *toppath* "/runconfigs.config"))) ;; no rush but it would be good to convert this call to use runconfig:read
	 		     (if (file-exists? runconfigf)
	 		     (if (common:file-exists? runconfigf)
	 			 (handle-exceptions
                                   exn
                                   #f  ;; do nothing, just keep on trucking ....
                                   (setup-env-defaults runconfigf run-id (make-hash-table) keydat environ-patt: keystring))
	 			 (make-hash-table))))
	       (testconfig    (begin
				;; (runs:set-megatest-env-vars run-id inrunname: runname testname: test-name itempath: item-path)
				(runs:set-megatest-env-vars run-id inkeyvals: keydat inrunname: runname intarget: keystring testname: testname itempath: item-path) ;; these may be needed by the launching process
				(handle-exceptions
				 exn  ;; NOTE: I've no idea why this was written this way. Research, study and fix needed!
				 (tests:get-testconfig (db:test-get-testname testdat) (db:test-get-item-path testdat) test-registry #f)
				 (tests:get-testconfig (db:test-get-testname testdat) item-path test-registry #t))))
	       (viewlog    (lambda (x)
			     (if (file-exists? logfile)
			     (if (common:file-exists? logfile)
					;(system (conc "firefox " logfile "&"))
				 (dcommon:run-html-viewer logfile)
				 (message-window (conc "File " logfile " not found")))))
	       (view-a-log (lambda (lfile) 
			     (let ((lfilename (conc rundir "/" lfile)))
			       ;; (print "lfilename: " lfilename)
			       (if (file-exists? lfilename)
			       (if (common:file-exists? lfilename)
					;(system (conc "firefox " logfile "&"))
				   (dcommon:run-html-viewer lfilename)
				   (message-window (conc "File " lfilename " not found"))))))
	       (xterm      (lambda (x)
			     (if (directory-exists? rundir)
				 (let ((shell (if (get-environment-variable "SHELL") 
						  (conc "-e " (get-environment-variable "SHELL"))
						  "")))
				   (common:without-vars
				    (conc "cd " rundir 
					  ";mt_xterm -T \"" (string-translate testfullname "()" "  ") "\" " shell "&")
				    "MT_.*"))
				 (message-window  (conc "Directory " rundir " not found")))))
	       (widgets    (make-hash-table))
	       (refreshdat (lambda ()
			     (let* ((curr-mod-time (file-modification-time db-path))
				                   ;;     (max ..... (if (file-exists? testdat-path)
				                   ;;     (max ..... (if (common:file-exists? testdat-path)
						   ;;      	      (file-modification-time testdat-path)
						   ;;      	      (begin
						   ;;      		(set! testdat-path (conc rundir "/testdat.db"))
						   ;;      		0))))
				    (need-update   (or (and (>= curr-mod-time db-mod-time)
							    (> (current-milliseconds)(+ last-update 250))) ;; every half seconds if db touched
						       (> (current-milliseconds)(+ last-update 10000))     ;; force update even 10 seconds

(define (dboard:add-external-tab commondat view-name views-cfgdat tabs tab-num)
  (let* ((success #t) ;; at any stage of the process set this flag to #f to skip downstream steps. Intention here is to recover gracefully if user provided tabs fail to load.
	 (source  (configf:lookup views-cfgdat view-name "source"))
	 (viewgen (configf:lookup views-cfgdat view-name "viewgen"))
	 (updater (configf:lookup views-cfgdat view-name "updater"))
	 (result-child #f))
    (if (and (file-exists? source)
    (if (and (common:file-exists? source)
	     (file-read-access? source))
	(handle-exceptions
	 exn
	 (begin
	   (print-call-chain)
	   (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
	   (debug:print 0 *default-log-port* "ERROR: failed to load " source ", try loading in the repl: megatest -repl")

   (common:max (map (lambda (filen)
		      (file-modification-time filen))
		    (glob (conc dbdir "/*.db*"))))))

(define (dashboard:monitor-changed? commondat tabdat)
  (let* ((run-update-time (current-seconds))
	 (monitor-db-path (dboard:tabdat-monitor-db-path tabdat))
	 (monitor-modtime (if (and monitor-db-path (file-exists? monitor-db-path))
	 (monitor-modtime (if (and monitor-db-path (common:file-exists? monitor-db-path))
			      (file-modification-time monitor-db-path)
			      -1)))
    (if (and (eq? (dboard:commondat-curr-tab-num commondat) 0)
	     (or (> monitor-modtime *last-monitor-update-time*)
		 (> (- run-update-time *last-monitor-update-time*) 5))) ;; update every 1/2 minute just in case
	(begin
	  (set! *last-monitor-update-time* run-update-time) ;; monitor-modtime)

;;======================================================================
;; The heavy lifting starts here
;;======================================================================

(define (main)
  (let ((mtdb-path (conc *toppath* "/megatest.db"))) ;; 
    (if (and (file-exists? mtdb-path)
    (if (and (common:file-exists? mtdb-path)
	     (file-write-access? mtdb-path))
	(if (not (args:get-arg "-skip-version-check"))
            (common:exit-on-version-changed)))
    (let* ((commondat       (dboard:commondat-make)))
      ;; Move this stuff to db.scm? I'm not sure that is the right thing to do...
      (cond 
       ((args:get-arg "-test") ;; run-id,test-id

				) "update buttons once"))
	    (th2 (make-thread iup:main-loop "Main loop")))
	(thread-start! th2)
	(thread-join! th2)))))

;; ease debugging by loading ~/.dashboardrc
(let ((debugcontrolf (conc (get-environment-variable "HOME") "/.dashboardrc")))
  (if (file-exists? debugcontrolf)
  (if (common:file-exists? debugcontrolf)
      (load debugcontrolf)))

(if (args:get-arg "-repl")
    (repl)
    (main))

(define (datashare:open-db configdat) 
  (let ((path (configf:lookup configdat "database" "location")))
    (if (and path
	     (directory? path)
	     (file-read-access? path))
	(let* ((dbpath    (conc path "/datashare.db"))
	       (writeable (file-write-access? dbpath))
	       (dbexists  (file-exists? dbpath))
	       (dbexists  (common:file-exists? dbpath))
	       (handler   (make-busy-timeout 136000)))
	  (handle-exceptions
	   exn
	   (begin
	     (debug:print 2 *default-log-port* "ERROR: problem accessing db " dbpath
			  ((condition-property-accessor 'exn 'message) exn))
	     (exit))

		res)))
	(cons 0 #f)
	paths))

;; remove existing link and if possible ...
;; create path to next of tip of target, create link back to source
(define (datashare:build-dir-make-link source target)
  (if (file-exists? target)(datashare:backup-move target))
  (if (common:file-exists? target)(datashare:backup-move target))
  (create-directory (pathname-directory target) #t)
  (create-symbolic-link source target))

(define (datashare:backup-move path)
  (let* ((trashdir  (conc (pathname-directory path) "/.trash"))
	 (trashfile (conc trashdir "/" (current-seconds) "-" (pathname-file path))))
    (create-directory trashdir #t)

  (conc "/" (string-intersperse (map conc pathlst) "/")))

(define (datashare:path->lst path)
  (string-split path "/"))

(define (datashare:pathdat-apply-heuristics configdat path)
  (cond
   ((file-exists? path) "found")
   ((common:file-exists? path) "found")
   (else (conc path " not installed"))))

(define (datashare:get-view configdat)
  (iup:vbox
   (iup:hbox
    (let* ((label-size     "60x")
	   ;; filter elements

        (set! (current-effective-user-id) eid))))

(define (datashare:find name paths)
  (if (null? paths)
      #f
      (let loop ((hed (car paths))
		 (tal (cdr paths)))
	(if (file-exists? (conc hed "/" name))
	(if (common:file-exists? (conc hed "/" name))
	    hed
	    (if (null? tal)
		#f
		(loop (car tal)(cdr tal)))))))

;;======================================================================
;; MAIN
;;======================================================================

(define (datashare:load-config exe-dir exe-name)
  (let* ((fname   (conc exe-dir "/." exe-name ".config")))
    (ini:property-separator-patt " *  *")
    (ini:property-separator #\space)
    (if (file-exists? fname)
    (if (common:file-exists? fname)
	;; (ini:read-ini fname)
	(read-config fname #f #t)
	(make-hash-table))))

(define (datashare:process-action configdat action . args)
  (case (string->symbol action)
    ((get)

			  (conc "\"" (vector-ref x 4) "\""))
		  (print (vector-ref x 0))))
	    versions)
       (sqlite3:finalize! db)))))

;; ease debugging by loading ~/.dashboardrc - REMOVE FROM PRODUCTION!
(let ((debugcontrolf (conc (get-environment-variable "HOME") "/.datasharerc")))
  (if (file-exists? debugcontrolf)
  (if (common:file-exists? debugcontrolf)
      (load debugcontrolf)))

(define (main)
  (let* ((args      (argv))
	 (prog      (car args))
	 (rema      (cdr args))
	 (exe-name  (pathname-file (car (argv))))

;;     (filedb:get-path db id)))

;; NB// #f => return dbdir only
;;      (was planned to be;  zeroth db with name=main.db)
;; 
;; If run-id is #f return to create and retrieve the path where the db will live.
;;
(define (db:dbfile-path . junk) ;;  run-id)
  (let* ((dbdir           (common:get-db-tmp-area)))
(define db:dbfile-path common:get-db-tmp-area)
    (handle-exceptions
     exn
     (begin
       (debug:print-error 0 *default-log-port* "Couldn't create path to " dbdir)
       (exit 1))
     (if (not (directory? dbdir))(create-directory dbdir #t)))
    dbdir))

(define (db:set-sync db)
  (let ((syncprag (configf:lookup *configdat* "setup" "sychronous")))
    (sqlite3:execute db (conc "PRAGMA synchronous = " (or syncprag 0) ";")))) 

;; open an sql database inside a file lock
;; returns: db existed-prior-to-opening
;; RA => Returns a db handler; sets the lock if opened in writable mode
;;
;;(define *db-open-mutex* (make-mutex))

(define (db:lock-create-open fname initproc)
  (let* ((parent-dir   (or (pathname-directory fname)(current-directory))) ;; no parent? go local
         (raw-fname    (pathname-file fname))
	 (dir-writable (file-write-access? parent-dir))
	 (file-exists  (file-exists? fname))
	 (file-exists  (common:file-exists? fname))
	 (file-write   (if file-exists
			   (file-write-access? fname)
			   dir-writable )))
    ;;(mutex-lock! *db-open-mutex*) ;; tried this mutex, not clear it helped.
    (if file-write ;; dir-writable
	(condition-case
         (let* ((lockfname   (conc fname ".lock"))
                (readyfname  (conc parent-dir "/.ready-" raw-fname))
                (readyexists (file-exists? readyfname)))
                (readyexists (common:file-exists? readyfname)))
           (if (not readyexists)
               (common:simple-file-lock-and-wait lockfname))
           (let ((db      (sqlite3:open-database fname)))
             (sqlite3:set-busy-handler! db (make-busy-timeout 136000))
             (sqlite3:execute db "PRAGMA synchronous = 0;")
             (if (not file-exists)
                 (begin

;; ;; This routine creates the db. It is only called if the db is not already opened
;; ;; 
;; (define (db:open-rundb dbstruct run-id #!key (attemptnum 0)(do-not-open #f)) ;;  (conc *toppath* "/megatest.db") (car *configinfo*)))
;;   (let* ((dbfile       (db:dbfile-path run-id)) ;; (conc toppath "/db/" run-id ".db"))
;;          (dbexists     (file-exists? dbfile))
;;          (dbexists     (common:file-exists? dbfile))
;;          (db           (db:lock-create-open dbfile (lambda (db)
;;                                                      (handle-exceptions
;;                                                       exn
;;                                                       (begin
;;                                                         ;; (release-dot-lock dbpath)
;;                                                         (if (> attemptnum 2)
;;                                                             (debug:print-error 0 *default-log-port* "tried twice, cannot create/initialize db for run-id " run-id ", at path " dbpath)

;; This routine creates the db if not already present. It is only called if the db is not already opened
;;
(define (db:open-db dbstruct #!key (areapath #f)(do-sync #t)) ;; TODO: actually use areapath
  (let ((tmpdb-stack (dbr:dbstruct-dbstack dbstruct))) ;; RA => Returns the first reference in dbstruct
    (if (stack? tmpdb-stack)
	(db:get-db tmpdb-stack) ;; get previously opened db (will create new db handle if all in the stack are already used
        (let* ((dbpath       (db:dbfile-path ))      ;; path to tmp db area
               (dbexists     (file-exists? dbpath))
               (dbexists     (common:file-exists? dbpath))
	       (tmpdbfname   (conc dbpath "/megatest.db"))
	       (dbfexists    (file-exists? tmpdbfname))  ;; (conc dbpath "/megatest.db")))
               (mtdbexists   (file-exists? (conc *toppath* "/megatest.db")))
	       (dbfexists    (common:file-exists? tmpdbfname))  ;; (conc dbpath "/megatest.db")))
               (mtdbexists   (common:file-exists? (conc *toppath* "/megatest.db")))
               
               (mtdb         (db:open-megatest-db))
               (mtdbpath     (db:dbdat-get-path mtdb))
               (tmpdb        (db:open-megatest-db path: dbpath)) ;; lock-create-open dbpath db:initialize-main-db))
               (refndb       (db:open-megatest-db path: dbpath name: "megatest_ref.db"))
               (write-access (file-write-access? mtdbpath))
	       (mtdbmodtime  (if mtdbexists (common:lazy-sqlite-db-modification-time mtdbpath)   #f))

;; Open the classic megatest.db file (defaults to open in toppath)
;;
;;   NOTE: returns a dbdat not a dbstruct!
;;
(define (db:open-megatest-db #!key (path #f)(name #f))
  (let* ((dbdir        (or path *toppath*))
         (dbpath       (conc  dbdir "/" (or name "megatest.db")))
	 (dbexists     (file-exists? dbpath))
	 (dbexists     (common:file-exists? dbpath))
	 (db           (db:lock-create-open dbpath
					    (lambda (db)
                                              (db:initialize-main-db db)
					      ;;(db:initialize-run-id-db db)
					      )))
	 (write-access (file-write-access? dbpath)))
    (debug:print-info 13 *default-log-port* "db:open-megatest-db "dbpath)

      (mutex-unlock! *db-multi-sync-mutex*)
      (db:sync-tables (db:sync-all-tables-list dbstruct) update_info tmpdb refndb mtdb))
    (mutex-lock! *db-multi-sync-mutex*)
    (set! *db-last-sync* start-t)
    (set! *db-last-access* start-t)
    (mutex-unlock! *db-multi-sync-mutex*)
    (stack-push! (dbr:dbstruct-dbstack dbstruct) tmpdb)))

(define (db:safely-close-sqlite3-db db #!key (try-num 3))
  (if (<= try-num 0)
      #f
      (handle-exceptions
	  exn
	  (begin
	    (thread-sleep! 3)
	    (sqlite3:interrupt! db)
	    (db:safely-close-sqlite3-db db try-num: (- try-num 1)))
	(if (sqlite3:database? db)
	    (begin
	      (sqlite3:finalize! db)
	      #t)
	    #f))))

;; close all opened run-id dbs
(define (db:close-all dbstruct)
  (if (dbr:dbstruct? dbstruct)
      (handle-exceptions
	  exn
	  (begin
	    (debug:print 0 *default-log-port* "WARNING: Finalizing failed, "  ((condition-property-accessor 'exn 'message) exn))
	    (print-call-chain *default-log-port*))
	;; (db:sync-touched dbstruct 0 force-sync: #t) ;; NO. Do not do this here. Instead we rely on a server to be started when there are writes, even if the server itself is not going to be used as a server.
        (let ((tdbs (map db:dbdat-get-db 
                         (stack->list (dbr:dbstruct-dbstack dbstruct))))
              (mdb (db:dbdat-get-db (dbr:dbstruct-mtdb   dbstruct)))
              (rdb (db:dbdat-get-db (dbr:dbstruct-refndb dbstruct))))
          (map (lambda (db)
		 (db:safely-close-sqlite3-db db))
		 (if (sqlite3:database? db)
		     (sqlite3:finalize! db)))
;; 		 (if (sqlite3:database? db)
;; 		     (sqlite3:finalize! db)))
	       tdbs)
          (if (sqlite3:database? mdb) (sqlite3:finalize! mdb))
          (if (sqlite3:database? rdb) (sqlite3:finalize! rdb))))))
          (db:safely-close-sqlite3-db mdb)     ;; (if (sqlite3:database? mdb) (sqlite3:finalize! mdb))
          (db:safely-close-sqlite3-db rdb))))) ;; (if (sqlite3:database? rdb) (sqlite3:finalize! rdb))))))

;;   (let ((locdbs (dbr:dbstruct-locdbs dbstruct)))
;;     (if (hash-table? locdbs)
;; 	(for-each (lambda (run-id)
;; 		    (db:close-run-db dbstruct run-id))
;; 		  (hash-table-keys locdbs)))))

	 (fname    (pathname-strip-directory dbpath))
	 (fnamejnl (conc fname "-journal"))
	 (tmpname  (conc fname "." (current-process-id)))
	 (tmpjnl   (conc fnamejnl "." (current-process-id))))
    (debug:print-error 0 *default-log-port* "" fname " appears corrupted. Making backup \"old/" fname "\"")
    (system (conc "cd " dbdir ";mkdir -p old;cat " fname " > old/" tmpname))
    (system (conc "rm -f " dbpath))
    (if (file-exists? fnamejnl)
    (if (common:file-exists? fnamejnl)
	(begin
	  (debug:print-error 0 *default-log-port* "" fnamejnl " found, moving it to old dir as " tmpjnl)
	  (system (conc "cd " dbdir ";mkdir -p old;cat " fnamejnl " > old/" tmpjnl))
	  (system (conc "rm -f " dbdir "/" fnamejnl))))
    ;; attempt to recreate database
    (system (conc "cd " dbdir ";sqlite3 old/" tmpname " .dump | sqlite3 " fname))))
    

(define (db:sync-tables tbls last-update fromdb todb . slave-dbs)
  (handle-exceptions
   exn
   (begin
     (debug:print 0 *default-log-port* "EXCEPTION: database probably overloaded or unreadable in db:sync-tables.")
     (print-call-chain (current-error-port))
     (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
     (print "exn=" (condition->list exn))
     (debug:print 5 *default-log-port* "exn=" (condition->list exn))
     (debug:print 0 *default-log-port* " status:  " ((condition-property-accessor 'sqlite3 'status) exn))
     (debug:print 0 *default-log-port* " src db:  " (db:dbdat-get-path fromdb))
     (for-each (lambda (dbdat)
		 (let ((dbpath (db:dbdat-get-path dbdat)))
		   (debug:print 0 *default-log-port* " dbpath:  " dbpath)
		   (if (not (db:repair-db dbdat))
		       (begin

     (let ((stmts       (make-hash-table)) ;; table-field => stmt
	   (all-stmts   '())              ;; ( ( stmt1 value1 ) ( stml2 value2 ))
	   (numrecs     (make-hash-table))
	   (start-time  (current-milliseconds))
	   (tot-count   0))
       (for-each ;; table
	(lambda (tabledat)
	  (let* ((tablename  (car tabledat))
		 (fields     (cdr tabledat))
		 (use-last-update  (if last-update
				       (if (pair? last-update)
					   (member (car last-update)    ;; last-update field name
						   (map car fields))
					   (begin
					     (debug:print 0 *default-log-port* "ERROR: parameter last-update for db:sync-tables must be a pair, received: " last-update) ;; found in fields
					     #f))
				       #f))
	  (let* ((tablename        (car tabledat))
		 (fields           (cdr tabledat))
		 (has-last-update  (member "last_update" fields))
		 (use-last-update  (cond
				    ((and has-last-update
					  (member "last_update" fields))
				     #t) ;; if given a number, just use it for all fields
				    ((number? last-update) #f) ;; if not matched first entry then ignore last-update for this table
				    ((and (pair? last-update)
					  (member (car last-update)    ;; last-update field name
						  (map car fields))) #t)
				    (last-update
				     (debug:print 0 *default-log-port* "ERROR: parameter last-update for db:sync-tables must be a pair or a number, received: " last-update) ;; found in fields
				     #f)
				    (else
				     #f)))
		 (last-update-value (if use-last-update ;; no need to check for has-last-update - it is already accounted for
					(if (number? last-update)
					    last-update
					    (cdr last-update))
					#f))
		 (last-update-field (if use-last-update
					(if (number? last-update)
					    "last_update"
					    (car last-update))
					#f))
		 (num-fields (length fields))
		 (field->num (make-hash-table))
		 (num->field (apply vector (map car fields)))
		 (full-sel   (conc "SELECT " (string-intersperse (map car fields) ",") 
				   " FROM " tablename (if use-last-update ;; apply last-update criteria
							  (conc " " (car last-update) ">=" (cdr last-update))
							  (conc " WHERE " last-update-field " >= " last-update-value)
							  "")
				   ";"))
		 (full-ins   (conc "INSERT OR REPLACE INTO " tablename " ( " (string-intersperse (map car fields) ",") " ) "
				   " VALUES ( " (string-intersperse (make-list num-fields "?") ",") " );"))
		 (fromdat    '())
		 (fromdats   '())
		 (totrecords 0)

;; return the target db handle so it can be used
;;
(define (db:cache-for-read-only source target #!key (use-last-update #f))
  (if (and (hash-table-ref/default *global-db-store* target #f)
	   (>= (file-modification-time target)(file-modification-time source)))
      (hash-table-ref *global-db-store* target)
      (let* ((toppath   (launch:setup))
	     (targ-db-last-mod (if (file-exists? target)
	     (targ-db-last-mod (if (common:file-exists? target)
				   (file-modification-time target)
				   0))
	     (cache-db  (or (hash-table-ref/default *global-db-store* target #f)
			    (db:open-megatest-db path: target)))
	     (source-db (db:open-megatest-db path: source))
	     (curr-time (current-seconds))
	     (res      '())
	     (last-update (if use-last-update (cons "last_update" targ-db-last-mod) #f)))
	(db:sync-tables (db:sync-main-list source-db) last-update source-db cache-db)
	(db:sync-tables db:sync-tests-only last-update source-db cache-db)
	(hash-table-set! *global-db-store* target cache-db)
	cache-db)))

;; call a proc with a cached db
;;
(define (db:call-with-cached-db proc . params)
  ;; first cache the db in /tmp
  (let* ((cname-part (conc "megatest_cache/" (common:get-testsuite-name)))
	 (fname      (conc  (common:get-area-path-signature) ".db"))
	 (cache-dir  (common:get-create-writeable-dir
		      (list (conc "/tmp/" (current-user-name) "/" cname-part)
			    (conc "/tmp/" (current-user-name) "-" cname-part)
			     (conc "/tmp/" (current-user-name) "_" cname-part))))
	 (megatest-db (conc *toppath* "/megatest.db")))
    ;; (debug:print-info 0 *default-log-port* "Using cache dir " cache-dir)
    (if (not cache-dir)
	(begin
	  (debug:print 0 *default-log-port* "ERROR: Failed to find an area to write the cache db")
	  (exit 1))
	(let* ((th1      (make-thread
			  (lambda ()
			    (if (and (file-exists? megatest-db)
				     (file-write-access? megatest-db))
				(begin
				  (common:sync-to-megatest.db 'timestamps) ;; internally mutexes on *db-local-sync*
				  (debug:print-info 2 *default-log-port* "Done syncing to megatest.db"))))
			  "call-with-cached-db sync-to-megatest.db"))
	       (cache-db (db:cache-for-read-only
			  megatest-db
			  (conc cache-dir "/" fname)
			  use-last-update: #t)))
	  (thread-start! th1)
	  (apply proc cache-db params)
	  ))))
;; ;; call a proc with a cached db
;; ;;
;; (define (db:call-with-cached-db proc . params)
;;   ;; first cache the db in /tmp
;;   (let* ((cname-part (conc "megatest_cache/" (common:get-testsuite-name)))
;; 	 (fname      (conc  (common:get-area-path-signature) ".db"))
;; 	 (cache-dir  (common:get-create-writeable-dir
;; 		      (list (conc "/tmp/" (current-user-name) "/" cname-part)
;; 			    (conc "/tmp/" (current-user-name) "-" cname-part)
;; 			     (conc "/tmp/" (current-user-name) "_" cname-part))))
;; 	 (megatest-db (conc *toppath* "/megatest.db")))
;;     ;; (debug:print-info 0 *default-log-port* "Using cache dir " cache-dir)
;;     (if (not cache-dir)
;; 	(begin
;; 	  (debug:print 0 *default-log-port* "ERROR: Failed to find an area to write the cache db")
;; 	  (exit 1))
;; 	(let* ((th1      (make-thread
;; 			  (lambda ()
;; 			    (if (and (common:file-exists? megatest-db)
;; 				     (file-write-access? megatest-db))
;; 				(begin
;; 				  (db:sync-to-megatest.db dbstruct 'timestamps) ;; internally mutexes on *db-local-sync*
;; 				  (debug:print-info 2 *default-log-port* "Done syncing to megatest.db"))))
;; 			  "call-with-cached-db sync-to-megatest.db"))
;; 	       (cache-db (db:cache-for-read-only
;; 			  megatest-db
;; 			  (conc cache-dir "/" fname)
;; 			  use-last-update: #t)))
;; 	  (thread-start! th1)
;; 	  (apply proc cache-db params)
;; 	  ))))

;; options:
;;
;;  'killservers  - kills all servers
;;  'dejunk       - removes junk records
;;  'adj-testids  - move test-ids into correct ranges
;;  'old2new      - sync megatest.db to /tmp/.../megatest.db and /tmp/.../megatest_ref.db
;;  'new2old      - sync /tmp/.../megatest.db to megatest.db and /tmp/.../megatest_ref.db (and update data_synced)
;;  'closeall     - close all opened dbs
;;  'schema       - attempt to apply schema changes
;;  run-ids: '(1 2 3 ...) or #f (for all)
;;
(define (db:multi-db-sync dbstruct . options)
  (if (not (launch:setup))
      (debug:print 0 *default-log-port* "ERROR: not able to setup up for megatest.")
      (let* ((mtdb     (dbr:dbstruct-mtdb dbstruct))
	     (tmpdb    (db:get-db dbstruct))
             (refndb   (dbr:dbstruct-refndb dbstruct))
	     (allow-cleanup #t) ;; (if run-ids #f #t))
	     (servers (server:get-list *toppath*)) ;; (tasks:get-all-servers (db:delay-if-busy tdbdat)))
	     (data-synced 0)) ;; count of changed records (I hope)

	(for-each
	 (lambda (option)

	   (case option
	     ;; kill servers
	     ((killservers)
	      (for-each
	       (lambda (server)
		 (match-let (((mod-time host port start-time pid) server))
			    (if (and host pid)
				(tasks:kill-server host pid))))
	       servers))

	     ;; clear out junk records
	     ;;
	     ((dejunk)
	      (db:delay-if-busy mtdb) ;; ok to delay on mtdb
	      (db:clean-up mtdb)
	      (db:clean-up tmpdb)
              (db:clean-up refndb))
  ;; (if (not (launch:setup))
  ;;    (debug:print 0 *default-log-port* "ERROR: not able to setup up for megatest.")
  (let* ((mtdb     (dbr:dbstruct-mtdb dbstruct))
	 (tmpdb    (db:get-db dbstruct))
	 (refndb   (dbr:dbstruct-refndb dbstruct))
	 (allow-cleanup #t) ;; (if run-ids #f #t))
	 (servers (server:get-list *toppath*)) ;; (tasks:get-all-servers (db:delay-if-busy tdbdat)))
	 (data-synced 0)) ;; count of changed records (I hope)
    
    (for-each
     (lambda (option)
       
       (case option
	 ;; kill servers
	 ((killservers)
	  (for-each
	   (lambda (server)
	     (match-let (((mod-time host port start-time pid) server))
	       (if (and host pid)
		   (tasks:kill-server host pid))))
	   servers))
	 
	 ;; clear out junk records
	 ;;
	 ((dejunk)
	  (db:delay-if-busy mtdb) ;; ok to delay on mtdb
	  (db:clean-up mtdb)
	  (db:clean-up tmpdb)
	  (db:clean-up refndb))

	     ;; sync runs, test_meta etc.
	     ;;
	     ((old2new)
	      (set! data-synced
		    (+ (db:sync-tables (db:sync-all-tables-list dbstruct) #f mtdb tmpdb refndb)
		       data-synced)))
	     
	     ;; now ensure all newdb data are synced to megatest.db
	     ;; do not use the run-ids list passed in to the function
	     ;;
	     ((new2old)
	      (set! data-synced
		    (+ (db:sync-tables (db:sync-all-tables-list dbstruct) #f tmpdb refndb mtdb)
		       data-synced)))
	 ;; sync runs, test_meta etc.
	 ;;
	 ((old2new)
	  (set! data-synced
	    (+ (db:sync-tables (db:sync-all-tables-list dbstruct) #f mtdb tmpdb refndb)
	       data-synced)))
	 
	 ;; now ensure all newdb data are synced to megatest.db
	 ;; do not use the run-ids list passed in to the function
	 ;;
	 ((new2old)
	  (set! data-synced
	    (+ (db:sync-tables (db:sync-all-tables-list dbstruct) #f tmpdb refndb mtdb)
	       data-synced)))

	     ((adj-target)
	      (db:adj-target (db:dbdat-get-db mtdb))
	      (db:adj-target (db:dbdat-get-db tmpdb))
	      (db:adj-target (db:dbdat-get-db refndb)))
	   
	     ((schema)
              (db:patch-schema-maindb (db:dbdat-get-db mtdb))
              (db:patch-schema-maindb (db:dbdat-get-db tmpdb))
              (db:patch-schema-maindb (db:dbdat-get-db refndb))
              (db:patch-schema-rundb  (db:dbdat-get-db mtdb))
              (db:patch-schema-rundb  (db:dbdat-get-db tmpdb))
              (db:patch-schema-rundb  (db:dbdat-get-db refndb))))
	
	   (stack-push! (dbr:dbstruct-dbstack dbstruct) tmpdb))
	 options)
	data-synced)))
	 ((adj-target)
	  (db:adj-target (db:dbdat-get-db mtdb))
	  (db:adj-target (db:dbdat-get-db tmpdb))
	  (db:adj-target (db:dbdat-get-db refndb)))
	 
	 ((schema)
	  (db:patch-schema-maindb (db:dbdat-get-db mtdb))
	  (db:patch-schema-maindb (db:dbdat-get-db tmpdb))
	  (db:patch-schema-maindb (db:dbdat-get-db refndb))
	  (db:patch-schema-rundb  (db:dbdat-get-db mtdb))
	  (db:patch-schema-rundb  (db:dbdat-get-db tmpdb))
	  (db:patch-schema-rundb  (db:dbdat-get-db refndb))))
       
       (stack-push! (dbr:dbstruct-dbstack dbstruct) tmpdb))
     options)
    data-synced))

(define (db:tmp->megatest.db-sync dbstruct last-update)
  (let* ((mtdb        (dbr:dbstruct-mtdb dbstruct))
	 (tmpdb       (db:get-db dbstruct))
	 (refndb      (dbr:dbstruct-refndb dbstruct)))
    (db:sync-tables (db:sync-all-tables-list dbstruct) last-update tmpdb refndb mtdb)))

;;;; run-ids
;;    if #f use *db-local-sync* : or 'local-sync-flags
;;    if #t use timestamps      : or 'timestamps
(define (db:sync-to-megatest.db dbstruct #!key (no-sync-db #f)) 
  (let* ((start-time         (current-seconds))
	 (last-update        (if no-sync-db
				 (db:no-sync-get/default no-sync-db "LAST_UPDATE" 0)
				 0)) ;; (or (db:get-var dbstruct "LAST_UPDATE") 0))
	 (sync-needed        (> (- start-time last-update) 6))
	 (res                (if sync-needed ;; don't sync if a sync already occurred in the past 6 seconds
				 (begin
				   (if no-sync-db
				       (db:no-sync-set no-sync-db "LAST_UPDATE" start-time))
				   (db:tmp->megatest.db-sync dbstruct last-update))
				 0))
	 (sync-time           (- (current-seconds) start-time)))
      (debug:print-info 3 *default-log-port* "Sync of newdb to olddb completed in " sync-time " seconds pid="(current-process-id))
      (if (common:low-noise-print 30 "sync new to old")
          (if sync-needed
              (debug:print-info 0 *default-log-port* "Sync of " res " records from newdb to olddb completed in " sync-time " seconds pid="(current-process-id))
              (debug:print-info 0 *default-log-port* "No sync needed, last updated " (- start-time last-update) " seconds ago")))
      res))

;; keeping it around for debugging purposes only
(define (open-run-close-no-exception-handling  proc idb . params)
  (debug:print-info 11 *default-log-port* "open-run-close-no-exception-handling START given a db=" (if idb "yes " "no ") ", params=" params)
  (print "I don't work anymore. open-run-close-no-exception-handling needs fixing or removing...")
  (exit)
  (if (or *db-write-access*

	 (err-status ((condition-property-accessor 'sqlite3 'status #f) exn)))
     (case err-status
       ((busy)
	(thread-sleep! sleep-time))
       (else
	(debug:print 0 *default-log-port* "EXCEPTION: database probably overloaded or unreadable.")
	(debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
	(print "exn=" (condition->list exn))
	(debug:print 5 *default-log-port* "exn=" (condition->list exn))
	(debug:print 0 *default-log-port* " status:  " ((condition-property-accessor 'sqlite3 'status) exn))
	(print-call-chain (current-error-port))
	(thread-sleep! sleep-time)
	(debug:print-info 0 *default-log-port* "trying db call one more time....this may never recover, if necessary kill process " (current-process-id) " on host " (get-host-name) " to clean up")))
     (apply open-run-close-exception-handling proc idb params))
   (apply open-run-close-no-exception-handling proc idb params)))

;;======================================================================
;; L O G G I N G    D B 
;;======================================================================

(define (open-logging-db)
  (let* ((dbpath    (conc (if *toppath* (conc *toppath* "/") "") "logging.db")) ;; fname)
	 (dbexists  (file-exists? dbpath))
	 (dbexists  (common:file-exists? dbpath))
	 (db        (sqlite3:open-database dbpath))
	 (handler   (make-busy-timeout (if (args:get-arg "-override-timeout")
					   (string->number (args:get-arg "-override-timeout"))
					   136000)))) ;; 136000)))
    (sqlite3:set-busy-handler! db handler)
    (if (not dbexists)
	(begin

       ;; These are defunct tests, do not do all the overhead of set-state-status. Force them to INCOMPLETE.
       ;;
       ;; (db:delay-if-busy dbdat)
       (let* (;; (min-incompleted (filter (lambda (x)
              ;;      		      (let* ((testpath (cadr x))
              ;;      			     (tdatpath (conc testpath "/testdat.db"))
              ;;      			     (dbexists (file-exists? tdatpath)))
              ;;      			     (dbexists (common:file-exists? tdatpath)))
              ;;      			(or (not dbexists) ;; if no file then something wrong - mark as incomplete
              ;;      			    (> (- (current-seconds)(file-modification-time tdatpath)) 600)))) ;; no change in 10 minutes to testdat.db - she's dead Jim
              ;;      		    incompleted))
              (min-incompleted-ids (map car incompleted)) ;; do 'em all
              (all-ids             (append min-incompleted-ids (map car oldlaunched))))
         (if (> (length all-ids) 0)
             (begin

	      (lambda (db)
		(sqlite3:execute db "INSERT OR REPLACE INTO metadat (var,val) VALUES (?,?);" var val))))

(define (db:del-var dbstruct var)
  (db:with-db dbstruct #f #t 
	      (lambda (db)
		(sqlite3:execute db "DELETE FROM metadat WHERE var=?;" var))))

;;======================================================================
;; no-sync.db - small bits of data to be shared between servers
;;======================================================================

(define (db:open-no-sync-db)
  (let* ((dbpath (db:dbfile-path))
	 (dbname (conc dbpath "/no-sync.db"))
	 (db     (sqlite3:open-database dbname)))
    (sqlite3:set-busy-handler! db (make-busy-timeout 136000))
    (sqlite3:execute db "PRAGMA synchronous = 0;")
    (sqlite3:execute db "CREATE TABLE IF NOT EXISTS no_sync_metadat (var TEXT,val TEXT, CONSTRAINT no_sync_metadat_constraint UNIQUE (var));")
    db))

;; if we are not a server create a db handle. this is not finalized
;; so watch for problems. I'm still not clear if it is needed to manually
;; finalize sqlite3 dbs with the sqlite3 egg.
;;
(define (db:no-sync-db db-in)
  (if db-in
      db-in
      (let ((db (db:open-no-sync-db)))
	(set! *no-sync-db* db)
	db)))

(define (db:no-sync-set db var val)
  (sqlite3:execute (db:no-sync-db db) "INSERT OR REPLACE INTO no_sync_metadat (var,val) VALUES (?,?);" var val))

(define (db:no-sync-del! db var)
  (sqlite3:execute (db:no-sync-db db) "DELETE FROM no_sync_metadat WHERE var=?;" var))

(define (db:no-sync-get/default db var default)
  (let ((res default))
    (sqlite3:for-each-row
     (lambda (val)
       (set! res val))
     (db:no-sync-db db)
     "SELECT val FROM no_sync_metadat WHERE var=?;"
     var)
    (if res
        (let ((newres (if (string? res)
			  (string->number res)
			  #f)))
          (if newres
              newres
              res))
        res)))

(define (db:no-sync-close-db db)
  (db:safely-close-sqlite3-db db))

;; use a global for some primitive caching, it is just silly to
;; re-read the db over and over again for the keys since they never
;; change

;; why get the keys from the db? why not get from the *configdat*
;; using keys:config-get-fields?

	res)))

;; look up values in a header/data structure
(define (db:get-value-by-header row header field)
  (if (or (null? header) (not row))
      #f
      (let loop ((hed (car header))
		 (tal (cdr header))
		 (n   0))
	(if (equal? hed field)
	    (vector-ref row n)
                 (tal (cdr header))
                 (n   0))
        (if (equal? hed field)
            (handle-exceptions
             exn
             (begin
               (debug:print 0 *default-log-port* "WARNING: attempt to read non-existant field, row=" row " header=" header " field=" field)
               #f)
             (vector-ref row n))
	    (if (null? tal) #f (loop (car tal)(cdr tal)(+ n 1)))))))

;; Accessors for the header/data structure
;; get rows and header from 
(define (db:get-header vec)(vector-ref vec 0))
(define (db:get-rows   vec)(vector-ref vec 1))

			     0                                              ;; 2 ;; Tolerance
			     "n/a"					    ;; 3 ;; Units
			     (configf:lookup dat entry-name "message")      ;; 4 ;; Comment
			     (configf:lookup dat entry-name "exit-status")  ;; 5 ;; Status
			     "logpro"                                       ;; 6 ;; Type
			     ))))
	   (let* ((value     (or (configf:lookup dat entry-name "measured")  "n/a"))
		  (expected  (or (configf:lookup dat entry-name "expected")  "n/a"))
		  (tolerance (or (configf:lookup dat entry-name "tolerance") "n/a"))
		  (expected  (or (configf:lookup dat entry-name "expected")  0.0))
		  (tolerance (or (configf:lookup dat entry-name "tolerance") 0.0))
		  (comment   (or (configf:lookup dat entry-name "comment")
				 (configf:lookup dat entry-name "desc")      "n/a"))
		  (status    (or (configf:lookup dat entry-name "status")    "n/a"))
		  (type      (or (configf:lookup dat entry-name "expected")  "n/a")))
	     (set! res (append
			res  
			(list (list stepname

     (lambda (db)
       (sqlite3:for-each-row 
	(lambda (id test_id category variable value expected tol units comment status type)
	  (set! res (cons (vector id test_id category variable value expected tol units comment status type) res)))
	db
	"SELECT id,test_id,category,variable,value,expected,tol,units,comment,status,type FROM test_data WHERE test_id=? AND category LIKE ? ORDER BY category,variable;" test-id categorypatt)
       (reverse res)))))

;; This routine moved from tdb.scm, tdb:read-test-data
;;
(define (db:read-test-data* dbstruct run-id test-id categorypatt varpatt)
  (let* ((res '()))
    (db:with-db
     dbstruct #f #f
     (lambda (db)
       (sqlite3:for-each-row 
	(lambda (id test_id category variable value expected tol units comment status type)
	  (set! res (cons (vector id test_id category variable value expected tol units comment status type) res)))
	db
	"SELECT id,test_id,category,variable,value,expected,tol,units,comment,status,type FROM test_data WHERE test_id=? AND category LIKE ? AND variable LIKE ? ORDER BY category,variable;" test-id categorypatt varpatt)
       (reverse res)))))


;;======================================================================
;; Misc. test related queries
;;======================================================================

(define (db:get-run-ids-matching-target dbstruct keynames target res runname testpatt statepatt statuspatt)
  (db:with-db

;; state is the priority rollup of all states
;; status is the priority rollup of all completed statesfu
;;
;; if test-name is an integer work off that instead of test-name test-path
;;
(define (db:set-state-status-and-roll-up-items dbstruct run-id test-name item-path state status comment)
  ;; establish info on incoming test followed by info on top level test
  ;; BBnote - for mode itemwait, linkage between upstream test & matching item status is propagated to run queue in db:prereqs-not-met
  
  (let* ((testdat      (if (number? test-name)
			   (db:get-test-info-by-id dbstruct run-id test-name) ;; test-name is actually a test-id
			   (db:get-test-info       dbstruct run-id test-name item-path)))
	 (test-id      (db:test-get-id testdat))
	 (test-name    (if (number? test-name)
			   (db:test-get-testname testdat)
			   test-name))
	 (item-path    (db:test-get-item-path testdat))
         (tl-testdat   (db:get-test-info dbstruct run-id test-name ""))
         (tl-test-id   (if tl-testdat
         (tl-test-id   (db:test-get-id tl-testdat)))
			   (db:test-get-id tl-testdat)
			   #f)))
    (if (member state '("LAUNCHED" "REMOTEHOSTSTART"))
	(db:general-call dbstruct 'set-test-start-time (list test-id)))
    (mutex-lock! *db-transaction-mutex*)
    (db:with-db
     dbstruct #f #f
     (lambda (db)
       (let ((tr-res

                                                                                      *common:not-started-ok-statuses*))))
								  state-status-counts)))
                            ;; (non-completes        (filter (lambda (x)
                            ;;                                 (not (equal? (dbr:counts-state x) "COMPLETED")))
                            ;;                               state-status-counts))
                            (all-curr-states      (common:special-sort  ;; worst -> best (sort of)
                                                   (delete-duplicates
                                                    (if (not (equal? state "DELETED"))
                                                    (cons state (map dbr:counts-state state-status-counts)))
                                                        (cons state (map dbr:counts-state state-status-counts))
                                                        (map dbr:counts-state state-status-counts)))
                                                   *common:std-states* >))
                            (all-curr-statuses    (common:special-sort  ;; worst -> best
                                                   (delete-duplicates
                                                    (if (not (equal? state "DELETED"))
                                                    (cons status (map dbr:counts-status state-status-counts)))
                                                        (cons status (map dbr:counts-status state-status-counts))
                                                        (map dbr:counts-status state-status-counts)))
                                                   *common:std-statuses* >))
			    (non-completes     (filter (lambda (x)
							 (not (equal? x "COMPLETED")))
						       all-curr-states))
			    (num-non-completes (length non-completes))
                            
                            (newstate          (cond
						((> running 0)
						 "RUNNING") ;; anything running, call the situation running
						((> bad-not-started 0)  ;; we have an ugly situation, it is completed in the sense we cannot do more.
						 "COMPLETED") 
						((> num-non-completes 0) ;;
						 (car non-completes))  ;;  (remove (lambda (x)(equal? "COMPLETED" x)) all-curr-states)))
                                                ;; only rollup DELETED if all DELETED
						(else
						 (car all-curr-states))))
			                       ;; (if (> running 0)
                                               ;;     "RUNNING"
                                               ;;     (if (> bad-not-started 0)
                                               ;;         "COMPLETED"
                                               ;;         (car all-curr-states))))
                            (newstatus            (if (or (> bad-not-started 0)
							  (and (equal? newstate "NOT_STARTED")
							       (> num-non-completes 0)))
						      "CHECK"
						      "STARTED"
                                                      (car all-curr-statuses))))
                       ;; (print "bad-not-supported: " bad-not-support " all-curr-states: " all-curr-states " all-curr-statuses: " all-curr-states)
                       ;;      " newstate: " newstate " newstatus: " newstatus)
                       ;; NB// Pass the db so it is part of the transaction
                       (if tl-test-id
                       (db:test-set-state-status db run-id tl-test-id newstate newstatus #f)))))))
			   (db:test-set-state-status db run-id tl-test-id newstate newstatus #f))))))))
         (mutex-unlock! *db-transaction-mutex*)
         (if (and test-id state status (equal? status "AUTO")) 
             (db:test-data-rollup dbstruct run-id test-id status))
         tr-res)))))

;; BBnote: db:get-all-state-status-counts-for-test returns dbr:counts object aggregating state and status of items of a given test, *not including rollup state/status*
(define (db:get-all-state-status-counts-for-test dbstruct run-id test-name item-path)
  (db:with-db
   dbstruct #f #f
   (lambda (db)
     (sqlite3:map-row
      (lambda (state status count)
	(make-dbr:counts state: state status: status count: count))

		 (dbfj   (conc dbpath "-journal")))
	    (if (handle-exceptions
		 exn
		 (begin
		   (debug:print-info 0 *default-log-port* "WARNING: failed to test for existance of " dbfj)
		   (thread-sleep! 1)
		   (db:delay-if-busy count (- count 1))) 
		 (file-exists? dbfj))
		 (common:file-exists? dbfj))
		(case count
		  ((6)
		   (thread-sleep! 0.2)
		   (db:delay-if-busy count: 5))
		  ((5)
		   (thread-sleep! 0.4)
		   (db:delay-if-busy count: 4))

;; Note: mode 'normal means that tests must be COMPLETED and ok (i.e. PASS, WARN, CHECK, SKIP or WAIVED)
;;       mode 'toplevel means that tests must be COMPLETED only
;;       mode 'itemmatch or 'itemwait means that tests items must be COMPLETED and (PASS|WARN|WAIVED|CHECK) [[ NB// NOT IMPLEMENTED YET ]]
;;       mode 'exclusive means this test/item cannot run if the same test/item is LAUNCHED,REMOTEHOSTSTART or RUNNING
;; 
;; (define (db:get-prereqs-not-met dbstruct run-id waitons ref-item-path mode)
(define (db:get-prereqs-not-met dbstruct run-id waitons ref-test-name ref-item-path mode itemmaps) ;; #!key (mode '(normal))(itemmap #f))
  ;; BBnote - rollup of an itemized test's overall state/status done in db:set-state-status-and-roll-up-items
  (append
   (if (member 'exclusive mode)
       (let ((running-tests (db:get-tests-for-run dbstruct
						  #f  ;; run-id of #f means for all runs. 
						  (if (string=? ref-item-path "")   ;; testpatt
						      ref-test-name
						      (conc ref-test-name "/" ref-item-path))

	    ;; and related sub items
	    ;; next should be using mt:get-tests-for-run?
	    (let ((tests             (db:get-tests-for-run-state-status dbstruct run-id waitontest-name))
		  (ever-seen         #f)
		  (parent-waiton-met #f)
		  (item-waiton-met   #f))
	      (for-each 
	       (lambda (test)
	       (lambda (test) ;; BB- this is the upstream test
		 ;; (if (equal? waitontest-name (db:test-get-testname test)) ;; by defintion this had better be true ...
		 (let* ((state             (db:test-get-state test))
			(status            (db:test-get-status test))
			(item-path         (db:test-get-item-path test))
			(item-path         (db:test-get-item-path test)) ;; BB- this is the upstream itempath
			(is-completed      (equal? state "COMPLETED"))
			(is-running        (equal? state "RUNNING"))
			(is-killed         (equal? state "KILLED"))
			(is-ok             (member status '("PASS" "WARN" "CHECK" "WAIVED" "SKIP")))
			;;                                       testname-b    path-a    path-b
			(same-itempath     (db:compare-itempaths ref-test-name item-path ref-item-path itemmaps))) ;; (equal? ref-item-path item-path)))
		   (set! ever-seen #t)

		     (set! parent-waiton-met #t))
		    ;; For itemwait mode IFF the previous matching item is good the set parent-waiton-met
		    ((and (not (null? (lset-intersection eq? mode '(itemmatch itemwait)))) ;; how is that different from (member mode '(itemmatch itemwait)) ?????
			  ;; (not (equal? item-path "")) ;; this applies to both top level (to allow launching of next batch) and items
			  same-itempath)
		     (if (and is-completed is-ok)
			 (set! item-waiton-met #t))
		     (if (and (equal? item-path "")
			      (or is-completed is-running));; this is the parent, set it to run if completed or running
		     (if (and (equal? item-path "") ;; if upstream rollup test is completed, parent-waiton-met is set
			      (or is-completed is-running));; this is the parent, set it to run if completed or running ;; BB1
			 (set! parent-waiton-met #t)))
		    ;; normal checking of parent items, any parent or parent item not ok blocks running
		    ((and is-completed
			  (or is-ok 
			      (member 'toplevel mode))              ;; toplevel does not block on FAIL
			  (and is-ok (member 'itemmatch mode))) ;; itemmatch blocks on not ok
		     (set! item-waiton-met #t)))))

								      (append res (list (vector-ref vb (+ i 2))))))))
					       (runname   (vector-ref vb 1))
					       (testname  (vector-ref vb (+  2 numkeys)))
					       (item-path (vector-ref vb (+  3 numkeys)))
					       (final-log (vector-ref vb (+  7 numkeys)))
					       (run-dir   (vector-ref vb (+ 18 numkeys)))
					       (log-fpath (conc run-dir "/"  final-log))) ;; (string-intersperse keyvals "/") "/" testname "/" item-path "/"
					  (debug:print 4 *default-log-port* "log: " log-fpath " exists: " (file-exists? log-fpath))
					  (vector-set! vb (+ 7 numkeys) (if (file-exists? log-fpath)
					  (debug:print 4 *default-log-port* "log: " log-fpath " exists: " (common:file-exists? log-fpath))
					  (vector-set! vb (+ 7 numkeys) (if (common:file-exists? log-fpath)
									    (let ((newpath (conc pathmod "/"
												 (string-intersperse keyvals "/")
												 "/" runname "/" testname "/"
												 (if (string=? item-path "") "" (conc "/" item-path))
												 final-log)))
									      ;; for now throw away newpath and use the log-fpath conc'd with pathmod
									      (set! newpath (conc pathmod log-fpath))

(use regex typed-records matchable)

(declare (unit dcommon))

(declare (uses megatest-version))
(declare (uses gutils))
(declare (uses db))
(declare (uses synchash))
;; (declare (uses synchash))

(include "common_records.scm")
(include "db_records.scm")
(include "key_records.scm")
(include "run_records.scm")

;; yes, this is non-ideal 

;;  1. Make "data" hash-table hierarchial store of all displayed data
;;  2. Update synchash to understand "get-runs", "get-tests" etc.
;;  3. Add extraction of filters to synchash calls
;;
;;    NOTE: Used in newdashboard
;;
;; Mode is 'full or 'incremental for full refresh or incremental refresh
(define (dcommon:run-update keys data runname keypatts testpatt states statuses mode window-id)
  (let* (;; count and offset => #f so not used
	 ;; the synchash calls modify the "data" hash
	 (changed         #f)
	 (get-runs-sig    (conc (client:get-signature) " get-runs"))
	 (get-tests-sig   (conc (client:get-signature) " get-tests"))
	 (get-details-sig (conc (client:get-signature) " get-test-details"))

	 ;; test-ids to get and display are indexed on window-id in curr-test-ids hash
	 (test-ids        (hash-table-values (dboard:tabdat-curr-test-ids data)))
	 ;; run-id is #f in next line to send the query to server 0
 	 (run-changes     (synchash:client-get 'db:get-runs get-runs-sig (length keypatts) data #f runname #f #f keypatts))
	 (tests-detail-changes (if (not (null? test-ids))
				   (synchash:client-get 'db:get-test-info-by-ids get-details-sig 0  data #f test-ids)
				   '()))

	 ;; Now can calculate the run-ids
	 (run-hash    (hash-table-ref/default data get-runs-sig #f))
	 (run-ids     (if run-hash (filter number? (hash-table-keys run-hash)) '()))

	 (all-test-changes (let ((res (make-hash-table)))
			     (for-each (lambda (run-id)
					 (if (> run-id 0)
					     (hash-table-set! res run-id (synchash:client-get 'db:get-tests-for-run-mindata get-tests-sig 0 data run-id 1 testpatt states statuses #f))))
				       run-ids)
			     res))
	 (runs-hash    (hash-table-ref/default data get-runs-sig #f))
	 (header       (hash-table-ref/default runs-hash "header" #f))
	 (run-ids      (sort (filter number? (hash-table-keys runs-hash))
			     (lambda (a b)
			       (let* ((record-a (hash-table-ref runs-hash a))
				      (record-b (hash-table-ref runs-hash b))
				      (time-a   (db:get-value-by-header record-a header "event_time"))
				      (time-b   (db:get-value-by-header record-b header "event_time")))
				 (> time-a time-b)))
			     ))
	 (runid-to-col    (hash-table-ref *cachedata* "runid-to-col"))
	 (testname-to-row (hash-table-ref *cachedata* "testname-to-row")) 
	 (colnum       1)
	 (rownum       0)
	 (cellname (conc rownum ":" colnum))) ;; rownum = 0 is the header
;; (debug:print 0 *default-log-port* "test-ids " test-ids ", tests-detail-changes " tests-detail-changes)
    
	 ;; tests related stuff
	 ;; (all-testnames (delete-duplicates (map db:test-get-testname test-changes))))

    ;; Given a run-id and testname/item_path calculate a cell R:C

    ;; NOTE: Also build the test tree browser and look up table
    ;;
    ;; Each run is unique on its keys and runname or run-id, store in hash on colnum
    (for-each (lambda (run-id)
		(let* ((run-record (hash-table-ref/default runs-hash run-id #f))
		       (key-vals   (map (lambda (key)(db:get-value-by-header run-record header key))
					keys))
		       (run-name   (db:get-value-by-header run-record header "runname"))
		       (col-name   (conc (string-intersperse key-vals "\n") "\n" run-name))
		       (run-path   (append key-vals (list run-name))))
		  (hash-table-set! (dboard:tabdat-run-keys data) run-id run-path)
		  ;; modify cell - but only if changed
		  (set! changed (dcommon:modifiy-if-different (dboard:tabdat-runs-matrix data) cellname col-name changed))
		  (hash-table-set! runid-to-col run-id (list colnum run-record))
		  ;; Here we update the tests treebox and tree keys
		  (tree:add-node (dboard:tabdat-tests-tree data) "Runs" (append key-vals (list run-name))
				 userdata: (conc "run-id: " run-id))
		  (set! colnum (+ colnum 1))))
	      run-ids)

    ;; Scan all tests to be displayed and organise all the test names, respecting what is in the hash table
    ;; Do this analysis in the order of the run-ids, the most recent run wins
    (for-each (lambda (run-id)
		(let* ((run-path       (hash-table-ref (dboard:tabdat-run-keys data) run-id))
		       (test-changes   (hash-table-ref all-test-changes run-id))
		       (new-test-dat   (car test-changes))
		       (removed-tests  (cadr test-changes))
		       (tests          (sort (map cadr (filter (lambda (testrec)
								 (eq? run-id (db:mintest-get-run_id (cadr testrec))))
							       new-test-dat))
					     (lambda (a b)
					       (let ((time-a (db:mintest-get-event_time a))
						     (time-b (db:mintest-get-event_time b)))
						 (> time-a time-b)))))
		       ;; test-changes is a list of (( id record ) ... )
		       ;; Get list of test names sorted by time, remove tests
		       (test-names (delete-duplicates (map (lambda (t)
							     (let ((i (db:mintest-get-item_path t))
								   (n (db:mintest-get-testname  t)))
							       (if (string=? i "")
								   (conc "   " i)
								   n)))
							   tests)))
		       (colnum     (car (hash-table-ref runid-to-col run-id))))
		  ;; for each test name get the slot if it exists and fill in the cell
		  ;; or take the next slot and fill in the cell, deal with items in the
		  ;; run view panel? The run view panel can have a tree selector for
		  ;; browsing the tests/items

		  ;; SWITCH THIS TO USING CHANGED TESTS ONLY
		  (for-each (lambda (test)
			      (let* ((test-id   (db:mintest-get-id test))
				     (state     (db:mintest-get-state test))
				     (status    (db:mintest-get-status test))
				     (testname  (db:mintest-get-testname test))
				     (itempath  (db:mintest-get-item_path test))
				     (fullname  (conc testname "/" itempath))
				     (dispname  (if (string=? itempath "") testname (conc "   " itempath)))
				     (rownum    (hash-table-ref/default testname-to-row fullname #f))
				     (test-path (append run-path (if (equal? itempath "") 
								     (list testname)
								     (list testname itempath))))
				     (tb         (dboard:tabdat-tests-tree data)))
				(print "INFONOTE: run-path: " run-path)
				(tree:add-node (dboard:tabdat-tests-tree data) "Runs" 
					       test-path
					       userdata: (conc "test-id: " test-id))
				(let ((node-num (tree:find-node tb (cons "Runs" test-path)))
				      (color    (car (gutils:get-color-for-state-status state status))))
				  (debug:print 0 *default-log-port* "node-num: " node-num ", color: " color)

				  (set! changed (dcommon:modifiy-if-different 
						 tb
						 (conc "COLOR" node-num)
						 color changed))

				  ;; (iup:attribute-set! tb (conc "COLOR" node-num) color)
;; (define (dcommon:run-update keys data runname keypatts testpatt states statuses mode window-id)
;;   (let* (;; count and offset => #f so not used
;; 	 ;; the synchash calls modify the "data" hash
;; 	 (changed         #f)
;; 	 (get-runs-sig    (conc (client:get-signature) " get-runs"))
;; 	 (get-tests-sig   (conc (client:get-signature) " get-tests"))
;; 	 (get-details-sig (conc (client:get-signature) " get-test-details"))
;; 
;; 	 ;; test-ids to get and display are indexed on window-id in curr-test-ids hash
;; 	 (test-ids        (hash-table-values (dboard:tabdat-curr-test-ids data)))
;; 	 ;; run-id is #f in next line to send the query to server 0
;;  	 (run-changes     (synchash:client-get 'db:get-runs get-runs-sig (length keypatts) data #f runname #f #f keypatts))
;; 	 (tests-detail-changes (if (not (null? test-ids))
;; 				   (synchash:client-get 'db:get-test-info-by-ids get-details-sig 0  data #f test-ids)
;; 				   '()))
;; 
;; 	 ;; Now can calculate the run-ids
;; 	 (run-hash    (hash-table-ref/default data get-runs-sig #f))
;; 	 (run-ids     (if run-hash (filter number? (hash-table-keys run-hash)) '()))
;; 
;; 	 (all-test-changes (let ((res (make-hash-table)))
;; 			     (for-each (lambda (run-id)
;; 					 (if (> run-id 0)
;; 					     (hash-table-set! res run-id (synchash:client-get 'db:get-tests-for-run-mindata get-tests-sig 0 data run-id 1 testpatt states statuses #f))))
;; 				       run-ids)
;; 			     res))
;; 	 (runs-hash    (hash-table-ref/default data get-runs-sig #f))
;; 	 (header       (hash-table-ref/default runs-hash "header" #f))
;; 	 (run-ids      (sort (filter number? (hash-table-keys runs-hash))
;; 			     (lambda (a b)
;; 			       (let* ((record-a (hash-table-ref runs-hash a))
;; 				      (record-b (hash-table-ref runs-hash b))
;; 				      (time-a   (db:get-value-by-header record-a header "event_time"))
;; 				      (time-b   (db:get-value-by-header record-b header "event_time")))
;; 				 (> time-a time-b)))
;; 			     ))
;; 	 (runid-to-col    (hash-table-ref *cachedata* "runid-to-col"))
;; 	 (testname-to-row (hash-table-ref *cachedata* "testname-to-row")) 
;; 	 (colnum       1)
;; 	 (rownum       0)
;; 	 (cellname (conc rownum ":" colnum))) ;; rownum = 0 is the header
;; ;; (debug:print 0 *default-log-port* "test-ids " test-ids ", tests-detail-changes " tests-detail-changes)
;;     
;; 	 ;; tests related stuff
;; 	 ;; (all-testnames (delete-duplicates (map db:test-get-testname test-changes))))
;; 
;;     ;; Given a run-id and testname/item_path calculate a cell R:C
;; 
;;     ;; NOTE: Also build the test tree browser and look up table
;;     ;;
;;     ;; Each run is unique on its keys and runname or run-id, store in hash on colnum
;;     (for-each (lambda (run-id)
;; 		(let* ((run-record (hash-table-ref/default runs-hash run-id #f))
;; 		       (key-vals   (map (lambda (key)(db:get-value-by-header run-record header key))
;; 					keys))
;; 		       (run-name   (db:get-value-by-header run-record header "runname"))
;; 		       (col-name   (conc (string-intersperse key-vals "\n") "\n" run-name))
;; 		       (run-path   (append key-vals (list run-name))))
;; 		  (hash-table-set! (dboard:tabdat-run-keys data) run-id run-path)
;; 		  ;; modify cell - but only if changed
;; 		  (set! changed (dcommon:modifiy-if-different (dboard:tabdat-runs-matrix data) cellname col-name changed))
;; 		  (hash-table-set! runid-to-col run-id (list colnum run-record))
;; 		  ;; Here we update the tests treebox and tree keys
;; 		  (tree:add-node (dboard:tabdat-tests-tree data) "Runs" (append key-vals (list run-name))
;; 				 userdata: (conc "run-id: " run-id))
;; 		  (set! colnum (+ colnum 1))))
;; 	      run-ids)
;; 
;;     ;; Scan all tests to be displayed and organise all the test names, respecting what is in the hash table
;;     ;; Do this analysis in the order of the run-ids, the most recent run wins
;;     (for-each (lambda (run-id)
;; 		(let* ((run-path       (hash-table-ref (dboard:tabdat-run-keys data) run-id))
;; 		       (test-changes   (hash-table-ref all-test-changes run-id))
;; 		       (new-test-dat   (car test-changes))
;; 		       (removed-tests  (cadr test-changes))
;; 		       (tests          (sort (map cadr (filter (lambda (testrec)
;; 								 (eq? run-id (db:mintest-get-run_id (cadr testrec))))
;; 							       new-test-dat))
;; 					     (lambda (a b)
;; 					       (let ((time-a (db:mintest-get-event_time a))
;; 						     (time-b (db:mintest-get-event_time b)))
;; 						 (> time-a time-b)))))
;; 		       ;; test-changes is a list of (( id record ) ... )
;; 		       ;; Get list of test names sorted by time, remove tests
;; 		       (test-names (delete-duplicates (map (lambda (t)
;; 							     (let ((i (db:mintest-get-item_path t))
;; 								   (n (db:mintest-get-testname  t)))
;; 							       (if (string=? i "")
;; 								   (conc "   " i)
;; 								   n)))
;; 							   tests)))
;; 		       (colnum     (car (hash-table-ref runid-to-col run-id))))
;; 		  ;; for each test name get the slot if it exists and fill in the cell
;; 		  ;; or take the next slot and fill in the cell, deal with items in the
;; 		  ;; run view panel? The run view panel can have a tree selector for
;; 		  ;; browsing the tests/items
;; 
;; 		  ;; SWITCH THIS TO USING CHANGED TESTS ONLY
;; 		  (for-each (lambda (test)
;; 			      (let* ((test-id   (db:mintest-get-id test))
;; 				     (state     (db:mintest-get-state test))
;; 				     (status    (db:mintest-get-status test))
;; 				     (testname  (db:mintest-get-testname test))
;; 				     (itempath  (db:mintest-get-item_path test))
;; 				     (fullname  (conc testname "/" itempath))
;; 				     (dispname  (if (string=? itempath "") testname (conc "   " itempath)))
;; 				     (rownum    (hash-table-ref/default testname-to-row fullname #f))
;; 				     (test-path (append run-path (if (equal? itempath "") 
;; 								     (list testname)
;; 								     (list testname itempath))))
;; 				     (tb         (dboard:tabdat-tests-tree data)))
;; 				(print "INFONOTE: run-path: " run-path)
;; 				(tree:add-node (dboard:tabdat-tests-tree data) "Runs" 
;; 					       test-path
;; 					       userdata: (conc "test-id: " test-id))
;; 				(let ((node-num (tree:find-node tb (cons "Runs" test-path)))
;; 				      (color    (car (gutils:get-color-for-state-status state status))))
;; 				  (debug:print 0 *default-log-port* "node-num: " node-num ", color: " color)
;; 
;; 				  (set! changed (dcommon:modifiy-if-different 
;; 						 tb
;; 						 (conc "COLOR" node-num)
;; 						 color changed))
;; 
;; 				  ;; (iup:attribute-set! tb (conc "COLOR" node-num) color)
				  )
				(hash-table-set! (dboard:tabdat-path-test-ids data) test-path test-id)
				(if (not rownum)
				    (let ((rownums (hash-table-values testname-to-row)))
				      (set! rownum (if (null? rownums)
;; 				  )
;; 				(hash-table-set! (dboard:tabdat-path-test-ids data) test-path test-id)
;; 				(if (not rownum)
;; 				    (let ((rownums (hash-table-values testname-to-row)))
;; 				      (set! rownum (if (null? rownums)
						       1
						       (+ 1 (common:max rownums))))
				      (hash-table-set! testname-to-row fullname rownum)
				      ;; create the label
				      (set! changed (dcommon:modifiy-if-different 
						     (dboard:tabdat-runs-matrix data)
						     (conc rownum ":" 0)
						     dispname
						     changed))
				      ;; (iup:attribute-set! (dboard:tabdat-runs-matrix data)
				      ;;   		  (conc rownum ":" 0) dispname)
				      ))
				;; set the cell text and color
				;; (debug:print 2 *default-log-port* "rownum:colnum=" rownum ":" colnum ", state=" status)
				(set! changed (dcommon:modifiy-if-different 
						     (dboard:tabdat-runs-matrix data)
						     (conc rownum ":" colnum)
						     (if (member state '("ARCHIVED" "COMPLETED"))
							 status
							 state)
						     changed))
				;; (iup:attribute-set! (dboard:tabdat-runs-matrix data)
				;; 		    (conc rownum ":" colnum)
				;; 		    (if (member state '("ARCHIVED" "COMPLETED"))
				;; 			status
				;; 			state))
				(set! changed (dcommon:modifiy-if-different 
					       (dboard:tabdat-runs-matrix data)
					       (conc "BGCOLOR" rownum ":" colnum)
					       (car (gutils:get-color-for-state-status state status))
					       changed))
				;; (iup:attribute-set! (dboard:tabdat-runs-matrix data)
				;; 		    (conc "BGCOLOR" rownum ":" colnum)
				;; 		    (car (gutils:get-color-for-state-status state status)))
				))
			    tests)))
	      run-ids)

    (let ((updater (hash-table-ref/default  (dboard:commondat-updaters commondat) window-id #f)))
      (if updater (updater (hash-table-ref/default data get-details-sig #f))))

    (if changed (iup:attribute-set! (dboard:tabdat-runs-matrix data) "REDRAW" "ALL"))
    ;; (debug:print 2 *default-log-port* "run-changes: " run-changes)
    ;; (debug:print 2 *default-log-port* "test-changes: " test-changes)
    (list run-changes all-test-changes)))
;; 						       1
;; 						       (+ 1 (common:max rownums))))
;; 				      (hash-table-set! testname-to-row fullname rownum)
;; 				      ;; create the label
;; 				      (set! changed (dcommon:modifiy-if-different 
;; 						     (dboard:tabdat-runs-matrix data)
;; 						     (conc rownum ":" 0)
;; 						     dispname
;; 						     changed))
;; 				      ;; (iup:attribute-set! (dboard:tabdat-runs-matrix data)
;; 				      ;;   		  (conc rownum ":" 0) dispname)
;; 				      ))
;; 				;; set the cell text and color
;; 				;; (debug:print 2 *default-log-port* "rownum:colnum=" rownum ":" colnum ", state=" status)
;; 				(set! changed (dcommon:modifiy-if-different 
;; 						     (dboard:tabdat-runs-matrix data)
;; 						     (conc rownum ":" colnum)
;; 						     (if (member state '("ARCHIVED" "COMPLETED"))
;; 							 status
;; 							 state)
;; 						     changed))
;; 				;; (iup:attribute-set! (dboard:tabdat-runs-matrix data)
;; 				;; 		    (conc rownum ":" colnum)
;; 				;; 		    (if (member state '("ARCHIVED" "COMPLETED"))
;; 				;; 			status
;; 				;; 			state))
;; 				(set! changed (dcommon:modifiy-if-different 
;; 					       (dboard:tabdat-runs-matrix data)
;; 					       (conc "BGCOLOR" rownum ":" colnum)
;; 					       (car (gutils:get-color-for-state-status state status))
;; 					       changed))
;; 				;; (iup:attribute-set! (dboard:tabdat-runs-matrix data)
;; 				;; 		    (conc "BGCOLOR" rownum ":" colnum)
;; 				;; 		    (car (gutils:get-color-for-state-status state status)))
;; 				))
;; 			    tests)))
;; 	      run-ids)
;; 
;;     (let ((updater (hash-table-ref/default  (dboard:commondat-updaters commondat) window-id #f)))
;;       (if updater (updater (hash-table-ref/default data get-details-sig #f))))
;; 
;;     (if changed (iup:attribute-set! (dboard:tabdat-runs-matrix data) "REDRAW" "ALL"))
;;     ;; (debug:print 2 *default-log-port* "run-changes: " run-changes)
;;     ;; (debug:print 2 *default-log-port* "test-changes: " test-changes)
;;     (list run-changes all-test-changes)))

(define (dcommon:runsdat-get-col-num dat target runname force-set)
  (let* ((runs-index (dboard:runsdat-runs-index dat))
	 (col-name   (conc target "/" runname))
	 (res        (hash-table-ref/default runs-index col-name #f)))
    (if res
	res

integrating and or running a complex suite of tests for release
qualification.</p></div>
</div>
</div>
<div class="sect1">
<h2 id="_megatest_design_philosophy">Megatest Design Philosophy</h2>
<div class="sectionbody">
<div class="paragraph"><p>Megatest is intended to provide the minimum needed resources to make
writing a suite of tests and tasks for implementing continuous build
for software, design engineering or process control (via owlfs for
example) without being specialized for any specific problem
space. Megatest in of itself does not know what constitutes a PASS or
FAIL of a test or task. In most cases megatest is best used in
conjunction with logpro or a similar tool to parse, analyze and decide
<div class="paragraph"><p>Megatest is a distributed system intended to provide the minimum needed
resources to make writing a suite of tests and tasks for implementing
continuous build for software, design engineering or process control (via
owlfs for example) without being specialized for any specific problem
space. Megatest in of itself does not know what constitutes a PASS or FAIL
of a test or task. In most cases megatest is best used in conjunction with
logpro or a similar tool to parse, analyze and decide on the test outcome.</p></div>
on the test outcome.</p></div>
<div class="ulist"><ul>
<li>
<p>
Self-checking -Repeatable strive for directed or self-checking test
   as opposed to delta based tests
</p>
</li>

<h3 id="_architecture_refactor">Architecture Refactor</h3>
<div class="sect3">
<h4 id="_goals">Goals</h4>
<div class="olist arabic"><ol class="arabic">
<li>
<p>
Reduce load on the file system. Sqlite3 files on network filesystem can be
  a burden.
  a burden. <span class="green">[DONE]</span>
</p>
</li>
<li>
<p>
Reduce number of servers and frequency of start/stop. This is mostly an
  issue of clutter but also a reduction in "moving parts".
  issue of clutter but also a reduction in "moving parts". <span class="green">[DONE]</span>
</p>
</li>
<li>
<p>
Coalesce activities to a single home host where possible. Give the user
  feedback that they have started the dashboard on a host other than the
  home host.
  home host. <span class="green">[DONE]</span>
</p>
</li>
<li>
<p>
Reduce number of processes involved in managing running tests.
</p>
</li>
</ol></div>
</div>
<div class="sect3">
<h4 id="_changes_needed">Changes Needed</h4>
<div class="olist arabic"><ol class="arabic">
<li>
<p>
ACID compliant db will be on /tmp and synced to megatest.db with a five
  second max delay.
  second max delay. <span class="green">[DONE]</span>
</p>
</li>
<li>
<p>
Read/writes to db for processes on homehost will go direct to /tmp
  megatest.db file.
  megatest.db file. <span class="green">[DONE]</span>
</p>
</li>
<li>
<p>
Read/wites fron non-homehost processes will go through one server. Bulk
  reads (e.g. for dashboard or list-runs) will be cached on the current host
  in /tmp and synced from the home megatest.db in the testsuite area.
  in /tmp and synced from the home megatest.db in the testsuite area. <span class="green">[DONE]</span>
</p>
</li>
<li>
<p>
Db syncs rely on the target db file timestame minus some margin.
Db syncs rely on the target db file timestame minus some margin. <span class="green">[DONE]</span>
</p>
</li>
<li>
<p>
Since bulk reads do not use the server we can switch to simple RPC for the
  network transport.
  network transport. <span class="green">[DONE]</span>
</p>
</li>
<li>
<p>
Test running manager process extended to manage multiple running tests.
</p>
</li>
</ol></div>
</div>
</div>
<div class="sect2">
<h3 id="_current_items">Current Items</h3>
<div class="sect3">
<h4 id="_ww05_migrate_to_inmem_db">ww05 - migrate to inmem-db</h4>
<div class="olist arabic"><ol class="arabic">
<li>
<p>
Switch to inmem db with fast sync to on disk db&#8217;s [DONE]
Switch to inmem db with fast sync to on disk db&#8217;s <span class="green">[DONE]</span>
</p>
</li>
<li>
<p>
Server polls tasks table for next action
</p>
<div class="olist loweralpha"><ol class="loweralpha">
<li>
<p>
Task table used for tracking runner process [DONE]
Task table used for tracking runner process <span class="red">[Replaced by mtutil]</span>
</p>
</li>
<li>
<p>
Task table used for jobs to run
Task table used for jobs to run <span class="red">[Replaced by mtutil]</span>
</p>
</li>
<li>
<p>
Task table used for queueing runner actions (remove runs, cleanRunExecute, etc)
Task table used for queueing runner actions (remove runs,
   cleanRunExecute, etc)  <span class="red">[Replaced by mtutil</span>]
</p>
</li>
</ol></div>
</li>
</ol></div>
<div class="paragraph"><p>shifting, note that the preceding blank line is needed.</p></div>
</div>

<div class="listingblock">
<div class="content monospaced">
<pre>[items]
A a b c
B d e f</pre>
</div></div>
<div class="paragraph"><p>Then the config file would effectively appear to contain an items section
exactly like the output from the script. This is extremely useful when
dynamically creating items, itemstables and other config structures. You can
see the expansion of the call by looking in the cached files (look in your
linktree for megatest.config and runconfigs.config cache files and in your
test run areas for the expanded and cached testconfig).</p></div>
exactly like the output from the script. This is useful when dynamically
creating items, itemstables and other config structures. You can see the
expansion of the call by looking in the cached files (look in your linktree
for megatest.config and runconfigs.config cache files and in your test run
areas for the expanded and cached testconfig).</p></div>
<div class="paragraph"><p>Wildcards and regexes in Targets</p></div>
<div class="listingblock">
<div class="content monospaced">
<pre>[a/2/b]
VAR1 VAL1

[a/%/b]
VAR1 VAL2</pre>
</div></div>
<div class="paragraph"><p>Will result in:</p></div>
<div class="listingblock">
<div class="content monospaced">
<pre>[a/2/b]
VAR1 VAL2</pre>
</div></div>
<div class="paragraph"><p>Can use either wildcard of "%" or a regular expression:</p></div>
<div class="listingblock">
<div class="content monospaced">
<pre>[/abc.*def/]</pre>
</div></div>
<div class="sect3">
<h4 id="_disk_space_checks">Disk Space Checks</h4>
<div class="paragraph"><p>Some parameters you can put in the [setup] section of megatest.config:</p></div>
<div class="listingblock">
<div class="content monospaced">
<pre># minimum space required in a run disk
minspace 10000000

<h5 id="_miscellaneous_setup_items">Miscellaneous Setup Items</h5>
<div class="paragraph"><p>Attempt to rerun tests in "STUCK/DEAD", "n/a", "ZERO_ITEMS" states.</p></div>
<div class="listingblock">
<div class="title">In megatest.config</div>
<div class="content monospaced">
<pre>[setup]
reruns 5</pre>
</div></div>
<div class="paragraph"><p>Replace the default blacklisted environment variables with user supplied
list.</p></div>
<div class="paragraph"><p>Default list: USER HOME DISPLAY LS_COLORS XKEYSYMDB EDITOR MAKEFLAGS MAKEF MAKEOVERRIDES</p></div>
<div class="paragraph"><div class="title">Add a "bad" variable "PROMPT" to the variables that will be commented out</div><p>in the megatest.sh and megatest.csh files:</p></div>
<div class="listingblock">
<div class="content monospaced">
<pre>[setup]
blacklistvars USER HOME DISPLAY LS_COLORS XKEYSYMDB EDITOR MAKEFLAGS PROMPT</pre>
</div></div>
</div>
<div class="sect4">
<h5 id="_run_time_limit">Run time limit</h5>
<div class="listingblock">
<div class="content monospaced">
<pre>[setup]

<div class="listingblock">
<div class="content monospaced">
<pre>$MT_MEGATEST -env2file .ezsteps/${stepname}</pre>
</div></div>
</div>
<div class="sect2">
<h3 id="_triggers">Triggers</h3>
<div class="paragraph"><p>In your testconfig triggers can be specified</p></div>
<div class="paragraph"><p>In your testconfig or megatest.config triggers can be specified</p></div>
<div class="listingblock">
<div class="content monospaced">
<pre>[triggers]

# Call script running.sh when test goes to state=RUNNING, status=PASS
RUNNING/PASS running.sh

# Call script running.sh any time state goes to RUNNING
RUNNING/ running.sh

# Call script onpass.sh any time status goes to PASS
PASS/ onpass.sh</pre>
</div></div>
<div class="paragraph"><p>Scripts called will have; test-id test-rundir trigger, added to the commandline.</p></div>
<div class="paragraph"><p>Scripts called will have; test-id test-rundir trigger test-name item-path state status event-time, added to the commandline.</p></div>
<div class="paragraph"><p>HINT</p></div>
<div class="paragraph"><p>To start an xterm (useful for debugging), use a command line like the following:</p></div>
<div class="listingblock">
<div class="content monospaced">
<pre>[triggers]
COMPLETED/ xterm -e bash -s --</pre>
</div></div>
<div class="admonitionblock">
<table><tr>
<td class="icon">
<img src="/nfs/pdx/disks/ice.disk.002/icfadm/pkgs/asciidoc/8.6.7/images/icons/note.png" alt="Note">
</td>
<td class="content">There is a trailing space after the --</td>
</tr></table>
</div>
<div class="paragraph"><p>There are a number of environment variables available to the trigger script
but since triggers can be called in various contexts not all variables are
available at all times. The trigger script should check for the variable and
fail gracefully if it doesn&#8217;t exist.</p></div>
<table class="tableblock frame-topbot grid-all"
style="
width:90%;
">
<caption class="title">Table 4. Environment variables visible to the trigger script</caption>
<col style="width:33%;">
<col style="width:66%;">
<thead>
<tr>
<th class="tableblock halign-center valign-top" >Variable             </th>
<th class="tableblock halign-left valign-top" > Purpose</th>
</tr>
</thead>
<tbody>
<tr>
<td class="tableblock halign-center valign-top" ><p class="tableblock">MT_TEST_RUN_DIR</p></td>
<td class="tableblock halign-left valign-top" ><p class="tableblock monospaced">The directory where Megatest ran this test</p></td>
</tr>
<tr>
<td class="tableblock halign-center valign-top" ><p class="tableblock">MT_CMDINFO</p></td>
<td class="tableblock halign-left valign-top" ><p class="tableblock monospaced">Encoded command data for the test</p></td>
</tr>
<tr>
<td class="tableblock halign-center valign-top" ><p class="tableblock">MT_DEBUG_MODE</p></td>
<td class="tableblock halign-left valign-top" ><p class="tableblock monospaced">Used to pass the debug mode to nested calls to Megatest</p></td>
</tr>
<tr>
<td class="tableblock halign-center valign-top" ><p class="tableblock">MT_RUN_AREA_HOME</p></td>
<td class="tableblock halign-left valign-top" ><p class="tableblock monospaced">Megatest home area</p></td>
</tr>
<tr>
<td class="tableblock halign-center valign-top" ><p class="tableblock">MT_TESTSUITENAME</p></td>
<td class="tableblock halign-left valign-top" ><p class="tableblock monospaced">The name of this testsuite or area</p></td>
</tr>
<tr>
<td class="tableblock halign-center valign-top" ><p class="tableblock">MT_TEST_NAME</p></td>
<td class="tableblock halign-left valign-top" ><p class="tableblock monospaced">The name of this test</p></td>
</tr>
<tr>
<td class="tableblock halign-center valign-top" ><p class="tableblock">MT_ITEM_INFO</p></td>
<td class="tableblock halign-left valign-top" ><p class="tableblock monospaced">The variable and values for the test item</p></td>
</tr>
<tr>
<td class="tableblock halign-center valign-top" ><p class="tableblock">MT_MEGATEST</p></td>
<td class="tableblock halign-left valign-top" ><p class="tableblock monospaced">Which Megatest binary is being used by this area</p></td>
</tr>
<tr>
<td class="tableblock halign-center valign-top" ><p class="tableblock">MT_TARGET</p></td>
<td class="tableblock halign-left valign-top" ><p class="tableblock monospaced">The target variable values, separated by <em>/</em></p></td>
</tr>
<tr>
<td class="tableblock halign-center valign-top" ><p class="tableblock">MT_LINKTREE</p></td>
<td class="tableblock halign-left valign-top" ><p class="tableblock monospaced">The base of the link tree where all run tests can be found</p></td>
</tr>
<tr>
<td class="tableblock halign-center valign-top" ><p class="tableblock">MT_ITEMPATH</p></td>
<td class="tableblock halign-left valign-top" ><p class="tableblock monospaced">The values of the item path variables, separated by <em>/</em></p></td>
</tr>
<tr>
<td class="tableblock halign-center valign-top" ><p class="tableblock">MT_RUNNAME</p></td>
<td class="tableblock halign-left valign-top" ><p class="tableblock monospaced">The name of the run</p></td>
</tr>
</tbody>
</table>
</div>
<div class="sect2">
<h3 id="_override_the_toplevel_html_file">Override the Toplevel HTML File</h3>
<div class="paragraph"><p>Megatest generates a simple html file summary for top level tests of
iterated tests. The generation can be overridden. NOTE: the output of
the script is captured from stdout to create the html.</p></div>
<div class="listingblock">

<h2 id="_programming_api">Programming API</h2>
<div class="sectionbody">
<div class="paragraph"><p>These routines can be called from the megatest repl.</p></div>
<table class="tableblock frame-topbot grid-all"
style="
width:70%;
">
<caption class="title">Table 4. API Keys Related Calls</caption>
<caption class="title">Table 5. API Keys Related Calls</caption>
<col style="width:14%;">
<col style="width:28%;">
<col style="width:28%;">
<col style="width:28%;">
<thead>
<tr>
<th class="tableblock halign-center valign-top" >API Call                        </th>

</div>
</div>
</div>
<div id="footnotes"><hr></div>
<div id="footer">
<div id="footer-text">
Version 1.0<br>
Last updated 2016-12-12 13:03:08 PST
Last updated 2017-05-15 15:18:21 PDT
</div>
</div>
</body>
</html>

tool, flexible enough to meet the needs of any team doing continuous
integrating and or running a complex suite of tests for release
qualification.

Megatest Design Philosophy
--------------------------

Megatest is intended to provide the minimum needed resources to make
writing a suite of tests and tasks for implementing continuous build
for software, design engineering or process control (via owlfs for
example) without being specialized for any specific problem
space. Megatest in of itself does not know what constitutes a PASS or
FAIL of a test or task. In most cases megatest is best used in
conjunction with logpro or a similar tool to parse, analyze and decide
Megatest is a distributed system intended to provide the minimum needed
resources to make writing a suite of tests and tasks for implementing
continuous build for software, design engineering or process control (via
owlfs for example) without being specialized for any specific problem
space. Megatest in of itself does not know what constitutes a PASS or FAIL
of a test or task. In most cases megatest is best used in conjunction with
logpro or a similar tool to parse, analyze and decide on the test outcome.
on the test outcome. 

 * Self-checking -Repeatable strive for directed or self-checking test
   as opposed to delta based tests

 * Traceable - environment variables, host OS and other possibly influential
   variables are captured and kept recorded.

-------------------------
[a/2/b]
VAR1 VAL2
-------------------------

Can use either wildcard of "%" or a regular expression:

-------------------------
[/abc.*def/]
-------------------------

Disk Space Checks
^^^^^^^^^^^^^^^^^

Some parameters you can put in the [setup] section of megatest.config:

-------------------

Attempt to rerun tests in "STUCK/DEAD", "n/a", "ZERO_ITEMS" states.

.In megatest.config
------------------
[setup]
reruns 5
------------------

Replace the default blacklisted environment variables with user supplied
list.

Default list: USER HOME DISPLAY LS_COLORS XKEYSYMDB EDITOR MAKEFLAGS MAKEF MAKEOVERRIDES

.Add a "bad" variable "PROMPT" to the variables that will be commented out
in the megatest.sh and megatest.csh files:
-----------------
[setup]
blacklistvars USER HOME DISPLAY LS_COLORS XKEYSYMDB EDITOR MAKEFLAGS PROMPT
-----------------

Run time limit
++++++++++++++

-----------------
[setup]
# this will automatically kill the test if it runs for more than 1h 2m and 3s

----------------------------
$MT_MEGATEST -env2file .ezsteps/${stepname}	  
----------------------------

Triggers
~~~~~~~~

In your testconfig triggers can be specified 
In your testconfig or megatest.config triggers can be specified 

-----------------
[triggers]

# Call script running.sh when test goes to state=RUNNING, status=PASS
RUNNING/PASS running.sh

# Call script running.sh any time state goes to RUNNING
RUNNING/ running.sh

# Call script onpass.sh any time status goes to PASS
PASS/ onpass.sh
-----------------

Scripts called will have; test-id test-rundir trigger, added to the commandline.
Scripts called will have; test-id test-rundir trigger test-name item-path state status event-time, added to the commandline.

HINT

To start an xterm (useful for debugging), use a command line like the following:

-----------------
[triggers]
COMPLETED/ xterm -e bash -s -- 
-----------------

NOTE: There is a trailing space after the --

There are a number of environment variables available to the trigger script
but since triggers can be called in various contexts not all variables are
available at all times. The trigger script should check for the variable and
fail gracefully if it doesn't exist.

.Environment variables visible to the trigger script
[width="90%",cols="^,2m",frame="topbot",options="header"]
|======================
|Variable             | Purpose            
| MT_TEST_RUN_DIR     | The directory where Megatest ran this test                   
| MT_CMDINFO          | Encoded command data for the test                   
| MT_DEBUG_MODE       | Used to pass the debug mode to nested calls to Megatest                   
| MT_RUN_AREA_HOME    | Megatest home area 
| MT_TESTSUITENAME    | The name of this testsuite or area                   
| MT_TEST_NAME        | The name of this test
| MT_ITEM_INFO        | The variable and values for the test item
| MT_MEGATEST         | Which Megatest binary is being used by this area
| MT_TARGET           | The target variable values, separated by '/'
| MT_LINKTREE         | The base of the link tree where all run tests can be found
| MT_ITEMPATH         | The values of the item path variables, separated by '/'
| MT_RUNNAME          | The name of the run
|======================


Override the Toplevel HTML File
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Megatest generates a simple html file summary for top level tests of
iterated tests. The generation can be overridden. NOTE: the output of

Road Map
--------

Note 1: This road-map is still evolving and subject to change without notice.

Architecture Refactor
~~~~~~~~~~~~~~~~~~~~~

Goals
^^^^^

. Reduce load on the file system. Sqlite3 files on network filesystem can be
  a burden.
  a burden. [green]#[DONE]#
. Reduce number of servers and frequency of start/stop. This is mostly an
  issue of clutter but also a reduction in "moving parts".
  issue of clutter but also a reduction in "moving parts". [green]#[DONE]#
. Coalesce activities to a single home host where possible. Give the user
  feedback that they have started the dashboard on a host other than the
  home host.
  home host. [green]#[DONE]#
. Reduce number of processes involved in managing running tests.

Changes Needed
^^^^^^^^^^^^^^

. ACID compliant db will be on /tmp and synced to megatest.db with a five
  second max delay.
  second max delay. [green]#[DONE]#
. Read/writes to db for processes on homehost will go direct to /tmp
  megatest.db file.
  megatest.db file. [green]#[DONE]#
. Read/wites fron non-homehost processes will go through one server. Bulk
  reads (e.g. for dashboard or list-runs) will be cached on the current host
  in /tmp and synced from the home megatest.db in the testsuite area.
. Db syncs rely on the target db file timestame minus some margin.
  in /tmp and synced from the home megatest.db in the testsuite area. [green]#[DONE]#
. Db syncs rely on the target db file timestame minus some margin. [green]#[DONE]#
. Since bulk reads do not use the server we can switch to simple RPC for the
  network transport.
  network transport. [green]#[DONE]#
. Test running manager process extended to manage multiple running tests.

Current Items
~~~~~~~~~~~~~

ww05 - migrate to inmem-db
^^^^^^^^^^^^^^^^^^^^^^^^^^

. Switch to inmem db with fast sync to on disk db's [DONE]
. Switch to inmem db with fast sync to on disk db's [green]#[DONE]#
. Server polls tasks table for next action
.. Task table used for tracking runner process [DONE]
.. Task table used for jobs to run
.. Task table used for queueing runner actions (remove runs, cleanRunExecute, etc)
.. Task table used for tracking runner process [red]#[Replaced by mtutil]#
.. Task table used for jobs to run [red]#[Replaced by mtutil]#
.. Task table used for queueing runner actions (remove runs,
   cleanRunExecute, etc)  [red]#[Replaced by mtutil#]


// ww32
// ~~~~
// 
// . Rerun step and or subsequent steps from gui
// . Refresh test area files from gui

;; These are called by the server on recipt of /api calls
;;    - keep it simple, only return the actual result of the call, i.e. no meta info here
;;
;;    - returns #( flag result )
;;
(define (api:execute-requests dbstruct dat)
  (handle-exceptions
   exn
   (let ((call-chain (get-call-chain)))
     (debug:print 0 *default-log-port* "WARNING: api:execute-requests received an exception from peer, dat=" dat)
     (print-call-chain (current-error-port))
     (debug:print 0 *default-log-port* " message: "  ((condition-property-accessor 'exn 'message) exn))       
     (vector #f (vector exn call-chain dat))) ;; return some stuff for debug if an exception happens
   (cond
    ((not (vector? dat))                    ;; it is an error to not receive a vector
     (vector #f (vector #f "remote must be called with a vector")))
    ((> *api-process-request-count* 20) ;; 20)
     (debug:print 0 *default-log-port* "WARNING: api:execute-requests received an overloaded message.")
     (set! *server-overloaded* #t)
     (vector #f (vector #f 'overloaded))) ;; the inner vector is what gets returned. nope, don't know why. please refactor!
    (else  
     (let* ((cmd-in            (vector-ref dat 0))
            (cmd               (if (symbol? cmd-in)
				   cmd-in
				   (string->symbol cmd-in)))
            (params            (vector-ref dat 1))
            (start-t           (current-milliseconds))
            (readonly-mode     (dbr:dbstruct-read-only dbstruct))
            (readonly-command  (member cmd api:read-only-queries))
            (writecmd-in-readonly-mode (and readonly-mode (not readonly-command)))
            (res    
             (if writecmd-in-readonly-mode
                 (conc "attempt to run write command "cmd" on a read-only database")
                 (case cmd
                   ;;===============================================
                   ;; READ/WRITE QUERIES
                   ;;===============================================

                   ((get-keys-write)                        (db:get-keys dbstruct)) ;; force a dummy "write" query to force server; for debug in -repl
                   
                   ;; SERVERS
                   ((start-server)                    (apply server:kind-run params))
                   ((kill-server)                     (set! *server-run* #f))

                   ;; TESTS

                   ;;((test-set-state-status-by-id)     (apply mt:test-set-state-status-by-id dbstruct params))
                   ;;BB - commented out above because it was calling below, eventually, incorrectly (dbstruct passed to mt:test-set-state-status-by-id, which previosly did more, but now only passes thru to db:set-state-status-and-roll-up-items.
                   ((test-set-state-status-by-id)

                    ;; (define (db:set-state-status-and-roll-up-items dbstruct run-id test-name item-path state status comment)
                    (db:set-state-status-and-roll-up-items
                     dbstruct
                     (list-ref params 0) ; run-id
                     (list-ref params 1) ; test-name
                     #f                  ; item-path
                     (list-ref params 2) ; state
                     (list-ref params 3) ; status
                     (list-ref params 4) ; comment
                     ))
                   
                   ((delete-test-records)             (apply db:delete-test-records dbstruct params))
                   ((delete-old-deleted-test-records) (apply db:delete-old-deleted-test-records dbstruct params))
                   ((test-set-state-status)           (apply db:test-set-state-status dbstruct params))
                   ((test-set-top-process-pid)        (apply db:test-set-top-process-pid dbstruct params))
                   ((set-state-status-and-roll-up-items) (apply db:set-state-status-and-roll-up-items dbstruct params))
                   ((top-test-set-per-pf-counts)      (apply db:top-test-set-per-pf-counts dbstruct params))
                   ((test-set-archive-block-id)       (apply db:test-set-archive-block-id dbstruct params))

                   ;; RUNS
                   ((register-run)                 (apply db:register-run dbstruct params))
                   ((set-tests-state-status)       (apply db:set-tests-state-status dbstruct params))
                   ((delete-run)                   (apply db:delete-run dbstruct params))
                   ((lock/unlock-run)              (apply db:lock/unlock-run dbstruct params))
                   ((update-run-event_time)        (apply db:update-run-event_time dbstruct params))
                   ((update-run-stats)             (apply db:update-run-stats dbstruct params))
                   ((set-var)                      (apply db:set-var dbstruct params))
                   ((del-var)                      (apply db:del-var dbstruct params))

                   ;; STEPS
                   ((teststep-set-status!)         (apply db:teststep-set-status! dbstruct params))

                   ;; TEST DATA
                   ((test-data-rollup)             (apply db:test-data-rollup dbstruct params))
                   ((csv->test-data)               (apply db:csv->test-data dbstruct params))

                   ;; MISC
                   ((sync-inmem->db)               (let ((run-id (car params)))
                                                     (db:sync-touched dbstruct run-id force-sync: #t)))
                   ((mark-incomplete)              (apply db:find-and-mark-incomplete dbstruct params))

                   ;; TESTMETA
                   ((testmeta-add-record)       (apply db:testmeta-add-record dbstruct params))
                   ((testmeta-update-field)     (apply db:testmeta-update-field dbstruct params))
                   ((get-tests-tags)            (db:get-tests-tags dbstruct))

                   ;; TASKS
                   ((tasks-add)                 (apply tasks:add dbstruct params))   
                   ((tasks-set-state-given-param-key) (apply tasks:set-state-given-param-key dbstruct params))
                   ((tasks-get-last)            (apply tasks:get-last dbstruct params))

		   ;; NO SYNC DB
		   ((no-sync-set)               (apply db:no-sync-set         *no-sync-db* params))
		   ((no-sync-get/default)       (apply db:no-sync-get/default *no-sync-db* params))
		   ((no-sync-del!)              (apply db:no-sync-del!        *no-sync-db* params))
		 
                   ;; ARCHIVES
                   ;; ((archive-get-allocations)   
                   ((archive-register-disk)     (apply db:archive-register-disk dbstruct params))
                   ((archive-register-block-name)(apply db:archive-register-block-name dbstruct params))
                   ((archive-allocate-testsuite/area-to-block)(apply db:archive-allocate-testsuite/area-to-block dbstruct block-id testsuite-name areakey))

                   ;;======================================================================
                   ;; READ ONLY QUERIES
                   ;;======================================================================

                   ;; KEYS
                   ((get-key-val-pairs)               (apply db:get-key-val-pairs dbstruct params))
                   ((get-keys)                        (db:get-keys dbstruct))
                   ((get-key-vals)                    (apply db:get-key-vals dbstruct params))
                   ((get-target)                      (apply db:get-target dbstruct params))
                   ((get-targets)                     (db:get-targets dbstruct))

                   ;; ARCHIVES
                   ((test-get-archive-block-info)     (apply db:test-get-archive-block-info dbstruct params))
                   
                   ;; TESTS
                   ((test-toplevel-num-items)         (apply db:test-toplevel-num-items dbstruct params))
                   ((get-test-info-by-id)	       (apply db:get-test-info-by-id dbstruct params))
                   ((test-get-rundir-from-test-id)    (apply db:test-get-rundir-from-test-id dbstruct params))
                   ((get-count-tests-running-for-testname) (apply db:get-count-tests-running-for-testname dbstruct params))
                   ((get-count-tests-running)         (apply db:get-count-tests-running dbstruct params))
                   ((get-count-tests-running-in-jobgroup) (apply db:get-count-tests-running-in-jobgroup dbstruct params))
                   ;; ((delete-test-step-records)        (apply db:delete-test-step-records dbstruct params))
                   ((get-previous-test-run-record)    (apply db:get-previous-test-run-record dbstruct params))
                   ((get-matching-previous-test-run-records)(apply db:get-matching-previous-test-run-records dbstruct params))
                   ((test-get-logfile-info)           (apply db:test-get-logfile-info dbstruct params))
                   ((test-get-records-for-index-file)  (apply db:test-get-records-for-index-file dbstruct params))
                   ((get-testinfo-state-status)       (apply db:get-testinfo-state-status dbstruct params))
                   ((test-get-top-process-pid)        (apply db:test-get-top-process-pid dbstruct params))
                   ((test-get-paths-matching-keynames-target-new) (apply db:test-get-paths-matching-keynames-target-new dbstruct params))
                   ((get-prereqs-not-met)             (apply db:get-prereqs-not-met dbstruct params))
                   ((get-count-tests-running-for-run-id) (apply db:get-count-tests-running-for-run-id dbstruct params))
                   ((synchash-get)                    (apply synchash:server-get dbstruct params))
                   ((get-raw-run-stats)               (apply db:get-raw-run-stats dbstruct params))

                   ;; RUNS
                   ((get-run-info)                 (apply db:get-run-info dbstruct params))
                   ((get-run-status)               (apply db:get-run-status dbstruct params))
                   ((set-run-status)               (apply db:set-run-status dbstruct params))
                   ((get-tests-for-run)            (apply db:get-tests-for-run dbstruct params))
                   ((get-test-id)                  (apply db:get-test-id dbstruct params))
                   ((get-tests-for-run-mindata)    (apply db:get-tests-for-run-mindata dbstruct params))
                   ((get-runs)                     (apply db:get-runs dbstruct params))
                   ((get-num-runs)                 (apply db:get-num-runs dbstruct params))
                   ((get-all-run-ids)              (db:get-all-run-ids dbstruct))
                   ((get-prev-run-ids)             (apply db:get-prev-run-ids dbstruct params))
                   ((get-run-ids-matching-target)  (apply db:get-run-ids-matching-target dbstruct params))
                   ((get-runs-by-patt)             (apply db:get-runs-by-patt dbstruct params))
                   ((get-run-name-from-id)         (apply db:get-run-name-from-id dbstruct params))
                   ((get-main-run-stats)           (apply db:get-main-run-stats dbstruct params))
                   ((get-var)                      (apply db:get-var dbstruct params))
                   ((get-run-stats)                (apply db:get-run-stats dbstruct params))

                   ;; STEPS
                   ((get-steps-data)               (apply db:get-steps-data dbstruct params))
                   ((get-steps-for-test)           (apply db:get-steps-for-test dbstruct params))

                   ;; TEST DATA
                   ((read-test-data)               (apply db:read-test-data dbstruct params))
                   ((read-test-data*)              (apply db:read-test-data* dbstruct params))

                   ;; MISC
                   ((get-latest-host-load)         (apply db:get-latest-host-load dbstruct params))
                   ((have-incompletes?)            (apply db:have-incompletes? dbstruct params))
                   ((login)                        (apply db:login dbstruct params))
                   ((general-call)                 (let ((stmtname   (car params))
                                                         (run-id     (cadr params))
                                                         (realparams (cddr params)))
                                                     (db:general-call dbstruct stmtname realparams)))
                   ((sdb-qry)                      (apply sdb:qry params))
                   ((ping)                         (current-process-id))
		   ((get-changed-record-ids)       (apply db:get-changed-record-ids dbstruct params))
		   
                   ;; TESTMETA
                   ((testmeta-get-record)       (apply db:testmeta-get-record dbstruct params))

                   ;; TASKS 
                   ((find-task-queue-records)   (apply tasks:find-task-queue-records dbstruct params))
		   (else
		    (debug:print 0 *default-log-port* "ERROR: bad api call " cmd)
		    (conc "ERROR: BAD api call " cmd))))))
       
       ;; save all stats
       (let ((delta-t (- (current-milliseconds)
			 start-t)))
	 (hash-table-set! *db-api-call-time* cmd
			  (cons delta-t (hash-table-ref/default *db-api-call-time* cmd '()))))
       (if writecmd-in-readonly-mode
	   (vector #f res)
           (vector #t res)))))))

(include "common_records.scm")
(include "key_records.scm")
(include "db_records.scm")
(include "run_records.scm")
(include "test_records.scm")

(define (common:wait-for-cpuload maxload numcpus waitdelay #!key (count 1000) (msg #f)(remote-host #f))
  (let* ((loadavg (common:get-cpu-load remote-host))
	 (first   (car loadavg))
	 (next    (cadr loadavg))
	 (adjload (* maxload numcpus))
	 (loadjmp (- first next)))
    (cond
     ((and (> first adjload)
	   (> count 0))
      (debug:print-info 0 *default-log-port* "waiting " waitdelay " seconds due to load " first " exceeding max of " adjload " " (if msg msg ""))
      (thread-sleep! waitdelay)
      (common:wait-for-cpuload maxload numcpus waitdelay count: (- count 1)))
     ((and (> loadjmp numcpus)
	   (> count 0))
      (debug:print-info 0 *default-log-port* "waiting " waitdelay " seconds due to load jump " loadjmp " > numcpus " numcpus (if msg msg ""))
      (thread-sleep! waitdelay)
      (common:wait-for-cpuload maxload numcpus waitdelay count: (- count 1))))))

(define (common:wait-for-homehost-load maxload msg)
  (let* ((hh-dat (if (common:on-homehost?) ;; if we are on the homehost then pass in #f so the calls are local.
                     #f
                     (common:get-homehost)))
         (hh     (if hh-dat (car hh-dat) #f))
         (numcpus (common:get-num-cpus hh)))
    (common:wait-for-normalized-load maxload msg: msg remote-host: hh)))

;; wait for normalized cpu load to drop below maxload
;;
(define (common:wait-for-normalized-load maxload #!key (msg #f)(remote-host #f))
  (let ((num-cpus (common:get-num-cpus remote-host)))
    (common:wait-for-cpuload maxload num-cpus 15 msg: msg remote-host: remote-host)))

;;  hed tal reg reruns reglen regfull test-record runname test-name item-path jobgroup max-concurrent-jobs run-id waitons item-path testmode test-patts required-tests test-registry registry-mutex flags keyvals run-info newtal all-tests-registry itemmaps)
(define (runs:process-expanded-tests runsdat testdat)
  ;; unroll the contents of runsdat and testdat (due to ongoing refactoring).
  (let* ((hed                    (runs:testdat-hed testdat))
	 (tal                    (runs:testdat-tal testdat))
	 (reg                    (runs:testdat-reg testdat))
	 (reruns                 (runs:testdat-reruns testdat))
	 (test-name              (runs:testdat-test-name testdat))
	 (item-path              (runs:testdat-item-path testdat))
	 (jobgroup               (runs:testdat-jobgroup testdat))
	 (waitons                (runs:testdat-waitons testdat))
	 (item-path              (runs:testdat-item-path testdat))
	 (testmode               (runs:testdat-testmode testdat))
	 (newtal                 (runs:testdat-newtal testdat))
	 (itemmaps               (runs:testdat-itemmaps testdat))
	 (test-record            (runs:testdat-test-record testdat))
	 (prereqs-not-met        (runs:testdat-prereqs-not-met testdat))

	 (reglen                 (runs:dat-reglen runsdat))
	 (regfull                (runs:dat-regfull runsdat))
	 (runname                (runs:dat-runname runsdat))
	 (max-concurrent-jobs    (runs:dat-max-concurrent-jobs runsdat))
	 (run-id                 (runs:dat-run-id runsdat))
	 (test-patts             (runs:dat-test-patts runsdat))
	 (required-tests         (runs:dat-required-tests runsdat))
	 (test-registry          (runs:dat-test-registry runsdat))
	 (registry-mutex         (runs:dat-registry-mutex runsdat))
	 (flags                  (runs:dat-flags runsdat))
	 (keyvals                (runs:dat-keyvals runsdat))
	 (run-info               (runs:dat-run-info runsdat))
	 (all-tests-registry     (runs:dat-all-tests-registry runsdat))
	 (run-limits-info        (runs:dat-can-run-more-tests runsdat))
	 ;; (runs:can-run-more-tests run-id jobgroup max-concurrent-jobs)) ;; look at the test jobgroup and tot jobs running
	 (have-resources         (car run-limits-info))
	 (num-running            (list-ref run-limits-info 1))
	 (num-running-in-jobgroup(list-ref run-limits-info 2)) 
	 (max-concurrent-jobs    (list-ref run-limits-info 3))
	 (job-group-limit        (list-ref run-limits-info 4))
	 ;; (prereqs-not-met        (rmt:get-prereqs-not-met run-id waitons hed item-path mode: testmode itemmaps: itemmaps))
	 ;; (prereqs-not-met         (mt:lazy-get-prereqs-not-met run-id waitons item-path mode: testmode itemmap: itemmap))
	 (fails                  (if (list? prereqs-not-met)
				      (runs:calc-fails prereqs-not-met)
				      (begin
					(debug:print-error 0 *default-log-port* "prereqs-not-met is not a list! " prereqs-not-met)
					'())))
	 (non-completed           (filter (lambda (x)             ;; remove hed from not completed list, duh, of course it is not completed!
					    (not (equal? x hed)))
					  (runs:calc-not-completed prereqs-not-met)))
	 (loop-list               (list hed tal reg reruns))
	 ;; configure the load runner
	 (numcpus                 (common:get-num-cpus #f))
	 (maxload                 (string->number (or (configf:lookup *configdat* "jobtools" "maxload") "3.0")))         ;; use a non-number string to disable
         (maxhomehostload         (string->number (or (configf:lookup *configdat* "jobtools" "maxhomehostload") "1.2"))) ;; use a non-number string to disable
         (waitdelay               (string->number (or (configf:lookup *configdat* "jobtools" "waitdelay") "60"))))
    (debug:print-info 4 *default-log-port* "have-resources: " have-resources " prereqs-not-met: (" 
		      (string-intersperse 
		       (map (lambda (t)
			      (if (vector? t)
				  (conc (db:test-get-state t) "/" (db:test-get-status t))
				  (conc " WARNING: t is not a vector=" t )))
			    prereqs-not-met)
		       ", ") ") fails: " fails
		       "\nregistered? " (hash-table-ref/default test-registry (db:test-make-full-name test-name item-path) #f))
			    

    
    (if (and (not (null? prereqs-not-met))
	     (runs:lownoise (conc "waiting on tests " prereqs-not-met hed) 60))
	(debug:print-info 2 *default-log-port* "waiting on tests; " (string-intersperse (runs:mixed-list-testname-and-testrec->list-of-strings prereqs-not-met) ", ")))

    ;; Don't know at this time if the test have been launched at some time in the past
    ;; i.e. is this a re-launch?
    (debug:print-info 4 *default-log-port* "run-limits-info = " run-limits-info)
    
    (cond
     
     ;; Check item path against item-patts, 
     ;;
     ((not (tests:match test-patts (tests:testqueue-get-testname test-record) item-path required: required-tests)) ;; This test/itempath is not to be run
      ;; else the run is stuck, temporarily or permanently
      ;; but should check if it is due to lack of resources vs. prerequisites
      (debug:print-info 1 *default-log-port* "Skipping " (tests:testqueue-get-testname test-record) " " item-path " as it doesn't match " test-patts)
      (if (or (not (null? tal))(not (null? reg)))
	  (list (runs:queue-next-hed tal reg reglen regfull)
		(runs:queue-next-tal tal reg reglen regfull)
		(runs:queue-next-reg tal reg reglen regfull)
		reruns)
	  #f))
     
     ;; Register tests 
     ;;
     ((not (hash-table-ref/default test-registry (db:test-make-full-name test-name item-path) #f))
      (debug:print-info 4 *default-log-port* "Pre-registering test " test-name "/" item-path " to create placeholder" )
      ;; always do firm registration now in v1.60 and greater ;; (eq? *transport-type* 'fs) ;; no point in parallel registration if use fs
      (let register-loop ((numtries 15))
	(rmt:register-test run-id test-name item-path)
	(if (rmt:get-test-id run-id test-name item-path)
	    (hash-table-set! test-registry (db:test-make-full-name test-name item-path) 'done)
	    (if (> numtries 0)
		(begin
		  (thread-sleep! 0.5)
		  (register-loop (- numtries 1)))
		(debug:print-error 0 *default-log-port* "failed to register test " (db:test-make-full-name test-name item-path)))))
      (if (not (eq? (hash-table-ref/default test-registry (db:test-make-full-name test-name "") #f) 'done))
	  (begin
	    (rmt:register-test run-id test-name "")
	    (if (rmt:get-test-id run-id test-name "")
		(hash-table-set! test-registry (db:test-make-full-name test-name "") 'done))))
      (runs:shrink-can-run-more-tests-count runsdat)   ;; DELAY TWEAKER (still needed?)
      (if (and (null? tal)(null? reg))
	  (list hed tal (append reg (list hed)) reruns)
	  (list (runs:queue-next-hed tal reg reglen regfull)
		(runs:queue-next-tal tal reg reglen regfull)
		;; NB// Here we are building reg as we register tests
		;; if regfull we must pop the front item off reg
		(if regfull
		    (append (cdr reg) (list hed))
		    (append reg (list hed)))
		reruns)))
     
     ;; At this point hed test registration must be completed.
     ;;
     ((eq? (hash-table-ref/default test-registry (db:test-make-full-name test-name item-path) #f)
	   'start)
      (debug:print-info 0 *default-log-port* "Waiting on test registration(s): "
			(string-intersperse 
			 (filter (lambda (x)
				   (eq? (hash-table-ref/default test-registry x #f) 'start))
				 (hash-table-keys test-registry))
			 ", "))
      (thread-sleep! 0.051)
      (list hed tal reg reruns))
     
     ;; If no resources are available just kill time and loop again
     ;;
     ((not have-resources) ;; simply try again after waiting a second
      (if (runs:lownoise "no resources" 60)
	  (debug:print-info 1 *default-log-port* "no resources to run new tests, waiting ..."))
      ;; Have gone back and forth on this but db starvation is an issue.
      ;; wait one second before looking again to run jobs.
      (thread-sleep! 1)
      ;; could have done hed tal here but doing car/cdr of newtal to rotate tests
      (list (car newtal)(cdr newtal) reg reruns))
     
     ;; This is the final stage, everything is in place so launch the test
     ;;
     ((and have-resources
	   (or (null? prereqs-not-met)
	       (and (member 'toplevel testmode) ;;  'toplevel)
		    (null? non-completed)
		    (not (member 'exclusive testmode)))))
      ;; (hash-table-delete! *max-tries-hash* (db:test-make-full-name test-name item-path))
      ;; we are going to reset all the counters for test retries by setting a new hash table
      ;; this means they will increment only when nothing can be run
      (set! *max-tries-hash* (make-hash-table))
      ;; well, first lets see if cpu load throttling is enabled. If so wait around until the
      ;; average cpu load is under the threshold before continuing
      (if maxload ;; only gate if maxload is specified
          (common:wait-for-cpuload maxload numcpus waitdelay))
      (if maxhomehostload
          (common:wait-for-homehost-load maxhomehostload (conc "Waiting for homehost load to drop below normalized value of " maxhomehostload)))
      
      (run:test run-id run-info keyvals runname test-record flags #f test-registry all-tests-registry)
      (runs:incremental-print-results run-id)
      (hash-table-set! test-registry (db:test-make-full-name test-name item-path) 'running)
      (runs:shrink-can-run-more-tests-count runsdat)  ;; DELAY TWEAKER (still needed?)
      ;; (thread-sleep! *global-delta*)
      (if (or (not (null? tal))(not (null? reg)))
	  (list (runs:queue-next-hed tal reg reglen regfull)
		(runs:queue-next-tal tal reg reglen regfull)
		(runs:queue-next-reg tal reg reglen regfull)
		reruns)
	  #f))
     
     ;; must be we have unmet prerequisites
     ;;
     (else
      (debug:print 4 *default-log-port* "FAILS: " fails)
      ;; If one or more of the prereqs-not-met are FAIL then we can issue
      ;; a message and drop hed from the items to be processed.
      ;; (runs:mixed-list-testname-and-testrec->list-of-strings prereqs-not-met)
      (if (and (not (null? prereqs-not-met))
	       (runs:lownoise (conc "waiting on tests " prereqs-not-met hed) 60))
	  (debug:print-info 1 *default-log-port* "waiting on tests; " (string-intersperse 
						    (runs:mixed-list-testname-and-testrec->list-of-strings 
						     prereqs-not-met) ", ")))
      (if (or (null? fails)
	      (member 'toplevel testmode))
	  (begin
	    ;; couldn't run, take a breather
	    (if  (runs:lownoise "Waiting for more work to do..." 60)
		 (debug:print-info 0 *default-log-port* "Waiting for more work to do..."))
	    (thread-sleep! 1)
	    (list (car newtal)(cdr newtal) reg reruns))
	  ;; the waiton is FAIL so no point in trying to run hed ever again
	  (if (or (not (null? reg))(not (null? tal)))
	      (if (vector? hed)
		  (begin
		    (debug:print 1 *default-log-port* "WARNING: Dropping test " test-name "/" item-path
				 " from the launch list as it has prerequistes that are FAIL")
		    (let ((test-id (rmt:get-test-id run-id hed "")))
		      (if test-id (mt:test-set-state-status-by-id run-id test-id "NOT_STARTED" "PREQ_FAIL" "Failed to run due to failed prerequisites")))
		    (runs:shrink-can-run-more-tests-count runsdat) ;; DELAY TWEAKER (still needed?)
		    ;; (thread-sleep! *global-delta*)
		    ;; This next is for the items
		    (mt:test-set-state-status-by-testname run-id test-name item-path "NOT_STARTED" "BLOCKED" #f)
		    (hash-table-set! test-registry (db:test-make-full-name test-name item-path) 'removed)
		    (list (runs:queue-next-hed tal reg reglen regfull)
			  (runs:queue-next-tal tal reg reglen regfull)
			  (runs:queue-next-reg tal reg reglen regfull)
			  reruns ;; WAS: (cons hed reruns) ;; but that makes no sense?
			  ))
		  (let ((nth-try (hash-table-ref/default test-registry hed 0)))
		    (cond
		     ((member "RUNNING" (map db:test-get-state prereqs-not-met))
		      (if (runs:lownoise (conc "possible RUNNING prerequistes " hed) 60)
			  (debug:print 0 *default-log-port* "WARNING: test " hed " has possible RUNNING prerequisites, don't give up on it yet."))
		      (thread-sleep! 4)
		      (list (runs:queue-next-hed newtal reg reglen regfull)
			    (runs:queue-next-tal newtal reg reglen regfull)
			    (runs:queue-next-reg newtal reg reglen regfull)
			    reruns))
		     ((or (not nth-try)
			  (and (number? nth-try)
			       (< nth-try 10)))
		      (hash-table-set! test-registry hed (if (number? nth-try)
							     (+ nth-try 1)
							     0))
		      (if (runs:lownoise (conc "not removing test " hed) 60)
			  (debug:print 1 *default-log-port* "WARNING: not removing test " hed " from queue although it may not be runnable due to FAILED prerequisites"))
		      ;; may not have processed correctly. Could be a race condition in your test implementation? Dropping test " hed) ;;  " as it has prerequistes that are FAIL. (NOTE: hed is not a vector)")
		      (runs:shrink-can-run-more-tests-count runsdat) ;; DELAY TWEAKER (still needed?)
		      ;; (list hed tal reg reruns)
		      ;; (list (car newtal)(cdr newtal) reg reruns)
		      ;; (hash-table-set! test-registry hed 'removed)
		      (list (runs:queue-next-hed newtal reg reglen regfull)
			    (runs:queue-next-tal newtal reg reglen regfull)
			    (runs:queue-next-reg newtal reg reglen regfull)
			    reruns))
		     ((symbol? nth-try)
		      (if (eq? nth-try 'removed) ;; removed is removed - drop it NOW
			  (if (null? tal)
			      #f ;; yes, really
			      (list (car tal)(cdr tal) reg reruns))
			  (begin
			    (if (runs:lownoise (conc "FAILED prerequisites or other issue" hed) 60)
				(debug:print 0 *default-log-port* "WARNING: test " hed " has FAILED prerequisites or other issue. Internal state " nth-try " will be overridden and we'll retry."))
			    (mt:test-set-state-status-by-testname run-id test-name item-path "NOT_STARTED" "KEEP_TRYING" #f)
			    (hash-table-set! test-registry hed 0)
			    (list (runs:queue-next-hed newtal reg reglen regfull)
				  (runs:queue-next-tal newtal reg reglen regfull)
				  (runs:queue-next-reg newtal reg reglen regfull)
				  reruns))))
		     (else
		      (if (runs:lownoise (conc "FAILED prerequitests and we tried" hed) 60)
			  (debug:print 0 *default-log-port* "WARNING: test " hed " has FAILED prerequitests and we've tried at least 10 times to run it. Giving up now."))
		      ;; (debug:print 0 *default-log-port* "         prereqs: " prereqs-not-met)
		      (hash-table-set! test-registry hed 'removed)
		      (mt:test-set-state-status-by-testname run-id test-name item-path "NOT_STARTED" "TEN_STRIKES" #f)
		      ;; I'm unclear on if this roll up is needed - it may be the root cause of the "all set to FAIL" bug.
		      (rmt:set-state-status-and-roll-up-items run-id test-name item-path #f "FAIL" #f) ;; treat as FAIL
		      (list (if (null? tal)(car newtal)(car tal))
			    tal
			    reg
			    reruns)))))
	      ;; can't drop this - maybe running? Just keep trying
	      (let ((runable-tests (runs:runable-tests prereqs-not-met)))
		(if (null? runable-tests)
		    #f   ;; I think we are truly done here
		    (list (runs:queue-next-hed newtal reg reglen regfull)
			    (runs:queue-next-tal newtal reg reglen regfull)
			    (runs:queue-next-reg newtal reg reglen regfull)
			    reruns)))))))))

;;======================================================================

(declare (unit env))

(use sql-de-lite) ;; srfi-1 posix regex regex-case srfi-69 srfi-18 call-with-environment-variables)

(define (env:open-db fname)
  (let* ((db-exists (file-exists? fname))
  (let* ((db-exists (common:file-exists? fname))
	 (db        (open-database fname)))
    (if (not db-exists)
	(begin
	  (exec (sql db "CREATE TABLE envvars (
                    id INTEGER PRIMARY KEY,
                    context TEXT NOT NULL,
                    var TEXT NOT NULL,

	 (rollup-status 0)
	 (exit-info     (vector #t #t #t))
	 (test-id       (db:test-get-id testdat))
	 (run-id        (db:test-get-run_id testdat))
	 (test-name     (db:test-get-testname testdat))
	 (kill-job      #f)) ;; for future use (on re-factoring with launch.scm code
    (let loop ((count 5))
      (if (file-exists? test-run-dir)
      (if (common:file-exists? test-run-dir)
	  (push-directory test-run-dir)
	  (if (> count 0)
	      (begin
		(debug:print 0 *default-log-port* "WARNING: ezsteps attempting to run but test run directory " test-run-dir " is not there. Waiting and trying again " count " more times")
		(sleep 3)
		(loop (- count 1))))))
    (debug:print-info 0 *default-log-port* "Running in directory " test-run-dir)
    (if (not (file-exists? ".ezsteps"))(create-directory ".ezsteps"))
    (if (not (common:file-exists? ".ezsteps"))(create-directory ".ezsteps"))
    ;; if ezsteps was defined then we are sure to have at least one step but check anyway
    (if (not (> (length ezstepslst) 0))
	(message-window "ERROR: You can only re-run steps defined via ezsteps")
	(begin
	  (let loop ((ezstep   (car ezstepslst))
		     (tal      (cdr ezstepslst))
		     (prevstep #f)

			  (set! runflag #t) ;; and continue
			  (if (not (null? tal))
			      (loop (car tal)(cdr tal) stepname #f))))

		  (debug:print 4 *default-log-port* "ezsteps:\n stepname: " stepname " stepinfo: " stepinfo " stepparts: " stepparts
			       " stepparms: " stepparms " stepcmd: " stepcmd)
		  
		  (if (file-exists? (conc stepname ".logpro"))(set! logpro-used #t))
		  (if (common:file-exists? (conc stepname ".logpro"))(set! logpro-used #t))
		  
		  ;; call the command using mt_ezstep
		  (set! script (conc "mt_ezstep " stepname " " (if prevstep prevstep "-") " " stepcmd))
		  
		  (debug:print 4 *default-log-port* "script: " script)
		  (rmt:teststep-set-status! run-id test-id stepname "start" "-" #f #f)
		  ;; now launch

(use (prefix sqlite3 sqlite3:) posix typed-records) 

(define (open-tail-db )
  (let* ((basedir   (create-directory (conc "/tmp/" (current-user-name))))
	 (dbpath    (conc basedir "/megatest_logs.db"))
	 (dbexists  (common:file-exists? dbpath))
	 (db        (sqlite3:open-database dbpath))
	 (handler   (sqlite3:make-busy-timeout 136000)))
    (sqlite3:set-busy-handler! db handler)
    (sqlite3:execute db "PRAGMA synchronous = 0;")
    (if (not dbexists)
	(begin
	  (sqlite3:execute db "CREATE TABLE IF NOT EXISTS log_files (id INTEGER PRIMARY KEY,filename TEXT,event_time TIMESTAMP DEFAULT (strftime('%s','now')));")
	  (sqlite3:execute db "CREATE TABLE IF NOT EXISTS log_data  (id INTEGER PRIMARY KEY,fid INTEGER,line TEXT,event_time TIMESTAMP DEFAULT (strftime('%s','now')));")
	  ))
    db))

(define (tail-write db fid lines)
  (sqlite3:with-transaction
   db
   (lambda ()
     (for-each
      (lambda (line)
	(sqlite3:execute db "INSERT INTO log_data (fid,line) VALUES (?,?);" fid line))
      lines))))

(define (tail-get-fid db fname)
  (let ((fid   (handle-exceptions
		   exn
		   #f
		 (sqlite3:first-result db "SELECT id FROM log_files WHERE filename=?;" fname))))
    (if fid
	fid
	(begin
	  (sqlite3:execute db "INSERT INTO log_files (filename) VALUES (?);" fname)
	  (tail-get-fid db fname)))))

(define (file-tail fname #!key (db-in #f))
  (let* ((inp (open-input-file fname))
	 (db  (or db-in (open-tail-db)))
	 (fid (tail-get-fid db fname)))
    (let loop ((inl    (read-line inp))
	       (lines '())
	       (lastwr (current-seconds)))
      (if (eof-object? inl)
	  (let ((timed-out (> (- (current-seconds) lastwr) 60)))
	    (if timed-out (tail-write db fid (reverse lines)))
	    (sleep 1)
	    (if timed-out
		(loop (read-line inp) '() (current-seconds))
		(loop (read-line inp) lines lastwr)))
	  (let* ((savelines (> (length lines) 19)))
	    ;; (print inl)
	    (if savelines (tail-write db fid (reverse lines)))
	    (loop (read-line inp)
		  (if savelines
		      '()
		      (cons inl lines))
		  (if savelines
		      (current-seconds)
		      lastwr)))))))

;; offset -20 means get last 20 lines
;;
(define (tail-get-lines db fid offset count)
  (if (> offset 0)
      (map-row (lambda (id line)
		 (vector id line))
	       db
	       "SELECT id,line FROM log_data WHERE fid=? OFFSET ? LIMIT ?;" fid offset count)
      (reverse ;; get N from the end
       (map-row (lambda (id line)
		  (vector id line))
		db
		"SELECT id,line FROM log_data WHERE fid=? ORDER BY id DESC LIMIT ?;" fid (abs offset)))))

(declare (unit filedb))

(include "fdb_records.scm")
;; (include "settings.scm")

(define (filedb:open-db dbpath)
  (let* ((fdb      (make-filedb:fdb))
	 (dbexists (file-exists? dbpath))
	 (dbexists (common:file-exists? dbpath))
	 (db (sqlite3:open-database dbpath)))
    (filedb:fdb-set-db!        fdb db)
    (filedb:fdb-set-dbpath!    fdb dbpath)
    (filedb:fdb-set-pathcache! fdb (make-hash-table))
    (filedb:fdb-set-idcache!   fdb (make-hash-table))
    (filedb:fdb-set-partcache! fdb (make-hash-table))
    (sqlite3:set-busy-handler!  db (make-busy-timeout 136000))

    (if (not (directory? path))
	(begin
	  (print "The path " path " does not exist or is not a directory. Attempting to create it now")
	  (create-directory path #t)))

    ;; First check that the directory is empty!
    (if (and (file-exists? path)
    (if (and (common:file-exists? path)
	     (not (null? (glob (conc path "/*")))))
	(begin
	  (print "WARNING: directory " path " is not empty, are you sure you want to continue?")
	  (display "Enter y/n: ")
	  (if (equal? "y" (read-line))
	      (print "Using directory " path " for your Megatest area.")
	      (begin

	(description #f)
	(steps    '())
	(scripts  '())
	(items    '())
	(rel-path #f))

    (cond
     ((file-exists? "megatest.config")         (set! rel-path "./"))
     ((file-exists? "../megatest.config")      (set! rel-path "../"))
     ((file-exists? "../../megatest.config")   (set! rel-path "../../"))
     ((file-exists? "../../../megatest.config")(set! rel-path "../../../"))) ;; good enough dang it.
     ((common:file-exists? "megatest.config")         (set! rel-path "./"))
     ((common:file-exists? "../megatest.config")      (set! rel-path "../"))
     ((common:file-exists? "../../megatest.config")   (set! rel-path "../../"))
     ((common:file-exists? "../../../megatest.config")(set! rel-path "../../../"))) ;; good enough dang it.

    ;; Don't gather data or continue if a) megatest.config can't be found or b) testconfig already exists
    (if (not rel-path)
	(begin
	  (print "ERROR: I could not find megatest.config, please run -create-test in the top dir of your megatest area")
	  (exit 1)))

    (if (file-exists? (conc rel-path "tests/" testname "/testconfig"))
    (if (common:file-exists? (conc rel-path "tests/" testname "/testconfig"))
	(begin
	  (print "WARNING: You already have a testconfig in " rel-path "tests/" testname ", do you want to clobber your files?")
	  (display "Enter y/n: ")
	  (if (not (equal? "y" (read-line)))
	      (begin
		(print "INFO: user abort of creation of test " testname)
		(exit 1)))))

  '((PASS . "70 249 73")
    (FAIL . "253 33 49")
    (SKIP . "230 230 0")))

(define (gutils:get-color-spec effective-state)
  (or (alist-ref effective-state gutils:colors)
      (alist-ref 'FAIL gutils:colors)))
 

;; BBnote - state status dashboard button color / text defined here
(define (gutils:get-color-for-state-status state status);; #!key (get-label #f))
  ;; ((if get-label cadr car)
  (case (string->symbol state)
    ((COMPLETED) ;; ARCHIVED)
     (case (string->symbol status)
       ((PASS)        (list "70  249 73" status))
       ((WARN WAIVED) (list "255 172 13" status))

    ;;	  (if (or (equal? status "WARN")
    ;;		  (equal? status "WAIVED"))
    ;;	      (list "255 172 13" status)
    ;;	      (list "223 33 49"  status)))) ;; greenish orangeish redish
    ((LAUNCHED)         (list "101 123 142"  state))
    ((CHECK)            (list "255 100 50"   state))
    ((REMOTEHOSTSTART)  (list "50 130 195"   state))
    ((RUNNING)          (list "9 131 232"    state))
    ((RUNNING STARTED)          (list "9 131 232"    state))
    ((KILLREQ)          (list "39 82 206"    state))
    ((KILLED)           (list "234 101 17"   state))
    ((NOT_STARTED)      (case (string->symbol status)
			  ((CHECK)(list (gutils:get-color-spec 'SKIP) state))
			  ((CHECK STARTED)(list (gutils:get-color-spec 'SKIP) state))
			  (else   (list "240 240 240"                 state))))
    ;; for xor mode below
    ;;
    ((CLEAN)
     (case (string->symbol status)
       ((CLEAN-FAIL CLEAN-CHECK CLEAN-ABORT)  (list "200 130 13" status)) ;; orange requested for these
       (else  (list "60  235 63" status))))
    ((DIRTY-BETTER)     (list "160  255 153" status))
    ((DIRTY-WORSE)      (list "165 42  42" status))
    ((BOTH-BAD)         (list "180 33 49" status))

    (else               (list "192 192 192"  state))))

	exn
	(begin
	  (print-error-message exn)
	  (if (< portnum 64000)
	      (begin 
		(debug:print 0 *default-log-port* "WARNING: attempt to start server failed. Trying again ...")
		(debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
		(debug:print 0 *default-log-port* "exn=" (condition->list exn))
		(debug:print 5 *default-log-port* "exn=" (condition->list exn))
		(portlogger:open-run-close portlogger:set-failed portnum)
		(debug:print 0 *default-log-port* "WARNING: failed to start on portnum: " portnum ", trying next port")
		(thread-sleep! 0.1)
		
		;; get_next_port goes here
		(http-transport:try-start-server ipaddrstr
						 (portlogger:open-run-close portlogger:find-port)))

(define (http-transport:close-connections #!key (area-dat #f))
  (let* ((runremote  (or area-dat *runremote*))
	 (server-dat (if runremote
                         (remote-conndat runremote)
                         #f))) ;; (hash-table-ref/default *runremote* run-id #f)))
    (if (vector? server-dat)
	(let ((api-dat (http-transport:server-dat-get-api-uri server-dat)))
	  (handle-exceptions
	    exn
	    (begin
	      (print-call-chain *default-log-port*)
	      (debug:print-error 0 *default-log-port* " closing connection failed with error: " ((condition-property-accessor 'exn 'message) exn)))
	  (close-connection! api-dat)
	  #t)
	    (close-connection! api-dat)
	    #t))
	#f)))


(define (make-http-transport:server-dat)(make-vector 6))
(define (http-transport:server-dat-get-iface         vec)    (vector-ref  vec 0))
(define (http-transport:server-dat-get-port          vec)    (vector-ref  vec 1))
(define (http-transport:server-dat-get-api-uri       vec)    (vector-ref  vec 2))

	    (debug:print 0 *default-log-port* "SERVER STARTED: " iface ":" port " AT " (current-seconds))
	    (flush-output *default-log-port*)))
      (if (common:low-noise-print 60 "dbstats")
	  (begin
	    (debug:print 0 *default-log-port* "Server stats:")
	    (db:print-current-query-stats)))
      (let* ((hrs-since-start  (/ (- (current-seconds) server-start-time) 3600))
	     (adjusted-timeout (if (> hrs-since-start 1)
	     (adjusted-timeout (if (> hrs-since-start 1)  ;; never used!
				   (- server-timeout (inexact->exact (round (* hrs-since-start 60))))  ;; subtract 60 seconds per hour
				   server-timeout)))
	(if (common:low-noise-print 120 "server timeout")
	    (debug:print-info 0 *default-log-port* "Adjusted server timeout: " adjusted-timeout))
	(cond
         ((and *server-run*
	       (> (+ last-access server-timeout)
		  (current-seconds))
	       (< (- (current-seconds) server-start-time) 3600)) ;; do not update log or touch log if we've been running for more than one hour.
          (if (common:low-noise-print 120 "server continuing")
              (debug:print-info 0 *default-log-port* "Server continuing, seconds since last db access: " (- (current-seconds) last-access))
	      (let ((curr-time (current-seconds)))
		(handle-exceptions
		    exn
		    (debug:print 0 *default-log-port* "ERROR: Failed to change timestamp on log file " server-log-file ". Are you out of space on that disk?")
		  (if (not *server-overloaded*)
		  (change-file-times server-log-file curr-time curr-time))))
		      (change-file-times server-log-file curr-time curr-time)))))
          (loop 0 server-state bad-sync-count (current-milliseconds)))
         (else
          (debug:print-info 0 *default-log-port* "Server timed out. seconds since last db access: " (- (current-seconds) last-access))
          (http-transport:server-shutdown port)))))))

(define (http-transport:server-shutdown port)
  (begin

    (exit)))

;; all routes though here end in exit ...
;;
;; start_server? 
;;
(define (http-transport:launch)
  ;; lets not even bother to start if there are already three or more server files ready to go
  (let* ((num-alive   (server:get-num-alive (server:get-list *toppath*))))
  ;; (if (args:get-arg "-daemonize")
  ;;     (begin
    (if (> num-alive 3)
	(begin
  ;; 	(daemon:ize)
  ;; 	(if *alt-log-file* ;; we should re-connect to this port, I think daemon:ize disrupts it
  ;; 	    (begin
	  (debug:print 0 *default-log-port* "ERROR: Aborting server start because there are already " num-alive " possible servers either running or starting up")
  ;; 	      (current-error-port *alt-log-file*)
  ;; 	      (current-output-port *alt-log-file*)))))
	  (exit))))
  (let* ((th2 (make-thread (lambda ()
			     (debug:print-info 0 *default-log-port* "Server run thread started")
			     (http-transport:run 
			      (if (args:get-arg "-server")
				  (args:get-arg "-server")
				  "-")
			      )) "Server run"))

(defstruct launch:einf (pid #t)(exit-status #t)(exit-code #t)(rollup-status 0))

;; return (conc status ": " comment) from the final section so that
;;   the comment can be set in the step record in launch.scm
;;
(define (launch:load-logpro-dat run-id test-id stepname)
  (let ((cname (conc stepname ".dat")))
    (if (file-exists? cname)
    (if (common:file-exists? cname)
	(let* ((dat  (read-config cname #f #f))
	       (csvr (db:logpro-dat->csv dat stepname))
	       (csvt (let-values (((fmt-cell fmt-record fmt-csv) (make-format ",")))
		       (fmt-csv (map list->csv-record csvr))))
	       (status (configf:lookup dat "final" "exit-status"))
	       (msg     (configf:lookup dat "final" "message")))
          (if csvt  ;; this if blocked stack dump caused by .dat file from logpro being 0-byte.  fixed by upgrading logpro

	 (stepparms      (list-ref stepparts 2)) ;; for future use, {VAR=1,2,3}, run step for each 
	 (stepcmd        (list-ref stepparts 3))
	 (script         "") ; "#!/bin/bash\n") ;; yep, we depend on bin/bash FIXME!!!\
	 (logpro-file    (conc stepname ".logpro"))
	 (html-file      (conc stepname ".html"))
	 (dat-file       (conc stepname ".dat"))
	 (tconfig-logpro (configf:lookup testconfig "logpro" stepname))
	 (logpro-used    (file-exists? logpro-file)))
	 (logpro-used    (common:file-exists? logpro-file)))

    (if (and tconfig-logpro
	     (not logpro-used)) ;; no logpro file found but have a defn in the testconfig
	(begin
	  (with-output-to-file logpro-file
	    (lambda ()
	      (print ";; logpro file extracted from testconfig\n"
		     ";;")
	      (print tconfig-logpro)))
	  (set! logpro-used #t)))
    
    ;; NB// can safely assume we are in test-area directory
    (debug:print 4 *default-log-port* "ezsteps:\n stepname: " stepname " stepinfo: " stepinfo " stepparts: " stepparts
		 " stepparms: " stepparms " stepcmd: " stepcmd)
    
    ;; ;; first source the previous environment
    ;; (let ((prev-env (conc ".ezsteps/" prevstep (if (string-search (regexp "csh") 
    ;;      							 (get-environment-variable "SHELL")) ".csh" ".sh"))))
    ;;   (if (and prevstep (file-exists? prev-env))
    ;;   (if (and prevstep (common:file-exists? prev-env))
    ;;       (set! script (conc script "source " prev-env))))
    
    ;; call the command using mt_ezstep
    ;; (set! script (conc "mt_ezstep " stepname " " (if prevstep prevstep "x") " " stepcmd))
    
    (debug:print 4 *default-log-port* "script: " script)
    (rmt:teststep-set-status! run-id test-id stepname "start" "-" #f #f)
    ;; now launch the actual process
    (call-with-environment-variables 
     (list (cons "PATH" (conc (get-environment-variable "PATH") ":.")))
     (lambda () ;; (process-run "/bin/bash" "-c" "exec ls -l /tmp/foobar > /tmp/delme-more.log 2>&1")
       (let* ((cmd (conc stepcmd " > " stepname ".log 2>&1")) ;; >outfile 2>&1 
	      (pid (process-run "/bin/bash" (list "-c" cmd))))

         (with-output-to-file "Makefile.ezsteps"
           (lambda ()
             (print stepname ".log :")
             (print "\t" cmd)
             (if (file-exists? (conc stepname ".logpro"))
             (if (common:file-exists? (conc stepname ".logpro"))
                 (print "\tlogpro " stepname ".logpro " stepname ".html < " stepname ".log"))
             (print)
             (print stepname " : " stepname ".log")
             (print))
           #:append)

	 (rmt:test-set-top-process-pid run-id test-id pid)

    
    (let ((exinfo (launch:einf-exit-code exit-info)) ;; (vector-ref exit-info 2))
	  (logfna (if logpro-used (conc stepname ".html") ""))
	  (comment #f))
      (if logpro-used
	  (let ((datfile (conc stepname ".dat")))
	    ;; load the .dat file into the test_data table if it exists
	    (if (file-exists? datfile)
	    (if (common:file-exists? datfile)
		(set! comment (launch:load-logpro-dat run-id test-id stepname)))
	    (rmt:test-set-log! run-id test-id (conc stepname ".html"))))
      (rmt:teststep-set-status! run-id test-id stepname "end" exinfo comment logfna))
    ;; set the test final status
    (let* ((process-exit-status (launch:einf-exit-code exit-info)) ;; (vector-ref exit-info 2))
	   (this-step-status (cond
			      ((and (eq? process-exit-status 2) logpro-used) 'warn)   ;; logpro 2 = warnings

	      (debug:print 0 *default-log-port* "WARNING: no testconfig found for " test-name " in search path:\n  "
			   (string-intersperse (tests:get-tests-search-path *configdat*) "\n  "))))
	;; after all that, still no testconfig? Time to abort
	(if (not testconfig)
	    (begin
	      (debug:print-error 0 *default-log-port* "Failed to resolve megatest.config, runconfigs.config and testconfig issues. Giving up now")
	      (exit 1)))
	(if (not (file-exists? ".ezsteps"))(create-directory ".ezsteps"))
	(if (not (common:file-exists? ".ezsteps"))(create-directory ".ezsteps"))
	;; if ezsteps was defined then we are sure to have at least one step but check anyway
	(if (not (> (length ezstepslst) 0))
	    (debug:print-error 0 *default-log-port* "ezsteps defined but ezstepslst is zero length")
	    (let loop ((ezstep (car ezstepslst))
		       (tal    (cdr ezstepslst))
		       (prevstep #f))
	      ;; check exit-info (vector-ref exit-info 1)
	      (if (launch:einf-exit-status exit-info) ;; (vector-ref exit-info 1)
		  (let ((logpro-used (launch:runstep ezstep run-id test-id exit-info m tal testconfig))
			(stepname    (car ezstep)))
		    ;; if logpro-used read in the stepname.dat file
		    (if (and logpro-used (file-exists? (conc stepname ".dat")))
		    (if (and logpro-used (common:file-exists? (conc stepname ".dat")))
			(launch:load-logpro-dat run-id test-id stepname))
		    (if (steprun-good? logpro-used (launch:einf-exit-code exit-info))
			(if (not (null? tal))
			    (loop (car tal) (cdr tal) stepname))
			(debug:print 4 *default-log-port* "WARNING: step " (car ezstep) " failed. Stopping")))
		  (debug:print 4 *default-log-port* "WARNING: a prior step failed, stopping at " ezstep)))))))

(define (launch:monitor-job run-id test-id item-path fullrunscript ezsteps test-name tconfigreg exit-info m work-area runtlim misc-flags)
  (let* ((update-period (string->number (or (configf:lookup *configdat* "setup" "test-stats-update-period") "30")))
  (let* ((start-seconds (current-seconds))
         (start-seconds (current-seconds))
	 (calc-minutes  (lambda ()
			  (inexact->exact 
			   (round 
			    (- 
			     (current-seconds) 
			     start-seconds)))))
	 (kill-tries 0))
    ;; (tests:set-full-meta-info #f test-id run-id (calc-minutes) work-area)
    ;; (tests:set-full-meta-info test-id run-id (calc-minutes) work-area)
    (tests:set-full-meta-info #f test-id run-id (calc-minutes) work-area 10)
    (let loop ((minutes   (calc-minutes))
	       (cpu-load  (alist-ref 'adj-core-load (common:get-normalized-cpu-load #f)))
	       (disk-free (get-df (current-directory))))
      (let ((new-cpu-load (let* ((load  (alist-ref 'adj-core-load (common:get-normalized-cpu-load #f)))
				 (delta (abs (- load cpu-load))))
			    (if (> delta 0.1) ;; don't bother updating with small changes
				load
				#f)))
	    (new-disk-free (let* ((df    (get-df (current-directory)))
				  (delta (abs (- df disk-free))))
			     (if (> delta 200) ;; ignore changes under 200 Meg
				 df
				 #f))))
	       (disk-free (get-df (current-directory)))
               (last-sync (current-seconds)))
      (let* ((over-time     (> (current-seconds) (+ last-sync update-period)))
             (new-cpu-load  (let* ((load  (alist-ref 'adj-core-load (common:get-normalized-cpu-load #f)))
                                   (delta (abs (- load cpu-load))))
                              (if (> delta 0.1) ;; don't bother updating with small changes
                                  load
                                  #f)))
             (new-disk-free (let* ((df    (if over-time ;; only get df every 30 seconds
                                              (get-df (current-directory))
                                              disk-free))
                                   (delta (abs (- df disk-free))))
                              (if (and (> df 0)
                                       (> (/ delta df) 0.1)) ;; (> delta 200) ;; ignore changes under 200 Meg
                                  df
                                  #f)))
             (do-sync       (or new-cpu-load new-disk-free over-time)))
        (debug:print 4 *default-log-port* "cpu: " new-cpu-load " disk: " new-disk-free " last-sync: " last-sync " do-sync: " do-sync)
	(set! kill-job? (or (test-get-kill-request run-id test-id) ;; run-id test-name itemdat))
			    (and runtlim (let* ((run-seconds   (- (current-seconds) start-seconds))
						(time-exceeded (> run-seconds runtlim)))
					   (if time-exceeded
					       (begin
						 (debug:print-info 0 *default-log-port* "KILLING TEST DUE TO TIME LIMIT EXCEEDED! Runtime=" run-seconds " seconds, limit=" runtlim)
						 #t)
					       #f)))))
        (if do-sync
	(tests:update-central-meta-info run-id test-id new-cpu-load new-disk-free (calc-minutes) #f #f)
            (tests:update-central-meta-info run-id test-id new-cpu-load new-disk-free (calc-minutes) #f #f))
	(if kill-job? 
	    (begin
	      (mutex-lock! m)
	      ;; NOTE: The pid can change as different steps are run. Do we need handshaking between this
	      ;;       section and the runit section? Or add a loop that tries three times with a 1/4 second
	      ;;       between tries?
	      (let* ((pid1 (launch:einf-pid exit-info)) ;; (vector-ref exit-info 0))

	      (mutex-unlock! m)
	      ;; no point in sticking around. Exit now.
	      (exit)))
	(if (hash-table-ref/default misc-flags 'keep-going #f)
	    (begin
	      (thread-sleep! 3) ;; (+ 3 (random 6))) ;; add some jitter to the call home time to spread out the db accesses
	      (if (hash-table-ref/default misc-flags 'keep-going #f)  ;; keep originals for cpu-load and disk-free unless they change more than the allowed delta
		  (loop (calc-minutes) (or new-cpu-load cpu-load) (or new-disk-free disk-free)))))))
		  (loop (calc-minutes)
                        (or new-cpu-load cpu-load)
                        (or new-disk-free disk-free)
                        (if do-sync (current-seconds) last-sync)))))))
    (tests:update-central-meta-info run-id test-id (get-cpu-load) (get-df (current-directory))(calc-minutes) #f #f))) ;; NOTE: Checking twice for keep-going is intentional


(define (launch:execute encoded-cmd)
  (let* ((cmdinfo    (common:read-encoded-string encoded-cmd))
	 (tconfigreg #f))
    (setenv "MT_CMDINFO" encoded-cmd)

	       (keys      #f)
	       (keyvals   #f)
	       (fullrunscript (if (not runscript)
                                  #f
                                  (if (substring-index "/" runscript)
                                      runscript ;; use unadultered if contains slashes
                                      (let ((fulln (conc testpath "/" runscript)))
	                                  (if (and (file-exists? fulln)
	                                  (if (and (common:file-exists? fulln)
                                                   (file-execute-access? fulln))
                                              fulln
                                              runscript))))) ;; assume it is on the path
	       ) ;; (rollup-status 0)

	  (if contour (setenv "MT_CONTOUR" contour))
	  

		  (handle-exceptions
		      exn
		      (debug:print 0 *default-log-port* "Failed to create directory " logdir " expect problems, message: " ((condition-property-accessor 'exn 'message) exn))
		    (create-directory logdir #t)))))
		  
	  ;; NFS might not have propagated the directory meta data to the run host - give it time if needed
	  (let loop ((count 0))
	    (if (or (file-exists? top-path)
	    (if (or (common:file-exists? top-path)
		    (> count 10))
		(change-directory top-path)
		(begin
		  (debug:print 0 *default-log-port* "INFO: Not starting job yet - directory " top-path " not found")
		  (thread-sleep! 10)
		  (loop (+ count 1)))))
	  (launch:setup) ;; should be properly in the top-path now

	    (set-signal-handler! signal/term sighand)
	    ) ;; (set-signal-handler! signal/stop sighand)
	  
	  ;; Do not run the test if it is REMOVING, RUNNING, KILLREQ or REMOTEHOSTSTART,
	  ;; Mark the test as REMOTEHOSTSTART *IMMEDIATELY*
	  ;;
	  (let* ((test-info (rmt:get-test-info-by-id run-id test-id))
		 (test-host (if test-info
		 (test-host (db:test-get-host        test-info))
				(db:test-get-host        test-info)
				(begin
				  (debug:print 0 *default-log-port* "ERROR: failed to find a record for test-id " test-id ", exiting.")
				  (exit))))
		 (test-pid  (db:test-get-process_id  test-info)))
	    (cond
	     ((member (db:test-get-state test-info) '("INCOMPLETE" "KILLED" "UNKNOWN" "KILLREQ" "STUCK")) ;; prior run of this test didn't complete, go ahead and try to rerun
	      (debug:print 0 *default-log-port* "INFO: test is INCOMPLETE or KILLED, treat this execute call as a rerun request")
	      ;; (tests:test-force-state-status! run-id test-id "REMOTEHOSTSTART" "n/a")
	      (rmt:test-set-state-status run-id test-id "REMOTEHOSTSTART" "n/a" #f)
	      ) ;; prime it for running

					  (debug:print-error 0 *default-log-port* "bad variable spec, " var "=" val))))
				  (configf:get-section rconfig section)))
		      (list "default" target)))
          ;;(bb-check-path msg: "launch:execute post block 1")

	  ;; NFS might not have propagated the directory meta data to the run host - give it time if needed
	  (let loop ((count 0))
	    (if (or (file-exists? work-area)
	    (if (or (common:file-exists? work-area)
		    (> count 10))
		(change-directory work-area)
		(begin
		  (debug:print 0 *default-log-port* "INFO: Not starting job yet - directory " work-area " not found")
		  (thread-sleep! 10)
		  (loop (+ count 1)))))
          ;;(bb-check-path msg: "launch:execute post block 1.5")

	  ;; environment overrides are done *before* the remaining critical envars.
	  (alist->env-vars env-ovrd)
          ;;(bb-check-path msg: "launch:execute post block 41")
	  (runs:set-megatest-env-vars run-id inkeys: keys inkeyvals: keyvals)
          ;;(bb-check-path msg: "launch:execute post block 42")
	  (set-item-env-vars itemdat)
          ;;(bb-check-path msg: "launch:execute post block 43")
          (let ((blacklist (configf:lookup *configdat* "setup" "blacklistvars")))
            (if blacklist
                (save-environment-as-files "megatest" ignorevars: (string-split blacklist))
	  (save-environment-as-files "megatest")
                (save-environment-as-files "megatest")))
          ;;(bb-check-path msg: "launch:execute post block 44")
	  ;; open-run-close not needed for test-set-meta-info
	  ;; (tests:set-full-meta-info #f test-id run-id 0 work-area)
	  ;; (tests:set-full-meta-info test-id run-id 0 work-area)
	  (tests:set-full-meta-info #f test-id run-id 0 work-area 10)

	  ;; (thread-sleep! 0.3) ;; NFS slowness has caused grief here

	  (if (args:get-arg "-xterm")
	      (set! fullrunscript "xterm")
	      (if (and fullrunscript 
		       (file-exists? fullrunscript)
		       (common:file-exists? fullrunscript)
		       (not (file-execute-access? fullrunscript)))
		  (system (conc "chmod ug+x " fullrunscript))))

	  ;; We are about to actually kick off the test
	  ;; so this is a good place to remove the records for 
	  ;; any previous runs
	  ;; (db:test-remove-steps db run-id testname itemdat)

	    (mutex-unlock! m)
	    (debug:print 2 *default-log-port* "Output from running " fullrunscript ", pid " (launch:einf-pid exit-info) " in work area " 
			 work-area ":\n====\n exit code " (launch:einf-exit-code exit-info) "\n" "====\n")
	    (if (not (launch:einf-exit-status exit-info))
		(exit 4))))
        )))

;; DO NOT USE - caching of configs is handled in launch:setup now.
;;
(define (launch:cache-config)
  ;; if we have a linktree and -runtests and -target and the directory exists dump the config
  ;; to megatest-(current-seconds).cfg and symlink it to megatest.cfg
  (if (and *configdat* 
	   (or (args:get-arg "-run")
	       (args:get-arg "-runtests")
	       (args:get-arg "-execute")))
      (let* ((linktree (common:get-linktree)) ;; (get-environment-variable "MT_LINKTREE"))
	     (target   (common:args-get-target exit-if-bad: #t))
	     (runname  (or (args:get-arg "-runname")
			   (args:get-arg ":runname")
			   (getenv "MT_RUNNAME")))
	     (fulldir  (conc linktree "/"
			     target "/"
			     runname)))
	(if (and linktree (file-exists? linktree)) ;; can't proceed without linktree
	(if (and linktree (common:file-exists? linktree)) ;; can't proceed without linktree
	    (begin
	      (debug:print-info 0 *default-log-port* "Have -run with target=" target ", runname=" runname ", fulldir=" fulldir ", testpatt=" (or (args:get-arg "-testpatt") "%"))
	      (if (not (file-exists? fulldir))
	      (if (not (common:file-exists? fulldir))
		  (create-directory fulldir #t)) ;; need to protect with exception handler 
	      (if (and target
		       runname
		       (file-exists? fulldir))
		       (common:file-exists? fulldir))
		  (let ((tmpfile  (conc fulldir "/.megatest.cfg." (current-seconds)))
			(targfile (conc fulldir "/.megatest.cfg-"  megatest-version "-" megatest-fossil-hash))
			(rconfig  (conc fulldir "/.runconfig." megatest-version "-" megatest-fossil-hash)))
		    (if (file-exists? rconfig) ;; only cache megatest.config AFTER runconfigs has been cached
		    (if (common:file-exists? rconfig) ;; only cache megatest.config AFTER runconfigs has been cached
			(begin
			  (debug:print-info 0 *default-log-port* "Caching megatest.config in " tmpfile)
                          (if (not (common:in-running-test?))
                              (configf:write-alist *configdat* tmpfile))
			  (system (conc "ln -sf " tmpfile " " targfile))))
		    )))
	    (debug:print-info 1 *default-log-port* "No linktree yet, no caching configs.")))))

      *toppath*   ;; return toppath
      (let* ((use-cache (common:use-cache?)) ;; BB- use-cache checks *configdat* for use-cache setting.  We do not have *configdat*.  Bootstrapping problem here.
	     (toppath  (or *toppath* areapath (getenv "MT_RUN_AREA_HOME"))) ;; preserve toppath
	     (target   (common:args-get-target))
	     (sections (if target (list "default" target) #f)) ;; for runconfigs
	     (mtconfig (or (args:get-arg "-config") "megatest.config")) ;; allow overriding megatest.config 
             (cachefiles (launch:get-cache-file-paths areapath toppath target mtconfig))
	     ;; checking for null cachefiles should not be necessary, I was seeing error car of '(), might be a chicken bug or a red herring ...
	     (mtcachef   (if (null? cachefiles)
			     #f
	     (mtcachef   (car cachefiles)) ;; (and cachedir (conc cachedir "/" ".megatest.cfg-"  megatest-version "-" megatest-fossil-hash)))
	     (rccachef   (cdr cachefiles)) ;; (and cachedir (conc cachedir "/" ".runconfigs.cfg-"  megatest-version "-" megatest-fossil-hash)))
	     ) ;; (cancreate (and cachedir (common:file-exists? cachedir)(file-write-access? cachedir) (not (common:in-running-test?)))))
			     (car cachefiles))) ;; (and cachedir (conc cachedir "/" ".megatest.cfg-"  megatest-version "-" megatest-fossil-hash)))
	     (rccachef   (if (null? cachefiles)
			     #f
			     (cdr cachefiles)))) ;; (and cachedir (conc cachedir "/" ".runconfigs.cfg-"  megatest-version "-" megatest-fossil-hash)))
	      ;; (cancreate (and cachedir (common:file-exists? cachedir)(file-write-access? cachedir) (not (common:in-running-test?)))))
	(set! *toppath* toppath) ;; This is needed when we are running as a test using CMDINFO as a datasource
        ;;(BB> "launch:setup-body -- cachefiles="cachefiles)
	(cond
	 ;; if mtcachef exists just read it, however we need to assume toppath is available in $MT_RUN_AREA_HOME
	 ((and (not force-reread)
	       mtcachef  rccachef
	       use-cache
	 ((and (not force-reread) mtcachef (common:file-exists? mtcachef) (get-environment-variable "MT_RUN_AREA_HOME") use-cache)
	       (get-environment-variable "MT_RUN_AREA_HOME")
	       (common:file-exists? mtcachef)
	       (common:file-exists? rccachef))
          ;;(BB> "launch:setup-body -- cond branch 1 - use-cache")
          (set! *configdat*    (configf:read-alist mtcachef))
          ;;(BB> "launch:setup-body -- 1 set! *configdat*="*configdat*)
	  (set! *runconfigdat* (configf:read-alist rccachef))
	  (set! *configinfo*   (list *configdat*  (get-environment-variable "MT_RUN_AREA_HOME")))
	  (set! *configstatus* 'fulldata)
	  (set! *toppath*      (get-environment-variable "MT_RUN_AREA_HOME"))
	  *toppath*)
	 ;; there are no existing cached configs, do full reads of the configs and cache them
	 ;; we have all the info needed to fully process runconfigs and megatest.config
	 ((and ;; (not force-reread) ;; force-reread is irrelevant in the AND, could however OR it?
	       mtcachef
	 ((and (not force-reread) mtcachef) ;; BB- why are we doing this without asking if caching is desired?
	       rccachef) ;; BB- why are we doing this without asking if caching is desired?
          ;;(BB> "launch:setup-body -- cond branch 2")
	  (let* ((first-pass    (find-and-read-config        ;; NB// sets MT_RUN_AREA_HOME as side effect
				 mtconfig
				 environ-patt: "env-override"
				 given-toppath: toppath
				 pathenvvar: "MT_RUN_AREA_HOME"))
		 (first-rundat  (let ((toppath (if toppath 

						   key-vals)
					 (read-config (conc toppath "/runconfigs.config") *runconfigdat* #t ;; consider using runconfig:read some day ...
						      sections: sections)))
                         (cachefiles   (launch:get-cache-file-paths areapath toppath target mtconfig))
                         (mtcachef     (car cachefiles))
                         (rccachef     (cdr cachefiles)))
		    (if rccachef (configf:write-alist runconfigdat rccachef))
		    (set! *runconfigdat* runconfigdat)
		    (if mtcachef (configf:write-alist *configdat* mtcachef))
		    (set! *runconfigdat* runconfigdat)
		    (if (and rccachef mtcachef) (set! *configstatus* 'fulldata))))
		;; no configs found? should not happen but let's try to recover gracefully, return an empty hash-table
		(begin (set! *configdat* (make-hash-table))
                       ;;(BB> "launch:setup-body -- 3 set! *configdat*="*configdat*)
                       )
		(set! *configdat* (make-hash-table))
		)))

		)))
	 ;; else read what you can and set the flag accordingly
	 ;; here we don't have either mtconfig or rccachef
	 (else
          ;;(BB> "launch:setup-body -- cond branch 3 - else")
	  (let* ((cfgdat   (find-and-read-config 
			    (or (args:get-arg "-config") "megatest.config")
			    environ-patt: "env-override"
			    given-toppath: (get-environment-variable "MT_RUN_AREA_HOME")
			    pathenvvar: "MT_RUN_AREA_HOME")))

            (if (and cfgdat (list? cfgdat) (> (length cfgdat) 0) (hash-table? (car cfgdat)))
		(let* ((toppath  (or (get-environment-variable "MT_RUN_AREA_HOME")(cadr cfgdat)))
		       (rdat     (read-config (conc toppath  ;; convert this to use runconfig:read!
						    "/runconfigs.config") *runconfigdat* #t sections: sections)))
		  (set! *configinfo*   cfgdat)
		  (set! *configdat*    (car cfgdat))
		  (set! *runconfigdat* rdat)
		  (set! *toppath*      toppath)
		  (set! *configstatus* 'partial))
		(begin
		  (debug:print-error 0 *default-log-port* "No " mtconfig " file found. Giving up.")
		  (exit 2))))))
	;; COND ends here.
	
	;; additional house keeping
	(let* ((linktree (common:get-linktree)))
	  (if linktree
	      (begin
		(if (not (common:file-exists? linktree))
		    (begin
		      (handle-exceptions
			  exn
			  (begin
			    (debug:print-error 0 *default-log-port* "Something went wrong when trying to create linktree dir at " linktree)
			    (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
			    (exit 1))
			(create-directory linktree #t))))
		(handle-exceptions
		    exn
		    (begin
		      (debug:print-error 0 *default-log-port* "Something went wrong when trying to create link to linktree at " *toppath*)
		      (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn)))
		  (let ((tlink (conc *toppath* "/lt")))
		    (if (not (file-exists? tlink))
		    (if (not (common:file-exists? tlink))
			(create-symbolic-link linktree tlink)))))
	      (begin
		(debug:print-error 0 *default-log-port* "linktree not defined in [setup] section of megatest.config")
		)))
	(if (and *toppath*
		 (directory-exists? *toppath*))
	    (begin
	      (setenv "MT_RUN_AREA_HOME" *toppath*)
	      (setenv "MT_TESTSUITENAME" (common:get-testsuite-name)))
	    (begin
	      (debug:print-error 0 *default-log-port* "failed to find the top path to your Megatest area.")
	      ;;(exit 1)
	      (set! *toppath* #f) ;; force it to be false so we return #f
	      #f
	      ))
	      #f))
	
        ;; one more attempt to cache the configs for future reading
        (let* ((cachefiles   (launch:get-cache-file-paths areapath toppath target mtconfig))
               (mtcachef     (car cachefiles))
               (rccachef     (cdr cachefiles)))
          (if (and rccachef *runconfigdat*) (configf:write-alist *runconfigdat* rccachef))
          (if (and mtcachef *configdat*)    (configf:write-alist *configdat* mtcachef))
          (if (and rccachef *runconfigdat* (not (common:file-exists? rccachef))) (configf:write-alist *runconfigdat* rccachef))
          (if (and mtcachef *configdat*    (not (common:file-exists? mtcachef))) (configf:write-alist *configdat* mtcachef))
          (if (and rccachef mtcachef *runconfigdat* *configdat*)
              (set! *configstatus* 'fulldata)))

	;; if have -append-config then read and append here
	(let ((cfname (args:get-arg "-append-config")))
	  (if (and cfname
		   (file-read-access? cfname))

	(handle-exceptions
	 exn
	 (begin
	   (debug:print-error 0 *default-log-port* " Failed to remove symlink " lnkpath ((condition-property-accessor 'exn 'message) exn) ", exiting")
	   (exit 1))
	 (delete-file lnkpath)))

    (if (not (or (file-exists? lnkpath)
    (if (not (or (common:file-exists? lnkpath)
		 (symbolic-link? lnkpath)))
	(handle-exceptions
	 exn
	 (begin
	   (debug:print-error 0 *default-log-port* " Failed to create symlink " lnkpath ((condition-property-accessor 'exn 'message) exn) ", exiting")
	   (exit 1))
	 (create-symbolic-link toptest-path lnkpath)))

							     ;; (db:get-path dbstruct
				   ;; (rmt:sdb-qry 'getstr 
				   (db:test-get-rundir testinfo) ;; ) ;; )
				   #f)))
	  (hash-table-set! *toptest-paths* testname curr-test-path)
	  ;; NB// Was this for the test or for the parent in an iterated test?
	  (rmt:general-call 'test-set-rundir-shortdir run-id lnkpath 
			    (if (file-exists? lnkpath)
			    (if (common:file-exists? lnkpath)
				;; (resolve-pathname lnkpath)
				(common:nice-path lnkpath)
				lnkpath)
			    testname "" run-id)
	  ;; (rmt:general-call 'test-set-rundir run-id lnkpath testname "") ;; toptest-path)
	  (if (or (not curr-test-path)
		  (not (directory-exists? toptest-path)))

	  ;; If there is already a symlink delete it and recreate it.
	  (handle-exceptions
	   exn
	   (begin
	     (debug:print-error 0 *default-log-port* " Failed to re-create link " lnktarget ((condition-property-accessor 'exn 'message) exn) ", exiting")
	     (exit))
	   (if (symbolic-link? lnktarget)     (delete-file lnktarget))
	   (if (not (file-exists? lnktarget)) (create-symbolic-link test-path lnktarget)))))
	   (if (not (common:file-exists? lnktarget)) (create-symbolic-link test-path lnktarget)))))

    (if (not (directory? test-path))
	(create-directory test-path #t)) ;; this is a hack, I don't know why out of the blue this path does not exist sometimes

    (if (and test-src-path (directory? test-path))
	(begin
	  (let* ((ovrcmd (let ((cmd (config-lookup *configdat* "setup" "testcopycmd")))

;;    - could be netbatch
;;      (launch-test db (cadr status) test-conf))
(define (launch-test test-id run-id run-info keyvals runname test-conf test-name test-path itemdat params)
  (mutex-lock! *launch-setup-mutex*) ;; setting variables and processing the testconfig is NOT thread-safe, reuse the launch-setup mutex
  (let* ((item-path       (item-list->path itemdat))
	 (contour         #f)) ;; NOT READY FOR THIS (args:get-arg "-contour")))
    (let loop ((delta        (- (current-seconds) *last-launch*))
	       (launch-delay (string->number (or (configf:lookup *configdat* "setup" "launch-delay") "0"))))
	       (launch-delay (configf:lookup-number *configdat* "setup" "launch-delay" default: 1)))
      (if (> launch-delay delta)
	  (begin
	    (if (common:low-noise-print 1200 "test launch delay") ;; every two hours or so remind the user about launch delay.
	    (debug:print-info 0 *default-log-port* "Delaying launch of " test-name " for " (- launch-delay delta) " seconds")
		(debug:print-info 0 *default-log-port* "NOTE: test launches are delayed by " launch-delay " seconds. See megatest.config launch-delay setting to adjust.")) ;; launch of " test-name " for " (- launch-delay delta) " seconds"))
	    (thread-sleep! (- launch-delay delta))
	    (loop (- (current-seconds) *last-launch*) launch-delay))))
    (change-directory *toppath*)
    (alist->env-vars ;; consolidate this code with the code in megatest.scm for "-execute", *maybe* - the longer they are set the longer each launch takes (must be non-overlapping with the vars)
     (append
      (list
       (list "MT_RUN_AREA_HOME" *toppath*)

					(list 'set-vars  (if params (hash-table-ref/default params "-setvars" #f)))
					(list 'runname   runname)
					(list 'mt-bindir-path mt-bindir-path))))))))
      
      ;; clean out step records from previous run if they exist
      ;; (rmt:delete-test-step-records run-id test-id)
      ;; if the dir does not exist we may have a itempath where individual variables are a path, launch anyway
      (if (file-exists? work-area)
      (if (common:file-exists? work-area)
	  (change-directory work-area)) ;; so that log files from the launch process don't clutter the test dir
      (cond
       ;; ((and launcher hosts) ;; must be using ssh hostname
       ;;    (set! fullcmd (append launcher (car hosts)(list remote-megatest "-m" test-sig "-execute" cmdparms) debug-param)))
       ;; (set! fullcmd (append launcher (car hosts)(list remote-megatest test-sig "-execute" cmdparms))))
       (launcher
	(set! fullcmd (append launcher (list remote-megatest "-m" test-sig "-execute" cmdparms) debug-param)))

(define (lock-queue:delete-lock-db dbdat)
  (let ((fname (lock-queue:db-dat-get-path dbdat)))
    (system (conc "rm -f " fname "*"))))

(define (lock-queue:open-db fname #!key (count 10))
  (let* ((actualfname (conc fname ".lockdb"))
	 (dbexists (file-exists? actualfname))
	 (dbexists (common:file-exists? actualfname))
	 (db       (sqlite3:open-database actualfname))
	 (handler  (make-busy-timeout 136000)))
    (if dbexists
	(vector db actualfname)
	(begin
	  (handle-exceptions
	   exn

	     (lock-queue:release-lock fname test-id count: (- count 1)))
	   (let ((journal (conc fname "-journal")))
	     ;; If we've tried ten times and failed there is a serious problem
	     ;; try to remove the lock db and allow it to be recreated
	     (handle-exceptions
	      exn
	      #f
	      (if (file-exists? journal)(delete-file journal))
	      (if (file-exists? fname)  (delete-file fname))
	      (if (common:file-exists? journal)(delete-file journal))
	      (if (common:file-exists? fname)  (delete-file fname))
	      #f))))
     (sqlite3:execute (lock-queue:db-dat-get-db dbdat) "DELETE FROM runlocks WHERE test_id=?;" test-id)
     (sqlite3:finalize! (lock-queue:db-dat-get-db dbdat)))))

(define (lock-queue:steal-lock dbdat test-id #!key (count 10))
  (debug:print-info 0 *default-log-port* "Attempting to steal lock at " (lock-queue:db-dat-get-path dbdat))
  (tasks:wait-on-journal (lock-queue:db-dat-get-path dbdat) 1200 "lock-queue:steal-lock; waiting on journal")

;; Always use two or four digit decimal
;; 1.01, 1.02...1.10,1.11,1.1101 ... 1.99,2.00..

(declare (unit megatest-version))

(define megatest-version 1.6501)
(define megatest-version 1.6502)

## commented out due to a bug in v1.6501 in mtutil
# [fields]
# a text
# b text
# c text
## [fields]
## a text
## b text
## c text

[defaults]
usercode    .mtutil.scm
areafilter  area-to-run
targtrans   generic-target-translator
runtrans    generic-runname-translator

[setup]
pktsdirs /tmp/mt_pkts /some/other/source

[areas]
#         path-to-area   map-target-script(future, optional)
# someqa     path=../megatestqa/someqa; targtrans=somefunc; areafilter=area-to-run
fullrun   path=tests/fullrun; 
# targtrans is name of scheme proc stored in .mtutil.scm, which lives in PWD where mtutil is run
#           the target translator can return: a/target OR (list/of targets/to apply/run) OR #f i.e. run nothing
# ext-tests path=ext-tests; targtrans=prefix-contour;
ext-tests path=ext-tests
ext       path=ext-tests

[contours]
#     mode-patt/tag-expr
quick areas=ext;    selector=/QUICKPATT
quick areas=ext-tests;   selector=/QUICKPATT
# quick areas=fullrun,ext-tests;   selector=QUICKPATT/quick
quick2 areafn=check-area; selector=/QUICKPATT
# quick areas=fullrun,ext-tests; selector=QUICKPATT/quick
# full  areas=fullrun,ext-tests; selector=MAXPATT/
# short areas=fullrun,ext-tests; selector=MAXPATT/
# all   areas=fullrun,ext-tests
# snazy selector=QUICKPATT/

[nopurpose]

[access]
ext matt:admin mattw:owner

[accesstypes]
admin run rerun resume remove set-ss
owner run rerun resume remove
jerk  set-ss

[setup]
maxload 1.2

(include "common_records.scm")
(include "key_records.scm")
(include "db_records.scm")
(include "run_records.scm")
(include "megatest-fossil-hash.scm")

(let ((debugcontrolf (conc (get-environment-variable "HOME") "/.megatestrc")))
  (if (file-exists? debugcontrolf)
  (if (common:file-exists? debugcontrolf)
      (load debugcontrolf)))

;; Disabled help items
;;  -rollup                 : (currently disabled) fill run (set by :runname)  with latest test(s)
;;                            from prior runs with same keys
;;  -daemonize              : fork into background and disconnect from stdin/out

  -reqtarg key1/key2/...  : run for key1, key2, etc. but key1/key2 must be in runconfigs
  -testpatt patt1/patt2,patt3/...  : % is wildcard
  -runname                : required, name for this particular test run
  -state                  : Applies to runs, tests or steps depending on context
  -status                 : Applies to runs, tests or steps depending on context
  --modepatt key          : load testpatt from <key> in runconfigs instead of default TESTPATT if -testpatt and -tagexpr are not specified
  -tagexpr tag1,tag2%,..  : select tests with tags matching expression
  

Test helpers (for use inside tests)
  -step stepname
  -test-status            : set the state and status of a test (use :state and :status)
  -setlog logfname        : set the path/filename to the final log relative to the test
                            directory. may be used with -test-status
  -set-toplog logfname    : set the overall log for a suite of sub-tests

  -dumpmode MODE          : dump in MODE format instead of sexpr, MODE=json,ini,sexp etc. (add -debug 0,9 to see which file contributes each line)
  -show-cmdinfo           : dump the command info for a test (run in test environment)
  -section sectionName
  -var varName            : for config and runconfig lookup value for sectionName varName
  -since N                : get list of runs changed since time N (Unix seconds)
  -fields fieldspec       : fields to include in json dump; runs:id,runame+tests:testname+steps
  -sort fieldname         : in -list-runs sort tests by this field
  -testdata-csv [categorypatt/]varpatt  : dump testdata for given category

Misc 
  -start-dir path         : switch to this directory before running megatest
  -contour cname          : add a level of hierarcy to the linktree and run paths
  -rebuild-db             : bring the database schema up to date
  -cleanup-db             : remove any orphan records, vacuum the db
  -import-megatest.db     : push data from megatest.db to cache db files in /tmp/$USER

			":runname"
			"-runname"
			":state"  
			"-state"
			":status"
			"-status"
			"-list-runs"
                        "-testdata-csv"
			"-testpatt"
                        "--modepatt"
                        "-tagexpr"
			"-itempatt"
			"-setlog"
			"-set-toplog"
			"-runstep"

	       )
	      ))
    (debug:print-error 0 *default-log-port* "Unrecognised arguments: " (string-intersperse (if (list? remargs) remargs (argv))  " ")))

;; before doing anything else change to the start-dir if provided
;;
(if (args:get-arg "-start-dir")
    (if (file-exists? (args:get-arg "-start-dir"))
    (if (common:file-exists? (args:get-arg "-start-dir"))
        (let ((fullpath (common:real-path (args:get-arg "-start-dir"))))
          (setenv "PWD" fullpath)
          (change-directory fullpath))
	(begin
	  (debug:print-error 0 *default-log-port* "non-existant start dir " (args:get-arg "-start-dir") " specified, exiting.")
	  (exit 1))))

;; immediately set MT_TARGET if -reqtarg or -target are available
;;
(let ((targ (or (args:get-arg "-reqtarg")(args:get-arg "-target"))))
  (if targ (setenv "MT_TARGET" targ)))

;; The watchdog is to keep an eye on things like db sync etc.
;;

;; TODO: for multiple areas, we will have multiple watchdogs; and multiple threads to manage
(define *watchdog* (make-thread common:watchdog "Watchdog thread"))
(define *watchdog* (make-thread
		    (lambda ()
		      (handle-exceptions
			  exn
			  (begin
			    (print-call-chain)
			    (print " message: " ((condition-property-accessor 'exn 'message) exn)))
			(common:watchdog)))
		    "Watchdog thread"))

;;(if (not (args:get-arg "-server"))
;;    (thread-start! *watchdog*)) ;; if starting a server; wait till we get to running state before kicking off watchdog
(let* ((no-watchdog-args
       '("-list-runs"
         "-testdata-csv"
         "-list-servers"
         "-server"
         "-list-disks"
         "-list-targets"
         "-show-runconfig"
         ;;"-list-db-targets"
         "-show-runconfig"

(if (args:get-arg "-manual")
    (let* ((htmlviewercmd (or (configf:lookup *configdat* "setup" "htmlviewercmd")
			      (common:which '("firefox" "arora"))))
	   (install-home  (common:get-install-area))
	   (manual-html   (conc install-home "/share/docs/megatest_manual.html")))
      (if (and install-home
	       (file-exists? manual-html))
	       (common:file-exists? manual-html))
	  (system (conc "(" htmlviewercmd " " manual-html " ) &"))
	  (system (conc "(" htmlviewercmd " http://www.kiatoa.com/cgi-bin/fossils/megatest/doc/tip/docs/manual/megatest_manual.html ) &")))
      (exit)))

(if (args:get-arg "-version")
    (begin
      (print (common:version-signature)) ;; (print megatest-version)

      (set! *didsomething* #t)))

;; handle a clean-cache request as early as possible
;;
(if (args:get-arg "-clean-cache")
    (let ((toppath  (launch:setup)))
      (set! *didsomething* #t) ;; suppress the help output.
      (runs:clean-cache (or (getenv "MT_TARGET")
			    (args:get-arg "-target")
			    (args:get-arg "-remtarg"))
      (runs:clean-cache (getenv "MT_TARGET")(args:get-arg "-runname") toppath)))
			(args:get-arg "-runname")
			toppath)))
	  
(if (args:get-arg "-env2file")
    (begin
      (save-environment-as-files (args:get-arg "-env2file"))
      (set! *didsomething* #t)))

(if (args:get-arg "-list-disks")

				  (display "\n")
				  (loop (+ row 1) 0 '() (append result (list curr-row))))
				 (else
				  (loop row (+ col 1) (append curr-row (list val)) result)))))))))
		    (hash-table-keys results))))
		((sqlite3)
		 (let* ((db-file   (or out-file (pathname-file input-db)))
			(db-exists (file-exists? db-file))
			(db-exists (common:file-exists? db-file))
			(db        (sqlite3:open-database db-file)))
		   (if (not db-exists)(sqlite3:execute db "CREATE TABLE data (sheet,section,var,val);"))
		   (configf:map-all-hier-alist
		    data
		    (lambda (sheetname sectionname varname val)
		      (sqlite3:execute db
				       "INSERT OR REPLACE INTO data (sheet,section,var,val) VALUES (?,?,?,?);"

;;	*runconfigdat*)))

  (let* ((rundir (if (and (getenv "MT_LINKTREE")(getenv "MT_TARGET")(getenv "MT_RUNNAME"))
		     (conc (getenv "MT_LINKTREE") "/" (getenv "MT_TARGET") "/" (getenv "MT_RUNNAME"))
		     #f))
	 (cfgf   (if rundir (conc rundir "/.runconfig." megatest-version "-" megatest-fossil-hash) #f)))
    (if (and cfgf
	     (file-exists? cfgf)
	     (common:file-exists? cfgf)
	     (file-write-access? cfgf)
	     (common:use-cache?))
	(configf:read-alist cfgf)
	(let* ((keys   (rmt:get-keys))
	       (target (common:args-get-target))
	       (key-vals (if target (keys:target->keyval keys target) #f))
	       (sections (if target (list "default" target) #f))

		   (directory-exists? rundir)
		   (file-write-access? rundir))
	      (begin
                (if (not (common:in-running-test?))
                    (configf:write-alist data cfgf))
		;; force re-read of megatest.config - this resolves circular references between megatest.config
		(launch:setup force-reread: #t)
		;; (launch:cache-config) ;; there are two independent config cache locations, turning this one off for now. MRW.
		(launch:cache-config))) ;; we can safely cache megatest.config since we have a valid runconfig
		)) ;; we can safely cache megatest.config since we have a valid runconfig
	  data))))

(if (args:get-arg "-show-runconfig")
    (let ((tl (launch:setup)))
      (push-directory *toppath*)
      (let ((data (full-runconfigs-read)))
	;; keep this one local

(define (get-value-by-fieldname datavec test-field-index fieldname)
  (let ((indx (hash-table-ref/default test-field-index fieldname #f)))
    (if indx
	(if (>= indx (vector-length datavec))
	    #f ;; index too high, should raise an error I suppose
	    (vector-ref datavec indx))
	#f)))





(when (args:get-arg "-testdata-csv")
  (if (launch:setup)
      (let* ((keys        (rmt:get-keys)) ;; (db:get-keys dbstruct))
             (runpatt     (or (args:get-arg "-runname") "%"))
             (testpatt    (common:args-get-testpatt #f))
             (datapatt    (args:get-arg "-testdata-csv"))
             (match-data  (string-match "^([^/]+)/(.*)" (args:get-arg "-testdata-csv")))
             (categorypatt (if match-data (list-ref match-data 1) "%"))
             (setvarpatt  (if match-data
                              (list-ref match-data 2)
                              (args:get-arg "-testdata-csv")))
             (runsdat     (rmt:get-runs-by-patt keys (or runpatt "%") 
                                                (common:args-get-target) #f #f '("id" "runname" "state" "status" "owner" "event_time" "comment") 0))
             (header      (db:get-header runsdat))
             (access-mode (db:get-access-mode))
             (testpatt    (common:args-get-testpatt #f))
             (fields-spec (if (args:get-arg "-fields")
                              (extract-fields-constraints (args:get-arg "-fields"))
                              (list (cons "runs" (append keys (list "id" "runname" "state" "status" "owner" "event_time" "comment" "fail_count" "pass_count")))
                                    (cons "tests"  db:test-record-fields) ;; "id" "testname" "test_path")
                                    (list "steps" "id" "stepname"))))
             (tests-spec  (let ((t (alist-ref "tests" fields-spec equal?)))
                            (if (and t (null? t)) ;; all fields
                                db:test-record-fields
                                t)))
             (adj-tests-spec (delete-duplicates (if tests-spec (cons "id" tests-spec) db:test-record-fields))) 
             (test-field-index (make-hash-table))
             (runs (db:get-rows runsdat))
             )
        (if (and tests-spec (not (null? tests-spec))) ;; do some validation and processing of the test-spec
            (let ((invalid-tests-spec (filter (lambda (x)(not (member x db:test-record-fields))) tests-spec)))
              (if (null? invalid-tests-spec)
                  ;; generate the lookup map test-field-name => index-number
                  (let loop ((hed (car adj-tests-spec))
                             (tal (cdr adj-tests-spec))
                             (idx 0))
                    (hash-table-set! test-field-index hed idx)
                    (if (not (null? tal))(loop (car tal)(cdr tal)(+ idx 1))))
                  (begin
                    (debug:print-error 0 *default-log-port* "Invalid test fields specified: " (string-intersperse invalid-tests-spec ", "))
                    (exit)))))
        (let* ((table-header (string-split "target,run,test,itempath,category,var,value,comment" ","))
               (table-rows
                (apply append (map  
                               (lambda (run)
                                 (let* ((target (string-intersperse (map (lambda (x)
							 (db:get-value-by-header run header x))
						       keys) "/"))
                                        (statuses (string-split (or (args:get-arg "-status") "") ","))
                                        (run-id  (db:get-value-by-header run header "id"))
                                        (runname (db:get-value-by-header run header "runname")) 
                                        (states  (string-split (or (args:get-arg "-state") "") ","))
                                        (tests   (if tests-spec
                                                     (db:dispatch-query access-mode rmt:get-tests-for-run db:get-tests-for-run run-id testpatt states statuses #f #f #f 'testname 'asc ;; (db:get-tests-for-run dbstruct run-id testpatt '() '() #f #f #f 'testname 'asc 
                                                                        ;; use qryvals if test-spec provided
                                                                        (if tests-spec
                                                                            (string-intersperse adj-tests-spec ",")
                                                                            ;; db:test-record-fields
                                                                            #f)
                                                                        #f
                                                                        'normal)
                                                     '())))
                                   (apply append
                                          (map
                                           (lambda (test)
                                             (let* (
                                                    (test-id      (if (member "id"           tests-spec)(get-value-by-fieldname test test-field-index "id"          ) #f)) ;; (db:test-get-id         test))
                                                    (testname     (if (member "testname"     tests-spec)(get-value-by-fieldname test test-field-index "testname"    ) #f)) ;; (db:test-get-testname   test))
                                                    (itempath     (if (member "item_path"    tests-spec)(get-value-by-fieldname test test-field-index "item_path"   ) #f)) ;; (db:test-get-item-path  test))
                                                    (fullname     (conc testname
                                                                        (if (equal? itempath "")
                                                                            "" 
                                                                            (conc "/" itempath ))))
                                                    (testdat-raw (map vector->list (rmt:read-test-data* run-id test-id categorypatt setvarpatt)))
                                                    (testdat (filter
                                                              (lambda (x)
                                                                (not (equal? "logpro"
                                                                             (list-ref x 10))))
                                                              testdat-raw)))
                                               (map 
                                                (lambda (item)
                                                  (receive (id test_id category
                                                               variable value expected
                                                               tol units comment status type)
                                                      (apply values item)
                                                    (list target runname testname itempath category variable value comment)))
                                                testdat)))
                                           tests))))
                               runs))))
          (print (string-join table-header ","))
          (for-each (lambda(table-row)
                      (print (string-join (map ->string table-row) ",")))

                    
                            table-rows))))
  (set! *didsomething* #t)
  (set! *time-to-exit* #t))



;; NOTE: list-runs and list-db-targets operate on local db!!!
;;
;; IDEA: megatest list -runname blah% ...
;;
(if (or (args:get-arg "-list-runs")
	(args:get-arg "-list-db-targets"))

		       
		     (for-each 
		      (lambda (test)
		      	(common:debug-handle-exceptions #f
			 exn
			 (begin
			   (debug:print-error 0 *default-log-port* "Bad data in test record? " test)
			   (print "exn=" (condition->list exn))
			   (debug:print-error 5 *default-log-port* "exn=" (condition->list exn))
			   (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
			   (print-call-chain (current-error-port)))
			 (let* ((test-id      (if (member "id"           tests-spec)(get-value-by-fieldname test test-field-index "id"          ) #f)) ;; (db:test-get-id         test))
				(testname     (if (member "testname"     tests-spec)(get-value-by-fieldname test test-field-index "testname"    ) #f)) ;; (db:test-get-testname   test))
				(itempath     (if (member "item_path"    tests-spec)(get-value-by-fieldname test test-field-index "item_path"   ) #f)) ;; (db:test-get-item-path  test))
				(comment      (if (member "comment"      tests-spec)(get-value-by-fieldname test test-field-index "comment"     ) #f)) ;; (db:test-get-comment    test))
				(tstate       (if (member "state"        tests-spec)(get-value-by-fieldname test test-field-index "state"       ) #f)) ;; (db:test-get-state      test))

		(debug:print 0 *default-log-port* "Failed to setup, giving up on -test-paths or -test-files, exiting")
		(exit 1)))
	  (let* ((keys     (rmt:get-keys))
		 ;; db:test-get-paths must not be run remote
		 (paths    (tests:test-get-paths-matching keys target (args:get-arg "-test-files"))))
	    (set! *didsomething* #t)
	    (for-each (lambda (path)
			(if (file-exists? path)
			(if (common:file-exists? path)
			(print path)))	
		      paths)))
	;; else do a general-run-call
	(general-run-call 
	 "-test-files"
	 "Get paths to test"
	 (lambda (target runname keys keyvals)

(if (args:get-arg "-rebuild-db")
    (begin
      (if (not (launch:setup))
	  (begin
	    (debug:print 0 *default-log-port* "Failed to setup, exiting") 
	    (exit 1)))
      ;; keep this one local
      (open-run-close patch-db #f)
      ;; (open-run-close patch-db #f)
      (let ((dbstruct (db:setup #f areapath: *toppath*)))
        (common:cleanup-db dbstruct full: #t))
      (set! *didsomething* #t)))

(if (args:get-arg "-cleanup-db")
    (begin
      (if (not (launch:setup))
	  (begin
	    (debug:print 0 *default-log-port* "Failed to setup, exiting") 

			    (debug:print 0 *default-log-port* "Discarding test " testn "(" test-dat ") due to " failed-test)
			    res)
			  (cons testn res)))))))))

;;======================================================================
;;  T R I G G E R S
;;======================================================================

(define (mt:run-trigger cmd test-id test-rundir trigger logname test-name item-path event-time actual-state actual-status)
  ;; Putting the commandline into ( )'s means no control over the shell. 
  ;; stdout and stderr will be caught in the NBFAKE or mt_launch.log files
  ;; or equivalent. No need to do this. Just run it?
  (let* ((fullcmd (conc "nbfake "
			cmd           " "
			test-id       " "
			test-rundir   " "
			trigger       " "
			test-name     " "
			item-path     " " ;; has / prepended to deal with toplevel tests
			actual-state  " "
			actual-status " "
			event-time
			))
	 (prev-nbfake-log (get-environment-variable "NBFAKE_LOG")))
    (setenv "NBFAKE_LOG" (conc (cond
				((and (directory-exists? test-rundir)
				      (file-write-access? test-rundir))
				 test-rundir)
				((and (directory-exists? *toppath*)
				      (file-write-access? *toppath*))
				 *toppath*)
				(else (conc "/tmp/" (current-user-name))))
			       "/" logname))
    (debug:print-info 0 *default-log-port* "TRIGGERED on " trigger ", running command " fullcmd " output at " (get-environment-variable "NBFAKE_LOG"))
    ;; (call-with-environment-variables
    ;;  `(("NBFAKE_LOG" . ,(conc test-rundir "/" logname)))
    ;;  (lambda ()
    (process-run fullcmd)
    (if prev-nbfake-log
	(setenv "NBFAKE_LOG" prev-nbfake-log)
	(unsetenv "NBFAKE_LOG"))
    )) ;; ))

(define (mt:process-triggers dbstruct run-id test-id newstate newstatus)
  (if test-id 
  (let* ((test-dat      (db:get-test-info-by-id dbstruct run-id test-id)))
    (if test-dat
      (let* ((test-dat      (db:get-test-info-by-id dbstruct run-id test-id)))
	(if test-dat
	(let* ((test-rundir   ;; (rmt:sdb-qry 'getstr ;; (filedb:get-path *fdb*
		(db:test-get-rundir test-dat)) ;; ) ;; )
	       (test-name     (db:test-get-testname test-dat))
	       (tconfig       #f)
	       (state         (if newstate  newstate  (db:test-get-state  test-dat)))
	       (status        (if newstatus newstatus (db:test-get-status test-dat))))
	  (if (and test-name
		   test-rundir   ;; #f means no dir set yet
		   (file-exists? test-rundir)
		   (directory? test-rundir))
	      (call-with-environment-variables
	       (list (cons "MT_TEST_NAME" test-name)
		     (cons "MT_TEST_RUN_DIR" test-rundir)
		     (cons "MT_ITEMPATH"     (db:test-get-item-path test-dat)))
	       (lambda ()
		 (push-directory test-rundir)
		 (set! tconfig (mt:lazy-read-test-config test-name))
		 (for-each (lambda (trigger)
			     (let ((cmd  (configf:lookup tconfig "triggers" trigger))
				   (logf (conc  test-rundir "/last-trigger.log")))
			       (if cmd
	    (let* ((test-rundir   (db:test-get-rundir       test-dat)) ;; ) ;; )
		   (test-name     (db:test-get-testname     test-dat))
		   (item-path     (db:test-get-item-path    test-dat))
		   (duration      (db:test-get-run_duration test-dat))
		   (comment       (db:test-get-comment      test-dat))
		   (event-time    (db:test-get-event_time   test-dat))
		   (tconfig       #f)
		   (state         (if newstate  newstate  (db:test-get-state  test-dat)))
		   (status        (if newstatus newstatus (db:test-get-status test-dat))))
	      ;; (mutex-lock! *triggers-mutex*)
	      (if (and test-name
		       test-rundir)   ;; #f means no dir set yet
		       ;; (common:file-exists? test-rundir)
		       ;; (directory? test-rundir))
		  (call-with-environment-variables
		   (list (cons "MT_TEST_NAME"    (or test-name "no such test"))
			 (cons "MT_TEST_RUN_DIR" (or test-rundir "no test directory yet"))
			 (cons "MT_ITEMPATH"     (or item-path "")))
		   (lambda ()
		     (if (directory-exists? test-rundir)
			 (push-directory test-rundir)
			 (push-directory *toppath*))
		     (set! tconfig (mt:lazy-read-test-config test-name))
		     (for-each (lambda (trigger)
				 (let* ((munged-trigger (string-translate trigger "/ " "--"))
					(logname        (conc "last-trigger-" munged-trigger ".log")))
				   ;; first any triggers from the testconfig
				   (let ((cmd  (configf:lookup tconfig "triggers" trigger)))
				     (if cmd (mt:run-trigger cmd test-id test-rundir trigger (conc "tconfig-" logname) test-name item-path event-time state status)))
				   ;; next any triggers from megatest.config
				   (let ((cmd  (configf:lookup *configdat* "triggers" trigger)))
				     (if cmd (mt:run-trigger cmd test-id test-rundir trigger (conc "mtconfig-" logname) test-name item-path event-time state status)))))
				   ;; Putting the commandline into ( )'s means no control over the shell. 
				   ;; stdout and stderr will be caught in the NBFAKE or mt_launch.log files
				   ;; or equivalent. No need to do this. Just run it?
				   (let ((fullcmd (conc cmd " " test-id " " test-rundir " " trigger "&")))
				     (debug:print-info 0 *default-log-port* "TRIGGERED on " trigger ", running command " fullcmd)
				     (process-run fullcmd)))))
			   (list
			    (conc state "/" status)
			    (conc state "/")
			    (conc "/" status)))
		 (pop-directory))
	       ))))))
			       (list
				(conc state "/" status)
				(conc state "/")
				(conc "/" status)))
		     (pop-directory))
		   ))
	      ;; (mutex-unlock! *triggers-mutex*)
	      )))))

;;======================================================================
;;  S T A T E   A N D   S T A T U S   F O R   T E S T S 
;;======================================================================

;; speed up for common cases with a little logic
(define (mt:test-set-state-status-by-id run-id test-id newstate newstatus newcomment)

    (if tconf
	tconf
	(let ((test-dirs (tests:get-tests-search-path *configdat*)))
	  (let loop ((hed (car test-dirs))
		     (tal (cdr test-dirs)))
	    ;; Setting MT_LINKTREE here is almost certainly unnecessary. 
	    (let ((tconfig-file (conc hed "/" test-name "/testconfig")))
	      (if (and (file-exists? tconfig-file)
	      (if (and (common:file-exists? tconfig-file)
		       (file-read-access? tconfig-file))
		  (let ((link-tree-path (common:get-linktree)) ;; (configf:lookup *configdat* "setup" "linktree"))
			(old-link-tree  (get-environment-variable "MT_LINKTREE")))
		    (if link-tree-path (setenv "MT_LINKTREE" link-tree-path))
		    (let ((newtcfg (read-config tconfig-file #f #f))) ;; NOTE: Does NOT run [system ...]
		      (hash-table-set! *testconfigs* test-name newtcfg)
		      (if old-link-tree 

;; (include "megatest-version.scm")

;; fake out readline usage of toplevel-command
(define (toplevel-command . a) #f)

(use srfi-1 posix srfi-69 readline ;;  regex regex-case srfi-69 apropos json http-client directory-utils rpc typed-records;; (srfi 18) extras)
     srfi-18 extras format pkts regex regex-case
     (prefix dbi dbi:)) ;;  zmq extras)
     (prefix dbi dbi:)
     (prefix nanomsg nmsg:))

(declare (uses common))
(declare (uses megatest-version))
(declare (uses margs))
(declare (uses configf))
;; (declare (uses rmt))

;;   
(define (map-targets mtconf aval-alist runkey area contour #!key (xlatr-key-in #f))
  (let* ((xlatr-key (or xlatr-key-in
                        (conf-get/default mtconf aval-alist 'targtrans)))
         (proc      (hash-table-ref/default *target-mappers* xlatr-key #f)))
    (if proc
        (begin
          (print "Using target mapper: " area-xlatr)
          (print "Using target mapper: " xlatr-key)
          (handle-exceptions
           exn
           (begin
             (print "FAILED TO RUN TARGET MAPPER FOR " area ", called " area-xlatr)
             (print "FAILED TO RUN TARGET MAPPER FOR " area ", called " xlatr-key)
             (print "   function is: " (hash-table-ref/default *target-mappers* xlatr-key #f ) )
             (print " message: " ((condition-property-accessor 'exn 'message) exn))
             runkey)
           (proc runkey area contour)))
        (begin
          (if xlatr-key 
              (print "ERROR: Failed to find named target translator " xlatr-key ", using original target."))

;;
;;   i. Check that owner of the file and calling user are same?
;;  ii. Check that we are in a legal megatest area?
;; iii. Have some form of authentication or record of the md5sum or similar of the file?
;;  iv. Use compiled version in preference to .scm version. Thus there is a manual "blessing"
;;      required to use .mtutil.scm.
;;
(if (file-exists? "megatest.config")
    (if (file-exists? ".mtutil.so")
(if (common:file-exists? "megatest.config")
    (if (common:file-exists? ".mtutil.so")
	(load ".mtutil.so")
	(if (file-exists? ".mtutil.scm")
	(if (common:file-exists? ".mtutil.scm")
            (load ".mtutil.scm"))))

;; Disabled help items
;;  -rollup                 : (currently disabled) fill run (set by :runname)  with latest test(s)
;;                            from prior runs with same keys
;; Contour actions
;;    import                  : import pkts

Called as " (string-intersperse (argv) " ") "
Version " megatest-version ", built from " megatest-fossil-hash ))

;; args and pkt key specs
;;
(define *arg-keys*
  ;; used keys
  ;;    a  - action
  '(
    ("-area"            . G) ;; maps to group
    ("-contour"         . c)
    ("-append-config"   . d)
    ("-state"           . e)
    ("-item-patt"       . i)
    ("-sync-to"         . k)
    ("-new"             . l) ;; l (see below) is new-ss
    ("-run-name"        . n)
    ("-mode-patt"       . o)
    ("-test-patt"       . p)  ;; idea, enhance margs ("-test-patt" "-testpatt") => yields one value in "-test-patt"
    ("-status"          . s)
    ("-target"          . t)
    ("-tag-expr"        . x)
    ;; misc

    ))

;; alist to map actions to old megatest commands
(define *action-keys*
  '((run         . "-run")
    (sync        . "")
    (archive     . "-archive")
    (set-ss      . "-set-state-status")))
    (set-ss      . "-set-state-status")
    (remove      . "-remove-runs")))

;; Card types:
;;
;; A action
;; U username (Unix)
;; D timestamp
;; T card type

    (D  . timestamp )
    (T  . cardtype  )
    (U  . user      ) ;; username
    (Z  . shar1sum  )

    ;; Extras
    (a  . runkey    ) ;; needed for matching up pkts with target derived from runkey
    ;; (l  . new-ss    ) ;; new state/status
    ))

;; inlst is an alternative input
;;
(define (lookup-param-by-key key #!key (inlst #f))
  (fold (lambda (a res)
	  (if (eq? (cdr a) key)

;;
(define (param-translate param)
  (or (alist-ref (string->symbol param)
		 '((-tag-expr  . "-tagexpr")
		   (-mode-patt . "--modepatt")
		   (-run-name  . "-runname")
		   (-test-patt . "-testpatt")
		   (-msg       . "-m")))
		   (-msg       . "-m")
		   (-new       . "-set-state-status")))
      param))

(define (val->alist val)
  (let ((val-list (string-split-fields ";\\s*" val #:infix)))
    (if val-list
	(map (lambda (x)
	       (let ((f (string-split-fields "\\s*=\\s*" x #:infix)))

    (handle-exceptions
	exn
	(print "ERROR: failed to create directory " dest-dir " message: " ((condition-property-accessor 'exn 'message) exn))
      (create-directory dest-dir #t))
    (handle-exceptions
	exn
	(print "ERROR: failed to clone or sync 1ossil " url " message: " ((condition-property-accessor 'exn 'message) exn))
      (if (file-exists? targ-file)
      (if (common:file-exists? targ-file)
	  (system (conc "fossil pull --once " url " -R " targ-file))
	  (system (conc "fossil clone " url " " targ-file))
	  ))))

(define (fossil:last-change-node-and-time fossils-dir fossil-name branch)
  (let* ((fossil-file   (conc fossils-dir "/" fossil-name))
	 (timeline-port (if (file-read-access? fossil-file)

(if (or (args:any? "-h" "help" "-help" "--help")
	(member *action* '("-h" "-help" "--help" "help")))
    (begin
      (print help)
      (exit 1)))

;;======================================================================
;; pkts
;;======================================================================

(define (with-queue-db mtconf proc)
  (let* ((pktsdirs (configf:lookup mtconf "setup"  "pktsdirs"))
	 (pktsdir  (if pktsdirs (car (string-split pktsdirs " ")) #f))
	 (toppath  (configf:lookup mtconf "dyndat" "toppath"))
	 (pdbpath  (or (configf:lookup mtconf "setup"  "pdbpath") pktsdir)))
    (if (not (and  pktsdir toppath pdbpath))
	(begin
	  (print "ERROR: settings are missing in your megatest.config for area management.")
	  (print "  you need to have pktsdir in the [setup] section."))
	(let* ((pdb  (open-queue-db pdbpath "pkts.db"
				    schema: '("CREATE TABLE groups (id INTEGER PRIMARY KEY,groupname TEXT, CONSTRAINT group_constraint UNIQUE (groupname));")))
	       (res  (proc pktsdirs pktsdir pdb)))
	  (dbi:close pdb)
	  res
	  ))))

(define (load-pkts-to-db mtconf)
  (with-queue-db
   mtconf
   (lambda (pktsdirs pktsdir pdb)
     (for-each
      (lambda (pktsdir) ;; look at all
	(if (and (file-exists? pktsdir)
		 (directory? pktsdir)
		 (file-read-access? pktsdir))
	    (let ((pkts (glob (conc pktsdir "/*.pkt"))))
	      (for-each
	       (lambda (pkt)
		 (let* ((uuid    (cadr (string-match ".*/([0-9a-f]+).pkt" pkt)))
			(exists  (lookup-by-uuid pdb uuid #f)))
		   (if (not exists)
		       (let* ((pktdat (string-intersperse
				       (with-input-from-file pkt read-lines)
				       "\n"))
			      (apkt   (pkt->alist pktdat))
			      (ptype  (alist-ref 'T apkt))
			      (parent (alist-ref 'P apkt)))
			 (add-to-queue pdb pktdat uuid (or ptype 'cmd) parent 0)
			 (debug:print 4 *default-log-port* "Added " uuid " of type " ptype " to queue"))
		       (debug:print 4 *default-log-port* "pkt: " uuid " exists, skipping...")
		       )))
	       pkts))))
      (string-split pktsdirs)))))

(define (get-pkt-alists pkts)
  (map (lambda (x)
	 (alist-ref 'apkt x)) ;; 'pkta pulls out the alist from the read pkt
       pkts))

;; given list of pkts (alist mode) return list of D cards as Unix epoch, sorted descending
;; also delete duplicates by target i.e. (car pkt)
(define (get-pkt-times pkts)
  (delete-duplicates
   (sort 
    (map (lambda (x)
	   `(,(alist-ref 't x) . ,(string->number (alist-ref 'D x))))
	 pkts)
    (lambda (a b)(> (cdr a)(cdr b))))      ;; sort descending
   (lambda (a b)(equal? (car a)(car b))))) ;; remove duplicates by target

;;======================================================================
;; Runs
;;======================================================================

;; make a runname
;;
(define (make-runname pre post)
 (time->string
  (seconds->local-time (current-seconds)) "%Yw%V.%w-%H%M"))

;; collect, translate, collate and assemble a pkt from the command-line
;;
;; sched => force the run start time to be recorded as sched Unix
;; epoch. This aligns times properly for triggers in some cases.
;;
;;  extra-dat format is ( 'x xval 'y yval .... )
;;
(define (command-line->pkt action args-alist sched-in #!key (extra-dat '()))
(define (command-line->pkt action args-alist sched-in #!key (extra-dat '())(area-path #f)(new-ss #f))
  (let* ((sched     (cond
		     ((vector? sched-in)(local-time->seconds sched-in)) ;; we recieved a time
		     ((number? sched-in) sched-in)
		     (else     (current-seconds))))
	 (args-data (if args-alist
			(if (hash-table? args-alist) ;; seriously?
			    (hash-table->alist args-alist)
			    args-alist)
			(hash-table->alist args:arg-hash))) ;; if no args-alist then we assume this is a call driven directly by commandline
	 (alldat    (apply append (list 'T "cmd"
					'A action
					'U (current-user-name)
					'D sched)
                           extra-dat
	 (alldat    (apply append
			   (list 'A action
				 'U (current-user-name)
				 'D sched)
			   (if area-path
			       (list 'S area-path) ;; the area-path is mapped to the start-dir
			       '())
                           (if (list? extra-dat)
			       extra-dat
			       (begin
				 (debug:print 0 *default-log-port* "ERROR: command-line->pkt received bad extra-dat " extra-dat)
				 '()))
			   (map (lambda (x)
				  (let* ((param (car x))
					 (value (cdr x))
					 (pmeta (assoc param *arg-keys*))    ;; translate the card key to a megatest switch or parameter
					 (smeta (assoc param *switch-keys*)) ;; first lookup the key in arg-keys or switch-keys
					 (meta  (if (or pmeta smeta)
						    (cdr (or pmeta smeta))   ;; found it?

	 (mtconfdat (find-and-read-config        ;; NB// sets MT_RUN_AREA_HOME as side effect
		     mtconfig
		     ;; environ-patt: "env-override"
		     given-toppath: start-dir
		     ;; pathenvvar: "MT_RUN_AREA_HOME"
		     ))
	 (mtconf    (if mtconfdat (car mtconfdat) #f)))
    ;; we set some dynamic data in a section called "dyndata"
    ;; we set some dynamic data in a section called "scratchdata"
    (if mtconf
	(begin
	  (configf:section-var-set! mtconf "dyndat" "toppath" start-dir)))
    ;; (print "TOPPATH: " (configf:lookup mtconf "dyndat" "toppath"))
	  (configf:section-var-set! mtconf "scratchdat" "toppath" start-dir)))
    ;; (print "TOPPATH: " (configf:lookup mtconf "scratchdat" "toppath"))
    mtconfdat))


;; NEED TIMESTAMP ON PKTS for efficient loading of packets into db.


;; make a run request pkt from basic data, this seriously needs to be refactored

(define (val-alist->areas val-alist)
  (let ((areas-string   (alist-ref 'areas  val-alist))
	(areas-procname (alist-ref 'areafn val-alist)))
    (if areas-procname ;; areas-procname take precedence
	areas-procname
	(string-split (or areas-string "") ","))))

;; area   - the current area under consideration
;; areas  - the list of allowed areas from the contour spec -OR-
;;          if it is a string then it is the function to use to
;;          lookup in *area-checkers*
;;
(define (area-allowed? area areas runkey contour)
(define (area-allowed? area areas runkey contour mode-patt)
  (cond
   ((not areas) #t) ;; no spec
   ((string? areas) ;; 
    (let ((check-fn (hash-table-ref/default *area-checkers* areas #f)))
    (let ((check-fn (hash-table-ref/default *area-checkers* (string->symbol areas) #f)))
      (if check-fn
	  (check-fn area runkey contour)
	  (check-fn area runkey contour mode-patt)
	  #f)))
   ((list? areas)(member area areas))
   (else #f))) ;; shouldn't get here 

;; (use trace)(trace create-run-pkt)

;; collect all needed data and create run pkts for contours with changed inputs
;;
(define (generate-run-pkts mtconf toppath)
  (let ((std-runname (conc "sched"  (time->string (seconds->local-time (current-seconds)) "%M%H%d"))))
    (with-queue-db
    (common:with-queue-db
     mtconf
     (lambda (pktsdirs pktsdir pdb)
       (let* ((rgconfdat (find-and-read-config (conc toppath "/runconfigs.config")))
	      (rgconf    (car rgconfdat))
	      (all-areas (map car (configf:get-section mtconf "areas")))
	      (contours  (configf:get-section mtconf "contours"))
	      (torun     (make-hash-table)) ;; target => ( ... info ... )

			(areas      (val-alist->areas    val-alist)) ;; areas can be a single string (a reference to call an areas function), or a list of area names.
			(dbdest     (alist-ref 'dbdest   val-alist))
			(appendconf (alist-ref 'appendconf val-alist))
			(file-globs (alist-ref 'glob val-alist))
			
			(runstarts  (find-pkts pdb '(runstart) `((o . ,contour)
								 (t . ,runkey))))
			(rspkts     (get-pkt-alists runstarts))
			(rspkts     (common:get-pkt-alists runstarts))
			;; starttimes is for run start times and is used to know when the last run was launched
			(starttimes (get-pkt-times rspkts)) ;; sort by age (youngest first) and delete duplicates by target
			(starttimes (common:get-pkt-times rspkts)) ;; sort by age (youngest first) and delete duplicates by target
			(last-run   (if (null? starttimes) ;; if '() then it has never been run, else get the max
					0
					(apply max (map cdr starttimes))))
			;; synctimes is for figuring out the last time a sync was done
			(syncstarts (find-pkts pdb '(syncstart) '())) ;; no qualifiers, a sync does all tarets etc.
			(sspkts     (get-pkt-alists syncstarts))
			(synctimes  (get-pkt-times  sspkts))
			(sspkts       (common:get-pkt-alists syncstarts))
			(synctimes    (common:get-pkt-times  sspkts))
			(last-sync  (if (null? synctimes) ;; if '() then it has never been run, else get the max
					0
					(apply max (map cdr synctimes))))
			)

		   (let ((delta (lambda (x)
				  (round (/ (- (current-seconds) x) 60)))))

		   (areas      (val-alist->areas cval-alist))
		   (selector   (alist-ref 'selector cval-alist))
		   (mode-tag   (and selector (string-split-fields "/" selector #:infix)))
		   (mode-patt  (and mode-tag (if (eq? (length mode-tag) 2)(cadr mode-tag) #f)))
		   (tag-expr   (and mode-tag (if (null? mode-tag) #f (car mode-tag)))))
	      (print "contour: " contour " areas=" areas " cval=" cval)
	      (for-each
	       (lambda (runkeydatset)
	       (lambda (runkeydatset) 
		 ;; (print "runkeydatset: ")(pp runkeydatset)
		 (let ((runkey     (car runkeydatset))
		       (runkeydats (cadr runkeydatset)))
		   (for-each
		    (lambda (runkeydat)
		      (for-each
		       (lambda (area)
			 (if (area-allowed? area areas runkey contour) ;; is this area to be handled (from areas=a,b,c OR using areafn=abcfn and *area-checks* ...)
			 (if (area-allowed? area areas runkey contour mode-patt) ;; is this area to be handled (from areas=a,b,c OR using areafn=abcfn and *area-checks* ...)
                             (let* ((aval       (or (configf:lookup mtconf "areas" area) ""))
                                    (aval-alist (val->alist aval))
                                    (runname    (alist-ref 'runname runkeydat))
                                    (runtrans   (alist-ref 'runtrans runkeydat))
                                    
                                    (reason     (alist-ref 'message runkeydat))
                                    (sched      (alist-ref 'sched   runkeydat))

		    runkeydats)))
	       (let ((res (configf:get-section torun contour))) ;; each contour / target
		 ;; (print "res=" res)
		 res))))
	  (hash-table-keys torun)))))))

(define (pkt->cmdline pkta)
  (let ((action (or (lookup-action-by-key (alist-ref 'A pkta)) "noaction")))
  (let* ((action (or (lookup-action-by-key (alist-ref 'A pkta)) "noaction"))
	 (action-param (case (string->symbol action)
			 ((-set-state-status) (conc (alist-ref 'l pkta) " "))
			 (else ""))))
    (fold (lambda (a res)
	    (let* ((key (car a)) ;; get the key name
		   (val (cdr a))
		   (par (or (lookup-param-by-key key)  ;; need to check also if it is a switch
			    (lookup-param-by-key key inlst: *switch-keys*))))
	      ;; (print "key: " key " val: " val " par: " par)
	      (if par
		  (conc res " " (param-translate par) " " val)
		  (if (alist-ref key *additional-cards*) ;; these cards do not translate to parameters or switches
		      res
		      (begin
			(print "ERROR: Unknown key in packet \"" key "\" with value \"" val "\"")
			res)))))
	  (conc "megatest " (if (not (member action '("sync")))
				(conc action " ")
				(conc action " " action-param)
				""))
	  pkta)))

;; (use trace)(trace pkt->cmdline)

(define (write-pkt pktsdir uuid pkt)
  (if pktsdir

		 (handle-exceptions
		     exn
		     #f
		   (create-directory "logs")
		   #t)
		 #t)
	     "logs"
	     "/tmp")))
    (with-queue-db
	     "/tmp"))
	(cpuload (alist-ref 'adj-proc-load (common:get-normalized-cpu-load #f)))
	(maxload (string->number (or (configf:lookup mtconf "setup" "maxload")
				     (configf:lookup mtconf "jobtools" "maxload") ;; respect value used by Megatest calls
				     "1.1"))))
    (common:with-queue-db
     mtconf
     (lambda (pktsdirs pktsdir pdb)
       (let* ((rgconfdat (find-and-read-config (conc toppath "/runconfigs.config")))
	      (rgconf    (car rgconfdat))
	      (areas     (configf:get-section mtconf "areas"))
	      (contours  (configf:get-section mtconf "contours"))
	      (pkts      (find-pkts pdb '(cmd) '()))
	      (torun     (make-hash-table)) ;; target => ( ... info ... )
	      (rgentargs (hash-table-keys rgconf))) ;; these are the targets registered for automatically triggering
	 (for-each
	  (lambda (pktdat)
	    (let* ((pkta    (alist-ref 'apkt pktdat))
		   (action  (alist-ref 'A pkta))
		   (cmdline (pkt->cmdline pkta))
		   (uuid    (alist-ref 'Z pkta))
		   (user    (alist-ref 'U pkta))
		   (area    (alist-ref 'G pkta))
		   (logf    (conc logdir "/" uuid "-run.log"))
		   (fullcmd (conc "NBFAKE_LOG=" logf " nbfake " cmdline)))
	      (if (check-access user mtconf action area)
		  (if (and (> cpuload maxload)
			   (member action '("run" "archive"))) ;; do not run archive or run if load is over the specified limit
		      (print "WARNING: cpuload too high, skipping processing of " uuid)
		      (begin
	      (print "RUNNING: " fullcmd)
	      (system fullcmd)
	      (mark-processed pdb (list (alist-ref 'id pktdat)))
	      (let-values (((ack-uuid ack-pkt)
			    (add-z-card
			     (construct-sdat 'P uuid
					     'T (case (string->symbol action)
						  ((run) "runstart")
						  ((sync) "syncstart")    ;; example of translating run -> runstart
						  (else   action))
					     'c (alist-ref 'o pkta) ;; THIS IS WRONG! SHOULD BE 'c
					     't (alist-ref 't pkta)))))
		(write-pkt pktsdir ack-uuid ack-pkt))))
			(print "RUNNING: " fullcmd)
			(system fullcmd)
			(mark-processed pdb (list (alist-ref 'id pktdat)))
			(let-values (((ack-uuid ack-pkt)
				      (add-z-card
				       (construct-sdat 'P uuid
						       'T (case (string->symbol action)
							    ((run) "runstart")
							    ((sync) "syncstart")    ;; example of translating run -> runstart
							    (else   action))
						       'c (alist-ref 'o pkta) ;; THIS IS WRONG! SHOULD BE 'c
						       't (alist-ref 't pkta)))))
			  (write-pkt pktsdir ack-uuid ack-pkt))))
		  (begin ;; access denied! Mark as such
		    (mark-processed pdb (list (alist-ref 'id pktdat)))
		    (let-values (((ack-uuid ack-pkt)
				  (add-z-card
				   (construct-sdat 'P uuid
						   'T "access-denied"
						   'c (alist-ref 'o pkta) ;; THIS IS WRONG! SHOULD BE 'c
						   't (alist-ref 't pkta)))))
		      (write-pkt pktsdir ack-uuid ack-pkt))))))
	  pkts))))))
  

(define (check-access user mtconf action area)
  ;; NOTE: Need control over defaults. E.g. default might be no access
  (let* ((access-ctrl (hash-table-exists? mtconf "access"))  ;; if there is an access section the default is to REQUIRE enablement/access
	 (access-list (map (lambda (x)
			     (string-split x ":"))
			   (string-split (or (configf:lookup mtconf "access" area) ;; userid:rightstype userid2:rightstype2 ...
					     (if access-ctrl
						 "*:none"  ;; nobody has access by default
						 "*:all")))))
	 (access-types-dat (configf:get-section mtconf "accesstypes")))
    (debug:print 0 *default-log-port* "Checking access in " access-list " with access-ctrl " access-ctrl " for area " area)
    (if access-ctrl
	(let* ((user-access     (or (assoc user access-list)
				    (assoc "*"  access-list)))
	       (access-type     (cadr user-access))
	       (access-types    (let ((res (alist-ref access-type access-types-dat equal?)))
				  (if res (car res) res)))
	       (allowed-actions (string-split (or access-types ""))))
	  (print "Got " allowed-actions " for user " user " where access-types=" access-types " access-type=" access-type)
	  (cond
	   ((and access-types (member action allowed-actions))
	    ;; (print "Access granted for " user " for " action)
	    #t)
	   (else
	    ;; (print "Access denied for " user " for " action)
	    #f))))))

(define (get-pkts-dir mtconf)
  (let ((pktsdirs  (configf:lookup mtconf "setup" "pktsdirs"))
	(pktsdir   (if pktsdirs (car (string-split pktsdirs " ")) #f)))
    pktsdir))

(let ((debugcontrolf (conc (get-environment-variable "HOME") "/.mtutilrc")))
  (if (file-exists? debugcontrolf)
  (if (common:file-exists? debugcontrolf)
      (load debugcontrolf)))

(if *action*
    (case (string->symbol *action*)
      ((run remove rerun set-ss archive kill list)
       (let* ((mtconfdat (simple-setup (args:get-arg "-start-dir")))
	      (mtconf    (car mtconfdat))
	      (area      (args:get-arg "-area")) ;; look up the area to dispatch to from [areas] section
	      (areasec   (if area (configf:lookup mtconf "areas" area) #f))
	      (areadat   (if areasec (val->alist areasec) #f))
	      (area-path (if areadat (alist-ref 'path areadat) #f))
	      (pktsdirs  (configf:lookup mtconf "setup" "pktsdirs"))
	      (pktsdir   (if pktsdirs (car (string-split pktsdirs " ")) #f))
	      (adjargs   (hash-table-copy args:arg-hash)))
	      (adjargs   (hash-table-copy args:arg-hash))
	      (new-ss    (args:get-arg "-new")))
	 ;; check a few things
	 (cond
	  ((and area (not area-path))
	   (print "ERROR: the specified area was not found in the [areas] table. Area name=" area)
	   (exit 1))
	  ((not area)
	   (print "ERROR: no area specified. Use -area <areaname>")
	   (exit 1))
	  (else
	   (let ((user (current-user-name)))
	     (if (check-access user mtconf *action* area);; check rights
		 (print "Access granted for " *action* " action by " user)
		 (begin
		   (print "Access denied for " *action* " action by " user)
		   (exit 1))))))
	 
	 ;; (for-each
	 ;;  (lambda (key)
	 ;;    (if (not (member key *legal-params*))
	 ;; 	(hash-table-delete! adjargs key))) ;; we need to delete any params intended for mtutil
	 ;;  (hash-table-keys adjargs))
	 (let-values (((uuid pkt)
		       (command-line->pkt *action* adjargs #f)))
		       (command-line->pkt *action* adjargs #f area-path: area-path new-ss: new-ss)))
	   (write-pkt pktsdir uuid pkt))))
      ((dispatch import rungen process)
       (let* ((mtconfdat (simple-setup (args:get-arg "-start-dir")))
	      (mtconf    (car mtconfdat))
	      (toppath   (configf:lookup mtconf "dyndat" "toppath")))
	      (toppath   (configf:lookup mtconf "scratchdat" "toppath")))
	 (case (string->symbol *action*)
	   ((process)  (begin
			 (load-pkts-to-db mtconf)
			 (common:load-pkts-to-db mtconf)
			 (generate-run-pkts mtconf toppath)
			 (load-pkts-to-db mtconf)
			 (common:load-pkts-to-db mtconf)
			 (dispatch-commands mtconf toppath)))
	   ((import)   (load-pkts-to-db mtconf)) ;; import pkts
	   ((import)   (common:load-pkts-to-db mtconf)) ;; import pkts
	   ((rungen)   (generate-run-pkts mtconf toppath))
	   ((dispatch) (dispatch-commands mtconf toppath)))))
      ;; misc
      ((show)
       (if (> (length remargs) 0)
	   (let* ((mtconfdat (simple-setup (args:get-arg "-start-dir")))
		  (mtconf    (car mtconfdat))
		  (sect-dat (configf:get-section mtconf (car remargs))))
	     (if sect-dat
		 (for-each
		  (lambda (entry)
		    (if (> (length entry) 1)
			(print (car entry) "   " (cadr entry))
			(print (car entry))))
		  sect-dat)
		 (print "No section \"" (car remargs) "\" found")))
	   (print "ERROR: list requires section parameter; areas, setup or contours")))
      ((gendot)
       (let* ((mtconfdat (simple-setup (args:get-arg "-start-dir")))
	      (mtconf    (car mtconfdat)))
	 (with-queue-db
	 (common:with-queue-db
	  mtconf
	  (lambda (pktsdirs pktsdir conn)
	    ;;                       pktspec display-fields 
	    (make-report "out.dot" conn '())))))
	    (make-report "out.dot" conn
			 '((cmd      . ((parent . P)
					(user   . M)
					(target . t)))
			   (runstart . ((parent . P)
					(target . t)))
			   (runtype . ((parent . P)))) ;; pktspec
			 '(P U t)                                                     ;; 
			 )))))  ;; no ptypes listed (ptypes are strings of pkt types to read from db
      ((db)
       (if (null? remargs)
	   (print "ERROR: missing sub command for db command")
	   (let ((subcmd (car remargs)))
	     (case (string->symbol subcmd)
	       ((pgschema)
		(let* ((install-home (common:get-install-area))
		       (schema-file  (conc install-home "/share/db/mt-pg.sql")))
		  (if (file-exists? schema-file)
		  (if (common:file-exists? schema-file)
		      (system (conc "/bin/cat " schema-file)))))
	       ((sqlite3schema)
		(let* ((install-home (common:get-install-area))
		       (schema-file  (conc install-home "/share/db/mt-sqlite3.sql")))
		  (if (file-exists? schema-file)
		  (if (common:file-exists? schema-file)
		      (system (conc "/bin/cat " schema-file)))))
	       ((junk)
		(rmt:get-keys))))))))

;; If HTTP_HOST is defined then we must be in the cgi environment
;; so run stml and exit
;;

(if (args:get-arg "-h")
    (begin
      (print help)
      (exit)))

;; ease debugging by loading ~/.dashboardrc
(let ((debugcontrolf (conc (get-environment-variable "HOME") "/.dashboardrc")))
  (if (file-exists? debugcontrolf)
  (if (common:file-exists? debugcontrolf)
      (load debugcontrolf)))

(debug:setup)

(define *tim* (iup:timer))
(define *ord* #f)

  (if (not (null? path))
      (hash-table-ref/default (dboard:data-path-test-ids *data*) path #f)
      #f))

(define (test-panel window-id)
  (let* ((curr-row-num 0)
	 (viewlog    (lambda (x)
		       (if (file-exists? logfile)
		       (if (common:file-exists? logfile)
					;(system (conc "firefox " logfile "&"))
			   (iup:send-url logfile)
			   (message-window (conc "File " logfile " not found")))))
	 (xterm      (lambda (x)
		       (if (directory-exists? rundir)
			   (let ((shell (if (get-environment-variable "SHELL") 
					    (conc "-e " (get-environment-variable "SHELL"))

    (iup:callback-set! *tim*
		       "ACTION_CB"
		       (lambda (x)
			 ;; Want to dedicate no more than 50% of the time to this so skip if
			 ;; 2x delta time has not passed since last query
			 (if (< nextmintime (current-milliseconds))
			     (let* ((starttime (current-milliseconds))
				    (changes   (dcommon:run-update keys data runname keypatts testpatt states statuses 'full my-window-id))
				    ;; (changes   (dcommon:run-update keys data runname keypatts testpatt states statuses 'full my-window-id))
				    (endtime   (current-milliseconds)))
			       (set! nextmintime (+ endtime (* 2 (- endtime starttime))))
			       (debug:print 11 *default-log-port* "CHANGE(S): " (car changes) "..."))
			       ;; (debug:print 11 *default-log-port* "CHANGE(S): " (car changes) "..."))
                               )
			     (debug:print-info 11 *default-log-port* "Server overloaded"))))))

;; (dboard:data-updaters-set! *data* (make-hash-table))
(newdashboard #f) ;; *dbstruct-local*)    
(iup:main-loop)