Megatest

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

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

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

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

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

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

                (dbstruct (db:setup #t)))
	    (debug:print 0 *default-log-port*
			 "WARNING: Version mismatch!\n"
			 "   expected: " (common:version-signature) "\n"
			 "   got:      " (common:get-last-run-version))
            (cond
             ((get-environment-variable "MT_SKIP_DB_MIGRATE") #t)
             ((and (common:file-exists? mtconf) (common:file-exists? dbfile) (not read-only)
                   (eq? (current-user-id)(file-owner mtconf))) ;; safe to run -cleanup-db
              (debug:print 0 *default-log-port* "   I see you are the owner of megatest.config, attempting to cleanup and reset to new version")
              (handle-exceptions
               exn
               (begin
                 (debug:print 0 *default-log-port* "Failed to switch versions.")
                 (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
                 (print-call-chain (current-error-port))
                 (exit 1))
               (common:cleanup-db dbstruct)))
             ((not (common:file-exists? mtconf))
              (debug:print 0 *default-log-port* "   megatest.config does not exist in this area.  Cannot proceed with megatest version migration.")
              (exit 1))
             ((not (common:file-exists? dbfile))
              (debug:print 0 *default-log-port* "   megatest.db does not exist in this area.  Cannot proceed with megatest version migration.")
              (exit 1))
             ((not (eq? (current-user-id)(file-owner mtconf)))
              (debug:print 0 *default-log-port* "   You do not own megatest.db in this area.  Cannot proceed with megatest version migration.")
              (exit 1))
             (read-only
              (debug:print 0 *default-log-port* "   You have read-only access to this area.  Cannot proceed with megatest version migration.")

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

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

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

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

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

(include "common_records.scm")

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

;;
;;(define *db-open-mutex* (make-mutex))

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

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

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

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

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

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

;;     ;; (debug:print-info 0 *default-log-port* "Using cache dir " cache-dir)
;;     (if (not cache-dir)
;; 	(begin
;; 	  (debug:print 0 *default-log-port* "ERROR: Failed to find an area to write the cache db")
;; 	  (exit 1))
;; 	(let* ((th1      (make-thread
;; 			  (lambda ()
;; 			    (if (and (common:file-exists? megatest-db)
;; 				     (file-write-access? megatest-db))
;; 				(begin
;; 				  (db:sync-to-megatest.db dbstruct 'timestamps) ;; internally mutexes on *db-local-sync*
;; 				  (debug:print-info 2 *default-log-port* "Done syncing to megatest.db"))))
;; 			  "call-with-cached-db sync-to-megatest.db"))
;; 	       (cache-db (db:cache-for-read-only
;; 			  megatest-db

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

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

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

	res)))

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

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

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

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

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

(declare (unit env))

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

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

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

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

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

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

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

(declare (unit filedb))

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

	  (if contour (setenv "MT_CONTOUR" contour))
	  

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

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

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

	  (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)

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

			(create-directory linktree #t))))
		(handle-exceptions
		    exn
		    (begin
		      (debug:print-error 0 *default-log-port* "Something went wrong when trying to create link to linktree at " *toppath*)
		      (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn)))
		  (let ((tlink (conc *toppath* "/lt")))
		    (if (not (common:file-exists? tlink))
			(create-symbolic-link linktree tlink)))))
	      (begin
		(debug:print-error 0 *default-log-port* "linktree not defined in [setup] section of megatest.config")
		)))
	(if (and *toppath*
		 (directory-exists? *toppath*))
	    (begin
	      (setenv "MT_RUN_AREA_HOME" *toppath*)
	      (setenv "MT_TESTSUITENAME" (common:get-testsuite-name)))
	    (begin
	      (debug:print-error 0 *default-log-port* "failed to find the top path to your Megatest area.")
	      (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))
               (mtcachef     (car cachefiles))
               (rccachef     (cdr cachefiles)))
          (if (and rccachef *runconfigdat* (not (common:file-exists? rccachef))) (configf:write-alist *runconfigdat* rccachef))
          (if (and mtcachef *configdat*    (not (common:file-exists? mtcachef))) (configf:write-alist *configdat* mtcachef))
          (if (and rccachef mtcachef *runconfigdat* *configdat*)
              (set! *configstatus* 'fulldata)))

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

;;	*runconfigdat*)))

  (let* ((rundir (if (and (getenv "MT_LINKTREE")(getenv "MT_TARGET")(getenv "MT_RUNNAME"))
		     (conc (getenv "MT_LINKTREE") "/" (getenv "MT_TARGET") "/" (getenv "MT_RUNNAME"))
		     #f))
	 (cfgf   (if rundir (conc rundir "/.runconfig." megatest-version "-" megatest-fossil-hash) #f)))
    (if (and cfgf
	     (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))

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

		   (event-time    (db:test-get-event_time   test-dat))
		   (tconfig       #f)
		   (state         (if newstate  newstate  (db:test-get-state  test-dat)))
		   (status        (if newstatus newstatus (db:test-get-status test-dat))))
	      ;; (mutex-lock! *triggers-mutex*)
	      (if (and test-name
		       test-rundir)   ;; #f means no dir set yet
		       ;; (common:file-exists? test-rundir)
		       ;; (directory? test-rundir))
		  (call-with-environment-variables
		   (list (cons "MT_TEST_NAME"    (or test-name "no such test"))
			 (cons "MT_TEST_RUN_DIR" (or test-rundir "no test directory yet"))
			 (cons "MT_ITEMPATH"     (or item-path "")))
		   (lambda ()
		     (if (directory-exists? test-rundir)

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

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

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

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

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

(define (load-pkts-to-db mtconf)
  (with-queue-db
   mtconf
   (lambda (pktsdirs pktsdir pdb)
     (for-each
      (lambda (pktsdir) ;; look at all
	(if (and (common:file-exists? pktsdir)
		 (directory? pktsdir)
		 (file-read-access? pktsdir))
	    (let ((pkts (glob (conc pktsdir "/*.pkt"))))
	      (for-each
	       (lambda (pkt)
		 (let* ((uuid    (cadr (string-match ".*/([0-9a-f]+).pkt" pkt)))
			(exists  (lookup-by-uuid pdb uuid #f)))

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

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

(if *action*
    (case (string->symbol *action*)
      ((run remove rerun set-ss archive kill)
       (let* ((mtconfdat (simple-setup (args:get-arg "-start-dir")))
	      (mtconf    (car mtconfdat))

       (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 (common:file-exists? schema-file)
		      (system (conc "/bin/cat " schema-file)))))
	       ((sqlite3schema)
		(let* ((install-home (common:get-install-area))
		       (schema-file  (conc install-home "/share/db/mt-sqlite3.sql")))
		  (if (common:file-exists? schema-file)
		      (system (conc "/bin/cat " schema-file)))))
	       ((junk)
		(rmt:get-keys))))))))

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

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

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

(debug:setup)

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

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

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

;; data format:
;;   '( (sheet1 (r1c1 r1c2 r1c3 ...)
;;              (r2c1 r2c3 r2c3 ...) )
;;      (sheet2 ( ... )
;;              ( ... ) ) )
(define (ods:list->ods path fname data)
  (if (not (common:file-exists? path))
      (print "ERROR: path to create ods data must pre-exist")
      (begin
	(with-output-to-file (conc path "/content.xml")
	  (lambda ()
	    (ods:construct-dir path)
	    (ods:add-non-content-files path)
	    (ods:make-thumbnail path)

(declare (uses db))

;; lsof -i


(define (portlogger:open-db fname)
  (let* ((avail    (tasks:wait-on-journal fname 5 remove: #t)) ;; wait up to about 10 seconds for the journal to go away
	 (exists   (common:file-exists? fname))
	 (db       (if avail 
		       (sqlite3:open-database fname)
		       (begin
			 (system (conc "rm -f " fname))
			 (sqlite3:open-database fname))))
	 (handler  (make-busy-timeout 136000))
	 (canwrite (file-write-access? fname)))

    (handle-exceptions
     exn
     (begin
       ;; (release-dot-lock fname)
       (debug:print-error 0 *default-log-port* "portlogger:open-run-close failed. " proc " " params)
       (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
       (debug:print 5 *default-log-port* "exn=" (condition->list exn))
       (if (common:file-exists? fname)(delete-file fname)) ;; brutally get rid of it
       (print-call-chain (current-error-port)))
     (let* (;; (lock   (obtain-dot-lock fname 2 9 10))
	    (db     (portlogger:open-db fname))
	    (res    (apply proc db params)))
       (sqlite3:finalize! db)
       ;; (release-dot-lock fname)
       res))))

	     (if proc (proc pid))
	     (loop (read-line) (cons pid res))))))))

(define (process:alive? pid)
  (handle-exceptions
   exn
   ;; possibly pid is a process not a child, look in /proc to see if it is running still
   (common:file-exists? (conc "/proc/" pid))
   (let-values (((rpid exit-type exit-signal)(process-wait pid #t)))
       (and (number? rpid)
	    (equal? rpid pid)))))

(define (process:alive-on-host? host pid)
  (let ((cmd (conc "ssh " host " ps -o pid= -p " pid)))
    (handle-exceptions

(define (rmt:test-set-state-status-by-id run-id test-id newstate newstatus newcomment)
  (rmt:send-receive 'test-set-state-status-by-id run-id (list run-id test-id newstate newstatus newcomment)))

(define (rmt:set-tests-state-status run-id                      testnames currstate currstatus newstate newstatus)
  (rmt:send-receive 'set-tests-state-status run-id (list run-id testnames currstate currstatus newstate newstatus)))

(define (rmt:get-tests-for-run run-id testpatt states statuses offset limit not-in sort-by sort-order qryvals last-update mode)
  ;; (if (number? run-id)
  (rmt:send-receive 'get-tests-for-run run-id (list run-id testpatt states statuses offset limit not-in sort-by sort-order qryvals last-update mode)))
  ;;    (begin
  ;;	(debug:print-error 0 *default-log-port* "rmt:get-tests-for-run called with bad run-id=" run-id)
  ;;	(print-call-chain (current-error-port))
  ;;	'())))

;; get stuff via synchash 
(define (rmt:synchash-get run-id proc synckey keynum params)
  (rmt:send-receive 'synchash-get run-id (list run-id proc synckey keynum params)))

;; IDEA: Threadify these - they spend a lot of time waiting ...
;;

(define (set-run-config-vars run-id keyvals targ-from-db)
  (push-directory *toppath*) ;; the push/pop doesn't appear to do anything ...
  (let ((runconfigf (conc  *toppath* "/runconfigs.config"))
	(targ       (or (common:args-get-target)
			targ-from-db
			(get-environment-variable "MT_TARGET"))))
    (pop-directory)
    (if (common:file-exists? runconfigf)
	(setup-env-defaults runconfigf run-id #t keyvals
			    environ-patt: (conc "(default"
						(if targ
						    (conc "|" targ ")")
						    ")")))
	(debug:print 0 *default-log-port* "WARNING: You do not have a run config file: " runconfigf))))

      (set-signal-handler! signal/term sighand))

    ;; force the starting of a server
    (debug:print 0 *default-log-port* "waiting on server...")
    (server:start-and-wait *toppath*)
    
    (runs:set-megatest-env-vars run-id inkeys: keys inrunname: runname) ;; these may be needed by the launching process
    (set! runconf (if (common:file-exists? runconfigf)
		      (setup-env-defaults runconfigf run-id *already-seen-runconfig-info* keyvals target)
		      (begin
			(debug:print 0 *default-log-port* "WARNING: You do not have a run config file: " runconfigf)
			#f)))

    (if (not test-patts) ;; first time in - adjust testpatt
	(set! test-patts (common:args-get-testpatt runconf)))

		(begin
		  (debug:print-info 0 *default-log-port* "WARNING: server is overloaded, trying again in one second")
		  (thread-sleep! 1)
		  (loop)))))
      (if (not testdat) ;; should NOT happen
	  (debug:print-error 0 *default-log-port* "failed to get test record for test-id " test-id))
      (set! test-id (db:test-get-id testdat))
      (if (common:file-exists? test-path)
	  (change-directory test-path)
	  (begin
	    (debug:print-error 0 *default-log-port* "test run path not created before attempting to run the test. Perhaps you are running -remove-runs at the same time?")
	    (change-directory *toppath*)))
      (case (if force ;; (args:get-arg "-force")
		'NOT_STARTED
		(if testdat

				      (debug:print-info 1 *default-log-port* "Run completed according to zero tests matching provided criteria.")
				      (loop (car new-tests)(cdr new-tests)))))
			       ((archive)
				(if (and run-dir (not toplevel-with-children))
				    (let ((ddir (conc run-dir "/")))
				      (case (string->symbol (args:get-arg "-archive"))
					((save save-remove keep-html)
					 (if (common:file-exists? ddir)
					     (debug:print-info 0 *default-log-port* "Estimating disk space usage for " test-fulln ": " (common:get-disk-space-used ddir)))))))
				(if (not (null? tal))
				    (loop (car tal)(cdr tal))))
			       )))
		       )
		     (if worker-thread (thread-join! worker-thread))))))
	   ;; remove the run if zero tests remain
	   (if (eq? action 'remove-runs)
	       (let* ((run-id   (db:get-value-by-header run header "id")) ;; NB// masks run-id from above?
                      (remtests (mt:get-tests-for-run run-id #f '("DELETED") '("n/a") not-in: #t)))
		 (if (null? remtests) ;; no more tests remaining
		     (let* ((dparts  (string-split lasttpath "/"))
			    (runpath (conc "/" (string-intersperse 
						(take dparts (- (length dparts) 1))
						"/"))))
		       (debug:print 1 *default-log-port* "Removing run: " runkey " " (db:get-value-by-header run header "runname") " and related record")
		       (rmt:delete-run run-id)

     runs)
    ;; (sqlite3:finalize! (db:delay-if-busy tdbdat))
    )
  #t)

(define (runs:remove-test-directory test mode) ;; remove-data-only)
  (let* ((run-dir       (db:test-get-rundir test))    ;; run dir is from the link tree
	 (real-dir      (if (common:file-exists? run-dir)
			    ;; (resolve-pathname run-dir)
			    (common:nice-path run-dir)
			    #f)))
    (case mode
      ((remove-data-only)(mt:test-set-state-status-by-id (db:test-get-run_id test)(db:test-get-id test) "CLEANING" "LOCKED" #f))
      ((remove-all)      (mt:test-set-state-status-by-id (db:test-get-run_id test)(db:test-get-id test) "REMOVING" "LOCKED" #f))
      ((archive-remove)  (mt:test-set-state-status-by-id (db:test-get-run_id test)(db:test-get-id test) "ARCHIVE_REMOVING" #f #f)))
    (debug:print-info 1 *default-log-port* "Attempting to remove " (if real-dir (conc " dir " real-dir " and ") "") " link " run-dir)
    (if (and real-dir 
	     (> (string-length real-dir) 5)
	     (common:file-exists? real-dir)) ;; bad heuristic but should prevent /tmp /home etc.
	(begin ;; let* ((realpath (resolve-pathname run-dir)))
	  (debug:print-info 1 *default-log-port* "Recursively removing " real-dir)
	  (if (common:file-exists? real-dir)
	      (runs:safe-delete-test-dir real-dir)
	      (debug:print 0 *default-log-port* "WARNING: test dir " real-dir " appears to not exist or is not readable")))
	(if real-dir 
	    (debug:print 0 *default-log-port* "WARNING: directory " real-dir " does not exist")
	    (debug:print 0 *default-log-port* "WARNING: no real directory corrosponding to link " run-dir ", nothing done")))
    (if (symbolic-link? run-dir)
	(begin

     
;; clean cache files
(define (runs:clean-cache target runname toppath)
  (if target
      (if runname
	  (let* ((linktree (common:get-linktree)) ;; (if toppath (configf:lookup *configdat* "setup" "linktree")))
		 (runtop   (conc linktree "/" target "/" runname))
		 (files    (if (common:file-exists? runtop)
			       (append (glob (conc runtop "/.megatest*"))
				       (glob (conc runtop "/.runconfig*")))
			       '())))
	    (if (null? files)
		(debug:print-info 0 *default-log-port* "No cached megatest or runconfigs files found. None removed.")
		(begin
		  (debug:print-info 0 *default-log-port* "Removing cached files:\n    " (string-intersperse files "\n    "))

(import (prefix base64 base64:))

(declare (unit sdb))

;; 
(define (sdb:open fname)
  (let* ((dbpath    (pathname-directory fname))
	 (dbexists  (let ((fe (common:file-exists? fname)))
		      (if fe 
			  fe
			  (begin
			    (create-directory dbpath #t)
			    #f))))
	 (sdb        (sqlite3:open-database fname))
	 (handler   (make-busy-timeout 136000)))

	 exn
	 (begin
	   (print-call-chain (current-error-port))
	   (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
	   (debug:print 5 *default-log-port* " exn=" (condition->list exn))
	   (debug:print 0 *default-log-port* "tasks:wait-on-journal failed. Continuing on, you can ignore this call-chain")
	   #t) ;; if stuff goes wrong just allow it to move on
	 (let loop ((journal-exists (common:file-exists? fullpath))
		    (count          n)) ;; wait ten times ...
	   (if journal-exists
	       (begin
		 (if (and waiting-msg
			  (eq? (modulo n 30) 0))
		     (debug:print 0 *default-log-port* waiting-msg))
		 (if (> count 0)
		     (begin
		       (thread-sleep! 1)
		       (loop (common:file-exists? fullpath)
			     (- count 1)))
		     (begin
		       (debug:print 0 *default-log-port* "ERROR: removing the journal file " fullpath ", this is not good. Look for disk full, write access and other issues.")
		       (if remove (system (conc "rm -rf " fullpath)))
		       #f)))
	       #t))))))

	   (begin
	     (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      (make-busy-timeout 36000)))

         (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"))))
    
 
;;======================================================================

;;
(define (open-test-db work-area) 
  (debug:print-info 11 *default-log-port* "open-test-db " work-area)
  (if (and work-area 
	   (directory? work-area)
	   (file-read-access? work-area))
      (let* ((dbpath              (conc work-area "/testdat.db"))
	     (dbexists            (common:file-exists? dbpath))
	     (work-area-writeable (file-write-access? work-area))
	     (db                  (handle-exceptions  ;; open the db if area writeable or db pre-existing. open in-mem otherwise. if exception, open in-mem
				   exn
				   (begin
				     (print-call-chain (current-error-port))
				     (debug:print 2 *default-log-port* "ERROR: problem accessing test db " work-area ", you probably should clean and re-run this test"
						  ((condition-property-accessor 'exn 'message) exn))

	    (append paths (list (conc *toppath* "/tests"))))))

(define (tests:get-valid-tests test-registry tests-paths)
  (if (null? tests-paths) 
      test-registry
      (let loop ((hed (car tests-paths))
		 (tal (cdr tests-paths)))
	(if (common:file-exists? hed)
	    (for-each (lambda (test-path)
			(let* ((tname   (last (string-split test-path "/")))
			       (tconfig (conc test-path "/testconfig")))
			  (if (and (not (hash-table-ref/default test-registry tname #f))
				   (common:file-exists? tconfig))
			      (hash-table-set! test-registry tname test-path))))
		      (glob (conc hed "/*"))))
	(if (null? tal)
	    test-registry
	    (loop (car tal)(cdr tal))))))

(define (tests:filter-test-names test-names test-patts)

	  (db:test-get-rundir testdat)) ;; )
	 (prev-rundir ;; (sdb:qry 'passstr 
	  (db:test-get-rundir prev-testdat)) ;; )
	 (waivers     (if testconfig (configf:section-vars testconfig "waivers") '()))
	 (waiver-rx   (regexp "^(\\S+)\\s+(.*)$"))
	 (diff-rule   "diff %file1% %file2%")
	 (logpro-rule "diff %file1% %file2% | logpro %waivername%.logpro %waivername%.html"))
    (if (not (common:file-exists? test-rundir))
	(begin
	  (debug:print-error 0 *default-log-port* "test run directory is gone, cannot propagate waiver")
	  #f)
	(begin
	  (push-directory test-rundir)
	  (let ((result (if (null? waivers)
			    #f
			    (let loop ((hed (car waivers))
				       (tal (cdr waivers)))
			      (debug:print 0 *default-log-port* "INFO: Applying waiver rule \"" hed "\"")
			      (let* ((waiver      (configf:lookup testconfig "waivers" hed))
				     (wparts      (if waiver (string-match waiver-rx waiver) #f))
				     (waiver-rule (if wparts (cadr wparts)  #f))
				     (waiver-glob (if wparts (caddr wparts) #f))
				     (logpro-file (if waiver
						      (let ((fname (conc hed ".logpro")))
							(if (common:file-exists? fname)
							    fname 
							    (begin
							      (debug:print 0 *default-log-port* "INFO: No logpro file " fname " falling back to diff")
							      #f)))
						      #f))
				     ;; if rule by name of waiver-rule is found in testconfig - use it
				     ;; else if waivername.logpro exists use logpro-rule

				  "Runs"
				  (common:htree->html runs-htree
						      '()
						      (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)

                                                (full-name  (db:test-make-full-name test-name item-path))
                                                (path-parts (string-split full-name)))
                                           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

                                                                   (let* ((targ-path (string-intersperse p "/"))
                                                                          (test-name (car p))
                                                                          (item-path ;; (if (> (length p) 2) ;; test-name + run-name
                                                                           (string-intersperse p "/"))
                                                                          (full-targ (conc html-dir "/" targ-path))
                                                                          (std-file  (conc full-targ "/test-summary.html"))
                                                                          (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)
                                                                           (conc "No summary for " run-name)))))
                                                                 ))))))
                     (close-output-port oup)))))
           runs)

			      
;;======================================================================
;; Gather data from test/task specifications
;;======================================================================

;; (define (tests:get-valid-tests testsdir test-patts) ;;  #!key (test-names '()))
;;   (let ((tests (glob (conc testsdir "/tests/*")))) ;; " (string-translate patt "%" "*")))))
;;     (set! tests (filter (lambda (test)(common:file-exists? (conc test "/testconfig"))) tests))
;;     (delete-duplicates
;;      (filter (lambda (testname)
;; 	       (tests:match test-patts testname #f))
;; 	     (map (lambda (testp)
;; 		    (last (string-split testp "/")))
;; 		  tests)))))

;;
(define (tests:get-testconfig test-name item-path test-registry system-allowed #!key (force-create #f))
  (let* ((use-cache    (common:use-cache?))
	 (cache-path   (tests:get-test-path-from-environment))
	 (cache-file   (and cache-path (conc cache-path "/.testconfig")))
	 (cache-exists (and cache-file
			    (not force-create)  ;; if force-create then pretend there is no cache to read
			    (common:file-exists? cache-file)))
	 (cached-dat   (if (and (not force-create)
				cache-exists
				use-cache)
			   (handle-exceptions
			    exn
			    #f ;; any issues, just give up with the cached version and re-read
			    (configf:read-alist cache-file))

	      dat
	      ;; no cached data available
	      (let* ((treg         (or test-registry
				       (tests:get-all)))
		     (test-path    (or (hash-table-ref/default treg test-name #f)
				       (conc *toppath* "/tests/" test-name)))
		     (test-configf (conc test-path "/testconfig"))
		     (testexists   (and (common:file-exists? test-configf)(file-read-access? test-configf)))
		     (tcfg         (if testexists
				       (read-config test-configf #f system-allowed
						    environ-patt: (if system-allowed
								      "pre-launch-env-vars"
								      #f))
				       #f)))
		(if (and tcfg cache-file) (hash-table-set! tcfg "have fulldata" #t)) ;; mark this as fully read data

;; read data from tmp file or create if not exists
;; if exists regen in background
;;
(define (tests:lazy-dot testrecords  outtype sizex sizey)
  (let ((dfile (conc "/tmp/." (current-user-name) "-" (server:mk-signature) ".dot"))
	(fname (conc "/tmp/." (current-user-name) "-" (server:mk-signature) ".dotdat")))
    (tests:write-dot-file testrecords dfile sizex sizey)
    (if (common:file-exists? fname)
	(let ((res (with-input-from-file fname
		     (lambda ()
		       (read-lines)))))
	  (system (conc "env -i PATH=$PATH dot -T " outtype " < " dfile " > " fname "&"))
	  res)
	(begin
	  (system (conc "env -i PATH=$PATH dot -T " outtype " < " dfile " > " fname))