Megatest

;;
;; NB// Runs on the server as part of the server loop
;;
(define (api:process-request dbstruct $) ;; the $ is the request vars proc
  (set! *api-process-request-count* (+ *api-process-request-count* 1))
  (let* ((cmd     ($ 'cmd))
	 (paramsj ($ 'params))
	 (params  (db:string->obj paramsj transport: 'http)) ;; (rmt:json-str->dat area-dat paramsj))
	 (resdat  (api:execute-requests dbstruct (vector cmd params))) ;; #( flag result )
	 (res     (vector-ref resdat 1)))
    (if (> *api-process-request-count* *max-api-process-requests*)
	(set! *max-api-process-requests* *api-process-request-count*))
    (set! *api-process-request-count* (- *api-process-request-count* 1))
    ;; This can be here but needs controls to ensure it doesn't run more than every 4 seconds
    ;; (rmt:dat->json-str area-dat
    ;;  (if (or (string? res)
    ;;          (list?   res)
    ;;          (number? res)
    ;;          (boolean? res))
    ;;      res 
    ;;      (list "ERROR, not string, list, number or boolean" 1 cmd params res)))))
    (db:obj->string res transport: 'http)))

    (if section
	section
	'())))

;; look for the best candidate archive area, else create new 
;; area
;;
(define (archive:get-archive area-dat testname itempath dused)
  ;; look up in archive_allocations if there is a pre-used archive
  ;; with adequate diskspace
  ;;
  (let* ((existing-blocks (rmt:archive-get-allocations area-dat testname itempath dused))
	 (candidate-disks (map (lambda (block)
				 (list
				  (vector-ref block 1)   ;; archive-area-name
				  (vector-ref block 2))) ;; disk-path
			       existing-blocks)))
    (or (common:get-disk-with-most-free-space candidate-disks dused)
	(archive:allocate-new-archive-block #f #f #f)))) ;; BROKEN. testname itempath))))

;; allocate a new archive area
;;
(define (archive:allocate-new-archive-block area-dat run-area-home testsuite-name dneeded)
  (let* ((adisks    (archive:get-archive-disks))
	 (best-disk (common:get-disk-with-most-free-space adisks dneeded)))
    (if best-disk
	(let* ((bdisk-name    (car best-disk))
	       (bdisk-path    (cdr best-disk))
	       (area-key      (substring (message-digest-string (md5-primitive) run-area-home) 0 5))
	       (bdisk-id      (rmt:archive-register-disk area-dat bdisk-name bdisk-path (get-df bdisk-path)))
	       (archive-name  (let ((sec (current-seconds)))
				(conc (time->string (seconds->local-time sec) "%Y")
				      "_q" (seconds->quarter sec) "/"
				      testsuite-name "_" area-key)))
	       (archive-path  (conc bdisk-path "/" archive-name))
	       (block-id      (rmt:archive-register-block-name area-dat bdisk-id archive-path)))
	  ;;   (allocation-id (rmt:archive-allocate-testsuite/area-to-block area-dat block-id testsuite-name area-key)))
	  (if block-id ;; (and block-id allocation-id)
	      (cons block-id archive-path)
	      #f))
	#f)))

;; archive - run bup
;;
;; 1. create the bup dir if not exists
;; 2. start the du of each directory
;; 3. gen index
;; 4. save
;;
(define (archive:run-bup area-dat archive-command run-id run-name tests rp-mutex bup-mutex)
  ;; move the getting of archive space down into the below block so that a single run can 
  ;; allocate as needed should a disk fill up
  ;;
  (let* ((min-space    (string->number (or (configf:lookup *configdat* "archive" "minspace") "1000")))
	 (archive-info (archive:allocate-new-archive-block *toppath* (common:get-testsuite-name) min-space))
	 (archive-dir  (if archive-info (cdr archive-info) #f))
	 (archive-id   (if archive-info (car archive-info) -1))

    ;;
    (for-each
     (lambda (test-dat)
       (let* ((item-path         (db:test-get-item-path test-dat))
	      (test-name         (db:test-get-testname  test-dat))
	      (test-id           (db:test-get-id        test-dat))
	      (run-id            (db:test-get-run_id    test-dat))
	      (target            (string-intersperse (map cadr (rmt:get-key-val-pairs area-dat run-id)) "/"))
	      
	      (toplevel/children (and (db:test-get-is-toplevel test-dat)
				      (> (rmt:test-toplevel-num-items area-dat run-id test-name) 0)))
	      (test-partial-path (conc target "/" run-name "/" (db:test-make-full-name test-name item-path)))
	      ;; note the trailing slash to get the dir inspite of it being a link
	      (test-path         (conc linktree "/" test-partial-path))
	      (mutex-lock! rp-mutex)
	      (test-physical-path (if (file-exists? test-path) 
				      (common:real-path test-path)
				      #f))

	   test-path))))
     tests)
    ;; for each disk-group
    (for-each 
     (lambda (disk-group)
       (debug:print 0 *default-log-port* "Processing disk-group " disk-group)
       (let* ((test-paths (hash-table-ref disk-groups disk-group))
	      ;; ((string-intersperse (map cadr (rmt:get-key-val-pairs area-dat 1)) "-")
	      (bup-init-params  (list "-d" archive-dir "init"))
	      (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))

	 (debug:print-info 0 *default-log-port* "Archiving data with bup")
	 (run-n-wait bup-exe params: bup-save-params print-cmd: print-prefix)
	 ;; (mutex-unlock! bup-mutex)
	 (for-each
	  (lambda (test-dat)
	    (let ((test-id           (db:test-get-id        test-dat))
		  (run-id            (db:test-get-run_id    test-dat)))
	      (rmt:test-set-archive-block-id area-dat run-id test-id archive-id)
	      (if (member archive-command '("save-remove"))
		  (runs:remove-test-directory area-dat test-dat 'archive-remove))))
	  (hash-table-ref test-groups disk-group))))
     (hash-table-keys disk-groups))
    #t))

(define (archive:bup-restore area-dat archive-command run-id run-name tests rp-mutex bup-mutex)  ;; move the getting of archive space down into the below block so that a single run can 
  ;; allocate as needed should a disk fill up
  ;;
  (let* ((bup-exe      (or (configf:lookup *configdat* "archive" "bup") "bup"))
	 (linktree     (configf:lookup *configdat* "setup" "linktree")))

    ;; from the test info bin the path to the test by stem
    ;;
    (for-each
     (lambda (test-dat)
       ;; When restoring test-dat will initially contain an old and invalid path to the test
       (let* ((best-disk         (get-best-disk *configdat* #f)) ;; BUG: get the testconfig and use it here. Otherwise data pulled out of archive could end up on the wrong kind of disk.
	      (item-path         (db:test-get-item-path test-dat))
	      (test-name         (db:test-get-testname  test-dat))
	      (test-id           (db:test-get-id        test-dat))
	      (run-id            (db:test-get-run_id    test-dat))
	      (keyvals           (rmt:get-key-val-pairs area-dat run-id))
	      (target            (string-intersperse (map cadr keyvals) "/"))
	      
	      (toplevel/children (and (db:test-get-is-toplevel test-dat)
				      (> (rmt:test-toplevel-num-items area-dat run-id test-name) 0)))
	      (test-partial-path (conc target "/" run-name "/" (db:test-make-full-name test-name item-path)))
	      ;; note the trailing slash to get the dir inspite of it being a link
	      (test-path         (conc linktree "/" test-partial-path))
	      ;; if the old path was not deleted then prev-test-physical-path will end up pointing to a real directory
	      (mutex-lock! rp-mutex)
	      (prev-test-physical-path (if (file-exists? test-path)
					   ;; (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 area-dat 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
	 ;;

	       ;; 1. Get the block id from the test info
	       ;; 2. Get the block data given the block id
	       ;; 3. Construct the paths etc. for the following command:
	       ;; 
	       ;; bup -d /tmp/matt/adisk1/2015_q1/fullrun_e1a40/ restore -C /tmp/seeme fullrun-30/latest/ubuntu/nfs/none/w02.1.20.54_b/

	       ;; DO BUP RESTORE
	       (let* ((new-test-dat        (rmt:get-test-info-by-id area-dat run-id test-id))
		      (new-test-path       (if (vector? new-test-dat )
					       (db:test-get-rundir new-test-dat)
					       (begin
						 (debug:print-error 0 *default-log-port* "unable to get data for run-id=" run-id ", test-id=" test-id)
						 (exit 1))))
		      ;; new-test-path won't work - must use best-disk instead? Nope, new-test-path but tack on /..
		      (bup-restore-params  (list "-d" archive-path "restore" "-C" (conc new-test-path "/..") archive-internal-path)))

;;      *transport-type* and *runremote* from the monitor.db
;;
;; client:setup
;;
;; lookup_server, need to remove *runremote* stuff
;;

(define (client:setup-http area-dat areapath #!key (remaining-tries 100) (failed-connects 0))
  (debug:print-info 2 *default-log-port* "client:setup remaining-tries=" remaining-tries)
  (server:start-and-wait area-dat areapath)
  (if (<= remaining-tries 0)
      (begin
	(debug:print-error 0 *default-log-port* "failed to start or connect to server")
	(exit 1))
      ;;
      ;; Alternatively here, we can get the list of candidate servers and work our way
      ;; through them searching for a good one.
      ;;
      (let* ((server-dat (server:get-first-best areapath)))
	(if (not server-dat) ;; no server found
	    (client:setup-http areapath remaining-tries: (- remaining-tries 1))
	    (let ((host  (cadr  server-dat))
		  (port  (caddr server-dat)))
	      (debug:print-info 4 *default-log-port* "client:setup server-dat=" server-dat ", remaining-tries=" remaining-tries)
	      (if (not *runremote*)(set! *runremote* (make-remote)))
	      (if (and host port)
		  (let* ((start-res (case *transport-type*
				      ((http)(http-transport:client-connect host port))))
			 (ping-res  (case *transport-type* 
				      ((http)(rmt:login-no-auto-client-setup area-dat start-res)))))
		    (if (and start-res
			     ping-res)
			(begin
			  (remote-conndat-set! *runremote* start-res) ;; (hash-table-set! *runremote* run-id start-res)
			  (debug:print-info 2 *default-log-port* "connected to " (http-transport:server-dat-make-url start-res))
			  start-res)
			(begin    ;; login failed but have a server record, clean out the record and try again
			  (debug:print-info 0 *default-log-port* "client:setup, login failed, will attempt to start server ... start-res=" start-res ", run-id=" run-id ", server-dat=" server-dat)
			  (case *transport-type* 
			    ((http)(http-transport:close-connections)))
			  (remote-conndat-set! *runremote* #f)  ;; (hash-table-delete! *runremote* run-id)
			  (thread-sleep! 1)
			  (client:setup-http areapath remaining-tries: (- remaining-tries 1))
			  )))
		  (begin    ;; no server registered
		    (server:kind-run areapath)
		    (debug:print-info 0 *default-log-port* "client:setup, no server registered, remaining-tries=" remaining-tries)
		    (thread-sleep! 1) ;; (+ 5 (random (- 20 remaining-tries))))  ;; give server a little time to start up, randomize a little to avoid start storms.
		    (server:start-and-wait area-dat areapath)
		    (client:setup-http areapath remaining-tries: (- remaining-tries 1)))))))))

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

(defstruct remote
  (hh-dat            (common:get-homehost)) ;; homehost record ( addr . hhflag )
  (server-url        #f)
  (last-server-check 0)  ;; last time we checked to see if the server was alive
  (conndat           #f)
  (transport         *transport-type*)
  (server-timeout    (or (server:get-timeout) 100))) ;; default to 100 seconds

;; launching and hosts
(defstruct host

  (conc megatest-version "-" megatest-fossil-hash))

(define (common:version-signature)
  (conc megatest-version "-" (substring megatest-fossil-hash 0 4)))

;; from metadat lookup MEGATEST_VERSION
;;
(define (common:get-last-run-version area-dat) ;; RADT => How does this work in send-receive function??; assume it is the value saved in some DB
  (rmt:get-var area-dat "MEGATEST_VERSION"))

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

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

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

;; Move me elsewhere ...
;; RADT => Why do we meed the version check here, this is called only if version misma

		      (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)))))))

(define (std-exit-procedure area-dat)
  (on-exit (lambda () 0))
  ;;(BB> "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))
	(rmt:print-db-stats area-dat))
    (let ((th1 (make-thread (lambda () ;; thread for cleaning up, give it five seconds
                              (if *dbstruct-db* (db:close-all *dbstruct-db*)) ;; one second allocated
			      (if *task-db*    
				  (let ((db (cdr *task-db*)))
				    (if (sqlite3:database? db)
					(begin
					  (sqlite3:interrupt! db)

(define (common:args-get-state)
  (or (args:get-arg "-state")(args:get-arg ":state")))

(define (common:args-get-status)
  (or (args:get-arg "-status")(args:get-arg ":status")))

(define (common:args-get-testpatt area-dat rconf)
  (let* ((tagexpr (args:get-arg "-tagexpr"))
         (tags-testpatt (if tagexpr (string-join (runs:get-tests-matching-tags area-dat tagexpr) ",") #f))
         (testpatt-key  (if (args:get-arg "--modepatt") (args:get-arg "--modepatt") "TESTPATT"))
         (args-testpatt (or (args:get-arg "-testpatt") (args:get-arg "-runtests") "%"))
         (rtestpatt     (if rconf (runconfigs-get rconf testpatt-key) #f)))
    (cond
     (tags-testpatt
      (debug:print-info 0 *default-log-port* "-tagexpr "tagexpr" selects testpatt "tags-testpatt)
      tags-testpatt)

  (let ((res (system (conc "ping -c 1 " hostname " > /dev/null"))))
    (eq? res 0)))

;; ideally put all this info into the db, no need to preserve it across moving homehost
;;
;; return list of
;;  ( reachable? cpuload update-time )
(define (common:get-host-info hostname area-dat)
  (let* ((loadinfo (rmt:get-latest-host-load area-dat hostname))
         (load (car loadinfo))
         (load-sample-time (cdr loadinfo))
         (load-sample-age (- (current-seconds) load-sample-time))
         (loadinfo-timeout-seconds 20)
         (host-last-update-timeout-seconds 10)
         (host-rec (hash-table-ref/default *host-loads* hostname #f))
         )

			   (test-meta-panel-get-description testmeta)))
	    )))))


;;======================================================================
;; Run info panel
;;======================================================================
(define (run-info-panel db area-dat keydat testdat runname)
  (let* ((run-id     (db:test-get-run_id testdat))
	 (rundat     (rmt:get-run-info area-dat run-id))
	 (header     (db:get-header rundat))
	 (event_time (db:get-value-by-header (db:get-rows rundat)
					     (db:get-header rundat)
					     "event_time")))
    (iup:frame 
     #:title "Megatest Run Info" ; #:expand "YES"
     (iup:hbox ; #:expand "YES"

(define *dashboard-test-db* #t)
(define *dashboard-comment-share-slot* #f)

;;======================================================================
;; Set fields 
;;======================================================================
(define (set-fields-panel area-dat run-id test-id testdat #!key (db #f))
  (let ((newcomment #f)
	(newstatus  #f)
	(newstate   #f)
	(wtxtbox    #f))
    (iup:frame
     #:title "Set fields"
     (iup:vbox
      (iup:hbox (iup:label "Comment:")
		(let ((txtbox (iup:textbox #:action (lambda (val a b)
						      ;; (rmt:test-set-state-status-by-id area-dat run-id test-id #f #f b)
						      (rmt:test-set-state-status area-dat run-id test-id #f #f b)
						      ;; IDEA: Just set a variable with the proc to call?
						      ;; (rmt:test-set-state-status-by-id area-dat run-id test-id #f #f b)
						      (set! newcomment b))
					   #:value (db:test-get-comment testdat)
					   #:expand "HORIZONTAL")))
		  (set! wtxtbox txtbox)
		  txtbox))
		  
      (apply iup:hbox
	     (iup:label "STATE:" #:size "30x")
	     (let* ((btns  (map (lambda (state)
				  (let ((btn (iup:button state
							 #:expand "HORIZONTAL" #:size "50x" #:font "Courier New, -10"
							 #:action (lambda (x)
								    ;; (rmt:test-set-state-status-by-id area-dat run-id test-id state #f #f)
								    (rmt:set-state-status-and-roll-up-items area-dat run-id test-id #f state #f #f) ;; test-name passed in as test-id is respected
								    (db:test-set-state! testdat state)))))
				    btn))
				(map cadr *common:std-states*)))) ;; (list "COMPLETED" "NOT_STARTED" "RUNNING" "REMOTEHOSTSTART" "LAUNCHED" "KILLED" "KILLREQ"))))
	       (vector-set! *state-status* 0
			    (lambda (state color)
			      (for-each 
			       (lambda (btn)

													      (if wtxtbox 
														  (begin
														    (iup:attribute-set! wtxtbox "VALUE" c)
														    (if (not *dashboard-comment-share-slot*)
															(set! *dashboard-comment-share-slot* wtxtbox)))
														  ))))
									  (begin
									    ;; (rmt:test-set-state-status-by-id area-dat run-id test-id #f status #f)
									    (rmt:set-state-status-and-roll-up-items area-dat run-id test-id #f #f status #f) ;; test-name passed in as test-id is respected
									    (db:test-set-status! testdat status))))))))
				    btn))
				(map cadr *common:std-statuses*)))) ;; (list  "PASS" "WARN" "FAIL" "CHECK" "n/a" "WAIVED" "SKIP"))))
	       (vector-set! *state-status* 1
			    (lambda (status color)
			      (for-each 
			       (lambda (btn)

					 (conc "ezstep run from step " stepname)))))
    ;; (iup:button "Refresh test data"
    ;;     	#:expand "HORIZONTAL"
    ;;     	#:action (lambda (obj)
    ;;     		   (print "Refresh test data " stepname))
    )))

(define (dashboard-tests:waiver area-dat run-id testdat ovrdval cmtcmd)
  (let* ((wpatt (configf:lookup *configdat* "setup" "waivercommentpatt"))
	 (wregx (if (string? wpatt)(regexp wpatt) #f))
	 (wmesg (iup:label (if wpatt (conc "Comment must match pattern " wpatt) "")))
	 (comnt (iup:textbox #:action (lambda (val a b)
					(if wpatt
					    (if (string-match wregx b)
						(iup:attribute-set! wmesg "TITLE" (conc "Comment matches " wpatt))

			      #:expand "HORIZONTAL"
			      #: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 area-dat run-id test-id #f "WAIVED" comment)
						 (rmt:test-set-state-status-by area-dat 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)))))))
    dlog))


;;======================================================================
;;
;;======================================================================
(define (dashboard-tests:examine-test area-dat run-id test-id) ;; run-id run-key origtest)
  (let* ((db-path       (db:dbfile-path)) ;; (conc (configf:lookup *configdat* "setup" "linktree") "/db/" run-id ".db"))
	 (dbstruct      #f) ;; NOT ACTUALLY USED (db:setup)) ;; (make-dbr:dbstruct path:  (db:dbfile-path #f) ;; (configf:lookup *configdat* "setup" "linktree") 
			    ;;		   local: #t))
	 (testdat        (rmt:get-test-info-by-id area-dat run-id test-id)) ;; (db:get-test-info-by-id dbstruct run-id test-id))
	 (db-mod-time   0) ;; (file-modification-time db-path))
	 (last-update   0) ;; (current-seconds))
	 (request-update #t))
    (if (not testdat)
	(begin
	  (debug:print 2 *default-log-port* "ERROR: No test data found for test " test-id ", exiting")
	  (exit 1))
	(let* (;; (run-id        (if testdat (db:test-get-run_id testdat) #f))
	       (test-registry (tests:get-all))
	       (keydat        (if testdat (rmt:get-key-val-pairs area-dat run-id) #f))
	       (rundat        (if testdat (rmt:get-run-info area-dat run-id) #f))
	       (runname       (if testdat (db:get-value-by-header (db:get-rows rundat)
								  (db:get-header rundat)
								  "runname") #f))

	       ;; These next two are intentional bad values to ensure errors if they should not
	       ;; get filled in properly.
	       (logfile       "/this/dir/better/not/exist")
	       (rundir        (if testdat 
				  (db:test-get-rundir testdat)
				  logfile))
	       ;; (testdat-path  (conc rundir "/testdat.db")) ;; this gets recalculated until found 
	       (teststeps     (if testdat (tests:get-compressed-steps area-dat run-id test-id) '()))
	       (testfullname  (if testdat (db:test-get-fullname testdat) "Gathering data ..."))
	       (testname      (if testdat (db:test-get-testname testdat) "n/a"))
	       ;; (tests:get-testconfig testdat testname 'return-procs))
	       (testmeta      (if testdat 
				  (let ((tm (rmt:testmeta-get-record area-dat testname)))
				    (if tm tm (make-db:testmeta)))
				  (make-db:testmeta)))

	       (keystring  (string-intersperse 
			    (map (lambda (keyval)
				   ;; (conc ":" (car keyval) " " (cadr keyval)))
				   (cadr keyval))

	 			 (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 area-dat run-id inkeyvals: keydat inrunname: runname intarget: keystring testname: testname itempath: item-path) ;; these may be needed by the launching process
				(handle-exceptions
				 exn
				 (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) test-registry #t))))
	       (viewlog    (lambda (x)
			     (if (file-exists? logfile)
					;(system (conc "firefox " logfile "&"))

						       (> (current-milliseconds)(+ last-update 10000))     ;; force update even 10 seconds
						       request-update))
				    (newtestdat (if need-update 
						    ;; NOTE: BUG HIDER, try to eliminate this exception handler
						    (handle-exceptions
						     exn 
						     (debug:print-info 0 *default-log-port* "test db access issue in examine test for run-id " run-id ", test-id " test-id ": " ((condition-property-accessor 'exn 'message) exn))
						     (rmt:get-test-info-by-id area-dat run-id test-id )))))
			       ;; (print "INFO: need-update= " need-update " curr-mod-time = " curr-mod-time)
			       (cond
				((and need-update newtestdat)
				 (set! testdat newtestdat)
				 (set! teststeps    (tests:get-compressed-steps area-dat run-id test-id))
				 (set! logfile      (conc (db:test-get-rundir testdat) "/" (db:test-get-final_logf testdat)))
				 (set! rundir       ;; (filedb:get-path *fdb* 
				       (db:test-get-rundir testdat)) ;; )
				 (set! testfullname (db:test-get-fullname testdat))
				 ;; (debug:print 0 *default-log-port* "INFO: teststeps=" (intersperse teststeps "\n    "))
				 
				 ;; I don't see why this was implemented this way. Please comment it ...

											      (db:test-data-get-value    x)
											      (db:test-data-get-expected x)
											      (db:test-data-get-tol      x)
											      (db:test-data-get-status   x)
											      (db:test-data-get-units    x)
											      (db:test-data-get-type     x)
											      (db:test-data-get-comment  x)))
										    (rmt:read-test-data area-dat run-id test-id "%")))
									      "\n")))
							       (if (not (equal? currval newval))
								   (iup:attribute-set! test-data "VALUE" newval ))))) ;; "TITLE" newval)))))
					  test-data))
				       ;;(dashboard:run-controls)
				       )))
				 (iup:attribute-set! tabs "TABTITLE0" "Steps")

  please-update  
  tabdats
  update-mutex
  updaters 
  updating
  uidat ;; needs to move to tabdat at some time
  hide-not-hide-tabs
  (area-dat (make-remote))
  )

(define (dboard:commondat-make)
  (make-dboard:commondat
   curr-tab-num:         0
   tabdats:              (make-hash-table)
   please-update:        #t

(define (dboard:tabdat-make-data)
  (let ((dat (make-dboard:tabdat)))
    (dboard:setup-tabdat dat)
    (dboard:setup-num-rows dat)
    dat))

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

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

;; RADT => Matrix defstruct addition
(defstruct dboard:graph-dat
    ((id           #f) : string)
    ((color        #f) : vector)
    ((flag         #t) : boolean)

;; 	newdat)))

;; this calls dboard:get-tests-for-run-duplicate for each run
;;
;; create a virtual table of all the tests
;; keypatts: ( (KEY1 "abc%def")(KEY2 "%") )
;;
(define (update-rundat tabdat area-dat runnamepatt numruns testnamepatt keypatts)
  (let* ((access-mode      (dboard:tabdat-access-mode tabdat))
         (keys             ;; (db:dispatch-query access-mode rmt:get-keys db:get-keys))
	  (rmt:get-keys area-dat))
	 (last-runs-update (- (dboard:tabdat-last-runs-update tabdat) 2))
         (allruns          (rmt:get-runs area-dat runnamepatt numruns (dboard:tabdat-start-run-offset tabdat) keypatts))
	  ;; (db:dispatch-query access-mode rmt:get-runs db:get-runs
	  ;;     runnamepatt numruns (dboard:tabdat-start-run-offset tabdat) keypatts))
         (allruns-tree    (rmt:get-runs-by-patt area-dat keys "%" #f #f #f #f))
         ;; (allruns-tree    (db:dispatch-query access-mode rmt:get-runs-by-patt db:get-runs-by-patt
         ;;                                     keys "%" #f #f #f #f last-runs-update));;'("id" "runname")
	 (header      (db:get-header allruns))
	 (runs        (db:get-rows   allruns)) ;; RA => Filtered as per runpatt selected
         (runs-tree   (db:get-rows   allruns-tree)) ;; RA => Returns complete list of runs
	 (start-time  (current-seconds))
	 (runs-hash   (let ((ht (make-hash-table)))
			 (for-each (lambda (run)
				     (hash-table-set! ht (db:get-value-by-header run header "id") run))

	(let loop ((run      (car runs))
		   (tal      (cdr runs))
		   (res     '())
		   (maxtests 0))
	  (let* ((run-id       (db:get-value-by-header run header "id"))
		 (run-struct   (hash-table-ref/default (dboard:tabdat-allruns-by-id tabdat) run-id #f))
		 (last-update  (if run-struct (dboard:rundat-last-update run-struct) 0))
		 (key-vals     (rmt:get-key-vals area-dat run-id))
		 (tests-ht     (dboard:get-tests-for-run-duplicate tabdat run-id run testnamepatt key-vals))
		 ;; GET RID OF dboard:get-tests-dat - it is superceded by dboard:get-tests-for-run-duplicate
		 ;;  dboard:get-tests-for-run-duplicate - returns a hash table
		 ;;  (dboard:get-tests-dat tabdat run-id last-update))
		 (all-test-ids (hash-table-keys tests-ht))
		 (num-tests    (length all-test-ids)))
	    ;; (print "run-struct: " run-struct)

    (dboard:update-tree tabdat runs-hash header tb)))

;; this calls dboard:get-tests-for-run-duplicate for each run
;;
;; create a virtual table of all the tests
;; keypatts: ( (KEY1 "abc%def")(KEY2 "%") )
;;
(define (dboard:update-rundat tabdat area-dat runnamepatt numruns testnamepatt keypatts)
  (let* ((access-mode      (dboard:tabdat-access-mode tabdat))
         (keys             (dboard:tabdat-keys tabdat)) ;; (db:dispatch-query access-mode rmt:get-keys db:get-keys)))
	 (last-runs-update (- (dboard:tabdat-last-runs-update tabdat) 2))
         (allruns          (db:dispatch-query access-mode rmt:get-runs db:get-runs
                                              runnamepatt numruns (dboard:tabdat-start-run-offset tabdat) keypatts))
         (allruns-tree    (rmt:get-runs-by-patt area-dat keys "%" #f #f #f #f))
         ;; (allruns-tree    (db:dispatch-query access-mode rmt:get-runs-by-patt db:get-runs-by-patt
         ;;                                     keys "%" #f #f #f #f 0)) ;; last-runs-update));;'("id" "runname")
	 (header      (db:get-header allruns))
	 (runs        (db:get-rows   allruns)) ;; RA => Filtered as per runpatt selected
         (runs-tree   (db:get-rows   allruns-tree)) ;; RA => Returns complete list of runs
	 (start-time  (current-seconds))
	 (runs-hash   (let ((ht (make-hash-table)))
			 (for-each (lambda (run)
				     (hash-table-set! ht (db:get-value-by-header run header "id") run))

	(let loop ((run      (car runs))
		   (tal      (cdr runs))
		   (res     '())
		   (maxtests 0))
	  (let* ((run-id       (db:get-value-by-header run header "id"))
		 (run-struct   (hash-table-ref/default (dboard:tabdat-allruns-by-id tabdat) run-id #f))
		 ;; (last-update  (if run-struct (dboard:rundat-last-update run-struct) 0))
		 (key-vals     (rmt:get-key-vals area-dat run-id))
		 (tests-ht     (dboard:get-tests-for-run-duplicate tabdat run-id run testnamepatt key-vals))
		 ;; GET RID OF dboard:get-tests-dat - it is superceded by dboard:get-tests-for-run-duplicate
		 ;;  dboard:get-tests-for-run-duplicate - returns a hash table
		 ;;  (dboard:get-tests-dat tabdat run-id last-update))
		 (all-test-ids (hash-table-keys tests-ht))
		 (num-tests    (length all-test-ids)))
	    ;; (print "run-struct: " run-struct)

	      (let ((newval (car values)))
		(iup:attribute-set! lb "VALUE" newval)
		newval))))))

(define (dashboard:update-target-selector tabdat #!key (action-proc #f))
  (let* ((runconf-targs (common:get-runconfig-targets))
	 (key-lbs       (dboard:tabdat-key-listboxes tabdat))
	 (db-target-dat (rmt:get-targets area-dat))
	 (header        (vector-ref db-target-dat 0))
	 (db-targets    (vector-ref db-target-dat 1))
	 (munge-target  (lambda (x)            ;; create a target vector from a string. Pad with na if needed.
			  (list->vector
			   (take (append (string-split x "/")
					 (make-list (length header) "na"))
				 (length header)))))

    (dashboard:update-run-command tabdat)))

;; used by run-controls
;;
(define (dashboard:update-tree-selector tabdat #!key (action-proc #f))
  (let* ((tb            (dboard:tabdat-runs-tree tabdat))
	 (runconf-targs (common:get-runconfig-targets))
	 (db-target-dat (rmt:get-targets area-dat))
         (runs-tree-ht  (dboard:tabdat-runs-tree-ht tabdat))
	 (header        (vector-ref db-target-dat 0))
	 (db-targets    (vector-ref db-target-dat 1))
	 (munge-target  (lambda (x)            ;; create a target vector from a string. Pad with na if needed.
			  (take (append (string-split x "/")
					(make-list (length header) "na"))
				(length header))))

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

(define (dboard:update-tree tabdat area-dat runs-hash runs-header tb)
  (let* ((access-mode   (dboard:tabdat-access-mode tabdat))
         (run-ids (sort (filter number? (hash-table-keys runs-hash))
			(lambda (a b)
			  (let* ((record-a (hash-table-ref runs-hash a))
				 (record-b (hash-table-ref runs-hash b))
				 (time-a   (db:get-value-by-header record-a runs-header "event_time"))
				 (time-b   (db:get-value-by-header record-b runs-header "event_time")))
			    (< time-a time-b)))))
         (changed      #f)
	 (last-runs-update  (dboard:tabdat-last-runs-update tabdat))
	 (runs-dat          (rmt:get-runs-by-patt area-dat (dboard:tabdat-keys tabdat) "%" #f #f #f #f last-runs-update)))
	  ;; (db:dispatch-query access-mode rmt:get-runs-by-patt db:get-runs-by-patt
          ;;                                (dboard:tabdat-keys tabdat) "%" #f #f #f #f last-runs-update)))
    (dboard:tabdat-last-runs-update-set! tabdat (- (current-seconds) 2))
    (for-each (lambda (run-id)
		(let* ((run-record (hash-table-ref/default runs-hash run-id #f))
		       (key-vals   (map (lambda (key)(db:get-value-by-header run-record runs-header key))
					(dboard:tabdat-keys tabdat)))
		       (run-name   (db:get-value-by-header run-record runs-header "runname"))
		       (col-name   (conc (string-intersperse key-vals "\n") "\n" run-name))

         ((> 0 (string-compare3 a-test-name b-test-name)) #f)
         ((< 0 (string-compare3 a-item-path b-item-path)) #t)
         (else #f)))))))


(define (dashboard:run-id->tests-mindat run-id tabdat runs-hash)
  (let* ((run          (hash-table-ref/default runs-hash run-id #f))
         (key-vals     (rmt:get-key-vals area-dat run-id))
         (testnamepatt (or (dboard:tabdat-test-patts tabdat) "%/%"))
         (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))

        (dcommon:xor-tests-mindat 
         (dashboard:run-id->tests-mindat src-run-id tabdat runs-hash)
         (dashboard:run-id->tests-mindat dest-run-id tabdat runs-hash)
         hide-clean: hide-clean)
        #f)))


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

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

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

               (nonsel-color "170 170 170")
               (current-mode (dboard:tabdat-runs-summary-mode tabdat)))
          (if (eq? this-mode current-mode)
              (iup:attribute-set! this-button "BGCOLOR" sel-color)
              (iup:attribute-set! this-button "BGCOLOR" nonsel-color))
          (loop (cdr buttons-left) (cdr modes-left))))))

(define (dboard:runs-summary-xor-labels-updater tabdat area-dat)
  (let ((source-runname-label (dboard:tabdat-runs-summary-source-runname-label tabdat))
        (dest-runname-label (dboard:tabdat-runs-summary-dest-runname-label tabdat))
        (mode (dboard:tabdat-runs-summary-mode tabdat)))
    (when (and source-runname-label dest-runname-label)
      (case mode
        ((xor-two-runs xor-two-runs-hide-clean)
         (let* ((curr-run-id          (dboard:tabdat-curr-run-id tabdat))
                (prev-run-id          (dboard:tabdat-prev-run-id tabdat))
                (curr-runname (if curr-run-id
                                  (rmt:get-run-name-from-id area-dat curr-run-id)
                                  "None"))
                (prev-runname (if prev-run-id
                                  (rmt:get-run-name-from-id area-dat prev-run-id)
                                  "None")))
           (iup:attribute-set! source-runname-label "TITLE" (conc " SRC: "prev-runname"  "))
           (iup:attribute-set! dest-runname-label "TITLE" (conc "DEST: "curr-runname"  "))))
        (else
         (iup:attribute-set! source-runname-label "TITLE" "")
         (iup:attribute-set! dest-runname-label "TITLE" ""))))))

;;
;; display and manage a single run at a time

;; This is the Run Summary tab
;; 
(define (dashboard:runs-summary commondat tabdat #!key (tab-num #f))
  (let* ((update-mutex (dboard:commondat-update-mutex commondat))
	 (area-dat     (dboard:commondat-area-dat commondat))
	 (tb      (iup:treebox
		   #:value 0
		   #:name "Runs"
		   #:expand "YES"
		   #:addexpanded "NO"
		   #:selection-cb
		   (lambda (obj id state)

                           (BB> "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 area-dat run-id))
                                  (target   (rmt:get-target area-dat run-id))
                                  (runname  (db:get-value-by-header (db:get-rows run-info)
                                                                    (db:get-header run-info) "runname"))
                                  (test-name (db:test-get-testname (rmt:get-test-info-by-id area-dat run-id test-id)))
                                  (testpatt  (let ((tlast (rmt:tasks-get-last area-dat target runname)))
                                                (if tlast
                                                    (let ((tpatt (tasks:task-get-testpatt tlast)))
                                                      (if (member tpatt '("0" 0)) ;; known bad historical value - remove in 2017
                                                          "%"
                                                          tpatt))
                                                    "%")))
                                  (item-path (db:test-get-item-path (rmt:get-test-info-by-id area-dat 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 " &")))
                             (BB> "status-chars=["status-chars"] status=["status"]")
                             (cond

    (lambda (obj)
      (dcommon:examine-xterm run-id test-id)))

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

   

               " -runname " runname
	       " -testpatt " item-test-path
	       " -preclean -clean-cache"))))
     (iup:menu-item
      (conc "Kill " item-test-path)
      #:action
      (lambda (obj)
        ;; (rmt:test-set-state-status-by-id area-dat run-id test-id "KILLREQ" #f #f)
	(common:run-a-command
	 (conc "megatest -set-state-status KILLREQ,n/a -target " target
               " -runname " runname
	       " -testpatt " item-test-path 
	       " -state RUNNING,REMOTEHOSTSTART,LAUNCHED"))))
     (iup:menu-item
      (conc "Clean "item-test-path)

	 (lftlst          '())
	 (hdrlst          '())
	 (bdylst          '())
	 (result          '())
	 (i               0)
	 (btn-height      (dboard:tabdat-runs-btn-height runs-dat))
	 (btn-fontsz      (dboard:tabdat-runs-btn-fontsz runs-dat))
	 (cell-width      (dboard:tabdat-runs-cell-width runs-dat))
	 (area-dat        (dboard:commondat-area-dat commondat)))
    ;; controls (along bottom)
    ;; (set! controls (dboard:make-controls commondat runs-dat))
    
    ;; create the left most column for the run key names and the test names 
    (set! lftlst (list (iup:hbox
			(iup:label) ;; (iup:valuator)
			(apply iup:vbox 

			      ;; (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)))
					      (buttndat (hash-table-ref (dboard:tabdat-buttondat runs-dat) button-key))
					      (test-id  (db:test-get-id (vector-ref buttndat 3)))
					      (run-id   (db:test-get-run_id (vector-ref buttndat 3)))
					      (run-info (rmt:get-run-info area-dat run-id))
					      (target   (rmt:get-target area-dat run-id))
					      (runname  (db:get-value-by-header (db:get-rows run-info)
										(db:get-header run-info) "runname"))
					      (test-name (db:test-get-testname (rmt:get-test-info-by-id area-dat run-id test-id)))
					      (testpatt  (let ((tlast (rmt:tasks-get-last area-dat target runname)))
							   (if tlast
							       (let ((tpatt (tasks:task-get-testpatt tlast)))
								 (if (member tpatt '("0" 0)) ;; known bad historical value - remove in 2017
								     "%"
								     tpatt))
							       "%")))
                                              (item-path (db:test-get-item-path (rmt:get-test-info-by-id area-dat run-id test-id)))
                                              (item-test-path (conc test-name "/" (if (equal? item-path "")
									"%" 
									item-path))))
					 (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")

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

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

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

  (let* ((keyvals (db:get-key-vals dbstruct run-id))
	 (thekey  (string-intersperse (map (lambda (x)(if x x "-na-")) keyvals) "/")))
    thekey))

;; Get run-ids for runs with same target but different runnames and NOT run-id
;;
(define (db:get-prev-run-ids dbstruct run-id)
  (let* ((keyvals (db:get-key-val-pairs dbstruct run-id))
	 (kvalues (map cadr keyvals))
	 (keys    (db:get-keys dbstruct))
	 (qrystr  (string-intersperse (map (lambda (x)(conc x "=?")) keys) " AND ")))
    (let ((prev-run-ids '()))
      (if (null? keyvals)
          '()
          (begin
            (db:with-db dbstruct #f #f ;; #f means work with the zeroth db - i.e. the runs db
                        (lambda (db)

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

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

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

	 (all-test-changes (let ((res (make-hash-table)))
			     (for-each (lambda (run-id)
					 (if (> run-id 0)
					     (hash-table-set! res run-id (synchash:client-get area-dat 'db:get-tests-for-run-mindata get-tests-sig 0 data run-id 1 testpatt states statuses #f))))
				       run-ids)
			     res))
	 (runs-hash    (hash-table-ref/default data get-runs-sig #f))
	 (header       (hash-table-ref/default runs-hash "header" #f))
	 (run-ids      (sort (filter number? (hash-table-keys runs-hash))
			     (lambda (a b)
			       (let* ((record-a (hash-table-ref runs-hash a))

             (not
              (equal?
               "CLEAN"
               (list-ref (list-ref item 2) 1))))
           res)
          res))))

(define (dcommon:examine-xterm area-dat run-id test-id)
  (let* ((testdat (rmt:get-test-info-by-id area-dat run-id test-id)))
    (if (not testdat)
        (begin
          (debug:print 2 "ERROR: No test data found for test " test-id ", exiting")
          (exit 1))
        (let*
            ((rundir        (if testdat 
                                (db:test-get-rundir testdat)

    ;; (iup:attribute-set! general-matrix "2:1" *toppath*)
    ;; Megatest version
    (iup:attribute-set! general-matrix "2:0" "Version")
    (iup:attribute-set! general-matrix "2:1" (conc megatest-version "-" (substring megatest-fossil-hash 0 4)))

    general-matrix))

(define (dcommon:run-stats area-dat commondat tabdat #!key (tab-num #f))
  (let* ((stats-matrix (iup:matrix expand: "YES"))
	 (changed      #f)
	 (stats-updater (lambda ()
			 (if (dashboard:database-changed? commondat tabdat context-key: 'run-stats)
			     (let* ((run-stats    (rmt:get-run-stats area-dat))
				    (indices      (common:sparse-list-generate-index run-stats)) ;;  proc: set-cell))
				    (row-indices  (car indices))
				    (col-indices  (cadr indices))
				    (max-row      (if (null? row-indices) 1 (common:max (map cadr row-indices))))
				    (max-col      (if (null? col-indices) 1 
						      (common:max (map cadr col-indices))))
				    (max-visible  (max (- (dboard:tabdat-num-tests tabdat) 15) 3))

					       (iup:attribute-set! tb "VALUE" val)
					       (dboard:tabdat-run-name-set! tabdat val)
					       (dashboard:update-run-command tabdat))))
				       "command-runname-selector lb action"))))
	  (refresh-runs-list (lambda ()
			       (if (dashboard:database-changed? commondat tabdat context-key: 'runname-selector-runs-list)
				   (let* (;; (target        (dboard:tabdat-target-string tabdat))
					  (runs-for-targ (rmt:get-runs-by-patt area-dat (dboard:tabdat-keys tabdat) "%" #f #f #f #f 0))
					  (runs-header   (vector-ref runs-for-targ 0))
					  (runs-dat      (vector-ref runs-for-targ 1))
					  (run-names     (cons default-run-name 
							       (map (lambda (x)
								      (db:get-value-by-header x runs-header "runname"))
								    runs-dat))))
				     ;; (print "DEBUGINFO: run-names=" run-names)

             (not
              (equal?
               "CLEAN"
               (list-ref (list-ref item 2) 1))))
           res)
          res))))

(define (diff:run-name->run-id area-dat run-name)
  (if (number? run-name)
      run-name
      (let* ((qry-res (rmt:get-runs area-dat run-name 1 0 '())))
        (if (eq? 2 (vector-length qry-res))
            (vector-ref (car (vector-ref qry-res 1)) 1)
            #f))))

(define (diff:target+run-name->run-id area-dat target run-name)
  (let* ((keys (rmt:get-keys area-dat))
         (target-parts (if target (string-split target "/") (map (lambda (x) "%") keys))))
    (if (not (eq? (length keys) (length keys)))
        (begin
          (print "Error: Target ("target") item count does not match fields count target tokens="target-parts" fields="keys)
          #f)
        (let* ((target-map (zip keys target-parts))
               (qry-res (rmt:get-runs area-dat run-name 1 0 target-map)))

          (if (eq? 2 (vector-length qry-res))
              (let ((first-ent (vector-ref qry-res 1)))
                (if (> (length first-ent) 0)
                    (vector-ref (car first-ent) 1)
                    #f))
              #f)))))

(define (diff:run-id->tests-mindat area-dat run-id #!key (testpatt "%/%"))
  (let* ((states '())
         (statuses '())
         (offset #f)
         (limit #f)
         (not-in #t)
         (sort-by #f)
         (sort-order #f)

       (let* ((id        (vector-ref row 0))
              (test-name  (vector-ref row 1))
              (item-path (vector-ref row 2))
              (state     (vector-ref row 3))
              (status    (vector-ref row 4)))
             (list test-name item-path (list id state status))))
     
     (rmt:get-tests-for-run area-dat run-id
                            testpatt states statuses
                            offset limit
                            not-in sort-by sort-order
                            qryvals
                            last-update
                            mode))))

| |\\/| |/ _ \\/ _` |/ _` | __/ _ \\/ __| __| | | | | | |_| |_
| |  | |  __/ (_| | (_| | ||  __/\\__ \\ |_  | |_| | |  _|  _|
|_|  |_|\\___|\\__, |\\__,_|\\__\\___||___/\\__| |____/|_|_| |_|
             |___/
</pre>")


(define (diff:run-id->target+run-name+starttime area-dat run-id)
  (let* ((target      (rmt:get-target area-dat run-id))
         (runinfo     (rmt:get-run-info area-dat run-id)) ; vector of header (list) and result (vector)
         (info-hash   (alist->hash-table
                       (map (lambda (x) (cons (car x) (cadr x)))  ; make it a useful hash
                            (zip (vector-ref runinfo 0) (vector->list (vector-ref runinfo 1))))))
         (run-name    (hash-table-ref/default info-hash "runname" "N/A"))
         (start-time  (hash-table-ref/default info-hash "event_time" 0)))
    (list target run-name start-time)))

;; (declare (uses filedb))

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

(define (ezsteps:run-from area-dat testdat start-step-name run-one)
  (let* ((test-run-dir  ;; (filedb:get-path *fdb* 
	  (db:test-get-rundir testdat)) ;; )
	 (testconfig    (read-config (conc test-run-dir "/testconfig") #f #t environ-patt: "pre-launch-env-vars"))
	 (ezstepslst    (hash-table-ref/default testconfig "ezsteps" '()))
	 (run-mutex     (make-mutex))
	 (rollup-status 0)
	 (exit-info     (vector #t #t #t))

		  
		  (if (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! area-dat run-id test-id stepname "start" "-" #f #f)
		  ;; now launch
		  (let ((pid (process-run script)))
		    (let processloop ((i 0))
		      (let-values (((pid-val exit-status exit-code)(process-wait pid #t)))
				  (mutex-lock! run-mutex)
				  (vector-set! exit-info 0 pid)
				  (vector-set! exit-info 1 exit-status)
				  (vector-set! exit-info 2 exit-code)
				  (mutex-unlock! run-mutex)
				  (if (eq? pid-val 0)
				      (begin
					(thread-sleep! 1)
					(processloop (+ i 1))))
				  ))
		    (let ((exinfo (vector-ref exit-info 2))
			  (logfna (if logpro-used (conc stepname ".html") "")))
		      (rmt:teststep-set-status! area-dat run-id test-id stepname "end" exinfo #f logfna))
		    (if logpro-used
			(rmt:test-set-log! area-dat test-id (conc stepname ".html")))
		    ;; set the test final status
		    (let* ((this-step-status (cond
					      ((and (eq? (vector-ref exit-info 2) 2) logpro-used) 'warn)
					      ((eq? (vector-ref exit-info 2) 0)                   'pass)
					      (else 'fail)))
			   (overall-status   (cond
					      ((eq? rollup-status 2) 'warn)

		      (debug:print 4 *default-log-port* "Exit value received: " (vector-ref exit-info 2) " logpro-used: " logpro-used 
				   " this-step-status: " this-step-status " overall-status: " overall-status 
				   " next-status: " next-status " rollup-status: " rollup-status)
		      (case next-status
			((warn)
			 (set! rollup-status 2)
			 ;; NB// test-set-status! does rdb calls under the hood
			 (tests:test-set-status! area-dat test-id "RUNNING" "WARN" 
						 (if (eq? this-step-status 'warn) "Logpro warning found" #f)
						 #f))
			((pass)
			 (tests:test-set-status! area-dat test-id "RUNNING" "PASS" #f #f))
			(else ;; 'fail
			 (set! rollup-status 1) ;; force fail
			 (tests:test-set-status! area-dat test-id "RUNNING" "FAIL" (conc "Failed at step " stepname) #f)
			 ))))
		  (if (and (steprun-good? logpro-used (vector-ref exit-info 2))
			   (not (null? tal)))
		      (if (not run-one) ;; if we got here we completed the step, if run-one is true, stop
			  (loop (car tal) (cdr tal) stepname runflag))))
		(debug:print 4 *default-log-port* "WARNING: a prior step failed, stopping at " ezstep)))
	  
	  ;; Once done with step/steps update the test record
	  ;;
	  (let* ((item-path (db:test-get-item-path testdat)) ;; (item-list->path itemdat))
		 (testinfo  (rmt:get-testinfo-by-id area-dat run-id test-id))) ;; refresh the testdat, call it iteminfo in case need prev/curr
	    ;; Am I completed?
	    (if (equal? (db:test-get-state testinfo) "RUNNING") ;; (not (equal? (db:test-get-state testinfo) "COMPLETED"))
		(let ((new-state  (if kill-job "KILLED" "COMPLETED") ;; (if (eq? (vector-ref exit-info 2) 0) ;; exited with "good" status
				  ;; "COMPLETED"
				  ;; (db:test-get-state testinfo)))   ;; else preseve the state as set within the test
				  )
		      (new-status (cond
				   ((not (vector-ref exit-info 1)) "FAIL") ;; job failed to run
				   ((eq? rollup-status 0)
				    ;; if the current status is AUTO the defer to the calculated value (i.e. leave this AUTO)
				    (if (equal? (db:test-get-status testinfo) "AUTO") "AUTO" "PASS"))
				   ((eq? rollup-status 1) "FAIL")
				   ((eq? rollup-status 2)
				    ;; if the current status is AUTO the defer to the calculated value but qualify (i.e. make this AUTO-WARN)
				    (if (equal? (db:test-get-status testinfo) "AUTO") "AUTO-WARN" "WARN"))
				   (else "FAIL")))) ;; (db:test-get-status testinfo)))
		  (debug:print-info 2 *default-log-port* "Test NOT logged as COMPLETED, (state=" (db:test-get-state testinfo) "), updating result, rollup-status is " rollup-status)
		  (tests:test-set-status! area-dat test-id 
					  new-state
					  new-status
					  (args:get-arg "-m") #f)
		  ;; need to update the top test record if PASS or FAIL and this is a subtest
		  (if (not (equal? item-path ""))
		      (cdb:set-state-status-and-roll-up-items *runremote* run-id test-name item-path new-status))))
	    ;; for automated creation of the rollup html file this is a good place...
	    (if (not (equal? item-path ""))
		(tests:summarize-items area-dat #f run-id test-id test-name #f)) ;; don't force - just update if no
	    )))
    (pop-directory)
    rollup-status))

;;                       0           1              2              3
(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 area-dat run-id test-id stepname)
  (let ((cname (conc stepname ".dat")))
    (if (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
              (rmt:csv->test-data area-dat 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"))
	  ;;  (BB> "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)))

(define (launch:runstep area-dat ezstep run-id test-id exit-info m tal testconfig)
  (let* ((stepname       (car ezstep))  ;; do stuff to run the step
	 (stepinfo       (cadr ezstep))
	 (stepparts      (string-match (regexp "^(\\{([^\\}]*)\\}\\s*|)(.*)$") stepinfo))
	 (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"))

    ;;   (if (and prevstep (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! area-dat run-id test-id stepname "start" "-" #f #f)
    ;; now launch the actual process
    (call-with-environment-variables 
     (list (cons "PATH" (conc (get-environment-variable "PATH") ":.")))
     (lambda () ;; (process-run "/bin/bash" "-c" "exec ls -l /tmp/foobar > /tmp/delme-more.log 2>&1")
       (let* ((cmd (conc stepcmd " > " stepname ".log 2>&1")) ;; >outfile 2>&1 
	      (pid (process-run "/bin/bash" (list "-c" cmd))))

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

	 (rmt:test-set-top-process-pid area-dat run-id test-id pid)
	 (let processloop ((i 0))
	   (let-values (((pid-val exit-status exit-code)(process-wait pid #t)))
		       (mutex-lock! m)
		       (launch:einf-pid-set!         exit-info pid)         ;; (vector-set! exit-info 0 pid)
		       (launch:einf-exit-status-set! exit-info exit-status) ;; (vector-set! exit-info 1 exit-status)
		       (launch:einf-exit-code-set!   exit-info exit-code)   ;; (vector-set! exit-info 2 exit-code)
		       (mutex-unlock! m)

	  (logfna (if logpro-used (conc stepname ".html") ""))
	  (comment #f))
      (if logpro-used
	  (let ((datfile (conc stepname ".dat")))
	    ;; load the .dat file into the test_data table if it exists
	    (if (file-exists? datfile)
		(set! comment (launch:load-logpro-dat run-id test-id stepname)))
	    (rmt:test-set-log! area-dat run-id test-id (conc stepname ".html"))))
      (rmt:teststep-set-status! area-dat run-id test-id stepname "end" exinfo comment logfna))
    ;; set the test final status
    (let* ((process-exit-status (launch:einf-exit-code exit-info)) ;; (vector-ref exit-info 2))
	   (this-step-status (cond
			      ((and (eq? process-exit-status 2) logpro-used) 'warn)   ;; logpro 2 = warnings
			      ((and (eq? process-exit-status 3) logpro-used) 'check)  ;; logpro 3 = check
			      ((and (eq? process-exit-status 4) logpro-used) 'waived) ;; logpro 4 = waived
			      ((and (eq? process-exit-status 5) logpro-used) 'abort)  ;; logpro 5 = abort

      (debug:print 4 *default-log-port* "Exit value received: " (launch:einf-exit-code exit-info) " logpro-used: " logpro-used 
		   " this-step-status: " this-step-status " overall-status: " overall-status 
		   " next-status: " next-status " rollup-status: "  (launch:einf-rollup-status exit-info)) ;; (vector-ref exit-info 3))
      (case next-status
	((warn)
	 (launch:einf-rollup-status-set! exit-info 2) ;; (vector-set! exit-info 3 2) ;; rollup-status
	 ;; NB// test-set-status! does rdb calls under the hood
	 (tests:test-set-status! area-dat run-id test-id next-state "WARN" 
				 (if (eq? this-step-status 'warn) "Logpro warning found" #f)
				 #f))
	((check)
	 (launch:einf-rollup-status-set! exit-info 3) ;; (vector-set! exit-info 3 3) ;; rollup-status
	 ;; NB// test-set-status! does rdb calls under the hood
	 (tests:test-set-status! area-dat run-id test-id next-state "CHECK" 
				 (if (eq? this-step-status 'check) "Logpro check found" #f)
				 #f))
	((waived)
	 (launch:einf-rollup-status-set! exit-info 4) ;; (vector-set! exit-info 3 3) ;; rollup-status
	 ;; NB// test-set-status! does rdb calls under the hood
	 (tests:test-set-status! area-dat run-id test-id next-state "WAIVED" 
				 (if (eq? this-step-status 'check) "Logpro waived found" #f)
				 #f))
	((abort)
	 (launch:einf-rollup-status-set! exit-info 5) ;; (vector-set! exit-info 3 4) ;; rollup-status
	 ;; NB// test-set-status! does rdb calls under the hood
	 (tests:test-set-status! area-dat run-id test-id next-state "ABORT" 
				 (if (eq? this-step-status 'abort) "Logpro abort found" #f)
				 #f))
	((skip)
	 (launch:einf-rollup-status-set! exit-info 6) ;; (vector-set! exit-info 3 4) ;; rollup-status
	 ;; NB// test-set-status! does rdb calls under the hood
	 (tests:test-set-status! area-dat run-id test-id next-state "SKIP" 
				 (if (eq? this-step-status 'skip) "Logpro skip found" #f)
				 #f))
	((pass)
	 (tests:test-set-status! area-dat run-id test-id next-state "PASS" #f #f))
	(else ;; 'fail
	 (launch:einf-rollup-status-set! exit-info 1) ;; (vector-set! exit-info 3 1) ;; force fail, this used to be next-state but that doesn't make sense. should always be "COMPLETED" 
	 (tests:test-set-status! area-dat run-id test-id "COMPLETED" "FAIL" (conc "Failed at step " stepname) #f)
	 )))
    logpro-used))

(define (launch:manage-steps area-dat run-id test-id item-path fullrunscript ezsteps test-name tconfigreg exit-info m)
  ;; (let-values
  ;;  (((pid exit-status exit-code)
  ;;    (run-n-wait fullrunscript)))
  ;; (tests:test-set-status! test-id "RUNNING" "n/a" #f #f)
  ;; Since we should have a clean slate at this time there is no need to do 
  ;; any of the other stuff that tests:test-set-status! does. Let's just 
  ;; force RUNNING/n/a

  ;; (thread-sleep! 0.3)
  ;; (tests:test-force-state-status! run-id test-id "RUNNING" "n/a")
  (rmt:set-state-status-and-roll-up-items area-dat run-id test-name item-path "RUNNING" #f #f) 
  ;; (thread-sleep! 0.3) ;; NFS slowness has caused grief here

  ;; if there is a runscript do it first
  (if fullrunscript
      (let ((pid (process-run fullrunscript)))
	(rmt:test-set-top-process-pid area-dat run-id test-id pid)
	(let loop ((i 0))
	  (let-values
	   (((pid-val exit-status exit-code) (process-wait pid #t)))
	   (mutex-lock! m)
	   (launch:einf-pid-set!           exit-info  pid)         ;; (vector-set! exit-info 0 pid)
	   (launch:einf-exit-status-set!   exit-info  exit-status) ;; (vector-set! exit-info 1 exit-status)
	   (launch:einf-exit-code-set!     exit-info  exit-code)   ;; (vector-set! exit-info 2 exit-code)

			(launch:load-logpro-dat run-id test-id stepname))
		    (if (steprun-good? logpro-used (launch:einf-exit-code exit-info))
			(if (not (null? tal))
			    (loop (car tal) (cdr tal) stepname))
			(debug:print 4 *default-log-port* "WARNING: step " (car ezstep) " failed. Stopping")))
		  (debug:print 4 *default-log-port* "WARNING: a prior step failed, stopping at " ezstep)))))))

(define (launch:monitor-job area-dat run-id test-id item-path fullrunscript ezsteps test-name tconfigreg exit-info m work-area runtlim misc-flags)
  (let* ((start-seconds (current-seconds))
	 (calc-minutes  (lambda ()
			  (inexact->exact 
			   (round 
			    (- 
			     (current-seconds) 
			     start-seconds)))))
	 (kill-tries 0))
    ;; (tests:set-full-meta-info #f test-id run-id (calc-minutes) work-area)
    ;; (tests:set-full-meta-info test-id run-id (calc-minutes) work-area)
    (tests:set-full-meta-info #f area-dat test-id run-id (calc-minutes) work-area 10)
    (let loop ((minutes   (calc-minutes))
	       (cpu-load  (alist-ref 'adj-core-load (common:get-normalized-cpu-load #f)))
	       (disk-free (get-df (current-directory))))
      (let ((new-cpu-load (let* ((load  (alist-ref 'adj-core-load (common:get-normalized-cpu-load #f)))
				 (delta (abs (- load cpu-load))))
			    (if (> delta 0.1) ;; don't bother updating with small changes
				load
				#f)))
	    (new-disk-free (let* ((df    (get-df (current-directory)))
				  (delta (abs (- df disk-free))))
			     (if (> delta 200) ;; ignore changes under 200 Meg
				 df
				 #f))))
	(set! kill-job? (or (test-get-kill-request area-dat run-id test-id) ;; run-id test-name itemdat))
			    (and runtlim (let* ((run-seconds   (- (current-seconds) start-seconds))
						(time-exceeded (> run-seconds runtlim)))
					   (if time-exceeded
					       (begin
						 (debug:print-info 0 *default-log-port* "KILLING TEST DUE TO TIME LIMIT EXCEEDED! Runtime=" run-seconds " seconds, limit=" runtlim)
						 #t)
					       #f)))))
	(tests:update-central-meta-info area-dat run-id test-id new-cpu-load new-disk-free (calc-minutes) #f #f)
	(if kill-job? 
	    (begin
	      (mutex-lock! m)
	      ;; NOTE: The pid can change as different steps are run. Do we need handshaking between this
	      ;;       section and the runit section? Or add a loop that tries three times with a 1/4 second
	      ;;       between tries?
	      (let* ((pid1 (launch:einf-pid exit-info)) ;; (vector-ref exit-info 0))
		     (pid2 (rmt:test-get-top-process-pid area-dat run-id test-id))
		     (pids (delete-duplicates (filter number? (list pid1 pid2)))))
		(if (not (null? pids))
		    (begin
		      (for-each
		       (lambda (pid)
			 (handle-exceptions
			  exn

				 (handle-exceptions
				  exn
				  #f
				  (process-signal pid-num signal/kill)))
			       (process:get-sub-pids pid))))
		       ;;    (debug:print-info 0 *default-log-port* "not killing process " pid " as it is not alive"))))
		       pids)
		      (tests:test-set-status! area-dat run-id test-id "KILLED"  "KILLED" (args:get-arg "-m") #f))
		    (begin
		      (debug:print-error 0 *default-log-port* "Nothing to kill, pid1=" pid1 ", pid2=" pid2)
		      (tests:test-set-status! area-dat run-id test-id "KILLED"  "FAILED TO KILL" (args:get-arg "-m") #f)
		      )))
	      (mutex-unlock! m)
	      ;; no point in sticking around. Exit now.
	      (exit)))
	(if (hash-table-ref/default misc-flags 'keep-going #f)
	    (begin
	      (thread-sleep! 3) ;; (+ 3 (random 6))) ;; add some jitter to the call home time to spread out the db accesses
	      (if (hash-table-ref/default misc-flags 'keep-going #f)  ;; keep originals for cpu-load and disk-free unless they change more than the allowed delta
		  (loop (calc-minutes) (or new-cpu-load cpu-load) (or new-disk-free disk-free)))))))
    (tests:update-central-meta-info area-dat run-id test-id (get-cpu-load) (get-df (current-directory))(calc-minutes) #f #f))) ;; NOTE: Checking twice for keep-going is intentional

(define (launch:execute encoded-cmd)
  (let* ((cmdinfo    (common:read-encoded-string encoded-cmd))
	 (tconfigreg #f))
    (setenv "MT_CMDINFO" encoded-cmd)
    (if (list? cmdinfo) ;; ((testpath /tmp/mrwellan/jazzmind/src/example_run/tests/sqlitespeed)
	;; (test-name sqlitespeed) (runscript runscript.rb) (db-host localhost) (run-id 1))

                                  (if (substring-index "/" runscript)
                                      runscript ;; use unadultered if contains slashes
                                      (let ((fulln (conc testpath "/" runscript)))
	                                  (if (and (file-exists? fulln)
                                                   (file-execute-access? fulln))
                                              fulln
                                              runscript))))) ;; assume it is on the path
	       (area-dat (make-remote))) ;; (rollup-status 0)

	  ;; NFS might not have propagated the directory meta data to the run host - give it time if needed
	  (let loop ((count 0))
	    (if (or (file-exists? top-path)
		    (> count 10))
		(change-directory top-path)
		(begin
		  (debug:print 0 *default-log-port* "INFO: Not starting job yet - directory " top-path " not found")
		  (thread-sleep! 10)
		  (loop (+ count 1)))))
	  (launch:setup) ;; should be properly in the top-path now
	  (set! tconfigreg (tests:get-all))
	  (let ((sighand (lambda (signum)
			   ;; (signal-mask! signum) ;; to mask or not? seems to cause issues in exiting
			   (if (eq? signum signal/stop)
			       (debug:print-error 0 *default-log-port* "attempt to STOP process. Exiting."))
			   (set! *time-to-exit* #t)
			   (print "Received signal " signum ", cleaning up before exit. Please wait...")
			   (let ((th1 (make-thread (lambda ()
						     (rmt:test-set-state-status area-dat run-id test-id "INCOMPLETE" "KILLED" #f)
						     (print "Killed by signal " signum ". Exiting")
						     (thread-sleep! 1)
						     (exit 1))))
				 (th2 (make-thread (lambda ()
						     (thread-sleep! 2)
						     (debug:print 0 *default-log-port* "Done")
						     (exit 4)))))
			     (thread-start! th2)
			     (thread-start! th1)
			     (thread-join! th2)))))
	    (set-signal-handler! signal/int sighand)
	    (set-signal-handler! signal/term sighand)
	    ) ;; (set-signal-handler! signal/stop sighand)
	  
	  ;; Do not run the test if it is REMOVING, RUNNING, KILLREQ or REMOTEHOSTSTART,
	  ;; Mark the test as REMOTEHOSTSTART *IMMEDIATELY*
	  ;;
	  (let* ((test-info (rmt:get-test-info-by-id area-dat run-id test-id))
		 (test-host (db:test-get-host        test-info))
		 (test-pid  (db:test-get-process_id  test-info)))
	    (cond
	     ((member (db:test-get-state test-info) '("INCOMPLETE" "KILLED" "UNKNOWN" "KILLREQ" "STUCK")) ;; prior run of this test didn't complete, go ahead and try to rerun
	      (debug:print 0 *default-log-port* "INFO: test is INCOMPLETE or KILLED, treat this execute call as a rerun request")
	      ;; (tests:test-force-state-status! run-id test-id "REMOTEHOSTSTART" "n/a")
	      (rmt:test-set-state-status area-dat run-id test-id "REMOTEHOSTSTART" "n/a" #f)
	      ) ;; prime it for running
	     ((member (db:test-get-state test-info) '("RUNNING" "REMOTEHOSTSTART"))
	      (if (process:alive-on-host? test-host test-pid)
		  (debug:print-error 0 *default-log-port* "test state is "  (db:test-get-state test-info) " and process " test-pid " is still running on host " test-host ", cannot proceed")
		  ;; (tests:test-force-state-status! run-id test-id "REMOTEHOSTSTART" "n/a")
		  (rmt:test-set-state-status area-dat run-id test-id "REMOTEHOSTSTART" "n/a" #f)
		  ))
	     ((not (member (db:test-get-state test-info) '("REMOVING" "REMOTEHOSTSTART" "RUNNING" "KILLREQ")))
	      ;; (tests:test-force-state-status! run-id test-id "REMOTEHOSTSTART" "n/a")
	      (rmt:test-set-state-status area-dat run-id test-id "REMOTEHOSTSTART" "n/a" #f)
	      )
	     (else ;; (member (db:test-get-state test-info) '("REMOVING" "REMOTEHOSTSTART" "RUNNING" "KILLREQ"))
	      (debug:print-error 0 *default-log-port* "test state is " (db:test-get-state test-info) ", cannot proceed")
	      (exit))))
	  
	  (debug:print 2 *default-log-port* "Exectuing " test-name " (id: " test-id ") on " (get-host-name))
	  (set! keys       (rmt:get-keys area-dat))

	  ;; one of these is defunct/redundant ...
	  (if (not (launch:setup force: #t))
	      (begin
		(debug:print 0 *default-log-port* "Failed to setup, exiting") 
		;; (sqlite3:finalize! db)
		;; (sqlite3:finalize! tdb)
		(exit 1)))

	  ;; (change-directory top-path)
	  ;; Can setup as client for server mode now
	  ;; (client:setup)

	  
	  ;; environment overrides are done *before* the remaining critical envars.
	  (alist->env-vars env-ovrd)
	  (runs:set-megatest-env-vars area-dat run-id inkeys: keys inkeyvals: keyvals)
	  (set-item-env-vars itemdat)
	  (save-environment-as-files "megatest")
	  ;; open-run-close not needed for test-set-meta-info
	  ;; (tests:set-full-meta-info #f test-id run-id 0 work-area)
	  ;; (tests:set-full-meta-info test-id run-id 0 work-area)
	  (tests:set-full-meta-info #f area-dat test-id run-id 0 work-area 10)

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

	  (if (args:get-arg "-xterm")
	      (set! fullrunscript "xterm")
	      (if (and fullrunscript 
		       (file-exists? fullrunscript)

	    (debug:print-info 0 *default-log-port* "Megatest exectute of test " test-name ", item path " item-path " complete. Notifying the db ...")
	    (hash-table-set! misc-flags 'keep-going #f)
	    (thread-join! th1)
	    (thread-sleep! 1)       ;; givbe thread th1 a chance to be done TODO: Verify this is needed. At 0.1 I was getting fail to stop, increased to total of 1.1 sec.
	    (mutex-lock! m)
	    (let* ((item-path (item-list->path itemdat))
		   ;; only state and status needed - use lazy routine
		   (testinfo  (rmt:get-testinfo-state-status area-dat run-id test-id)))
	      ;; Am I completed?
	      (if (member (db:test-get-state testinfo) '("REMOTEHOSTSTART" "RUNNING")) ;; NOTE: It should *not* be REMOTEHOSTSTART but for reasons I don't yet understand it sometimes gets stuck in that state ;; (not (equal? (db:test-get-state testinfo) "COMPLETED"))
		  (let ((new-state  (if kill-job? "KILLED" "COMPLETED") ;; (if (eq? (vector-ref exit-info 2) 0) ;; exited with "good" status
				                                        ;; "COMPLETED"
							                ;; (db:test-get-state testinfo)))   ;; else preseve the state as set within the test
				    )
			(new-status (cond

				      (if (equal? (db:test-get-status testinfo) "AUTO") "AUTO-WARN" "WARN"))
				     ((eq? (launch:einf-rollup-status exit-info) 3) "CHECK")
				     ((eq? (launch:einf-rollup-status exit-info) 4) "WAIVED")
				     ((eq? (launch:einf-rollup-status exit-info) 5) "ABORT")
				     ((eq? (launch:einf-rollup-status exit-info) 6) "SKIP")
				     (else "FAIL")))) ;; (db:test-get-status testinfo)))
		    (debug:print-info 1 *default-log-port* "Test exited in state=" (db:test-get-state testinfo) ", setting state/status based on exit code of " (launch:einf-exit-status exit-info) " and rollup-status of " (launch:einf-rollup-status exit-info))
		    (tests:test-set-status! area-dat run-id 
					    test-id 
					    new-state
					    new-status
					    (args:get-arg "-m") #f)
		    ;; need to update the top test record if PASS or FAIL and this is a subtest
		    ;; NO NEED TO CALL set-state-status-and-roll-up-items HERE, THIS IS DONE IN set-state-status-and-roll-up-items called by tests:test-set-status!
		    ))
	      ;; for automated creation of the rollup html file this is a good place...
	      (if (not (equal? item-path ""))
		  (tests:summarize-items area-dat run-id test-id test-name #f))
	      (tests:summarize-test area-dat run-id test-id)  ;; don't force - just update if no
	      (rmt:update-run-stats area-dat run-id (rmt:get-raw-run-stats area-dat run-id)))
	    (mutex-unlock! m)
	    (debug:print 2 *default-log-port* "Output from running " fullrunscript ", pid " (launch:einf-pid exit-info) " in work area " 
			 work-area ":\n====\n exit code " (launch:einf-exit-code exit-info) "\n" "====\n")
	    (if (not (launch:einf-exit-status exit-info))
		(exit 4)))))))

(define (launch:cache-config)

	(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)))
	(mutex-unlock! *launch-setup-mutex*)
	res)))

(define (launch:setup-body area-dat #!key (force #f))
  (let* ((toppath  (or *toppath* (getenv "MT_RUN_AREA_HOME"))) ;; preserve toppath
	 (runname  (common:args-get-runname))
	 (target   (common:args-get-target))
	 (linktree (common:get-linktree))
	 (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 "/" target "/" runname) #f))

	      (set! toppath      *toppath*)
	      (if (not *toppath*)
		  (begin
		    (debug:print-error 0 *default-log-port* "you are not in a megatest area!")
		    (exit 1)))
	      (setenv "MT_RUN_AREA_HOME" *toppath*)
	      ;; the seed read is done, now read runconfigs, cache it then read megatest.config one more time and cache it
	      (let* ((keys         (rmt:get-keys area-dat))
		     (key-vals     (keys:target->keyval keys target))
		     (linktree     (or (getenv "MT_LINKTREE")
				       (if *configdat* (configf:lookup *configdat* "setup" "linktree") #f)))
		     (second-pass  (find-and-read-config
				    mtconfig
				    environ-patt: "env-override"
				    given-toppath: toppath

	 (lnkbase   (conc linktree "/" target "/" runname))
	 (lnkpath   (conc lnkbase "/" testname))
	 (lnkpathf  (conc lnkpath (if not-iterated "" "/") item-path))
	 (lnktarget (conc lnkpath "/" item-path)))

    ;; Update the rundir path in the test record for all, rundir=physical, shortdir=logical
    ;;                                                 rundir   shortdir
    (rmt:general-call area-dat 'test-set-rundir-shortdir run-id lnkpathf test-path testname item-path run-id)

    (debug:print 2 *default-log-port* "INFO:\n       lnkbase=" lnkbase "\n       lnkpath=" lnkpath "\n  toptest-path=" toptest-path "\n     test-path=" test-path)
    (if (not (file-exists? linktree))
	(begin
	  (debug:print 0 *default-log-port* "WARNING: linktree did not exist! Creating it now at " linktree)
	  (create-directory linktree #t))) ;; (system (conc "mkdir -p " linktree))))
    ;; create the directory for the tests dir links, this is needed no matter what...

    ;; NB - This was not working right - some top tests are not getting the path set!!!
    ;;
    ;; Do the setting of this record after the paths are created so that the shortdir can 
    ;; be set to the real directory location. This is safer for future clean up if the link
    ;; tree is damaged or lost.
    ;; 
    (if (not (hash-table-ref/default *toptest-paths* testname #f))
	(let* ((testinfo       (rmt:get-test-info-by-id area-dat run-id test-id)) ;;  run-id testname item-path))
	       (curr-test-path (if testinfo ;; (filedb:get-path *fdb*
							     ;; (db:get-path dbstruct
				   ;; (rmt:sdb-qry area-dat '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 area-dat 'test-set-rundir-shortdir run-id lnkpath 
			    (if (file-exists? lnkpath)
				;; (resolve-pathname lnkpath)
				(common:nice-path lnkpath)
				lnkpath)
			    testname "" run-id)
	  ;; (rmt:general-call area-dat 'test-set-rundir run-id lnkpath testname "") ;; toptest-path)
	  (if (or (not curr-test-path)
		  (not (directory-exists? toptest-path)))
	      (begin
		(debug:print-info 2 *default-log-port* "Creating " toptest-path " and link " lnkpath)
		(handle-exceptions
		 exn
		 #f ;; don't care to catch and deal with errors here for now.

	   (test-sig        (conc (common:get-testsuite-name) ":" test-name ":" item-path)) ;; (item-list->path itemdat))) ;; test-path is the full path including the item-path
	   (work-area       #f)
	   (toptest-work-area #f) ;; for iterated tests the top test contains data relevant for all
	   (diskpath   #f)
	   (cmdparms   #f)
	   (fullcmd    #f) ;; (define a (with-output-to-string (lambda ()(write x))))
	   (mt-bindir-path #f)
	   (testinfo   (rmt:get-test-info-by-id area-dat run-id test-id))
	   (mt_target  (string-intersperse (map cadr keyvals) "/"))
	   (debug-param (append (if (args:get-arg "-debug")  (list "-debug" (args:get-arg "-debug")) '())
				(if (args:get-arg "-logging")(list "-logging") '()))))
      ;; (if hosts (set! hosts (string-split hosts)))
      ;; set the megatest to be called on the remote host
      (if (not remote-megatest)(set! remote-megatest local-megatest)) ;; "megatest"))
      (set! mt-bindir-path (pathname-directory remote-megatest))
      (if launcher (set! launcher (string-split launcher)))
      ;; set up the run work area for this test
      (if (and (args:get-arg "-preclean") ;; user has requested to preclean for this run
	       (not (member (db:test-get-rundir testinfo)(list "n/a" "/tmp/badname")))) ;; n/a is a placeholder and thus not a read dir
	  (begin
	    (debug:print-info 0 *default-log-port* "attempting to preclean directory " (db:test-get-rundir testinfo) " for test " test-name "/" item-path)
	    (runs:remove-test-directory area-dat testinfo 'remove-data-only))) ;; remove data only, do not perturb the record
      
      ;; prevent overlapping actions - set to LAUNCHED as early as possible
      ;;
      ;; the following call handles waiver propogation. cannot yet condense into roll-up-pass-fail
      (tests:test-set-status! area-dat run-id test-id "LAUNCHED" "n/a" #f #f) ;; (if launch-results launch-results "FAILED"))
      (rmt:set-state-status-and-roll-up-items area-dat run-id test-name item-path #f "LAUNCHED" #f)
      ;; (pp (hash-table->alist tconfig))
      (set! diskpath (get-best-disk *configdat* tconfig))
      (if diskpath
	  (let ((dat  (create-work-area run-id run-info keyvals test-id test-path diskpath test-name itemdat)))
	    (set! work-area (car dat))
	    (set! toptest-work-area (cadr dat))
	    (debug:print-info 2 *default-log-port* "Using work area " work-area))

					(list 'runtlim   (if run-time-limit (common:hms-string->seconds run-time-limit) #f))
					(list 'env-ovrd  (hash-table-ref/default *configdat* "env-override" '())) 
					(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 area-dat run-id test-id)
      ;; if the dir does not exist we may have a itempath where individual variables are a path, launch anyway
      (if (file-exists? work-area)
	  (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))))

	(alist->env-vars testprevvals)
	(alist->env-vars commonprevvals)
	launch-results))
    (change-directory *toppath*)))

;; recover a test where the top controlling mtest may have died
;;
(define (launch:recover-test area-dat run-id test-id)
  ;; this function is called on the test run host via ssh
  ;;
  ;; 1. look at the process from pid
  ;;    - is it owned by calling user
  ;;    - it it's run directory correct for the test
  ;;    - is there a controlling mtest (maybe stuck)
  ;; 2. if recovery is needed watch pid
  ;;    - when it exits take the exit code and do the needful
  ;;
  (let* ((pid (rmt:test-get-top-process-id area-dat run-id test-id))
	 (psres (with-input-from-pipe
		 (conc "ps -F -u " (current-user-name) " | grep -E '" pid " ' | grep -v 'grep -E " pid "'")
		 (lambda ()
		   (read-line))))
	 (rundir (if (string? psres) ;; real process owned by user
		     (read-symbolic-link (conc "/proc/" pid "/cwd"))
		     #f)))
    ;; now wait on that process if all is correct
    ;; periodically update the db with runtime
    ;; when the process exits look at the db, if still RUNNING after 10 seconds set
    ;; state/status appropriately
    (process-wait pid)))

		 (pp data))))))
      (if out-file (close-output-port out-port))
      (exit) ;; yes, bending the rules here - need to exit since this is a utility
      ))

(if (args:get-arg "-ping")
    (let* ((server-id     (string->number (args:get-arg "-ping"))) ;; extract run-id (i.e. no ":"
	   (host:port     (args:get-arg "-ping"))
	   (area-dat      (make-remote)))
      (server:ping area-dat (or server-id host:port) do-exit: #t)))

;;======================================================================
;; Capture, save and manipulate environments
;;======================================================================

;; NOTE: Keep these above the section where the server or client code is setup

             (json-write targets))
            (else
             (debug:print-error 0 *default-log-port* "dump output format " (args:get-arg "-dumpmode") " not supported for -list-targets")))
          (set! *didsomething* #t))))

;; cache the runconfigs in $MT_LINKTREE/$MT_TARGET/$MT_RUNNAME/.runconfig
;;
(define (full-runconfigs-read area-dat)
;; in the envprocessing branch the below code replaces the further below code
;;  (if (eq? *configstatus* 'fulldata)
;;      *runconfigdat*
;;      (begin
;;	(launch:setup)
;;	*runconfigdat*)))

  (let* ((rundir (if (and (getenv "MT_LINKTREE")(getenv "MT_TARGET")(getenv "MT_RUNNAME"))
		     (conc (getenv "MT_LINKTREE") "/" (getenv "MT_TARGET") "/" (getenv "MT_RUNNAME"))
		     #f))
	 (cfgf   (if rundir (conc rundir "/.runconfig." megatest-version "-" megatest-fossil-hash) #f)))
    (if (and cfgf
	     (file-exists? cfgf)
	     (file-write-access? cfgf))
	(configf:read-alist cfgf)
	(let* ((keys   (rmt:get-keys area-dat))
	       (target (common:args-get-target))
	       (key-vals (if target (keys:target->keyval keys target) #f))
	       (sections (if target (list "default" target) #f))
	       (data     (begin
			   (setenv "MT_RUN_AREA_HOME" *toppath*)
			   (if key-vals
			       (for-each (lambda (kt)

;;======================================================================
;; Remove old run(s)
;;======================================================================

;; since several actions can be specified on the command line the removal
;; is done first
(define (operate-on area-dat action)
  (let* ((runrec (runs:runrec-make-record))
	 (target (common:args-get-target)))
    (cond
     ((not target)
      (debug:print-error 0 *default-log-port* "Missing required parameter for " action ", you must specify -target or -reqtarg")
      (exit 1))
     ((not (or (args:get-arg ":runname")

	  (begin
	    (debug:print-error 0 *default-log-port* "Attempted " action "on test(s) but run area config file not found")
	    (exit 1))
	  ;; put test parameters into convenient variables
	  (begin
	    ;; check for correct version, exit with message if not correct
	    (common:exit-on-version-changed)
	    (runs:operate-on  area-dat action
			      target
			      (common:args-get-runname)  ;; (or (args:get-arg "-runname")(args:get-arg ":runname"))
			      (common:args-get-testpatt #f) ;; (args:get-arg "-testpatt")
			      state: (common:args-get-state)
			      status: (common:args-get-status)
			      new-state-status: (args:get-arg "-set-state-status"))))
      (set! *didsomething* #t)))))

(if (args:get-arg "-remove-runs")
    (general-run-call 
     "-remove-runs"
     "remove runs"
     (lambda (area-dat target runname keys keyvals)
       (operate-on area-dat 'remove-runs))))

(if (args:get-arg "-set-state-status")
    (general-run-call 
     "-set-state-status"
     "set state and status"
     (lambda (area-dat target runname keys keyvals)
       (operate-on area-dat 'set-state-status))))

(if (or (args:get-arg "-set-run-status")
	(args:get-arg "-get-run-status"))
    (general-run-call
     "-set-run-status"
     "set run status"
     (lambda (area-dat target runname keys keyvals)
       (let* ((runsdat  (rmt:get-runs-by-patt area-dat keys runname 
					(common:args-get-target)
					#f #f #f #f))
	      (header   (vector-ref runsdat 0))
	      (rows     (vector-ref runsdat 1)))
	 (if (null? rows)
	     (begin
	       (debug:print-info 0 *default-log-port* "No matching run found.")
	       (exit 1))
	     (let* ((row      (car (vector-ref runsdat 1)))
		    (run-id   (db:get-value-by-header row header "id")))
	       (if (args:get-arg "-set-run-status")
		   (rmt:set-run-status area-dat run-id (args:get-arg "-set-run-status") msg: (args:get-arg "-m"))
		   (print (rmt:get-run-status area-dat run-id))
		   )))))))

;;======================================================================
;; Query runs
;;======================================================================

;; -fields runs:id,target,runname,comment+tests:id,testname,item_path+steps

(if (or (args:get-arg "-list-runs")
	(args:get-arg "-list-db-targets"))
    (if (launch:setup)
	(let* (;; (dbstruct    (make-dbr:dbstruct path: *toppath* local: (args:get-arg "-local")))
	       (runpatt     (args:get-arg "-list-runs"))
               (access-mode (db:get-access-mode))
	       (testpatt    (common:args-get-testpatt #f))
	       (area-dat    (make-remote))
	       ;; (if (args:get-arg "-testpatt") 
	       ;;  	        (args:get-arg "-testpatt") 
	       ;;  	        "%"))
	       (keys        (rmt:get-keys area-dat)) ;; (db:get-keys dbstruct))
	       ;; (runsdat  (db:get-runs dbstruct runpatt #f #f '()))
	;; (runsdat     (rmt:get-runs-by-patt area-dat keys (or runpatt "%") (common:args-get-target) ;; (db:get-runs-by-patt dbstruct keys (or runpatt "%") (common:args-get-target)
	;; 		           	 #f #f '("id" "runname" "state" "status" "owner" "event_time" "comment") 0))
	       (runsdat     (rmt:get-runs-by-patt area-dat keys (or runpatt "%") 
                                                  (common:args-get-target) #f #f '("id" "runname" "state" "status" "owner" "event_time" "comment") 0))
	       (runstmp     (db:get-rows runsdat))
	       (header      (db:get-header runsdat))
	       ;; this is "-since" support. This looks at last mod times of <run-id>.db files
	       ;; and collects those modified since the -since time.
	       (runs        runstmp)
                        ;; (if (and (not (null? runstmp))

;; 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)))
;;       ;; (rmt:get-tests-for-runs-mindata area-dat run-ids testpatt states status not-in)
;;       (print (sort run-ids <))
;;       (set! *didsomething* #t)))
      
      
;;======================================================================
;; full run
;;======================================================================

	(args:get-arg "-run")
	(args:get-arg "-rerun-clean")
	(args:get-arg "-rerun-all")
	(args:get-arg "-runtests"))
    (general-run-call 
     "-runall"
     "run all tests"
     (lambda (area-dat target runname keys keyvals)
       (if (args:get-arg "-rerun-clean") ;; first set states/statuses correct
	   (let ((states   (or (configf:lookup *configdat* "validvalues" "cleanrerun-states")
			       "KILLREQ,KILLED,UNKNOWN,INCOMPLETE,STUCK,NOT_STARTED"))
		 (statuses (or (configf:lookup *configdat* "validvalues" "cleanrerun-statuses")
			       "FAIL,INCOMPLETE,ABORT,CHECK")))
	     (hash-table-set! args:arg-hash "-preclean" #t)
	     (runs:operate-on area-dat 'set-state-status
			      target
			      (common:args-get-runname)  ;; (or (args:get-arg "-runname")(args:get-arg ":runname"))
			      "%" ;; (common:args-get-testpatt #f) ;; (args:get-arg "-testpatt")
			      state:  states
			      ;; status: statuses
			      new-state-status: "NOT_STARTED,n/a")
	     (runs:operate-on area-dat 'set-state-status
			      target
			      (common:args-get-runname)  ;; (or (args:get-arg "-runname")(args:get-arg ":runname"))
			      "%" ;; (common:args-get-testpatt #f) ;; (args:get-arg "-testpatt")
			      ;; state:  states
			      status: statuses
			      new-state-status: "NOT_STARTED,n/a")))
       ;; RERUN ALL
       (if (args:get-arg "-rerun-all") ;; first set states/statuses correct
	   (begin
	     (hash-table-set! args:arg-hash "-preclean" #t)
	     (runs:operate-on area-dat 'set-state-status
			      target
			      (common:args-get-runname)  ;; (or (args:get-arg "-runname")(args:get-arg ":runname"))
			      "%" ;; (common:args-get-testpatt #f) ;; (args:get-arg "-testpatt")
			      state:  #f
			      ;; status: statuses
			      new-state-status: "NOT_STARTED,n/a")
	     (runs:operate-on area-dat 'set-state-status
			      target
			      (common:args-get-runname)  ;; (or (args:get-arg "-runname")(args:get-arg ":runname"))
			      "%" ;; (common:args-get-testpatt #f) ;; (args:get-arg "-testpatt")
			      ;; state:  states
			      status: #f
			      new-state-status: "NOT_STARTED,n/a")))
       (runs:run-tests area-dat target
		       runname
		       #f ;; (common:args-get-testpatt #f)
		       ;; (or (args:get-arg "-testpatt")
		       ;;     "%")
		       user
		       args:arg-hash))))

;;    started and completed
;;    - step started, timestamp
;;    - step completed, exit status, timestamp
;; 6. test phone home
;;    - if test run time > allowed run time then kill job
;;    - if cannot access db > allowed disconnect time then kill job























;;======================================================================
;; Rollup into a run
;;======================================================================

(if (args:get-arg "-rollup")
    (general-run-call 
     "-rollup" 
     "rollup tests" 
     (lambda (area-dat target runname keys keyvals)
       (runs:rollup-run area-dat keys
			keyvals
			(or (args:get-arg "-runname")(args:get-arg ":runname") )
			user))))

;;======================================================================
;; Lock or unlock a run
;;======================================================================

(if (or (args:get-arg "-lock")(args:get-arg "-unlock"))
    (general-run-call 
     (if (args:get-arg "-lock") "-lock" "-unlock")
     "lock/unlock tests" 
     (lambda (area-dat target runname keys keyvals)
       (runs:handle-locking
	          area-dat
		  target
		  keys
		  (or (args:get-arg "-runname")(args:get-arg ":runname") )
		  (args:get-arg "-lock")
		  (args:get-arg "-unlock")
		  user))))

	       (runscript (assoc/default 'runscript cmdinfo))
	       (db-host   (assoc/default 'db-host   cmdinfo))
	       (run-id    (assoc/default 'run-id    cmdinfo))
	       (itemdat   (assoc/default 'itemdat   cmdinfo))
	       (state     (args:get-arg ":state"))
	       (status    (args:get-arg ":status"))
	       (target    (args:get-arg "-target"))
	       (toppath   (assoc/default 'toppath   cmdinfo))
	       (area-dat  (make-remote)))
	  (change-directory toppath)
	  (if (not target)
	      (begin
		(debug:print-error 0 *default-log-port* "-target is required.")
		(exit 1)))
	  (if (not (launch:setup))
	      (begin
		(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 area-dat))
		 ;; db:test-get-paths must not be run remote
		 (paths    (tests:test-get-paths-matching area-dat keys target (args:get-arg "-test-files"))))
	    (set! *didsomething* #t)
	    (for-each (lambda (path)
			(if (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)
	   (let* ((db       #f)
		  ;; DO NOT run remote
		  (paths    (tests:test-get-paths-matching area-dat keys target (args:get-arg "-test-files"))))
	     (for-each (lambda (path)
			 (print path))
		       paths))))))

;;======================================================================
;; Archive tests
;;======================================================================
;; Archive tests matching target, runname, and testpatt
(if (args:get-arg "-archive")
    ;; else do a general-run-call
    (general-run-call 
     "-archive"
     "Archive"
     (lambda (area-dat target runname keys keyvals)
       (operate-on area-dat 'archive))))

;;======================================================================
;; Extract a spreadsheet from the runs database
;;======================================================================

(if (args:get-arg "-extract-ods")
    (general-run-call

	     (test-name (assoc/default 'test-name cmdinfo))
	     (runscript (assoc/default 'runscript cmdinfo))
	     (db-host   (assoc/default 'db-host   cmdinfo))
	     (run-id    (assoc/default 'run-id    cmdinfo))
	     (test-id   (assoc/default 'test-id   cmdinfo))
	     (itemdat   (assoc/default 'itemdat   cmdinfo))
	     (work-area (assoc/default 'work-area cmdinfo))
	     (db        #f)
	     (area-dat  (make-remote)))
	(change-directory testpath)
	(if (not (launch:setup))
	    (begin
	      (debug:print 0 *default-log-port* "Failed to setup, exiting")
	      (exit 1)))
	(if (and state status)
	    (let ((comment (launch:load-logpro-dat run-id test-id step)))
	      ;; (rmt:test-set-log! area-dat run-id test-id (conc stepname ".html"))))
	      (rmt:teststep-set-status! area-dat run-id test-id step state status (or comment msg) logfile))
	    (begin
	      (debug:print-error 0 *default-log-port* "You must specify :state and :status with every call to -step")
	      (exit 6))))))

(if (args:get-arg "-step")
    (begin
      (megatest:step 

	       (run-id    (assoc/default 'run-id    cmdinfo))
	       (test-id   (assoc/default 'test-id   cmdinfo))
	       (itemdat   (assoc/default 'itemdat   cmdinfo))
	       (work-area (assoc/default 'work-area cmdinfo))
	       (db        #f) ;; (open-db))
	       (state     (args:get-arg ":state"))
	       (status    (args:get-arg ":status"))
	       (stepname  (args:get-arg "-step"))
	       (area-dat  (make-remote)))
	  (if (not (launch:setup))
	      (begin
		(debug:print 0 *default-log-port* "Failed to setup, exiting")
		(exit 1)))

	  (if (args:get-arg "-runstep")(debug:print-info 1 *default-log-port* "Running -runstep, first change to directory " work-area))
	  (change-directory work-area)
	  ;; can setup as client for server mode now
	  ;; (client:setup)

	  (if (args:get-arg "-load-test-data")
	      ;; has sub commands that are rdb:
	      ;; DO NOT put this one into either rmt: or open-run-close
	      (tdb:load-test-data area-dat run-id test-id))
	  (if (args:get-arg "-setlog")
	      (let ((logfname (args:get-arg "-setlog")))
		(rmt:test-set-log! area-dat run-id test-id logfname)))
	  (if (args:get-arg "-set-toplog")
	      ;; DO NOT run remote
	      (tests:test-set-toplog! area-dat run-id test-name (args:get-arg "-set-toplog")))
	  (if (args:get-arg "-summarize-items")
	      ;; DO NOT run remote
	      (tests:summarize-items run-id test-id test-name #t)) ;; do force here
	  (if (args:get-arg "-runstep")
	      (if (null? remargs)
		  (begin
		    (debug:print-error 0 *default-log-port* "nothing specified to run!")

				       ((tcsh csh ksh)    ">&")
				       ((zsh bash sh ash) "2>&1 >")
				       (else ">&")))
			 (fullcmd    (conc "(" (string-intersperse 
						(cons cmd params) " ")
					   ") " redir " " logfile)))
		    ;; mark the start of the test
		    (rmt:teststep-set-status! area-dat run-id test-id stepname "start" "n/a" (args:get-arg "-m") logfile)
		    ;; run the test step
		    (debug:print-info 2 *default-log-port* "Running \"" fullcmd "\" in directory \"" startingdir)
		    (change-directory startingdir)
		    (set! exitstat (system fullcmd))
		    (set! *globalexitstatus* exitstat)
		    ;; (change-directory testpath)
		    ;; run logpro if applicable ;; (process-run "ls" (list "/foo" "2>&1" "blah.log"))
		    (if logprofile
			(let* ((htmllogfile (conc stepname ".html"))
			       (oldexitstat exitstat)
			       (cmd         (string-intersperse (list "logpro" logprofile htmllogfile "<" logfile ">" (conc stepname "_logpro.log")) " ")))
			  (debug:print-info 2 *default-log-port* "running \"" cmd "\"")
			  (change-directory startingdir)
			  (set! exitstat (system cmd))
			  (set! *globalexitstatus* exitstat) ;; no necessary
			  (change-directory testpath)
			  (rmt:test-set-log! area-dat run-id test-id htmllogfile)))
		    (let ((msg (args:get-arg "-m")))
		      (rmt:teststep-set-status! area-dat run-id test-id stepname "end" exitstat msg logfile))
		    )))
	  (if (or (args:get-arg "-test-status")
		  (args:get-arg "-set-values"))
	      (let ((newstatus (cond
				((number? status)       (if (equal? status 0) "PASS" "FAIL"))
				((and (string? status)
				      (string->number status))(if (equal? (string->number status) 0) "PASS" "FAIL"))

		    (begin
		      (debug:print-error 0 *default-log-port* "You must specify :state and :status with every call to -test-status\n" help)
		      (if (sqlite3:database? db)(sqlite3:finalize! db))
		      (exit 6)))
		(let* ((msg    (args:get-arg "-m"))
		       (numoth (length (hash-table-keys otherdata))))
		  ;; Convert to rpc inside the tests:test-set-status! call, not here
		  (tests:test-set-status! area-dat run-id test-id state newstatus msg otherdata work-area: work-area))))
	  (if (sqlite3:database? db)(sqlite3:finalize! db))
	  (set! *didsomething* #t))))

;;======================================================================
;; Various helper commands can go below here
;;======================================================================

(if (or (args:get-arg "-showkeys")
        (args:get-arg "-show-keys"))
    (let ((db       #f)
	  (keys     #f)
	  (area-dat (make-remote)))
      (if (not (launch:setup))
	  (begin
	    (debug:print 0 *default-log-port* "Failed to setup, exiting")
	    (exit 1)))
      (set! keys (rmt:get-keys area-dat)) ;;  db))
      (debug:print 1 *default-log-port* "Keys: " (string-intersperse keys ", "))
      (if (sqlite3:database? db)(sqlite3:finalize! db))
      (set! *didsomething* #t)))

(if (args:get-arg "-gui")
    (begin
      (debug:print 0 *default-log-port* "Look at the dashboard for now")

;;======================================================================
;; Wait on a run to complete
;;======================================================================

(if (and (args:get-arg "-run-wait")
	 (not (or (args:get-arg "-run")
		  (args:get-arg "-runtests")))) ;; run-wait is built into runtests now
    (let ((area-dat (make-remote)))
      (if (not (launch:setup))
	  (begin
	    (debug:print 0 *default-log-port* "Failed to setup, exiting") 
	    (exit 1)))
      (operate-on area-dat 'run-wait)
      (set! *didsomething* #t)))

;; ;; ;; redo me ;; Not converted to use dbstruct yet
;; ;; ;; redo me ;;
;; ;; ;; redo me (if (args:get-arg "-convert-to-norm")
;; ;; ;; redo me     (let* ((toppath (setup-for-run))
;; ;; ;; redo me 	   (dbstruct (if toppath (make-dbr:dbstruct path: toppath local: #t))))

      (db:multi-db-sync 
       (db:setup)
       'new2old
       )
      (set! *didsomething* #t)))

(if (args:get-arg "-generate-html")
    (let* ((toppath (launch:setup))
	   (area-dat (make-remote)))
      (if (tests:create-html-tree area-dat #f)
          (debug:print-info 0 *default-log-port* "HTML output created in " toppath "/lt/page#.html")
          (debug:print 0 *default-log-port* "Failed to create HTML output in " toppath "/lt/runs-index.html"))
      (set! *didsomething* #t)))

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

;; runs:get-runs-by-patt
;; get runs by list of criteria
;; register a test run with the db
;;
;; Use: (db-get-value-by-header (db:get-header runinfo)(db:get-rows runinfo))
;;  to extract info from the structure returned
;;
(define (mt:get-runs-by-patt area-dat keys runnamepatt targpatt)
  (let loop ((runsdat  (rmt:get-runs-by-patt area-dat keys runnamepatt targpatt 0 500 #f 0))
	     (res      '())
	     (offset   0)
	     (limit    500))
    ;; (print "runsdat: " runsdat)
    (let* ((header    (vector-ref runsdat 0))
	   (runslst   (vector-ref runsdat 1))
	   (full-list (append res runslst))
	   (have-more (eq? (length runslst) limit)))
      ;; (debug:print 0 *default-log-port* "header: " header " runslst: " runslst " have-more: " have-more)
      (if have-more 
	  (let ((new-offset (+ offset limit))
		(next-batch (rmt:get-runs-by-patt area-dat keys runnamepatt targpatt offset limit #f 0)))
	    (debug:print-info 4 *default-log-port* "More than " limit " runs, have " (length full-list) " runs so far.")
	    (debug:print-info 0 *default-log-port* "next-batch: " next-batch)
	    (loop next-batch
		  full-list
		  new-offset
		  limit))
	 (vector header full-list)))))

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

(define (mt:get-tests-for-run area-dat run-id testpatt states status #!key (not-in #t) (sort-by 'event_time) (sort-order "ASC") (qryvals #f)(last-update #f))
  (let loop ((testsdat (rmt:get-tests-for-run area-dat run-id testpatt states status 0 500 not-in sort-by sort-order qryvals last-update 'normal))
	     (res      '())
	     (offset   0)
	     (limit    500))
    (let* ((full-list (append res testsdat))
	   (have-more (eq? (length testsdat) limit)))
      (if have-more 
	  (let ((new-offset (+ offset limit)))
	    (debug:print-info 4 *default-log-port* "More than " limit " tests, have " (length full-list) " tests so far.")
	    (loop (rmt:get-tests-for-run area-dat run-id testpatt states status new-offset limit not-in sort-by sort-order qryvals last-update 'normal)
		  full-list
		  new-offset
		  limit))
	  full-list))))

(define (mt:lazy-get-prereqs-not-met area-dat run-id waitons ref-item-path #!key (mode '(normal))(itemmaps #f) )
  (let* ((key    (list run-id waitons ref-item-path mode))
	 (res    (hash-table-ref/default *pre-reqs-met-cache* key #f))
	 (useres (let ((last-time (if (vector? res) (vector-ref res 0) #f)))
		   (if last-time
		       (< (current-seconds)(+ last-time 5))
		       #f))))
    (if useres
	(let ((result (vector-ref res 1)))
	  (debug:print 4 *default-log-port* "Using lazy value res: " result)
	  result)
	(let ((newres (rmt:get-prereqs-not-met area-dat run-id waitons ref-item-path mode: mode itemmaps: itemmaps)))
	  (hash-table-set! *pre-reqs-met-cache* key (vector (current-seconds) newres))
	  newres))))

(define (mt:get-run-stats dbstruct run-id)
;;  Get run stats from local access, move this ... but where?
  (db:get-run-stats dbstruct run-id))

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

(define (mt:process-triggers dbstruct run-id test-id newstate newstatus)
  (let* ((test-dat      (db:get-test-info-by-id dbstruct run-id test-id)))
    (if test-dat
	(let* ((test-rundir   ;; (rmt:sdb-qry area-dat 'getstr ;; (filedb:get-path *fdb*
		(db:test-get-rundir test-dat)) ;; ) ;; )
	       (test-name     (db:test-get-testname test-dat))
	       (tconfig       #f)
	       (state         (if newstate  newstate  (db:test-get-state  test-dat)))
	       (status        (if newstatus newstatus (db:test-get-status test-dat))))
	  (if (and test-name
		   test-rundir   ;; #f means no dir set yet

	       ))))))

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

;; speed up for common cases with a little logic
(define (mt:test-set-state-status-by-id area-dat run-id test-id newstate newstatus newcomment)
  (if (not (and run-id test-id))
      (begin
	(debug:print-error 0 *default-log-port* "bad data handed to mt:test-set-state-status-by-id, run-id=" run-id ", test-id=" test-id ", newstate=" newstate)
	(print-call-chain (current-error-port))
	#f)
      (begin
	;; cond
	;; ((and newstate newstatus newcomment)
	;;  (rmt:general-call area-dat 'state-status-msg run-id newstate newstatus newcomment test-id))
	;; ((and newstate newstatus)
	;;  (rmt:general-call area-dat 'state-status run-id newstate newstatus test-id))
	;; (else
	;;  (if newstate   (rmt:general-call area-dat 'set-test-state   run-id newstate   test-id))
	;;  (if newstatus  (rmt:general-call area-dat 'set-test-status  run-id newstatus  test-id))
	;;  (if newcomment (rmt:general-call area-dat 'set-test-comment run-id newcomment test-id))))
	(rmt:set-state-status-and-roll-up-items area-dat run-id test-id #f newstate newstatus newcomment)
	;; (mt:process-triggers run-id test-id newstate newstatus)
	#t)))

(define (mt:test-set-state-status-by-testname area-dat run-id test-name item-path new-state new-status new-comment)
  (let ((test-id (rmt:get-test-id area-dat run-id test-name item-path)))
    (rmt:set-state-status-and-roll-up-items area-dat run-id test-name item-path new-state new-status new-comment)
    ;; (mt:process-triggers run-id test-id new-state new-status)
    #t))
	;;(mt:test-set-state-status-by-id run-id test-id new-state new-status new-comment)))

(define (mt:lazy-read-test-config test-name)
  (let ((tconf (hash-table-ref/default *testconfigs* test-name #f)))
    (if tconf

;;======================================================================
;;  S U P P O R T   F U N C T I O N S
;;======================================================================

;; if a server is either running or in the process of starting call client:setup
;; else return #f to let the calling proc know that there is no server available
;;
(define (rmt:get-connection-info area-dat areapath) ;; TODO: push areapath down.
  (let ((cinfo (remote-conndat area-dat))
        (run-id 0))
    (if cinfo
	cinfo
	(if (server:check-if-running area-dat areapath)
	    (client:setup areapath)
	    #f))))

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

;; RA => e.g. usage (rmt:send-receive area-dat 'get-var #f (list varname))
;;
(define (rmt:send-receive area-dat cmd rid params #!key (attemptnum 1)) ;; start attemptnum at 1 so the modulo below works as expected

  ;; do all the prep locked under the rmt-mutex
  (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 area-dat
  ;; 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
    (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))
     ;; reset the connection if it has been unused too long
     ((and area-dat
           (remote-conndat area-dat)
	   (let ((expire-time (+ (- start-time (remote-server-timeout area-dat))(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 area-dat)) expire-time)))
      (debug:print-info 12 *default-log-port* "rmt:send-receive, case  8")
      (remote-conndat-set! area-dat #f)
      (mutex-unlock! *rmt-mutex*)
      (rmt:send-receive area-dat cmd rid params attemptnum: attemptnum))
     ;; ensure we have a record for our connection for given area
     ((not area-dat)
      (print "ERROR!!!!!!! SHOULD NEVER GET HERE NOW.")
      (set! area-dat (make-remote))
      (mutex-unlock! *rmt-mutex*)
      (debug:print-info 12 *default-log-port* "rmt:send-receive, case  1")
      (rmt:send-receive area-dat cmd rid params attemptnum: attemptnum))
     ;; ensure we have a homehost record
     ((not (pair? (remote-hh-dat area-dat)))  ;; not on homehost
      (thread-sleep! 0.1) ;; since we shouldn't get here, delay a little
      (remote-hh-dat-set! area-dat (common:get-homehost))
      (mutex-unlock! *rmt-mutex*)
      (debug:print-info 12 *default-log-port* "rmt:send-receive, case  2")
      (rmt:send-receive area-dat cmd rid params attemptnum: attemptnum))
     ;; on homehost and this is a read
     ((and (cdr (remote-hh-dat area-dat))   ;; on homehost
           (member cmd api:read-only-queries)) ;; this is a read
      (mutex-unlock! *rmt-mutex*)
      (debug:print-info 12 *default-log-port* "rmt:send-receive, case  3")
      (rmt:open-qry-close-locally area-dat cmd 0 params))

     ;; on homehost and this is a write, we already have a server, but server has died
     ((and (cdr (remote-hh-dat area-dat))         ;; on homehost
           (not (member cmd api:read-only-queries))  ;; this is a write
           (remote-server-url area-dat)           ;; have a server
           (not (server:check-if-running area-dat *toppath*)))  ;; server has died.
      (remote-server-url-set! area-dat #f) 
      (mutex-unlock! *rmt-mutex*)
      (debug:print-info 12 *default-log-port* "rmt:send-receive, case  4.1")
      (rmt:send-receive area-dat cmd rid params attemptnum: attemptnum))

     ;; on homehost and this is a write, we already have a server
     ((and (cdr (remote-hh-dat area-dat))         ;; on homehost
           (not (member cmd api:read-only-queries))  ;; this is a write
           (remote-server-url area-dat))          ;; have a server
      (mutex-unlock! *rmt-mutex*)
      (debug:print-info 12 *default-log-port* "rmt:send-receive, case  4")
      (rmt:open-qry-close-locally area-dat cmd 0 params))

     ;;  on homehost, no server contact made and this is a write, passively start a server 
     ((and (cdr (remote-hh-dat area-dat)) ; new
           (not (remote-server-url area-dat))
	   (not (member cmd api:read-only-queries)))
      (debug:print-info 12 *default-log-port* "rmt:send-receive, case  5")
      (let ((server-url  (server:check-if-running area-dat *toppath*))) ;; (server:read-dotserver->url *toppath*))) ;; (server:check-if-running *toppath*))) ;; Do NOT want to run server:check-if-running - very expensive to do for every write call
	(if server-url
	    (remote-server-url-set! area-dat server-url) ;; the string can be consumed by the client setup if needed
	    (server:kind-run *toppath*)))
      (mutex-unlock! *rmt-mutex*)
      (debug:print-info 12 *default-log-port* "rmt:send-receive, case  5.1")
      (rmt:open-qry-close-locally area-dat cmd 0 params))

     ((and (not (cdr (remote-hh-dat area-dat)))        ;; not on a homehost 
           (not (remote-conndat area-dat)))            ;; and no connection
      (debug:print-info 12 *default-log-port* "rmt:send-receive, case  6  hh-dat: " (remote-hh-dat area-dat) " conndat: " (remote-conndat area-dat))
      (mutex-unlock! *rmt-mutex*)
      (server:start-and-wait area-dat *toppath*)
      (remote-conndat-set! area-dat (rmt:get-connection-info area-dat *toppath*)) ;; calls client:setup which calls client:setup-http
      (rmt:send-receive area-dat cmd rid params attemptnum: attemptnum)) ;; TODO: add back-off timeout as
     ;; all set up if get this far, dispatch the query
     ((cdr (remote-hh-dat area-dat)) ;; we are on homehost
      (mutex-unlock! *rmt-mutex*)
      (debug:print-info 12 *default-log-port* "rmt:send-receive, case  7")
      (rmt:open-qry-close-locally area-dat cmd (if rid rid 0) params))

     ;; not on homehost, do server query
     (else
      (mutex-unlock! *rmt-mutex*)
      (debug:print-info 12 *default-log-port* "rmt:send-receive, case  9")
      (mutex-lock! *rmt-mutex*)
      (let* ((conninfo (remote-conndat area-dat))
	     (dat      (case (remote-transport area-dat)
			 ((http) (condition-case ;; handling here has caused a lot of problems. However it is needed to deal with attemtped communication to servers that have gone away
                                  (http-transport:client-api-send-receive 0 conninfo cmd params)
                                  ((commfail)(vector #f "communications fail"))
                                  ((exn)(vector #f "other fail" (print-call-chain)))))
			 (else
			  (debug:print 0 *default-log-port* "ERROR: transport " (remote-transport area-dat) " not supported")
			  (exit))))
	     (success  (if (vector? dat) (vector-ref dat 0) #f))
	     (res      (if (vector? dat) (vector-ref dat 1) #f)))
	(if (vector? conninfo)(http-transport:server-dat-update-last-access conninfo)) ;; refresh access time
	;; (mutex-unlock! *rmt-mutex*)
        (debug:print-info 12 *default-log-port* "rmt:send-receive, case  9. conninfo=" conninfo " dat=" dat " area-dat = "area-dat)
	(if success
	    (case (remote-transport area-dat)
	      ((http)
	       (mutex-unlock! *rmt-mutex*)
	       res)
	      (else
	       (debug:print 0 *default-log-port* "ERROR: transport " (remote-transport area-dat) " is unknown")
	       (mutex-unlock! *rmt-mutex*)
	       (exit 1)))
	    (begin
	      (debug:print 0 *default-log-port* "WARNING: communication failed. Trying again, try num: " attemptnum)
	      (remote-conndat-set!    area-dat #f)
	      (remote-server-url-set! area-dat #f)
              (debug:print-info 12 *default-log-port* "rmt:send-receive, case  9.1")
	      (mutex-unlock! *rmt-mutex*)
	      (server:start-and-wait area-dat *toppath*)
	      (rmt:send-receive area-dat cmd rid params attemptnum: (+ attemptnum 1)))))))))

;; (define (rmt:update-db-stats area-dat run-id rawcmd params duration)
;;   (mutex-lock! *db-stats-mutex*)
;;   (handle-exceptions
;;    exn
;;    (begin
;;      (debug:print 0 *default-log-port* "WARNING: stats collection failed in update-db-stats")
;;      (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
;;      (print "exn=" (condition->list exn))
;;      #f) ;; if this fails we don't care, it is just stats
;;    (let* ((cmd      (conc "run-id=" run-id " " (if (eq? rawcmd 'general-call) (car params) rawcmd)))
;; 	  (stat-vec (hash-table-ref/default *db-stats* cmd #f)))
;;      (if (not (vector? stat-vec))
;; 	 (let ((newvec (vector 0 0)))
;; 	   (hash-table-set! *db-stats* cmd newvec)
;; 	   (set! stat-vec newvec)))
;;      (vector-set! stat-vec 0 (+ (vector-ref stat-vec 0) 1))
;;      (vector-set! stat-vec 1 (+ (vector-ref stat-vec 1) duration))))
;;   (mutex-unlock! *db-stats-mutex*))

(define (rmt:print-db-stats area-dat)
  (let ((fmtstr "~40a~7-d~9-d~20,2-f")) ;; "~20,2-f"
    (debug:print 18 *default-log-port* "DB Stats\n========")
    (debug:print 18 *default-log-port* (format #f "~40a~8a~10a~10a" "Cmd" "Count" "TotTime" "Avg"))
    (for-each (lambda (cmd)
		(let ((cmd-dat (hash-table-ref *db-stats* cmd)))
		  (debug:print 18 *default-log-port* (format #f fmtstr cmd (vector-ref cmd-dat 0) (vector-ref cmd-dat 1) (/ (vector-ref cmd-dat 1)(vector-ref cmd-dat 0))))))
	      (sort (hash-table-keys *db-stats*)
		    (lambda (a b)
		      (> (vector-ref (hash-table-ref *db-stats* a) 0)
			 (vector-ref (hash-table-ref *db-stats* b) 0)))))))

(define (rmt:get-max-query-average area-dat run-id)
  (mutex-lock! *db-stats-mutex*)
  (let* ((runkey (conc "run-id=" run-id " "))
	 (cmds   (filter (lambda (x)
			   (substring-index runkey x))
			 (hash-table-keys *db-stats*)))
	 (res    (if (null? cmds)
		     (cons 'none 0)

			     (if (> tot 10)
				 (cons newmax-cmd currmax)
				 (cons 'none 0))
			     (loop (car tal)(cdr tal) newmax-cmd currmax)))))))
    (mutex-unlock! *db-stats-mutex*)
    res))

(define (rmt:open-qry-close-locally area-dat cmd run-id params #!key (remretries 5))
  (let* ((qry-is-write   (not (member cmd api:read-only-queries)))
	 (db-file-path   (db:dbfile-path)) ;;  0))
	 (dbstruct-local (db:setup))  ;; make-dbr:dbstruct path:  dbdir local: #t)))
	 (read-only      (not (file-write-access? db-file-path)))
	 (start          (current-milliseconds))
	 (resdat         (if (not (and read-only qry-is-write))
			     (api:execute-requests dbstruct-local (vector (symbol->string cmd) params))
			     (vector #t '())))
	 (success        (vector-ref resdat 0))
	 (res            (vector-ref resdat 1))
	 (duration       (- (current-milliseconds) start)))
    (if (and read-only qry-is-write)
        (debug:print 0 *default-log-port* "ERROR: attempt to write to read-only database ignored. cmd=" cmd))
    (if (not success)
	(if (> remretries 0)
	    (begin
	      (debug:print-error 0 *default-log-port* "local query failed. Trying again.")
	      (thread-sleep! (/ (random 5000) 1000)) ;; some random delay 
	      (rmt:open-qry-close-locally area-dat cmd run-id params remretries: (- remretries 1)))
	    (begin
	      (debug:print-error 0 *default-log-port* "too many retries in rmt:open-qry-close-locally, giving up")
	      #f))
	(begin
	  ;; (rmt:update-db-stats area-dat run-id cmd params duration)
	  ;; mark this run as dirty if this was a write, the watchdog is responsible for syncing it
	  (if qry-is-write
	      (let ((start-time (current-seconds)))
		(mutex-lock! *db-multi-sync-mutex*)
		(set! *db-last-access* start-time)  ;; THIS IS PROBABLY USELESS? (we are on a client)
                (mutex-unlock! *db-multi-sync-mutex*)))))
    res))

(define (rmt:send-receive-no-auto-client-setup area-dat connection-info cmd run-id params)
  (let* ((run-id   (if run-id run-id 0))
	 (res  	   (handle-exceptions
		    exn
		    #f
		    (http-transport:client-api-send-receive run-id connection-info cmd params))))
    (if (and res (vector-ref res 0))
	(vector-ref res 1) ;;; YES!! THIS IS CORRECT!! CHANGE IT HERE, THEN CHANGE rmt:send-receive ALSO!!!
	#f)))

;; ;; Wrap json library for strings (why the ports crap in the first place?)
;; (define (rmt:dat->json-str area-dat dat)
;;   (with-output-to-string 
;;     (lambda ()
;;       (json-write dat))))
;; 
;; (define (rmt:json-str->dat area-dat json-str)
;;   (with-input-from-string json-str
;;     (lambda ()
;;       (json-read))))

;;======================================================================
;;
;; A C T U A L   A P I   C A L L S  
;;
;;======================================================================

;;======================================================================
;;  S E R V E R
;;======================================================================

(define (rmt:kill-server area-dat run-id)
  (rmt:send-receive area-dat 'kill-server run-id (list run-id)))

(define (rmt:start-server area-dat run-id)
  (rmt:send-receive area-dat 'start-server 0 (list run-id)))

;;======================================================================
;;  M I S C
;;======================================================================

(define (rmt:login area-dat run-id)
  (rmt:send-receive area-dat 'login run-id (list *toppath* megatest-version *my-client-signature*)))

;; This login does no retries under the hood - it acts a bit like a ping.
;; Deprecated for nmsg-transport.
;;
(define (rmt:login-no-auto-client-setup area-dat connection-info)
  (case *transport-type* ;; run-id of 0 is just a placeholder
    ((http)(rmt:send-receive-no-auto-client-setup area-dat connection-info 'login 0 (list *toppath* megatest-version *my-client-signature*)))
    ;;((nmsg)(nmsg-transport:client-api-send-receive run-id connection-info 'login (list *toppath* megatest-version run-id *my-client-signature*)))
    ))

;; hand off a call to one of the db:queries statements
;; added run-id to make looking up the correct db possible 
;;
(define (rmt:general-call area-dat stmtname run-id . params)
  (rmt:send-receive area-dat 'general-call run-id (append (list stmtname run-id) params)))


;; given a hostname, return a pair of cpu load and update time representing latest intelligence from tests running on that host
(define (rmt:get-latest-host-load area-dat hostname)
  (rmt:send-receive area-dat 'get-latest-host-load 0 (list hostname)))

;; (define (rmt:sync-inmem->db area-dat run-id)
;;   (rmt:send-receive area-dat 'sync-inmem->db run-id '()))

(define (rmt:sdb-qry area-dat qry val run-id)
  ;; add caching if qry is 'getid or 'getstr
  (rmt:send-receive area-dat 'sdb-qry run-id (list qry val)))

;; NOT COMPLETED
(define (rmt:runtests area-dat user run-id testpatt params)
  (rmt:send-receive area-dat 'runtests run-id testpatt))

;;======================================================================
;;  T E S T   M E T A 
;;======================================================================

(define (rmt:get-tests-tags area-dat)
  (rmt:send-receive area-dat 'get-tests-tags #f '()))

;;======================================================================
;;  K E Y S 
;;======================================================================

;; These require run-id because the values come from the run!
;;
(define (rmt:get-key-val-pairs area-dat run-id)
  (rmt:send-receive area-dat 'get-key-val-pairs run-id (list run-id)))

(define (rmt:get-keys area-dat)
  (if *db-keys* *db-keys* 
     (let ((res (rmt:send-receive area-dat 'get-keys #f '())))
       (set! *db-keys* res)
       res)))

(define (rmt:get-keys-write area-dat) ;; dummy query to force server start
  (let ((res (rmt:send-receive area-dat 'get-keys-write #f '())))
    (set! *db-keys* res)
    res))

;; we don't reuse run-id's (except possibly *after* a db cleanup) so it is safe
;; to cache the resuls in a hash
;;
(define (rmt:get-key-vals area-dat run-id)
  (or (hash-table-ref/default *keyvals* run-id #f)
      (let ((res (rmt:send-receive area-dat 'get-key-vals #f (list run-id))))
        (hash-table-set! *keyvals* run-id res)
        res)))

(define (rmt:get-targets area-dat)
  (rmt:send-receive area-dat 'get-targets #f '()))

(define (rmt:get-target area-dat run-id)
  (rmt:send-receive area-dat 'get-target run-id (list run-id)))

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

;; Just some syntatic sugar
(define (rmt:register-test area-dat run-id test-name item-path)
  (rmt:general-call area-dat 'register-test run-id run-id test-name item-path))

(define (rmt:get-test-id area-dat run-id testname item-path)
  (rmt:send-receive area-dat 'get-test-id run-id (list run-id testname item-path)))

(define (rmt:get-test-info-by-id area-dat run-id test-id)
  (if (and (number? run-id)(number? test-id))
      (rmt:send-receive area-dat 'get-test-info-by-id run-id (list run-id test-id))
      (begin
	(debug:print 0 *default-log-port* "WARNING: Bad data handed to rmt:get-test-info-by-id run-id=" run-id ", test-id=" test-id)
	(print-call-chain (current-error-port))
	#f)))

(define (rmt:test-get-rundir-from-test-id area-dat run-id test-id)
  (rmt:send-receive area-dat 'test-get-rundir-from-test-id run-id (list run-id test-id)))

(define (rmt:open-test-db-by-test-id area-dat run-id test-id #!key (work-area #f))
  (let* ((test-path (if (string? work-area)
			work-area
			(rmt:test-get-rundir-from-test-id area-dat run-id test-id))))
    (debug:print 3 *default-log-port* "TEST PATH: " test-path)
    (open-test-db test-path)))

;; WARNING: This currently bypasses the transaction wrapped writes system
(define (rmt:test-set-state-status-by-id area-dat run-id test-id newstate newstatus newcomment)
  (rmt:send-receive area-dat 'test-set-state-status-by-id run-id (list run-id test-id newstate newstatus newcomment)))

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

(define (rmt:get-tests-for-run area-dat run-id testpatt states statuses offset limit not-in sort-by sort-order qryvals last-update mode)
  (if (number? run-id)
      (rmt:send-receive area-dat '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 area-dat run-id proc synckey keynum params)
  (rmt:send-receive area-dat 'synchash-get run-id (list run-id proc synckey keynum params)))

;; IDEA: Threadify these - they spend a lot of time waiting ...
;;
(define (rmt:get-tests-for-runs-mindata area-dat run-ids testpatt states status not-in)
  (let ((multi-run-mutex (make-mutex))
	(run-id-list (if run-ids
			 run-ids
			 (rmt:get-all-run-ids area-dat)))
	(result      '()))
    (if (null? run-id-list)
	'()
	(let loop ((hed     (car run-id-list))
		   (tal     (cdr run-id-list))
		   (threads '()))
	  (if (> (length threads) 5)
	      (loop hed tal (filter (lambda (th)(not (member (thread-state th) '(terminated dead)))) threads))
	      (let* ((newthread (make-thread
				 (lambda ()
				   (let ((res (rmt:send-receive area-dat 'get-tests-for-run-mindata hed (list hed testpatt states status not-in))))
				     (if (list? res)
					 (begin
					   (mutex-lock! multi-run-mutex)
					   (set! result (append result res))
					   (mutex-unlock! multi-run-mutex))
					 (debug:print-error 0 *default-log-port* "get-tests-for-run-mindata failed for run-id " hed ", testpatt " testpatt ", states " states ", status " status ", not-in " not-in))))
				 (conc "multi-run-thread for run-id " hed)))
		     (newthreads (cons newthread threads)))
		(thread-start! newthread)
		(thread-sleep! 0.05) ;; give that thread some time to start
		(if (null? tal)
		    newthreads
		    (loop (car tal)(cdr tal) newthreads))))))
    result))

;; ;; IDEA: Threadify these - they spend a lot of time waiting ...
;; ;;
;; (define (rmt:get-tests-for-runs-mindata area-dat run-ids testpatt states status not-in)
;;   (let ((run-id-list (if run-ids
;; 			 run-ids
;; 			 (rmt:get-all-run-ids area-dat))))
;;     (apply append (map (lambda (run-id)
;; 			 (rmt:send-receive area-dat 'get-tests-for-run-mindata run-id (list run-ids testpatt states status not-in)))
;; 		       run-id-list))))

(define (rmt:delete-test-records area-dat run-id test-id)
  (rmt:send-receive area-dat 'delete-test-records run-id (list run-id test-id)))

;; This is not needed as test steps are deleted on test delete call
;;
;; (define (rmt:delete-test-step-records area-dat run-id test-id)
;;   (rmt:send-receive area-dat 'delete-test-step-records run-id (list run-id test-id)))

(define (rmt:test-set-state-status area-dat run-id test-id state status msg)
  (rmt:send-receive area-dat 'test-set-state-status run-id (list run-id test-id state status msg)))

(define (rmt:test-toplevel-num-items area-dat run-id test-name)
  (rmt:send-receive area-dat 'test-toplevel-num-items run-id (list run-id test-name)))

;; (define (rmt:get-previous-test-run-record area-dat run-id test-name item-path)
;;   (rmt:send-receive area-dat 'get-previous-test-run-record run-id (list run-id test-name item-path)))

(define (rmt:get-matching-previous-test-run-records area-dat run-id test-name item-path)
  (rmt:send-receive area-dat 'get-matching-previous-test-run-records run-id (list run-id test-name item-path)))

(define (rmt:test-get-logfile-info area-dat run-id test-name)
  (rmt:send-receive area-dat 'test-get-logfile-info run-id (list run-id test-name)))

(define (rmt:test-get-records-for-index-file area-dat run-id test-name)
  (rmt:send-receive area-dat 'test-get-records-for-index-file run-id (list run-id test-name)))

(define (rmt:get-testinfo-state-status area-dat run-id test-id)
  (rmt:send-receive area-dat 'get-testinfo-state-status run-id (list run-id test-id)))

(define (rmt:test-set-log! area-dat run-id test-id logf)
  (if (string? logf)(rmt:general-call area-dat 'test-set-log run-id logf test-id)))

(define (rmt:test-set-top-process-pid area-dat run-id test-id pid)
  (rmt:send-receive area-dat 'test-set-top-process-pid run-id (list run-id test-id pid)))

(define (rmt:test-get-top-process-pid area-dat run-id test-id)
  (rmt:send-receive area-dat 'test-get-top-process-pid run-id (list run-id test-id)))

(define (rmt:get-run-ids-matching-target area-dat keynames target res runname testpatt statepatt statuspatt)
  (rmt:send-receive area-dat 'get-run-ids-matching-target #f (list keynames target res runname testpatt statepatt statuspatt)))

;; NOTE: This will open and access ALL run databases. 
;;
(define (rmt:test-get-paths-matching-keynames-target-new area-dat keynames target res testpatt statepatt statuspatt runname)
  (let ((run-ids (rmt:get-run-ids-matching-target area-dat keynames target res runname testpatt statepatt statuspatt)))
    (apply append 
	   (map (lambda (run-id)
		  (rmt:send-receive area-dat 'test-get-paths-matching-keynames-target-new run-id (list run-id keynames target res testpatt statepatt statuspatt runname)))
	   run-ids))))

;; (define (rmt:get-run-ids-matching area-dat keynames target res)
;;   (rmt:send-receive area-dat #f 'get-run-ids-matching (list keynames target res)))

(define (rmt:get-prereqs-not-met area-dat run-id waitons ref-test-name ref-item-path #!key (mode '(normal))(itemmaps #f))
  (rmt:send-receive area-dat 'get-prereqs-not-met run-id (list run-id waitons ref-test-name ref-item-path mode itemmaps)))

(define (rmt:get-count-tests-running-for-run-id area-dat run-id)
  (rmt:send-receive area-dat 'get-count-tests-running-for-run-id run-id (list run-id)))

;; Statistical queries

(define (rmt:get-count-tests-running area-dat run-id)
  (rmt:send-receive area-dat 'get-count-tests-running run-id (list run-id)))

(define (rmt:get-count-tests-running-for-testname area-dat run-id testname)
  (rmt:send-receive area-dat 'get-count-tests-running-for-testname run-id (list run-id testname)))

(define (rmt:get-count-tests-running-in-jobgroup area-dat run-id jobgroup)
  (rmt:send-receive area-dat 'get-count-tests-running-in-jobgroup run-id (list run-id jobgroup)))

;; state and status are extra hints not usually used in the calculation
;;
(define (rmt:set-state-status-and-roll-up-items area-dat run-id test-name item-path state status comment)
  (rmt:send-receive area-dat 'set-state-status-and-roll-up-items run-id (list run-id test-name item-path state status comment)))

(define (rmt:update-pass-fail-counts area-dat run-id test-name)
  (rmt:general-call area-dat 'update-pass-fail-counts run-id test-name test-name test-name))

(define (rmt:top-test-set-per-pf-counts area-dat run-id test-name)
  (rmt:send-receive area-dat 'top-test-set-per-pf-counts run-id (list run-id test-name)))

(define (rmt:get-raw-run-stats area-dat run-id)
  (rmt:send-receive area-dat 'get-raw-run-stats run-id (list run-id)))

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

(define (rmt:get-run-info area-dat run-id)
  (rmt:send-receive area-dat 'get-run-info run-id (list run-id)))

(define (rmt:get-num-runs area-dat runpatt)
  (rmt:send-receive area-dat 'get-num-runs #f (list runpatt)))

;; Use the special run-id == #f scenario here since there is no run yet
(define (rmt:register-run area-dat keyvals runname state status user)
  (rmt:send-receive area-dat 'register-run #f (list keyvals runname state status user)))
    
(define (rmt:get-run-name-from-id area-dat run-id)
  (rmt:send-receive area-dat 'get-run-name-from-id run-id (list run-id)))

(define (rmt:delete-run area-dat run-id)
  (rmt:send-receive area-dat 'delete-run run-id (list run-id)))

(define (rmt:update-run-stats area-dat run-id stats)
  (rmt:send-receive area-dat 'update-run-stats #f (list run-id stats)))

(define (rmt:delete-old-deleted-test-records area-dat)
  (rmt:send-receive area-dat 'delete-old-deleted-test-records #f '()))

(define (rmt:get-runs area-dat runpatt count offset keypatts)
  (rmt:send-receive area-dat 'get-runs #f (list runpatt count offset keypatts)))

(define (rmt:get-all-run-ids area-dat)
  (rmt:send-receive area-dat 'get-all-run-ids #f '()))

(define (rmt:get-prev-run-ids area-dat run-id)
  (rmt:send-receive area-dat 'get-prev-run-ids #f (list run-id)))

(define (rmt:lock/unlock-run area-dat run-id lock unlock user)
  (rmt:send-receive area-dat 'lock/unlock-run #f (list run-id lock unlock user)))

;; set/get status
(define (rmt:get-run-status area-dat run-id)
  (rmt:send-receive area-dat 'get-run-status #f (list run-id)))

(define (rmt:set-run-status area-dat run-id run-status #!key (msg #f))
  (rmt:send-receive area-dat 'set-run-status #f (list run-id run-status msg)))

(define (rmt:update-run-event_time area-dat run-id)
  (rmt:send-receive area-dat 'update-run-event_time #f (list run-id)))

(define (rmt:get-runs-by-patt area-dat  keys runnamepatt targpatt offset limit fields last-runs-update) ;; fields of #f uses default
  (rmt:send-receive area-dat 'get-runs-by-patt #f (list keys runnamepatt targpatt offset limit fields last-runs-update)))

(define (rmt:find-and-mark-incomplete area-dat run-id ovr-deadtime)
  ;; (if (rmt:send-receive area-dat 'have-incompletes? run-id (list run-id ovr-deadtime))
  (rmt:send-receive area-dat 'mark-incomplete run-id (list run-id ovr-deadtime))) ;; )

(define (rmt:get-main-run-stats area-dat run-id)
  (rmt:send-receive area-dat 'get-main-run-stats #f (list run-id)))

(define (rmt:get-var area-dat varname)
  (rmt:send-receive area-dat 'get-var #f (list varname)))

(define (rmt:set-var area-dat varname value)
  (rmt:send-receive area-dat 'set-var #f (list varname value)))

;;======================================================================
;; M U L T I R U N   Q U E R I E S
;;======================================================================

;; Need to move this to multi-run section and make associated changes
(define (rmt:find-and-mark-incomplete-all-runs area-dat #!key (ovr-deadtime #f))
  (let ((run-ids (rmt:get-all-run-ids area-dat)))
    (for-each (lambda (run-id)
	       (rmt:find-and-mark-incomplete area-dat run-id ovr-deadtime))
	     run-ids)))

;; get the previous record for when this test was run where all keys match but runname
;; returns #f if no such test found, returns a single test record if found
;; 
;; Run this at the client end since we have to connect to multiple run-id dbs
;;
(define (rmt:get-previous-test-run-record area-dat run-id test-name item-path)
  (let* ((keyvals (rmt:get-key-val-pairs area-dat run-id))
	 (keys    (rmt:get-keys area-dat))
	 (selstr  (string-intersperse  keys ","))
	 (qrystr  (string-intersperse (map (lambda (x)(conc x "=?")) keys) " AND ")))
    (if (not keyvals)
	#f
	(let ((prev-run-ids (rmt:get-prev-run-ids area-dat run-id)))
	  ;; for each run starting with the most recent look to see if there is a matching test
	  ;; if found then return that matching test record
	  (debug:print 4 *default-log-port* "selstr: " selstr ", qrystr: " qrystr ", keyvals: " keyvals ", previous run ids found: " prev-run-ids)
	  (if (null? prev-run-ids) #f
	      (let loop ((hed (car prev-run-ids))
			 (tal (cdr prev-run-ids)))
		(let ((results (rmt:get-tests-for-run area-dat hed (conc test-name "/" item-path) '() '() ;; run-id testpatt states statuses
						      #f #f #f               ;; offset limit not-in hide/not-hide
						      #f #f #f #f 'normal))) ;; sort-by sort-order qryvals last-update mode
		  (debug:print 4 *default-log-port* "Got tests for run-id " run-id ", test-name " test-name ", item-path " item-path ": " results)
		  (if (and (null? results)
			   (not (null? tal)))
		      (loop (car tal)(cdr tal))
		      (if (null? results) #f
			  (car results))))))))))

(define (rmt:get-run-stats area-dat)
  (rmt:send-receive area-dat 'get-run-stats #f '()))

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

;; Getting steps is more complicated.
;;
;; If given work area 
;;  1. Find the testdat.db file
;;  2. Open the testdat.db file and do the query
;; If not given the work area
;;  1. Do a remote call to get the test path
;;  2. Continue as above
;; 
;;(define (rmt:get-steps-for-test area-dat run-id test-id)
;;  (rmt:send-receive area-dat 'get-steps-data run-id (list test-id)))

(define (rmt:teststep-set-status! area-dat run-id test-id teststep-name state-in status-in comment logfile)
  (let* ((state     (items:check-valid-items "state" state-in))
	 (status    (items:check-valid-items "status" status-in)))
    (if (or (not state)(not status))
	(debug:print 3 *default-log-port* "WARNING: Invalid " (if status "status" "state")
		     " value \"" (if status state-in status-in) "\", update your validvalues section in megatest.config"))
    (rmt:send-receive area-dat 'teststep-set-status! run-id (list run-id test-id teststep-name state-in status-in comment logfile))))

(define (rmt:get-steps-for-test area-dat run-id test-id)
  (rmt:send-receive area-dat 'get-steps-for-test run-id (list run-id test-id)))

;;======================================================================
;;  T E S T   D A T A 
;;======================================================================

(define (rmt:read-test-data area-dat run-id test-id categorypatt #!key (work-area #f)) 
  (rmt:send-receive area-dat 'read-test-data run-id (list run-id test-id categorypatt)))
;;   (let ((tdb  (rmt:open-test-db-by-test-id area-dat run-id test-id work-area: work-area)))
;;     (if tdb
;; 	(tdb:read-test-data tdb test-id categorypatt)
;; 	'())))

(define (rmt:testmeta-add-record area-dat testname)
  (rmt:send-receive area-dat 'testmeta-add-record #f (list testname)))

(define (rmt:testmeta-get-record area-dat testname)
  (rmt:send-receive area-dat 'testmeta-get-record #f (list testname)))

(define (rmt:testmeta-update-field area-dat test-name fld val)
  (rmt:send-receive area-dat 'testmeta-update-field #f (list test-name fld val)))

(define (rmt:test-data-rollup area-dat run-id test-id status)
  (rmt:send-receive area-dat 'test-data-rollup run-id (list run-id test-id status)))

(define (rmt:csv->test-data area-dat run-id test-id csvdata)
  (rmt:send-receive area-dat 'csv->test-data run-id (list run-id test-id csvdata)))

;;======================================================================
;;  T A S K S
;;======================================================================

(define (rmt:tasks-find-task-queue-records area-dat target run-name test-patt state-patt action-patt)
  (rmt:send-receive area-dat 'find-task-queue-records #f (list target run-name test-patt state-patt action-patt)))

(define (rmt:tasks-add area-dat action owner target runname testpatt params)
  (rmt:send-receive area-dat 'tasks-add #f (list action owner target runname testpatt params)))

(define (rmt:tasks-set-state-given-param-key area-dat param-key new-state)
  (rmt:send-receive area-dat 'tasks-set-state-given-param-key #f (list  param-key new-state)))

(define (rmt:tasks-get-last area-dat target runname)
  (rmt:send-receive area-dat 'tasks-get-last #f (list target runname)))

;;======================================================================
;; A R C H I V E S
;;======================================================================

(define (rmt:archive-get-allocations area-dat  testname itempath dneeded)
  (rmt:send-receive area-dat 'archive-get-allocations #f (list testname itempath dneeded)))

(define (rmt:archive-register-block-name area-dat bdisk-id archive-path)
  (rmt:send-receive area-dat 'archive-register-block-name #f (list bdisk-id archive-path)))

(define (rmt:archive-allocate-testsuite/area-to-block area-dat block-id testsuite-name areakey)
  (rmt:send-receive area-dat 'archive-allocate-test-to-block #f (list  block-id testsuite-name areakey)))

(define (rmt:archive-register-disk area-dat bdisk-name bdisk-path df)
  (rmt:send-receive area-dat 'archive-register-disk #f (list bdisk-name bdisk-path df)))

(define (rmt:test-set-archive-block-id area-dat run-id test-id archive-block-id)
  (rmt:send-receive area-dat 'test-set-archive-block-id run-id (list run-id test-id archive-block-id)))

(define (rmt:test-get-archive-block-info area-dat archive-block-id)
  (rmt:send-receive area-dat 'test-get-archive-block-info #f (list archive-block-id)))

   ;;    (apply (eval (string->symbol (conc "remote:" procstr))) params)
   (apply (eval (string->symbol procstr)) params)))

;; all routes though here end in exit ...
;;
;; start_server? 
;;
(define (rpc-transport:launch area-dat run-id)
  (let* ((tdbdat (tasks:open-db)))
    (BB> "rpc-transport:launch fired for run-id="run-id)
    (set! *run-id*   run-id)
    (if (args:get-arg "-daemonize")
        (daemon:ize))
    (if (server:check-if-running area-dat run-id)
        (begin
          (debug:print 0 *default-log-port* "INFO: Server for run-id " run-id " already running")
          (exit 0)))
    (let loop ((server-id (tasks:server-lock-slot (db:delay-if-busy tdbdat) run-id))
               (remtries  4))
      (if (not server-id)
          (if (> remtries 0)

(defstruct runs:testdat
  hed tal reg reruns  test-record
  test-name item-path jobgroup
  waitons testmode  newtal itemmaps prereqs-not-met)

;; set up needed environment variables given a run-id and optionally a target, itempath etc.
;;
(define (runs:set-megatest-env-vars area-dat run-id #!key (inkeys #f)(inrunname #f)(inkeyvals #f)(intarget #f)(testname #f)(itempath #f))
  (let* ((target    (or intarget 
			(common:args-get-target)
			(get-environment-variable "MT_TARGET")))
	 (keys      (if inkeys    inkeys    (rmt:get-keys area-dat)))
	 (keyvals   (if inkeyvals inkeyvals (keys:target->keyval keys target)))
	 (vals      (hash-table-ref/default *env-vars-by-run-id* run-id #f))
	 (link-tree (configf:lookup *configdat* "setup" "linktree")))
    (if testname (setenv "MT_TEST_NAME" testname))
    (if itempath (setenv "MT_ITEMPATH"  itempath))

    ;; get the info from the db and put it in the cache

     vals
     (lambda (key val)
       (debug:print 2 *default-log-port* "setenv " key " " val)
       (safe-setenv key val)))
    (if (not (get-environment-variable "MT_TARGET"))(setenv "MT_TARGET" target))
    (alist->env-vars (hash-table-ref/default *configdat* "env-override" '()))
    ;; Lets use this as an opportunity to put MT_RUNNAME in the environment
    (let ((runname  (if inrunname inrunname (rmt:get-run-name-from-id area-dat run-id))))
      (if runname
	  (setenv "MT_RUNNAME" runname)
	  (debug:print-error 0 *default-log-port* "no value for runname for id " run-id)))
    (setenv "MT_RUN_AREA_HOME" *toppath*)
    ;; if a testname and itempath are available set the remaining appropriate variables
    (if testname (setenv "MT_TEST_NAME" testname))
    (if itempath (setenv "MT_ITEMPATH"  itempath))

	    itemdat))

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

(define (runs:shrink-can-run-more-tests-count runsdat)
  (runs:dat-can-run-more-tests-count-set! runsdat 0))

(define (runs:inc-can-run-more-tests-count runsdat)
  (runs:dat-can-run-more-tests-count-set!
   runsdat
   (+ (runs:dat-can-run-more-tests-count runsdat) 1)))



;; Temporary globals. Move these into the logic or into common
;;
(define *seen-cant-run-tests* (make-hash-table)) ;; use to track tests that we suspect cannot be run
(define (runs:inc-cant-run-tests testname)
  (hash-table-set! *seen-cant-run-tests* testname
		   (+ (hash-table-ref/default *seen-cant-run-tests* testname 0) 1)))

	(currtime (current-seconds)))
    (if (> (- currtime lasttime) waitval)
	(begin
	  (hash-table-set! *runs:denoise* key currtime)
	  #t)
	#f)))

(define (runs:can-run-more-tests area-dat runsdat run-id jobgroup max-concurrent-jobs)
  ;; Take advantage of a good place to exit if running the one-pass methodology
  (if (and (> (runs:dat-can-run-more-tests-count runsdat) 20)
	   (args:get-arg "-one-pass"))
      (exit 0))
  (thread-sleep! (cond
        	  ((> (runs:dat-can-run-more-tests-count runsdat) 20)
		   (if (runs:lownoise "waiting on tasks" 60)(debug:print-info 2 *default-log-port* "waiting for tasks to complete, sleeping briefly ..."))
		   2);; obviously haven't had any work to do for a while
        	  (else 0)))
  (let* ((num-running             (rmt:get-count-tests-running area-dat run-id))
	 (num-running-in-jobgroup (rmt:get-count-tests-running-in-jobgroup area-dat run-id jobgroup))
	 (job-group-limit         (let ((jobg-count (config-lookup *configdat* "jobgroups" jobgroup)))
				    (if (string? jobg-count)
					(string->number jobg-count)
					jobg-count))))
    (if (> (+ num-running num-running-in-jobgroup) 0)
	(runs:inc-can-run-more-tests-count runsdat))
    (if (not (eq? *last-num-running-tests* num-running))
	(begin
	  (debug:print 2 *default-log-port* "max-concurrent-jobs: " max-concurrent-jobs ", num-running: " num-running)
	  (set! *last-num-running-tests* num-running)))
    (if (not (eq? 0 *globalexitstatus*))
	(list #f num-running num-running-in-jobgroup max-concurrent-jobs job-group-limit)
	(let ((can-not-run-more (cond

	  (list (not can-not-run-more) num-running num-running-in-jobgroup max-concurrent-jobs job-group-limit)))))


;;  test-names: Comma separated patterns same as test-patts but used in selection 
;;              of tests to run. The item portions are not respected.
;;              FIXME: error out if /patt specified
;;            
(define (runs:run-tests area-dat 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 area-dat keyvals runname "new" "n/a" user))  ;;  test-name)))
	 ;; (deferred          '()) ;; delay running these since they have a waiton clause
	 (runconfigf         (conc  *toppath* "/runconfigs.config"))
	 (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))

    (let ((sighand (lambda (signum)
		     ;; (signal-mask! signum) ;; to mask or not? seems to cause issues in exiting
		     (set! *time-to-exit* #t)
		     (print "Received signal " signum ", cleaning up before exit. Please wait...")
		     (let ((th1 (make-thread (lambda ()
					       (let ((tdbdat (tasks:open-db)))
						 (rmt:tasks-set-state-given-param-key area-dat task-key "killed"))
					       (print "Killed by signal " signum ". Exiting")
					       (thread-sleep! 3)
					       (exit))))
			   (th2 (make-thread (lambda ()
					       (thread-sleep! 5)
					       (debug:print 0 *default-log-port* "Done")
					       (exit 4)))))
		       (thread-start! th2)
		       (thread-start! th1)
		       (thread-join! th2)))))
      (set-signal-handler! signal/int sighand)
      (set-signal-handler! signal/term sighand))

    (runs:set-megatest-env-vars area-dat run-id inkeys: keys inrunname: runname) ;; these may be needed by the launching process
    (set! runconf (if (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)))

    ;; register this run in monitor.db
    (rmt:tasks-add area-dat "run-tests" user target runname test-patts task-key) ;; params)
    (rmt:tasks-set-state-given-param-key area-dat task-key "running")

    ;; Now generate all the tests lists
    (set! all-tests-registry (tests:get-all))   ;; hash of testname => path-to-test
    (set! all-test-names     (hash-table-keys all-tests-registry))
    (set! test-names         (tests:filter-test-names all-test-names test-patts))

    ;; I think seeding required-tests with all test-names makes sense but lack analysis to back that up.

	  ;; should not cause problems here.
	  ;;
	  ;; have to delete test records where NOT_STARTED since they can cause -keepgoing to 
	  ;; get stuck due to becoming inaccessible from a failed test. I.e. if test B depends 
	  ;; on test A but test B reached the point on being registered as NOT_STARTED and test
	  ;; A failed for some reason then on re-run using -keepgoing the run can never complete.
	  ;;
	  ;; (rmt:general-call area-dat 'delete-tests-in-state run-id "NOT_STARTED")
	  
	  ;; Now convert anything in allow-auto-rerun to NOT_STARTED
	  ;;
	  (for-each (lambda (state)
		      (rmt:set-tests-state-status area-dat run-id test-names state #f "NOT_STARTED" state))
		    (string-split (or (configf:lookup *configdat* "setup" "allow-auto-rerun") "")))))

    ;; Ensure all tests are registered in the test_meta table
    (runs:update-all-test_meta area-dat)

    ;; now add non-directly referenced dependencies (i.e. waiton)
    ;;======================================================================
    ;; refactoring this block into tests:get-full-data
    ;;
    ;; What happended, this code is now duplicated in tests!?
    ;;

    ;; NOTE: these are all parent tests, items are not expanded yet.
    (debug:print-info 4 *default-log-port* "test-records=" (hash-table->alist test-records))
    (let ((reglen (configf:lookup *configdat* "setup" "runqueue")))
      (if (> (length (hash-table-keys test-records)) 0)
	  (let* ((keep-going        #t)
		 (run-queue-retries 5)
		 (th1        (make-thread (lambda ()
					    (runs:run-tests-queue area-dat run-id runname test-records keyvals flags test-patts required-tests (any->number reglen) all-tests-registry))










					  "runs:run-tests-queue"))
		 (th2        (make-thread (lambda ()				    
					    ;; (rmt:find-and-mark-incomplete-all-runs area-dat))))) CAN'T INTERRUPT IT ...
					    (let ((run-ids (rmt:get-all-run-ids area-dat)))
					      (for-each (lambda (run-id)
							  (if keep-going
							      (handle-exceptions
							       exn
							       (debug:print 0 *default-log-port* "error in calling find-and-mark-incomplete for run-id " run-id)
							       (rmt:find-and-mark-incomplete area-dat run-id #f)))) ;; ovr-deadtime)))
							run-ids)))
					  "runs: mark-incompletes")))
	    (thread-start! th1)
	    (thread-start! th2)
	    (thread-join! th1)
	    (set! keep-going #f)
	    (thread-join! th2)
	    ;; if run-count > 0 call, set -preclean and -rerun STUCK/DEAD
	    (if (> run-count 0) ;; handle reruns
		(begin
		  (if (not (hash-table-ref/default flags "-preclean" #f))
		      (hash-table-set! flags "-preclean" #t))
		  (if (not (hash-table-ref/default flags "-rerun" #f))
		      (hash-table-set! flags "-rerun" "STUCK/DEAD,n/a,ZERO_ITEMS"))
		  ;; recursive call to self
		  (runs:run-tests area-dat target runname test-patts user flags run-count: (- run-count 1)))))
	  (debug:print-info 0 *default-log-port* "No tests to run")))
    (debug:print-info 4 *default-log-port* "All done by here")
    (rmt:tasks-set-state-given-param-key area-dat task-key "done")
    ;; (sqlite3:finalize! tasks-db)
    ))


;; loop logic. These are used in runs:run-tests-queue to make it a bit more readable.
;;
;; If reg not full and have items in tal then loop with (car tal)(cdr tal) reg reruns

      (cdr reg)
      (if (null? tal) ;; if tal is null and reg not full then '() as reg contents moved to tal
	  '()
	  reg)))

(define runs:nothing-left-in-queue-count 0)

(define (runs:expand-items area-dat hed tal reg reruns regfull newtal jobgroup max-concurrent-jobs run-id waitons item-path testmode test-record can-run-more items runname tconfig reglen test-registry test-records itemmaps)
  (let* ((loop-list       (list hed tal reg reruns))
	 (prereqs-not-met (let ((res (rmt:get-prereqs-not-met area-dat run-id waitons hed item-path mode: testmode itemmaps: itemmaps)))
			    (if (list? res)
				res
				(begin
				  (debug:print 0 *default-log-port*
					       "ERROR: rmt:get-prereqs-not-met returned non-list!\n"
					       "  res=" res " run-id=" run-id " waitons=" waitons " hed=" hed " item-path=" item-path " testmode=" testmode " itemmaps=" itemmaps)
				  '()))))

     ((or (null? prereqs-not-met)
	  (and (member 'toplevel testmode)
	       (null? non-completed)))
      (debug:print-info 4 *default-log-port* "runs:expand-items: (or (null? prereqs-not-met) (and (member 'toplevel testmode)(null? non-completed)))")
      (let ((test-name (tests:testqueue-get-testname test-record)))
	(setenv "MT_TEST_NAME" test-name) ;; 
	(setenv "MT_RUNNAME"   runname)
	(runs:set-megatest-env-vars area-dat run-id inrunname: runname) ;; these may be needed by the launching process
	(let ((items-list (items:get-items-from-config tconfig)))
	  (if (list? items-list)
	      (begin
		(if (null? items-list)
		    (let ((test-id   (rmt:get-test-id area-dat run-id test-name ""))
			  (num-items (rmt:test-toplevel-num-items area-dat run-id test-name)))
		      (if (and test-id
			       (not (> num-items 0)))
			  (mt:test-set-state-status-by-id run-id test-id "NOT_STARTED" "ZERO_ITEMS" "Failed to run due to failed prerequisites"))))
		(tests:testqueue-set-items! test-record items-list)
		(list hed tal reg reruns))
	      (begin
		(debug:print-error 0 *default-log-port* "The proc from reading the items table did not yield a list - please report this")

	(if (and give-up
		 (not (and (null? tal)(null? reg))))
	    (let ((trimmed-tal (mt:discard-blocked-tests run-id hed tal test-records))
		  (trimmed-reg (mt:discard-blocked-tests run-id hed reg test-records)))
	      (debug:print 1 *default-log-port* "WARNING: test " hed " has discarded prerequisites, removing it from the queue")

	      (let ((test-id (rmt:get-test-id area-dat run-id hed "")))
		(if test-id (mt:test-set-state-status-by-id run-id test-id "NOT_STARTED" "PREQ_DISCARDED" "Failed to run due to discarded prerequisites")))
	      
	      (if (and (null? trimmed-tal)
		       (null? trimmed-reg))
		  #f
		  (list (runs:queue-next-hed trimmed-tal trimmed-reg reglen regfull)
			(runs:queue-next-tal trimmed-tal trimmed-reg reglen regfull)

	    ;; getting here likely means the system is way overloaded, kill a full minute before continuing
	    (thread-sleep! 60)
	    ;; num-retries code was here
	    ;; we use this opportunity to move contents of reg to tal
	    (list (car newtal)(append (cdr newtal) reg) '() reruns)) ;; an issue with prereqs not yet met?
	  (begin
	    (debug:print-info 1 *default-log-port* "no fails in prerequisites for " hed " but nothing seen running in a while, dropping test " hed " from the run queue")
	    (let ((test-id (rmt:get-test-id area-dat run-id hed "")))
	      (if test-id (mt:test-set-state-status-by-id run-id test-id "NOT_STARTED" "TIMED_OUT" "Nothing seen running in a while.")))
	    (list (runs:queue-next-hed tal reg reglen regfull)
		  (runs:queue-next-tal tal reg reglen regfull)
		  (runs:queue-next-reg tal reg reglen regfull)
		  reruns))))

     ((and 
       (or (not (null? fails))
	   (not (null? prereq-fails)))
       (member 'normal testmode))
      (debug:print-info 1 *default-log-port* "test "  hed " (mode=" testmode ") has failed prerequisite(s); "
			(string-intersperse (map (lambda (t)(conc (db:test-get-testname t) ":" (db:test-get-state t)"/"(db:test-get-status t))) fails) ", ")
			", removing it from to-do list")
      (let ((test-id (rmt:get-test-id area-dat run-id hed "")))
	(if test-id
	    (if (not (null? prereq-fails))
		(mt:test-set-state-status-by-id run-id test-id "NOT_STARTED" "PREQ_DISCARDED" "Failed to run due to prior failed prerequisites")
		(mt:test-set-state-status-by-id run-id test-id "NOT_STARTED" "PREQ_FAIL"      "Failed to run due to failed prerequisites"))))
      (if (or (not (null? reg))(not (null? tal)))
	  (begin
	    (hash-table-set! test-registry hed 'CANNOTRUN)

	       t)
	      (else 
	       (conc t))))
	   inlst)))


;;  hed tal reg reruns reglen regfull test-record runname test-name item-path jobgroup max-concurrent-jobs run-id waitons item-path testmode test-patts required-tests test-registry registry-mutex flags keyvals run-info newtal all-tests-registry itemmaps)
(define (runs:process-expanded-tests area-dat runsdat testdat)
  ;; unroll the contents of runsdat and testdat (due to ongoing refactoring).
  (let* ((hed                    (runs:testdat-hed testdat))
	 (tal                    (runs:testdat-tal testdat))
	 (reg                    (runs:testdat-reg testdat))
	 (reruns                 (runs:testdat-reruns testdat))
	 (test-name              (runs:testdat-test-name testdat))
	 (item-path              (runs:testdat-item-path testdat))

	 (test-registry          (runs:dat-test-registry runsdat))
	 (registry-mutex         (runs:dat-registry-mutex runsdat))
	 (flags                  (runs:dat-flags runsdat))
	 (keyvals                (runs:dat-keyvals runsdat))
	 (run-info               (runs:dat-run-info runsdat))
	 (all-tests-registry     (runs:dat-all-tests-registry runsdat))
	 (run-limits-info        (runs:dat-can-run-more-tests runsdat))

	 (have-resources         (car run-limits-info))
	 (num-running            (list-ref run-limits-info 1))
	 (num-running-in-jobgroup(list-ref run-limits-info 2)) 
	 (max-concurrent-jobs    (list-ref run-limits-info 3))
	 (job-group-limit        (list-ref run-limits-info 4))
	 ;; (prereqs-not-met        (rmt:get-prereqs-not-met area-dat run-id waitons hed item-path mode: testmode itemmaps: itemmaps))
	 ;; (prereqs-not-met         (mt:lazy-get-prereqs-not-met run-id waitons item-path mode: testmode itemmap: itemmap))
	 (fails                  (if (list? prereqs-not-met)
				      (runs:calc-fails prereqs-not-met)
				      (begin
					(debug:print-error 0 *default-log-port* "prereqs-not-met is not a list! " prereqs-not-met)
					'())))
	 (non-completed           (filter (lambda (x)             ;; remove hed from not completed list, duh, of course it is not completed!

     
     ;; Register tests 
     ;;
     ((not (hash-table-ref/default test-registry (db:test-make-full-name test-name item-path) #f))
      (debug:print-info 4 *default-log-port* "Pre-registering test " test-name "/" item-path " to create placeholder" )
      ;; always do firm registration now in v1.60 and greater ;; (eq? *transport-type* 'fs) ;; no point in parallel registration if use fs
      (let register-loop ((numtries 15))
	(rmt:register-test area-dat run-id test-name item-path)
	(if (rmt:get-test-id area-dat run-id test-name item-path)
	    (hash-table-set! test-registry (db:test-make-full-name test-name item-path) 'done)
	    (if (> numtries 0)
		(begin
		  (thread-sleep! 0.5)
		  (register-loop (- numtries 1)))
		(debug:print-error 0 *default-log-port* "failed to register test " (db:test-make-full-name test-name item-path)))))
      (if (not (eq? (hash-table-ref/default test-registry (db:test-make-full-name test-name "") #f) 'done))
	  (begin
	    (rmt:register-test area-dat run-id test-name "")
	    (if (rmt:get-test-id area-dat run-id test-name "")
		(hash-table-set! test-registry (db:test-make-full-name test-name "") 'done))))
      (runs:shrink-can-run-more-tests-count runsdat)   ;; DELAY TWEAKER (still needed?)
      (if (and (null? tal)(null? reg))
	  (list hed tal (append reg (list hed)) reruns)
	  (list (runs:queue-next-hed tal reg reglen regfull)
		(runs:queue-next-tal tal reg reglen regfull)
		;; NB// Here we are building reg as we register tests

      ;; we are going to reset all the counters for test retries by setting a new hash table
      ;; this means they will increment only when nothing can be run
      (set! *max-tries-hash* (make-hash-table))
      ;; well, first lets see if cpu load throttling is enabled. If so wait around until the
      ;; average cpu load is under the threshold before continuing
      (if (configf:lookup *configdat* "jobtools" "maxload") ;; only gate if maxload is specified
	  (common:wait-for-cpuload maxload numcpus waitdelay))
      (run:test area-dat run-id run-info keyvals runname test-record flags #f test-registry all-tests-registry)
      (runs:incremental-print-results area-dat run-id)
      (hash-table-set! test-registry (db:test-make-full-name test-name item-path) 'running)
      (runs:shrink-can-run-more-tests-count runsdat)  ;; DELAY TWEAKER (still needed?)
      ;; (thread-sleep! *global-delta*)
      (if (or (not (null? tal))(not (null? reg)))
	  (list (runs:queue-next-hed tal reg reglen regfull)
		(runs:queue-next-tal tal reg reglen regfull)
		(runs:queue-next-reg tal reg reglen regfull)

	    (list (car newtal)(cdr newtal) reg reruns))
	  ;; the waiton is FAIL so no point in trying to run hed ever again
	  (if (or (not (null? reg))(not (null? tal)))
	      (if (vector? hed)
		  (begin
		    (debug:print 1 *default-log-port* "WARNING: Dropping test " test-name "/" item-path
				 " from the launch list as it has prerequistes that are FAIL")
		    (let ((test-id (rmt:get-test-id area-dat run-id hed "")))
		      (if test-id (mt:test-set-state-status-by-id run-id test-id "NOT_STARTED" "PREQ_FAIL" "Failed to run due to failed prerequisites")))
		    (runs:shrink-can-run-more-tests-count runsdat) ;; DELAY TWEAKER (still needed?)
		    ;; (thread-sleep! *global-delta*)
		    ;; This next is for the items
		    (mt:test-set-state-status-by-testname run-id test-name item-path "NOT_STARTED" "BLOCKED" #f)
		    (hash-table-set! test-registry (db:test-make-full-name test-name item-path) 'removed)
		    (list (runs:queue-next-hed tal reg reglen regfull)

		     (else
		      (if (runs:lownoise (conc "FAILED prerequitests and we tried" hed) 60)
			  (debug:print 0 *default-log-port* "WARNING: test " hed " has FAILED prerequitests and we've tried at least 10 times to run it. Giving up now."))
		      ;; (debug:print 0 *default-log-port* "         prereqs: " prereqs-not-met)
		      (hash-table-set! test-registry hed 'removed)
		      (mt:test-set-state-status-by-testname run-id test-name item-path "NOT_STARTED" "TEN_STRIKES" #f)
		      ;; I'm unclear on if this roll up is needed - it may be the root cause of the "all set to FAIL" bug.
		      (rmt:set-state-status-and-roll-up-items area-dat run-id test-name item-path #f "FAIL" #f) ;; treat as FAIL
		      (list (if (null? tal)(car newtal)(car tal))
			    tal
			    reg
			    reruns)))))
	      ;; can't drop this - maybe running? Just keep trying
	      (let ((runable-tests (runs:runable-tests prereqs-not-met)))
		(if (null? runable-tests)

   inc-results-fmt: "~12a~12a~20a~12a~40a\n" ;; state status time duration test-name item-path
   run-info: #f
   runname: #f
   target: #f
   )
)

(define (runs:incremental-print-results area-dat run-id)
  (let ((curr-sec (current-seconds)))
    (if (> (- curr-sec (runs:gendat-inc-results-last-update *runs:general-data*)) 5) ;; at least five seconds since last update
	(let* ((run-dat  (or (runs:gendat-run-info *runs:general-data*)(rmt:get-run-info area-dat run-id)))
	       (runname  (or (runs:gendat-runname *runs:general-data*)
			     (db:get-value-by-header (db:get-rows run-dat)
						     (db:get-header run-dat) "runname")))
	       (target   (or (runs:gendat-target *runs:general-data*)(rmt:get-target area-dat run-id)))
	       (testsdat (rmt:get-tests-for-run area-dat run-id "%" '() '() ;; run-id testpatt states statuses
						#f #f ;; offset limit
						#f ;; not-in
						#f ;; sort-by
						#f ;; sort-order
						#f ;; get full data (not 'shortlist)
						(runs:gendat-inc-results-last-update *runs:general-data*) ;; last update time
						'dashboard)))

;; every time though the loop increment the test/itempatt val.
;; when the min is > max-allowed and none running then force exit
;;
(define *max-tries-hash* (make-hash-table))

;; test-records is a hash table testname:item_path => vector < testname testconfig waitons priority items-info ... >
(define (runs:run-tests-queue area-dat run-id runname test-records keyvals flags test-patts required-tests reglen-in all-tests-registry)
  ;; At this point the list of parent tests is expanded 
  ;; NB// Should expand items here and then insert into the run queue.
  (debug:print 5 *default-log-port* "test-records: " test-records ", flags: " (hash-table->alist flags))

  ;; Do mark-and-find clean up of db before starting runing of quue
  ;;
  ;; (rmt:find-and-mark-incomplete area-dat)

  (let* ((run-info              (rmt:get-run-info area-dat run-id))
	(tests-info            (mt:get-tests-for-run run-id #f '() '())) ;;  qryvals: "id,testname,item_path"))
	(sorted-test-names     (tests:sort-by-priority-and-waiton test-records))
	(test-registry         (make-hash-table))
	(registry-mutex        (make-mutex))
	(num-retries           0)
	(max-retries           (config-lookup *configdat* "setup" "maxretries"))
	(max-concurrent-jobs   (let ((mcj (config-lookup *configdat* "setup"     "max_concurrent_jobs")))

		  registry-mutex: registry-mutex
		  flags: flags
		  keyvals: keyvals
		  run-info: run-info
		  ;; newtal: newtal
		  all-tests-registry: all-tests-registry
		  ;; itemmaps: itemmaps
		  ;; prereqs-not-met: (rmt:get-prereqs-not-met area-dat run-id waitons hed item-path mode: testmode itemmaps: itemmaps)

		  )))

    ;; Initialize the test-registery hash with tests that already have a record
    ;; convert state to symbol and use that as the hash value
    (for-each (lambda (trec)
		(let ((id (db:test-get-id        trec))
		      (tn (db:test-get-testname  trec))
		      (ip (db:test-get-item-path trec))
		      (st (db:test-get-state     trec)))
		  (if (not (equal? st "DELETED"))
		      (hash-table-set! test-registry (db:test-make-full-name tn ip) (string->symbol st)))))
	      tests-info)
    (set! max-retries (if (and max-retries (string->number max-retries))(string->number max-retries) 100))

    (let loop ((hed         (car sorted-test-names))
	       (tal         (cdr sorted-test-names))
	       (reg         '()) ;; registered, put these at the head of tal 
	       (reruns      '()))

      (runs:incremental-print-results area-dat run-id)

      (if (not (null? reruns))(debug:print-info 4 *default-log-port* "reruns=" reruns))

      ;; Here we mark any old defunct tests as incomplete. Do this every fifteen minutes
      ;; moving this to a parallel thread and just run it once.
      ;;
      (if (> (current-seconds)(+ last-time-incomplete 900))
          (begin
            (set! last-time-incomplete (current-seconds))
            ;; (rmt:find-and-mark-incomplete-all-runs area-dat)
	    ))

      ;; (print "Top of loop, hed=" hed ", tal=" tal " ,reruns=" reruns)
      (let* ((test-record (hash-table-ref test-records hed))
	     (test-name   (tests:testqueue-get-testname test-record))
	     (tconfig     (tests:testqueue-get-testconfig test-record))
	     (jobgroup    (config-lookup tconfig "test_meta" "jobgroup"))
	     (testmode    (let ((m (config-lookup tconfig "requirements" "mode")))
			    (if m (map string->symbol (string-split m)) '(normal))))
	     (itemmaps    (tests:get-itemmaps tconfig)) ;;  (configf:lookup tconfig "requirements" "itemmap"))
	     (waitons     (tests:testqueue-get-waitons    test-record))
	     (priority    (tests:testqueue-get-priority   test-record))
	     (itemdat     (tests:testqueue-get-itemdat    test-record)) ;; itemdat can be a string, list or #f
	     (items       (tests:testqueue-get-items      test-record))
	     (item-path   (item-list->path itemdat))
	     (tfullname   (db:test-make-full-name test-name item-path))
	     (newtal      (append tal (list hed)))
	     (regfull     (>= (length reg) reglen))
	     (num-running (rmt:get-count-tests-running-for-run-id area-dat run-id))
	     (testdat     (make-runs:testdat
			   hed: hed
			   tal: tal
			   reg: reg
			   reruns: reruns
			   test-record: test-record
			   test-name:   test-name

	  (hash-table-set! *max-tries-hash* tfullname (+ (hash-table-ref/default *max-tries-hash* tfullname 0) 1)))
	;; (debug:print 0 *default-log-port* "max-tries-hash: " (hash-table->alist *max-tries-hash*))

	;; Ensure all top level tests get registered. This way they show up as "NOT_STARTED" on the dashboard
	;; and it is clear they *should* have run but did not.
	(if (not (hash-table-ref/default test-registry (db:test-make-full-name test-name "") #f))
	    (begin
	      (rmt:register-test area-dat run-id test-name "")
	      (hash-table-set! test-registry (db:test-make-full-name test-name "") 'done)))
	
	;; Fast skip of tests that are already "COMPLETED" - NO! Cannot do that as the items may not have been expanded yet :(
	;;
	(if (member (hash-table-ref/default test-registry tfullname #f) 
		    '(DONOTRUN removed)) ;; *common:cant-run-states-sym*) ;; '(COMPLETED KILLED WAIVED UNKNOWN INCOMPLETE))
	    (begin
	      (if (runs:lownoise (conc "been marked do not run " tfullname) 60)
		  (debug:print-info 0 *default-log-port* "Skipping test " tfullname " as it has been marked do not run due to being completed or not runnable"))
	      (if (or (not (null? tal))(not (null? reg)))
		  (loop (runs:queue-next-hed tal reg reglen regfull)
			(runs:queue-next-tal tal reg reglen regfull)
			(runs:queue-next-reg tal reg reglen regfull)
			reruns))))
		  ;; (loop (car tal)(cdr tal) reg reruns))))

	(runs:incremental-print-results area-dat run-id)
	(debug:print 4 *default-log-port* "TOP OF LOOP => "
		     "test-name: " test-name
		     "\n  test-record  " test-record
		     "\n  hed:         " hed
		     "\n  itemdat:     " itemdat
		     "\n  items:       " items
		     "\n  item-path:   " item-path

	 ;; items is #f then the test is ok to be handed off to launch (but not before)
	 ;; 
	 ((not items)
	  (debug:print-info 4 *default-log-port* "OUTER COND: (not items)")
	  (if (and (not (tests:match test-patts (tests:testqueue-get-testname test-record) item-path required: required-tests))
		   (not (null? tal)))
	      (loop (car tal)(cdr tal) reg reruns))
	  (runs:testdat-prereqs-not-met-set! testdat (rmt:get-prereqs-not-met area-dat run-id waitons hed item-path mode: testmode itemmaps: itemmaps))
	  (runs:dat-can-run-more-tests-set! runsdat (runs:can-run-more-tests area-dat runsdat run-id jobgroup max-concurrent-jobs)) ;; look at the test jobgroup and tot jobs running
	  (let ((loop-list (runs:process-expanded-tests area-dat runsdat testdat)))
	      (if loop-list (apply loop loop-list))))

	 ;; items processed into a list but not came in as a list been processed
	 ;;
	 ((and (list? items)     ;; thus we know our items are already calculated
	       (not   itemdat))  ;; and not yet expanded into the list of things to be done
	  (debug:print-info 4 *default-log-port* "OUTER COND: (and (list? items)(not itemdat))")

	      #f
	      (loop (car tal)(cdr tal) reg reruns)))
	    
	 ;; if items is a proc then need to run items:get-items-from-config, get the list and loop 
	 ;;    - but only do that if resources exist to kick off the job
	 ;; EXPAND ITEMS
	 ((or (procedure? items)(eq? items 'have-procedure))
	  (let ((can-run-more    (runs:can-run-more-tests area-dat runsdat run-id jobgroup max-concurrent-jobs)))
	    (if (and (list? can-run-more)
		     (car can-run-more))
		(let ((loop-list (runs:expand-items area-dat hed tal reg reruns regfull newtal jobgroup max-concurrent-jobs run-id waitons item-path testmode test-record can-run-more items runname tconfig reglen test-registry test-records itemmaps)))
		  (if loop-list
		      (apply loop loop-list)))
		;; if can't run more just loop with next possible test
		(loop (car newtal)(cdr newtal) reg reruns))))
	    
	 ;; this case should not happen, added to help catch any bugs
	 ((and (list? items) itemdat)
	  (debug:print-error 0 *default-log-port* "Should not have a list of items in a test and the itemspath set - please report this")
	  (exit 1))
	 ((not (null? reruns))
	  (let* ((newlst (tests:filter-non-runnable area-dat run-id tal test-records)) ;; i.e. not FAIL, WAIVED, INCOMPLETE, PASS, KILLED,
		 (junked (lset-difference equal? tal newlst)))
	    (debug:print-info 4 *default-log-port* "full drop through, if reruns is less than 100 we will force retry them, reruns=" reruns ", tal=" tal)
	    (if (< num-retries max-retries)
		(set! newlst (append reruns newlst)))
	    (set! num-retries (+ num-retries 1))
	    ;; (thread-sleep! (+ 1 *global-delta*))
	    (if (not (null? newlst))
		;; since reruns have been tacked on to newlst create new reruns from junked
		(loop (car newlst)(cdr newlst) reg (delete-duplicates junked)))))
	 ((not (null? tal))
	  (debug:print-info 4 *default-log-port* "I'm pretty sure I shouldn't get here."))
	 ((not (null? reg)) ;; could we get here with leftovers?
	  (debug:print-info 0 *default-log-port* "Have leftovers!")
	  (loop (car reg)(cdr reg) '() reruns))
	 (else
	  (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 area-dat run-id))
		    (prev-num-running 0))
      ;; (BB> "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))
		(begin
		  (debug:print-info 0 *default-log-port* "Marking stuck tests as INCOMPLETE while waiting for run " run-id ". Running as pid " (current-process-id) " on " (get-host-name))
		  (set! last-time-incomplete (current-seconds))
		  (rmt:find-and-mark-incomplete area-dat run-id #f)))
	    (if (not (eq? num-running prev-num-running))
		(debug:print-info 0 *default-log-port* "run-wait specified, waiting on " num-running " tests in RUNNING, REMOTEHOSTSTART or LAUNCHED state at " (time->string (seconds->local-time (current-seconds)))))
	    (thread-sleep! 5)
	    ;; (wait-loop (rmt:get-count-tests-running-for-run-id area-dat run-id) num-running))))
	    (wait-loop (rmt:get-count-tests-running-for-run-id area-dat run-id) num-running))))
    ;; LET* ((test-record
    ;; we get here on "drop through". All done!
    (debug:print-info 1 *default-log-port* "All tests launched")))

(define (runs:calc-fails prereqs-not-met)
  (filter (lambda (test)
	    (and (vector? test) ;; not (string? test))

	 (if (not (vector? t))
	     (conc t)
	     (conc (db:test-get-testname t) ":" (db:test-get-state t) "/" (db:test-get-status t))))
       lst))

;; parent-test is there as a placeholder for when parent-tests can be run as a setup step
;;
(define (run:test area-dat run-id run-info keyvals runname test-record flags parent-test test-registry all-tests-registry)
  ;; All these vars might be referenced by the testconfig file reader
  (let* ((test-name    (tests:testqueue-get-testname   test-record))
	 (test-waitons (tests:testqueue-get-waitons    test-record))
	 (test-conf    (tests:testqueue-get-testconfig test-record))
	 (itemdat      (tests:testqueue-get-itemdat    test-record))
	 (test-path    (hash-table-ref all-tests-registry test-name)) ;; (conc *toppath* "/tests/" test-name)) ;; could use tests:get-testconfig here ...
	 (force        (hash-table-ref/default flags "-force" #f))

		      "\n   test-config: " (hash-table->alist test-conf)
		      "\n   itemdat: " itemdat
		      )
    (debug:print 2 *default-log-port* "Attempting to launch test " full-test-name)
    ;; (setenv "MT_TEST_NAME" test-name) ;; 
    ;; (setenv "MT_ITEMPATH"  item-path)
    ;; (setenv "MT_RUNNAME"   runname)
    (runs:set-megatest-env-vars area-dat run-id inrunname: runname testname: test-name itempath: item-path) ;; these may be needed by the launching process
    (change-directory *toppath*)

    ;; Here is where the test_meta table is best updated
    ;; Yes, another use of a global for caching. Need a better way?
    ;;
    ;; There is now a single call to runs:update-all-test_meta and this 
    ;; per-test call is not needed. Given the delicacy of the move to 
    ;; v1.55 this code is being left in place for the time being.
    ;;
    (if (not (hash-table-ref/default *test-meta-updated* test-name #f))
        (begin
	   (hash-table-set! *test-meta-updated* test-name #t)
           (runs:update-test_meta area-dat test-name test-conf)))
    
    ;; itemdat => ((ripeness "overripe") (temperature "cool") (season "summer"))
    (let* ((new-test-path (string-intersperse (cons test-path (map cadr itemdat)) "/"))
	   (test-id       (rmt:get-test-id area-dat run-id test-name item-path))
	   (testdat       (if test-id (rmt:get-test-info-by-id area-dat run-id test-id) #f)))
      (if (not testdat)
	  (let loop ()
	    ;; ensure that the path exists before registering the test
	    ;; NOPE: Cannot! Don't know yet which disk area will be assigned....
	    ;; (system (conc "mkdir -p " new-test-path))
	    ;;
	    ;; (open-run-close tests:register-test db run-id test-name item-path)
	    ;;
	    ;; NB// for the above line. I want the test to be registered long before this routine gets called!
	    ;;
	    (if (not test-id)(set! test-id (rmt:get-test-id area-dat run-id test-name item-path)))
	    (if (not test-id)
		(begin
		  (debug:print 2 *default-log-port* "WARN: Test not pre-created? test-name=" test-name ", item-path=" item-path ", run-id=" run-id)
		  (rmt:register-test area-dat run-id test-name item-path)
		  (set! test-id (rmt:get-test-id area-dat run-id test-name item-path))))
	    (debug:print-info 4 *default-log-port* "test-id=" test-id ", run-id=" run-id ", test-name=" test-name ", item-path=\"" item-path "\"")
	    (set! testdat (rmt:get-test-info-by-id area-dat run-id test-id))
	    (if (not testdat)
		(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))

		     (skip-check  (configf:get-section test-conf "skip")))
		 (cond 
		  ;; Have to check for skip conditions. This one skips if there are same-named tests
		  ;; currently running
		  ((and skip-check
			(configf:lookup test-conf "skip" "prevrunning"))
		   ;; run-ids = #f means *all* runs
		   (let ((running-tests (rmt:get-tests-for-runs-mindata area-dat #f full-test-name '("RUNNING" "REMOTEHOSTSTART" "LAUNCHED") '() #f)))
		     (if (not (null? running-tests)) ;; have to skip 
			 (set! skip-test "Skipping due to previous tests running"))))
		  ((and skip-check
			(configf:lookup test-conf "skip" "fileexists"))
		   (if (file-exists? (configf:lookup test-conf "skip" "fileexists"))
		       (set! skip-test (conc "Skipping due to existance of file " (configf:lookup test-conf "skip" "fileexists")))))

		  ((and skip-check
			(configf:lookup test-conf "skip" "rundelay"))
		   ;; run-ids = #f means *all* runs
		   (let* ((numseconds      (common:hms-string->seconds (configf:lookup test-conf "skip" "rundelay")))
			  (running-tests   (rmt:get-tests-for-runs-mindata area-dat #f full-test-name '("RUNNING" "REMOTEHOSTSTART" "LAUNCHED") '() #f))
			  (completed-tests (rmt:get-tests-for-runs-mindata area-dat #f full-test-name '("COMPLETED" "INCOMPLETE") '("PASS" "FAIL" "ABORT") #f)) ;; ironically INCOMPLETE is same as COMPLETED in this contex
			  (last-run-times  (map db:mintest-get-event_time completed-tests))
			  (time-since-last (- (current-seconds) (if (null? last-run-times) 0 (common:max last-run-times)))))
		     (if (or (not (null? running-tests)) ;; have to skip if test is running
			     (> numseconds time-since-last))
			 (set! skip-test (conc "Skipping due to previous test run less than " (configf:lookup test-conf "skip" "rundelay") " ago"))))))
		 
		 (if skip-test
		     (begin
		       (mt:test-set-state-status-by-id run-id test-id "COMPLETED" "SKIP" skip-test)
		       (debug:print-info 1 *default-log-port* "SKIPPING Test " full-test-name " due to " skip-test))
		     (if (not (launch-test test-id run-id run-info keyvals runname test-conf test-name test-path itemdat flags))
			 (begin
			   (print "ERROR: Failed to launch the test. Exiting as soon as possible")
			   (set! *globalexitstatus* 1) ;; 
			   (process-signal (current-process-id) signal/kill))))))))
	((KILLED) 
	 (debug:print 1 *default-log-port* "NOTE: " full-test-name " is already running or was explictly killed, use -force to launch it.")
	 (hash-table-set! test-registry (db:test-make-full-name test-name test-path) 'DONOTRUN)) ;; KILLED))
	((LAUNCHED REMOTEHOSTSTART RUNNING)  
	 (debug:print 2 *default-log-port* "NOTE: " test-name " is already running"))









	(else      
	 (debug:print-error 0 *default-log-port* "Failed to launch test " full-test-name ". Unrecognised state " (test:get-state testdat))
	 (case (string->symbol (test:get-state testdat)) 
	   ((COMPLETED INCOMPLETE)
	    (hash-table-set! test-registry (db:test-make-full-name test-name test-path) 'DONOTRUN))
	   (else
	    (hash-table-set! test-registry (db:test-make-full-name test-name test-path) 'DONOTRUN))))))))

;; fields are passing in through 
;; action:
;;    'remove-runs
;;    'set-state-status
;;
;; NB// should pass in keys?
;;
(define (runs:operate-on area-dat action target runnamepatt testpatt #!key (state #f)(status #f)(new-state-status #f)(mode 'remove-all)(options '()))
  (common:clear-caches) ;; clear all caches
  (let* ((db           #f)
	 (tdbdat       (tasks:open-db))
	 (keys         (rmt:get-keys area-dat))
	 (rundat       (mt:get-runs-by-patt keys runnamepatt target))
	 (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))

	   (if (not (null? tests))
	       (begin
		 (case action
		   ((remove-runs)
		    ;; (if (tasks:need-server run-id)(tasks:start-and-wait-for-server tdbdat run-id 10))
		    ;; seek and kill in flight -runtests with % as testpatt here
		    ;; (if (equal? testpatt "%")
		    (tasks:kill-runner area-dat target run-name testpatt)
		    
		    ;; (debug:print 0 *default-log-port* "not attempting to kill any run launcher processes as testpatt is " testpatt))
		    (debug:print 1 *default-log-port* "Removing tests for run: " runkey " " (db:get-value-by-header run header "runname")))
		   ((set-state-status)
		    (if (tasks:need-server run-id)(tasks:start-and-wait-for-server tdbdat run-id 10))
		    (debug:print 1 *default-log-port* "Modifying state and staus for tests for run: " runkey " " (db:get-value-by-header run header "runname")))
		   ((print-run)

		   (else
		    (debug:print-info 0 *default-log-port* "action not recognised " action)))
		 
		 ;; actions that operate on one test at a time can be handled below
		 ;;
		 (let ((sorted-tests     (filter 
					  vector?
					  (sort tests (lambda (a b)(let ((dira ;; (rmt:sdb-qry area-dat 'getstr 
									  (db:test-get-rundir a)) ;; )  ;; (filedb:get-path *fdb* (db:test-get-rundir a)))
									 (dirb ;; (rmt:sdb-qry area-dat 'getstr 
									  (db:test-get-rundir b))) ;; ) ;; ((filedb:get-path *fdb* (db:test-get-rundir b))))
								     (if (and (string? dira)(string? dirb))
									 (> (string-length dira)(string-length dirb))
									 #f))))))
		       (toplevel-retries (make-hash-table)) ;; try three times to loop through and remove top level tests
		       (test-retry-time  (make-hash-table))
		       (allow-run-time   10)) ;; seconds to allow for killing tests before just brutally killing 'em
		   (let loop ((test (car sorted-tests))
			      (tal  (cdr sorted-tests)))
		     (let* ((test-id       (db:test-get-id test))
			    (new-test-dat  (rmt:get-test-info-by-id area-dat run-id test-id)))
		       (if (not new-test-dat)
			   (begin
			     (debug:print-error 0 *default-log-port* "We have a test-id of " test-id " but no record was found. NOTE: No locking of records is done between processes, do not simultaneously remove the same run from two processes!")
			     (if (not (null? tal))
				 (loop (car tal)(cdr tal))))
			   (let* ((item-path     (db:test-get-item-path new-test-dat))
				  (test-name     (db:test-get-testname new-test-dat))
				  (run-dir       ;;(filedb:get-path *fdb*
				   ;; (rmt:sdb-qry area-dat 'getid 
				   (db:test-get-rundir new-test-dat)) ;; )    ;; run dir is from the link tree
				  (test-state    (db:test-get-state new-test-dat))
				  (test-fulln    (db:test-get-fullname new-test-dat))
				  (uname         (db:test-get-uname    new-test-dat))
				  (toplevel-with-children (and (db:test-get-is-toplevel test)
							       (> (rmt:test-toplevel-num-items area-dat run-id test-name) 0))))
			     (case action
			       ((remove-runs)
				;; if the test is a toplevel-with-children issue an error and do not remove
				(if toplevel-with-children
				    (begin
				      (debug:print 0 *default-log-port* "WARNING: skipping removal of " test-fulln " with run-id " run-id " as it has sub tests")
				      (hash-table-set! toplevel-retries test-fulln (+ (hash-table-ref/default toplevel-retries test-fulln 0) 1))

					    (mt:test-set-state-status-by-id run-id (db:test-get-id test) "KILLREQ" "n/a" #f)
						  (thread-sleep! 1)))
					    ;; NOTE: This is suboptimal as the testdata will be used later and the state/status may have changed ...
					    (if (null? tal)
						(loop new-test-dat tal)
						(loop (car tal)(append tal (list new-test-dat)))))
					  (begin
					    (runs:remove-test-directory area-dat new-test-dat mode) ;; 'remove-all)
					    (if (not (null? tal))
						(loop (car tal)(cdr tal)))))))
				(rmt:update-run-stats area-dat run-id (rmt:get-raw-run-stats area-dat run-id)))
			       ((set-state-status)
				(debug:print-info 2 *default-log-port* "new state " (car state-status) ", new status " (cadr state-status))
				(mt:test-set-state-status-by-id run-id (db:test-get-id test) (car state-status)(cadr state-status) #f)
				(if (not (null? tal))
				    (loop (car tal)(cdr tal))))
			       ((run-wait)
				(debug:print-info 2 *default-log-port* "still waiting, " (length tests) " tests still running")

	       (let ((remtests (mt:get-tests-for-run (db:get-value-by-header run header "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 area-dat run-id)
		       (rmt:delete-old-deleted-test-records area-dat)
		       ;; (rmt:set-var area-dat "DELETED_TESTS" (current-seconds))
		       ;; need to figure out the path to the run dir and remove it if empty
		       ;;    (if (null? (glob (conc runpath "/*")))
		       ;;        (begin
		       ;; 	 (debug:print 1 *default-log-port* "Removing run dir " runpath)
		       ;; 	 (system (conc "rmdir -p " runpath))))
		       )))))
	 ))
     runs)
    ;; (sqlite3:finalize! (db:delay-if-busy tdbdat))
    )
  #t)

(define (runs:remove-test-directory area-dat test mode) ;; remove-data-only)
  (let* ((run-dir       (db:test-get-rundir test))    ;; run dir is from the link tree
	 (real-dir      (if (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))

		(debug:print 0 *default-log-port* "WARNING: not removing " run-dir " as it either doesn't exist or is not a symlink")
		(debug:print 0 *default-log-port* "NOTE: the run dir for this test is undefined. Test may have already been deleted."))
	    ))
    ;; Only delete the records *after* removing the directory. If things fail we have a record 
    (case mode
      ((remove-data-only)(mt:test-set-state-status-by-id (db:test-get-run_id test)(db:test-get-id test) "NOT_STARTED" "n/a" #f))
      ((archive-remove)  (mt:test-set-state-status-by-id (db:test-get-run_id test)(db:test-get-id test) "ARCHIVED" #f #f))
      (else (rmt:delete-test-records area-dat (db:test-get-run_id test) (db:test-get-id test))))))

;;======================================================================
;; Routines for manipulating runs
;;======================================================================

;; Since many calls to a run require pretty much the same setup 
;; this wrapper is used to reduce the replication of code

	;; (if db (sqlite3:finalize! db))
	(set! *didsomething* #t))))))

;;======================================================================
;; Lock/unlock runs
;;======================================================================

(define (runs:handle-locking area-dat target keys runname lock unlock user)
  (let* ((db       #f)
	 (rundat   (mt:get-runs-by-patt keys runname target))
	 (header   (vector-ref rundat 0))
	 (runs     (vector-ref rundat 1)))
    (for-each (lambda (run)
		(let ((run-id (db:get-value-by-header run header "id")))
		  (if (or lock
			  (and unlock
			       (begin
				 (print "Do you really wish to unlock run " run-id "?\n   y/n: ")
				 (equal? "y" (read-line)))))
		      (rmt:lock/unlock-run area-dat run-id lock unlock user)
		      (debug:print-info 0 *default-log-port* "Skipping lock/unlock on " run-id))))
	      runs)))
;;======================================================================
;; Rollup runs
;;======================================================================

;; Update the test_meta table for this test
(define (runs:update-test_meta area-dat test-name test-conf)
  (let ((currrecord (rmt:testmeta-get-record area-dat test-name)))
    (if (not currrecord)
	(begin
	  (set! currrecord (make-vector 11 #f))
	  (rmt:testmeta-add-record area-dat test-name)))
    (for-each 
     (lambda (key)
       (let* ((idx (cadr key))
	      (fld (car  key))
	      (val (config-lookup test-conf "test_meta" fld)))
	 ;; (debug:print 5 *default-log-port* "idx: " idx " fld: " fld " val: " val)
	 (if (and val (not (equal? (vector-ref currrecord idx) val)))
	     (begin
	       (print "Updating " test-name " " fld " to " val)
	       (rmt:testmeta-update-field area-dat test-name fld val)))))
     '(("author" 2)("owner" 3)("description" 4)("reviewed" 5)("tags" 9)("jobgroup" 10)))))

;; find tests with matching tags, tagpatt is a string "tagpatt1,tagpatt2%, ..."
;;
(define (runs:get-tests-matching-tags area-dat tagpatt)
  (let* ((tagdata (rmt:get-tests-tags area-dat))
         (res     '())) ;; list of tests that match one or more tags
    (for-each
     (lambda (tag)
       (if (patt-list-match tag tagpatt)
           (set! res (append (hash-table-ref tagdata tag)))))
     (hash-table-keys tagdata))
    res))
    

;; Update test_meta for all tests
(define (runs:update-all-test_meta area-dat)
  (let ((test-names (tests:get-all))) ;; (tests:get-valid-tests)))
    (for-each 
     (lambda (test-name)
       (let* ((test-conf    (mt:lazy-read-test-config test-name)))
	 (if test-conf (runs:update-test_meta area-dat test-name test-conf))))
     (hash-table-keys test-names))))

;; This could probably be refactored into one complex query ...
;; NOT PORTED - DO NOT USE YET
;;
(define (runs:rollup-run area-dat keys runname user keyvals)
  (debug:print 4 *default-log-port* "runs:rollup-run, keys: " keys " -runname " runname " user: " user)
  (let* ((db              #f)
	 ;; register run operates on the main db
	 (new-run-id      (rmt:register-run area-dat keyvals runname "new" "n/a" user))
	 (prev-tests      (rmt:get-matching-previous-test-run-records area-dat new-run-id "%" "%"))
	 (curr-tests      (mt:get-tests-for-run new-run-id "%/%" '() '()))
	 (curr-tests-hash (make-hash-table)))
    (rmt:update-run-event_time area-dat new-run-id)
    ;; index the already saved tests by testname and itemdat in curr-tests-hash
    (for-each
     (lambda (testdat)
       (let* ((testname  (db:test-get-testname testdat))
	      (item-path (db:test-get-item-path testdat))
	      (full-name (conc testname "/" item-path)))
	 (hash-table-set! curr-tests-hash full-name testdat)))
     curr-tests)
    ;; NOPE: Non-optimal approach. Try this instead.
    ;;   1. tests are received in a list, most recent first
    ;;   2. replace the rollup test with the new *always*
    (for-each 
     (lambda (testdat)
       (let* ((testname  (db:test-get-testname testdat))
	      (item-path (db:test-get-item-path testdat))
	      (full-name (conc testname "/" item-path))
	      (prev-test-dat (hash-table-ref/default curr-tests-hash full-name #f))
	      (test-steps    (rmt:get-steps-for-test area-dat (db:test-get-id testdat)))
	      (new-test-record #f))
	 ;; replace these with insert ... select
	 (apply sqlite3:execute 
		db 
		(conc "INSERT OR REPLACE INTO tests (run_id,testname,state,status,event_time,host,cpuload,diskfree,uname,rundir,item_path,run_duration,final_logf,comment) "
		      "VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?);")
		new-run-id (cddr (vector->list testdat)))

;; Given a run id start a server process    ### NOTE ### > file 2>&1 
;; if the run-id is zero and the target-host is set 
;; try running on that host
;;   incidental: rotate logs in logs/ dir.
;;
(define  (server:run areapath) ;; areapath is *toppath* for a given testsuite area
  (let* ((curr-host   (get-host-name))


	 (curr-ip     (server:get-best-guess-address curr-host))
	 (curr-pid    (current-process-id))
	 (homehost    (common:get-homehost)) ;; configf:lookup *configdat* "server" "homehost" ))
	 (target-host (car homehost))
	 (testsuite   (common:get-testsuite-name))
	 (logfile     (conc areapath "/logs/server.log")) ;; -" curr-pid "-" target-host ".log"))
	 (cmdln (conc (common:get-megatest-exe)

                            ((2)  300)
                            (else 600))
                          (random 5)))) ;; add a small random number just in case a lot of jobs hit the work hosts simultaneously
    (if	(> (- (current-seconds) when-run) run-delay)
        (server:run areapath))
    (hash-table-set! *server-kind-run* areapath (list (+ call-num 1)(current-seconds)))))

(define (server:start-and-wait area-dat areapath #!key (timeout 60))
  (let ((give-up-time (+ (current-seconds) timeout)))
    (let loop ((server-url (server:check-if-running area-dat areapath)))
      (if (or server-url
	      (> (current-seconds) give-up-time)) ;; server-url will be #f if no server available.
	  server-url
	  (let ((num-ok (length (server:get-best (server:get-list areapath)))))
	    (if (< num-ok 1) ;; if there are no decent candidates for servers then try starting a new one
		(server:kind-run areapath))
	    (thread-sleep! 5)
	    (loop (server:check-if-running area-dat areapath)))))))

(define server:try-running server:run) ;; there is no more per-run servers ;; REMOVE ME. BUG.

(define (server:dotserver-age-seconds areapath)
  (let ((server-file (conc areapath "/.server")))
    (begin
      (handle-exceptions
       exn
       #f
       (- (current-seconds)
          (file-modification-time server-file))))))
    
;; no longer care if multiple servers are started by accident. older servers will drop off in time.
;;
(define (server:check-if-running area-dat areapath)
  (let* ((servers       (server:get-best (server:get-list areapath))))
    (if (null? servers)
        #f
        (let loop ((hed (car servers))
                   (tal (cdr servers)))
          (let ((res (server:check-server area-dat hed)))
            (if res
                res
                (if (null? tal)
                    #f
                    (loop (car tal)(cdr tal)))))))))

;; ping the given server
;;
(define (server:check-server area-dat server-record)
  (let* ((server-url (server:record->url server-record))
         (res        (case *transport-type*
                       ((http)(server:ping area-dat server-url))
                       ;; ((nmsg)(nmsg-transport:ping (tasks:hostinfo-get-interface server)
                       )))
    (if res
        server-url
	#f)))

(define (server:kill servr)
  (match-let (((mod-time hostname port start-time pid)
	       servr))
    (tasks:kill-server hostname pid)))

;; called in megatest.scm, host-port is string hostname:port
;;
;; NOTE: This is NOT called directly from clients as not all transports support a client running
;;       in the same process as the server.
;;
(define (server:ping area-dat host-port-in #!key (do-exit #f))
  (let ((host:port (if (not host-port-in) ;; use read-dotserver to find
		       #f






		       host-port-in))) ;; )
    (let* ((host-port (if host:port
			  (let ((slst (string-split   host:port ":")))
			    (if (eq? (length slst) 2)
				(list (car slst)(string->number (cadr slst)))
				#f))
			  #f)))


      (if (not host-port)
	  (begin
	    (if host-port-in
		(debug:print 0 *default-log-port*  "ERROR: bad host:port"))
	    (if do-exit (exit 1))
	    #f)
	  (let* ((iface      (car host-port))
		 (port       (cadr host-port))
		 (server-dat (http-transport:client-connect iface port))
		 (login-res  (rmt:login-no-auto-client-setup area-dat server-dat)))
	    (if (and (list? login-res)
		     (car login-res))
		(begin
		  ;; (print "LOGIN_OK")
		  (if do-exit (exit 0))
		  #t)
		(begin

     orig-keys)
    (list changed deleted)
    ;; (list indat '()) ;; just for debugging
    ))
    
;; keynum => the field to use as the unique key (usually 0 but can be other field)
;;
(define (synchash:client-get area-dat proc synckey keynum synchash run-id . params)
  (let* ((data   (rmt:synchash-get area-dat run-id proc synckey keynum params))
	 (newdat (car data))
	 (removs (cadr data))
	 (myhash (hash-table-ref/default synchash synckey #f)))
    (if (not myhash)
	(begin
	  (set! myhash (make-hash-table))
	  (hash-table-set! synchash synckey myhash)))

     target run-name state-patt action-patt test-patt)
    res)) ;; )

;; kill any runner processes (i.e. processes handling -runtests) that match target/runname
;; 
;; do a remote call to get the task queue info but do the killing as self here.
;;
(define (tasks:kill-runner area-dat target run-name testpatt)
  (let ((records    (rmt:tasks-find-task-queue-records area-dat target run-name testpatt "running" "run-tests"))
	(hostpid-rx (regexp "\\s+(\\w+)\\s+(\\d+)$"))) ;; host pid is at end of param string
    (if (null? records)
	(debug:print 0 *default-log-port* "No run launching processes found for " target " / " run-name " with testpatt " (or testpatt "* no testpatt specified! *"))
	(debug:print 0 *default-log-port* "Found " (length records) " run(s) to kill."))
    (for-each 
     (lambda (record)
       (let* ((param-key (list-ref record 8))

      (let ((baddb (sqlite3:open-database ":memory:")))
 	(debug:print-info 11 *default-log-port* "open-test-db END (unsucessful)" work-area)
 	;; provide an in-mem db (this is dangerous!)
 	(tdb:testdb-initialize baddb)
 	baddb)))

;; find and open the testdat.db file for an existing test
(define (tdb:open-test-db-by-test-id area-dat test-id #!key (work-area #f))
  (let* ((test-path (if work-area
			work-area
			(rmt:test-get-rundir-from-test-id area-dat test-id))))
    (debug:print 3 *default-log-port* "TEST PATH: " test-path)
    (open-test-db test-path)))

;; find and open the testdat.db file for an existing test
(define (tdb:open-test-db-by-test-id-local dbstruct run-id test-id #!key (work-area #f))
  (let* ((test-path (if work-area
			work-area

;; 	   tdb
;; 	   "SELECT id,test_id,category,variable,value,expected,tol,units,comment,status,type FROM test_data WHERE test_id=? AND category LIKE ? ORDER BY category,variable;" test-id categorypatt)
;; 	  (sqlite3:finalize! tdb)
;; 	  (reverse res))
;; 	'())))

;; NOTE: Run this local with #f for db !!!
(define (tdb:load-test-data area-dat run-id test-id)
  (let loop ((lin (read-line)))
    (if (not (eof-object? lin))
	(begin
	  (debug:print 4 *default-log-port* lin)
          ;;(when lin ;; this when blocked stack dump caused by .dat file from logpro being 0-byte.  fixed by upgrading logpro
          (rmt:csv->test-data area-dat run-id test-id lin)
          ;;)
	  (loop (read-line)))))
  ;; roll up the current results.
  ;; FIXME: Add the status too 
  (rmt:test-data-rollup area-dat run-id test-id #f))

;; NOTE: Run this local with #f for db !!!
(define (tdb:load-logpro-data area-dat run-id test-id)
  (let loop ((lin (read-line)))
    (if (not (eof-object? lin))
	(begin
	  (debug:print 4 *default-log-port* lin)
          ;;(when lin  ;; this when blocked stack dump caused by .dat file from logpro being 0-byte.  fixed by upgrading logpro
          (rmt:csv->test-data area-dat run-id test-id lin)
          ;;)
	  (loop (read-line)))))
  ;; roll up the current results.
  ;; FIXME: Add the status too 
  (rmt:test-data-rollup area-dat run-id test-id #f))

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

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

			 #t
			 (if (eq? time-a time-b)
			     (string<? (conc (vector-ref a 2))
				       (conc (vector-ref b 2)))
			     #f))
		     (string<? (conc time-a)(conc time-b))))))))

;; NOT USED
;;
(define (tdb:remote-update-testdat-meta-info area-dat run-id test-id work-area cpuload diskfree minutes)
  (let ((tdb         (rmt:open-test-db-by-test-id area-dat run-id test-id work-area: work-area)))
    (if (sqlite3:database? tdb)
	(begin
	  (sqlite3:execute tdb "INSERT INTO test_rundat (update_time,cpuload,diskfree,run_duration) VALUES (strftime('%s','now'),?,?,?);"
			   cpuload diskfree minutes)
	  (sqlite3:finalize! tdb))
	(debug:print 2 *default-log-port* "Can't update testdat.db for test " test-id " read-only or non-existant"))))
    

					#t
					(loop (car tal)(cdr tal)))
				    #f))))))
	    (pop-directory)
	    result)))))

;; Do not rpc this one, do the underlying calls!!!
(define (tests:test-set-status! area-dat run-id test-id state status comment dat #!key (work-area #f))
  (let* ((real-status status)
	 (otherdat    (if dat dat (make-hash-table)))
	 (testdat     (rmt:get-test-info-by-id area-dat run-id test-id))
	 (test-name   (db:test-get-testname  testdat))
	 (item-path   (db:test-get-item-path testdat))
	 ;; before proceeding we must find out if the previous test (where all keys matched except runname)
	 ;; was WAIVED if this test is FAIL

	 ;; NOTES:
	 ;;  1. Is the call to test:get-previous-run-record remotified?
	 ;;  2. Add test for testconfig waiver propagation control here
	 ;;
	 (prev-test   (if (equal? status "FAIL")
			  (rmt:get-previous-test-run-record area-dat run-id test-name item-path)
			  #f))
	 (waived   (if prev-test
		       (if prev-test ;; true if we found a previous test in this run series
			   (let ((prev-status  (db:test-get-status  prev-test))
				 (prev-state   (db:test-get-state   prev-test))
				 (prev-comment (db:test-get-comment prev-test)))
			     (debug:print 4 *default-log-port* "prev-status " prev-status ", prev-state " prev-state ", prev-comment " prev-comment)

	(set! real-status "WAIVED"))

    (debug:print 4 *default-log-port* "real-status " real-status ", waived " waived ", status " status)

    ;; update the primary record IF state AND status are defined
    (if (and state status)
	(begin
	  (rmt:set-state-status-and-roll-up-items area-dat run-id test-id item-path state real-status (if waived waived comment))
	  ;; (mt:process-triggers run-id test-id state real-status) ;; triggers are called in test-set-state-status
	  ))
    
    ;; if status is "AUTO" then call rollup (note, this one modifies data in test
    ;; run area, it does remote calls under the hood.
    ;; (if (and test-id state status (equal? status "AUTO")) 
    ;; 	(rmt:test-data-rollup area-dat run-id test-id status))

    ;; add metadata (need to do this way to avoid SQL injection issues)

    ;; :first_err
    ;; (let ((val (hash-table-ref/default otherdat ":first_err" #f)))
    ;;   (if val
    ;;       (sqlite3:execute db "UPDATE tests SET first_err=? WHERE run_id=? AND testname=? AND item_path=?;" val run-id test-name item-path)))

			   value    ","
			   expected ","
			   tol      ","
			   units    ","
			   dcomment ",," ;; extra comma for status
			   type     )))
	    ;; This was run remote, don't think that makes sense. Perhaps not, but that is the easiest path for the moment.
	    (rmt:csv->test-data area-dat run-id test-id
				dat))))
      
    ;; need to update the top test record if PASS or FAIL and this is a subtest
    ;;;;;; (if (not (equal? item-path ""))
    ;;;;;;     (rmt:set-state-status-and-roll-up-items area-dat run-id test-name item-path state status #f) ;;;;;)

    (if (or (and (string? comment)
		 (string-match (regexp "\\S+") comment))
	    waived)
	(let ((cmt  (if waived waived comment)))
	  (rmt:general-call area-dat 'set-test-comment run-id cmt test-id)))))

(define (tests:test-set-toplog! area-dat run-id test-name logf) 
  (rmt:general-call area-dat 'tests:test-set-toplog run-id logf run-id test-name))

(define (tests:summarize-items area-dat run-id test-id test-name force)
  ;; if not force then only update the record if one of these is true:
  ;;   1. logf is "log/final.log
  ;;   2. logf is same as outputfilename
  (let* ((outputfilename (conc "megatest-rollup-" test-name ".html"))
	 (orig-dir       (current-directory))
	 (logf-info      (rmt:test-get-logfile-info area-dat run-id test-name))
	 (logf           (if logf-info (cadr logf-info) #f))
	 (path           (if logf-info (car  logf-info) #f)))
    ;; This query finds the path and changes the directory to it for the test
    (if (and (string? path)
	     (directory? path)) ;; can get #f here under some wierd conditions. why, unknown ...
	(begin
	  (debug:print 4 *default-log-port* "Found path: " path)
	  (change-directory path))
	;; (set! outputfilename (conc path "/" outputfilename)))
	(debug:print-error 0 *default-log-port* "summarize-items for run-id=" run-id ", test-name=" test-name ", no such path: " path))
    (debug:print 4 *default-log-port* "summarize-items with logf " logf ", outputfilename " outputfilename " and force " force)
    (if (or (equal? logf "logs/final.log")
	    (equal? logf outputfilename)
	    force)
	(let ((my-start-time (current-seconds))
	      (lockf         (conc outputfilename ".lock")))
	  (let loop ((have-lock  (common:simple-file-lock lockf)))
	    (if have-lock
		(let ((script (configf:lookup *configdat* "testrollup" test-name)))
		  (print "Obtained lock for " outputfilename)
		  (rmt:set-state-status-and-roll-up-items area-dat run-id test-name "" #f #f #f)
		  (if script
		      (system (conc script " > " outputfilename " & "))
		      (tests:generate-html-summary-for-iterated-test area-dat run-id test-id test-name outputfilename))
		  (common:simple-file-release-lock lockf)
		  (change-directory orig-dir)
		  ;; NB// tests:test-set-toplog! is remote internal...
		  (tests:test-set-toplog! area-dat run-id test-name outputfilename))
		;; didn't get the lock, check to see if current update started later than this 
		;; update, if so we can exit without doing any work
		(if (> my-start-time (file-modification-time lockf))
		    ;; we started since current re-gen in flight, delay a little and try again
		    (begin
		      (debug:print-info 1 *default-log-port* "Waiting to update " outputfilename ", another test currently updating it")
		      (thread-sleep! (+ 5 (random 5))) ;; delay between 5 and 10 seconds
		      (loop (common:simple-file-lock lockf))))))))))

(define (tests:generate-html-summary-for-iterated-test area-dat run-id test-id test-name outputfilename)
  (let ((counts              (make-hash-table))
	(statecounts         (make-hash-table))
	(outtxt              "")
	(tot                 0)
	(testdat             (rmt:test-get-records-for-index-file area-dat run-id test-name)))
    (with-output-to-file outputfilename
      (lambda ()
	(set! outtxt (conc outtxt "<html><title>Summary: " test-name 
			   "</title><body><h2>Summary for " test-name "</h2>"))
	(for-each
	 (lambda (testrecord)
	   (let ((id             (vector-ref testrecord 0))

	(print "<tr><td>Total</td><td>" tot "</td></tr></table>")
	(print "</td></td></tr></table>")
	
	(print "<table cellspacing=\"0\" border=\"1\">" 
	       "<tr><td>Item</td><td>State</td><td>Status</td><td>Comment</td>"
	       outtxt "</table></body></html>")
	;; (release-dot-lock outputfilename)
	;;(rmt:update-run-stats area-dat 
	;; run-id
	;; (hash-table-map
	;;  state-status-counts
	;;  (lambda (key val)
	;;	(append key (list val)))))
	))))

)

(define (tests:run-record->test-path run numkeys)
   (append (take (vector->list run) numkeys)
	   (list (vector-ref run (+ 1 numkeys)))))


(define (tests:get-rest-data area-dat runs header numkeys)
   (let ((resh (make-hash-table)))
   (for-each
     (lambda (run)
        (let* ((run-id (db:get-value-by-header run header "id"))
               (run-dir      (tests:run-record->test-path run numkeys))
	       (test-data    (rmt:get-tests-for-run area-dat
				   run-id
                                   "%"       ;; testnamepatt
				   '()        ;; states
				   '()        ;; statuses
				   #f         ;; offset
				   #f         ;; num-to-get
				   #f         ;; hide/not-hide

               (hash-table-set!  (hash-table-ref/default (hash-table-ref/default resh test-name  #f) test-item #f) run-id (list test-status test-html-path)))) 
        test-data)))
      runs)
   resh))

;; (tests:create-html-tree "test-index.html")
;;
(define (tests:create-html-tree area-dat outf)
   (let* ((lockfile  (conc outf ".lock"))
	 (runs-to-process '())
         (linktree  (common:get-linktree))
          (area-name (common:get-testsuite-name))
	  (keys      (rmt:get-keys area-dat))
	  (numkeys   (length keys))
         (total-runs  (rmt:get-num-runs area-dat "%"))
         (pg-size 10)   )
    (if (common:simple-file-lock lockfile)
        (begin
         (print total-runs)    
        (let loop ((page 0))
	(let* ((oup       (open-output-file (or outf (conc linktree "/page" page ".html"))))
               (start (* page pg-size)) 
	       (runsdat   (rmt:get-runs area-dat "%" pg-size start (map (lambda (x)(list x "%")) keys)))
	       (header    (vector-ref runsdat 0))
	       (runs      (vector-ref runsdat 1))
               (ctr 0)
               (test-runs-hash (tests:get-rest-data area-dat runs header numkeys))
               (test-list (hash-table-keys test-runs-hash))
               (get-prev-links (lambda (page linktree )   
                            (let* ((link  (if (not (eq? page 0))
                                            (s:a "<<prev" 'href (conc  linktree "/page" (- page 1) ".html"))
                                            (s:a "" 'href (conc  linktree "/page"  page ".html")))))
                               link)))
                (get-next-links (lambda (page linktree total-runs)   

	#f)))






(define (tests:create-html-tree-old area-dat outf)
   (let* ((lockfile  (conc outf ".lock"))
	 (runs-to-process '()))
    (if (common:simple-file-lock lockfile)
	(let* ((linktree  (common:get-linktree))
	       (oup       (open-output-file (or outf (conc linktree "/runs-index.html"))))
	       (area-name (common:get-testsuite-name))
	       (keys      (rmt:get-keys area-dat))
	       (numkeys   (length keys))
	       (runsdat   (rmt:get-runs area-dat "%" #f #f (map (lambda (x)(list x "%")) keys)))
	       (header    (vector-ref runsdat 0))
	       (runs      (vector-ref runsdat 1))
	       (runtreedat (map (lambda (x)
				  (tests:run-record->test-path x numkeys))
				runs))
	       (runs-htree (common:list->htree runtreedat)))
	  (set! runs-to-process runs)

	  (common:simple-file-release-lock lockfile)
               
	  (for-each
	   (lambda (run)
	     (let* ((test-subpath (tests:run-record->test-path run numkeys))
		    (run-id       (db:get-value-by-header run header "id"))
                    (run-dir      (tests:run-record->test-path run numkeys))
		    (test-dats    (rmt:get-tests-for-run area-dat
				   run-id
                                   "%/"       ;; testnamepatt
				   '()        ;; states
				   '()        ;; statuses
				   #f         ;; offset
				   #f         ;; num-to-get
				   #f         ;; hide/not-hide

                                                                          (html-file (if (file-exists? alt-file)
                                                                                         alt-file
                                                                                         std-file))
                                                                          (run-name  (car (reverse p))))
                                                                     (if (and (not (file-exists? full-targ))
                                                                              (directory? full-targ)
                                                                              (file-write-access? full-targ))
                                                                         (tests:summarize-test area-dat
                                                                          run-id 
                                                                          (rmt:get-test-id area-dat run-id test-name item-path)))
                                                                     (if (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)))))

		      ((eq? (tdb:step-get-event_time a)(tdb:step-get-event_time b)) 
		       (<   (tdb:step-get-id a)        (tdb:step-get-id b)))
		      (else #f)))))
      res))

;; 
;;
(define (tests:get-compressed-steps area-dat run-id test-id)
  (let* ((steps-data  (rmt:get-steps-for-test area-dat run-id test-id))
	 (comprsteps  (tests:process-steps-table steps-data))) ;; (open-run-close db:get-steps-table #f test-id work-area: work-area)))
    (map (lambda (x)
	   ;; take advantage of the \n on time->string
	   (vector
	    (vector-ref x 0)
	    (let ((s (vector-ref x 1)))
	      (if (number? s)(seconds->time-string s) s))

					 (conc (vector-ref b 2)))
			       #f))
		       (string<? (conc time-a)(conc time-b)))))))))


;; summarize test in to a file test-summary.html in the test directory
;;
(define (tests:summarize-test area-dat run-id test-id)
  (let* ((test-dat  (rmt:get-test-info-by-id area-dat run-id test-id))
	 (steps-dat (rmt:get-steps-for-test area-dat run-id test-id))
	 (test-name (db:test-get-testname test-dat))
	 (item-path (db:test-get-item-path test-dat))
	 (full-name (db:test-make-full-name test-name item-path))
	 (oup       (open-output-file (conc (db:test-get-rundir test-dat) "/test-summary.html")))
	 (status    (db:test-get-status   test-dat))
	 (color     (common:get-color-from-status status))
	 (logf      (db:test-get-final_logf test-dat))
	 (steps-dat (tests:get-compressed-steps area-dat run-id test-id)))
    ;; (dcommon:get-compressed-steps #f 1 30045)
    ;; (#("wasting_time" "23:36:13" "23:36:21" "0" "8.0s" "wasting_time.log"))

    (s:output-new
     oup
     (s:html
      (s:title "Summary for " full-name)

	     steps-dat))
	)))
    (close-output-port oup)))
	  
	  
;; MUST BE CALLED local!
;;
(define (tests:test-get-paths-matching area-dat keynames target fnamepatt #!key (res '()))
  ;; BUG: Move the values derived from args to parameters and push to megatest.scm
  (let* ((testpatt   (or (args:get-arg "-testpatt")(args:get-arg "-testpatt") "%"))
	 (statepatt  (or (args:get-arg "-state")   (args:get-arg ":state")    "%"))
	 (statuspatt (or (args:get-arg "-status")  (args:get-arg ":status")   "%"))
	 (runname    (or (args:get-arg "-runname") (args:get-arg ":runname")  "%"))
	 (paths-from-db (rmt:test-get-paths-matching-keynames-target-new area-dat keynames target res
					testpatt
					statepatt
					statuspatt
					runname)))
    (if fnamepatt
	(apply append 
	       (map (lambda (p)

	  (with-input-from-file fname
	    (lambda ()
	      (read-lines)))))))
	  

;; for each test:
;;   
(define (tests:filter-non-runnable area-dat run-id testkeynames testrecordshash)
  (let ((runnables '()))
    (for-each
     (lambda (testkeyname)
       (let* ((test-record (hash-table-ref testrecordshash testkeyname))
	      (test-name   (tests:testqueue-get-testname  test-record))
	      (itemdat     (tests:testqueue-get-itemdat   test-record))
	      (item-path   (tests:testqueue-get-item_path test-record))
	      (waitons     (tests:testqueue-get-waitons   test-record))
	      (keep-test   #t)
	      (test-id     (rmt:get-test-id area-dat run-id test-name item-path))
	      (tdat        (rmt:get-testinfo-state-status area-dat run-id test-id))) ;; (cdb:get-test-info-by-id *runremote* test-id)))
	 (if tdat
	     (begin
	       ;; Look at the test state and status
	       (if (or (and (member (db:test-get-status tdat) 
				    '("PASS" "WARN" "WAIVED" "CHECK" "SKIP"))
			    (equal? (db:test-get-state tdat) "COMPLETED"))
		       (member (db:test-get-state tdat)
				    '("INCOMPLETE" "KILLED")))
		   (set! keep-test #f))

	       ;; examine waitons for any fails. If it is FAIL or INCOMPLETE then eliminate this test
	       ;; from the runnable list
	       (if keep-test
		   (for-each (lambda (waiton)
			       ;; for now we are waiting only on the parent test
			       (let* ((parent-test-id (rmt:get-test-id area-dat run-id waiton ""))
				      (wtdat          (rmt:get-testinfo-state-status area-dat run-id test-id))) ;; (cdb:get-test-info-by-id *runremote* test-id)))
				 (if (or (and (equal? (db:test-get-state wtdat) "COMPLETED")
					      (member (db:test-get-status wtdat) '("FAIL" "ABORT")))
					 (member (db:test-get-status wtdat)  '("KILLED"))
					 (member (db:test-get-state wtdat)   '("INCOMPETE")))
				 ;; (if (or (member (db:test-get-status wtdat)
				 ;;        	 '("FAIL" "KILLED"))
				 ;;         (member (db:test-get-state wtdat)

;;======================================================================
;; test steps
;;======================================================================

;; teststep-set-status! used to be here

(define (test-get-kill-request area-dat run-id test-id) ;; run-id test-name itemdat)
  (let* ((testdat   (rmt:get-test-info-by-id area-dat run-id test-id)))
    (and testdat
	 (equal? (test:get-state testdat) "KILLREQ"))))

(define (test:tdb-get-rundat-count tdb)
  (if tdb
      (let ((res 0))
	(sqlite3:for-each-row
	 (lambda (count)
	   (set! res count))
	 tdb
	 "SELECT count(id) FROM test_rundat;")
	res))
  0)

(define (tests:update-central-meta-info area-dat run-id test-id cpuload diskfree minutes uname hostname)
  (rmt:general-call area-dat 'update-test-rundat run-id test-id (current-seconds) (or cpuload -1)(or diskfree -1) -1 (or minutes -1))
  (if (and cpuload diskfree)
      (rmt:general-call area-dat 'update-cpuload-diskfree run-id cpuload diskfree test-id))
  (if minutes 
      (rmt:general-call area-dat 'update-run-duration run-id minutes test-id))
  (if (and uname hostname)
      (rmt:general-call area-dat 'update-uname-host run-id uname hostname test-id)))
  
;; This one is for running with no db access (i.e. via rmt: internally)
(define (tests:set-full-meta-info db area-dat test-id run-id minutes work-area remtries)
;; (define (tests:set-full-meta-info test-id run-id minutes work-area)
;;  (let ((remtries 10))
  (let* ((cpuload  (get-cpu-load))
	 (diskfree (get-df (current-directory)))
	 (uname    (get-uname "-srvpio"))
	 (hostname (get-host-name)))
    (tests:update-central-meta-info area-dat run-id test-id cpuload diskfree minutes uname hostname)))
    
;; (define (tests:set-partial-meta-info test-id run-id minutes work-area)
(define (tests:set-partial-meta-info test-id run-id minutes work-area remtries)
  (let* ((cpuload  (get-cpu-load))
	 (diskfree (get-df (current-directory)))
	 (remtries 10))
    (handle-exceptions
     exn
     (if (> remtries 0)
	 (begin
	   (print-call-chain (current-error-port))
	   (debug:print-info 0 *default-log-port* "WARNING: failed to set meta info. Will try " remtries " more times")
	   (set! remtries (- remtries 1))
	   (thread-sleep! 10)
	   (tests:set-full-meta-info db area-dat test-id run-id minutes work-area (- remtries 1)))
	 (let ((err-status ((condition-property-accessor 'sqlite3 'status #f) exn)))
	   (debug:print-error 0 *default-log-port* "tried for over a minute to update meta info and failed. Giving up")
	   (debug:print 0 *default-log-port* "EXCEPTION: database probably overloaded or unreadable.")
	   (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
	   (print "exn=" (condition->list exn))
	   (debug:print 0 *default-log-port* " status:  " ((condition-property-accessor 'sqlite3 'status) exn))
	   (print-call-chain (current-error-port))))