Megatest

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

$(PREFIX)/share/db/mt-pg.sql : mt-pg.sql
	mkdir -p $(PREFIX)/share/db
	$(INSTALL) mt-pg.sql $(PREFIX)/share/db/mt-pg.sql

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

# 
# $(PREFIX)/bin/revtagfsl : utils/revtagfsl.scm
#	csc utils/revtagfsl.scm -o $(PREFIX)/bin/revtagfsl

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

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

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

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

(declare (uses db))
(declare (uses tasks))

;; allow these queries through without starting a server
;;
(define api:read-only-queries
  '(get-key-val-pairs
    get-var
    get-keys
    get-key-vals
    test-toplevel-num-items
    get-test-info-by-id
    test-get-rundir-from-test-id
    get-count-tests-running-for-testname
    get-count-tests-running

;;    - keep it simple, only return the actual result of the call, i.e. no meta info here
;;
;;    - returns #( flag result )
;;
(define (api:execute-requests dbstruct dat)
  (handle-exceptions
   exn
   (let ((call-chain (get-call-chain))
         )
     (debug:print 0 *default-log-port* "WARNING: api:execute-requests received an exception from peer")
     (print-call-chain (current-error-port))
     (debug:print 0 *default-log-port* " message: "  ((condition-property-accessor 'exn 'message) exn))       
     (vector #f (vector exn call-chain dat))) ;; return some stuff for debug if an exception happens
   (cond
    ((not (vector? dat))                    ;; it is an error to not receive a vector
     (vector #f #f "remote must be called with a vector")       )


    (else  
     (let* ((cmd-in (vector-ref dat 0))
            (cmd    (if (symbol? cmd-in)
                        cmd-in
                        (string->symbol cmd-in)))
            (params (vector-ref dat 1))
            (start-t (current-milliseconds))
            (readonly-mode (dbr:dbstruct-read-only dbstruct))
            (readonly-command (member cmd api:read-only-queries))
            (writecmd-in-readonly-mode (and readonly-mode (not readonly-command)))
            (res    
             (if writecmd-in-readonly-mode
                 (conc "attempt to run write command "cmd" on a read-only database")
                 (case cmd
                   ;;===============================================
                   ;; READ/WRITE QUERIES
                   ;;===============================================

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

                   ;; TESTS
                   ((test-set-state-status-by-id)     (apply db:test-set-state-status-by-id dbstruct params))
                   ((delete-test-records)             (apply db:delete-test-records dbstruct params))
                   ((delete-old-deleted-test-records) (apply db:delete-old-deleted-test-records dbstruct params))
                   ((test-set-state-status)           (apply db:test-set-state-status dbstruct params))
                   ((test-set-top-process-pid)        (apply db:test-set-top-process-pid dbstruct params))
                   ((set-state-status-and-roll-up-items) (apply db:set-state-status-and-roll-up-items dbstruct params))
                   ((top-test-set-per-pf-counts)      (apply db:top-test-set-per-pf-counts dbstruct params))
                   ((test-set-archive-block-id)       (apply db:test-set-archive-block-id dbstruct params))

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

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

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

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

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

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

                   ;; ARCHIVES
                   ;; ((archive-get-allocations)   
                   ((archive-register-disk)     (apply db:archive-register-disk dbstruct params))
                   ((archive-register-block-name)(apply db:archive-register-block-name dbstruct params))
                   ((archive-allocate-testsuite/area-to-block)(apply db:archive-allocate-testsuite/area-to-block dbstruct block-id testsuite-name areakey))

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

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

                   ((get-key-vals)                    (apply db:get-key-vals dbstruct params))
                   ((get-target)                      (apply db:get-target dbstruct params))
                   ((get-targets)                     (db:get-targets dbstruct))

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

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

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

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

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

                   ;; TESTMETA
                   ((testmeta-get-record)       (apply db:testmeta-get-record dbstruct params))

                   ;; TASKS 
                   ((find-task-queue-records)   (apply tasks:find-task-queue-records dbstruct params))))))
       (if (not writecmd-in-readonly-mode)
           (let ((delta-t (- (current-milliseconds)
                             start-t)))
             (hash-table-set! *db-api-call-time* cmd
                              (cons delta-t (hash-table-ref/default *db-api-call-time* cmd '())))
             (vector #t res))
           (vector #f res)))))))

;; http-server  send-response
;;                 api:process-request
;;                    db:*
;;
;; NB// Runs on the server as part of the server loop
;;

   "logs"))

;; Force a megatest cleanup-db if version is changed and skip-version-check not specified
;;
(define (common:exit-on-version-changed)
  (if (common:version-changed?)
      (if (common:on-homehost?)
	  (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)))
	    (debug:print 0 *default-log-port*
			 "WARNING: Version mismatch!\n"
			 "   expected: " (common:version-signature) "\n"
			 "   got:      " (common:get-last-run-version))
            (cond
             ((get-environment-variable "MT_SKIP_DB_MIGRATE") #t)
             ((and (file-exists? mtconf) (file-exists? dbfile) (not read-only)
                   (eq? (current-user-id)(file-owner mtconf))) ;; safe to run -cleanup-db

              (debug:print 0 *default-log-port* "   I see you are the owner of megatest.config, attempting to cleanup and reset to new version")
              (handle-exceptions
               exn
               (begin
                 (debug:print 0 *default-log-port* "Failed to switch versions.")
                 (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
                 (print-call-chain (current-error-port))
                 (exit 1))
               (common:cleanup-db dbstruct)))
             ((not (file-exists? mtconf))
              (debug:print 0 *default-log-port* "   megatest.config does not exist in this area.  Cannot proceed with megatest version migration.")
              (exit 1))
             ((not (file-exists? dbfile))
              (debug:print 0 *default-log-port* "   megatest.db does not exist in this area.  Cannot proceed with megatest version migration.")
              (exit 1))
             ((not (eq? (current-user-id)(file-owner mtconf)))
              (debug:print 0 *default-log-port* "   You do not own megatest.db in this area.  Cannot proceed with megatest version migration.")
              (exit 1))
             (read-only
              (debug:print 0 *default-log-port* "   You have read-only access to this area.  Cannot proceed with megatest version migration.")
              (exit 1))
             (else
              (debug:print 0 *default-log-port* " to switch versions you can run: \"megatest -cleanup-db\"")
              (exit 1))))
	  (begin
	    (debug:print 0 *default-log-port* "ERROR: cannot migrate version unless on homehost. Exiting.")
	    (exit 1)))))

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

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


(define (common:readonly-watchdog dbstruct)
  (thread-sleep! 0.05) ;; delay for startup
  (debug:print-info 13 *default-log-port* "common:readonly-watchdog entered.")
  ;; sync megatest.db to /tmp/.../megatst.db
  (let* ((sync-cool-off-duration   3)
        (golden-mtdb     (dbr:dbstruct-mtdb dbstruct))
        (golden-mtpath   (db:dbdat-get-path golden-mtdb))
        (tmp-mtdb        (dbr:dbstruct-tmpdb dbstruct))
        (tmp-mtpath      (db:dbdat-get-path tmp-mtdb)))
    (debug:print-info 0 *default-log-port* "Read-only periodic sync thread started.")
    (let loop ((last-sync-time 0))
      (debug:print-info 13 *default-log-port* "loop top tmp-mtpath="tmp-mtpath" golden-mtpath="golden-mtpath)
      (let* ((duration-since-last-sync (- (current-seconds) last-sync-time)))
        (debug:print-info 13 *default-log-port* "duration-since-last-sync="duration-since-last-sync)
        (if (and (not *time-to-exit*)
                 (< duration-since-last-sync sync-cool-off-duration))
            (thread-sleep! (- sync-cool-off-duration duration-since-last-sync)))
        (if (not *time-to-exit*)
            (let ((golden-mtdb-mtime (file-modification-time golden-mtpath))
                  (tmp-mtdb-mtime    (file-modification-time tmp-mtpath)))
              (if (> golden-mtdb-mtime tmp-mtdb-mtime)
                  (let ((res (db:multi-db-sync dbstruct 'old2new)))
                    (debug:print-info 13 *default-log-port* "rosync called, " res " records transferred.")))
              (loop (current-seconds)))
            #t)))
    (debug:print-info 0 *default-log-port* "Exiting readonly-watchdog timer, *time-to-exit* = " *time-to-exit*" pid="(current-process-id)" mtpath="golden-mtpath)))


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

		    (set! last-time start-time)
		    (debug:print-info 4 *default-log-port* "timestamp -> " (seconds->time-string (current-seconds)) ", time since start -> " (seconds->hr-min-sec (- (current-seconds) *time-zero*))))))
	    
	    ;; keep going unless time to exit
	    ;;
	    (if (not *time-to-exit*)
		(let delay-loop ((count 0))
                  ;;(debug:print-info 13 *default-log-port* "delay-loop top; count="count" pid="(current-process-id)" this-wd-num="this-wd-num" *time-to-exit*="*time-to-exit*)
                                                            
		  (if (and (not *time-to-exit*)
			   (< count 4)) ;; was 11, changing to 4. 
		      (begin
			(thread-sleep! 1)
			(delay-loop (+ count 1))))
		  (if (not *time-to-exit*) (loop))))
	    (if (common:low-noise-print 30)
		(debug:print-info 0 *default-log-port* "Exiting watchdog timer, *time-to-exit* = " *time-to-exit*" pid="(current-process-id)" this-wd-num="this-wd-num)))))))

;; TODO: for multiple areas, we will have multiple watchdogs; and multiple threads to manage
(define (common:watchdog)
  ;;#t)
  (debug:print-info 13 *default-log-port* "common:watchdog entered.")

 (let ((dbstruct (db:setup)))
   (debug:print-info 13 *default-log-port* "after db:setup with dbstruct="dbstruct)
   (cond
    ((dbr:dbstruct-read-only dbstruct)
     (debug:print-info 13 *default-log-port* "loading read-only watchdog")
     (common:readonly-watchdog dbstruct))
    (else
     (debug:print-info 13 *default-log-port* "loading writable-watchdog.")
     (common:writable-watchdog dbstruct))))
 (debug:print-info 13 *default-log-port* "watchdog done.");;)
 )


(define (std-exit-procedure)
  (on-exit (lambda () 0))
  ;;(debug:print-info 13 *default-log-port* "std-exit-procedure called; *time-to-exit*="*time-to-exit*)
  (let ((no-hurry  (if *time-to-exit* ;; hurry up
		       #f
		       (begin
			 (set! *time-to-exit* #t)
			 #t))))
    (debug:print-info 4 *default-log-port* "starting exit process, finalizing databases.")
    (if (and no-hurry (debug:debug-mode 18))

    )

  0)

(define (std-signal-handler signum)
  ;; (signal-mask! signum)
  (set! *time-to-exit* #t)
  ;;(debug:print-info 13 *default-log-port* "got signal "signum)
  (debug:print-error 0 *default-log-port* "Received signal " signum " exiting promptly")
  ;; (std-exit-procedure) ;; shouldn't need this since we are exiting and it will be called anyway
  (exit))

(set-signal-handler! signal/int  std-signal-handler)  ;; ^C
(set-signal-handler! signal/term std-signal-handler)
;; (set-signal-handler! signal/stop std-signal-handler)  ;; ^Z NO, do NOT handle ^Z!

	      (apply print params)
	      )))))

;; Brandon's debug printer shortcut (indulge me :)
(define *BB-process-starttime* (current-milliseconds))
(define (BB> . in-args)
  (let* ((stack (get-call-chain))
         (location "??"))
    (for-each
     (lambda (frame)
       (let* ((this-loc (vector-ref frame 0))
              (temp     (string-split (->string this-loc) " "))
              (this-func (if (and (list? temp) (> (length temp) 1)) (cadr temp) "???")))
         (if (equal? this-func "BB>")
             (set! location this-loc))))
     stack)
    (let ((dp-args
           (append
            (list 0 *default-log-port*
                  (conc location "@"(/ (- (current-milliseconds) *BB-process-starttime*) 1000)"   ")  )
            in-args)))
      (apply debug:print dp-args))))

(define *BBpp_custom_expanders_list* (make-hash-table))



;; register hash tables with BBpp.

			      #:action (lambda (obj)
					 (let ((comment (iup:attribute comnt "VALUE"))
					       (test-id (db:test-get-id testdat)))
					   (if (or (not wpatt)
						   (string-match wregx comment))
					       (begin
						 ;; (rmt:test-set-state-status-by-id run-id test-id #f "WAIVED" comment)
						 (rmt:test-set-state-status run-id test-id #f "WAIVED" comment)
						 (db:test-set-status! testdat "WAIVED")
						 (cmtcmd comment)
						 (iup:destroy! dlog))))))
		  (iup:button "Cancel"
			      #:expand "HORIZONTAL" 
			      #:action (lambda (obj)
					 (iup:destroy! dlog)))))))

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

(if (not (common:on-homehost?))
    (begin
      (debug:print 0 *default-log-port* "WARNING: Current policy requires running dashboard on homehost: " (common:get-homehost))))
    
;; TODO: Move this inside (main)
;;
(if (not (launch:setup))
    (begin
      (print "Failed to find megatest.config, exiting") 
      (exit 1)))

;; create a watch dog to move changes from lt/.db/*.db to megatest.db
;;
;;;(if (file-write-access? (conc *toppath* "/megatest.db"))
;;(debug:print-info 13 *default-log-port* "Before common:watchdog spawn")
(thread-start! (make-thread common:watchdog "Watchdog thread"))
;;(debug:print-info 13 *default-log-port* "After common:watchdog spawn")
(if (not (args:get-arg "-use-db-cache"))
    (begin
      (debug:print-info 0 *default-log-port* "Forcing db-cache mode due to read-only access to megatest.db")
      (hash-table-set! args:arg-hash "-use-db-cache" #t)));;;)

;; data common to all tabs goes here
;;
(defstruct dboard:commondat
  ((curr-tab-num 0) : number)
  please-update  
  tabdats

         (tests-ht     (dboard:get-tests-for-run-duplicate tabdat run-id run testnamepatt key-vals))
         (tests-dat    (dashboard:tests-ht->tests-dat tests-ht)) 
         (tests-mindat (dcommon:minimize-test-data tests-dat)))  ;; reduces data for display
    (dboard:tabdat-last-runs-update-set! tabdat (- (current-seconds) 2))
    (hash-table-set! (dboard:tabdat-last-test-dat tabdat) run-id tests-dat)
    (hash-table-set! (dboard:tabdat-run-update-times tabdat) run-id (- (current-seconds) 10))
    (when (not run)
        (debug:print-info 13 *default-log-port* "ERROR: NO RUN FOR RUN-ID run-id="run-id)
        (debug:print-info 13 *default-log-port* "runs-hash-> " (hash-table->alist runs-hash))
        )
    tests-mindat))

(define (dashboard:runs-summary-xor-matrix-content tabdat runs-hash #!key (hide-clean #f))
  (let* ((src-run-id (dboard:tabdat-prev-run-id tabdat))
         (dest-run-id (dboard:tabdat-curr-run-id tabdat)))
    (if (and src-run-id dest-run-id)

                        (debug:catch-and-dump
                         (lambda ()

                           ;; Bummer - we dont have the global get/set api mapped in chicken
                           ;; (let* ((modkeys (iup:global "MODKEYSTATE")))
                           ;;   (BB> "modkeys="modkeys))

                           (debug:print-info 13 *default-log-port* "click-cb: obj="obj" lin="lin" col="col" status="status)
                           ;; status is corrupted on Brandon's home machine.  will have to wait until after shutdown to see if it is still broken in PDX SLES
                           (let* ((toolpath (car (argv)))
                                  (key      (conc lin ":" col))
                                  (test-id   (hash-table-ref/default cell-lookup key -1))
                                  (run-id   (dboard:tabdat-curr-run-id tabdat))
                                  (run-info (rmt:get-run-info run-id))
                                  (target   (rmt:get-target run-id))

                                                    "%")))
                                  (item-path (db:test-get-item-path (rmt:get-test-info-by-id run-id test-id)))
                                  (item-test-path (conc test-name "/" (if (equal? item-path "")
									"%" 
									item-path)))
                                  (status-chars (char-set->list (string->char-set status)))
                                  (testpanel-cmd      (conc toolpath " -test " (dboard:tabdat-curr-run-id tabdat) "," test-id " &")))
                             (debug:print-info 13 *default-log-port* "status-chars=["status-chars"] status=["status"]")
                             (cond
                              ((member #\1 status-chars) ;; 1 is left mouse button
                               (system testpanel-cmd))
                              
                              ((member #\2 status-chars) ;; 2 is middle mouse button
                               
                               (debug:print-info 13 *default-log-port* "mmb- test-name="test-name" testpatt="testpatt)
                               (iup:show (dashboard:popup-menu run-id test-id target runname test-name testpatt item-test-path) ;; popup-menu
                                         #:x 'mouse
                                         #:y 'mouse
                                         #:modal? "NO")
                               )
                              (else
                               (debug:print-info 13 *default-log-port* "unhandled status in run-summary-click-cb.  Doing right click action. (status is corrupted on Brandon's ubuntu host - bad/buggy  iup install??" )
                               (iup:show (dashboard:popup-menu run-id test-id target runname test-name testpatt item-test-path) ;; popup-menu
                                         #:x 'mouse
                                         #:y 'mouse
                                         #:modal? "NO")
                               )
                              )
                            

      (cond ;; nb// no else for this approach.
       ((>= runnum nruns) #f)
       ((>= keynum nkeys) 
	(vector-set! header runnum keyvec)
	(set! hdrlst (cons (apply iup:vbox (reverse res)) hdrlst))
	(loop (+ runnum 1) 0 (make-vector nkeys) '()))
       (else
	(let ((labl  (iup:label "" #:size (conc cell-width btn-height)  #:fontsize btn-fontsz #:expand "HORIZONTAL"))) ;; #:expand "HORIZONTAL" "60x15" 
	  (vector-set! keyvec keynum labl)
	  (loop runnum (+ keynum 1) keyvec (cons labl res))))))
    ;; By here the hdrlst contains a list of vboxes containing nkeys labels
    (let loop ((runnum  0)
	       (testnum 0)
	       (testvec  (make-vector ntests))
	       (res    '()))
      (cond
       ((>= runnum nruns) #f) ;;  (vector tableheader runsvec))
       ((>= testnum ntests) 
	(vector-set! runsvec runnum testvec)
	(set! bdylst (cons (apply iup:vbox (reverse res)) bdylst))
	(loop (+ runnum 1) 0 (make-vector ntests) '()))
       (else
	(let* ((button-key (mkstr runnum testnum))
	       (butn       (iup:button
			    "" ;; button-key 
			    #:size (conc cell-width btn-height )
			    #:expand "HORIZONTAL"
			    #:fontsize btn-fontsz
			    #:button-cb
			    (lambda (obj a pressed x y btn . rem)
			      ;; (print "pressed= " pressed " x= " x " y= " y " rem=" rem " btn=" btn " string? " (string? btn))
			      (if  (substring-index "3" btn)
				   (if (eq? pressed 1)
				       (let* ((toolpath (car (argv)))

     (let* ((tabdat (dboard:common-get-tabdat commondat tab-num: tab-num))
	    (dbkeys (dboard:tabdat-dbkeys tabdat)))
       ;;(print "RA => calling runs-tab-updater with commondat " commondat " tab-num " tab-num)
       ;;(tabdat-values tabdat) ;;RA added 
       ;; (pp (dboard:tabdat->alist tabdat))
       ;; (if (dashboard:database-changed? commondat tabdat context-key: 'runs-rundat)      
       (dashboard:do-update-rundat tabdat)
       ;;(debug:print-info 13 *default-log-port* "dashboard:runs-tab-updater")
       ;;(inspect tabdat)

       (let ((uidat (dboard:commondat-uidat commondat)))
	 ;;(print "RA => Calling update-buttons with tabdat : " tabdat " uidat " uidat)
	 (update-buttons tabdat uidat (dboard:tabdat-numruns tabdat) (dboard:tabdat-num-tests tabdat)))
       ))
   "dashboard:runs-tab-updater"))

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

;; each db entry is a pair ( db . dbfilepath )
;; I propose this record evolves into the area record
;;
(defstruct dbr:dbstruct 
  (tmpdb       #f)
  (dbstack     #f) ;; stack for tmp db handles, do not initialize with a stack
  (mtdb        #f)
  (refndb      #f)
  (homehost    #f) ;; not used yet
  (on-homehost #f) ;; not used yet
  (read-only   #f)
  )                ;; goal is to converge on one struct for an area but for now it is too confusing
  

;; record for keeping state,status and count for doing roll-ups in
;; iterated tests
;;
(defstruct dbr:counts

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

;; This routine creates the db if not already present. It is only called if the db is not already opened
;;
(define (db:open-db dbstruct #!key (areapath #f)) ;; TODO: actually use areapath
  (let ((tmpdb-stack (dbr:dbstruct-dbstack dbstruct))) ;; RA => Returns the first reference in dbstruct
    (if (stack? tmpdb-stack)
	(db:get-db tmpdb-stack) ;; get previously opened db (will create new db handle if all in the stack are already used
        (let* ((dbpath       (db:dbfile-path ))      ;; path to tmp db area
               (dbexists     (file-exists? dbpath))
	       (tmpdbfname   (conc dbpath "/megatest.db"))
	       (dbfexists    (file-exists? tmpdbfname)) ;; (conc dbpath "/megatest.db")))
               (tmpdb        (db:open-megatest-db path: dbpath)) ;; lock-create-open dbpath db:initialize-main-db))
               (mtdb         (db:open-megatest-db))
               (mtdbpath     (db:dbdat-get-path mtdb))
               (mtdbexists   (file-exists? mtdbpath))
               (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))
	       (tmpdbmodtime (if dbfexists  (common:lazy-sqlite-db-modification-time tmpdbfname) #f))
	       (modtimedelta (and mtdbmodtime tmpdbmodtime (- mtdbmodtime tmpdbmodtime))))
	  
          ;;(debug:print-info 13 *default-log-port* "db:open-db>> mtdbpath="mtdbpath" mtdbexists="mtdbexists" and write-access="write-access)
          (if (and dbexists (not write-access))
              (begin
                (set! *db-write-access* #f)
                (dbr:dbstruct-read-only-set! dbstruct #t)))
          (dbr:dbstruct-mtdb-set!   dbstruct mtdb)
          (dbr:dbstruct-tmpdb-set!  dbstruct tmpdb)
          (dbr:dbstruct-dbstack-set! dbstruct (make-stack)) ;; BB: why a stack?  Why would the number of db's be indeterminate?  Is this a legacy of 1.db 2.db .. ?
          (stack-push! (dbr:dbstruct-dbstack dbstruct) tmpdb) ;; olddb is already a (cons db path)
          (dbr:dbstruct-refndb-set! dbstruct refndb)
          ;;	    (mutex-unlock! *rundb-mutex*)
          (if (and write-access
		   (or (not dbfexists)

		       (and modtimedelta
			    (> modtimedelta 10)))) ;; if db in tmp is over ten seconds older than the file in MTRA then do a sync back
	      (begin
		(debug:print 0 *default-log-port* "filling db " (db:dbdat-get-path tmpdb) " with data \n    from " (db:dbdat-get-path mtdb) " mod time delta: " modtimedelta)
		(db:sync-tables (db:sync-all-tables-list dbstruct) #f mtdb refndb tmpdb)
                (debug:print-info 13 *default-log-port* "db:sync-all-tables-list done.")
                )
	      (debug:print 0 *default-log-port* " db, " (db:dbdat-get-path tmpdb) " already exists or fresh enough, not propogating data from\n     " (db:dbdat-get-path mtdb) " mod time delta: " modtimedelta) )
	  ;; (db:multi-db-sync dbstruct 'old2new))  ;; migrate data from megatest.db automatically
          tmpdb))))

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

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

;; Open the classic megatest.db file (defaults to open in toppath)
;;
;;   NOTE: returns a dbdat not a dbstruct!
;;
(define (db:open-megatest-db #!key (path #f)(name #f))
  (let* ((dbpath       (conc (or path *toppath*) "/" (or name "megatest.db")))
	 (dbexists     (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:sync-touched dbstruct run-id #!key (force-sync #f))

			 (debug:print-error 0 *default-log-port* "Failed to rebuild " dbpath ", exiting now.")
			 (exit)))))
	       (cons todb slave-dbs))
     
     0)
   ;; this is the work to be done
   (cond
    ((not fromdb) (debug:print 3 *default-log-port* "WARNING: db:sync-tables called with fromdb missing")
     -1)
    ((not todb)   (debug:print 3 *default-log-port* "WARNING: db:sync-tables called with todb missing")
     -2)
    ((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)
    (else
     (let ((stmts       (make-hash-table)) ;; table-field => stmt
	   (all-stmts   '())              ;; ( ( stmt1 value1 ) ( stml2 value2 ))
	   (numrecs     (make-hash-table))
	   (start-time  (current-milliseconds))
	   (tot-count   0))
       (for-each ;; table

							  "")
				   ";"))
		 (full-ins   (conc "INSERT OR REPLACE INTO " tablename " ( " (string-intersperse (map car fields) ",") " ) "
				   " VALUES ( " (string-intersperse (make-list num-fields "?") ",") " );"))
		 (fromdat    '())
		 (fromdats   '())
		 (totrecords 0)
		 (batch-len  (string->number (or (configf:lookup *configdat* "sync" "batchsize") "100")))
		 (todat      (make-hash-table))
		 (count      0))

	    ;; set up the field->num table
	    (for-each
	     (lambda (field)
	       (hash-table-set! field->num field count)

	  ))))

;; options:
;;
;;  'killservers  - kills all servers
;;  'dejunk       - removes junk records
;;  'adj-testids  - move test-ids into correct ranges
;;  'old2new      - sync megatest.db to /tmp/.../megatest.db and /tmp/.../megatest_ref.db
;;  'new2old      - sync /tmp/.../megatest.db to megatest.db and /tmp/.../megatest_ref.db (and update data_synced)
;;  'closeall     - close all opened dbs
;;  'schema       - attempt to apply schema changes

;;  run-ids: '(1 2 3 ...) or #f (for all)
;;
(define (db:multi-db-sync dbstruct . options)
  (if (not (launch:setup))
      (debug:print 0 *default-log-port* "ERROR: not able to setup up for megatest.")
      (let* ((mtdb     (dbr:dbstruct-mtdb dbstruct))
	     (tmpdb    (db:get-db dbstruct))

	;;       (db:delay-if-busy mtdb)
	;;       (db:prep-megatest.db-for-migration mtdb)))

	;; sync runs, test_meta etc.
	;;
	(if (member 'old2new options)
	    ;; (begin
            (set! data-synced
                  (+ (db:sync-tables (db:sync-all-tables-list dbstruct) #f mtdb tmpdb refndb)
                     data-synced)))
			      ;; (db:sync-main-list mtdb) mtdb (db:get-db dbstruct #f))
;; 	      (for-each 
;; 	       (lambda (run-id)
;; 		 (db:delay-if-busy mtdb)
;; 		 (let ((testrecs (db:get-all-tests-info-by-run-id mtdb run-id)))
;; ;;		       (dbstruct (if toppath (make-dbr:dbstruct path: toppath local: #t) #f)))
;; 		   (debug:print 0 *default-log-port* "INFO: Propagating " (length testrecs) " records for run-id=" run-id " to run specific db")
;; 		   (db:replace-test-records dbstruct run-id testrecs)
;; 		   (sqlite3:finalize! (db:dbdat-get-db (dbr:dbstruct-rundb dbstruct)))))
;; 	       run-ids)))

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



        (if (member 'schema options)
            (begin
              (db:patch-schema-maindb (db:dbdat-get-db mtdb))
              (db:patch-schema-maindb (db:dbdat-get-db tmpdb))
              (db:patch-schema-maindb (db:dbdat-get-db refndb))

	       (csvt (let-values (((fmt-cell fmt-record fmt-csv) (make-format ",")))
		       (fmt-csv (map list->csv-record csvr))))
	       (status (configf:lookup dat "final" "exit-status"))
	       (msg     (configf:lookup dat "final" "message")))
          (if csvt  ;; this if blocked stack dump caused by .dat file from logpro being 0-byte.  fixed by upgrading logpro
              (rmt:csv->test-data run-id test-id csvt)
	      (debug:print 0 *default-log-port* "ERROR: no csvdat exists for run-id: " run-id " test-id: " test-id " stepname: " stepname ", check that logpro version is 1.15 or newer"))
	  ;;  (debug:print-info 13 *default-log-port* "Error: run-id/test-id/stepname="run-id"/"test-id"/"stepname" => bad csvr="csvr)
	  ;;  )
	  (cond
	   ((equal? status "PASS") "PASS") ;; skip the message part if status is pass
	   (status (conc (configf:lookup dat "final" "exit-status") ": " (if msg msg "no message")))
	   (else #f)))
	#f)))

;;   returns:
;;     *toppath*
;;   side effects:
;;     sets; *configdat*    (megatest.config info)
;;           *runconfigdat* (runconfigs.config info)
;;           *configstatus* (status of the read data)
;;
(define (launch:setup #!key (force #f) (areapath #f))
  (mutex-lock! *launch-setup-mutex*)
  (if (and *toppath*
	   (eq? *configstatus* 'fulldata)) ;; got it all
      (begin
	(debug:print 0 *default-log-port* "NOTE: skipping launch:setup-body call since we have fulldata")
	(mutex-unlock! *launch-setup-mutex*)
	*toppath*)
      (let ((res (launch:setup-body force: force areapath: areapath)))
	(mutex-unlock! *launch-setup-mutex*)
	res)))

(define (launch:setup-body #!key (force #f) (areapath #f))
  (let* ((toppath  (or *toppath* areapath (getenv "MT_RUN_AREA_HOME"))) ;; preserve toppath
	 (runname  (common:args-get-runname))
	 (target   (common:args-get-target))
	 (linktree (common:get-linktree))
	 (contour  (args:get-arg "-contour"))
	 (sections (if target (list "default" target) #f)) ;; for runconfigs
	 (mtconfig (or (args:get-arg "-config") "megatest.config")) ;; allow overriding megatest.config 
	 (rundir   (if (and runname target linktree)(conc linktree (if contour (conc "/" contour) "") "/" target "/" runname) #f))

	  (setenv "MT_RUN_AREA_HOME" *toppath*)
	  (setenv "MT_TESTSUITE_NAME" (common:get-testsuite-name)))
	(begin
	  (debug:print-error 0 *default-log-port* "failed to find the top path to your Megatest area.")
          ;;(exit 1)
          #f
          ))
    ;; if have -append-config then read and append here
    (let ((cfname (args:get-arg "-append-config")))
      (if (and cfname
	       (file-read-access? cfname))
	  (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")))
		     (string->number (or m "10000")))))

  -stop-server id         : stop server specified by id (see output of -list-servers), use
                            0 to kill all
  -repl                   : start a repl (useful for extending megatest)
  -load file.scm          : load and run file.scm
  -mark-incompletes       : find and mark incomplete tests
  -ping run-id|host:port  : ping server, exit with 0 if found
  -debug N|N,M,O...       : enable debug 0-N or N and M and O ...
  -config fname           : override the megatest.config file with fname
  -append-config fname    : append fname to the megatest.config file

Utilities
  -env2file fname         : write the environment to fname.csh and fname.sh
  -envcap fname=context   : save current variables labeled as context in file fname
  -refdb2dat refdb        : convert refdb to sexp or to format specified by -dumpmode
                            formats: perl, ruby, sqlite3, csv (for csv the -o param
                            will substitute %s for the sheet name in generating 

# Get test path, use '.' to get a single path or a specific path/file pattern
megatest -test-files 'logs/*.log' -target ubuntu/n%/no% -runname w49% -testpatt test_mt%

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

;;  -gui                    : start a gui interface


;; process args
(define remargs (args:get-args 
		 (argv)
		 (list  "-runtests"  ;; run a specific test
			"-config"    ;; override the config file name
			"-append-config"
			"-execute"   ;; run the command encoded in the base64 parameter
			"-step"
			"-target"
			"-reqtarg"
			":runname"
			"-runname"
			":state"  

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

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

;; TODO: for multiple areas, we will have multiple watchdogs; and multiple threads to manage
(define *watchdog* (make-thread common:watchdog "Watchdog thread"))

(if (not (args:get-arg "-server"))
    (thread-start! *watchdog*)) ;; if starting a server; wait till we get to running state before kicking off watchdog

;; bracket open-output-file with code to make leading directory if it does not exist and handle exceptions
(define (open-logfile logpath)

			       (tal (cdr adj-tests-spec))
			       (idx 0))
		      (hash-table-set! test-field-index hed idx)
		      (if (not (null? tal))(loop (car tal)(cdr tal)(+ idx 1))))
		    (begin
		      (debug:print-error 0 *default-log-port* "Invalid test fields specified: " (string-intersperse invalid-tests-spec ", "))
		      (exit)))))

	  ;; Each run
	  (for-each 
	   (lambda (run)
	     (let ((targetstr (string-intersperse (map (lambda (x)
							 (db:get-value-by-header run header x))
						       keys) "/")))
	       (if db-targets

				       outputfile
				       (begin
					 (debug:print 0 *default-log-port* "WARNING: path given, " outputfile " is relative, prefixing with current directory")
					 (conc (current-directory) "/" outputfile)))))
		  (create-directory tempdir #t)
		  (ods:list->ods tempdir ouf sheets))))
	  ;; (system (conc "rm -rf " tempdir))
	  (set! *didsomething* #t)
          (set! *time-to-exit* #t)
          ) ;; end if true branch (end of a let)
        ) ;; end if
    ) ;; end if -list-runs

;; Don't think I need this. Incorporated into -list-runs instead
;;
;; (if (and (args:get-arg "-since")
;; 	 (launch:setup))
;;     (let* ((since-time (string->number (args:get-arg "-since")))
;; 	   (run-ids    (db:get-changed-run-ids since-time)))

;;======================================================================
;; Exit and clean up
;;======================================================================

(if (not *didsomething*)
    (debug:print 0 *default-log-port* help))
;;(debug:print-info 13 *default-log-port* "thread-join! watchdog")

;; join the watchdog thread if it has been thread-start!ed  (it may not have been started in the case of a server that never enters running state)
;;   (symbols returned by thread-state: created ready running blocked suspended sleeping terminated dead)
;; TODO: for multiple areas, we will have multiple watchdogs; and multiple threads to manage
(if (thread? *watchdog*)
    (case (thread-state *watchdog*)
      ((ready running blocked sleeping terminated dead)
       (thread-join! *watchdog*))))

(set! *time-to-exit* #t)

CREATE TABLE IF NOT EXISTS keys (
       id INTEGER PRIMARY KEY,
       fieldname TEXT,
       fieldtype TEXT,
       CONSTRAINT keyconstraint UNIQUE (fieldname));
		    
CREATE TABLE IF NOT EXISTS areas (
       id INTEGER PRIMARY KEY,
       areaname TEXT DEFAULT 'local',
       areapath TEXT DEFAULT '.',
       last_sync INTEGER DEFAULT 0,
       CONSTRAINT areaconstraint UNIQUE (areaname));

INSERT INTO areas (id,areaname,areapath) VALUES (0,'local','.');

CREATE TABLE IF NOT EXISTS ttype (
       id INTEGER PRIMARY KEY,
       target_spec TEXT DEFAULT '');
       
CREATE TABLE IF NOT EXISTS runs (
       id INTEGER PRIMARY KEY,
       target     TEXT DEFAULT '',
       ttype_id   INTEGER DEFAULT 0,
       runname    TEXT DEFAULT 'norun',
       state      TEXT DEFAULT '',
       status     TEXT DEFAULT '',
       owner      TEXT DEFAULT '',
       event_time INTEGER DEFAULT extract(epoch from now()),
       comment    TEXT DEFAULT '',
       fail_count INTEGER DEFAULT 0,
       pass_count INTEGER DEFAULT 0,
       last_update INTEGER DEFAULT extract(epoch from now()),
       area_id     INTEGER DEFAULT 0,
       CONSTRAINT runsconstraint UNIQUE (runname));

CREATE TABLE IF NOT EXISTS run_stats (
       id     INTEGER PRIMARY KEY,
       run_id INTEGER,
       state  TEXT,
       status TEXT,
       count  INTEGER,
       last_update INTEGER DEFAULT extract(epoch from now()));

CREATE TABLE IF NOT EXISTS test_meta (
       id          INTEGER PRIMARY KEY,
       testname    TEXT DEFAULT '',
       author      TEXT DEFAULT '',
       owner       TEXT DEFAULT '',
       description TEXT DEFAULT '',
       reviewed    TEXT,
       iterated    TEXT DEFAULT '',
       avg_runtime REAL,
       avg_disk    REAL,
       tags        TEXT DEFAULT '',
       jobgroup    TEXT DEFAULT 'default',
       CONSTRAINT test_meta_constraint UNIQUE (testname));

CREATE TABLE IF NOT EXISTS tasks_queue (
       id INTEGER PRIMARY KEY,
       action TEXT DEFAULT '',
       owner TEXT,
       state TEXT DEFAULT 'new',
       target TEXT DEFAULT '',
       name TEXT DEFAULT '',
       testpatt TEXT DEFAULT '',
       keylock TEXT,
       params TEXT,
       creation_time INTEGER DEFAULT extract(epoch from now()),
       execution_time INTEGER);

CREATE TABLE IF NOT EXISTS archive_disks (
       id INTEGER PRIMARY KEY,
       archive_area_name TEXT,
       disk_path TEXT,
       last_df INTEGER DEFAULT -1,
       last_df_time INTEGER DEFAULT extract(epoch from now()),
       creation_time INTEGER DEFAULT extract(epoch from now()));

CREATE TABLE IF NOT EXISTS archive_blocks (
       id INTEGER PRIMARY KEY,
       archive_disk_id INTEGER,
       disk_path TEXT,
       last_du INTEGER DEFAULT -1,
       last_du_time INTEGER DEFAULT extract(epoch from now()),
       creation_time INTEGER DEFAULT extract(epoch from now()));

CREATE TABLE IF NOT EXISTS archive_allocations (
       id INTEGER PRIMARY KEY,
       archive_block_id INTEGER,
       testname TEXT,
       item_path TEXT,
       creation_time INTEGER DEFAULT extract(epoch from now()));

CREATE TABLE IF NOT EXISTS extradat (
       id INTEGER PRIMARY KEY,
       run_id INTEGER,
       key TEXT,
       val TEXT);

CREATE TABLE IF NOT EXISTS metadat (
       id INTEGER PRIMARY KEY,
       var TEXT,
       val TEXT);

CREATE TABLE IF NOT EXISTS access_log (
       id INTEGER PRIMARY KEY,
       "user" TEXT,
       accessed TIMESTAMP,
       args TEXT);

CREATE TABLE tests  (
       id INTEGER PRIMARY KEY,                                                                            
       run_id       INTEGER   DEFAULT -1,                                                                 
       testname     TEXT      DEFAULT 'noname',                                                           
       host         TEXT      DEFAULT 'n/a',                                                              
       cpuload      REAL      DEFAULT -1,                                                                 
       diskfree     INTEGER   DEFAULT -1,                                                                 
       uname        TEXT      DEFAULT 'n/a',                                                              
       rundir       TEXT      DEFAULT '/tmp/badname',                                                     
       shortdir     TEXT      DEFAULT '/tmp/badname',                                                     
       item_path    TEXT      DEFAULT '',                                                                 
       state        TEXT      DEFAULT 'NOT_STARTED',                                                      
       status       TEXT      DEFAULT 'FAIL',                                                             
       attemptnum   INTEGER   DEFAULT 0,                                                                  
       final_logf   TEXT      DEFAULT 'logs/final.log',                                                   
       logdat       TEXT      DEFAULT '',                                                                 
       run_duration INTEGER   DEFAULT 0,                                                                  
       comment      TEXT      DEFAULT '',                                                                 
       event_time   INTEGER DEFAULT extract(epoch from now()),                                             
       fail_count   INTEGER   DEFAULT 0,                                                                  
       pass_count   INTEGER   DEFAULT 0,                                                                  
       archived     INTEGER   DEFAULT 0, -- 0=no, > 1=archive block id where test data can be found       
       last_update  INTEGER DEFAULT extract(epoch from now()),                                               
       CONSTRAINT testsconstraint UNIQUE (run_id, testname, item_path));  

CREATE TABLE IF NOT EXISTS test_steps (
       id INTEGER PRIMARY KEY,
       test_id INTEGER, 
       stepname TEXT, 
       state TEXT DEFAULT 'NOT_STARTED', 
       status TEXT DEFAULT 'n/a',
       event_time INTEGER DEFAULT extract(epoch from now()),
       comment TEXT DEFAULT '',
       logfile TEXT DEFAULT '',
       last_update  INTEGER DEFAULT extract(epoch from now()),
       CONSTRAINT test_steps_constraint UNIQUE (test_id,stepname,state));

CREATE TABLE IF NOT EXISTS test_data (
       id INTEGER PRIMARY KEY,
       test_id INTEGER,
       category TEXT DEFAULT '',
       variable TEXT,
       value REAL,
       expected REAL,
       tol REAL,
       units TEXT,
       comment TEXT DEFAULT '',
       status TEXT DEFAULT 'n/a',
       type TEXT DEFAULT '',
       last_update  INTEGER DEFAULT extract(epoch from now()),
       CONSTRAINT test_data_constraint UNIQUE (test_id,category,variable));

CREATE TABLE IF NOT EXISTS test_rundat (
       id           INTEGER PRIMARY KEY,
       test_id      INTEGER,
       update_time  INTEGER,
       cpuload      INTEGER DEFAULT -1,
       diskfree     INTEGER DEFAULT -1,
       diskusage    INTEGER DEFAULT -1,
       run_duration INTEGER DEFAULT 0);

CREATE TABLE IF NOT EXISTS archives (
       id           INTEGER PRIMARY KEY,
       test_id      INTEGER,
       state        TEXT DEFAULT 'new',
       status       TEXT DEFAULT 'n/a',
       archive_type TEXT DEFAULT 'bup',
       du           INTEGER,
       archive_path TEXT);
 

TRUNCATE archive_blocks, archive_allocations, extradat, metadat, access_log, tests, test_steps, test_data, test_rundat, archives, keys, runs, run_stats, test_meta, tasks_queue, archive_disks;

     srfi-18 extras format pkts regex
     (prefix dbi dbi:)) ;;  zmq extras)

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

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

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

Actions:
   run                     : initiate runs
   remove                  : remove runs
   rerun                   : register action for processing
   set-ss                  : set state/status
   archive                 : compress and move test data to archive disk
   kill                    : stop tests or entire runs
   db                      : database utilities

Contour actions:
   process                 : runs import, rungen and dispatch 

Selectors 
  -immediate               : apply this action immediately, default is to queue up actions
  -area areapatt1,area2... : apply this action only to the specified areas

  -set-vars V1=1,V2=2      : Add environment variables to a run NB// these are
                                 overwritten by values set in config files.
  -log logfile             : send stdout and stderr to logfile
  -repl                    : start a repl (useful for extending megatest)
  -load file.scm           : load and run file.scm
  -debug N|N,M,O...        : enable debug messages 0-N or N and M and O ...

Utility
  pgschema                 : emit postgresql schema; do \"mtutil db pgschema | psql -d mydb\"

Examples:

# Start a megatest run in the area \"mytests\"
mtutil -area mytests -action run -target v1.63/aa3e -mode-patt MYPATT -tag-expr quick

# Start a contour
mtutil run -contour quick -target v1.63/aa3e 

;; Add args that use remargs here
;;
(if (and (not (null? remargs))
	 (not (or
	       (args:get-arg "-runstep")
	       (args:get-arg "-envcap")
	       (args:get-arg "-envdelta")
	       (member *action* '("db"))   ;; very loose checks on db.
	       )))
    (debug:print-error 0 *default-log-port* "Unrecognised arguments: " (string-intersperse (if (list? remargs) remargs (argv))  " ")))

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

	   ((process)  (begin
			 (load-pkts-to-db mtconf)
			 (generate-run-pkts mtconf toppath)
			 (load-pkts-to-db mtconf)
			 (dispatch-commands mtconf toppath)))
	   ((import)   (load-pkts-to-db mtconf)) ;; import pkts
	   ((rungen)   (generate-run-pkts mtconf toppath))
	   ((dispatch) (dispatch-commands mtconf toppath)))))
      ((db)
       (if (null? remargs)
	   (print "ERROR: missing sub command for db command")
	   (let ((subcmd (car remargs)))
	     (case (string->symbol subcmd)
	       ((pgschema)
		(let* ((install-home (common:get-install-area))
		       (schema-file  (conc install-home "/share/db/mt-pg.sql")))
		  (if (file-exists? schema-file)
		      (system (conc "/bin/cat " schema-file)))))
	       ((junk)
		(rmt:get-keys))))))))

(if (or (args:get-arg "-repl")
	(args:get-arg "-load"))
    (begin
      (import extras) ;; might not be needed
      ;; (import csi)
      (import readline)

  (mutex-lock! *rmt-mutex*)

  ;; 1. check if server is started IFF cmd is a write OR if we are not on the homehost, store in runremote
  ;; 2. check the age of the connections. refresh the connection if it is older than timeout-20 seconds.
  ;; 3. do the query, if on homehost use local access
  ;;
  (let* ((start-time (current-seconds)) ;; snapshot time so all use cases get same value
         (areapath *toppath*);; TODO - resolve from dbstruct to be compatible with multiple areas
         (dbfile (conc *toppath* "/megatest.db"))
         (readonly-mode (not (file-write-access? dbfile))) ;; TODO: use dbstruct or runremote to figure this out in future
	 (runremote  (or area-dat *runremote*)))
    ;;(print "BB> readonly-mode is "readonly-mode" dbfile is "dbfile)
    (cond
     ;; give up if more than 15 attempts
     ((> attemptnum 15)
      (debug:print 0 *default-log-port* "ERROR: 15 tries to start/connect to server. Giving up.")
      (exit 1))

     ;; readonly mode, read request-  handle it - case 20
     ((and readonly-mode
           (member cmd api:read-only-queries)) 
      (mutex-unlock! *rmt-mutex*)
      (debug:print-info 12 *default-log-port* "rmt:send-receive, case 20")
      (rmt:open-qry-close-locally cmd 0 params)
      )

     ;; readonly mode, write request.  Do nothing, return #f
     (readonly-mode
      (mutex-unlock! *rmt-mutex*)
      (debug:print-info 12 *default-log-port* "rmt:send-receive, case 21")
      (debug:print 0 *default-log-port* "WARNING: write transaction requested on a readonly area.  cmd="cmd" params="params)
      #f
      )

     ;; reset the connection if it has been unused too long
     ((and runremote
           (remote-conndat runremote)
	   (let ((expire-time (+ (- start-time (remote-server-timeout runremote))(random 30)))) ;; add 30 seconds of noise so that not all running tests expire at the same time causing a storm of server starts
	     (< (http-transport:server-dat-get-last-access (remote-conndat runremote)) expire-time)))
      (debug:print-info 12 *default-log-port* "rmt:send-receive, case  8")
      (remote-conndat-set! runremote #f)

;;            
(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
	 (task-key           (conc (hash-table->alist flags) " " (get-host-name) " " (current-process-id)))
	 (tdbdat             (tasks:open-db))
	 (config-reruns      (let ((x (configf:lookup *configdat* "setup" "reruns")))
			       (if x (string->number x) #f)))
	 (allowed-tests      #f))

    ;; check if readonly
    (when readonly-mode
      (debug:print-error 0 *default-log-port* "megatest.db is readonly.  Cannot proceed.")
      (exit 1))

    ;; per user request. If less than 100Meg space on dbdir partition, bail out with error
    ;; this will reduce issues in database corruption
    (common:check-db-dir-and-exit-if-insufficient)

    ;; override the number of reruns from the configs
    (if (and config-reruns
	     (> run-count config-reruns))
	(set! run-count config-reruns))
    
    ;; (if (tasks:need-server run-id)(tasks:start-and-wait-for-server tdbdat run-id 10))

	  (debug:print-info 4 *default-log-port* "Exiting loop with...\n  hed=" hed "\n  tal=" tal "\n  reruns=" reruns))
	 )))
    ;; now *if* -run-wait we wait for all tests to be done
    ;; Now wait for any RUNNING tests to complete (if in run-wait mode)
    (thread-sleep! 5) ;; I think there is a race condition here. Let states/statuses settle
    (let wait-loop ((num-running      (rmt:get-count-tests-running-for-run-id run-id))
		    (prev-num-running 0))
      ;; (debug:print-info 13 *default-log-port* "num-running=" num-running ", prev-num-running=" prev-num-running)
      (if (and (or (args:get-arg "-run-wait")
		   (equal? (configf:lookup *configdat* "setup" "run-wait") "yes"))
	       (> num-running 0))
	  (begin
	    ;; Here we mark any old defunct tests as incomplete. Do this every fifteen minutes
	    ;; (debug:print 0 *default-log-port* "Got here eh! num-running=" num-running " (> num-running 0) " (> num-running 0))
	    (if (> (current-seconds)(+ last-time-incomplete 900))

	 (header       (vector-ref rundat 0))
	 (runs         (vector-ref rundat 1))
	 (states       (if state  (string-split state  ",") '()))
	 (statuses     (if status (string-split status ",") '()))
	 (state-status (if (string? new-state-status) (string-split new-state-status ",") '(#f #f)))
	 (rp-mutex     (make-mutex))
	 (bup-mutex    (make-mutex)))

    (let* ((write-access-actions '(remove-runs set-state-status archive run-wait))
           (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))
	(begin
	  (debug:print-error 0 *default-log-port* "the parameter to -set-state-status is a comma delimited string. E.g. COMPLETED,FAIL")
	  (exit)))
    (for-each
     (lambda (run)

	(if (launch:setup)
	    (begin
	      (full-runconfigs-read) ;; cache the run config
	      (launch:cache-config)) ;; do not cache here - need to be sure runconfigs is processed
	    (begin 
	      (debug:print 0 *default-log-port* "Failed to setup, exiting")
	      (exit 1)))

        
	(set! keys (keys:config-get-fields *configdat*))
	;; have enough to process -target or -reqtarg here
	(if (args:get-arg "-reqtarg")
	    (let* ((runconfigf (conc  *toppath* "/runconfigs.config")) ;; DO NOT EVALUATE ALL 
		   (runconfig  (read-config runconfigf #f #t environ-patt: #f)))
	      (if (hash-table-ref/default runconfig (args:get-arg "-reqtarg") #f)
		  (keys:target-set-args keys (args:get-arg "-reqtarg") args:arg-hash)