Megatest

mofiles/mtconfigf.o : mtconfigf/mtconfigf.scm
mofiles/ulex.o      : ulex/ulex.scm
mofiles/mutils.o    : mutils/mutils.scm
mofiles/cookie.o    : stml2/cookie.scm
mofiles/stml2.o     : stml2/stml2.scm

# for the modularized stuff

mofiles/commonmod.o : megatest-fossil-hash.scm mofiles/stml2.o \
                      mofiles/mtargs.o mofiles/pkts.o mofiles/mtconfigf.o \
                      mofiles/processmod.o
mofiles/pgdbmod.o   : mofiles/commonmod.o
mofiles/dbmod.o     : mofiles/commonmod.o mofiles/keysmod.o \
                      mofiles/tasksmod.o mofiles/odsmod.o
mofiles/tasksmod.o  : mofiles/commonmod.o mofiles/pgdbmod.o
mofiles/rmtmod.o    : mofiles/commonmod.o \
                      mofiles/apimod.o mofiles/ulex.o mofiles/itemsmod.o
mofiles/apimod.o    : mofiles/dbmod.o mofiles/commonmod.o mofiles/servermod.o
mofiles/runsmod.o   : mofiles/testsmod.o mofiles/mtmod.o
mofiles/mtmod.o     : mofiles/mtconfigf.o mofiles/rmtmod.o mofiles/tasksmod.o \
                      mofiles/dbmod.o mofiles/pgdbmod.o mofiles/launchmod.o \
                      mofiles/subrunmod.o
mofiles/servermod.o : mofiles/commonmod.o mofiles/dbmod.o
mofiles/testsmod.o  : mofiles/servermod.o mofiles/dbmod.o
mofiles/launchmod.o : mofiles/subrunmod.o mofiles/testsmod.o




# Removed from megamod.o dep:   mofiles/ftail.o
mofiles/megamod.o   : \
   mofiles/rmtmod.o \
   mofiles/commonmod.o \
   mofiles/apimod.o \
   mofiles/archivemod.o \

# base64 dot-locking \
#             csv-xml z3

#  "(define (toplevel-command . a) #f)"
# if egrep 'version.*3.0' $(shell dirname $(shell dirname $(shell which csi)))/lib/chicken/7/readline.setup-info;then \

readline-fix.scm :
	if [[ $(shell chicken-status | grep readline | awk '{print $4}' | cut -d. -f1) -gt 3 ]];then \
	   echo "(define *use-new-readline* #f)" > readline-fix.scm; \
	else \
	   echo "(define *use-new-readline* #t)" > readline-fix.scm;\
	fi

altdb.scm :
	echo ";; optional alternate db setup" > altdb.scm
	echo "(define *available-db* (make-hash-table))" >> altdb.scm
	if  csi -ne '(use mysql-client)';then \
           echo "(use mysql-client)(hash-table-set! *available-db* 'mysql #t)" >> altdb.scm; \
	fi
	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 dashboard-tests.o dashboard-context-menu.o db.o dcommon.o ezsteps.o filedb.o genexample.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 dashboard-tests.o dashboard-context-menu.o db.o dcommon.o ezsteps.o filedb.o genexample.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

(declare (uses commonmod))
(declare (uses dbmod))
(declare (uses servermod))

(module apimod
	*
	
(import scheme chicken data-structures extras)
(import (prefix sqlite3 sqlite3:) posix typed-records srfi-18 srfi-69 format ports srfi-1 matchable
	s11n z3 (prefix base64 base64:) regex stack srfi-13
	irregex)
(import commonmod)
(import dbmod)
(import servermod)

;; (use (prefix ulex ulex:))

;; These are called by the server on recipt of /api calls

                   ;; STEPS
                   ((teststep-set-status!)         (apply db:teststep-set-status! dbstruct params))
                   ((delete-steps-for-test!)        (apply db:delete-steps-for-test! 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

(declare (unit archivemod))
(declare (uses commonmod))

(module archivemod
	*
	
(import scheme chicken data-structures extras)
(import (prefix sqlite3 sqlite3:) posix typed-records srfi-18 srfi-69 format ports srfi-1 matchable)
(import commonmod)
;; (use (prefix ulex ulex:))

(include "common_records.scm")


)

(declare (unit clientmod))
(declare (uses commonmod))

(module clientmod
	*
	
(import scheme chicken data-structures extras)
(import (prefix sqlite3 sqlite3:) posix typed-records srfi-18 srfi-69 format ports srfi-1 matchable)
(import commonmod)
;; (use (prefix ulex ulex:))

(include "common_records.scm")


)

;; 
;;     You should have received a copy of the GNU General Public License
;;     along with Megatest.  If not, see <http://www.gnu.org/licenses/>.
;;
;;======================================================================

;; (use trace)
(use typed-records)

;; moved to commonmod

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

(declare (unit commonmod))
(declare (uses mtargs))
;; (declare (uses stml2))
(declare (uses mtconfigf))
(declare (uses ulex))
(declare (uses pkts))
(module commonmod
	*
	
(import scheme chicken data-structures extras)

















(use (prefix sqlite3 sqlite3:) posix typed-records srfi-18
     srfi-1 files format srfi-13 matchable 
     srfi-69 ports
     (prefix base64 base64:)
     regex-case regex hostinfo srfi-4
     (prefix dbi dbi:)
     stack
     md5 
     message-digest
     z3
     directory-utils

     sparse-vectors)

(import pkts)
(import ulex)
(import	(prefix mtconfigf configf:))
(import	(prefix mtargs args:))

(include "common_records.scm")
(include "megatest-fossil-hash.scm")
(include "megatest-version.scm")

		      (get-environment-variable "MT_DEBUG_MODE"))))
    (set! *verbosity* (debug:calc-verbosity debugstr verbose quiet))
    (debug:check-verbosity *verbosity* debugstr)
    ;; if we were handed a bad verbosity rule then we will override it with 1 and continue
    (if (not *verbosity*)(set! *verbosity* 1))
    (if (or dmode                                            ;; (args:get-arg "-debug")
	    (not (get-environment-variable "MT_DEBUG_MODE")))
	(setenv "MT_DEBUG_MODE" (if (list? *verbosity*)
				    (string-intersperse (map conc *verbosity*) ",")
				    (conc *verbosity*))))))
  
(define (debug:print n e . params)
  (if (debug:debug-mode n)
      (with-output-to-port (or e (current-error-port))
	(lambda ()

;;
(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)))))

;;    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)))
;;======================================================================
;; T I M E   A N D   D A T E
;;======================================================================

;; Convert strings like "5s 2h 3m" => 60x60x2 + 3x60 + 5

  (if (or (substring-index "!" key) (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 \":\" or starting with \"!\"")
      (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"))

;; put any changed environment variables back to how they were - TODO - turn this into some sort of with-
(define (common:set-vars-back all-vars)
  (for-each
     (lambda (vardat)
       (let ((var (car vardat))
	     (val (cdr vardat)))
	 (if (not (equal? (get-environment-variable var) val))
	     (handle-exceptions
	      exn
	      (debug:print-error 0 *default-log-port* "Failed to set " var " to " val)
	      (setenv var val)))))
     all-vars))
  
  ;; 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/*")))))

    ;; TEST DATA
    test-data-rollup
    csv->test-data

    ;; MISC
    sync-inmem->db

    ;; TESTMETA
    testmeta-add-record
    testmeta-update-field

    ;; TASKS
    tasks-add
    tasks-set-state-given-param-key
    ))

;;======================================================================
;; ALLDATA
;;======================================================================
;;
;; attempt to consolidate a bunch of global information into one struct to toss around
(defstruct alldat
  ;; misc
  (denoise           (make-hash-table))
  (areapath          #f) ;; i.e. toppath
  (mtconfig          #f)
  (log-port          #f)
  (areadat           #f) ;; i.e. runremote
  (rmt-mutex         (make-mutex))
  (db-sync-mutex     (make-mutex))
  (db-with-db-mutex  (make-mutex))
  (read-only-queries api:read-only-queries)
  (write-queries     api:write-queries)
  (max-api-process-requests 0)
  (api-process-request-count 0)

(use posix-extras pathname-expand files)

;; this plugs a hole in posix-extras in recent chicken versions > 4.9)
(let-values (( (chicken-release-number chicken-major-version)
               (apply values
                      (map string->number
                           (take
                            (string-split (chicken-version) ".")
                            2)))))
  (let ((resolve-pathname-broken?
         (or (> chicken-release-number 4)
             (and (eq? 4 chicken-release-number) (> chicken-major-version 9)))))
    (if resolve-pathname-broken?
        (define ##sys#expand-home-path pathname-expand))))
      
(define (realpath x) (resolve-pathname  (pathname-expand (or x "/dev/null")) ))


(define (common:get-this-exe-fullpath #!key (argv (argv)))
  (let* ((this-script
          (cond
           ((and (> (length argv) 2)
                 (string-match "^(.*/csi|csi)$" (car argv))
                 (string-match "^-(s|ss|sx|script)$" (cadr argv)))

				 (begin
				   (mutex-unlock! *homehost-mutex*)
				   (debug:print 0 *default-log-port* "ERROR: ["(common:human-time)"] 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
					 (mutex-unlock! *homehost-mutex*)
					 (car (common:get-homehost))))

;; get values from cached info from dropping file in logs dir
;;  e.g. key is host and dtype is normalized-load
;;
(define (common:get-cached-info key dtype #!key (age 5))
  (let* ((fullpath (conc *toppath* "/logs/" key "-" dtype ".log")))
    (if (and (file-exists? fullpath)
	     (file-read-access? fullpath))
	(handle-exceptions
	 exn
	 #f
	 (debug:print 2 *default-log-port* "reading file " fullpath)
	 (let ((real-age (- (current-seconds)(file-change-time fullpath)))) 
	   (if (< real-age age)
	       (with-input-from-file fullpath read)

     (lambda (disk-num)
       (let* ((dirpath    (cadr (assoc disk-num disks)))
	      (freespc    (cond
			   ((not (directory? dirpath))
			    (if (common:low-noise-print 300 "disks not a dir " disk-num)
				(debug:print 0 *default-log-port* "WARNING: disk " disk-num " at path \"" dirpath "\" is not a directory - ignoring it."))
			    -1)
			   ((not (file-write-access? dirpath))
			    (if (common:low-noise-print 300 "disks not writeable " disk-num)
				(debug:print 0 *default-log-port* "WARNING: disk " disk-num " at path \"" dirpath "\" is not writeable - ignoring it."))
			    -1)
			   ((not (eq? (string-ref dirpath 0) #\/))
			    (if (common:low-noise-print 300 "disks not a proper path " disk-num)
				(debug:print 0 *default-log-port* "WARNING: disk " disk-num " at path \"" dirpath "\" is not a fully qualified path - ignoring it."))
			    -1)
			   (else
			    (get-df dirpath))))
	      (free-inodes (cond
			   ((not (directory? dirpath))
			    (if (common:low-noise-print 300 "disks not a dir " disk-num)
				(debug:print 0 *default-log-port* "WARNING: disk " disk-num " at path \"" dirpath "\" is not a directory - ignoring it."))
			    -1)
			   ((not (file-write-access? dirpath))
			    (if (common:low-noise-print 300 "disks not writeable " disk-num)
				(debug:print 0 *default-log-port* "WARNING: disk " disk-num " at path \"" dirpath "\" is not writeable - ignoring it."))
			    -1)
			   ((not (eq? (string-ref dirpath 0) #\/))
			    (if (common:low-noise-print 300 "disks not a proper path " disk-num)
				(debug:print 0 *default-log-port* "WARNING: disk " disk-num " at path \"" dirpath "\" is not a fully qualified path - ignoring it."))
			    -1)

	(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

                      (if (string-match "^MT_.*" (car x))
                          #f
                          x))
                    envvars))))

(define (common:with-orig-env proc)
  (let  ((current-env (get-environment-variables)))
    (for-each (lambda (x) (unsetenv (car x)))             current-env)
    (for-each (lambda (x) (setenv (car x) (cdr x))) *common:orig-env*)
    (let ((rv (cond
               ((string? proc)(system proc))
               (proc          (proc)))))
      (for-each (lambda (x) (unsetenv (car x))) *common:orig-env*)
      (for-each (lambda (x) (setenv (car x) (cdr x))) current-env)
      rv)))

(define (common:without-vars proc . var-patts)
  (let ((vars (make-hash-table)))
    (for-each
     (lambda (vardat) ;; each env var
       (for-each
	(lambda (var-patt)
	  (if (string-match var-patt (car vardat))
	      (let ((var (car vardat))
		    (val (cdr vardat)))
		(hash-table-set! vars var val)
		(unsetenv var))))
	var-patts))
     (get-environment-variables))
    (cond
     ((string? proc)(system proc))
     (proc          (proc)))
    (hash-table-for-each
     vars
     (lambda (var val)
       (setenv var val)))
    vars))


;;======================================================================
;; C O L O R S
;;======================================================================
      

     (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)))

           (map (lambda (env-pair)
                  (let* ((env-var     (car env-pair))
                         (new-val     (let ((tmp (cdr env-pair)))
                                        (if (list? tmp) (car tmp) tmp)))
                         (current-val (get-environment-variable env-var))
                         (restore-thunk
                          (cond
                           ((not current-val) (lambda () (unsetenv env-var)))
                           ((not (string? new-val)) #f)
                           ((eq? current-val new-val) #f)
                           (else 
                            (lambda () (setenv env-var current-val))))))
                    ;;(when (not (string? new-val))
                    ;;    (debug:print 0 *default-log-port* " PROBLEM: not a string: "new-val"\n from env-alist:\n"delta-env-alist)
                    ;;    (pp delta-env-alist)
                    ;;    (exit 1))
                        
                    
                    (cond
                     ((not new-val)  ;; modify env here
                      (unsetenv env-var))
                     ((string? new-val)
                      (setenv env-var new-val)))
                    restore-thunk))
                delta-env-alist))))
    (let ((rv (thunk)))
      (for-each (lambda (x) (x)) restore-thunks) ;; restore env to original state
      rv)))

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

(declare (unit configfmod))
;; (declare (uses commonmod))

(module configfmod
	*
	
(import scheme chicken data-structures extras)
(import (prefix sqlite3 sqlite3:) posix typed-records srfi-18
	srfi-69 format ports srfi-1 matchable regex)
;; (import commonmod)
;; (use (prefix ulex ulex:))

(include "common_records.scm")

(define (configf:lookup cfgdat section var)
  (if (hash-table? cfgdat)

(declare (uses ulex.import))
(declare (uses cookie.import))
(declare (uses stml2.import))
(declare (uses ducttape-lib.import))
(declare (uses mtconfigf.import))

(declare (uses gutilsmod))

(declare (uses megamod))
(declare (uses commonmod))
(declare (uses rmtmod))
(declare (uses runsmod))
(declare (uses dbmod))
(declare (uses testsmod))
(declare (uses tasksmod))
(declare (uses dcommonmod))
(declare (uses tasksmod))


(import gutilsmod)

(import megamod)
(import commonmod)
(import rmtmod)
(import runsmod)
(import dbmod)
(import testsmod)
(import tasksmod)

       (iup:hbox
        (iup:button "Hide All" #:action (lambda (obj)
                                          (for-each (lambda (graph-cell)
                                                      (let* ((graph-dat   (hash-table-ref (dboard:tabdat-graph-cell-table tabdat) graph-cell)))
                                                        (dboard:graph-dat-flag-set! graph-dat #f)))
                                                    (hash-table-keys (dboard:tabdat-graph-cell-table tabdat)))))))
      ))))






















































































;;======================================================================
;; R U N
;;======================================================================
;;
;; display and manage a single run at a time

(define (make-dashboard-buttons commondat) ;;  runs-sum-dat new-view-dat)
  (let* ((stats-dat       (dboard:tabdat-make-data))
	 (runs-dat        (dboard:tabdat-make-data))
	 (onerun-dat      (dboard:tabdat-make-data)) ;; name for run-summary structure 
	 (runcontrols-dat (dboard:tabdat-make-data))
	 (runtimes-dat    (dboard:tabdat-make-data))

	 (nruns           (dboard:tabdat-numruns runs-dat))
	 (ntests          (dboard:tabdat-num-tests runs-dat))
	 (keynames        (dboard:tabdat-dbkeys runs-dat))
	 (nkeys           (length keynames))
	 (runsvec         (make-vector nruns))
	 (header          (make-vector nruns))
	 (lftcol          (make-vector ntests))

			  (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")

	;; (iup:attribute-set! tabs "TABTITLE3" "New View")
	;; (iup:attribute-set! tabs "TABTITLE4" "Run Control")

	;; set the tab names for user added tabs
	(for-each
	 (lambda (tab-info)
	   (iup:attribute-set! tabs (conc "TABTITLE" (car tab-info)) (cdr tab-info)))
	 additional-tabnames)
	
	(iup:attribute-set! tabs "BGCOLOR" "190 190 190")
	;; make the iup tabs object available (for changing color for example)
	(dboard:commondat-hide-not-hide-tabs-set! commondat tabs)
	;; now set up the tabdat lookup
	(dboard:common-set-tabdat! commondat 0 stats-dat)
	(dboard:common-set-tabdat! commondat 1 runs-dat)
	(dboard:common-set-tabdat! commondat 2 onerun-dat)
	(dboard:common-set-tabdat! commondat 3 runcontrols-dat)
	(dboard:common-set-tabdat! commondat 4 runtimes-dat)


	(iup:vbox
	 tabs
	 ;; controls
	 ))))
    (vector keycol lftcol header runsvec)))

(declare (uses tasksmod))
;; (declare (uses servermod))
;; (declare (uses testsmod))

(module dbmod
	*
	
(import scheme chicken data-structures extras)
(import (prefix sqlite3 sqlite3:) posix typed-records srfi-18
	srfi-69 format ports srfi-1 matchable stack regex
	srfi-13 stack s11n
	(prefix base64 base64:)





	z3
	csv csv-xml
	directory-utils
	call-with-environment-variables)

(import commonmod)
(import keysmod)
(import files)
(import tasksmod)
(import odsmod)
;; (import testsmod)
(import (prefix mtargs args:))
(import (prefix mtconfigf configf:))
;; (import servermod)

	tconf
	(let ((test-dirs (tests:get-tests-search-path *configdat*)))
	  (let loop ((hed (car test-dirs))
		     (tal (cdr test-dirs)))
	    ;; Setting MT_LINKTREE here is almost certainly unnecessary. 
	    (let ((tconfig-file (conc hed "/" test-name "/testconfig")))
	      (if (and (common:file-exists? tconfig-file)
		       (file-read-access? tconfig-file))
		  (let ((link-tree-path (common:get-linktree)) ;; (configf:lookup *configdat* "setup" "linktree"))
			(old-link-tree  (get-environment-variable "MT_LINKTREE")))
		    (if link-tree-path (setenv "MT_LINKTREE" link-tree-path))
		    (let ((newtcfg (configf:read-config tconfig-file #f #f))) ;; NOTE: Does NOT run [system ...]
		      (hash-table-set! *testconfigs* test-name newtcfg)
		      (if old-link-tree 
			  (setenv "MT_LINKTREE" old-link-tree)
			  (unsetenv "MT_LINKTREE"))
		      newtcfg))
		  (if (null? tal)
		      (begin
			(debug:print-error 0 *default-log-port* "No readable testconfig found for " test-name)
			#f)
		      (loop (car tal)(cdr tal))))))))))

;; 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)))

							 
               (mtdbmodtime  (if mtdbexists (common:lazy-sqlite-db-modification-time (conc *toppath* "/megatest.db"))   #f))
	        		 (tmpdbmodtime (if dbfexists  (common:lazy-sqlite-db-modification-time tmpdbfname) #f)) 
               (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)) ; moving this before db:open-megatest-db is called. if wal mode is on -WAL and -shm file get created with causing the  tmpdbmodtime timestamp always greater than mtdbmodtime
	       ;(tmpdbmodtime (if dbfexists  (common:lazy-sqlite-db-modification-time tmpdbfname) #f)) 
					;if wal mode is on -WAL and -shm file get created when db:open-megatest-db is called. modtimedelta will always be < 10 so db in tmp not get synced
          ;(tmpdbmodtime (if dbfexists (db:get-last-update-time (car tmpdb)) #f))    
          ;(fmt (file-modification-time tmpdbfname))
	       (modtimedelta (and mtdbmodtime tmpdbmodtime (- mtdbmodtime tmpdbmodtime))))

         (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)
    (if (and dbexists (not write-access))
	(set! *db-write-access* #f))
    (cons db dbpath)))

;; sync run to disk if touched
;;

;;
(define (db:repair-db dbdat #!key (numtries 1))
  (let* ((dbpath   (db:dbdat-get-path        dbdat))
	 (dbdir    (pathname-directory       dbpath))
	 (fname    (pathname-strip-directory dbpath)))
    (debug:print-info 0 *default-log-port* "Checking db " dbpath " for errors.")
    (cond
     ((not (file-write-access? dbdir))
      (debug:print 0 *default-log-port* "WARNING: can't write to " dbdir ", can't fix " fname)
      #f)

     ;; handle special cases, megatest.db and monitor.db
     ;; 
     ;;  NOPE: apply this same approach to all db files
     ;;

    ((not (sqlite3:database? (db:dbdat-get-db fromdb)))
     (debug:print-error 0 *default-log-port* "db:sync-tables called with fromdb not a database " fromdb)
     -3)
    ((not (sqlite3:database? (db:dbdat-get-db todb)))
     (debug:print-error 0 *default-log-port* "db:sync-tables called with todb not a database " todb)
     -4)

    ((not (file-write-access? (db:dbdat-get-path todb)))
     (debug:print-error 0 *default-log-port* "db:sync-tables called with todb not a read-only database " todb)
     -5)
    ((not (null? (let ((readonly-slave-dbs
                        (filter
                         (lambda (dbdat)
                           (not (file-write-access? (db:dbdat-get-path todb))))
                         slave-dbs)))
                   (for-each
                    (lambda (bad-dbdat)
                      (debug:print-error
                       0 *default-log-port* "db:sync-tables called with todb not a read-only database " bad-dbdat))
                    readonly-slave-dbs)
                   readonly-slave-dbs))) -6)

	(debug:print-info 11 *default-log-port* "open-run-close-no-exception-handling END" )
	res)
      #f))

#;(define (open-run-close-exception-handling proc idb . params)
  (handle-exceptions
   exn
   (let ((sleep-time (random 30))
	 (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))

(define (tdb:get-prev-tol-for-test tdb test-id category variable)
  ;; Finish me?
  (values #f #f #f))


(define (db:csv->test-data dbstruct run-id test-id csvdata)
  (debug:print 4 *default-log-port* "test-id " test-id ", csvdata: " csvdata)
  (db:with-db
   dbstruct #f #f
   (lambda (db)
     (let* ((csvlist (csv->list (make-csv-reader
				 (open-input-string csvdata)
				 '((strip-leading-whitespace? #t)

  (let* ((keysstr  (string-intersperse (map car keypatt-alist) ","))
	 (keyqry   (string-intersperse (map (lambda (p)(conc (car p) " LIKE ? ")) keypatt-alist) " AND "))
	 (numkeys  (length keypatt-alist))
	 (test-ids '())
	 (dbdat    (db:get-db dbstruct))
	 (db       (db:dbdat-get-db dbdat))
	 (windows  (and pathmod (substring-index "\\" pathmod)))
	 (tempdir  (conc "/tmp/" (current-user-name) "/" runspatt "_" (random 10000) "_" (current-process-id)))
	 (runsheader (append (list "Run Id" "Runname") ; 0 1
			     (map car keypatt-alist)   ; + N = length keypatt-alist
			     (list "Testname"          ; 2
				   "Item Path"         ; 3 
				   "Description"       ; 4 
				   "State"             ; 5 
				   "Status"            ; 6  

			test-name     " "
			item-path     " " ;; has / prepended to deal with toplevel tests
			actual-state  " "
			actual-status " "
			event-time
			))
	 (prev-nbfake-log (get-environment-variable "NBFAKE_LOG")))
    (setenv "NBFAKE_LOG" (conc (cond
				((and (directory-exists? test-rundir)
				      (file-write-access? test-rundir))
				 test-rundir)
				((and (directory-exists? *toppath*)
				      (file-write-access? *toppath*))
				 *toppath*)
				(else (conc "/tmp/" (current-user-name))))
			       "/" logname))
    (debug:print-info 0 *default-log-port* "TRIGGERED on " trigger ", running command " fullcmd " output at " (get-environment-variable "NBFAKE_LOG"))
    ;; (call-with-environment-variables
    ;;  `(("NBFAKE_LOG" . ,(conc test-rundir "/" logname)))
    ;;  (lambda ()
    (process-run fullcmd)
    (if prev-nbfake-log
	(setenv "NBFAKE_LOG" prev-nbfake-log)
	(unsetenv "NBFAKE_LOG"))
    )) ;; ))

(define (mt:process-triggers dbstruct run-id test-id newstate newstatus)
  (if test-id 
      (let* ((test-dat      (db:get-test-info-by-id dbstruct run-id test-id)))
	(if test-dat
	    (let* ((test-rundir   (db:test-get-rundir       test-dat)) ;; ) ;; )

(declare (uses rmtmod))
(declare (uses dbmod))
(declare (uses servermod))

(module dcommonmod
	*
	
(import scheme chicken data-structures extras)
(import (prefix sqlite3 sqlite3:)
	posix typed-records srfi-18 srfi-69 format ports srfi-1
	matchable (prefix iup iup:)
	canvas-draw
	;; blindly copied from megamod
	(prefix base64 base64:)
	(prefix dbi dbi:)
	(prefix nanomsg nmsg:)
	(prefix sqlite3 sqlite3:)
	call-with-environment-variables
	csv
	csv-xml
	data-structures
	directory-utils
	dot-locking
	extras
	files

	fmt
	format
	hostinfo
	http-client
	intarweb
	irregex
	matchable
	md5
	message-digest
	pathname-expand
	pkts
	ports
	posix
	;; queue
	regex
	regex-case
	s11n
	sparse-vectors
	spiffy
	spiffy-directory-listing
	spiffy-request-vars
	sql-de-lite
	srfi-1
	srfi-4
	srfi-13
	srfi-18
	srfi-69
	stack
	stml2
	tcp
	typed-records
	udp
	uri-common
	z3
	)

(import (prefix mtconfigf configf:))
(import gutilsmod)
(import commonmod)
(import servermod)
(import testsmod)
(import megamod)
(import subrunmod)
(import runsmod)
(import rmtmod)
(import dbmod)
(import canvas-draw)
(import canvas-draw-iup)
(use (prefix iup iup:))
(import (prefix mtargs args:))

(define *tim* (iup:timer))

;; (use (prefix ulex ulex:))

(include "common_records.scm")

module-hierarchy.pdf : module-hierarchy.dot
	dot -Tpdf module-hierarchy.dot -o module-hierarchy.pdf

  node [shape=box,style=filled,fontname="clear",fontsize="10"];

  subgraph cluster_megatest {
    label="megatest";

    rmtmod [label="rmt mod"];
    // httpmod [label="http-transportmod"];
    // commonmod
// archivemod.scm

"commonmod" -> "archivemod";
// clientmod.scm
"commonmod" -> "clientmod";





// configfmod.scm
// subrunmod.scm

"commonmod" -> "subrunmod";
// ezstepsmod.scm
"commonmod" -> "ezstepsmod";
// itemsmod.scm

"commonmod" -> "itemsmod";
// gutilsmod.scm


// testsmod.scm

"commonmod" -> "testsmod";
"servermod" -> "testsmod";
"itemsmod" -> "testsmod";
"dbmod" -> "testsmod";
// runconfigmod.scm
"commonmod" -> "runconfigmod";
// keysmod.scm
"srfi-13" -> "keysmod";
// launchmod.scm

"commonmod" -> "launchmod";
// megamod.scm

"spiffy" -> "megamod";
"stml2" -> "megamod";

"commonmod" -> "megamod";
"keysmod" -> "megamod";
"pgdbmod" -> "megamod";
"tasksmod" -> "megamod";
"dbmod" -> "megamod";
"apimod" -> "megamod";
"ducttape-lib" -> "megamod";
"itemsmod" -> "megamod";
"pkts" -> "megamod";
"rmtmod" -> "megamod";
"mtmod" -> "megamod";
"testsmod" -> "megamod";
"servermod" -> "megamod";
"subrunmod" -> "megamod";
"itemsmod" -> "megamod";
"runsmod" -> "megamod";
// odsmod.scm

"commonmod" -> "odsmod";

// envmod.scm

"commonmod" -> "envmod";


// http-transportmod.scm

"commonmod" -> "http-transportmod";

"stml2" -> "http-transportmod";

"apimod" -> "http-transportmod";





"dbmod" -> "http-transportmod";
"testsmod" -> "http-transportmod";





// processmod.scm


// mtmod.scm



"commonmod" -> "mtmod";
"dbmod" -> "mtmod";
"pgdbmod" -> "mtmod";
"rmtmod" -> "mtmod";
"servermod" -> "mtmod";
"stml2" -> "mtmod";
"subrunmod" -> "mtmod";
"tasksmod" -> "mtmod";
"testsmod" -> "mtmod";
// pgdbmod.scm



"commonmod" -> "pgdbmod";
// runsmod.scm

"commonmod" -> "runsmod";
"dbmod" -> "runsmod";
"itemsmod" -> "runsmod";
"mtmod" -> "runsmod";
"pgdbmod" -> "runsmod";


"rmtmod" -> "runsmod";
"servermod" -> "runsmod";
"stml2" -> "runsmod";
"subrunmod" -> "runsmod";
"tasksmod" -> "runsmod";
"testsmod" -> "runsmod";
// apimod.scm
"commonmod" -> "apimod";
"dbmod" -> "apimod";
"servermod" -> "apimod";
// dbmod.scm
"commonmod" -> "dbmod";
"keysmod" -> "dbmod";
"files" -> "dbmod";
"tasksmod" -> "dbmod";
"odsmod" -> "dbmod";
// dcommonmod.scm
"gutilsmod" -> "dcommonmod";
"commonmod" -> "dcommonmod";
"testsmod" -> "dcommonmod";
"megamod" -> "dcommonmod";
"canvas-draw" -> "dcommonmod";
"canvas-draw-iup" -> "dcommonmod";
// tasksmod.scm
"commonmod" -> "tasksmod";
"pgdbmod" -> "tasksmod";
// template-mod.scm
"commonmod" -> "template-mod";
// servermod.scm
"commonmod" -> "servermod";
"dbmod" -> "servermod";
"tasksmod" -> "servermod";
// treemod.scm
// commonmod.scm
"pkts" -> "commonmod";
// vgmod.scm
// rmtmod.scm
"commonmod" -> "rmtmod";
"itemsmod" -> "rmtmod";
"apimod" -> "rmtmod";
"dbmod" -> "rmtmod";







































    
}
  
}

     current-wwdate
     current-isodate
     *this-exe-dir*
     *this-exe-name*
     *this-exe-fullpath*
     )

  (import scheme chicken extras ports data-structures )
  (use posix regex ansi-escape-sequences test srfi-1 irregex slice srfi-13 rfc3339)
  ;;scsh-process ;; dropping scsh-process, it was clobbering posix's process and process*
  (use directory-utils uuid-lib filepath srfi-19 ) ; linenoise


    ;; plugs a hole in posix-extras in latter chicken versions
  (use posix-extras pathname-expand files)
  (define ##sys#expand-home-path pathname-expand)
  (define (realpath x) (resolve-pathname  (pathname-expand (or x "/dev/null")) ))


  ;; (include "mimetypes.scm") ; provides ext->mimetype
  ;; (include "workweekdate.scm")

  ;; gathered from macosx:
;;   cat /etc/apache2/mime.types | grep -v '^#' | perl -ne 'print "(\"$2\" . \"$1\")\n" if /(\S+)\s+(\S+)/' > mimetypes.scm
;; + manual manipulation

("wmx" . "video/x-ms-wmx")
("wvx" . "video/x-ms-wvx")
("avi" . "video/x-msvideo")
("movie" . "video/x-sgi-movie")
("smv" . "video/x-smv")
("ice" . "x-conference/x-cooltalk")))

(use srfi-19)
(use test)
;;(use format)
(use regex)
;(declare (unit wwdate))
;; utility procedures to convert among
;; different ways to express date (wwdate, seconds since epoch, isodate)
;;
;; samples:
;; isodate   -> "2016-01-01"
;; wwdate -> "16ww01.5"

      (let loop ((rest-path-items path-items))
        (if (null? rest-path-items)
            #f
            (let* ((this-dir (car rest-path-items))
                   (next-rest (cdr rest-path-items))
                   (candidate (conc this-dir "/" exe)))
              (if (file-execute-access? candidate)
                  candidate
                  (loop next-rest)))))))


  
;;;; define some handy globals
  ;; resolve fullpath to this script or binary.

     (let ( (raw-debug-level (get-environment-variable "DUCTTAPE_DEBUG_LEVEL")) )
       (if raw-debug-level
           (let ((num-debug-level (runs-ok (string->number raw-debug-level))))
             (if (integer? num-debug-level)
                 (begin
                   (let ((new-num-debug-level (- num-debug-level 1)))
                     (if (> new-num-debug-level 0) ;; decrement
                         (setenv "DUCTTAPE_DEBUG_LEVEL" (number->string new-num-debug-level))
                         (unsetenv "DUCTTAPE_DEBUG_LEVEL")))
                   num-debug-level) ; it was set and > 0, mode is value
                 (begin
                   (unsetenv "DUCTTAPE_DEBUG_LEVEL") ;; value was invalid, unset it
                   #f))) ; value was invalid, mode is f
           #f)))) ; var not set, mode is f


  (define ducttape-debug-mode (if (ducttape-debug-level)  #t  #f))

  ;; ducttape-debug-regex-filter suppresses non-matching debug messages

            " "))
          (pwd (or (get-environment-variable "PWD") "nopwd"))
          (user (or (get-environment-variable "USER") "nouser"))
          (host (or (get-environment-variable "HOST") "nohost")))
      (if logfile
          (begin
            (ducttape-log-file logfile)
            (setenv "DUCTTAPE_LOG_FILE" (ducttape-log-file))))
      (ducttape-append-logfile 'note (format #f "START - pid=~A ppid=~A argv=(~A) pwd=~A user=~A host=~A" pid ppid argv pwd user host) #t)))         


  ;; log exit code
  (define (set-ducttape-log-exit-handler)
    (let ((orig-exit-handler (exit-handler)))
      (exit-handler 

                    )

    (define (sendmail-proc sendmail-port)
      (define (wl line-str)
        (write-line line-str sendmail-port))

      (define (get-uuid)

        (string-upcase (uuid->string (uuid-generate))))

      (let ((mailpart-uuid (get-uuid))
            (mailpart-body-uuid (get-uuid)))
        
        (define (boundary)
          (wl (conc "--" mailpart-uuid)))

  ;; are sure they can coexist.
  (define (ducttape-process-command-line)

    ;; --quiet
    (let ((quiet-opts (skim-cmdline-opts-noarg-by-regex "--?quiet")))
      (if (not (null? quiet-opts))
          (begin
            (setenv "DUCTTAPE_QUIET_MODE" "1")
            (ducttape-quiet-mode "1"))))

    ;; --silent
    (let ((silent-opts (skim-cmdline-opts-noarg-by-regex "--?silent")))
      (if (not (null? silent-opts))
          (begin
            (setenv "DUCTTAPE_SILENT_MODE" "1")
            (ducttape-silent-mode "1"))))

    ;; -color
    (let ((color-opts (skim-cmdline-opts-noarg-by-regex "--?colou?r(ize)?")))
      (if (not (null? color-opts))
          (begin
            (setenv "DUCTTAPE_COLORIZE" "1")
            (ducttape-color-mode "1"))))

    ;; -nocolor
    (let ((nocolor-opts (skim-cmdline-opts-noarg-by-regex "--?nocolou?r(ize)?")))
      (if (not (null? nocolor-opts))
          (begin
            (unsetenv "DUCTTAPE_COLORIZE" )
            (ducttape-color-mode #f))))

    ;; -logfile
    (let ((logfile-opts (skim-cmdline-opts-withargs-by-regex "--?log(-?file)?")))
      (if (not (null? logfile-opts))
          (begin
            (ducttape-log-file (car (reverse logfile-opts)))
            (setenv "DUCTTAPE_LOG_FILE" (ducttape-log-file)))))

    ;; -d -dd -d#
    (let ((debug-opts (skim-cmdline-opts-noarg-by-regex "-d(d*|\\d+)"))
          (initial-debuglevel (if (ducttape-debug-level) (ducttape-debug-level) 0) ))
      (if (not (null? debug-opts))
          (begin
            (ducttape-debug-level
             (let loop ((opts debug-opts) (debuglevel initial-debuglevel))
               (if (null? opts)
                   debuglevel
                   (let*
                       ( (curopt (car opts))
                         (restopts (cdr opts))
                         (ds (string-match "-(d+)" curopt))
                         (dnum (string-match "-d(\\d+)" curopt)))
                     (cond
                      (ds (loop restopts (+ debuglevel (string-length (cadr ds)))))
                      (dnum  (loop restopts (string->number (cadr dnum)))))))))
            (setenv "DUCTTAPE_DEBUG_LEVEL" (number->string (ducttape-debug-level))))))


    ;; -dp <pat> / --debug-pattern <pat>
    (let ((debugpat-opts (skim-cmdline-opts-withargs-by-regex "--?(debug-pattern|dp)")))
      (if (not (null? debugpat-opts))
          (begin
            (ducttape-debug-regex-filter (string-join debugpat-opts "|"))
            (setenv "DUCTTAPE_DEBUG_PATTERN" (ducttape-debug-regex-filter)))))) 


  ;;; following code commented out; side effects not wanted on startup
  ;; immediately activate logfile (will be noop if logfile disabled)
  ;;(ducttape-activate-logfile)
  ;;(set-ducttape-log-exit-handler)
  

;;     (print "BEFORE:   " (string-intersperse patha-parts  "\n       "))
;;     (print "AFTER:    " (string-intersperse pathb-parts  "\n       "))
;;     (print "COMMON:   " (string-intersperse common-parts "\n       "))
    (string-intersperse final separator)))

(define (env:process-path-envvar varname separator patha pathb)
  (let ((newpath (env:merge-path-envvar separator patha pathb)))
    (setenv varname newpath)))

(define (env:have-context db context)
  (> (query fetch-value (sql db "SELECT count(id) FROM envvars WHERE context=?") context)
     0))

;; this is so the calling block does not need to import sql-de-lite
(define (env:close-database db)

(declare (unit envmod))
(declare (uses commonmod))

(module envmod
	*
	
(import scheme chicken data-structures extras)
(import (prefix sqlite3 sqlite3:) posix typed-records srfi-18 srfi-69 format ports srfi-1 matchable)
(import commonmod)
;; (use (prefix ulex ulex:))

(include "common_records.scm")


)

(declare (unit ezstepsmod))
(declare (uses commonmod))

(module ezstepsmod
	*
	
(import scheme chicken data-structures extras)
(import (prefix sqlite3 sqlite3:) posix typed-records srfi-18 srfi-69 format ports srfi-1 matchable)
(import commonmod)
;; (use (prefix ulex ulex:))

(include "common_records.scm")


)

;;     You should have received a copy of the GNU General Public License
;;     along with Megatest.  If not, see <http://www.gnu.org/licenses/>.
;;
;;======================================================================

(require-library iup)
(import (prefix iup iup:))
(use canvas-draw)

(use srfi-1 regex regex-case srfi-69)
(declare (unit gutilsmod))

(module gutilsmod
  *

(import  scheme chicken data-structures extras srfi-1)
(include "gutils-inc.scm")
)

(declare (uses mtargs))
(declare (uses mtconfigf))
(declare (uses testsmod))

(module apimod
	*
	
(import scheme chicken data-structures extras posix files
	srfi-13 srfi-18 spiffy http-client
	spiffy-directory-listing
	spiffy-request-vars tcp
	)
(import commonmod)
(import stml2)
(import apimod)

(declare (unit itemsmod))
(declare (uses commonmod))
(declare (uses mtconfigf))

(module itemsmod
	*
	
(import scheme chicken data-structures extras)
(use (prefix sqlite3 sqlite3:) posix typed-records srfi-18 srfi-69 format ports srfi-1 matchable)
(import commonmod)
(import(prefix mtconfigf configf:))
;; (use (prefix ulex ulex:))

;; (include "common_records.scm")
;; (include "items-inc.scm")

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

(declare (unit keysmod))

(module keysmod
	*
	
(import scheme chicken data-structures extras)
(import (prefix sqlite3 sqlite3:) posix typed-records srfi-18 srfi-69 format ports srfi-1 matchable)
;; (import commonmod)

;; (use (prefix ulex ulex:))
(import srfi-13)

;; (include "common_records.scm")

(declare (unit launchmod))

(declare (uses commonmod))

(module launchmod
	*
	
(import scheme chicken data-structures extras files)
(import (prefix sqlite3 sqlite3:) posix typed-records
	srfi-18 srfi-69 format ports srfi-1 matchable
	z3 (prefix base64 base64:) regex
	call-with-environment-variables csv)
(import commonmod)

)

(declare (uses subrunmod))
(declare (uses itemsmod))
(declare (uses runsmod))

(module megamod
	*
	
(import scheme chicken data-structures extras)
(use 
 (prefix base64 base64:)
 (prefix dbi dbi:)
 (prefix nanomsg nmsg:)
 (prefix sqlite3 sqlite3:)
 call-with-environment-variables
 csv
 csv-xml
 data-structures
 directory-utils
 dot-locking
 extras
 files






 fmt
 format

 hostinfo
 http-client
 intarweb
 irregex
 matchable
 md5
 message-digest
 pathname-expand
 ;; pkts
 ports
 posix
 ;; queue
 regex
 regex-case
 s11n
 sparse-vectors
 spiffy
 spiffy-directory-listing
 spiffy-request-vars
 sql-de-lite
 srfi-1
 srfi-4
 srfi-13
 srfi-18
 srfi-69
 stack
 tcp
 typed-records
 udp
 uri-common
 z3
 )

(import (prefix mtconfigf configf:))
(define read-config configf:read-config)
(define find-and-read-config configf:find-and-read-config)

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

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

(use (prefix sqlite3 sqlite3:) srfi-1 posix regex regex-case srfi-69 (prefix base64 base64:)
     readline apropos json http-client directory-utils typed-records
     http-client srfi-18 extras format)

;; Added for csv stuff - will be removed
;;
(use sparse-vectors)

(declare (uses mtargs))
(declare (uses mtconfigf))

	    ;; (exit)
	    ;; EOF

	    (repl))
	   (else
	    (begin
	      (set! *db* dbstruct)
	      (import extras) ;; might not be needed
	      ;; (import csi)
	      (import readline)
	      (import apropos)
	      (import dbmod)
	      (import rmtmod)
	      (import commonmod)
	      ;; (import (prefix sqlite3 sqlite3:)) ;; doesn't work ...

	      (if *use-new-readline*

     usage
     get-args
     print-args
     any-defined?
     help
     )

(import scheme chicken data-structures extras posix ports files)
(use srfi-69 srfi-1)

(define arg-hash (make-hash-table))
(define help "")

(define (get-arg arg . default)
  (if (null? default)
      (hash-table-ref/default arg-hash arg #f)

	 get-eval-string
         squelch-debug-prints
	 ;; misc
	 realpath
	 find-chicken-lib
         )

(import scheme chicken data-structures extras ports files)
(use posix typed-records srfi-18 pathname-expand posix-extras)
(use regex regex-case srfi-69 srfi-1 directory-utils extras srfi-13 )
(use srfi-69)

(import posix)

;; stub debug printers overridden by set-debug-printers
(define (debug:print n e . args)
  (apply print args))
(define (debug:print-info n e . args)
  (apply print "INFO: " args))
(define (debug:print-error n e . args)
  (apply print "ERROR: " args))

;;(import (prefix mtdebug debug:))
;;(define args:any? args:any-defined?) ;; cannot name it any? in mtargs module


;; FROM common.scm
;;
;; this plugs a hole in posix-extras in recent chicken versions > 4.9)
(let-values (( (chicken-release-number chicken-major-version)
               (apply values
                      (map string->number
                           (take
                            (string-split (chicken-version) ".")
                            2)))))
  (if (or (> chicken-release-number 4)
	  (and (eq? 4 chicken-release-number) (> chicken-major-version 9)))
      (define ##sys#expand-home-path pathname-expand)))


 ;;(define (set-verbosity v)(debug:set-verbosity v))

 (define *default-log-port* (current-error-port))

 (define (debug:print-error n . args) ;;; n available to end-users but ignored for

  (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))))

;; accept an alist or hash table containing envvar/env value pairs (value of #f causes unset) 
;;   execute thunk in context of environment modified as per this list
;;   restore env to prior state then return value of eval'd thunk.
;;   ** this is not thread safe **
(define (with-env-vars delta-env-alist-or-hash-table thunk)
  (let* ((delta-env-alist (if (hash-table? delta-env-alist-or-hash-table)
                              (hash-table->alist delta-env-alist-or-hash-table)
                              delta-env-alist-or-hash-table))
         (restore-thunks
          (filter
           identity
           (map (lambda (env-pair)
                  (let* ((env-var     (car env-pair))
                         (new-val     (let ((tmp (cdr env-pair)))
                                        (if (list? tmp) (car tmp) tmp)))
                         (current-val (get-environment-variable env-var))
                         (restore-thunk
                          (cond
                           ((not current-val) (lambda () (unsetenv env-var)))
                           ((not (string? new-val)) #f)
                           ((eq? current-val new-val) #f)
                           (else 
                            (lambda () (setenv env-var current-val))))))
                    ;;(when (not (string? new-val))
                    ;;    (debug:print 0 *default-log-port* " PROBLEM: not a string: "new-val"\n from env-alist:\n"delta-env-alist)
                    ;;    (pp delta-env-alist)
                    ;;    (exit 1))
                        
                    
                    (cond
                     ((not new-val)  ;; modify env here
                      (unsetenv env-var))
                     ((string? new-val)
                      (setenv env-var new-val)))
                    restore-thunk))
                delta-env-alist))))
    (let ((rv (thunk)))
      (for-each (lambda (x) (x)) restore-thunks) ;; restore env to original state
      rv)))

(define (cmd-run->list cmd #!key (delta-env-alist-or-hash-table '()))

				       (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 (safe-file-exists? include-script)(file-execute-access? include-script))
                                      (let* ((local-allow-system  (calc-allow-system allow-system curr-section-name sections))
                                             (env-delta  (cfgdat->env-alist curr-section-name res local-allow-system))
                                             (new-inp-port
                                              (with-env-vars
                                               env-delta
                                               (lambda ()
                                                 (open-input-pipe (conc include-script " " params))))))

;;
(define (find-and-read-config fname #!key (environ-patt #f)(given-toppath #f)(pathenvvar #f)(set-fields #f)(keep-filenames #f))
  (let* ((curr-dir   (current-directory))
         (configinfo (find-config fname toppath: given-toppath))
	 (toppath    (car configinfo))
	 (configfile (cadr configinfo)))
    (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: (if set-fields (list (cons "^fields$" set-fields)   ) '())
				       #f
                                       keep-filenames: keep-filenames))))
      (if toppath (change-directory curr-dir))
      (list configdat toppath configfile fname))))

;; reads a refdb into an assoc array of assoc arrays
;;   returns (list dat msg)
(define (read-refdb refdb-path)
  (let ((sheets-file  (conc refdb-path "/sheet-names.cfg")))
    (if (not (safe-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 '()))
				 (if (eof-object? inl)
				     (reverse res)

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

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

(use srfi-1 posix srfi-69 readline ;;  regex regex-case srfi-69 apropos json http-client directory-utils rpc typed-records;; (srfi 18) extras)
   srfi-19  srfi-18 extras format pkts regex regex-case
     (prefix dbi dbi:)
     )

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

    (begin
      (print help)
      (exit)))

(if (or (args:get-arg "-repl")
	(args:get-arg "-load"))
    (begin
      (import extras) ;; might not be needed
      ;; (import csi)
      (import readline)
      (import apropos)
      ;; (import (prefix sqlite3 sqlite3:)) ;; doesn't work ...
      
      (install-history-file (get-environment-variable "HOME") ".mtutil_history") ;;  [homedir] [filename] [nlines])
      (current-input-port (make-readline-port "mtutil> "))
      (if (args:get-arg "-repl")
	  (repl)

(declare (uses subrunmod))
(declare (uses tasksmod))
(declare (uses testsmod))

(module mtmod
	*
	
(import scheme chicken data-structures extras posix ports files)

(use (prefix sqlite3 sqlite3:)
     srfi-69 regex srfi-18 srfi-13 srfi-1
     call-with-environment-variables z3 (prefix base64 base64:)
     typed-records
     csv directory-utils)

(import	(prefix mtargs args:))
(import (prefix mtconfigf configf:))
(import	commonmod)
(import	dbmod)
(import	pgdbmod)
(import	rmtmod)

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

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

(use srfi-1 posix srfi-69 readline ;;  regex regex-case srfi-69 apropos json http-client directory-utils rpc typed-records;; (srfi 18) extras)
   srfi-19  srfi-18 extras format pkts regex regex-case
     (prefix dbi dbi:)
     nanomsg)

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

    (begin
      (stml:main #f)
      (exit)))

(if (or (args:get-arg "-repl")
	(args:get-arg "-load"))
    (begin
      (import extras) ;; might not be needed
      ;; (import csi)
      (import readline)
      (import apropos)
      ;; (import (prefix sqlite3 sqlite3:)) ;; doesn't work ...
      
      (install-history-file (get-environment-variable "HOME") ".mtutil_history") ;;  [homedir] [filename] [nlines])
      (current-input-port (make-readline-port "mtutil> "))
      (if (args:get-arg "-repl")
	  (repl)

;; This is from the perl world, a hash of hashes is a super easy way to keep a handle on
;; lots of disparate data
;;

(module mutils
    *

  (import chicken scheme
	  ;; data-structures posix
	  srfi-1
	  ;; srfi-13
	  srfi-69
	  ;; ports
	  extras
	  regex
	  )

(define (mutils:hierhash-ref hh . keys)
  (if (null? keys)
      #f
      (let loop ((ht   hh)

      (if (eof-object? l)
	  (reverse res)
	  (if (or (string-match comment l)
		  (string-match blank l))
	      (loop (read-line fh) res)
	      (loop (read-line fh) (cons l res)))))))

(use sparse-vectors)

;; this is a simple two dimensional sparse array

;; ONLY TWO DIMENSIONS!!! SEE ARRAY-LIB IF YOUR NEEDS ARE GREATER!!
;;
(define (mutils:make-sparse-array)
  (let ((a (make-sparse-vector)))

(declare (unit odsmod))
(declare (uses commonmod))

(module odsmod
	*
	
(import scheme chicken data-structures extras csv-xml regex)
(import (prefix sqlite3 sqlite3:) posix typed-records srfi-18 srfi-69
	format ports srfi-1 matchable srfi-13)
(import commonmod)
;; (use (prefix ulex ulex:))

(define ods:dirs
  '("Configurations2"
    "Configurations2/toolpanel"
    "Configurations2/menubar"

(declare (uses mtconfigf))
(declare (uses commonmod))
(declare (uses mtargs))

(module pgdbmod
	*
	
(import scheme chicken data-structures extras files)
(use posix
     (prefix dbi dbi:)
     srfi-69 srfi-1 typed-records)

(import (prefix mtconfigf configf:))
(import	(prefix mtargs    args:))
(import	commonmod)

pktdb-pktspec

;; utility procs
increment-string ;; used to get indexes for strings in ref pkts
make-report      ;; make a .dot file 
)


(import chicken scheme data-structures posix srfi-1 regex srfi-13 srfi-69 ports extras)
(use crypt sha1 message-digest (prefix dbi dbi:) typed-records)

;;======================================================================
;; DATA MANIPULATION UTILS
;;======================================================================

(define-inline (unescape-data data)
  (string-translate* data '(("\\n" . "\n") ("\\\\" . "\\"))))

     (for-each
      (lambda (pktsdir) ;; look at all
	(cond
	 ((not (file-exists? pktsdir))
	  (print "ERROR: packets directory " pktsdir " does not exist."))
	 ((not (directory? pktsdir))
	  (print "ERROR: packets directory path " pktsdir " is not a directory."))
	 ((not (file-read-access? pktsdir))
	  (print "ERROR: packets directory path " pktsdir " is not readable."))
	 (else
	  ;; (print "INFO: 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   (common:file-exists? fname))
	 (db       (if avail 
		       (sqlite3:open-database fname)
		       (begin
			 (system (conc "rm -f " fname))
			 (sqlite3:open-database fname))))
	 (handler  (sqlite3:make-busy-timeout 136000))
	 (canwrite (file-write-access? fname)))
	 ;; (db-init  (lambda ()
	 ;;             (sqlite3:execute 
	 ;;              db
	 ;;              "CREATE TABLE IF NOT EXISTS ports (
         ;;                 port INTEGER PRIMARY KEY,
         ;;                 state TEXT DEFAULT 'not-used',
         ;;                 fail_count INTEGER DEFAULT 0,

  (let* ((lowport (let ((val (configf:lookup *configdat* "server" "lowport")))
		    (if (and val 
			     (string->number val))
			(string->number val)
			32768)))
	 (portnum (or (portlogger:get-prev-used-port db)
		      (+ lowport ;; top of registered ports is 49152 but lets use ports in the registered range
			 (random (- 64000 lowport))))))
    (handle-exceptions
     exn
     (begin
       (debug:print 0 *default-log-port* "EXCEPTION: portlogger database probably overloaded or unreadable. If you see this message again remove /tmp/.$USER-portlogger.db")
       (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
       (debug:print 5 *default-log-port* "exn=" (condition->list exn))
       (print-call-chain (current-error-port))

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

(declare (unit processmod))

(module processmod
	*
	
(import scheme chicken data-structures extras)
(import (prefix sqlite3 sqlite3:) posix typed-records srfi-18 srfi-69
	format ports srfi-1 matchable regex directory-utils)

;;======================================================================
;; supporting functions
;;======================================================================

)

(declare (uses dbmod))
(declare (uses itemsmod))
(declare (uses ulex))

(module rmtmod
	*
	
(import scheme chicken data-structures extras)
(import (prefix sqlite3 sqlite3:) posix typed-records srfi-18 srfi-69 format ports srfi-1 matchable)

(import (prefix ulex ulex:))

(import commonmod)
(import itemsmod)
(import apimod)
(import dbmod)

;; (include "rmt-inc.scm")

;;
;; THESE ARE ALL CALLED ON THE CLIENT SIDE!!!
;;

;; generate entries for ~/.megatestrc with the following
;;
;;  grep define ../rmt.scm | grep rmt: |perl -pi -e 's/\(define\s+\((\S+)\W.*$/\1/'|sort -u

			#f)))
	  (if cinfo
	      cinfo
	      (if (server:check-if-running areapath)
		  (client:setup areapath)
		  #f))))










(define *send-receive-mutex* (make-mutex)) ;; should have separate mutex per run-id













(define (rmt:send-receive cmd rid params #!key (attemptnum 1)(area-dat #f)) ;; start attemptnum at 1 so the modulo below works as expected









  (rmt:open-qry-close-locally cmd 0 params))

;;   
;; ;; #;(common:telemetry-log (conc "rmt:"(->string cmd))
;; ;; #;(define (rmt:send-receive cmd rid params #!key (attemptnum 1)(area-dat #f)) ;; start attemptnum at 1 so the modulo below works as expected
;; ;; 
;; ;; #;(common:telemetry-log (conc "rmt:"(->string cmd))
;; ;; payload: `((rid . ,rid)
;; ;; (params . ,params)))

    (mutex-unlock! *db-stats-mutex*)
    res))

(define (rmt:open-qry-close-locally cmd run-id params #!key (remretries 5))
  (let* ((qry-is-write   (not (member cmd api:read-only-queries)))
	 (db-file-path   (db:dbfile-path)) ;;  0))
	 (dbstruct-local (db:setup #t))  ;; make-dbr:dbstruct path:  dbdir local: #t)))
	 (read-only      (not (file-write-access? db-file-path)))
	 (start          (current-milliseconds))
	 (resdat         (if (not (and read-only qry-is-write))
			     (let ((v (api:execute-requests dbstruct-local (vector (symbol->string cmd) params))))
			       (handle-exceptions ;; there has been a long history of receiving strange errors from values returned by the client when things go wrong..
				exn               ;;  This is an attempt to detect that situation and recover gracefully
				(begin
				  (debug:print 0 *default-log-port* "ERROR: bad data from server " v " message: "  ((condition-property-accessor 'exn 'message) exn))

	 (duration       (- (current-milliseconds) start)))
    (if (and read-only qry-is-write)
        (debug:print 0 *default-log-port* "ERROR: attempt to write to read-only database ignored. cmd=" cmd))
    (if (not success)
	(if (> remretries 0)
	    (begin
	      (debug:print-error 0 *default-log-port* "local query failed. Trying again.")
	      (thread-sleep! (/ (random 5000) 1000)) ;; some random delay 
	      (rmt:open-qry-close-locally cmd run-id params remretries: (- remretries 1)))
	    (begin
	      (debug:print-error 0 *default-log-port* "too many retries in rmt:open-qry-close-locally, giving up")
	      #f))
	(begin
	  ;; (rmt:update-db-stats run-id cmd params duration)
	  ;; mark this run as dirty if this was a write, the watchdog is responsible for syncing it

	       remote-ro-mode-checked-set!            remote-ro-mode-checked)

(define (rmtmod:calc-ro-mode runremote *toppath*)
  (if (and runremote
	   (remote-ro-mode-checked runremote))
      (remote-ro-mode runremote)
      (let* ((dbfile  (conc *toppath* "/megatest.db"))
	     (ro-mode (not (file-write-access? dbfile)))) ;; TODO: use dbstruct or runremote to figure this out in future
	(if runremote
	    (begin
	      (remote-ro-mode-set! runremote ro-mode)
	      (remote-ro-mode-checked-set! runremote #t)
	      ro-mode)
	    ro-mode))))

(declare (unit runconfigmod))
(declare (uses commonmod))

(module runconfigmod
	*
	
(import scheme chicken data-structures extras)
(import (prefix sqlite3 sqlite3:) posix typed-records srfi-18 srfi-69 format ports srfi-1 matchable)
(import commonmod)
;; (use (prefix ulex ulex:))

(include "common_records.scm")


)

(declare (uses tasksmod))
(declare (uses testsmod))
(declare (uses itemsmod))

(module runsmod
	*
	
(import scheme chicken data-structures extras ports files)



(use (prefix base64 base64:)
     (prefix sqlite3 sqlite3:)
     call-with-environment-variables
     csv
     directory-utils
     format
     matchable
     message-digest
     md5
     ports
     posix
     regex
     srfi-1
     srfi-1
     srfi-13
     srfi-18
     srfi-18
     srfi-69

  hed tal reg reruns  test-record
  test-name item-path jobgroup
  waitons testmode  newtal itemmaps prereqs-not-met)

(define (set-item-env-vars itemdat)
  (for-each (lambda (item)
	      (debug:print 2 *default-log-port* "setenv " (car item) " " (cadr item))
	      (setenv (car item) (cadr item)))
	    itemdat))

;; Every time can-run-more-tests is called increment the delay
;;
;; NOTE: We run this server-side!! Do not use this global except in the runs:can-run-more-tests routine
;;
(define *last-num-running-tests* 0)

(define (runs:run-tests target runname test-patts user flags #!key (run-count 1)) ;; test-names
  (let* ((keys               (keys:config-get-fields *configdat*))
	 (keyvals            (keys:target->keyval keys target))
	 (run-id             (rmt:register-run keyvals runname "new" "n/a" user (args:get-arg "-contour")))  ;;  test-name)))
	 ;; (deferred          '()) ;; delay running these since they have a waiton clause
	 (runconfigf         (conc  *toppath* "/runconfigs.config"))
         (dbfile             (conc  *toppath* "/megatest.db"))
         (readonly-mode      (not (file-write-access? dbfile)))
	 (test-records       (make-hash-table))
	 ;; need to process runconfigs before generating these lists
	 (all-tests-registry #f)  ;; (tests:get-all)) ;; (tests:get-valid-tests (make-hash-table) test-search-path)) ;; all valid tests to check waiton names
	 (all-test-names     #f)  ;; (hash-table-keys all-tests-registry))
	 (test-names         #f)  ;; Generated by a call to (tests:filter-test-names all-test-names test-patts))
	 (required-tests     #f)  ;; Put fully qualified test/testpath names in this list to be done
         (waitors-upon       (make-hash-table)) ;; given a test, return list of tests waiting upon this test.

    ;;======================================================================
    
    (if (not (null? test-names)) ;; BEGIN test-names loop
	(let loop ((hed (car test-names))   ;; NOTE: This is the main loop that iterates over the test-names
		   (tal (cdr test-names)))         ;; 'return-procs tells the config reader to prep running system but return a proc
          (debug:print-info 4 *default-log-port* "\n\ntestpatt elaboration loop => hed="hed " tal="tal" test-patts="test-patts" test-names="test-names)
	  (change-directory *toppath*) ;; PLEASE OPTIMIZE ME!!! I think this should be a no-op but there are several places where change-directories could be happening.
	  (setenv "MT_TEST_NAME" hed) ;; 
	  (let*-values (((waitons waitors config) (tests:get-waitons hed all-tests-registry))

			;; NOTE: Have the config - can extract [waitons] section
			
                        ((hed-mode)
                         (let ((m (configf:lookup config "requirements" "mode")))
                           (if m (map string->symbol (string-split m)) '(normal))))

     ;;       - prereqs are completed and passed (we could consider removing "and passed" -- it would change behavior from current)          
    ((or (null? prereqs-not-met)
	  (and (member 'toplevel testmode)
	       (null? non-completed)))
      (debug:print-info 4 *default-log-port* "cond branch - "  "ei-2")
      (debug:print-info 4 *default-log-port* "runs:expand-items: (or (null? prereqs-not-met) (and (member 'toplevel testmode)(null? non-completed)))")
      (let ((test-name (tests:testqueue-get-testname test-record)))
	(setenv "MT_TEST_NAME" test-name) ;; 
	(setenv "MT_RUNNAME"   runname)
	(runs:set-megatest-env-vars run-id inrunname: runname) ;; these may be needed by the launching process
	(let ((items-list (items:get-items-from-config tconfig)))
	  (if (list? items-list)
	      (begin
		(if (null? items-list)
		    (let ((test-id   (rmt:get-test-id run-id test-name ""))
			  (num-items (rmt:test-toplevel-num-items run-id test-name)))

     (lambda (vardat)
       (let ((var (car vardat))
	     (val (cdr vardat)))
	 (if (not (equal? (get-environment-variable var) val))
	     (handle-exceptions
	      exn
	      (debug:print-error 0 *default-log-port* "Failed to set " var " to " val)
	      (setenv var val)))))
     all-vars)
        ))

;;======================================================================
;; END OF NEW STUFF
;;======================================================================

	 (state-status (if (string? new-state-status) (string-split new-state-status ",") '(#f #f)))
	 (rp-mutex     (make-mutex))
	 (bup-mutex    (make-mutex))
         (keep-records (args:get-arg "-keep-records"))) ;; used in conjunction with -remove-runs to keep the records, TODO: consolidate this with "mode".

    (let* ((write-access-actions '(remove-runs set-state-status archive run-wait kill-runs))
           (dbfile             (conc  *toppath* "/megatest.db"))
           (readonly-mode      (not (file-write-access? dbfile))))
      (when (and readonly-mode
                 (member action write-access-actions))
        (debug:print-error 0 *default-log-port* "megatest.db is readonly.  Cannot proceed with action ["action"] in which write-access isrequired .")
        (exit 1)))
    
    (debug:print-info 4 *default-log-port* "runs:operate-on => Header: " header " action: " action " new-state-status: " new-state-status)
    (if (> 2 (length state-status))

			    #t)
			  (process-signal pid signal/int)
			  (thread-sleep! 5)
			  (if (process:alive? pid)
			      (process-signal pid signal/kill)))))
		   ;;  (call-with-environment-variables
		   (let ((old-targethost (getenv "TARGETHOST")))
		     (setenv "TARGETHOST" hostname)
		     (setenv "TARGETHOST_LOGF" "server-kills.log")
		     (system (conc "nbfake kill " pid))
		     (if old-targethost (setenv "TARGETHOST" old-targethost))
		     (unsetenv "TARGETHOST")
		     (unsetenv "TARGETHOST_LOGF"))))
	     (debug:print-error 0 *default-log-port* "no record or improper record for " target "/" run-name " in tasks_queue in main.db"))))
     records)))

(define (task:get-run-times)
   (let* ( 
           (run-patt (if (args:get-arg "-run-patt")
                        (args:get-arg "-run-patt")

	 (maxload (if force-maxload
		      maxload-in
		      (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))
         (adjwait (min (+ 300 (random 10)) (abs (* (+ (random 10)(/ (- 1000 count) 10) waitdelay) (- first adjload) ))  ))) ;; add some randomness to the time to break any alignment where netbatch dumps many jobs to machines simultaneously
    (debug:print-info 1 *default-log-port* "Checking cpuload on " (or remote-host "localhost") ", maxload: " maxload
		      ", load: " first ", adjload: " adjload ", loadjmp: " loadjmp)
    (cond
     ((and (> first adjload)
	   (> count 0))
      (debug:print-info 0 *default-log-port* "server start delayed " adjwait " 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! adjwait)

;; 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)

  (let* ((ns            (server:get-num-servers))
	 (servers       (server:get-best (server:get-list areapath))))
    ;; (print "servers: " servers " ns: " ns)
    (if (or (and servers
		 (null? servers))
	    (not servers)
	    (and (list? servers)
		 (< (length servers) (random ns)))) ;; somewhere between 0 and numservers
        #f
        (let loop ((hed (car servers))
                   (tal (cdr servers)))
          (let ((res (server:check-server hed)))
            (if res
                res
                (if (null? tal)

    (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)
    (thread-sleep! (/ (random 5000) 1000)) ;; add about a random (up to 5 seconds) initial delay. It seems pretty common that many running tests request a server at the same time
    (common:wait-for-normalized-load load-limit " delaying server start due to load" target-host) ;; 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)))

;; kind start up of servers, wait 40 seconds before allowing another server for a given
;; run-id to be launched
#;(define (server:kind-run areapath)
  (if (not (server:check-if-running areapath)) ;; why try if there is already a server running?
      (let* ((last-run-dat (hash-table-ref/default *server-kind-run* areapath '(0 0))) ;; callnum, whenrun
	     (call-num     (car last-run-dat))
	     (when-run     (cadr last-run-dat))
	     (run-delay    (+ (case call-num
				((0)    0)
				((1)   20)
				((2)  300)
				(else 600))
			      (random 5)))   ;; add a small random number just in case a lot of jobs hit the work hosts simultaneously
	     (lock-file    (conc areapath "/logs/server-start.lock")))
	(if	(> (- (current-seconds) when-run) run-delay)
		(begin
		  (common:simple-file-lock-and-wait lock-file expire-time: 15)
		  (server:run areapath)
		  (thread-sleep! 2) ;; don't release the lock for at least a few seconds
		  (common:simple-file-release-lock lock-file)))

		(if (> my-start-time (handle-exceptions
					 exn
					 0
				       (file-modification-time lockf)))
		    ;; we started since current re-gen in flight, delay a little and try again
		    (begin
		      (debug:print-info 1 *default-log-port* "Waiting to update " outputfilename ", another test currently updating it")
		      (thread-sleep! (+ 5 (random 5))) ;; delay between 5 and 10 seconds
		      (loop (common:simple-file-lock lockf))))))))))

(define (tests:generate-html-summary-for-iterated-test run-id test-id test-name outputfilename)
  (let ((counts              (make-hash-table))
	(statecounts         (make-hash-table))
	(outtxt              "")
	(tot                 0)

						      '()
						      (lambda (x p)
							(let* ((targ-path (string-intersperse p "/"))
                                                               (full-path (conc linktree "/" targ-path))
                                                               (run-name  (car (reverse p))))
                                                          (if (and (common:file-exists? full-path)
                                                                   (directory?   full-path)
                                                                   (file-write-access? full-path))
                                                              (s:a run-name 'href (conc targ-path "/run-summary.html"))
                                                              (begin
                                                                (debug:print 0 *default-log-port* "INFO: Can't create " targ-path "/run-summary.html")
                                                                (conc run-name " (Not able to create summary at " targ-path ")")))))))))))
          (close-output-port oup)
	  (common:simple-file-release-lock lockfile)
               

                                           path-parts))
                                       test-dats))
                    (tests-htree (common:list->htree tests-tree-dat))
                    (html-dir    (conc linktree "/" (string-intersperse run-dir "/")))
                    (html-path   (conc html-dir "/run-summary.html"))
                    (oup         (if (and (common:file-exists? html-dir)
                                          (directory?   html-dir)
                                          (file-write-access? html-dir))
                                     (open-output-file  html-path)
                                     #f)))
               ;; (print "run-dir: " run-dir ", tests-tree-dat: " tests-tree-dat)
               (if oup
                   (begin
                     (s:output-new
                      oup

                                                                          (alt-file  (conc full-targ "/megatest-rollup-" test-name ".html"))
                                                                          (html-file (if (common:file-exists? alt-file)
                                                                                         alt-file
                                                                                         std-file))
                                                                          (run-name  (car (reverse p))))
                                                                     (if (and (not (common:file-exists? full-targ))
                                                                              (directory? full-targ)
                                                                              (file-write-access? full-targ))
                                                                         (tests:summarize-test 
                                                                          run-id 
                                                                          (rmt:get-test-id run-id test-name item-path)))
                                                                     (if (common:file-exists? full-targ)
                                                                         (s:a run-name 'href html-file)
                                                                         (begin
                                                                           (debug:print 0 *default-log-port* "ERROR: can't access " full-targ)

;; summarize test in to a file test-summary.html in the test directory
;;
(define (tests:summarize-test run-id test-id)
  (let* ((test-dat  (rmt:get-test-info-by-id run-id test-id))
	 (out-dir   (db:test-get-rundir test-dat))
	 (out-file  (conc out-dir "/test-summary.html")))
    ;; first verify we are able to write the output file
    (if (not (file-write-access? out-dir))
	(debug:print 0 *default-log-port* "ERROR: cannot write test-summary.html to " out-dir)
	(let* (;; (steps-dat (rmt:get-steps-for-test run-id test-id))
	       (test-name (db:test-get-testname test-dat))
	       (item-path (db:test-get-item-path test-dat))
	       (full-name (db:test-make-full-name test-name item-path))
	       (oup       (open-output-file out-file))
	       (status    (db:test-get-status   test-dat))

    (if enccmd
	(common:read-encoded-string enccmd)
	'())))

;; 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  (configf: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"))

    
    (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
			      ((and (eq? process-exit-status 3) logpro-used) 'check)  ;; logpro 3 = check

			     (prevstep #f))
                    (debug:print-info 0 *default-log-port* "Processing ezstep \"" (string-intersperse ezstep " ") "\"")
		    ;; 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 0 *default-log-port* "WARNING: step " (car ezstep) " failed. Stopping")))
			(debug:print 0 *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* ((target    (or intarget 
			(common:args-get-target)
			(get-environment-variable "MT_TARGET")))
	 (keys      (if inkeys    inkeys    (rmt:get-keys)))
	 (keyvals   (if inkeyvals inkeyvals (keys:target->keyval keys target)))
	 (vals      (hash-table-ref/default *env-vars-by-run-id* run-id #f))
	 (link-tree (common:get-linktree))) ;; (configf:lookup *configdat* "setup" "linktree")))
    (if testname (setenv "MT_TEST_NAME" testname))
    (if itempath (setenv "MT_ITEMPATH"  itempath))

    ;; get the info from the db and put it in the cache
    (if link-tree
	(setenv "MT_LINKTREE" link-tree)
	(debug:print-error 0 *default-log-port* "linktree not set, should be set in megatest.config in [setup] section."))
    (if (not vals)
	(let ((ht (make-hash-table)))
	  (hash-table-set! *env-vars-by-run-id* run-id ht)
	  (set! vals ht)
	  (for-each
	   (lambda (key)
	     (hash-table-set! vals (car key) (cadr key)))
	   keyvals)))
    ;; from the cached data set the vars
    
    (hash-table-for-each
     vals
     (lambda (key val)
       (debug:print 2 *default-log-port* "setenv " key " " val)
       (safe-setenv key val)))
    ;;(bb-check-path msg: "runs:set-megatest-env-vars block 1")
    ;;(BB> "*env-vars-by-run-id*/runid("run-id" vals="(hash-table->alist vals))

    (if (not (get-environment-variable "MT_TARGET"))(setenv "MT_TARGET" target))
    ;; we had a case where there was an exception generated by the hash-table-ref
    ;; due to *configdat* being #f Adding a handle and exit
    (let fatal-loop ((count 0)) 
      (handle-exceptions
	  exn
	  (let ((call-chain (get-call-chain))
		(msg        ((condition-property-accessor 'exn 'message) exn)))

              (thread-sleep! 2) ;; assuming nfs lag.
              (launch:setup force-reread: #t))
          (alist->env-vars (hash-table-ref/default *configdat* "env-override" '())))) ;;;; environment is tainted HERE in this let block.
    ;;(bb-check-path msg: "runs:set-megatest-env-vars block 2")
    ;; Lets use this as an opportunity to put MT_RUNNAME in the environment
    (let ((runname  (if inrunname inrunname (rmt:get-run-name-from-id run-id))))
      (if runname
	  (setenv "MT_RUNNAME" runname)
	  (debug:print-error 0 *default-log-port* "no value for runname for id " run-id)))
    (setenv "MT_RUN_AREA_HOME" *toppath*)
    ;; if a testname and itempath are available set the remaining appropriate variables
    (if testname (setenv "MT_TEST_NAME" testname))
    (if itempath (setenv "MT_ITEMPATH"  itempath))
    ;;(bb-check-path msg: "runs:set-megatest-env-vars block 3")
    (if (and testname link-tree)
	(setenv "MT_TEST_RUN_DIR" (conc (getenv "MT_LINKTREE")  "/"
					(getenv "MT_TARGET")    "/"
					(getenv "MT_RUNNAME")   "/"
					(getenv "MT_TEST_NAME")
					(if (and itempath
						 (not (equal? itempath "")))
					    (conc "/" itempath)
					    ""))))))

(define (launch:execute encoded-cmd)
  (let* ((cmdinfo    (common:read-encoded-string encoded-cmd))
	 (tconfigreg #f))
    (setenv "MT_CMDINFO" encoded-cmd)
    ;;(bb-check-path msg: "launch:execute incoming")
    (if (list? cmdinfo) ;; ((testpath /tmp/mrwellan/jazzmind/src/example_run/tests/sqlitespeed)
	;; (test-name sqlitespeed) (runscript runscript.rb) (db-host localhost) (run-id 1))
	(let* ((testpath  (assoc/default 'testpath  cmdinfo))  ;; testpath is the test spec area
	       (top-path  (assoc/default 'toppath   cmdinfo))
	       (work-area (assoc/default 'work-area cmdinfo))  ;; work-area is the test run area
	       (test-name (assoc/default 'test-name cmdinfo))

	       (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
               (check-work-area           (lambda ()
                                            ;; 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))

                                                    (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)))
	       ) ;; let*

	  (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)
          (change-directory *toppath*) ;; temporarily switch to the run area home
	  (setenv "MT_TEST_RUN_DIR"  work-area)

	  (launch:setup) ;; should be properly in the run area home now

	  (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)
          (change-directory *toppath*) ;; temporarily switch to the run area home
	  (setenv "MT_TEST_RUN_DIR"  work-area)

	  (launch:setup) ;; should be properly in the run area home now
          
	  (set! tconfigreg (tests:get-all)) ;; mapping of testname => test source path
	  (let ((sighand (lambda (signum)
			   ;; (signal-mask! signum) ;; to mask or not? seems to cause issues in exiting
			   (if (eq? signum signal/stop)

		(debug:print 4 *default-log-port* "varpairs: " varpairs)
		(map (lambda (varpair)
		       (let ((varval (string-split varpair "=")))
			 (if (eq? (length varval) 2)
			     (let ((var (car varval))
				   (val (cadr varval)))
			       (debug:print 1 *default-log-port* "Adding pre-var/val " var " = " val " to the environment")
			       (setenv var val)))))
		     varpairs)))
          ;;(bb-check-path msg: "launch:execute post block 2")
	  (for-each
	   (lambda (varval)
	     (let ((var (car varval))
		   (val (cadr varval)))
	       (if val
		   (setenv var val)
		   (begin
		     (debug:print-error 0 *default-log-port* "required variable " var " does not have a valid value. Exiting")
		     (exit)))))
	     (list 
	      (list  "MT_TEST_RUN_DIR" work-area)
	      (list  "MT_TEST_NAME" test-name)
	      (list  "MT_ITEM_INFO" (conc itemdat))
	      (list  "MT_ITEMPATH"  item-path)
	      (list  "MT_RUNNAME"   runname)
	      (list  "MT_MEGATEST"  megatest)
	      (list  "MT_TARGET"    target)
	      (list  "MT_LINKTREE"  (common:get-linktree)) ;; (configf:lookup *configdat* "setup" "linktree"))
	      (list  "MT_TESTSUITENAME" (common:get-testsuite-name))))
          ;;(bb-check-path msg: "launch:execute post block 3")

	  (if mt-bindir-path (setenv "PATH" (conc (getenv "PATH") ":" mt-bindir-path)))
          ;;(bb-check-path msg: "launch:execute post block 4")
	  ;; (change-directory top-path)
	  ;; Can setup as client for server mode now
	  ;; (client:setup)

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

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

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

	  ;; We are about to actually kick off the test
	  ;; so this is a good place to remove the records for 
	  ;; any previous runs
	  ;; (db:test-remove-steps db run-id testname itemdat)
	  ;; now is also a good time to write the .testconfig file
	  (let* ((tconfig-fname   (conc work-area "/.testconfig"))
		 (tconfig-tmpfile (conc tconfig-fname ".tmp"))
		 (tconfig         (tests:get-testconfig test-name item-path tconfigreg #t force-create: #t))) ;; 'return-procs)))
	    (configf:write-alist tconfig tconfig-tmpfile)
	    (file-move tconfig-tmpfile tconfig-fname #t))
	  ;; 
	  (let* ((m            (make-mutex))
		 (kill-job?    #f)
		 (exit-info    (make-launch:einf pid: #t exit-status: #t exit-code: #t rollup-status: 0)) ;; pid exit-status exit-code (i.e. process was successfully run) rollup-status
		 (job-thread   #f)
		 ;; (keep-going   #t)
		 (misc-flags   (let ((ht (make-hash-table)))

		  (set! *configinfo* first-pass)
		  (set! *toppath*    (or toppath (cadr first-pass))) ;; use the gathered data unless already have it
		  (set! toppath      *toppath*)
		  (if (not *toppath*)
		      (begin
			(debug:print-error 0 *default-log-port* "you are not in a megatest area!")
			(exit 1)))
		  (setenv "MT_RUN_AREA_HOME" *toppath*)
		  ;; the seed read is done, now read runconfigs, cache it then read megatest.config one more time and cache it
		  (let* ((keys         (rmt:get-keys))
			 (key-vals     (keys:target->keyval keys target))
			 (linktree     (common:get-linktree)) ;; (or (getenv "MT_LINKTREE")(if *configdat* (configf:lookup *configdat* "setup" "linktree") #f)))
					;     (if *configdat*
					; 	   (configf:lookup *configdat* "setup" "linktree")
					; 	   (conc *toppath* "/lt"))))
			 (second-pass  (configf:find-and-read-config
					mtconfig
					environ-patt: "env-override"
					given-toppath: toppath
					pathenvvar: "MT_RUN_AREA_HOME"))
			 (runconfigdat (begin     ;; this read of the runconfigs will see any adjustments made by re-reading megatest.config
					 (for-each (lambda (kt)
						     (setenv (car kt) (cadr kt)))
						   key-vals)
					 (configf:read-config (conc toppath "/runconfigs.config") *runconfigdat* #t ;; consider using runconfig:read some day ...
						      sections: sections)))
                         (cachefiles   (launch:get-cache-file-paths areapath toppath target mtconfig))
                         (mtcachef     (car cachefiles))
                         (rccachef     (cdr cachefiles)))
                    ;;  trap exception due to stale NFS handle -- Error: (open-output-file) cannot open file - Stale NFS file handle: "/p/fdk/gwa/lefkowit/mtTesting/qa/primbeqa/links/p1222/11/PDK_r1.1.1/prim/clean/pcell_testgen/.runconfigs.cfg-1.6427-7d1e789cb3f62f9cde719a4865bb51b3c17ea853" - ticket 220546342

			(create-symbolic-link linktree tlink)))))
	      (begin
		(debug:print-error 0 *default-log-port* "linktree not defined in [setup] section of megatest.config")
		)))
	(if (and *toppath*
		 (directory-exists? *toppath*))
	    (begin
	      (setenv "MT_RUN_AREA_HOME" *toppath*)
	      (setenv "MT_TESTSUITENAME" (common:get-testsuite-name)))
	    (begin
	      (debug:print-error 0 *default-log-port* "failed to find the top path to your Megatest area.")
	      (set! *toppath* #f) ;; force it to be false so we return #f
	      #f))
	
        ;; one more attempt to cache the configs for future reading
        (let* ((cachefiles   (launch:get-cache-file-paths areapath toppath target mtconfig))

              )
          (if (and rccachef mtcachef *runconfigdat* *configdat*)
              (set! *configstatus* 'fulldata)))

	;; if have -append-config then read and append here
	(let ((cfname (args:get-arg "-append-config")))
	  (if (and cfname
		   (file-read-access? cfname))
	      (configf:read-config cfname *configdat* #t))) ;; values are added to the hash, no need to do anything special.
	*toppath*)))

(define (get-best-disk confdat testconfig)
  (let* ((disks   (or (and testconfig (hash-table-ref/default testconfig "disks" #f))
		      (hash-table-ref/default confdat "disks" #f)))
	 (minspace (let ((m (configf:lookup confdat "setup" "minspace")))

  (let* ((rundir (if (and (getenv "MT_LINKTREE")(getenv "MT_TARGET")(getenv "MT_RUNNAME"))
		     (conc (getenv "MT_LINKTREE") "/" (getenv "MT_TARGET") "/" (getenv "MT_RUNNAME"))
		     #f))
	 (cfgf   (if rundir (conc rundir "/.runconfig." megatest-version "-" megatest-fossil-hash) #f)))
    (if (and cfgf
	     (common:file-exists? cfgf)
	     (file-write-access? cfgf)
	     (common:use-cache?))
	(configf:read-alist cfgf)
	(let* ((keys   (rmt:get-keys))
	       (target (common:args-get-target))
	       (key-vals (if target (keys:target->keyval keys target) #f))
	       (sections (if target (list "default" target) #f))
	       (data     (begin
			   (setenv "MT_RUN_AREA_HOME" *toppath*)
			   (if key-vals
			       (for-each (lambda (kt)
					   (setenv (car kt) (cadr kt)))
					 key-vals))
			   ;; (read-config (conc *toppath* "/runconfigs.config") #f #t sections: sections))))
                           (runconfig:read (conc *toppath* "/runconfigs.config") target #f))))
	  (if (and rundir ;; have all needed variabless
		   (directory-exists? rundir)
		   (file-write-access? rundir))
	      (begin
                (if (not (common:in-running-test?))
                    (configf:write-alist data cfgf))
		;; force re-read of megatest.config - this resolves circular references between megatest.config
		(launch:setup force-reread: #t)
		;; (launch:cache-config) ;; there are two independent config cache locations, turning this one off for now. MRW.
		)) ;; we can safely cache megatest.config since we have a valid runconfig

(use z3)
(use base64)

(let* ((env-str        (with-output-to-string (lambda () (pp (get-environment-variables)))))
       (zipped-env-str (z3:encode-buffer env-str))
       (b64-env-str    (base64-encode zipped-env-str)))
  (print b64-env-str))

(declare (uses mtconfigf))
(declare (uses mtargs))
(declare (uses tasksmod))

(module servermod
	*
	
(import scheme chicken data-structures extras files)
(import (prefix sqlite3 sqlite3:) posix typed-records srfi-18
	srfi-69 format ports srfi-1 matchable
	directory-utils md5 message-digest regex



	stack)
(import commonmod)
(import dbmod)
(import tasksmod)
(import (prefix mtargs args:))
(import (prefix mtconfigf configf:))

          (delete-file* (common:get-sync-lock-filepath))
          )
	 
	 ;; clear out junk records
	 ;;
	 ((dejunk)
	  (db:delay-if-busy mtdb) ;; ok to delay on mtdb
	  (when (file-write-access? (db:dbdat-get-path mtdb)) (db:clean-up mtdb))
	  (db:clean-up tmpdb)
	  (db:clean-up refndb))

	 ;; sync runs, test_meta etc.
	 ;;
	 ((old2new)
	  (set! data-synced

  (let ((fname-rx    (regexp "^(|.*/)server-(\\d+)-(\\S+).log$"))
	(day-seconds (* 24 60 60)))
    ;; if the directory exists continue to get the list
    ;; otherwise attempt to create the logs dir and then
    ;; continue
    (if (if (directory-exists? (conc areapath "/logs"))
	    '()
	    (if (file-write-access? areapath)
		(begin
		  (condition-case
		      (create-directory (conc areapath "/logs") #t)
		    (exn (i/o file)(debug:print 0 *default-log-port* "ERROR: Cannot create directory at " (conc areapath "/logs")))
		    (exn ()(debug:print 0 *default-log-port* "ERROR: Unknown error attemtping to get server list.")))
		  (directory-exists? (conc areapath "/logs")))
		'()))

				;; (print "start-time: " start-time " mod-time: " mod-time)
				(and start-time mod-time
				     (> (- now start-time) 0)    ;; been running at least 0 seconds
				     (< (- now mod-time)   16)   ;; still alive - file touched in last 16 seconds
				     (< (- now start-time)       
					(+ (- (string->number (or (configf:lookup *configdat* "server" "runtime") "3600"))
					      180)
					   (random 360))) ;; under one hour running time +/- 180
				     ))
			      #f))
			srvlst)
		(lambda (a b)
		  (< (list-ref a 3)
		     (list-ref b 3))))))
    (if (> (length slst) nums)

	#f)))

(define (server:get-rand-best areapath)
  (let ((srvrs (server:get-best (server:get-list areapath))))
    (if (and (list? srvrs)
	     (not (null? srvrs)))
	(let* ((len (length srvrs))
	       (idx (random len)))
	  (list-ref srvrs idx))
	#f)))


(define (server:record->url servr)
  (match-let (((mod-time host port start-time pid)
	       servr))

	(begin
	  (print "[database]\nlocation /some/path\n\n Is missing from the config file!")
	  (exit 1)))
    (if (and path
	     (directory? path)
	     (file-read-access? path))
	(let* ((dbpath    (conc path "/spublish.db"))
	       (writeable (file-write-access? dbpath))
	       (dbexists  (file-exists? dbpath)))
	  (handle-exceptions
	   exn
	   (begin
	     (debug:print 2 *default-log-port* "ERROR: problem accessing db " dbpath
			  ((condition-property-accessor 'exn 'message) exn))
	     (exit 1))

;;   <http://www.netscape.com/newsref/std/cookie_spec.html>

;; (declare (unit cookie))

(module cookie
    *

(import chicken scheme data-structures extras srfi-13 ports posix)
  
(require-extension srfi-1 srfi-13 srfi-14 regex)
;; (use  srfi-1 srfi-13 srfi-14 regex)
;; (declare (export parse-cookie-string construct-cookie-string))

;; #>
;; #include <time.h>

;; stml is a list of html strings

;; (declare (unit stml))

(module stml2
    *

(import chicken scheme data-structures extras srfi-13 ports posix srfi-69 files srfi-1) 

(import cookie)
(use (prefix dbi dbi:) (prefix crypt c:) typed-records)

;; (declare (uses misc-stml))
(use regex)

;; The (usually global) sdat contains everything about the session
;;
(defstruct sdat
  ;; database
  (dbtype 'pg)
  (dbinit #f)

;; to obscure and indirect database ids use one time keys
;;
;;  (s:get-key 'n 1)     => "n99e1882" n=number 99e is the week number since 1970, remainder is random
;;  (s:key->val "n1882") => 1
;;
;;  first letter is a type: n=number, s=string, b=boolean
(define (s:get-key key-type val)
  (let ((mkrandstr (lambda (innum)(number->string (random innum) 16)))
	(week      (number->string (quotient (current-seconds) (* 7 24 60 60)) 16)))
    (let loop ((siz 1000)
	       (key (conc key-type week (mkrandstr 100)))
	       (num 0))
      (if (s:session-var-get key) ;; have a collision
	  (loop (cond                 ;; in the unlikey event we have trouble getting a new var, keep increasing the size of the number
		 ((< num 50)  100)

#;(define session:valid-chars "abcdefghijklmnopqrstuvwxyz0123456789") ;; cookies are case insensitive.
#;(define session:num-valid-chars (string-length session:valid-chars))

#;(define (session:get-nth-char nth)
  (substring session:valid-chars nth  (+ nth 1)))

#;(define (session:get-rand-char)
  (session:get-nth-char (random session:num-valid-chars)))

#;(define (session:make-rand-string len)
  (let loop ((res "")
             (n   1))
    (if (> n len) res
        (loop (string-append res (session:get-rand-char))
              (+ n 1)))))

;; maybe replace above make-rand-string with this someday?
;;
#;(define (session:generic-make-rand-string len seed-string)
  (let ((num-chars (string-length seed-string)))
    (let loop ((res "")
	       (n   1))
      (let ((char-num (random num-chars)))
	(if (> n len) res
	    (loop (string-append res (substring seed-string char-num (+ char-num 1)))
		  (+ n 1)))))))

;; Rely on crypt egg's default settings being secure enough, accept
;; backwards-compatible OpenSSL crypt passwords too.
;;

(define session:valid-chars "abcdefghijklmnopqrstuvwxyz0123456789") ;; cookies are case insensitive.
(define session:num-valid-chars (string-length session:valid-chars))

(define (session:get-nth-char nth)
  (substring session:valid-chars nth  (+ nth 1)))

(define (session:get-rand-char)
  (session:get-nth-char (random session:num-valid-chars)))

(define (session:make-rand-string len)
  (let loop ((res "")
             (n   1))
    (if (> n len) res
        (loop (string-append res (session:get-rand-char))
              (+ n 1)))))

;; maybe replace above make-rand-string with this someday?
;;
(define (session:generic-make-rand-string len seed-string)
  (let ((num-chars (string-length seed-string)))
    (let loop ((res "")
	       (n   1))
      (let ((char-num (random num-chars)))
	(if (> n len) res
	    (loop (string-append res (substring seed-string char-num (+ char-num 1)))
		  (+ n 1)))))))


;;======================================================================
;; P A R A M S

      (if debugmode (session:log self "session:setup dbfname=" dbfname ", dbtype=" dbtype ", dbinit=" dbinit))
      (if (eq? dbtype 'sqlite3)
	  ;; The 'auto method will distribute dbs across the disk using hash
	  ;; of user host and user. TODO
	  ;; (if (eq? dbfname 'auto) ;; This is the auto assignment of a db based on hash of IP
	  (let ((dbpath (pathname-directory dbfname)))  ;; do a couple sanity checks here to make setting up easier
	    (if debugmode (session:log self "INFO: setting up for sqlite3 db access to " dbfname))
	    (if (not (file-write-access? dbpath))
		(session:log self "WARNING: Cannot write to " dbpath)
		(if debugmode (session:log self "INFO: " dbpath " is writeable")))
	    (if (file-exists? dbfname)
		(begin
		  ;; (session:log self "setting dbexists to #t")
		  (set! dbexists #t))))
	  (if debugmode (session:log self "INFO: setting up for pg db access to account info " dbinit)))

(declare (uses commonmod))
(declare (uses mtconfigf))

(module subrunmod
	*
	
(import scheme chicken data-structures extras)
(use (prefix sqlite3 sqlite3:) posix typed-records srfi-18
	srfi-69 format ports srfi-1 matchable irregex

	call-with-environment-variables)

(import commonmod)
(import  (prefix mtconfigf configf:))

;; (use (prefix ulex ulex:))

(declare (uses commonmod))
(declare (uses pgdbmod))
(declare (uses mtconfigf))

(module tasksmod
	*
	
(import scheme chicken data-structures extras)
(use (prefix sqlite3 sqlite3:) posix typed-records srfi-18 srfi-69 format ports srfi-1 matchable
	regex)

(import commonmod)
(import	(prefix mtconfigf configf:))
(import	pgdbmod)

(include "common_records.scm")

	     (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))))
       (let* ((dbpath        (db:dbfile-path )) ;; (tasks:get-task-db-path))
	      (dbfile       (conc dbpath "/monitor.db"))
	      (avail        (tasks:wait-on-journal dbpath 10)) ;; wait up to about 10 seconds for the journal to go away
	      (exists       (common:file-exists? dbpath))
	      (write-access (file-write-access? dbpath))
	      (mdb          (cond ;; what the hek is *toppath* doing here?
			     ((and (string? *toppath*)(file-write-access? *toppath*))
			      (sqlite3:open-database dbfile))
			     ((file-read-access? dbpath)    (sqlite3:open-database dbfile))
			     (else (sqlite3:open-database ":memory:")))) ;; (never-give-up-open-db dbpath))
	      (handler      (sqlite3:make-busy-timeout 36000)))
	 (if (and exists
		  (not write-access))
	     (set! *db-write-access* write-access)) ;; only unset so other db's also can use this control
	 (sqlite3:set-busy-handler! mdb handler)
	 (db:set-sync mdb) ;; (sqlite3:execute mdb (conc "PRAGMA synchronous = 0;"))

(define (tasks:need-server run-id)
  (equal? (configf:lookup *configdat* "server" "required") "yes"))

;; no elegance here ...
;;
(define (tasks:kill-server hostname pid #!key (kill-switch ""))
  (debug:print-info 0 *default-log-port* "Attempting to kill server process " pid " on host " hostname)
  (setenv "TARGETHOST" hostname)
  (let* ((logdir (if (directory-exists? "logs")
                    "logs/"
                    ""))
         (logfile (if logdir (conc "logs/server-"pid"-"hostname".log") #f))
         (gzfile  (if logfile (conc logfile ".gz"))))
    (setenv "TARGETHOST_LOGF" (conc logdir "server-kills.log"))

    (system (conc "nbfake kill "kill-switch" "pid))

    (when logfile
      (thread-sleep! 0.5)
      (if (common:file-exists? gzfile) (delete-file gzfile))
      (system (conc "gzip " logfile))
      
      (unsetenv "TARGETHOST_LOGF")
      (unsetenv "TARGETHOST"))))
    
 
;;======================================================================
;; M O N I T O R S
;;======================================================================

(define (tasks:remove-monitor-record mdb)

(declare (unit rmtmod))
(declare (uses commonmod))

(module rmtmod
	*
	
(import scheme chicken data-structures extras)
(import (prefix sqlite3 sqlite3:) posix typed-records srfi-18 srfi-69 format ports srfi-1 matchable)
(import commonmod)
;; (use (prefix ulex ulex:))

(include "common_records.scm")


)

(declare (uses mtconfigf))
(declare (uses itemsmod))
(declare (uses dbmod))

(module testsmod
	*
	
(import scheme chicken data-structures extras files)

(use (prefix sqlite3 sqlite3:) posix typed-records srfi-18 srfi-69
	format ports srfi-1 matchable
	directory-utils
	regex srfi-13)


(import	commonmod)
(import	servermod)
(import	itemsmod)

;; The setting of :keyfield in args should be turned off ASAP
;;
(define (keys:target-set-args keys target ht)
  (if target
      (let ((vals (string-split target "/")))
	(if (eq? (length vals)(length keys))
	    (for-each (lambda (key val)
			(setenv key val)
			(if ht (hash-table-set! ht (conc ":" key) val)))
		      keys
		      vals)
	    (debug:print-error 0 *default-log-port* "wrong number of values in " target ", should match " keys))
	vals)
      (debug:print 4 *default-log-port* "ERROR: keys:target-set-args called with no target.")))

                                             local-tcdir
                                             #f))
				       (conc *toppath* "/tests/" test-name)))
		     (test-configf (conc test-path "/testconfig"))
		     (testexists   (let loopa ((tries-left 30))
                                     (cond
                                      (
                                       (and (common:file-exists? test-configf)(file-read-access? test-configf))
                                       #t)
                                      (
                                       (common:file-exists? test-configf)
                                       (debug:print 0 *default-log-port* "WARNING: Cannot read testconfig file: "test-configf)
                                       #f)
                                      (
                                       (and wait-a-minute (> tries-left 0))

								      "pre-launch-env-vars"
								      #f))
				       #f)))
		(if (and tcfg cache-file) (hash-table-set! tcfg "have fulldata" #t)) ;; mark this as fully read data
		(if tcfg (hash-table-set! *testconfigs* test-full-name tcfg))
		(if (and testexists
			 cache-file
			 (file-write-access? cache-path)
			 allow-write-cache)
		    (let ((tpath (conc cache-path "/.testconfig")))
		      (debug:print-info 1 *default-log-port* "Caching testconfig for " test-name " in " tpath)
                      (if (and tcfg (not (common:in-running-test?)))
                          (configf:write-alist tcfg tpath))))
		tcfg))))))
  

       (conc "env -i PATH=$PATH dot -T" outtype " < " temp-path)
       (lambda ()
	 (let ((res (read-lines)))
	   ;; (delete-file temp-path)
	   res))))))

(define (tests:write-dot-file test-records fname sizex sizey)
  (if (file-write-access? (pathname-directory fname))
      (with-output-to-file fname
	(lambda ()
	  (map print (tests:tests->dot test-records sizex sizey))))))

(define (tests:tests->dot test-records sizex sizey)
  (let ((all-testnames (hash-table-keys test-records)))
    (if (null? all-testnames)

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

(declare (unit treemod))

(module treemod
	*
	
(import scheme chicken data-structures extras)
(import (prefix iup iup:)) ;; (prefix sqlite3 sqlite3:) posix typed-records srfi-18 srfi-69 format ports srfi-1 matchable)
;; (import commonmod) ;;; DO NOT ALLOW rmt*scm TO DEPEND ON common*scm!!!!
;; (include "common_records.scm")

)

;;   See README in the distribution at https://www.kiatoa.com/fossils/ulex
;; NOTES:
;;   Why sql-de-lite and not say, dbi?  - performance mostly, then simplicity.
;;
;;======================================================================

;; (use rpc pkts mailbox sqlite3)
  
(module ulex
    *
     ;;    #;(
     ;;     ;; areas
     ;;     make-area
     ;;     area->alist
     ;;     ;; server
     ;;     launch
     ;;     update-known-servers
     ;;     shutdown
     ;;     get-best-server
     ;;     ;; client side
     ;;     call
     ;;     ;; queries, procs and system commands (i.e. workers)
     ;;     register
     ;;     register-batch
     ;;     ;; database - note: most database stuff need not be exposed, these calls may be removed from exports in future
     ;;     open-db
     ;;     ;; ports
     ;;     pl-find-port
     ;;     pl-get-prev-used-port
     ;;     pl-open-db
     ;;     pl-open-run-close
     ;;     pl-release-port
     ;;     pl-set-port
     ;;     pl-take-port
     ;;     pl-is-port-available
     ;;     pl-get-port-state
     ;;     ;; system
     ;;     get-normalized-cpu-load
	 
     ;; )

(import scheme posix chicken data-structures ports extras files mailbox)
(import rpc srfi-18 pkts matchable regex
	typed-records srfi-69 srfi-1
	srfi-4 regex-case
	(prefix sqlite3 sqlite3:)
	foreign
	tcp) ;; ulex-netutil)

;;======================================================================
;; D E B U G   H E L P E R S
;;======================================================================
    
(define (dbg> . args)
  (with-output-to-port (current-error-port)
    (lambda ()
      (apply print "dbg> " args))))

(define (debug-pp . args)
  (if (get-environment-variable "ULEX_DEBUG")
      (with-output-to-port (current-error-port)
	(lambda ()
	  (apply pp args)))))

(define *default-debug-port* (current-error-port))

(define (sdbg> fn stage-name stage-start stage-end start-time . message)
  (if (get-environment-variable "ULEX_DEBUG")
      (with-output-to-port *default-debug-port* 
	(lambda ()
	  (apply print "ulex:" fn " " stage-name " took " (- (if stage-end stage-end (current-milliseconds)) stage-start) " ms. "
		 (if start-time
		     (conc "total time " (- (current-milliseconds) start-time)
			   " ms.")
		     "")
		 message
		 )))))

;;======================================================================
;; M A C R O S
;;======================================================================
;; iup callbacks are not dumping the stack, this is a work-around
;;

;; Some of these routines use:
;;
;;     http://www.cs.toronto.edu/~gfb/scheme/simple-macros.html
;;
;; Syntax for defining macros in a simple style similar to function definiton,
;;  when there is a single pattern for the argument list and there are no keywords.
;;
;; (define-simple-syntax (name arg ...) body ...)
;;

(define-syntax define-simple-syntax
  (syntax-rules ()
    ((_ (name arg ...) body ...)
     (define-syntax name (syntax-rules () ((name arg ...) (begin body ...)))))))

(define-simple-syntax (catch-and-dump proc procname)
  (handle-exceptions
   exn
   (begin
     (print-call-chain (current-error-port))
     (with-output-to-port (current-error-port)
       (lambda ()
         (print ((condition-property-accessor 'exn 'message) exn))
         (print "Callback error in " procname)
         (print "Full condition info:\n" (condition->list exn)))))
   (proc)))


;;======================================================================
;;  R E C O R D S
;;======================================================================

;; information about me as a server
;;
(defstruct area
  ;; about this area
  (useportlogger #f)
  (lowport       32768)
  (server-type   'auto)  ;; auto=create up to five servers/pkts, main=create pkts, passive=no pkt (unless there are no pkts at all)
  (conn          #f)
  (port          #f)
  (myaddr        (get-my-best-address))
  pktid          ;; get pkt from hosts table if needed
  pktfile
  pktsdir
  dbdir
  (dbhandles     (make-hash-table)) ;; fname => list-of-dbh, NOTE: Should really never need more than one?
  (mutex         (make-mutex))
  (rtable        (make-hash-table)) ;; registration table of available actions
  (dbs           (make-hash-table)) ;; filename => random number, used for choosing what dbs I serve
  ;; about other servers
  (hosts         (make-hash-table)) ;; key => hostdat
  (hoststats     (make-hash-table)) ;; key => alist of fname => ( qcount . qtime )
  (reqs          (make-hash-table)) ;; uri => queue
  ;; work queues
  (wqueues       (make-hash-table)) ;; fname => qdat
  (stats         (make-hash-table)) ;; fname => totalqueries
  (last-srvup    (current-seconds)) ;; last time we updated the known servers
  (cookie2mbox   (make-hash-table)) ;; map cookie for outstanding request to mailbox of awaiting call
  (ready #f)
  (health        (make-hash-table)) ;; ipaddr:port => num failed pings since last good ping
  )

;; host stats
;;
(defstruct hostdat
  (pkt      #f)
  (dbload   (make-hash-table))  ;; "dbfile.db" => queries/min
  (hostload #f)                 ;; normalized load ( 5min load / numcpus )
  )

;; dbdat
;;
(defstruct dbdat
  (dbh    #f)
  (fname  #f)
  (write-access #f)
  (sths   (make-hash-table))  ;; hash mapping query strings to handles
  )

;; qdat
;;
(defstruct qdat
  (writeq  (make-queue))
  (readq   (make-queue))
  (rwq     (make-queue))
  (logq    (make-queue)) ;; do we need a queue for logging? yes, if we use sqlite3 db for logging
  (osshort (make-queue))
  (oslong  (make-queue))
  (misc    (make-queue)) ;; used for things like ping-full
  )

;; calldat
;;
(defstruct calldat
  (ctype 'dbwrite)
  (obj   #f)              ;; this would normally be an SQL statement e.g. SELECT, INSERT etc.
  (rtime (current-milliseconds)))

;; make it a global? Well, it is local to area module

(define *pktspec*
  `((server (hostname . h)
	    (port     . p)
	    (pid      . i)
	    (ipaddr   . a)
	    )
    (data   (hostname . h)  ;; sender hostname
	    (port     . p)  ;; sender port
	    (ipaddr   . a)  ;; sender ip
	    (hostkey  . k)  ;; sending host key - store info at server under this key
	    (servkey  . s)  ;; server key - this needs to match at server end or reject the msg
	    (format   . f)  ;; sb=serialized-base64, t=text, sx=sexpr, j=json
	    (data     . d)  ;; base64 encoded slln data
	    )))

;; work item
;;
(defstruct witem
  (rhost #f)   ;; return host
  (ripaddr #f) ;; return ipaddr
  (rport #f)   ;; return port
  (servkey #f) ;; the packet representing the client of this workitem, used by final send-message
  (rdat  #f)   ;; the request - usually an sql query, type is rdat
  (action #f)  ;; the action: immediate, dbwrite, dbread,oslong, osshort
  (cookie #f)  ;; cookie id for response
  (data   #f)  ;; the data payload, i.e. parameters
  (result #f)  ;; the result from processing the data
  (caller #f)) ;; the calling peer according to rpc itself

(define (trim-pktid pktid)
  (if (string? pktid)
      (substring pktid 0 4)
      "nopkt"))

(define (any->number num)
  (cond
   ((number? num) num)
   ((string? num) (string->number num))
   (else num)))

(use trace)
(trace-call-sites #t)

;;======================================================================
;; D A T A B A S E   H A N D L I N G 
;;======================================================================

;; look in dbhandles for a db, return it, else return #f
;;
(define (get-dbh acfg fname)
  (let ((dbh-lst (hash-table-ref/default (area-dbhandles acfg) fname '())))
    (if (null? dbh-lst)
	(begin
	  ;; (print "opening db for " fname)
	  (open-db acfg fname)) ;; Note that the handles get put back in the queue in the save-dbh calls
	(let ((rem-lst (cdr dbh-lst)))
	  ;; (print "re-using saved connection for " fname)
	  (hash-table-set! (area-dbhandles acfg) fname rem-lst)
	  (car dbh-lst)))))

(define (save-dbh acfg fname dbdat)
    ;; (print "saving dbh for " fname)
    (hash-table-set! (area-dbhandles acfg) fname (cons dbdat (hash-table-ref/default (area-dbhandles acfg) fname '()))))

;; open the database, if never before opened init it. put the handle in the
;; open db's hash table
;; returns: the dbdat
;;
(define (open-db acfg fname)
  (let* ((fullname     (conc (area-dbdir acfg) "/" fname))
	 (exists       (file-exists? fullname))
	 (write-access (if exists
			   (file-write-access? fullname)
			   (file-write-access? (area-dbdir acfg))))
	 (db           (sqlite3:open-database fullname))
	 (handler      (sqlite3:make-busy-timeout 136000))
	 )
    (sqlite3:set-busy-handler! db handler)
    (sqlite3:execute db "PRAGMA synchronous = 0;")
    (if (not exists) ;; need to init the db
	(if write-access
	    (let ((isql (get-rsql acfg 'dbinitsql))) ;; get the init sql statements
	      ;; (sqlite3:with-transaction
	      ;;  db
	      ;;  (lambda ()
		 (if isql
		     (for-each
		      (lambda (sql)
			(sqlite3:execute db sql))
		      isql)))
	    (print "ERROR: no write access to " (area-dbdir acfg))))
    (make-dbdat dbh: db fname: fname write-access: write-access)))

;; This is a low-level command to retrieve or to prepare, save and return a prepared statment
;; you must extract the db handle
;;
(define (get-sth db cache stmt)
  (if (hash-table-exists? cache stmt)
      (begin
	;; (print "Reusing cached stmt for " stmt)
	(hash-table-ref/default cache stmt #f))
      (let ((sth (sqlite3:prepare db stmt)))
	(hash-table-set! cache stmt sth)
	;; (print "prepared stmt for " stmt)
	sth)))

;; a little more expensive but does all the tedious deferencing - only use if you don't already
;; have dbdat and db sitting around
;;
(define (full-get-sth acfg fname stmt)
  (let* ((dbdat  (get-dbh acfg fname))
	 (db     (dbdat-dbh dbdat))
	 (sths   (dbdat-sths dbdat)))
    (get-sth db sths stmt)))

;; write to a db
;; acfg: area data
;; rdat: request data
;; hdat: (host . port)
;;
;; (define (dbwrite acfg rdat hdat data-in)
;;   (let* ((dbname (car data-in))
;; 	 (dbdat  (get-dbh acfg dbname))
;; 	 (db     (dbdat-dbh dbdat))
;; 	 (sths   (dbdat-sths dbdat))
;; 	 (stmt   (calldat-obj rdat))
;; 	 (sth    (get-sth db sths stmt))
;; 	 (data   (cdr data-in)))
;;     (print "dbname: " dbname " acfg: " acfg " rdat: " (calldat->alist rdat) " hdat: " hdat " data: " data)
;;     (print "dbdat: " (dbdat->alist dbdat))
;;     (apply sqlite3:execute sth data)
;;     (save-dbh acfg dbname dbdat)
;;     #t
;;     ))

(define (finalize-all-db-handles acfg)
  (let* ((dbhandles (area-dbhandles acfg))  ;; dbhandles is hash of fname ==> dbdat
	 (num       0))
    (for-each
     (lambda (area-name)
       (print "Closing handles for " area-name)
       (let ((dbdats (hash-table-ref/default dbhandles area-name '())))
	 (for-each
	  (lambda (dbdat)
	    ;; first close all statement handles
	    (for-each
	     (lambda (sth)
	       (sqlite3:finalize! sth)
	       (set! num (+ num 1)))
	     (hash-table-values (dbdat-sths dbdat)))
	    ;; now close the dbh
	    (set! num (+ num 1))
	    (sqlite3:finalize! (dbdat-dbh dbdat)))
	  dbdats)))
     (hash-table-keys dbhandles))
    (print "FINALIZED " num " dbhandles")))

;;======================================================================
;; W O R K   Q U E U E   H A N D L I N G 
;;======================================================================

(define (register-db-as-mine acfg dbname)
  (let ((ht (area-dbs acfg)))
    (if (not (hash-table-ref/default ht dbname #f))
	(hash-table-set! ht dbname (random 10000)))))
	
(define (work-queue-add acfg fname witem)
  (let* ((work-queue-start (current-milliseconds))
	 (action           (witem-action witem)) ;; NB the action is the index into the rdat actions
	 (qdat             (or (hash-table-ref/default (area-wqueues acfg) fname #f)
			       (let ((newqdat (make-qdat)))
				 (hash-table-set! (area-wqueues acfg) fname newqdat)
				 newqdat)))
	 (rdat             (hash-table-ref/default (area-rtable acfg) action #f)))
    (if rdat
	(queue-add!
	 (case (calldat-ctype rdat)
	   ((dbwrite)   (register-db-as-mine acfg fname)(qdat-writeq qdat))
	   ((dbread)    (register-db-as-mine acfg fname)(qdat-readq  qdat))
	   ((dbrw)      (register-db-as-mine acfg fname)(qdat-rwq    qdat))
	   ((oslong)    (qdat-oslong qdat))
	   ((osshort)   (qdat-osshort qdat))
	   ((full-ping) (qdat-misc  qdat))
	   (else
	    (print "ERROR: no queue for " action ". Adding to dbwrite queue.")
	    (qdat-writeq qdat)))
	 witem)
	(case action
	  ((full-ping)(qdat-misc qdat))
	  (else
	   (print "ERROR: No action " action " was registered"))))
    (sdbg> "work-queue-add" "queue-add" work-queue-start #f #f)
    #t)) ;; for now, simply return #t to indicate request got to the queue

(define (doqueue acfg q fname dbdat dbh)
  ;; (print "doqueue: " fname)
  (let* ((start-time (current-milliseconds))
	 (qlen       (queue-length q)))
    (if (> qlen 1)
	(print "Processing queue of length " qlen))
    (let loop ((count      0)
	       (responses '()))
      (let ((delta (- (current-milliseconds) start-time)))
	(if (or (queue-empty? q)
		(> delta 400)) ;; stop working on this queue after 400ms have passed
	    (list count delta responses) ;; return count, delta and responses list
	    (let* ((witem  (queue-remove! q))
		   (action (witem-action witem))
		   (rdat   (witem-rdat   witem))
		   (stmt   (calldat-obj rdat))
		   (sth    (full-get-sth acfg fname stmt))
		   (ctype  (calldat-ctype rdat))
		   (data   (witem-data   witem))
		   (cookie (witem-cookie witem)))
	      ;; do the processing and save the result in witem-result
	      (witem-result-set!
	       witem
	       (case ctype ;; action
		 ((noblockwrite) ;; blind write, no ack of success returned
		  (apply sqlite3:execute sth data)
		  (sqlite3:last-insert-rowid dbh))
		 ((dbwrite)      ;; blocking write   
		  (apply sqlite3:execute sth data)
		  #t)
		 ((dbread) ;; TODO: consider breaking this up and shipping in pieces for large query
		  (apply sqlite3:map-row (lambda x x) sth data))
		 ((full-ping)  'full-ping)
		 (else (print "Not ready for action " action) #f)))
	      (loop (add1 count)
		    (if cookie
			(cons witem responses)
			responses))))))))

;; do up to 400ms of processing on each queue
;; - the work-queue-processor will allow the max 1200ms of work to complete but it will flag as overloaded
;; 
(define (process-db-queries acfg fname)
  (if (hash-table-exists? (area-wqueues acfg) fname)
      (let* ((process-db-queries-start-time (current-milliseconds))
	     (qdat             (hash-table-ref/default (area-wqueues acfg) fname #f))
	     (queue-sym->queue (lambda (queue-sym)
				 (case queue-sym  ;; lookup the queue from qdat given a name (symbol)
				   ((wqueue)  (qdat-writeq qdat))
				   ((rqueue)  (qdat-readq  qdat))
				   ((rwqueue) (qdat-rwq    qdat))
				   ((misc)    (qdat-misc   qdat))
				   (else #f))))
	     (dbdat   (get-dbh acfg fname))
	     (dbh     (if (dbdat? dbdat)(dbdat-dbh dbdat) #f))
	     (nowtime (current-seconds)))
	;; handle the queues that require a transaction
	;;
	(map ;; 
	 (lambda (queue-sym)
	   ;; (print "processing queue " queue-sym)
	   (let* ((queue (queue-sym->queue queue-sym)))
	     (if (not (queue-empty? queue))
		 (let ((responses
			(sqlite3:with-transaction ;; todo - catch exceptions...
			 dbh
			 (lambda ()
			   (let* ((res (doqueue acfg queue fname dbdat dbh))) ;; this does the work!
			     ;; (print "res=" res)
			     (match res
			      ((count delta responses)
			       (update-stats acfg fname queue-sym delta count)
			       (sdbg> "process-db-queries" "sqlite3-transaction" process-db-queries-start-time #f #f)
			       responses) ;; return responses
			      (else
			       (print "ERROR: bad return data from doqueue " res)))
			     )))))
		   ;; having completed the transaction, send the responses.
		   ;; (print "INFO: sending " (length responses) " responses.")
		   (let loop ((responses-left responses))
		     (cond
		      ((null? responses-left)  #t)
		      (else
		       (let* ((witem    (car responses-left))
			      (response (cdr responses-left)))  
			 (call-deliver-response acfg (witem-ripaddr witem)(witem-rport witem)
						(witem-cookie witem)(witem-result witem)))
		       (loop (cdr responses-left))))))
		 )))
	 '(wqueue rwqueue rqueue))
	
	;; handle misc queue
	;;
	;; (print "processing misc queue")
	(let ((queue (queue-sym->queue 'misc)))
	  (doqueue acfg queue fname dbdat dbh))
	;; ....
	(save-dbh acfg fname dbdat)
	#t ;; just to let the tests know we got here
	)
      #f ;; nothing processed
      ))

;; run all queues in parallel per db but sequentially per queue for that db.
;;  - process the queues every 500 or so ms
;;  - allow for long running queries to continue but all other activities for that
;;    db will be blocked.
;;
(define (work-queue-processor acfg)
  (let* ((threads (make-hash-table))) ;; fname => thread
    (let loop ((fnames      (hash-table-keys (area-wqueues acfg)))
	       (target-time (+ (current-milliseconds) 50)))
      ;;(if (not (null? fnames))(print "Processing for these databases: " fnames))
      (for-each
       (lambda (fname)
	 ;; (print "processing for " fname)
	 ;;(process-db-queries acfg fname))
	 (let ((th (hash-table-ref/default threads fname #f)))
	   (if (and th (not (member (thread-state th) '(dead terminated))))
	       (begin
		 (print "WARNING: worker thread for " fname " is taking a long time.")
		 (print "Thread is in state " (thread-state th)))
	       (let ((th1 (make-thread (lambda ()
					 ;; (catch-and-dump
					 ;;  (lambda ()
					    ;; (print "Process queries for " fname)
					    (let ((start-time (current-milliseconds)))
					      (process-db-queries acfg fname)
					      ;; (thread-sleep! 0.01) ;; need the thread to take at least some time
					      (hash-table-delete! threads fname)) ;; no mutexes?
					    fname)
					  "th1"))) ;; ))
		 (hash-table-set! threads fname th1)
		 (thread-start! th1)))))
       fnames)
      ;; (thread-sleep! 0.1) ;; give the threads some time to process requests
      ;; burn time until 400ms is up
      (let ((now-time (current-milliseconds)))
	(if (< now-time target-time)
	    (let ((delta (- target-time now-time)))
	      (thread-sleep! (/ delta 1000)))))
      (loop (hash-table-keys (area-wqueues acfg))
	    (+ (current-milliseconds) 50)))))

;;======================================================================
;; S T A T S   G A T H E R I N G

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


(defstruct stat
  (qcount-avg  0)                  ;; coarse running average
  (qtime-avg   0)                  ;; coarse running average
  (qcount      0)                  ;; total
  (qtime       0)                  ;; total
  (last-qcount 0)                  ;; last 
  (last-qtime  0)                  ;; last
  (dbs        '())                 ;; list of db files handled by this node
  (when        0))                 ;; when the last query happened - seconds


(define (update-stats acfg fname bucket duration numqueries)
  (let* ((key   fname) ;; for now do not use bucket. Was: (conc fname "-" bucket)) ;; lazy but good enough
	 (stats (or (hash-table-ref/default (area-stats acfg) key #f)
		    (let ((newstats (make-stat)))
		      (hash-table-set! (area-stats acfg) key newstats)
		      newstats))))
    ;; when the last query happended (used to remove the fname from the active list)
    (stat-when-set! stats (current-seconds))
    ;; last values
    (stat-last-qcount-set! stats numqueries)
    (stat-last-qtime-set!  stats duration)
    ;; total over process lifetime
    (stat-qcount-set! stats (+ (stat-qcount stats) numqueries))
    (stat-qtime-set!  stats (+ (stat-qtime  stats) duration))
    ;; coarse average
    (stat-qcount-avg-set! stats (/ (+ (stat-qcount-avg stats) numqueries) 2))
    (stat-qtime-avg-set!  stats (/ (+ (stat-qtime-avg  stats) duration)   2))

    ;; here is where we add the stats for a given dbfile
    (if (not (member fname (stat-dbs stats)))
	(stat-dbs-set! stats (cons fname (stat-dbs stats))))

    ))

;;======================================================================
;; S E R V E R   S T U F F 
;;======================================================================

;; this does NOT return!
;;
(define (find-free-port-and-open acfg)
  (let ((port (or (area-port acfg) 3200)))
    (handle-exceptions
	exn
	(begin
	  (print "INFO: cannot bind to port " (rpc:default-server-port) ", trying next port")
	  (area-port-set! acfg (+ port 1))
	  (find-free-port-and-open acfg))
      (rpc:default-server-port port)
      (area-port-set! acfg port)
      (tcp-read-timeout 120000)
      ;; ((rpc:make-server (tcp-listen port)) #t)
      (tcp-listen (rpc:default-server-port)
      ))))

;; register this node by putting a packet into the pkts dir.
;; look for other servers
;; contact other servers and compile list of servers
;; there are two types of server
;;     main servers - dashboards, runners and dedicated servers - need pkt
;;     passive servers - test executers, step calls, list-runs - no pkt
;;
(define (register-node acfg hostip port-num)
  ;;(mutex-lock! (area-mutex acfg))
  (let* ((server-type  (area-server-type acfg)) ;; auto, main, passive (no pkt created)
	 (best-ip      (or hostip (get-my-best-address)))
	 (mtdir        (area-dbdir acfg))
	 (pktdir       (area-pktsdir acfg))) ;; conc mtdir "/.server-pkts")))
    (print "Registering node " best-ip ":" port-num)
    (if (not mtdir) ;; require a home for this node to put or find databases
	#f
	(begin
	  (if  (not (directory? pktdir))(create-directory pktdir))
	  ;; server is started, now create pkt if needed
	  (print "Starting server in " server-type " mode with port " port-num)
	  (if (member server-type '(auto main)) ;; TODO: if auto, count number of servers registers, if > 3 then don't put out a pkt
	      (begin
		(area-pktid-set! acfg
				 (write-alist->pkt
				  pktdir 
				  `((hostname . ,(get-host-name))
				    (ipaddr   . ,best-ip)
				    (port     . ,port-num)
				    (pid      . ,(current-process-id)))
				  pktspec: *pktspec*
				  ptype:   'server))
		(area-pktfile-set! acfg (conc pktdir "/" (area-pktid acfg) ".pkt"))))
	  (area-port-set!    acfg port-num)
	  #;(mutex-unlock! (area-mutex acfg))))))

(define *cookie-seqnum* 0)
(define (make-cookie key)
  (set! *cookie-seqnum* (add1 *cookie-seqnum*))
  ;;(print "MAKE COOKIE CALLED -- on "servkey"-"*cookie-seqnum*)
  (conc key "-" *cookie-seqnum*)
  )

;; dispatch locally if possible
;;
(define (call-deliver-response acfg ipaddr port cookie data)
  (if (and (equal? (area-myaddr acfg) ipaddr)
	   (equal? (area-port     acfg) port))
      (deliver-response acfg cookie data)
      ((rpc:procedure 'response ipaddr port) cookie data)))

(define (deliver-response acfg cookie data)
  (let ((deliver-response-start (current-milliseconds)))
    (thread-start! (make-thread
		    (lambda ()
		      (let loop ((tries-left 5))
			;;(print "TOP OF DELIVER_RESPONSE LOOP; triesleft="tries-left)
			;;(pp (hash-table->alist (area-cookie2mbox acfg)))
			(let* ((mbox (hash-table-ref/default (area-cookie2mbox acfg) cookie #f)))
			  (cond
			   ((eq? 0 tries-left)
			    (print "ulex:deliver-response: I give up. Mailbox never appeared. cookie="cookie)
			    )
			   (mbox
			    ;;(print "got mbox="mbox"  got data="data"  send.")
			    (mailbox-send! mbox data))
			   (else
			    ;;(print "no mbox yet.  look for "cookie)
			    (thread-sleep! (/ (- 6 tries-left) 10))
			    (loop (sub1 tries-left))))))
		      ;; (debug-pp (list (conc "ulex:deliver-response took " (- (current-milliseconds) deliver-response-start) " ms, cookie=" cookie " data=") data))
		      (sdbg> "deliver-response" "mailbox-send" deliver-response-start #f #f cookie)
		      )
		    (conc "deliver-response thread for cookie="cookie))))
  #t)

;; action:
;;   immediate - quick actions, no need to put in queues
;;   dbwrite   - put in dbwrite queue
;;   dbread    - put in dbread queue
;;   oslong    - os actions, e.g. du, that could take a long time
;;   osshort   - os actions that should be quick, e.g. df
;;
(define (request acfg from-ipaddr from-port servkey action cookie fname params) ;; std-peer-handler
  ;; NOTE: Use rpc:current-peer for getting return address
  (let* ((std-peer-handler-start (current-milliseconds))
	 ;; (raw-data               (alist-ref 'data     dat))
	 (rdat                   (hash-table-ref/default
				  (area-rtable acfg) action #f)) ;; this looks up the sql query or other details indexed by the action
	 (witem                  (make-witem ripaddr: from-ipaddr ;; rhost:   from-host   
					     rport:   from-port   action:  action
					     rdat:    rdat        cookie:  cookie
					     servkey: servkey     data:    params ;; TODO - rename data to params
					     caller:  (rpc:current-peer))))
    (if (not (equal? servkey (area-pktid acfg)))
	`(#f . ,(conc "I don't know you servkey=" servkey ", pktid=" (area-pktid acfg))) ;; immediately return this
	(let* ((ctype (if rdat 
			  (calldat-ctype rdat) ;; is this necessary? these should be identical
			  action)))
	  (sdbg> "std-peer-handler" "immediate" std-peer-handler-start #f #f)
	  (case ctype
	    ;; (dbwrite acfg rdat (cons from-ipaddr from-port) data)))
	    ((full-ping)  `(#t  "ack to full ping"        ,(work-queue-add acfg fname witem) ,cookie))
	    ((response)   `(#t  "ack from requestor"      ,(deliver-response acfg fname params)))
	    ((dbwrite)    `(#t  "db write submitted"      ,(work-queue-add acfg fname witem) ,cookie))
	    ((dbread)     `(#t  "db read submitted"       ,(work-queue-add acfg fname witem) ,cookie  ))
	    ((dbrw)       `(#t  "db read/write submitted" ,cookie))
	    ((osshort)    `(#t  "os short submitted"      ,cookie))
	    ((oslong)     `(#t  "os long submitted"       ,cookie))
	    (else         `(#f  "unrecognised action"     ,ctype)))))))

;; Call this to start the actual server
;;
;; start_server
;;
;;   mode: '
;;   handler: proc which takes pktrecieved as argument
;;

(define (start-server acfg)
  (let* ((conn (find-free-port-and-open acfg))
	 (port (area-port acfg)))
    (rpc:publish-procedure!
     'delist-db
     (lambda (fname)
       (hash-table-delete! (area-dbs acfg) fname)))
    (rpc:publish-procedure!
     'calling-addr
     (lambda ()
       (rpc:current-peer)))
    (rpc:publish-procedure!
     'ping
     (lambda ()(real-ping acfg)))
    (rpc:publish-procedure!
     'request
     (lambda (from-addr from-port servkey action cookie dbname params)
       (request acfg from-addr from-port servkey action cookie dbname params)))
    (rpc:publish-procedure!
     'response
     (lambda (cookie res-dat)
       (deliver-response acfg cookie res-dat)))
    (area-ready-set! acfg #t)
    (area-conn-set! acfg conn)
    ((rpc:make-server conn) #f)));; ((tcp-listen (rpc:default-server-port)) #t)


(define (launch acfg) ;;  #!optional (proc std-peer-handler))
  (print "starting launch")
  (update-known-servers acfg) ;; gotta do this on every start (thus why limit number of publicised servers)
  #;(let ((original-handler (current-exception-handler))) ;; is th
    (lambda (exception)
      (server-exit-procedure)
      (original-handler exception)))
  (on-exit (lambda ()
	     (shutdown acfg))) ;; (finalize-all-db-handles acfg)))
  ;; set up the rpc handler
  (let* ((th1  (make-thread
		(lambda ()(start-server acfg))
		"server thread"))
	 (th2   (make-thread
		 (lambda ()
		   (print "th2 starting")
		   (let loop ()
		     (work-queue-processor acfg)
		     (print "work-queue-processor crashed!")
		     (loop)))
		 "work queue thread")))
    (thread-start! th1)
    (thread-start! th2)
    (let loop ()
      (thread-sleep! 0.025)
      (if (area-ready acfg)
	  #t
	  (loop)))
    ;; attempt to fix my address
    (let* ((all-addr (get-all-ips-sorted)))	     ;; could use (tcp-addresses conn)?
      (let loop ((rem-addrs all-addr))
	(if (null? rem-addrs)
	    (begin
	      (print "ERROR: Failed to figure out the ip address of myself as a server. Giving up.")
	      (exit 1)) ;; BUG Changeme to raising an exception
		
	    (let* ((addr      (car rem-addrs))
		   (good-addr (handle-exceptions
				  exn
				  #f
				((rpc:procedure 'calling-addr addr (area-port acfg))))))
	      (if good-addr
		  (begin
		    (print "Got good-addr of " good-addr)
		    (area-myaddr-set! acfg good-addr))
		  (loop (cdr rem-addrs)))))))
    (register-node acfg (area-myaddr acfg)(area-port acfg))
    (print "INFO: Server started on " (area-myaddr acfg) ":" (area-port acfg))
    ;; (update-known-servers acfg) ;; gotta do this on every start (thus why limit number of publicised servers)
    ))

(define (clear-server-pkt acfg)
  (let ((pktf (area-pktfile acfg)))
    (if pktf (delete-file* pktf))))

(define (shutdown acfg)
  (let (;;(conn (area-conn    acfg))
	(pktf (area-pktfile acfg))
	(port (area-port    acfg)))
    (if pktf (delete-file* pktf))
    (send-all "imshuttingdown")
    ;; (rpc:close-all-connections!) ;; don't know if this is actually needed
    (finalize-all-db-handles acfg)))

(define (send-all msg)
  #f)

;; given a area record look up all the packets
;;
(define (get-all-server-pkts acfg)
  (let ((all-pkt-files (glob (conc (area-pktsdir acfg) "/*.pkt"))))
    (map (lambda (pkt-file)
	   (read-pkt->alist pkt-file pktspec: *pktspec*))
	 all-pkt-files)))

#;((Z . "9a0212302295a19610d5796fce0370fa130758e9")
  (port . "34827")
  (pid . "28748")
  (hostname . "zeus")
  (T . "server")
  (D . "1549427032.0"))

#;(define (get-my-best-address)
  (let ((all-my-addresses (get-all-ips))) ;; (vector->list (hostinfo-addresses (hostname->hostinfo (get-host-name))))))
    (cond
     ((null? all-my-addresses)
      (get-host-name))                                          ;; no interfaces?
     ((eq? (length all-my-addresses) 1)
      (ip->string (car all-my-addresses)))                      ;; only one to choose from, just go with it
     (else 
      (ip->string (car (filter (lambda (x)                      ;; take any but 127.
				 (not (eq? (u8vector-ref x 0) 127)))
			       all-my-addresses)))))))

;; whoami? I am my pkt
;;
(define (whoami? acfg)
  (hash-table-ref/default (area-hosts acfg)(area-pktid acfg) #f))

;;======================================================================
;; "Client side" operations
;;======================================================================

(define (safe-call call-key host port . params)
  (handle-exceptions
   exn
   (begin
     (print "Call " call-key " to " host ":" port " failed")
     #f)
   (apply (rpc:procedure call-key host port) params)))

;; ;; convert to/from string / sexpr
;; 
;; (define (string->sexpr str)
;;   (if (string? str)
;;       (with-input-from-string str read)
;;       str))
;; 
;; (define (sexpr->string s)
;;   (with-output-to-string (lambda ()(write s))))

;; is the server alive?
;;
(define (ping acfg host port)
  (let* ((myaddr     (area-myaddr acfg))
	 (myport     (area-port   acfg))
	 (start-time (current-milliseconds))
	 (res        (if (and (equal? myaddr host)
			      (equal? myport port))
			 (real-ping acfg)
			 ((rpc:procedure 'ping host port)))))
    (cons (- (current-milliseconds) start-time)
	  res)))

;; returns ( ipaddr port alist-fname=>randnum )
(define (real-ping acfg)
  `(,(area-myaddr acfg) ,(area-port acfg) ,(get-host-stats acfg)))

;; is the server alive AND the queues processing?
;;
#;(define (full-ping acfg servpkt)
  (let* ((start-time (current-milliseconds))
	 (res        (send-message acfg servpkt '(full-ping) 'full-ping)))
    (cons (- (current-milliseconds) start-time)
	  res))) ;; (equal? res "got ping"))))


;; look up all pkts and get the server id (the hash), port, host/ip
;; store this info in acfg
;; return the number of responsive servers found
;;
;; DO NOT VERIFY THAT THE SERVER IS ALIVE HERE. This is called at times where the current server is not yet alive and cannot ping itself
;;
(define (update-known-servers acfg)
  ;; readll all pkts
  ;; foreach pkt; if it isn't me ping the server; if alive, add to hosts hash, else rm the pkt
  (let* ((start-time (current-milliseconds))
	 (all-pkts  (delete-duplicates
		     (append (get-all-server-pkts acfg)
			     (hash-table-values (area-hosts acfg)))))
	 (hostshash (area-hosts acfg))
	 (my-id     (area-pktid acfg))
	 (pktsdir   (area-pktsdir acfg)) ;; needed to remove pkts from non-responsive servers
	 (numsrvs   0)
	 (delpkt    (lambda (pktsdir sid)
		      (print "clearing out server " sid)
		      (delete-file* (conc pktsdir "/" sid ".pkt"))
		      (hash-table-delete! hostshash sid))))
    (area-last-srvup-set! acfg (current-seconds))
    (for-each
     (lambda (servpkt)
       (if (list? servpkt)
	   ;; (pp servpkt)
	   (let* ((shost (alist-ref 'ipaddr servpkt))
		  (sport (any->number (alist-ref 'port servpkt)))
		  (res   (handle-exceptions
			  exn
			  (begin
			    ;; (print "INFO: bad server on " shost ":" sport)
			    #f)
			  (ping acfg shost sport)))
		  (sid   (alist-ref 'Z servpkt)) ;; Z code is our name for the server
		  (url   (conc shost ":" sport))
		  )
	     #;(if (or (not res)
		     (null? res))
		 (begin
		   (print "STRANGE: ping of " url " gave " res)))
	     
	     ;; (print "Got " res " from " shost ":" sport)
	     (match res
		    ((qduration . payload)
		     ;; (print "Server pkt:" (alist-ref 'ipaddr servpkt) ":" (alist-ref 'port servpkt)
		     ;;        (if payload
		     ;;            "Success" "Fail"))
		     (match payload
			    ((host port stats)
			     ;; (print "From " host ":" port " got stats: " stats)
			     (if (and host port stats)
				 (let ((url (conc host ":" port)))
				   (hash-table-set! hostshash sid servpkt)
				   ;; store based on host:port
				   (hash-table-set! (area-hoststats acfg) sid stats))
				 (print "missing data from the server, not sure what that means!"))
			     (set! numsrvs (+ numsrvs 1)))
			    (#f
			     (print "Removing pkt " sid " due to #f from server or failed ping")
			     (delpkt pktsdir sid))
			    (else
			     (print "Got ")(pp res)(print " from server ")(pp servpkt) " but response did not match (#f/#t . msg)")))
		    (else
		     ;; here we delete the pkt - can't reach the server, remove it
		     ;; however this logic is inadequate. we should mark the server as checked
		     ;; and not good, if it happens a second time - then remove the pkt
		     ;; or something similar. I.e. don't be too quick to assume the server is wedged or dead
		     ;; could be it is simply too busy to reply
		     (let ((bad-pings (hash-table-ref/default (area-health acfg) url 0)))
		       (if (> bad-pings 1) ;; two bad pings - remove pkt
			   (begin
			     (print "INFO: " bad-pings " bad responses from " url ", deleting pkt " sid)
			     (delpkt pktsdir sid))
			   (begin
			     (print "INFO: " bad-pings " bad responses from " shost ":" sport " not deleting pkt yet")
			     (hash-table-set! (area-health acfg)
					      url
					      (+ (hash-table-ref/default (area-health acfg) url 0) 1))
			     ))
		       ))))
	   ;; servpkt is not actually a pkt?
	   (begin
	     (print "Bad pkt " servpkt))))
     all-pkts)
    (sdbg> "update-known-servers" "end" start-time #f #f " found " numsrvs
	   " servers, pkts: " (map (lambda (p)
				     (alist-ref 'Z p))
				   all-pkts))
    numsrvs))

(defstruct srvstat
  (numfiles 0)   ;; number of db files handled by this server - subtract 1 for the db being currently looked at
  (randnum  #f)  ;; tie breaker number assigned to by the server itself - applies only to the db under consideration
  (pkt      #f)) ;; the server pkt

;;(define (srv->srvstat srvpkt)
  
;; Get the server best for given dbname and key
;;
;;   NOTE: key is not currently used. The key points to the kind of query, this may be useful for directing read-only queries.
;;
(define (get-best-server acfg dbname key)
  (let* (;; (servers (hash-table-values (area-hosts acfg)))
	 (servers     (area-hosts acfg))
	 (skeys       (sort (hash-table-keys servers) string>=?)) ;; a stable listing
	 (start-time  (current-milliseconds))
	 (srvstats    (make-hash-table))  ;; srvid => srvstat
	 (url         (conc (area-myaddr acfg) ":" (area-port acfg))))
    ;; (print "scores for " dbname ": " (map (lambda (k)(cons k (calc-server-score acfg dbname k))) skeys))
    (if (null? skeys)
	(if (> (update-known-servers acfg) 0)
	    (get-best-server acfg dbname key) ;; some risk of infinite loop here, TODO add try counter
	    (begin
	      (print "ERROR: no server found!") ;; since this process is also a server this should never happen
	      #f))
	(begin
	  ;; (print "in get-best-server with skeys=" skeys)
	  (if (> (- (current-seconds) (area-last-srvup acfg)) 10)
	      (begin
		(update-known-servers acfg)
		(sdbg> "get-best-server" "update-known-servers" start-time #f #f)))

	  ;; for each server look at the list of dbfiles, total number of dbs being handled
	  ;; and the rand number, save the best host
	  ;; also do a delist-db for each server dbfile not used
	  (let* ((best-server       #f)
		 (servers-to-delist (make-hash-table)))
	    (for-each
	     (lambda (srvid)
	       (let* ((server    (hash-table-ref/default servers srvid #f))
		      (stats     (hash-table-ref/default (area-hoststats acfg) srvid '(()))))
		 ;; (print "stats: " stats)
 		 (if server
		     (let* ((dbweights (car stats))
			    (srvload   (length (filter (lambda (x)(not (equal? dbname (car x)))) dbweights)))
			    (dbrec     (alist-ref dbname dbweights equal?))  ;; get the pair with fname . randscore
			    (randnum   (if dbrec
					   dbrec ;; (cdr dbrec)
					   0)))
		       (hash-table-set! srvstats srvid (make-srvstat numfiles: srvload randnum: randnum pkt: server))))))
	     skeys)
	    
	    (let* ((sorted    (sort (hash-table-values srvstats) 
				    (lambda (a b)
				      (let ((numfiles-a (srvstat-numfiles a))
					    (numfiles-b (srvstat-numfiles b))
					    (randnum-a  (srvstat-randnum a))
					    (randnum-b  (srvstat-randnum b)))
					(if (< numfiles-a numfiles-b) ;; Note, I don't think adding an offset works here. Goal was only move file handling to a different server if it has 2 less
					    #t
					    (if (and (equal? numfiles-a numfiles-b)
						     (< randnum-a randnum-b))
						#t
						#f))))))
		   (best      (if (null? sorted)
				  (begin
				    (print "ERROR: should never be null due to self as server.")
				    #f)
				  (srvstat-pkt (car sorted)))))
	      #;(print "SERVER(" url "): " dbname ": " (map (lambda (srv)
							    (let ((p (srvstat-pkt srv)))
							      (conc (alist-ref 'ipaddr p) ":" (alist-ref 'port p)
								    "(" (srvstat-numfiles srv)","(srvstat-randnum srv)")")))
							    sorted))
	      best))))))
    
    ;; send out an "I'm about to exit notice to all known servers"
    ;;
(define (death-imminent acfg)
  '())

;;======================================================================
;; U L E X  -  T H E   I N T E R E S T I N G   S T U F F ! !
;;======================================================================

;; register a handler
;;   NOTES:
;;     dbinitsql   is reserved for a list of sql statements for initializing the db
;;     dbinitfn    is reserved for a db init function, if exists called after dbinitsql
;;     
(define (register acfg key obj #!optional (ctype 'dbwrite))
  (let ((ht (area-rtable acfg)))
    (if (hash-table-exists? ht key)
	(print "WARNING: redefinition of entry " key))
    (hash-table-set! ht key (make-calldat obj: obj ctype: ctype))))

;; usage: register-batch acfg '((key1 . sql1) (key2 . sql2) ... )
;; NB// obj is often an sql query
;;
(define (register-batch acfg ctype data)
  (let ((ht (area-rtable acfg)))
    (map (lambda (dat)
	   (hash-table-set! ht (car dat)(make-calldat obj: (cdr dat) ctype: ctype)))
	 data)))

(define (initialize-area-calls-from-specfile area specfile)
  (let* ((callspec (with-input-from-file specfile read )))
    (for-each (lambda (group)
                (register-batch
                 area
                 (car group)
                 (cdr group)))
              callspec)))

;; get-rentry
;;
(define (get-rentry acfg key)
  (hash-table-ref/default (area-rtable acfg) key #f))

(define (get-rsql acfg key)
  (let ((cdat (get-rentry acfg key)))
    (if cdat
	(calldat-obj cdat)
	#f)))



;; blocking call:
;;    client                         server
;;    ------                         ------
;;    call()
;;    send-message()
;;    nmsg-send()
;;                                   nmsg-receive()
;;                                   nmsg-respond(ack,cookie)
;;    ack, cookie
;;    mbox-thread-wait(cookie)
;;                                   nmsg-send(client,cookie,result)
;;        nmsg-respond(ack)
;;        return result
;;
;; reserved action:
;;    'immediate
;;    'dbinitsql
;;
(define (call acfg dbname action params #!optional (count 0))
  (let* ((call-start-time     (current-milliseconds))
	 (srv                 (get-best-server acfg dbname action))
	 (post-get-start-time (current-milliseconds))
	 (rdat                (hash-table-ref/default (area-rtable acfg) action #f))
	 (myid                (trim-pktid (area-pktid acfg)))
	 (srvid               (trim-pktid (alist-ref 'Z srv)))
	 (cookie              (make-cookie myid)))
    (sdbg> "call" "get-best-server" call-start-time #f call-start-time " from: " myid " to server: " srvid " for " dbname " action: " action " params: " params " rdat: " rdat)
    (print "INFO: call to " (alist-ref 'ipaddr srv) ":" (alist-ref 'port srv) " from " (area-myaddr acfg) ":" (area-port acfg) " for " dbname)
    (if (and srv rdat) ;; need both to dispatch a request
	(let* ((ripaddr  (alist-ref 'ipaddr srv))
	       (rsrvid   (alist-ref 'Z srv))
	       (rport    (any->number (alist-ref 'port   srv)))
	       (res-full (if (and (equal? ripaddr (area-myaddr acfg))
				  (equal? rport   (area-port acfg)))
			     (request acfg ripaddr rport (area-pktid acfg) action cookie dbname params)
			     (safe-call 'request ripaddr rport
					(area-myaddr acfg)
					(area-port   acfg)
					#;(area-pktid acfg)
					rsrvid
					action cookie dbname params))))
	  ;; (print "res-full: " res-full)
	  (match res-full
	    ((response-ok response-msg rem ...)
	     (let* ((send-message-time (current-milliseconds))
		    ;; (match res-full
		    ;;  ((response-ok response-msg)
		    ;; (response-ok  (car res-full))
		    ;; (response-msg (cadr res-full)
		    )
	       ;; (res (take res-full 3))) ;; ctype == action, TODO: converge on one term <<=== what was this? BUG 
	       ;; (print "ulex:call: send-message took " (- send-message-time post-get-start-time) " ms params=" params)
	       (sdbg> "call" "send-message" post-get-start-time #f call-start-time)
	       (cond
		((not response-ok) #f)
		((member response-msg '("db read submitted" "db write submitted"))

		 (let* ((cookie-id   (cadddr res-full))
			(mbox        (make-mailbox))
			(mbox-time   (current-milliseconds)))
		   (hash-table-set! (area-cookie2mbox acfg) cookie-id mbox)
		   (let* ((mbox-timeout-secs    20)
			  (mbox-timeout-result 'MBOX_TIMEOUT)
			  (res                  (mailbox-receive! mbox mbox-timeout-secs mbox-timeout-result))
			  (mbox-receive-time    (current-milliseconds)))
		     (hash-table-delete! (area-cookie2mbox acfg) cookie-id)
		     (sdbg> "call" "mailbox-receive" mbox-time #f call-start-time " from: " myid " to server: " srvid " for " dbname)
		     ;; (print "ulex:call mailbox-receive took " (- mbox-receive-time mbox-time) "ms params=" params)
		     res)))
		(else
		 (print "Unhandled response \""response-msg"\"")
		 #f))
	       ;; depending on what action (i.e. ctype) is we will block here waiting for
	       ;; all the data (mechanism to be determined)
	       ;;
	       ;; if res is a "working on it" then wait
	       ;;    wait for result
	       ;; mailbox thread wait on 
	       
	       ;; if res is a "can't help you" then try a different server
	       ;; if res is a "ack" (e.g. for one-shot requests) then return res
	       ))
	    (else
	     (if (< count 10)
		 (let* ((url (conc (alist-ref 'ipaddr srv) ":" (alist-ref 'port srv))))
		   (thread-sleep! 1)
		   (print "ERROR: Bad result from " url ", dbname: " dbname ", action: " action ", params: " params ". Trying again in 1 second.")
		   (call acfg dbname action params (+ count 1)))
		 (begin
		   (error (conc "ERROR: " count " tries, still have improper response res-full=" res-full)))))))
	(begin
	  (if (not rdat)
	      (print "ERROR: action " action " not registered.")
	      (if (< count 10)
		 (begin
		   (thread-sleep! 1)
		   (area-hosts-set! acfg (make-hash-table)) ;; clear out all known hosts
		   (print "ERROR: no server found, srv=" srv ", trying again in 1 seconds")
		   (call acfg dbname action params (+ count 1)))
		 (begin
		   (error (conc "ERROR: no server found after 10 tries, srv=" srv ", giving up."))
		   #;(error "No server available"))))))))


;;======================================================================
;; U T I L I T I E S 
;;======================================================================

;; get a signature for identifing this process
;;
(define (get-process-signature)
  (cons (get-host-name)(current-process-id)))

;;======================================================================
;; S Y S T E M   S T U F F
;;======================================================================

;; get normalized cpu load by reading from /proc/loadavg and
;; /proc/cpuinfo return all three values and the number of real cpus
;; and the number of threads returns alist '((adj-cpu-load
;; . normalized-proc-load) ... etc.  keys: adj-proc-load,
;; adj-core-load, 1m-load, 5m-load, 15m-load
;;
(define (get-normalized-cpu-load)
  (let ((res (get-normalized-cpu-load-raw))
	(default `((adj-proc-load . 2) ;; there is no right answer
		   (adj-core-load . 2)
		   (1m-load       . 2)
		   (5m-load       . 0) ;; causes a large delta - thus causing default of throttling if stuff goes wrong
		   (15m-load      . 0)
		   (proc          . 1)
		   (core          . 1)
		   (phys          . 1)
		   (error         . #t))))
    (cond
     ((and (list? res)
	   (> (length res) 2))
      res)
     ((eq? res #f)   default) ;; add messages?
     ((eq? res #f) default)   ;; this would be the #eof
     (else default))))

(define (get-normalized-cpu-load-raw)
  (let* ((actual-host           (get-host-name))) ;; #f is localhost
    (let ((data  (append 
		  (with-input-from-file "/proc/loadavg" read-lines)
		  (with-input-from-file "/proc/cpuinfo" read-lines)
		  (list "end")))
	  (load-rx  (regexp "^([\\d\\.]+)\\s+([\\d\\.]+)\\s+([\\d\\.]+)\\s+.*$"))
	  (proc-rx  (regexp "^processor\\s+:\\s+(\\d+)\\s*$"))
	  (core-rx  (regexp "^core id\\s+:\\s+(\\d+)\\s*$"))
	  (phys-rx  (regexp "^physical id\\s+:\\s+(\\d+)\\s*$"))
	  (max-num  (lambda (p n)(max (string->number p) n))))
      ;; (print "data=" data)
      (if (null? data) ;; something went wrong
	  #f
	  (let loop ((hed      (car data))
		     (tal      (cdr data))
		     (loads    #f)
		     (proc-num 0)  ;; processor includes threads
		     (phys-num 0)  ;; physical chip on motherboard
		     (core-num 0)) ;; core
	    ;; (print hed ", " loads ", " proc-num ", " phys-num ", " core-num)
	    (if (null? tal) ;; have all our data, calculate normalized load and return result
		(let* ((act-proc (+ proc-num 1))
		       (act-phys (+ phys-num 1))
		       (act-core (+ core-num 1))
		       (adj-proc-load (/ (car loads) act-proc))
		       (adj-core-load (/ (car loads) act-core))
		       (result
			(append (list (cons 'adj-proc-load adj-proc-load)
				      (cons 'adj-core-load adj-core-load))
				(list (cons '1m-load (car loads))
				      (cons '5m-load (cadr loads))
				      (cons '15m-load (caddr loads)))
				(list (cons 'proc act-proc)
				      (cons 'core act-core)
				      (cons 'phys act-phys)))))
		  result)
		(regex-case
		    hed
		  (load-rx  ( x l1 l5 l15 ) (loop (car tal)(cdr tal)(map string->number (list l1 l5 l15)) proc-num phys-num core-num))
		  (proc-rx  ( x p         ) (loop (car tal)(cdr tal) loads           (max-num p proc-num) phys-num core-num))
		  (phys-rx  ( x p         ) (loop (car tal)(cdr tal) loads           proc-num (max-num p phys-num) core-num))
		  (core-rx  ( x c         ) (loop (car tal)(cdr tal) loads           proc-num phys-num (max-num c core-num)))
		  (else 
		   (begin
		     ;; (print "NO MATCH: " hed)
		     (loop (car tal)(cdr tal) loads proc-num phys-num core-num))))))))))

(define (get-host-stats acfg)
  (let ((stats-hash (area-stats acfg)))
    ;; use this opportunity to remove references to dbfiles which have not been accessed in a while
    (for-each
     (lambda (dbname)
       (let* ((stats       (hash-table-ref stats-hash dbname))
	      (last-access (stat-when stats)))
	 (if (and (> last-access 0)                             ;; if zero then there has been no access
		  (> (- (current-seconds) last-access) 10))     ;; not used in ten seconds
	     (begin
	       (print "Removing " dbname " from stats list")
	       (hash-table-delete! stats-hash dbname) ;; remove from stats hash
	       (stat-dbs-set! stats (hash-table-keys stats))))))
     (hash-table-keys stats-hash))
    
    `(,(hash-table->alist (area-dbs acfg)) ;; dbname => randnum
      ,(map (lambda (dbname)  ;; dbname is the db name
	      (cons dbname (stat-when (hash-table-ref stats-hash dbname))))
	    (hash-table-keys stats-hash))
      (cpuload . ,(get-normalized-cpu-load)))))
    #;(stats   . ,(map (lambda (k) ;; create an alist from the stats data
		       (cons k (stat->alist (hash-table-ref (area-stats acfg) k))))
		     (hash-table-keys (area-stats acfg))))

#;(trace
 ;; assv
 ;; cdr
 ;; caar
 ;; ;; cdr
 ;; call
 ;; finalize-all-db-handles
 ;; get-all-server-pkts
 ;; get-normalized-cpu-load
 ;; get-normalized-cpu-load-raw
 ;; launch
 ;; nmsg-send
 ;; process-db-queries
 ;; receive-message
 ;; std-peer-handler
 ;; update-known-servers
 ;; work-queue-processor
 )

;;======================================================================
;; netutil
;;   move this back to ulex-netutil.scm someday?
;;======================================================================

;; #include <stdio.h>
;; #include <netinet/in.h>
;; #include <string.h>
;; #include <arpa/inet.h>

(foreign-declare "#include \"sys/types.h\"")
(foreign-declare "#include \"sys/socket.h\"")
(foreign-declare "#include \"ifaddrs.h\"")
(foreign-declare "#include \"arpa/inet.h\"")

;; get IP addresses from ALL interfaces
(define get-all-ips
  (foreign-safe-lambda* scheme-object ()
    "

// from https://stackoverflow.com/questions/17909401/linux-c-get-default-interfaces-ip-address :


    C_word lst = C_SCHEME_END_OF_LIST, len, str, *a;
//    struct ifaddrs *ifa, *i;
//    struct sockaddr *sa;

    struct ifaddrs * ifAddrStruct = NULL;
    struct ifaddrs * ifa = NULL;
    void * tmpAddrPtr = NULL;

    if ( getifaddrs(&ifAddrStruct) != 0)
      C_return(C_SCHEME_FALSE);

//    for (i = ifa; i != NULL; i = i->ifa_next) {
    for (ifa = ifAddrStruct; ifa != NULL; ifa = ifa->ifa_next) {
        if (ifa->ifa_addr->sa_family==AF_INET) { // Check it is
            // a valid IPv4 address
            tmpAddrPtr = &((struct sockaddr_in *)ifa->ifa_addr)->sin_addr;
            char addressBuffer[INET_ADDRSTRLEN];
            inet_ntop(AF_INET, tmpAddrPtr, addressBuffer, INET_ADDRSTRLEN);
//            printf(\"%s IP Address %s\\n\", ifa->ifa_name, addressBuffer);
            len = strlen(addressBuffer);
            a = C_alloc(C_SIZEOF_PAIR + C_SIZEOF_STRING(len));
            str = C_string(&a, len, addressBuffer);
            lst = C_a_pair(&a, str, lst);
        } 

//        else if (ifa->ifa_addr->sa_family==AF_INET6) { // Check it is
//            // a valid IPv6 address
//            tmpAddrPtr = &((struct sockaddr_in6 *)ifa->ifa_addr)->sin6_addr;
//            char addressBuffer[INET6_ADDRSTRLEN];
//            inet_ntop(AF_INET6, tmpAddrPtr, addressBuffer, INET6_ADDRSTRLEN);
////            printf(\"%s IP Address %s\\n\", ifa->ifa_name, addressBuffer);
//            len = strlen(addressBuffer);
//            a = C_alloc(C_SIZEOF_PAIR + C_SIZEOF_STRING(len));
//            str = C_string(&a, len, addressBuffer);
//            lst = C_a_pair(&a, str, lst);
//       }

//       else {
//         printf(\" not an IPv4 address\\n\");
//       }

    }

    freeifaddrs(ifa);
    C_return(lst);

"))

;; Change this to bias for addresses with a reasonable broadcast value?
;;
(define (ip-pref-less? a b)
  (let* ((rate (lambda (ipstr)
                 (regex-case ipstr
                             ( "^127\\." _ 0 )
                             ( "^(10\\.0|192\\.168\\.)\\..*" _ 1 )
                             ( else 2 ) ))))
    (< (rate a) (rate b))))
  

(define (get-my-best-address)
  (let ((all-my-addresses (get-all-ips))
        ;;(all-my-addresses-old (vector->list (hostinfo-addresses (hostname->hostinfo (get-host-name)))))
        )
    (cond