Megatest

	  last-name))))

(define (str-first-char->number str)
  (char->integer (string-ref str 0)))
 
;; example of how to set up and write target mappers
;;

(add-target-mapper 'prefix-contour
		   (lambda (target run-name area area-path reason contour mode-patt)
		     (conc contour "/" target)))

(add-target-mapper 'prefix-area-contour
		   (lambda (target run-name area area-path reason contour mode-patt)
		     (conc area "/" contour "/" target)))
  

(add-runname-mapper 'corporate-ww
		    (lambda (target run-name area area-path reason contour mode-patt)
		      (print "corporate-ww called with: target=" target " run-name=" run-name " area=" area " area-path=" area-path " reason=" reason " contour=" contour " mode-patt=" mode-patt)
		      (let* ((last-name   (get-last-runname area-path target))
			     (last-letter (let* ((ch (if (string? last-name)
							 (let ((len (string-length last-name)))
							   (substring last-name (- len 1) len))
							 "a"))
						 (chnum (str-first-char->number ch))
						 (a     (str-first-char->number "a"))
						 (z     (str-first-char->number "z")))
					    (if (and (>= chnum a)(<= chnum z))
						chnum
						#f)))
			     (next-letter (if last-letter
					      (list->string
					       (list
						(integer->char
						 (+ last-letter 1)))) ;; surely there is an easier way?
					      "a")))
			;; (print "last-name: " last-name " last-letter: " last-letter " next-letter: " next-letter)
			(conc (seconds->wwdate (current-seconds)) next-letter))))

(add-runname-mapper 'auto

		    (lambda (target run-name area area-path reason contour mode-patt)
		      "auto-eh"))

;; run only areas where first letter of area name is "a"
;;
(add-area-checker 'first-letter-a
                  (lambda (area target contour)
                    (string-match "^a.*$" area)))

# 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 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 \
json md5 awful http-client spiffy uri-common intarweb spiffy-request-vars \
spiffy-directory-listing ssax sxml-serializer sxml-modifications iup canvas-draw sqlite3

# BASH_MACHTYPE=$(shell bash -c "echo \$$MACHTYPE")
# ARCHSTR=$(BASH_MACHTYPE)_$(shell lsb_release -sr)
ARCHSTR=$(shell lsb_release -sr)
# ARCHSTR=$(shell bash -c "echo \$$MACHTYPE")

PNGFILES = $(shell cd docs/manual;ls *png)

#all : $(PREFIX)/bin/.$(ARCHSTR) mtest dboard mtut ndboard
all : $(PREFIX)/bin/.$(ARCHSTR) mtest dboard mtut

mtest: $(OFILES) readline-fix.scm megatest.o
	csc $(CSCOPTS) $(OFILES) megatest.o -o mtest

dboard : $(OFILES) $(GOFILES) dashboard.scm
	csc $(CSCOPTS) $(OFILES) dashboard.scm $(GOFILES) -o dboard

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)/bin/tcmt $(PREFIX)/share/db/mt-pg.sql
#         $(PREFIX)/bin/.$(ARCHSTR)/ndboard

# $(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))
                   ;;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))
		   ((no-sync-get-lock)          (apply db:no-sync-get-lock    *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))

;; Copyright 2006-2014, 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.

;;  strftime('%m/%d/%Y %H:%M:%S','now','localtime')


(use (prefix sqlite3 sqlite3:) srfi-1 posix regex regex-case srfi-69 format md5 message-digest srfi-18)

(declare (unit archive))
(declare (uses db))
(declare (uses common))

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

(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 (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 (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 (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 (common:file-exists? archive-dir))
	     (create-directory archive-dir #t))
	 (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 (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
		  (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

      state=?,status=?,owner=?,event_time=?,comment=?,fail_count=?,pass_count=?
     WHERE id=?;"
   state status owner event-time comment fail-count pass-count run-id))

;; given all needed info create run record
;;
(define (pgdb:insert-run dbh ttype-id target run-name state status owner event-time comment fail-count pass-count)
    (dbi:exec
   dbh
   "INSERT INTO runs (ttype_id,target,run_name,state,status,owner,event_time,comment,fail_count,pass_count)
      VALUES (?,?,?,?,?,?,?,?,?,?);"
    ttype-id target run-name state status owner event-time comment fail-count pass-count))

;;======================================================================
;;  T E S T S

;;  implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
;;  PURPOSE.

;;======================================================================
;; C L I E N T S
;;======================================================================


(use srfi-18 extras tcp s11n srfi-1 posix regex srfi-69 hostinfo md5
     message-digest matchable spiffy uri-common intarweb http-client




     spiffy-request-vars uri-common intarweb directory-utils)

(declare (unit client))

(declare (uses common))
(declare (uses db))
(declare (uses tasks)) ;; tasks are where stuff is maintained about what is running.

;; gotta compile with csc, doesn't work with csi -s for whatever reason

(use srfi-69)
(use matchable)
(use utils)
(use ports)
(use extras)
(use srfi-1)
(use posix)
(use srfi-12)

;; turn scheme file to a list of sexps, sexps of interest will be in the form of (define (<procname> <args>) <body> )
(define (load-scm-file scm-file)
  ;;(print "load "scm-file)
  (handle-exceptions
   exn
   '()
   (with-input-from-string
       (conc "("
             (with-input-from-file scm-file read-all)
             ")" )
     read)))

;; extract a list of procname, filename, args and body of procedures defined in filename, input from load-scm-file
;;   -- be advised:
;;      * this may be fooled by macros, since this code does not take them into account.
;;      * this code does only checks for form (define (<procname> ... ) <body>)
;;           so it excludes from reckoning
;;               - generated functions, as in things like foo-set! from defstructs,
;;               - define-inline, (
;;               - define procname (lambda ..
;;               - etc...
(define (get-toplevel-procs+file+args+body filename)
  (let* ((scm-tree (load-scm-file filename))
         (procs
          (filter identity
                  (map
                   (match-lambda 
                    [('define ('uses args ...) body ...) #f] ;; filter out (define (uses ...
                    [('define ('unit args ...) body ...) #f] ;; filter out (define (unit ...
                    [('define ('prefix args ...) body ...) #f] ;; filter out (define (prefix ...
                    [('define (defname args ...) body ...) ;; match (define (procname <args>) <body>)
                     (if (atom? defname) ;; filter out things we dont understand (procname is a list, what??)
                         (list defname filename args body)
                         #f)]
                    [else #f] ) scm-tree))))
    procs))


;; given a sexp, return a flat lost of atoms in that sexp
(define (get-atoms-in-body body)
  (cond
   ((null? body) '())
   ((atom? body) (list body))
   (else
    (apply append (map get-atoms-in-body body)))))

;;  given a file, return a list of procname, file, list of atoms in said procname
(define (get-procs+file+atoms file)
  (let* ((toplevel-proc-items (get-toplevel-procs+file+args+body file))
         (res
          (map
           (lambda (item)
             (let* ((proc (car item))
                    (file (cadr item))
                    (args (caddr item))
                    (body (cadddr item))
                    (atoms (append (get-atoms-in-body args) (get-atoms-in-body body))))
               (list proc file atoms)))
           toplevel-proc-items)))
    res))

;; uniquify a list of atoms 
(define (unique-atoms lst)
  (let loop ((lst (flatten lst)) (res '()))
    (if (null? lst)
        (reverse res)
        (let ((c (car lst)))
          (loop (cdr lst) (if (member c res) res (cons c res)))))))

;; given a list of procname, filename, list of procs called from procname, cross reference and reverse
;; returning alist mapping procname to procname that calls said procname
(define (get-callers-alist all-procs+file+calls)
  (let* ((all-procs (map car all-procs+file+calls))
         (caller-ht (make-hash-table))) 
    ;; let's cross reference with a hash table
    (for-each (lambda (proc) (hash-table-set! caller-ht proc '())) all-procs)
    (for-each (lambda (item)
               (let* ((proc (car item))
                      (file (cadr item))
                      (calls (caddr item)))
                 (for-each (lambda (callee)
                             (hash-table-set! caller-ht callee
                                              (cons proc
                                                    (hash-table-ref caller-ht callee))))
                           calls)))
              all-procs+file+calls)
    (map (lambda (x)
           (let ((k (car x))
                 (r (unique-atoms (cdr x))))
             (cons k r)))                    
         (hash-table->alist caller-ht))))

;; create a handy cross-reference of callees to callers in the form of an alist.
(define (get-xref all-scm-files)
  (let* ((all-procs+file+atoms
          (apply append (map get-procs+file+atoms all-scm-files)))
         (all-procs (map car all-procs+file+atoms))
         (all-procs+file+calls  ; proc calls things in calls list
          (map (lambda (item)
                 (let* ((proc (car item))
                        (file (cadr item))
                        (atoms (caddr item))
                        (calls
                         (filter identity
                                 (map
                                  (lambda (x)
                                    (if (and ;; (not (equal? x proc))  ;; uncomment to prevent listing self
                                         (member x all-procs))
                                        x
                                        #f))
                                  atoms))))
                   (list proc file calls)))
               all-procs+file+atoms))
         (callers (get-callers-alist all-procs+file+calls))) 
    callers))

;;======================================================================
;; 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 data-structures 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

     pkts (prefix dbi dbi:))


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

(declare (unit common))


(include "common_records.scm")

;; (require-library margs)
;; (include "margs.scm")

;; (define old-exit exit)
;; 
;; (define (exit . code)
;;   (if (null? code)
;;       (old-exit)
;;       (old-exit code)))


;; execute thunk, return value.  If exception thrown, trap exception, return #f, and emit nonfatal condition note to *default-log-port* .
;; arguments - thunk, message
(define (common:fail-safe thunk warning-message-on-exception)
  (handle-exceptions
   exn
   (begin
     (debug:print-info 0 *default-log-port* "notable but nonfatal condition - "warning-message-on-exception)
     (debug:print-info 0 *default-log-port*
                       (string-substitute "\n?Error:" "nonfatal condition:"
                                          (with-output-to-string
                                            (lambda ()
                                              (print-error-message exn) ))))
     (debug:print-info 0 *default-log-port* "    -- continuing after nonfatal condition...")
     #f)
   (thunk)))

(define getenv get-environment-variable)
(define (safe-setenv key val)
  (if (substring-index ":" key) ;; variables containing : are for internal use and cannot be environment variables.
      (debug:print-error 4 *default-log-port* "skip setting internal use only variables containing \":\"")
      (if (and (string? val)
	       (string? key))
	  (handle-exceptions
	      exn
	      (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"))


;; returns list of fd count, socket count
(define (get-file-descriptor-count #!key  (pid (current-process-id )))
  (list
    (length (glob (conc "/proc/" pid "/fd/*")))
    (length  (filter identity (map socket? (glob (conc "/proc/" pid "/fd/*")))))
  )
)


;; 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:expiration-timeout))
  (force-server      #f)
  (ro-mode           #f)  
  (ro-mode-checked   #f)) ;; flag that indicates we have checked for ro-mode

;; launching and hosts
(defstruct host
  (reachable    #f)

(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))))

(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 #!key (full #f))
  (apply db:multi-db-sync 
   dbstruct
   'schema
   ;; 'new2old
   'killservers

   'adj-target
   ;; 'old2new
   'new2old
   ;; (if full
       '(dejunk)
       ;; '())
       )
  (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 (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: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 (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 (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 (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))))

;;======================================================================
;; 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 (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 (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*   ;; 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 "DEAD")
    (9 "FAIL")
    (10 "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" "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 . junk)
  (if *db-cache-path*
      *db-cache-path*
      (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)
















(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)
		  (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)))







































































;; 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.")
	      (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 (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)
		   (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 (common:file-exists? exe-path)
	(handle-exceptions
	 exn
	 #f
	 (pathname-directory
	  (pathname-directory 
	   (pathname-directory exe-path))))
	#f)))

;; return first path that can be created or already exists and is writable
;;
(define (common:get-create-writeable-dir dirs)
  (if (null? dirs)
      #f
      (let loop ((hed (car dirs))
		 (tal (cdr dirs)))
	(let ((res (or (and (directory? hed)
			    (file-write-access? hed)
			    hed)
		       (handle-exceptions
			   exn
			   (begin
			     (debug:print-info 0 *default-log-port* "could not create " hed ", this might cause problems down the road.")
			     #f)
			(create-directory hed #t)))))
	  (if (and (string? res)
		   (directory? res))
	      res
	      (if (null? tal)
		  #f
		  (loop (car tal)(cdr tal))))))))

(define (common:args-get-testpatt rconf)
  (let* (;; (tagexpr       (args:get-arg "-tagexpr"))
         ;; (tags-testpatt (if tagexpr (string-join (runs:get-tests-matching-tags tagexpr) ",") #f))
         (testpatt-key  (if (args:get-arg "--modepatt") (args:get-arg "--modepatt") "TESTPATT"))
         (args-testpatt (or (args:get-arg "-testpatt") (args:get-arg "-runtests") "%"))
         (rtestpatt     (if rconf (runconfigs-get rconf testpatt-key) #f)))
    (cond
     ((args:get-arg "--modepatt") ;; modepatt is a forced setting, when set it MUST refer to an existing PATT in the runconfig
      (if rconf
	  (runconfigs-get rconf testpatt-key)
	  #f))     ;; We do NOT fall back to "%"
     ;; (tags-testpatt
     ;;  (debug:print-info 0 *default-log-port* "-tagexpr "tagexpr" selects testpatt "tags-testpatt)
     ;;  tags-testpatt)
     ((and (equal? args-testpatt "%") rtestpatt)
      (debug:print-info 0 *default-log-port* "testpatt defined in "testpatt-key" from runconfigs: " rtestpatt)
      rtestpatt)
     (else args-testpatt))))

(define (common:directory-exists? path-string)
  ;;;; TODO: catch permission denied exceptions and emit appropriate warnings, eg:  system error while trying to access file: "/nfs/pdx/disks/icf_env_disk001/bjbarcla/gwa/issues/mtdev/randy-slow/reproduce/q...
  (common:false-on-exception (lambda () (directory-exists? path-string))
                             message: (conc "Unable to access path: " path-string)
                             ))

;; does the directory exist and do we have write access?
;;
;;    returns the directory or #f
;;
(define (common:directory-writable? path-string)
  (handle-exceptions
   exn
   #f
   (if (and (directory-exists? path-string)
            (file-write-access? path-string))
       path-string
       #f)))

(define (common:get-linktree)
  (or (getenv "MT_LINKTREE")
      (if *configdat*
	  (configf:lookup *configdat* "setup" "linktree")
	  (if *toppath*
	      (conc *toppath* "/lt")


	      #f))))

(define (common:args-get-runname)
  (let ((res (or (args:get-arg "-runname")
		 (args:get-arg ":runname")
		 (getenv "MT_RUNNAME"))))
    ;; (if res (set-environment-variable "MT_RUNNAME" res)) ;; not sure if this is a good idea. side effect and all ...
    res))

(define (common:get-fields cfgdat)
  (let ((fields (hash-table-ref/default cfgdat "fields" '())))
    (map car fields)))

(define (common:args-get-target #!key (split #f)(exit-if-bad #f))
  (let* ((keys    (if (hash-table? *configdat*) (common:get-fields *configdat*) '()))
	 (numkeys (length keys))
	 (target  (or (args:get-arg "-reqtarg")
		      (args:get-arg "-target")
		      (getenv "MT_TARGET")))
	 (tlist   (if target (string-split target "/" #t) '()))
	 (valid   (if target
		      (or (null? keys) ;; probably don't know our keys yet

	    target)
	(if target
	    (begin
	      (debug:print-error 0 *default-log-port* "Invalid target, spaces or blanks not allowed \"" target "\", target should be: " (string-intersperse keys "/") ", have " tlist " for elements")
	      (if exit-if-bad (exit 1))
	      #f)
	    #f))))

;; looking only (at least for now) at the MT_ variables craft the full testname
;;
(define (common:get-full-test-name)
  (if (getenv "MT_TEST_NAME")
      (if (and (getenv "MT_ITEMPATH")
               (not (equal? (getenv "MT_ITEMPATH") "")))
          (getenv "MT_TEST_NAME")
          (conc (getenv "MT_TEST_NAME") "/" (getenv "MT_ITEMPATH")))
      #f))

;; logic for getting homehost. Returns (host . at-home)
;; IF *toppath* is not set, wait up to five seconds trying every two seconds
;; (this is to accomodate the watchdog)
;;
(define (common:get-homehost #!key (trynum 5))
  ;; called often especially at start up. use mutex to eliminate collisions

				   (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 (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

    (if hh
	(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* ;; 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")
	    (set! res #t)
	    (if (equal? (getenv "MT_USE_CACHE") "no")
		(set! res #f))))    ;; overrides -no-cache switch
    res))
  
;; force use of server?
;;
(define (common:force-server?)
  (let* ((force-setting (configf:lookup *configdat* "server" "force"))
	 (force-type    (if force-setting (string->symbol force-setting) #f))
	 (force-result  (case force-type
			  ((#f)     #f)
			  ((always) #t)
			  ((test)   (if (args:get-arg "-execute") ;; we are in a test
					#t
					#f))
			  (else
			   (debug:print 0 *default-log-port* "ERROR: Bad server force setting " force-setting ", forcing server.")
			   #t)))) ;; default to requiring server
    (if force-result
	(begin
	  (debug:print-info 0 *default-log-port* "forcing use of server, force setting is \"" force-setting "\".")
	  #t)
	#f)))








;;======================================================================
;; M I S C   L I S T S
;;======================================================================

;; items in lista are matched value and position in listb
;; return the remaining items in listb or #f
;;

          ((< (+ 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-in numcpus-in waitdelay #!key (count 1000) (msg #f)(remote-host #f))
  (let* ((loadavg (common:get-cpu-load remote-host))
	 (numcpus (if (< 1 numcpus-in) ;; not possible
		      (common:get-num-cpus remote-host)
		      numcpus-in))
	 (maxload (max maxload-in 0.5)) ;; so maxload must be greater than 0.5 for now BUG - FIXME?
	 (first   (car loadavg))
	 (next    (cadr loadavg))
	 (adjload (* maxload (max 1 numcpus))) ;; possible bug where numcpus (or could be maxload) is zero, crude fallback is to at least use 1
	 (loadjmp (- first next)))
    (cond
     ((and (> first adjload)
	   (> count 0))
      (debug:print-info 0 *default-log-port* "server start delayed " waitdelay " seconds due to load " first " exceeding max of " adjload " on server " (or remote-host (get-host-name)) " (normalized load-limit: " maxload ") " (if msg msg ""))
      (thread-sleep! waitdelay)
      (common:wait-for-cpuload maxload numcpus waitdelay count: (- count 1) msg: msg remote-host: remote-host))
     ((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) msg: msg remote-host: remote-host)))))

(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 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 msg remote-host)
  (let ((num-cpus (common:get-num-cpus remote-host)))
    (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 (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 
			  (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)))
   "/"))

;;======================================================================
;; L O C K I N G   M E C H A N I S M S 
;;======================================================================

;; faux-lock is deprecated. Please use simple-lock below
;;
(define (common:faux-lock keyname #!key (wait-time 8)(allow-lock-steal #t))
  (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: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:no-sync-get/default keyname #f))))
      (begin
        (if (rmt:no-sync-get/default keyname #f) (rmt:no-sync-del! keyname))
        #t)
      #f))

;; simple lock. improve and converge on this one.
;;
(define (common:simple-lock keyname)
  (rmt:no-sync-get-lock keyname))

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

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

(define (common:get-color-from-status status)
  (cond
   ((equal? status "PASS")    "green")
   ((equal? status "FAIL")    "red")
   ((equal? status "WARN")    "orange")
   ((equal? status "KILLED")  "orange")
   ((equal? status "KILLREQ") "purple")
   ((equal? status "RUNNING") "blue")
   ((equal? status "ABORT")   "brown")
   (else "black")))

;; ;;======================================================================
;; ;; N A N O M S G   C L I E N T
;; ;;======================================================================
;; 












;; 
;; 
;; (define (common:send-dboard-main-changed)
;;   (let* ((dashboard-ips (mddb:get-dashboards)))
;;     (for-each
;;      (lambda (ipadr)
;;        (let* ((soc (common:open-nm-req (conc "tcp://" ipadr)))
;; 	      (msg (conc "main " *toppath*))
;; 	      (res (common:nm-send-receive-timeout soc msg)))
;; 	 (if (not res) ;; couldn't reach that dashboard - remove it from db
;; 	     (print "ERROR: couldn't reach dashboard " ipadr))
;; 	 res))
;;      dashboard-ips)))
;;     
;;     
;; ;;======================================================================
;; ;; D A S H B O A R D   D B 
;; ;;======================================================================
;; 
;; (define (mddb:open-db)
;;   (let* ((db (open-database (conc (get-environment-variable "HOME") "/.dashboard.db"))))
;;     (set-busy-handler! db (busy-timeout 10000))
;;     (for-each
;;      (lambda (qry)
;;        (exec (sql db qry)))
;;      (list 
;;       "CREATE TABLE IF NOT EXISTS vars       (id INTEGER PRIMARY KEY,key TEXT, val TEXT, CONSTRAINT varsconstraint UNIQUE (key));"
;;       "CREATE TABLE IF NOT EXISTS dashboards (
;;           id         INTEGER PRIMARY KEY,
;;           pid        INTEGER,
;;           username   TEXT,
;;           hostname   TEXT,
;;           ipaddr     TEXT,
;;           portnum    INTEGER,
;;           start_time TIMESTAMP DEFAULT (strftime('%s','now')),
;;              CONSTRAINT hostport UNIQUE (hostname,portnum)
;;         );"
;;       ))
;;     db))
;; 
;; ;; register a dashboard 
;; ;;
;; (define (mddb:register-dashboard port)
;;   (let* ((pid      (current-process-id))
;; 	 (hostname (get-host-name))
;; 	 (ipaddr   (server:get-best-guess-address hostname))
;; 	 (username (current-user-name)) ;; (car userinfo)))
;; 	 (db      (mddb:open-db)))
;;     (print "Register monitor, pid: " pid ", hostname: " hostname ", port: " port ", username: " username)
;;     (exec (sql db "INSERT OR REPLACE INTO dashboards (pid,username,hostname,ipaddr,portnum) VALUES (?,?,?,?,?);")
;; 	   pid username hostname ipaddr port)
;;     (close-database db)))
;; 
;; ;; unregister a monitor
;; ;;
;; (define (mddb:unregister-dashboard host port)
;;   (let* ((db      (mddb:open-db)))
;;     (print "Register unregister monitor, host:port=" host ":" port)
;;     (exec (sql db "DELETE FROM dashboards WHERE hostname=? AND portnum=?;") host port)
;;     (close-database db)))
;; 
;; ;; get registered dashboards
;; ;;
;; (define (mddb:get-dashboards)
;;   (let ((db (mddb:open-db)))
;;     (query fetch-column
;; 	   (sql db "SELECT ipaddr || ':' || portnum FROM dashboards;"))))
    
;;======================================================================
;;  T E S T   L A U N C H I N G   P E R   I T E M   W I T H   H O S T   T Y P E S
;;======================================================================
;; 
;; [hosts]
;; arm cubie01 cubie02

;; 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 (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 (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 (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 (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 (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 (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 (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

								 (proc (cdr dat)))
							     (if (string-match patt curr-section-name)
								 (proc curr-section-name section-name res path))))
							 post-section-procs)
                                                        ;; after gathering the vars for a section and if apply-wildcards is true and if there is a wildcard in the section name process wildcards
                                                        ;; NOTE: we are processing the curr-section-name, NOT section-name.
                                                        (process-wildcards res curr-section-name)
							(if (not (hash-table-ref/default res section-name #f))(hash-table-set! res section-name '())) ;; ensure that mere mention of a section is not lost
							(loop (configf:read-line inp res (calc-allow-system allow-system curr-section-name sections) settings)
							      ;; if we have the sections list then force all settings into "" and delete it later?
							      ;; (if (or (not sections) 
							      ;;	      (member section-name sections))
							      ;;	  section-name "") ;; stick everything into "". NOPE: We need new strategy. Put stuff in correct sections and then delete all sections later.
							      section-name
							      #f #f)))

;; pathenvvar will set the named var to the path of the config
(define (find-and-read-config fname #!key (environ-patt #f)(given-toppath #f)(pathenvvar #f))
  (let* ((curr-dir   (current-directory))
         (configinfo (find-config fname toppath: given-toppath))
	 (toppath    (car configinfo))
	 (configfile (cadr configinfo))
	 (set-fields (lambda (curr-section next-section ht path)
		       (let ((field-names (if ht (common:get-fields ht) '()))
			     (target      (or (getenv "MT_TARGET")(args:get-arg "-reqtarg")(args:get-arg "-target"))))
			 (debug:print-info 9 *default-log-port* "set-fields with field-names=" field-names " target=" target " curr-section=" curr-section " next-section=" next-section " path=" path " ht=" ht)
			 (if (not (null? field-names))(keys:target-set-args field-names target #f))))))
    (if toppath (change-directory toppath)) 
    (if (and toppath pathenvvar)(setenv pathenvvar toppath))
    (let ((configdat  (if configfile 
			  (read-config configfile #f #t environ-patt: environ-patt post-section-procs: (list (cons "^fields$" set-fields)) #f))))

	    (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 (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 (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 (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))



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

;;======================================================================
;; C O M M O N
;;======================================================================

(define *dashboard-comment-share-slot* #f)

(define (dtests:get-pre-command #!key (default-override #f))
  (let* ((orig-pre-command "export CMD='")
         (viewscreen-pre-command  "viewscreen ")
         (use-viewscreen (configf:lookup *configdat* "dashboard" "use-viewscreen"))
         (default-pre-command (if use-viewscreen viewscreen-pre-command orig-pre-command))
         (cfg-ovrd (configf:lookup *configdat* "dashboard" "pre-command")))
    (or cfg-ovrd default-override default-pre-command))) ;; "xterm -geometry 180x20 -e \""))

  
(define (dtests:get-post-command #!key (default-override #f))
  (let* ((orig-post-command (conc "';xterm -geometry 180x20 -e \"(echo; echo -n START:;date +ww%U.%w-$H:%M:%S;echo;echo $CMD;echo;$CMD)|&"
                                 "tee -a runlog-`date +ww%U.%w-%H:%M`.log;echo Press any key to continue;bash -c 'read -n 1 -s'\" &"))
         (viewscreen-post-command  "")
         (use-viewscreen (configf:lookup *configdat* "dashboard" "use-viewscreen"))
         (default-post-command (if use-viewscreen viewscreen-post-command orig-post-command))
         (cfg-ovrd (configf:lookup *configdat* "dashboard" "post-command")))
    (or cfg-ovrd default-override default-post-command))) ;; ";echo Press any key to continue;bash -c 'read -n 1 -s'\" &")))


(define (test-info-panel testdat store-label widgets)
  (iup:frame 
   #:title "Test Info" ; #:expand "YES"
   (iup:hbox ; #:expand "YES"
    (apply iup:vbox ; #:expand "YES"

			   (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 (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 (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 allow-write-cache: #f)
				 (tests:get-testconfig (db:test-get-testname testdat) item-path test-registry #t allow-write-cache: #f))))
	       (viewlog    (lambda (x)
			     (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 (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 (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

(include "task_records.scm")
(include "megatest-fossil-hash.scm")
(include "vg_records.scm")

(define help (conc 
	      "Megatest Dashboard, documentation at http://www.kiatoa.com/fossils/megatest
  version " megatest-version "
  license GPL, Copyright (C) Matt Welland 2012-2017

Usage: dashboard [options]
  -h                    : this help
  -test run-id,test-id  : control test identified by testid
  -skip-version-check   : skip the version check
  -use-db-cache         : access database via cache 

			"-guimonitor"
			"-main"
			"-v"
			"-q"
			"-use-db-cache"
			"-skip-version-check"
			"-repl"
                        "-rh5.11" ;; fix to allow running on rh5.11
			)
		 args:arg-hash
		 0))

(if (not (null? remargs))
    (begin
      (print "Unrecognised arguments: " (string-intersperse remargs " "))
      (exit)))

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

;; TODO: Move this inside (main)
;;
(if (not (launch:setup))
    (begin
      (print "Failed to find megatest.config, exiting") 
      (exit 1)))

;; deal with RH 5.11 gtk lib or iup lib missing detachbox feature
;; first check for the switch
;;
(if (or (args:get-arg "-rh5.11")
	(configf:lookup *configdat* "dashboard" "no-detachbox"))
    (set! iup:detachbox iup:vbox))

(if (not (common:on-homehost?))
    (begin
      (debug:print 0 *default-log-port* "WARNING: Current policy requires running dashboard on homehost: " (common:get-homehost))))
    
;; RA => Might require revert for filters 
;; create a watch dog to move changes from lt/.db/*.db to megatest.db

  ;; runs
  ((allruns         '())                 : list)        ;; list of dboard:rundat records
  ((allruns-by-id    (make-hash-table))  : hash-table)  ;; hash of run-id -> dboard:rundat records
  ((done-runs       '())                 : list)        ;; list of runs already drawn
  ((not-done-runs   '())                 : list)        ;; list of runs not yet drawn
  (header            #f)                                ;; header for decoding the run records
  (keys              #f)                                ;; keys for this run (i.e. target components)
  ((numruns          (string->number (or (args:get-arg "-cols")
					 (configf:lookup *configdat* "dashboard" "cols")
					 "8")))                 : number)      ;; 
  ((tot-runs          0)                 : number)
  ((last-data-update  0)                 : number)      ;; last time the data in allruns was updated
  ((last-runs-update  0)                 : number)      ;; last time we pulled the runs info to update the tree
  (runs-mutex         (make-mutex))                     ;; use to prevent parallel access to draw objects
  ((run-update-times  (make-hash-table)) : hash-table)  ;; update times indexed by run-id
  ((last-test-dat      (make-hash-table)) : hash-table)  ;; cache last tests dat by run-id
  ((run-db-paths      (make-hash-table)) : hash-table)  ;; cache the paths to the run db files

  ;; Runs view
  ((buttondat         (make-hash-table)) : hash-table)  ;;     
  ((item-test-names  '())                : list)        ;; list of itemized tests
  ((run-keys          (make-hash-table)) : hash-table)
  (runs-matrix        #f)                               ;; used in newdashboard
  ((start-run-offset   0)                : number)      ;; left-right slider value
  ((start-test-offset  0)                : number)      ;; up-down slider value
  ((runs-btn-height    (or (configf:lookup *configdat* "dashboard" "btn-height") "x16")) : string)  ;; was 12
  ((runs-btn-fontsz    (or (configf:lookup *configdat* "dashboard" "btn-fontsz") "10")) : string)   ;; was 8
  ((runs-cell-width    (or (configf:lookup *configdat* "dashboard" "cell-width") "50")) : string)   ;; was 50
  ((all-test-names     '())              : list)
  
  ;; Canvas and drawing data
  (cnv                #f)
  (cnv-obj            #f)
  (drawing            #f)
  ((run-start-row     0)                 : number)

  (dboard:tabdat-dbdir-set! tabdat (db:dbfile-path)) ;; (conc (configf:lookup *configdat* "setup" "linktree") "/.db"))
  (dboard:tabdat-dbfpath-set! tabdat (db:dbfile-path))
  (dboard:tabdat-monitor-db-path-set! tabdat (conc (dboard:tabdat-dbdir tabdat) "/monitor.db"))

  ;; HACK ALERT: this is a hack, please fix.
  (dboard:tabdat-ro-set! tabdat (not (file-read-access? (dboard:tabdat-dbfpath tabdat))))
  
  (dboard:tabdat-keys-set! tabdat (rmt:get-keys))
  (dboard:tabdat-dbkeys-set! tabdat (append (dboard:tabdat-keys tabdat) (list "runname")))
  (dboard:tabdat-tot-runs-set! tabdat (rmt:get-num-runs "%"))
  )

;; RADT => Matrix defstruct addition
(defstruct dboard:graph-dat
    ((id           #f) : string)

			     db-pth)))
	 (db-mod-time  (common:lazy-sqlite-db-modification-time db-path))
	 (db-modified  (>= db-mod-time last-db-time))
	 (multi-get    (> (dboard:rundat-run-data-offset run-dat) 0))  ;; multi-get in progress
	 (tmptests     (if (or do-not-use-db-file-timestamps
			       (dboard:tabdat-filters-changed tabdat)
			       db-modified)

			   (rmt:get-tests-for-run run-id testnamepatt states statuses     ;; run-id testpatt states statuses
					      (dboard:rundat-run-data-offset run-dat) ;; query offset
					      num-to-get
					      (dboard:tabdat-hide-not-hide tabdat) ;; no-in
					      sort-by                              ;; sort-by
					      sort-order                           ;; sort-order
					      #f ;; 'shortlist                     ;; qrytype
					      last-update                          ;; last-update

;; this calls dboard:get-tests-for-run-duplicate for each run
;;
;; create a virtual table of all the tests
;; keypatts: ( (KEY1 "abc%def")(KEY2 "%") )
;;
(define (update-rundat tabdat runnamepatt numruns testnamepatt keypatts)
  (let* ((access-mode      (dboard:tabdat-access-mode tabdat))
         (keys             (rmt:get-keys))
	 (last-runs-update (- (dboard:tabdat-last-runs-update tabdat) 2))

         (allruns          (rmt:get-runs runnamepatt numruns (dboard:tabdat-start-run-offset tabdat) keypatts))
         ;;(allruns-tree (rmt:get-runs-by-patt (dboard:tabdat-keys tabdat) "%" #f #f #f #f))

         (allruns-tree    (rmt:get-runs-by-patt keys "%" #f #f #f #f last-runs-update)) ;;'("id" "runname")
	 (header      (db:get-header allruns))
	 (runs        (db:get-rows   allruns)) ;; RA => Filtered as per runpatt selected
         (runs-tree   (db:get-rows   allruns-tree)) ;; RA => Returns complete list of runs
	 (start-time  (current-seconds))
	 (runs-hash   (let ((ht (make-hash-table)))
			 (for-each (lambda (run)
				     (hash-table-set! ht (db:get-value-by-header run header "id") run))

;; create a virtual table of all the tests
;; keypatts: ( (KEY1 "abc%def")(KEY2 "%") )
;;
(define (dboard:update-rundat tabdat runnamepatt numruns testnamepatt keypatts)
  (let* ((access-mode      (dboard:tabdat-access-mode tabdat))
         (keys             (dboard:tabdat-keys tabdat)) ;; (db:dispatch-query access-mode rmt:get-keys db:get-keys)))
	 (last-runs-update (- (dboard:tabdat-last-runs-update tabdat) 2))

         (allruns          (rmt:get-runs runnamepatt numruns (dboard:tabdat-start-run-offset tabdat) keypatts))
         ;;(allruns-tree (rmt:get-runs-by-patt (dboard:tabdat-keys tabdat) "%" #f #f #f #f))

         (allruns-tree    (rmt:get-runs-by-patt keys "%" #f #f #f #f 0)) ;; last-runs-update));;'("id" "runname")
	 (header      (db:get-header allruns))
	 (runs        (db:get-rows   allruns)) ;; RA => Filtered as per runpatt selected
         (runs-tree   (db:get-rows   allruns-tree)) ;; RA => Returns complete list of runs
	 (start-time  (current-seconds))
	 (runs-hash   (let ((ht (make-hash-table)))
			 (for-each (lambda (run)
				     (hash-table-set! ht (db:get-value-by-header run header "id") run))

		   (elapsed-time (- (current-seconds) start-time)))
	      (if (null? all-test-ids)
		  (hash-table-delete! (dboard:tabdat-allruns-by-id tabdat) run-id)
		  (hash-table-set!    (dboard:tabdat-allruns-by-id tabdat) run-id run-struct))
	      (if (or (null? tal)
		      (> elapsed-time 2)) ;; stop loading data after 5 seconds, on the next call more data *should* be loaded since get-tests-for-run uses last update
		  (begin
		    (when (> elapsed-time 2)   
                      (debug:print 0 *default-log-port* "NOTE: updates are taking a long time, " elapsed-time "s elapsed.")
                      (let* ((old-val (iup:attribute *tim* "TIME"))
                             (new-val (number->string (inexact->exact (floor (* 2  (string->number old-val)))))))
                        (debug:print 0 *default-log-port* "NOTE: increasing poll interval from "old-val" to "new-val)
                        (iup:attribute-set! *tim* "TIME" new-val))


                      )
		    (dboard:tabdat-allruns-set! tabdat new-res)
		    maxtests)
		  (if (> (dboard:rundat-run-data-offset run-struct) 0)
		      (loop run tal new-res newmaxtests) ;; not done getting data for this run
		      (loop (car tal)(cdr tal) new-res newmaxtests)))))))
    (dboard:tabdat-filters-changed-set! tabdat #f)
    (dboard:update-tree tabdat runs-hash header tb)))

			    "190 190 190"
			    ))
    (dboard:tabdat-filters-changed-set! tabdat #t)))

(define (update-search commondat tabdat x val)
  (hash-table-set! (dboard:tabdat-searchpatts tabdat) x val)
  (dboard:tabdat-filters-changed-set! tabdat #t)
  (mark-for-update tabdat)
  (set-bg-on-filter commondat tabdat))

;; force ALL updates to zero (effectively)
;;
(define (mark-for-update tabdat)
  (dboard:tabdat-last-db-update-set! tabdat (make-hash-table)))

	     ;;        #:value 300

	     ;; Target, testpatt, state and status input boxes
	     ;;
	     (iup:vbox
	      ;; Command to run, placed over the top of the canvas
	      (dcommon:command-action-selector commondat tabdat tab-num: tab-num)
              (dboard:runs-tree-browser commondat tabdat)
	      (dcommon:command-runname-selector commondat tabdat tab-num: tab-num)
	      (dcommon:command-testname-selector commondat tabdat update-keyvals))
	     ;;  key-listboxes))
	     (dcommon:command-tests-tasks-canvas tabdat test-records sorted-testnames tests-draw-state))))
	   (tb (dboard:tabdat-runs-tree tabdat)))
      (dboard:commondat-add-updater 
       commondat 
       (lambda ()
         (if (dashboard:database-changed? commondat tabdat context-key: 'run-control)
             (dashboard:update-tree-selector tabdat)))
       tab-num: tab-num)
      result)))

 ;;(iup:frame
 ;; #:title "Logs" ;; To be replaced with tabs
 ;; (let ((logs-tb (iup:textbox #:expand "YES"
 ;;				   #:multiline "YES")))
 ;;	 (dboard:tabdat-logs-textbox-set! tabdat logs-tb)
 ;;	 logs-tb))

;; browse runs as a tree. Used in both "Runs" tab and
;; in the runs control panel.
;;
(define (dboard:runs-tree-browser commondat tabdat)
  (let* (
	 (txtbox (iup:textbox #:action (lambda (val a b)
					 (debug:catch-and-dump
					  (lambda ()
					    ;; for the Runs view we put the list of keyvals into tabdat target
					    ;; for the Run Controls we put then update the run-command
					    (if b (dboard:tabdat-target-set! tabdat (string-split b "/")))
					    (dashboard:update-run-command tabdat))
					  "command-testname-selector tb action"))
			      #:value (dboard:test-patt->lines
				       (dboard:tabdat-test-patts-use tabdat))
			      #:expand "HORIZONTAL"
			      ;; #:size "10x30"
			      ))
	 (tb
          (iup:treebox
           #:value 0
           #:name "Runs"
           #:expand "YES"
           #:addexpanded "YES"
           #:size "10x"
           #:selection-cb
           (lambda (obj id state)
             (debug:catch-and-dump
              (lambda ()
                (let* ((run-path (tree:node->path obj id))
                       (run-id    (tree-path->run-id tabdat (cdr run-path))))

                        (dboard:tabdat-curr-run-id-set! tabdat run-id)
                        (dboard:tabdat-view-changed-set! tabdat #t))
                      (debug:print-error 5 *default-log-port* "tree-path->run-id returned non-number " run-id))))
              "treebox"))
           ;; (print "path: " (tree:node->path obj id) " run-id: " run-id)
           )))
    (dboard:tabdat-runs-tree-set! tabdat tb)
    (iup:detachbox
     (iup:vbox 
      tb
      txtbox))))

;;======================================================================
;; R U N   C O N T R O L S
;;======================================================================
;;
;; A gui for launching tests
;;

      (dcommon:command-runname-selector commondat tabdat tab-num: tab-num)
      (dcommon:command-testname-selector commondat tabdat update-keyvals))
     (iup:vbox
      (iup:split
       #:orientation "HORIZONTAL"
       #:value 800
      (let* ((cnv-obj (iup:canvas 
		       ;; #:size "250x250" ;; "500x400"
		       #:expand "YES"
		       #:scrollbar "YES"
		       #:posx "0.5"
		       #:posy "0.5"
		       #:action (make-canvas-action
				  (lambda (c xadj yadj)
				    (debug:catch-and-dump

		       )))
	cnv-obj)
      (let* ((hb1 (iup:hbox))
             (graph-cell-table (dboard:tabdat-graph-cell-table tabdat))
             (changed #f)
             (graph-matrix (iup:matrix
                           #:alignment1 "ALEFT"
                           ;; #:expand "YES" ;; "HORIZONTAL"
                           #:scrollbar "YES"
                           #:numcol 10
                           #:numlin 20
                           #:numcol-visible 5 ;; (min 8)
                           #:numlin-visible 1
                           #:click-cb
                           (lambda (obj row col status)
                             (let*
                                 ((graph-cell (conc row ":" col))
                                 (graph-dat   (hash-table-ref/default graph-cell-table graph-cell #f))
                                 (graph-flag  (dboard:graph-dat-flag graph-dat)))

;; display and manage a single run at a time

(define (tree-path->run-id tabdat path)
  (if (not (null? path))
      (hash-table-ref/default (dboard:tabdat-path-run-ids tabdat) path #f)
      #f))

;; (define (dboard:get-tests-dat tabdat run-id last-update)
;;   (let* ((access-mode     (dboard:tabdat-access-mode tabdat))
;;          (tdat (if run-id (db:dispatch-query access-mode rmt:get-tests-for-run db:get-tests-for-run
;;                                              run-id 
;; 					     (hash-table-ref/default (dboard:tabdat-searchpatts tabdat) "test-name" "%/%")
;; 					     (hash-table-keys (dboard:tabdat-state-ignore-hash tabdat))  ;; '()
;; 					     (hash-table-keys (dboard:tabdat-status-ignore-hash tabdat)) ;; '()
;; 					     #f #f                                                       ;; offset limit
;; 					     (dboard:tabdat-hide-not-hide tabdat)                        ;; not-in
;; 					     #f #f                                                       ;; sort-by sort-order
;; 					     #f ;; get all? "id,testname,item_path,state,status,event_time,run_duration"                        ;; qryval
;;                                              (if (dboard:tabdat-filters-changed tabdat)

;; 					         0
;; 					         last-update)
;; 					     *dashboard-mode*)
;; 		  '()))) ;; get 'em all
;;     ;; (debug:print 0 *default-log-port* "dboard:get-tests-dat: got " (length tdat) " test records for run " run-id)
;;     (sort tdat (lambda (a b)
;; 		 (let* ((aval (vector-ref a 2))
;; 			(bval (vector-ref b 2))
;; 			(anum (string->number aval))
;; 			(bnum (string->number bval)))
;; 		   (if (and anum bnum)
;; 		       (< anum bnum)
;; 		       (string<= aval bval)))))))


(define (dashboard:safe-cadr-assoc name lst)
  (let ((res (assoc name lst)))
    (if (and res (> (length res) 1))
	(cadr res)
	#f)))

(define (dboard:update-tree tabdat runs-hash runs-header tb)
  (let* ((access-mode   (dboard:tabdat-access-mode tabdat))
         (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 runs-header "event_time"))
				 (time-b   (db:get-value-by-header record-b runs-header "event_time")))
			    (< time-a time-b)))))
         (changed      #f)
	 (last-runs-update  (dboard:tabdat-last-runs-update tabdat))

	 (runs-dat     (rmt:get-runs-by-patt (dboard:tabdat-keys tabdat) "%" #f #f #f #f last-runs-update)))
    (dboard:tabdat-last-runs-update-set! tabdat (- (current-seconds) 2))
    (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 runs-header key))
					(dboard:tabdat-keys tabdat)))
		       (run-name   (db:get-value-by-header run-record runs-header "runname"))
		       (col-name   (conc (string-intersperse key-vals "\n") "\n" run-name))

         hide-clean: hide-clean)
        #f)))


(define (dashboard:get-runs-hash tabdat)
  (let* ((access-mode       (dboard:tabdat-access-mode tabdat))
         (last-runs-update  0);;(dboard:tabdat-last-runs-update tabdat))

	 (runs-dat     (rmt:get-runs-by-patt (dboard:tabdat-keys tabdat) "%" #f #f #f #f last-runs-update))
	 (runs-header  (vector-ref runs-dat 0)) ;; 0 is header, 1 is list of records
         (runs         (vector-ref runs-dat 1))
	 (run-id       (dboard:tabdat-curr-run-id tabdat))
         (runs-hash    (let ((ht (make-hash-table)))
			 (for-each (lambda (run)
				     (hash-table-set! ht (db:get-value-by-header run runs-header "id") run))
				   runs) ht)))
    runs-hash))
         

(define (dashboard:runs-summary-updater commondat tabdat tb cell-lookup run-matrix)
  ;; (if (dashboard:database-changed? commondat tabdat context-key: 'runs-summary-rundat)
  (dashboard:do-update-rundat tabdat) ;; )
  (dboard:runs-summary-control-panel-updater tabdat)
  (let* ((last-runs-update  (dboard:tabdat-last-runs-update tabdat))


	 (runs-dat     (rmt:get-runs-by-patt (dboard:tabdat-keys tabdat) "%" #f #f #f #f last-runs-update))
	 (runs-header  (vector-ref runs-dat 0)) ;; 0 is header, 1 is list of records
         (runs         (vector-ref runs-dat 1))
	 (run-id       (dboard:tabdat-curr-run-id tabdat))
         (runs-hash (dashboard:get-runs-hash tabdat))
         ;; (runs-hash    (let ((ht (make-hash-table)))
	 ;;        	 (for-each (lambda (run)
	 ;;        		     (hash-table-set! ht (db:get-value-by-header run runs-header "id") run))

;;
;; General info about the run(s) and megatest area
(define (dashboard:summary commondat tabdat #!key (tab-num #f))
  (let* ((rawconfig        (read-config (conc *toppath* "/megatest.config") #f #f)) ;; changed to #f since I want #{} to be expanded by [system ...] to NOT be expanded. WAS: 'return-string)))
	 (changed          #f))
    (iup:vbox
     (iup:split
      #:value 300
      (iup:frame 
       #:title "General Info"
       (iup:vbox
	(iup:hbox
	 (iup:label "Area Path")
	 (iup:textbox #:value *toppath* #:expand "HORIZONTAL"))
	(iup:hbox 

(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 (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")

;; 
(define (dashboard:runs-summary commondat tabdat #!key (tab-num #f))
  (let* ((update-mutex (dboard:commondat-update-mutex commondat))
	 (tb      (iup:treebox
		   #:value 0
		   #:name "Runs"
		   #:expand "YES"
		   #:addexpanded "YES"
		   #:selection-cb
		   (lambda (obj id state)
		     (debug:catch-and-dump
		      (lambda ()
			;; (print "obj: " obj ", id: " id ", state: " state)
			(let* ((run-path (tree:node->path obj id))
			       (run-id   (tree-path->run-id tabdat (cdr run-path))))

      run-matrix)
     (dboard:make-controls commondat tabdat extra-widget: runs-summary-control-panel))))

;;======================================================================
;; R U N S 
;;======================================================================

(define (dboard:squarify toggles size)
  (let loop ((hed (car toggles))
	     (tal (cdr toggles))
	     (cur '())
	     (res '()))
    (let* ((ovrflo (>= (length cur) size))
	   (newcur (if ovrflo
		       (list hed)
		       (cons hed cur)))
	   (newres (if ovrflo
		       (cons cur res)
		       res)))
      (if (null? tal)
	  (if ovrflo
	      newres
	      (cons newcur res))
	  (loop (car tal)(cdr tal) newcur newres)))))

(define (dboard:make-controls commondat tabdat #!key (extra-widget #f) )
  (let ((btn-fontsz  (dboard:tabdat-runs-btn-fontsz tabdat)))
    (iup:hbox
     (iup:vbox
      (iup:frame 
       #:title "filter test and items"
       (iup:vbox

				    (update-search commondat tabdat "test-name" val))
				  "make-controls")))
	 (iup:hbox
	  (iup:button "Quit"      #:action (lambda (obj)
					     (exit))
		      #:expand "NO" #:size "40x15")
	  (iup:button "Refresh"   #:action (lambda (obj)
                                             (dboard:tabdat-last-data-update-set! tabdat 0)
                                             (dboard:tabdat-last-runs-update-set! tabdat 0)
                                             (dboard:tabdat-run-update-times-set! tabdat (make-hash-table))
                                             (dboard:tabdat-last-test-dat-set!    tabdat (make-hash-table))
                                             (dboard:tabdat-allruns-set!          tabdat '())
                                             (dboard:tabdat-allruns-by-id-set!    tabdat (make-hash-table))
                                             (dboard:tabdat-done-runs-set!        tabdat '())
                                             (dboard:tabdat-not-done-runs-set!    tabdat '())
                                             (dboard:tabdat-view-changed-set!     tabdat #t)
                                             (dboard:commondat-please-update-set! commondat #t)
					     (mark-for-update tabdat))
		      #:expand "NO" #:size "40x15")
	  (iup:button "Collapse"  #:action (lambda (obj)
					     (debug:catch-and-dump 
					      (lambda ()
						(let ((myname (iup:attribute obj "TITLE")))
						  (if (equal? myname "Collapse")

					 #:action (lambda (obj val index lbstate)
						    (set! *tests-sort-reverse* index)
						    (mark-for-update tabdat))))
		(default-cmd (car (list-ref *tests-sort-type-index* *tests-sort-reverse*))))
                
	   (iuplistbox-fill-list sort-lb cmds-list selected-item: default-cmd)
	   
	   ;; (set! hide-empty (iup:button "HideEmpty"
	   ;; 				;; #:expand HORIZONTAL"
	   ;; 				#:expand "NO" #:size "80x15"
	   ;; 				#:action (lambda (obj)
	   ;; 					   (dboard:tabdat-hide-empty-runs-set! tabdat (not (dboard:tabdat-hide-empty-runs tabdat)))
	   ;; 					   (iup:attribute-set! obj "TITLE" (if (dboard:tabdat-hide-empty-runs tabdat) "+HideE" "-HideE"))
	   ;; 					   (mark-for-update tabdat))))
	   (set! hide (iup:button "Hide"
				  #:expand "NO" #:size "40x15" ;; #:expand "HORIZONTAL"
				  #:action (lambda (obj)
					     (dboard:tabdat-hide-not-hide-set! tabdat #t) ;; (not (dboard:tabdat-hide-not-hide tabdat)))
					     ;; (iup:attribute-set! obj "TITLE" (if (dboard:tabdat-hide-not-hide tabdat) "HideTests" "NotHide"))
					     (iup:attribute-set! hide "BGCOLOR" sel-color)
					     (iup:attribute-set! show "BGCOLOR" nonsel-color)

            (iup:hbox)) ;; empty widget

        

        
        )))






     (let* ((status-toggles (map (lambda (status)
				   (iup:toggle (conc status)
					       #:fontsize 8 ;; btn-fontsz ;; "10"
					       ;; #:expand "HORIZONTAL"
					       #:action   (lambda (obj val)
							    (mark-for-update tabdat)
							    (if (eq? val 1)
								(hash-table-set! (dboard:tabdat-status-ignore-hash tabdat) status #t)
								(hash-table-delete! (dboard:tabdat-status-ignore-hash tabdat) status))
							    (set-bg-on-filter commondat tabdat))))
				 (map cadr *common:std-statuses*))) ;; '("PASS" "FAIL" "WARN" "CHECK" "WAIVED" "STUCK/DEAD" "n/a" "SKIP")))


	    (state-toggles  (map (lambda (state)
				   (iup:toggle (conc state)
					       #:fontsize 8 ;; btn-fontsz
					       ;; #:expand "HORIZONTAL"
					       #:action   (lambda (obj val)
							    (mark-for-update tabdat)
							    (if (eq? val 1)
								(hash-table-set! (dboard:tabdat-state-ignore-hash tabdat) state #t)
								(hash-table-delete! (dboard:tabdat-state-ignore-hash tabdat) state))
							    (set-bg-on-filter commondat tabdat))))
				 (map cadr *common:std-states*))) ;; '("RUNNING" "COMPLETED" "INCOMPLETE" "LAUNCHED" "NOT_STARTED" "KILLED" "DELETED")))
	    (num-toggle-cols (inexact->exact (round (/ (max (length status-toggles)(length state-toggles)) 3)))))
       (iup:vbox
	(iup:hbox
	 (iup:frame
	  #:title "states"
	  (apply
	   iup:hbox
	   (map (lambda (colgrp)
		  (apply iup:vbox colgrp))
		(dboard:squarify state-toggles 3))))
	 (iup:frame
	  #:title "statuses"
	  (apply
	   iup:hbox
	   (map (lambda (colgrp)
		  (apply iup:vbox colgrp))
		(dboard:squarify status-toggles 3)))))
	;; 
	;; (iup:frame 
	;; 	#:title "state/status filter"
	;; 	(iup:vbox
	;; 	 (apply
	;; 	  iup:hbox
	;; 	  (map
	;; 	   (lambda (status-toggle state-toggle)
	;; 	     (iup:vbox
	;; 	      status-toggle
	;; 	      state-toggle))
	;; 	   status-toggles state-toggles))

	;; horizontal slider was here
	
	)))))

(define (dashboard:runs-horizontal-slider tabdat )
  (iup:valuator #:valuechanged_cb (lambda (obj)
				    (let ((val (inexact->exact (round (/ (string->number (iup:attribute obj "VALUE")) 10))))
					  (oldmax   (string->number (iup:attribute obj "MAX")))
					  (maxruns  (dboard:tabdat-tot-runs tabdat)))
				      (dboard:tabdat-start-run-offset-set! tabdat val)
				      (mark-for-update tabdat)
				      (debug:print 6 *default-log-port* "(dboard:tabdat-start-run-offset tabdat) " (dboard:tabdat-start-run-offset tabdat) " maxruns: " maxruns ", val: " val " oldmax: " oldmax)
				      (iup:attribute-set! obj "MAX" (* maxruns 10))))
		#:expand "HORIZONTAL"
		#:max (* 10 (max (hash-table-size (dboard:tabdat-allruns-by-id tabdat)) 10))
		#:min 0
		#:step 0.01))




(define (dashboard:popup-menu  run-id test-id target runname test-name testpatt item-test-path test-info)
  (iup:menu 
   (iup:menu-item
    "Test Control Panel"
    #:action
    (lambda (obj)

    #:action
    (lambda (obj)
      ;; (rmt:test-set-state-status-by-id run-id test-id "KILLREQ" #f #f)
      (common:run-a-command
       (conc "megatest -set-state-status KILLREQ,n/a -target " target
             " -runname " runname
             " -testpatt " item-test-path 
             " -state RUNNING,REMOTEHOSTSTART,LAUNCHED,NOT_STARTED"))))

   
   (iup:menu-item
    "Run"
    (iup:menu              
     (iup:menu-item
      (conc "Rerun " testpatt)

      "Kill Complete Run"
      #:action
      (lambda (obj)
        (common:run-a-command
         (conc "megatest -set-state-status KILLREQ,n/a -target " target
               " -runname " runname
               " -testpatt % "
               "  -state RUNNING,REMOTEHOSTSTART,LAUNCHED,NOT_STARTED"))))
     (iup:menu-item 
      "Delete Run Data"
      #:action
      (lambda (obj)
        (common:run-a-command
         (conc "megatest -remove-runs -target " target
               " -runname " runname
               " -testpatt % "
               "  -keep-records"))))))
   (iup:menu-item
    "Test"
    (iup:menu 
     (iup:menu-item
      (conc "Rerun " item-test-path)
      #:action
      (lambda (obj)

      (lambda (obj)
        ;; (rmt:test-set-state-status-by-id run-id test-id "KILLREQ" #f #f)
	(common:run-a-command
	 (conc "megatest -set-state-status KILLREQ,n/a -target " target
               " -runname " runname
	       " -testpatt " item-test-path 
	       " -state RUNNING,REMOTEHOSTSTART,LAUNCHED"))))
     (iup:menu-item
      (conc "Delete data : " item-test-path)
      #:action
      (lambda (obj)
	(common:run-a-command
	 (conc "megatest -remove-runs -target " target
               " -runname " runname
	       " -testpatt " item-test-path 
	       " -keep-records"))))
     (iup:menu-item
      (conc "Clean "item-test-path)
      #:action
      (lambda (obj)
	(common:run-a-command
	 (conc "megatest -remove-runs -target " target
               " -runname " runname

			(iup:label) ;; (iup:valuator)
			(apply iup:vbox 
			       (map (lambda (x)		
				      (let ((res (iup:hbox #:expand "HORIZONTAL"
							   (iup:label x #:size (conc 40 btn-height) #:fontsize btn-fontsz #:expand "NO") ;; "HORIZONTAL")
							   (iup:textbox #:size (conc 35 btn-height) #:fontsize btn-fontsz #:value "%" #:expand "NO" ;; "HORIZONTAL"
									#:action (lambda (obj unk val)
										   ;; each field (field name is "x" var) live updates
										   ;; the search filter as it is typed
										   (dboard:tabdat-target-set! runs-dat #f) ;; ensure the fields text boxes are used and not the info from the tree
										   (mark-for-update runs-dat)
										   (update-search commondat runs-dat x val))))))
					(set! i (+ i 1))
					res))
				    keynames)))))
    (let loop ((testnum  0)
	       (res      '()))

    (iup:show
     (iup:dialog 
      #:title (conc "Megatest dashboard " (current-user-name) ":" *toppath*)
      #:menu (dcommon:main-menu)
      (let* ((runs-view (iup:vbox
			 (iup:split
			  #:orientation "VERTICAL" ;; "HORIZONTAL"
			  #:value 100
			  (dboard:runs-tree-browser commondat runs-dat)
			  (iup:split
			   #:value 100
			   ;; left most block, including row names
			   (apply iup:vbox lftlst)
			   ;; right hand block, including cells
			   (iup:vbox
			    #:expand "YES"
			    ;; the header
			    (apply iup:hbox (reverse hdrlst))
			    (apply iup:hbox (reverse bdylst))
			    (dashboard:runs-horizontal-slider runs-dat))))
			 controls
			 ))
	     (views-cfgdat (common:load-views-config))
	     (additional-tabnames '())
	     (tab-start-num       5)   ;; DON'T FORGET TO UPDATE THIS WHEN CHANGING THE STANDARD TABS BELOW
	     ;; (data (dboard:tabdat-init (make-d:data)))
	     (additional-views 	;; process views-dat

					       "tabchangepos"))
			  (dashboard:summary commondat stats-dat tab-num: 0)
			  runs-view
			  (dashboard:runs-summary commondat onerun-dat tab-num: 2)
			  ;; (dashboard:new-view db data new-view-dat tab-num: 3)
			  (dashboard:run-controls commondat runcontrols-dat tab-num: 3)
			  (dashboard:run-times commondat runtimes-dat tab-num: 4)
			  ;; (dashboard:runs-summary commondat onerun-dat tab-num: 4)
			  additional-views)))
	;; (set! (iup:callback tabs tabchange-cb:) (lambda (a b c)(print "SWITCHED TO TAB: " a " " b " " c)))
	(iup:attribute-set! tabs "TABTITLE0" "Summary")
	(iup:attribute-set! tabs "TABTITLE1" "Runs")
	(iup:attribute-set! tabs "TABTITLE2" "Run Summary")
	(iup:attribute-set! tabs "TABTITLE3" "Run Control")
	(iup:attribute-set! tabs "TABTITLE4" "Run Times")

   (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 (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)

		   (db:test-get-event_time (hash-table-ref testsdat (car b))))))))))

;; run times tab data updater
;;
(define (dashboard:run-times-tab-run-data-updater commondat tabdat tab-num)
  (let* ((access-mode      (dboard:tabdat-access-mode tabdat))
         (last-runs-update (dboard:tabdat-last-runs-update tabdat))


         (runs-dat      (rmt:get-runs-by-patt (dboard:tabdat-keys tabdat) "%" #f #f #f #f last-runs-update))
	 (runs-header   (vector-ref runs-dat 0)) ;; 0 is header, 1 is list of records
	 (runs-hash     (let ((ht (make-hash-table)))
			  (for-each (lambda (run)
				      (hash-table-set! ht (db:get-value-by-header run runs-header "id") run))
				    (vector-ref runs-dat 1))
			  ht))
	 (run-ids       (sort (filter number? (hash-table-keys runs-hash))

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

(define (main)
  (let ((mtdb-path (conc *toppath* "/megatest.db"))) ;; 
    (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 (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  (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 (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
   ((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 (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 (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 (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 common:get-db-tmp-area)








(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  (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 (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
                   (if (and (configf:lookup *configdat* "setup" "use-wal")
                            (string-match "^/tmp/.*" fname)) ;; this is a file in /tmp
                       (sqlite3:execute db "PRAGMA journal_mode=WAL;")
                       (debug:print 2 *default-log-port* "Creating " fname " in NON-WAL mode."))
                   (initproc db)))
             (if (not readyexists)
                 (begin
                   (common:simple-file-release-lock lockfname)
                   (with-output-to-file
                       readyfname
                     (lambda ()

;; ;; 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     (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)

;;     (dbr:dbstruct-olddb-set!  dbstruct olddb)
;;     ;;; (mutex-unlock! *rundb-mutex*) ;;; why did we need a mutex on opening db's?
;;     (db:sync-tables db:sync-tests-only *megatest-db* db)
;;     db))

;; 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     (common:file-exists? dbpath))
	       (tmpdbfname   (conc dbpath "/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))

                (dbr:dbstruct-read-only-set! dbstruct #t)))
          (dbr:dbstruct-mtdb-set!   dbstruct mtdb)
          (dbr:dbstruct-tmpdb-set!  dbstruct tmpdb)
          (dbr:dbstruct-dbstack-set! dbstruct (make-stack)) ;; BB: why a stack?  Why would the number of db's be indeterminate?  Is this a legacy of 1.db 2.db .. ?
          (stack-push! (dbr:dbstruct-dbstack dbstruct) tmpdb) ;; olddb is already a (cons db path)
          (dbr:dbstruct-refndb-set! dbstruct refndb)
          ;;	    (mutex-unlock! *rundb-mutex*)
          (if (and  (or (not dbfexists)
			(and modtimedelta
			     (> modtimedelta 10))) ;; if db in tmp is over ten seconds older than the file in MTRA then do a sync back
		    do-sync)
	      (begin
		(debug:print 4 *default-log-port* "filling db " (db:dbdat-get-path tmpdb) " with data \n    from " (db:dbdat-get-path mtdb) " mod time delta: " modtimedelta)
		(db:sync-tables (db:sync-all-tables-list dbstruct) #f mtdb refndb tmpdb)
                (debug:print-info 13 *default-log-port* "db:sync-all-tables-list done.")
                )
	      (debug:print 4 *default-log-port* " db, " (db:dbdat-get-path tmpdb) " already exists or fresh enough, not propogating data from\n     " (db:dbdat-get-path mtdb) " mod time delta: " modtimedelta) )
	  ;; (db:multi-db-sync dbstruct 'old2new))  ;; migrate data from megatest.db automatically
          tmpdb))))

;; Make the dbstruct, setup up auxillary db's and call for main db at least once
;;
;; called in http-transport and replicated in rmt.scm for *local* access. 
;;
(define (db:setup do-sync #!key (areapath #f))
  ;;

  (cond
   (*dbstruct-db* *dbstruct-db*);; TODO: when multiple areas are supported, this optimization will be a hazard
   (else ;;(common:on-homehost?)
    (debug:print-info 13 *default-log-port* "db:setup entered (first time, not cached.)")
    (let* ((dbstruct (make-dbr:dbstruct)))
      (when (not *toppath*)
        (debug:print-info 13 *default-log-port* "in db:setup, *toppath* not set; calling launch:setup")
        (launch:setup areapath: areapath))
      (debug:print-info 13 *default-log-port* "Begin db:open-db")
      (db:open-db dbstruct areapath: areapath do-sync: do-sync)
      (debug:print-info 13 *default-log-port* "Done db:open-db")
      (set! *dbstruct-db* dbstruct)
      ;;(debug:print-info 13 *default-log-port* "new dbstruct = "(dbr:dbstruct->alist dbstruct))
      dbstruct))))
   ;; (else
   ;;  (debug:print 0 *default-log-port* "ERROR: attempt to open database when not on homehost. Exiting. Homehost: " (common:get-homehost))
   ;;  (exit 1))))

;; 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     (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)))
	       tdbs)
          (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)))))

	 '("final_logf"     #f)
	 '("logdat"         #f)
	 '("run_duration"   #f)
	 '("comment"        #f)
	 '("event_time"     #f)
	 '("fail_count"     #f)
	 '("pass_count"     #f)
	 '("archived"       #f)
         '("last_update"    #f))
  (list "test_steps"
	 '("id"             #f)
	 '("test_id"        #f)
	 '("stepname"       #f)
	 '("state"          #f)
	 '("status"         #f)
	 '("event_time"     #f)
	 '("comment"        #f)
	 '("logfile"        #f)
         '("last_update"    #f))
   (list "test_data"
	 '("id"             #f)
	 '("test_id"        #f)
	 '("category"       #f)
	 '("variable"       #f)
	 '("value"          #f)
	 '("expected"       #f)
	 '("tol"            #f)
	 '("units"          #f)
	 '("comment"        #f)
	 '("status"         #f)
	 '("type"           #f)
         '("last_update"    #f))))

;; needs db to get keys, this is for syncing all tables
;;
(define (db:sync-main-list dbstruct)
  (let ((keys  (db:get-keys dbstruct)))
    (list
     (list "keys"
	   '("id"        #f)
	   '("fieldname" #f)
	   '("fieldtype" #f))
     (list "metadat" '("var" #f) '("val" #f))
     (append (list "runs" 
		   '("id"  #f))
	     (map (lambda (k)(list k #f))
		  (append keys
			  (list "runname" "state" "status" "owner" "event_time" "comment" "fail_count" "pass_count" "contour" "last_update"))))
     (list "test_meta"
	   '("id"             #f)
	   '("testname"       #f)
	   '("owner"          #f)
	   '("description"    #f)
	   '("reviewed"       #f)
	   '("iterated"       #f)

	 (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 (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))
     (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))
		 (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 " 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)

                              count  INTEGER,
                              last_update INTEGER DEFAULT (strftime('%s','now')))")
  (sqlite3:execute maindb "CREATE TRIGGER  IF NOT EXISTS update_run_stats_trigger AFTER UPDATE ON run_stats
                             FOR EACH ROW
                               BEGIN 
                                 UPDATE run_stats SET last_update=(strftime('%s','now'))
                                   WHERE id=old.id;
                               END;")
  (sqlite3:execute maindb "CREATE TABLE IF NOT EXISTS test_rundat (
                              id           INTEGER PRIMARY KEY,
                              test_id      INTEGER,
                              update_time  TIMESTAMP,
                              cpuload      INTEGER DEFAULT -1,
                              diskfree     INTEGER DEFAULT -1,
                              diskusage    INTGER DEFAULT -1,
                              run_duration INTEGER DEFAULT 0);"))

(define (db:adj-target db)
  (let ((fields    (configf:get-section *configdat* "fields"))
	(field-num 0))
    ;; because we will be refreshing the keys table it is best to clear it here
    (sqlite3:execute db "DELETE FROM keys;")
    (for-each
     (lambda (field)
       (let ((column (car field))
	     (spec   (cadr field)))
	 (handle-exceptions
	  exn
	  (if (string-match ".*duplicate.*" ((condition-property-accessor 'exn 'message) exn))
	      (debug:print 0 *default-log-port* "Target field " column " already exists in the runs table")
	      (db:general-sqlite-error-dump exn "alter table runs ..." #f "none"))
	  ;; Add the column if needed
	  (sqlite3:execute
	   db
	   (conc "ALTER TABLE runs ADD COLUMN " column " " spec)))
	 ;; correct the entry in the keys column
	 (sqlite3:execute
	  db
	  "INSERT INTO keys (id,fieldname,fieldtype) VALUES (?,?,?);"
	  field-num column spec)
	 ;; fill in blanks (not allowed as it would be part of the path
	 (sqlite3:execute
	  db
	  (conc "UPDATE runs SET " column "='x' WHERE " column "='';"))
	 (set! field-num (+ field-num 1))))
     fields)))
  
(define *global-db-store* (make-hash-table))

(define (db:get-access-mode)
  (if (args:get-arg "-use-db-cache") 'cached 'rmt))

;; Add db direct
;;

;; 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 (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 (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))








	 ;; 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))

(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))
	 (res         (db:sync-tables (db:sync-all-tables-list dbstruct) last-update tmpdb refndb mtdb)))
    (stack-push! (dbr:dbstruct-dbstack dbstruct) tmpdb)
    res))


;;;; run-ids
;;    if #f use *db-local-sync* : or 'local-sync-flags
;;    if #t use timestamps      : or 'timestamps
;;
;;  NB// no-sync-db is the db handle, not a flag!
;;

(define (db:sync-to-megatest.db dbstruct #!key (no-sync-db #f)) 
  (let* ((start-time         (current-seconds))

	 (last-full-update   (if no-sync-db
				 (db:no-sync-get/default no-sync-db "LAST_FULL_UPDATE" 0)
				 0))
	 (full-sync-needed   (> (- start-time last-full-update) 3600)) ;; every hour do a full sync















	 (last-update        (if full-sync-needed


				 0

				 (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 (or sync-needed ;; don't sync if a sync already occurred in the past 6 seconds
				     full-sync-needed)
				 (begin
				   (if no-sync-db
				       (begin

					 (if full-sync-needed (db:no-sync-set no-sync-db "LAST_FULL_UPDATE" start-time))
					 (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))
	(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)))

;; (define open-run-close 
(define open-run-close open-run-close-exception-handling)
		;;	   open-run-close-no-exception-handling
;;			   open-run-close-exception-handling)
;;)

(define (db:initialize-main-db dbdat)
  (when (not *configinfo*)
           (launch:setup)) ;; added because Elena was getting stack dump because *configinfo* below was #f.
  (let* ((configdat (car *configinfo*))  ;; tut tut, global warning...
	 (keys     (keys:config-get-fields configdat))
	 (havekeys (> (length keys) 0))
	 (keystr   (keys->keystr keys))
	 (fieldstr (keys:make-key/field-string configdat))
	 (db       (db:dbdat-get-db dbdat)))
    (for-each (lambda (key)
		(let ((keyn key))
		  (if (member (string-downcase keyn)
			      (list "runname" "state" "status" "owner" "event_time" "comment" "fail_count"
				    "pass_count" "contour"))
		      (begin
			(print "ERROR: your key cannot be named " keyn " as this conflicts with the same named field in the runs table, you must remove your megatest.db and <linktree>/.db before trying again.")
			(exit 1)))))
	      keys)
    (sqlite3:with-transaction
     db
     (lambda ()

	 db 
	 (conc
	  "SELECT d.id,d.archive_area_name,disk_path,last_df,last_df_time FROM archive_disks AS d
             INNER JOIN archive_blocks AS b ON d.id=b.archive_disk_id
             WHERE b.id IN (" (string-intersperse (map conc res) ",") ") AND
         last_df > ?;")
	 dneeded))
    (stack-push! (dbr:dbstruct-dbstack dbstruct) dbdat)
    blocks))
    
;; returns id of the record, register a disk allocated to archiving and record it's last known
;; available space
;;
(define (db:archive-register-disk dbstruct bdisk-name bdisk-path df)
  (let* ((dbdat        (db:get-db dbstruct)) ;; archive tables are in main.db

                                          WHERE archive_disk_id=? AND disk_path=?;"
				   bdisk-id archive-path du))
	  res)
	(begin
	  (sqlite3:execute db "INSERT OR REPLACE INTO archive_blocks (archive_disk_id,disk_path,last_du)
                                                        VALUES (?,?,?);"
			   bdisk-id archive-path (or du 0))
	  (db:archive-register-block-name dbstruct bdisk-id archive-path du: du)))
    (stack-push! (dbr:dbstruct-dbstack dbstruct) dbdat)
    res))


;; The "archived" field in tests is overloaded; 0 = not archived, > 0 archived in block with given id
;;
(define (db:test-set-archive-block-id dbstruct run-id test-id archive-block-id)
  (db:with-db
   dbstruct

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

(define (open-logging-db)
  (let* ((dbpath    (conc (if *toppath* (conc *toppath* "/") "") "logging.db")) ;; fname)
	 (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 (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

	       "DELETE FROM tests WHERE state='DELETED';"
	       ;; delete all tests that have no run
	       "DELETE FROM tests WHERE run_id NOT IN (SELECT DISTINCT id FROM runs);"
	       ;; delete all runs that are state='deleted'
	       "DELETE FROM runs WHERE state='deleted';"
	       ;; delete empty runs
	       "DELETE FROM runs WHERE id NOT IN (SELECT DISTINCT r.id FROM runs AS r INNER JOIN tests AS t ON t.run_id=r.id);"
	       ;; remove orphaned test_rundat entries
	       "DELETE FROM test_rundat where test_id NOT IN (SELECT id FROM tests);"
	       ;; 
	       "DELETE FROM test_steps WHERE test_id NOT IN (SELECT id FROM tests);"
	       ))))
    ;; (db:delay-if-busy dbdat)
    (sqlite3:with-transaction 
     db
     (lambda ()
       (sqlite3:for-each-row (lambda (tot)
			       (debug:print-info 0 *default-log-port* "Records count before clean: " tot))

	      (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-exists (common:file-exists? dbname))
	 (db     (sqlite3:open-database dbname)))
    (sqlite3:set-busy-handler! db (make-busy-timeout 136000))
    (if (not db-exists)
	(begin
	  (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));")
	  (sqlite3:execute db "PRAGMA journal_mode=WAL;")))
    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)
  (mutex-lock! *db-access-mutex*)
  (let ((res (if db-in
                 db-in
                 (let ((db (db:open-no-sync-db)))
                   (set! *no-sync-db* db)
                   db))))
    (mutex-unlock! *db-access-mutex*)
    res))

(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))

;; transaction protected lock aquisition
;; either:
;;    fails    returns  (#f . lock-creation-time)
;;    succeeds (returns (#t . lock-creation-time)
;; use (db:no-sync-del! db keyname) to release the lock
;;
(define (db:no-sync-get-lock db-in keyname)
  (let ((db (db:no-sync-db db-in)))
    (sqlite3:with-transaction
     db
     (lambda ()
       (handle-exceptions
	   exn
	   (let ((lock-time (current-seconds)))
	     (sqlite3:execute db "INSERT INTO no_sync_metadat (var,val) VALUES(?,?);" keyname lock-time)
	     `(#t . ,lock-time))
	 `(#f . ,(sqlite3:first-result db "SELECT val FROM no_sync_metadat WHERE var=?;" keyname)))))))



;; 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)
            (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")  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
			   (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))
                                                        (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))
                                                        (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)))
						      "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))))))))
         (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))) 
		 (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) ;; 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)) ;; 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 "") ;; 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: " (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))

     (if (string-match (regexp "^[/~]+.*") outputfile) ;; full path?
	 outputfile
	 (begin
	   (debug:print 0 *default-log-port* "WARNING: path given, " outputfile " is relative, prefixing with current directory")
	   (conc (current-directory) "/" outputfile)))
     results)
    ;; brutal clean up
    (stack-push! (dbr:dbstruct-dbstack dbstruct) dbdat)
    (system "rm -rf tempdir")))

;; (db:extract-ods-file db "outputfile.ods" '(("sysname" "%")("fsname" "%")("datapath" "%")) "%")

(use regex typed-records matchable)

(declare (unit dcommon))

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

;; 				  )
;; 				(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)))

(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

;; Section to table
(define (dcommon:section-matrix rawconfig sectionname varcolname valcolname #!key (title #f))
  (let* ((curr-row-num    1)
         (key-vals        (configf:section-vars rawconfig sectionname))
         (section-matrix  (iup:matrix
                           #:alignment1 "ALEFT"
                           ;; #:expand "YES" ;; "HORIZONTAL"
                           #:numcol 1
                           #:numlin (length key-vals)
                           #:numcol-visible 1
                           #:numlin-visible (min 10 (length key-vals))
			   #:scrollbar "YES")))
    (iup:attribute-set! section-matrix "0:0" varcolname)
    (iup:attribute-set! section-matrix "0:1" valcolname)

				      (hash-table-set! tests-draw-state 'scalef (+ scalef
										   (if (> step 0)
										       (* scalef 0.01)
										       (* scalef -0.01))))
				      (if the-cnv
					  (dashboard:draw-tests the-cnv last-xadj last-yadj tests-draw-state sorted-testnames test-records))
				      ))
		       ;; #:size "250x250"
		       #:expand "YES"
		       #:scrollbar "YES"
		       #:posx "0.5"
		       #:posy "0.5"
		       #:button-cb (lambda (obj btn pressed x y status)
				     ;; (print "obj: " obj ", pressed " pressed ", status " status)
					; (print "canvas-origin: " (canvas-origin the-cnv))

html/megatest.html : megatest.lyx
	elyxer megatest.lyx html/megatest.html
	fossil add html/*

megatest.pdf : megatest.lyx
	lyx -e pdf2 megatest.lyx

pkts.pdf : pkts.dot
	dot -Tpdf pkts.dot -o pkts.pdf

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 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>

<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. <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". <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. <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. <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. <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. <span class="green">[DONE]</span>
</p>
</li>
<li>
<p>
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. <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 <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 <span class="red">[Replaced by mtutil]</span>
</p>
</li>
<li>
<p>
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)  <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 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 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 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 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 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 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.


 * 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 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 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

  "RUNNING" [
     shape="record";
     label="{RUNNING|{n/a|<here> PASS |<here> FAIL}}";
  ]

  "COMPLETED" [
      shape="record";
      label = "{COMPLETED|{PASS | SKIP | WAIVED | FAIL | CHECK| ABORT}}";
  ]


"RUNNING" -> "COMPLETED";
"RUNNING" -> "INCOMPLETE" [label="test dead for > 24hrs"];

digraph megatest_pkts {
  ranksep=0.05
  // rankdir=LR

node [shape=box,style=filled];
  
  "SENSORS" [ label = "{ Sensor Processing | { file | git | fossil | script }}"
	      shape = "record"; ];
  
  "RUNS"    [ label = "{ Runs Processing | { launch | clean | re-run | archive } | { dispatcher }}";
	      shape = "record"; ];

  "WORK"    [ label = "{ Work Items | { start task | task competed }}";
	      shape = "record"; ];

  "USERREQ" [ label = "{ User Requests (Unix and Web) | { launch | clean | re-run | archive }}";
	      shape = "record"; ];

  "MTAREA1" [ label = "{ Megatest Area 1 | { parallel job\nmanagement | test\nmanagement | data\nrollup }}";
	      shape = "record"; ];
  
  "MTAREA2" [ label = "{ Megatest Area 2 | { parallel job\nmanagement | test\nmanagement | data\nrollup }}";
	      shape = "record"; ];
  
  "MTAREA3" [ label = "More Megatest Areas ... ";
	      shape = "record"; ];

  "PGDB"    [ label = "postgres database";
	      shape = "cylinder"; ];

  "WEBAPP"  [ label = "{ Web View | { Runs | Contours | Control | Time View }}";
	      shape = "record"; ];

  // "WEBCTRL" [ label = "{ Web View \n(control) }";
  //	      shape = "record"; ];
  
  "SENSORS" -> "SPKTS";
  "RUNS"    -> "run pkts";
  "run pkts" -> "RUNS";
  "WORK"    -> "work pkts";
  "work pkts" -> "RUNS";
  "USERREQ" -> "user request pkts";
  "SPKTS"   -> "RUNS";
  "user request pkts" -> "RUNS";
  "RUNS"    -> "MTAREA1" -> "PGDB";
  "RUNS"    -> "MTAREA2" -> "PGDB";
  "RUNS"    -> "MTAREA3" -> "PGDB";
  "PGDB"    -> "WEBAPP";
  // "WEBCTRL" -> "run pkts";
  
  subgraph cluster_pkts {
    label="Packets";
    "SPKTS" [ label = "Sensor Packets" ];
    "run pkts";
    "work pkts";
    "user request pkts";
  }
}

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. [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". [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. [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. [green]#[DONE]#
. Read/writes to db for processes on homehost will go direct to /tmp
  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. [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. [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 [green]#[DONE]#
. Server polls tasks table for next action
.. 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)))))))))

 ;; To build patch:
 ;;;;;;;;;;;;;;;;;;;;;;;;;
 ;; ldd /p/foundry/env/pkgs/megatest/1.64/19/bin/.11/mtest
 ;;        linux-vdso.so.1 =>  (0x00002aaaaaaab000)
 ;;        libchicken.so.7 => /p/foundry/env/pkgs/megatest/1.64/chicken-4.10.0//lib/libchicken.so.7 (0x00002aaaaaaad000)
 ;;        libm.so.6 => /lib64/libm.so.6 (0x00002aaaab0a6000)
 ;;        libdl.so.2 => /lib64/libdl.so.2 (0x00002aaaab31f000)
 ;;        libc.so.6 => /lib64/libc.so.6 (0x00002aaaab523000)
 ;;        /lib64/ld-linux-x86-64.so.2 (0x0000555555554000)
 ;;
 ;;  /p/foundry/env/pkgs/megatest/1.64/chicken-4.10.0/bin/csc -s emergency-patch-3.scm
 ;;


 ;; to test patch:
 ;;;;;;;;;;;;;;;;;;;;;;;;;
 ;; in .megatestrc, add:
 ;; (if (and (> megatest-version 1.64)
 ;;         (< megatest-version 1.6421))
 ;;   (begin
 ;;      (load "/p/foundry/env/pkgs/megatest/1.64/19/share/epatch-1.so")
 ;;      (load "/p/foundry/env/pkgs/megatest/1.64/19/share/epatch-2.so"))) 
 ;;


 ;; to productize patch:
 ;;;;;;;;;;;;;;;;;;;;;;;;;
 ;; 
(use directory-utils regex)

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

;; Given a run id start a server process    ### NOTE ### > file 2>&1 
;; if the run-id is zero and the target-host is set 
;; try running on that host
;;   incidental: rotate logs in logs/ dir.
;;
(define  (server:run areapath) ;; areapath is *toppath* for a given testsuite area
  (let* ((curr-host   (get-host-name))
         ;; (attempt-in-progress (server:start-attempted? areapath))
         ;; (dot-server-url (server:check-if-running areapath))
	 (curr-ip     (server:get-best-guess-address curr-host))
	 (curr-pid    (current-process-id))
	 (homehost    (common:get-homehost)) ;; configf:lookup *configdat* "server" "homehost" ))
	 (target-host (car homehost))
	 (testsuite   (common:get-testsuite-name))
	 (logfile     (conc areapath "/logs/server.log")) ;; -" curr-pid "-" target-host ".log"))
	 (cmdln (conc (common:get-megatest-exe)
		      " -server " (or target-host "-") (if (equal? (configf:lookup *configdat* "server" "daemonize") "yes")
							   " -daemonize "
							   "")
		      ;; " -log " logfile
		      " -m testsuite:" testsuite)) ;; (conc " >> " logfile " 2>&1 &")))))
	 (log-rotate  (make-thread common:rotate-logs  "server run, rotate logs thread"))
         (load-limit  (configf:lookup-number *configdat* "server" "load-limit" default: 0.9)))
    ;; we want the remote server to start in *toppath* so push there
    (push-directory areapath)
    (debug:print 0 *default-log-port* "INFO: Trying to start server (" cmdln ") ...")
    (thread-start! log-rotate)
    
    ;; host.domain.tld match host?
    (if (and target-host 
	     ;; look at target host, is it host.domain.tld or ip address and does it 
	     ;; match current ip or hostname
	     (not (string-match (conc "("curr-host "|" curr-host"\\..*)") target-host))
	     (not (equal? curr-ip target-host)))
	(begin
	  (debug:print-info 0 *default-log-port* "Starting server on " target-host ", logfile is " logfile)
	  (setenv "TARGETHOST" target-host)))
      
    (setenv "TARGETHOST_LOGF" logfile)
    (common:wait-for-normalized-load load-limit " delaying server start due to load" remote-host: (get-environment-variable "TARGETHOST")) ;; do not try starting servers on an already overloaded machine, just wait forever
    (system (conc "nbfake " cmdln))
    (unsetenv "TARGETHOST_LOGF")
    (if (get-environment-variable "TARGETHOST")(unsetenv "TARGETHOST"))
    (thread-join! log-rotate)
    (pop-directory)))

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

(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 (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,

;; 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.

;;  strftime('%m/%d/%Y %H:%M:%S','now','localtime')

(use srfi-1 posix regex srfi-69 directory-utils)


(declare (unit ezsteps))
(declare (uses db))
(declare (uses common))
(declare (uses items))
(declare (uses runconfig))
;; (declare (uses sdb))

	 (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 (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 (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 (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 (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 (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
     ((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 (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)))))

 grep configf:lookup *.scm | sed 's/^.*:lookup//; s/^-number//; s/^ //' | grep -v '^\(section\|test-conf\|tconfig\|testconfig\|dat\|config\|views-cfgdat\)' | perl -pe 's/^\s*(\*configdat\*|configdat|mtconf)//; s/^\s+//; s/\).*$//; s/"//g' | awk '{print $1,$2}' | sort | grep -v section | sort | uniq

(define (gutils:colors-similar? color1 color2)
  (let* ((c1 (map string->number (string-split color1)))
	 (c2 (map string->number (string-split color2)))
	 (delta (map (lambda (a b)(abs (- a b))) c1 c2)))
    (null? (filter (lambda (x)(> x 3)) delta))))

(define gutils:colors
  '((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))
       ((SKIP)        (list (gutils:get-color-spec 'SKIP) status))
       ((ABORT)       (list "198 36 166" status))
       (else (list "253 33 49" status))))
    ((ARCHIVED)
     (case (string->symbol status)
       ((PASS)        (list "70  170 73" status))
       ((WARN WAIVED) (list "200 130 13" status))
       ((SKIP)        (list (gutils:get-color-spec 'SKIP) status))
       (else (list "180 33 49" status))))
    ;;      (if (equal? status "PASS")
    ;;	  '("70 249 73" "PASS")
    ;;	  (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 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 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))

;; 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.

(require-extension (srfi 18) extras tcp s11n)

(use  srfi-1 posix regex regex-case srfi-69 hostinfo md5 message-digest posix-extras)


(use spiffy uri-common intarweb http-client spiffy-request-vars intarweb spiffy-directory-listing)

;; Configurations for server
(tcp-buffer-size 2048)
(max-connections 2048) 

	 (hostname        (get-host-name))
	 (ipaddrstr       (let ((ipstr (if (string=? "-" hostn)
					   ;; (string-intersperse (map number->string (u8vector->list (hostname->ip hostname))) ".")
					   (server:get-best-guess-address hostname)
					   #f)))
			    (if ipstr ipstr hostn))) ;; hostname))) 
	 (start-port      (portlogger:open-run-close portlogger:find-port))
	 (link-tree-path  (common:get-linktree))
	 (tmp-area        (common:get-db-tmp-area))
	 (start-file      (conc tmp-area "/.server-start")))
    (debug:print-info 0 *default-log-port* "portlogger recommended port: " start-port)
    ;; set some parameters for the server
    (root-path     (if link-tree-path 
		       link-tree-path
		       (current-directory))) ;; WARNING: SECURITY HOLE. FIX ASAP!
    (handle-directory spiffy-directory-listing)
    (handle-exception (lambda (exn chain)
			(signal (make-composite-condition
				 (make-property-condition 

				   (send-response body: "hey there!\n"
						  headers: '((content-type text/plain))))
				  ((equal? (uri-path (request-uri (current-request))) 
					   '(/ "hey"))
				   (send-response body: "hey there!\n"
						  headers: '((content-type text/plain))))
				  (else (continue))))))))
    (with-output-to-file start-file (lambda ()(print (current-process-id))))
    (http-transport:try-start-server ipaddrstr start-port)))

;; This is recursively run by http-transport:run until sucessful
;;
(define (http-transport:try-start-server ipaddrstr portnum)
  (let ((config-hostname (configf:lookup *configdat* "server" "hostname"))
	(config-use-proxy (equal? (configf:lookup *configdat* "client" "use-http_proxy") "yes")))
    (if (not config-use-proxy)
	(determine-proxy (constantly #f)))
    (debug:print-info 0 *default-log-port* "http-transport:try-start-server time=" (seconds->time-string (current-seconds)) " ipaddrsstr=" ipaddrstr " portnum=" portnum " config-hostname=" config-hostname)
    (handle-exceptions
	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 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)))

					      exn
					      (let ((call-chain (get-call-chain))
						    (msg        ((condition-property-accessor 'exn 'message) exn)))
						(set! success #f)
						(debug:print 0 *default-log-port* "WARNING: failure in with-input-from-request to " fullurl ".")
						(debug:print 0 *default-log-port* " message: " msg)
						(debug:print 0 *default-log-port* " cmd: " cmd " params: " params)
                                                (debug:print 0 *default-log-port* " call-chain: " call-chain)
						(if runremote
						    (remote-conndat-set! runremote #f))
						;; Killing associated server to allow clean retry.")
						;; (tasks:kill-server-run-id run-id)  ;; better to kill the server in the logic that called this routine?
						(mutex-unlock! *http-mutex*)
					     ;;; (signal (make-composite-condition
					     ;;;          (make-property-condition 'commfail 'message "failed to connect to server")))
					     ;;; "communications failed"
						(db:obj->string #f))
					    (with-input-from-request ;; was dat
					     fullurl 
					     (list (cons 'key (or *server-id* "thekey"))
						   (cons 'cmd cmd)
						   (cons 'params sparams))
					     read-string))
					  transport: 'http)
                                         0)) ;; added this speculatively
			      ;; Shouldn't this be a call to the managed call-all-connections stuff above?
			      (close-all-connections!)

(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)
            ;;(close-idle-connections!)
	    #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))

;; used and to shutdown after sometime if it is not.
;;
(define (http-transport:keep-running) 
  ;; if none running or if > 20 seconds since 
  ;; server last used then start shutdown
  ;; This thread waits for the server to come alive
  (debug:print-info 0 *default-log-port* "Starting the sync-back, keep alive thread in server")
  (let* ((tmp-area          (common:get-db-tmp-area))
	 (started-file      (conc tmp-area "/.server-started"))
	 (server-start-time (current-seconds))
	 (server-info (let loop ((start-time (current-seconds))
				 (changed    #t)
				 (last-sdat  "not this"))
                        (let ((sdat #f))
			  (thread-sleep! 0.01)
			  (debug:print-info 0 *default-log-port* "Waiting for server alive signature")
                          (mutex-lock! *heartbeat-mutex*)

				      (exit))
				    (loop start-time
					  (equal? sdat last-sdat)
					  sdat)))))))
         (iface       (car server-info))
         (port        (cadr server-info))
         (last-access 0)
	 (server-timeout (server:expiration-timeout))
	 (server-going  #f)
	 (server-log-file (args:get-arg "-log"))) ;; always set when we are a server

    (with-output-to-file started-file (lambda ()(print (current-process-id))))

    (let loop ((count         0)
	       (server-state 'available)
	       (bad-sync-count 0)
	       (start-time     (current-milliseconds)))
      ;; Use this opportunity to sync the tmp db to megatest.db
      (if (not server-going) ;; *dbstruct-db* 
	  (begin
	    (debug:print 0 *default-log-port* "SERVER: dbprep")
	    (set! *dbstruct-db*  (db:setup #t)) ;;  run-id))
	    (set! server-going #t)
	    (debug:print 0 *default-log-port* "SERVER: running, megatest version: " (common:get-full-version)) ;; NOTE: the server is NOT yet marked as running in the log. We do that in the keep-running routine.
	    (thread-start! *watchdog*)))
      
      ;; when things go wrong we don't want to be doing the various queries too often
      ;; so we strive to run this stuff only every four seconds or so.
      (let* ((sync-time (- (current-milliseconds) start-time))

	  (begin
	    (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)))





	(cond
         ((and *server-run*
	       (> (+ last-access server-timeout)
		  (current-seconds)))

          (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)))))
          (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)
  ;; check that a server start is in progress, pause or exit if so
  (let* ((tmp-area            (common:get-db-tmp-area))
	 (server-start        (conc tmp-area "/.server-start"))
	 (server-started      (conc tmp-area "/.server-started"))
	 (start-time          (common:lazy-modification-time server-start))
	 (started-time        (common:lazy-modification-time server-started))
	 (server-starting     (< start-time started-time)) ;; if start-time is less than started-time then a server is still starting
	 (start-time-old      (> (- (current-seconds) start-time) 5))
         (cleanup-proc        (lambda (msg)
                                (let* ((serv-fname      (conc "server-" (current-process-id) "-" (get-host-name) ".log"))
                                       (full-serv-fname (conc *toppath* "/logs/" serv-fname))
                                       (new-serv-fname  (conc *toppath* "/logs/" "defunct-" serv-fname)))
                                  (debug:print 0 *default-log-port* msg)
                                  (if (common:file-exists? full-serv-fname)
                                      (system (conc "sleep 1;mv -f " full-serv-fname " " new-serv-fname))
                                      (debug:print 0 *default-log-port* "INFO: cannot move " full-serv-fname " to " new-serv-fname))
                                  (exit)))))
    (if (and (not start-time-old) ;; last server start try was less than five seconds ago
	     (not server-starting))
	(begin


	  (cleanup-proc "NOT starting server, there is either a recently started server or a server in process of starting")
	  (exit)))
    ;; 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 (> num-alive 3)
          (begin

            (cleanup-proc (conc "ERROR: Aborting server start because there are already " num-alive " possible servers either running or starting up"))
            (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"))
           (th3 (make-thread (lambda ()
                               (debug:print-info 0 *default-log-port* "Server monitor thread started")
                               (http-transport:keep-running)
                               "Keep running"))))
      (thread-start! th2)
      (thread-sleep! 0.25) ;; give the server time to settle before starting the keep-running monitor.
      (thread-start! th3)
      (set! *didsomething* #t)
      (thread-join! th2)
      (exit))))

;; (define (http-transport:server-signal-handler signum)
;;   (signal-mask! signum)
;;   (handle-exceptions
;;    exn
;;    (debug:print 0 *default-log-port* " ... exiting ...")
;;    (let ((th1 (make-thread (lambda ()
;; 			     (thread-sleep! 1))
;; 			   "eat response"))
;; 	 (th2 (make-thread (lambda ()
;; 			     (debug:print-error 0 *default-log-port* "Received ^C, attempting clean exit. Please be patient and wait a few seconds before hitting ^C again.")
;; 			     (thread-sleep! 3) ;; give the flush three seconds to do it's stuff
;; 			     (debug:print 0 *default-log-port* "       Done.")
;; 			     (exit 4))
;; 			   "exit on ^C timer")))
;;      (thread-start! th2)
;;      (thread-start! th1)
;;      (thread-join! th2))))

;;======================================================================
;; web pages
;;======================================================================

(define (http-transport:main-page)
  (let ((linkpath (root-path)))

			   item))
		     items))
    (set! itemstable (map (lambda (item)
			    (if (procedure? (cadr item))
				(list (car item)((cadr item)))  ;; evaluate the proc
				item))
			  itemstable))
    (if (and have-items  (null? items))     (debug:print 0 *default-log-port* "WARNING:[items] section in testconfig but no entries defined"))
    (if (and have-itable (null? itemstable))(debug:print 0 *default-log-port* "WARNNG:[itemstable] section in testconfig but no entries defined"))
    (if (or (not (null? items))(not (null? itemstable)))
	(append (item-assoc->item-list items)
		(item-table->item-list itemstable))
	'(()))))

;; (pp (item-assoc->item-list itemdat))

;;  implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
;;  PURPOSE.
;;======================================================================

(define-inline (keys->valslots keys) ;; => ?,?,? ....
  (string-intersperse (map (lambda (x) "?") keys) ","))

;; (define-inline (keys->key/field keys . additional)
;;   (string-join (map (lambda (k)(conc k " TEXT"))
;; 		    (append keys additional)) ","))

(define-inline (item-list->path itemdat)
  (if (list? itemdat)
      (string-intersperse  (map cadr itemdat) "/")
      ""))

	   (list key targ))
	 keys targtweaked)))

;;======================================================================
;; config file related routines
;;======================================================================

(define keys:config-get-fields common:get-fields)
(define (keys:make-key/field-string confdat)
  (let ((fields (configf:get-section confdat "fields")))
    (string-join
     (map (lambda (field)(conc (car field) " " (cadr field)))
	  fields)
     ",")))

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

(declare (unit launch))
(declare (uses common))
(declare (uses configf))
(declare (uses db))




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

;;======================================================================
;; ezsteps

(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 (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    (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 (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 (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 (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 (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 (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")))
         (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)))
               (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))
	(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)
                        (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)

	       (mt-bindir-path (assoc/default 'mt-bindir-path cmdinfo))
	       (keys      #f)
	       (keyvals   #f)
	       (fullrunscript (if (not runscript)
                                  #f
                                  (if (substring-index "/" runscript)
                                      runscript ;; use unadultered if contains slashes
                                      (let ((fulln (conc work-area "/" runscript)))
	                                  (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))
	  
	  ;; immediated set some key variables from CMDINFO data, yes, these will be set again below ...
	  ;;
	  (setenv "MT_TESTSUITENAME" areaname)
	  (setenv "MT_RUN_AREA_HOME" top-path)
	  (set! *toppath* top-path)
	  (setenv "MT_TEST_RUN_DIR"  work-area)

	;;    ;; On NFS it can be slow and unreliable to get needed startup information.
	;;    ;;  i. Check if we are on the homehost, if so, proceed
	;;    ;; ii. Check if host and port passed in via CMDINFO are valid and if
	;;    ;;     possible use them.
	;;    (let ((bestadrs (server:get-best-guess-address (get-host-name)))
	;;    	(needcare #f))
	;;      (if (equal? homehost bestadrs) ;; we are likely on the homehost
	;;    	(debug:print-info 0 *default-log-port* "test " test-name " appears to be running on the homehost " homehost)
	;;    	(let ((host-port (if serverurl (string-split serverurl ":") #f)))
	;;    	  (if (not *runremote*)(set! *runremote* (make-remote))) ;; init *runremote*
	;;    	  (if (string? homehost)
	;;    	      (if (and host-port
	;;    		       (> (length host-port) 1))
	;;    		  (let* ((host      (car host-port))
        ;;                           (port      (cadr host-port))
        ;;                           (start-res (http-transport:client-connect host port))
        ;;                           (ping-res  (rmt:login-no-auto-client-setup start-res)))
	;;    		    (if (and start-res
	;;    			     ping-res)
	;;    			;; (begin ;; let ((url  (http-transport:server-dat-make-url start-res)))
	;;    			(begin
	;;    			  (remote-conndat-set! *runremote* start-res)
	;;    			  ;; (remote-server-url-set! *runremote* url)
	;;    			  ;; (if (server:ping url)
	;;    			  (debug:print-info 0 *default-log-port* "connected to " host ":" port " using CMDINFO data."))
	;;    			(begin
	;;    			  (debug:print-info 0 *default-log-port* "have CMDINFO data but failed to connect to " host ":" port)
	;;    			  (set! *runremote* #f))
	;;    			  ;; (remote-conndat-set! *runremote* #f))
	;;    			))
	;;    		  (begin
	;;    		    (set! *runremote* #f)
	;;    		    (debug:print-info 0 *default-log-port* (if host-port
	;;    							       (conc "received invalid host-port information " host-port)
	;;    							       "no host-port information received"))
	;;    		    ;; potential for bad situation if simultaneous starting of hundreds of jobs on servers, set needcare.
	;;    		    (set! needcare #t)))
	;;    	      (begin
	;;    		(set! *runremote* #f)
	;;    		(debug:print-info 0 *default-log-port* "received no homehost information. Please report this to support as it should not happen.")
	;;    		(set! needcare #t)))))
	;;      (if needcare  ;; due to very slow NFS we will do a brute force mkdir to ensure that the directory inode it truly available on this host
	;;    	(let ((logdir (conc top-path "/logs"))) ;; we'll try to create this directory
	;;    	  (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 (common:directory-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)))))

          (if (not (string=?  (common:real-path work-area)(common:real-path (current-directory))))
              (begin
                (debug:print 0 *default-log-port*
                             "INFO: we are expecting to be in directory " work-area "\n"
                             "     but we are actually in the directory " (current-directory) "\n"
                             "     doing another change dir.")
                (change-directory work-area)))
          
	  ;; spot check that the files in testpath are available. Too often NFS delays cause problems here.
	  (let ((files      (glob (conc testpath "/*")))
		(bad-files '()))
	    (for-each
	     (lambda (fullname)
	       (let* ((fname (pathname-strip-directory fullname))
                      (targn (conc work-area "/" fname)))
		 (if (not (file-exists? targn))
		     (set! bad-files (cons fname bad-files)))))
	     files)
	    (if (not (null? bad-files))
                (begin
                  (debug:print 0 *default-log-port* "INFO: test data from " testpath " not copied properly or filesystem problems causing data to not be found. Re-running the copy command.")
                  (debug:print 0 *default-log-port* "INFO: missing files from " work-area ": " (string-intersperse bad-files ", "))
                  (launch:test-copy testpath work-area))))

          ;; one more time, change to the work-area directory
          (change-directory work-area)
          
	  (launch:setup) ;; should be properly in the top-path now
	  (set! tconfigreg (tests:get-all))
	  (let ((sighand (lambda (signum)
			   ;; (signal-mask! signum) ;; to mask or not? seems to cause issues in exiting
			   (if (eq? signum signal/stop)
			       (debug:print-error 0 *default-log-port* "attempt to STOP process. Exiting."))
			   (set! *time-to-exit* #t)