Megatest

	stml2
	directory-utils
	(prefix mtargs args:))

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

;; This is the Megatest API. All generally "useful" routines will be wrapped or extended
;; here.

;;                       0           1              2              3
(defstruct launch:einf (pid #t)(exit-status #t)(exit-code #t)(rollup-status 0))

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

;; 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 keys runnamepatt targpatt)
  (let loop ((runsdat  (rmt:get-runs-by-patt 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 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 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 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 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 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 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))

(define (mt:discard-blocked-tests run-id failed-test tests test-records)
  (if (null? tests)
      tests
      (begin
	(debug:print-info 1 *default-log-port* "Discarding tests from " tests " that are waiting on " failed-test)
	(let loop ((testn (car tests))
		   (remt  (cdr tests))
		   (res   '()))
	  (let* ((test-dat (hash-table-ref/default test-records testn (vector #f #f '())))
		 (waitons  (vector-ref test-dat 2)))
	    ;; (print "mt:discard-blocked-tests run-id: " run-id " failed-test: " failed-test " testn: " testn " with waitons: " waitons)
	    (if (null? remt)
		(let ((new-res (reverse res)))
		  ;; (print "       new-res: " new-res)
		  new-res)
		(loop (car remt)
		      (cdr remt)
		      (if (member failed-test waitons)
			  (begin
			    (debug:print 0 *default-log-port* "Discarding test " testn "(" test-dat ") due to " failed-test)
			    res)
			  (cons testn res)))))))))

;;======================================================================
;;  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 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 'state-status-msg run-id newstate newstatus newcomment test-id))
	;; ((and newstate newstatus)
	;;  (rmt:general-call 'state-status run-id newstate newstatus test-id))
	;; (else
	;;  (if newstate   (rmt:general-call 'set-test-state   run-id newstate   test-id))
	;;  (if newstatus  (rmt:general-call 'set-test-status  run-id newstatus  test-id))
	;;  (if newcomment (rmt:general-call 'set-test-comment run-id newcomment test-id))))
	(rmt:set-state-status-and-roll-up-items 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-id-unless-completed run-id test-id newstate newstatus newcomment)
  (let* ((test-vec   (rmt:get-testinfo-state-status run-id test-id))
         (state     (vector-ref test-vec 3)))
    (if (equal? state "COMPLETED")
        #t
        (rmt:set-state-status-and-roll-up-items run-id test-id #f newstate newstatus newcomment))))

  
(define (mt:test-set-state-status-by-testname run-id test-name item-path new-state new-status new-comment)
  ;(let ((test-id (rmt:get-test-id run-id test-name item-path)))
  (rmt:set-state-status-and-roll-up-items 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:test-set-state-status-by-testname-unless-completed run-id test-name item-path new-state new-status new-comment)
  (let ((test-id (rmt:get-test-id run-id test-name item-path)))
    (mt:test-set-state-status-by-id-unless-completed run-id test-id new-state new-status new-comment)))
    
;; 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 target run-name testpatt)
  (let ((records    (rmt:tasks-find-task-queue-records 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))
	      (match-dat (string-search hostpid-rx param-key)))
	 (if match-dat
	     (let ((hostname  (cadr match-dat))
		   (pid       (string->number (caddr match-dat))))
	       (debug:print 0 *default-log-port* "Sending SIGINT to process " pid " on host " hostname)
	       (if (equal? (get-host-name) hostname)
		   (if (process:alive? pid)
		       (begin
			 (handle-exceptions
			  exn
			  (begin
			    (debug:print 0 *default-log-port* "Kill of process " pid " on host " hostname " failed.")
			    (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
			    #t)
			  (process-signal pid signal/int)
			  (thread-sleep! 5)
			  (if (process:alive? pid)
			      (process-signal pid signal/kill)))))
		   ;;  (call-with-environment-variables
		   (let ((old-targethost (getenv "TARGETHOST")))
		     (setenv "TARGETHOST" hostname)
		     (setenv "TARGETHOST_LOGF" "server-kills.log")
		     (system (conc "nbfake kill " pid))
		     (if old-targethost (setenv "TARGETHOST" old-targethost))
		     (unsetenv "TARGETHOST")
		     (unsetenv "TARGETHOST_LOGF"))))
	     (debug:print-error 0 *default-log-port* "no record or improper record for " target "/" run-name " in tasks_queue in main.db"))))
     records)))

(define (task:get-run-times)
   (let* ( 
           (run-patt (if (args:get-arg "-run-patt")
                        (args:get-arg "-run-patt")
                        "%"))
           (target-patt (if (args:get-arg "-target-patt")
                        (args:get-arg "-target-patt")
                        "%"))
 
           (run-times  (rmt:get-run-times  run-patt target-patt )))
   (if (eq? (length run-times) 0)
     (begin
       (print "Data not found!!")
       (exit)))
   (if (equal? (args:get-arg "-dumpmode") "json")
       (task:print-runtime-as-json run-times)
         (if (equal? (args:get-arg "-dumpmode") "csv")
	     (task:print-runtime run-times ",")
	     (task:print-runtime run-times "  ")))))

 (define (task:get-test-times)
   (let* ((runname (if (args:get-arg "-runname")
                        (args:get-arg "-runname")
                        #f))
           (target (if (args:get-arg "-target")
                        (args:get-arg "-target")
                        #f))
 
           (test-times  (rmt:get-test-times  runname target )))
   (if (not runname)
      (begin
      (print "Error: Missing argument -runname")
      (exit))) 
    (if (string-contains runname "%")
      (begin
      (print "Error: Invalid runname, '%' not allowed  (" runname ") ")
      (exit)))
    (if (not target)
      (begin
      (print "Error: Missing argument -target")
      (exit)))
     (if  (string-contains target "%")
      (begin
      (print "Error: Invalid target, '%' not allowed  (" target ") ")
      (exit)))
 
   (if (eq? (length test-times) 0)
     (begin
       (print "Data not found!!")
       (exit)))
   (if (equal? (args:get-arg "-dumpmode") "json")
       (task:print-testtime-as-json test-times)
         (if (equal? (args:get-arg "-dumpmode") "csv")
	     (task:print-testtime test-times ",")
	     (task:print-testtime test-times "  ")))))



;; gets mtpg-run-id and syncs the record if different
;;
(define (tasks:run-id->mtpg-run-id dbh cached-info run-id area-info smallest-last-update-time)
  (let* ((runs-ht (hash-table-ref cached-info 'runs))
	 (runinf  (hash-table-ref/default runs-ht run-id #f))
         (area-id (vector-ref area-info 0)))
       (if runinf
	runinf ;; already cached
	(let* ((run-dat    (rmt:get-run-info run-id))               ;; NOTE: get-run-info returns a vector < row header >
	       (run-name   (rmt:get-run-name-from-id run-id))
	       (row        (db:get-rows run-dat))                   ;; yes, this returns a single row
	       (header     (db:get-header run-dat))
	       (state      (db:get-value-by-header row header "state"))
	       (status     (db:get-value-by-header row header "status"))
	       (owner      (db:get-value-by-header row header "owner"))
	       (event-time (db:get-value-by-header row header "event_time"))
	       (comment    (db:get-value-by-header row header "comment"))
	       (fail-count (db:get-value-by-header row header "fail_count"))
	       (pass-count (db:get-value-by-header row header "pass_count"))
         (db-contour (db:get-value-by-header row header "contour"))
	       (contour    (if (args:get-arg "-prepend-contour") 
                                 (if (and db-contour (not (equal? db-contour ""))  (string? db-contour )) 
                                           (begin 
                                            (debug:print-info 1 *default-log-port*  "db-contour") 
 						db-contour)
					    (args:get-arg "-contour"))))
         (run-tag (if (args:get-arg "-run-tag")
                            (args:get-arg "-run-tag")
									""))
         (last-update (db:get-value-by-header row header "last_update"))
	       (keytarg    (if (or (args:get-arg "-prepend-contour") (args:get-arg "-prefix-target"))
	       			(conc "MT_CONTOUR/MT_AREA/" (string-intersperse (rmt:get-keys) "/")) (string-intersperse (rmt:get-keys) "/"))) ;; e.g. version/iteration/platform
	       (target     (if (or (args:get-arg "-prepend-contour") (args:get-arg "-prefix-target")) 
	       			(conc (or (args:get-arg "-prefix-target") (conc contour "/" (common:get-area-name) "/")) (rmt:get-target run-id)) (rmt:get-target run-id)))                 ;; e.g. v1.63/a3e1/ubuntu
	       (spec-id    (pgdb:get-ttype dbh keytarg))
	       (publish-time (if (args:get-arg "-cp-eventtime-to-publishtime")
                            event-time
                           (current-seconds))) 
	       (new-run-id (pgdb:get-run-id dbh spec-id target run-name area-id)))
         (if new-run-id
	         (begin ;; let ((run-record (pgdb:get-run-info dbh new-run-id))
		        (hash-table-set! runs-ht run-id new-run-id)
		;; ensure key fields are up to date
     ;; if last_update == pgdb_last_update do not update smallest-last-update-time  
    (let* ((pgdb-last-update (pgdb:get-run-last-update dbh new-run-id))
           (smallest-time (hash-table-ref/default smallest-last-update-time "smallest-time" #f)))
     (if (and  (> last-update pgdb-last-update) (or (not smallest-time) (< last-update smallest-time)))
        (hash-table-set! smallest-last-update-time "smallest-time" last-update)))
		(pgdb:refresh-run-info
		 dbh
		 new-run-id
		 state status owner event-time comment fail-count pass-count area-id last-update publish-time)
     (debug:print-info 0 *default-log-port* "Working on run-id " run-id " pgdb-id "  new-run-id )
     (if (not (equal? run-tag ""))
      (task:add-run-tag dbh new-run-id run-tag))
		new-run-id) 
      
	      (if (equal? state "deleted")
          (begin 
          (debug:print-info 1 *default-log-port*  "Warning: Run with id " run-id " was created after previous sync and deleted before the sync") #f)
          (if (handle-exceptions
		        exn
		        (begin (print-call-chain)
              (print ((condition-property-accessor 'exn 'message) exn))     
			      #f)
            
            (pgdb:insert-run
		     dbh
		     spec-id target run-name state status owner event-time comment fail-count pass-count  area-id last-update publish-time))
		       (let* ((smallest-time (hash-table-ref/default smallest-last-update-time "smallest-time" #f)))
             (if (or (not smallest-time) (< last-update smallest-time))
        				(hash-table-set! smallest-last-update-time "smallest-time" last-update))
             (tasks:run-id->mtpg-run-id dbh cached-info run-id area-info smallest-last-update-time))
		  #f)))))))

(define (task:add-run-tag dbh run-id tag) 
  (let* ((tag-info (pgdb:get-tag-info-by-name dbh tag)))
   (if (not tag-info)
     (begin   
     (if (handle-exceptions
	   exn
	   (begin 
               (debug:print-info 1 *default-log-port*  ((condition-property-accessor 'exn 'message) exn))     
	   #f)
	   (pgdb:insert-tag  dbh   tag))
                       (set! tag-info (pgdb:get-tag-info-by-name dbh tag))
		  #f)))
     ;;add to area_tags
     (handle-exceptions
	   exn
	   (begin 
               (debug:print-info 1 *default-log-port*  ((condition-property-accessor 'exn 'message) exn))     
	   #f)
           (if (not (pgdb:is-run-taged-with-a-tag dbh (vector-ref tag-info 0)  run-id))  
	   (pgdb:insert-run-tag  dbh   (vector-ref tag-info 0)  run-id)))))


(define (tasks:sync-test-steps dbh cached-info test-step-ids smallest-last-update-time)
 ; (print "Sync Steps " test-step-ids )
  (let ((test-ht (hash-table-ref cached-info 'tests))
        (step-ht (hash-table-ref cached-info 'steps)))
    (for-each
     (lambda (test-step-id)
        (let* ((test-step-info  (rmt:get-steps-info-by-id test-step-id))
               (step-id (tdb:step-get-id test-step-info))
               (test-id  (tdb:step-get-test_id    test-step-info))   
	       (stepname (tdb:step-get-stepname  test-step-info))
	       (state (tdb:step-get-state test-step-info))	
	       (status (tdb:step-get-status test-step-info))	
	       (event_time (tdb:step-get-event_time  test-step-info))	
	       (comment  (tdb:step-get-comment test-step-info))	
	       (logfile (tdb:step-get-logfile test-step-info))	
         (last-update (tdb:step-get-last_update test-step-info))
	       (pgdb-test-id  (hash-table-ref/default test-ht test-id #f))
				 (smallest-time (hash-table-ref/default smallest-last-update-time "smallest-time" #f))
         (pgdb-step-id (if pgdb-test-id 
                         (pgdb:get-test-step-id dbh pgdb-test-id stepname state)
                          #f)))
    (if step-id
      (begin  
        (if pgdb-test-id
           (begin 
                (if  pgdb-step-id
                   (begin
                    (debug:print-info 1 *default-log-port*  "Updating existing test-step with test-id: " test-id " and step-id " step-id " pgdb test id: " pgdb-test-id " pgdb step id " pgdb-step-id )
										(let* ((pgdb-last-update (pgdb:get-test-step-last-update dbh pgdb-step-id)))
         (if (and  (> last-update pgdb-last-update) (or (not smallest-time) (< last-update smallest-time)))
        (hash-table-set! smallest-last-update-time "smallest-time" last-update))) 
                    (pgdb:update-test-step dbh pgdb-step-id pgdb-test-id stepname state status event_time comment logfile last-update))
                    (begin
 		      (debug:print-info 1 *default-log-port*  "Inserting test-step with test-id: " test-id " and step-id " step-id  " pgdb test id: " pgdb-test-id)
                     (if (or (not smallest-time) (< last-update smallest-time))
        				      (hash-table-set! smallest-last-update-time "smallest-time" last-update))
                      (pgdb:insert-test-step dbh pgdb-test-id stepname state status event_time comment logfile last-update )
                      (set! pgdb-step-id  (pgdb:get-test-step-id dbh pgdb-test-id stepname state))))
                (hash-table-set! step-ht step-id pgdb-step-id ))
           (debug:print-info 1 *default-log-port*  "Error: Test not cashed")))
      (debug:print-info 1 *default-log-port*  "Error: Could not get test step info for step id " test-step-id ))))	;; this is a wierd senario need to debug      	
   test-step-ids)))

(define (tasks:sync-test-gen-data dbh cached-info test-data-ids smallest-last-update-time)
  (let ((test-ht (hash-table-ref cached-info 'tests))
        (data-ht (hash-table-ref cached-info 'data)))
    (for-each
     (lambda (test-data-id)
        (let* ((test-data-info  (rmt:get-data-info-by-id test-data-id))
               (data-id (db:test-data-get-id  test-data-info))
               (test-id  (db:test-data-get-test_id   test-data-info))   
	       (category  (db:test-data-get-category  test-data-info))
	       (variable  (db:test-data-get-variable test-data-info))	
	       (value (db:test-data-get-value  test-data-info))	
               (expected (db:test-data-get-expected  test-data-info))
               (tol (db:test-data-get-tol  test-data-info))
               (units (db:test-data-get-units  test-data-info))     
	       (comment  (db:test-data-get-comment test-data-info))	
               (status (db:test-data-get-status test-data-info))	
	       (type (db:test-data-get-type test-data-info))
				 (last-update (db:test-data-get-last_update test-data-info))
				 (smallest-time (hash-table-ref/default smallest-last-update-time "smallest-time" #f))
   	
	       (pgdb-test-id  (hash-table-ref/default test-ht test-id #f))
               (pgdb-data-id (if pgdb-test-id 
                                 (pgdb:get-test-data-id dbh pgdb-test-id category variable)
                                  #f)))
    (if data-id
      (begin
        (if pgdb-test-id
           (begin 
                (if  pgdb-data-id
                   (begin
                    (debug:print-info 1 *default-log-port*  "Updating existing test-data with test-id: " test-id " and  data-id " data-id " pgdb test id: " pgdb-test-id " pgdb data id " pgdb-data-id)
                    (let* ((pgdb-last-update (pgdb:get-test-data-last-update dbh pgdb-data-id)))
         (if (and  (>  last-update pgdb-last-update) (or (not smallest-time) (< last-update smallest-time)))
        (hash-table-set! smallest-last-update-time "smallest-time" last-update))) 
                    (pgdb:update-test-data dbh pgdb-data-id pgdb-test-id  category variable value expected tol units comment status type last-update))
                    (begin
 		      (debug:print-info 1 *default-log-port*  "Inserting test-data with test-id: " test-id " and data-id " data-id " pgdb test id: " pgdb-test-id)
                       (if (handle-exceptions
		      exn
		      (begin (print-call-chain)
                              (print ((condition-property-accessor 'exn 'message) exn))     
			#f)
                     
                    (pgdb:insert-test-data dbh pgdb-test-id category variable value expected tol units comment status type last-update))
		       ;(tasks:run-id->mtpg-run-id dbh cached-info run-id area-info)
                      (begin
                      ;(pgdb:insert-test-data dbh pgdb-test-id category variable value expected tol units comment status type )
											(if (or (not smallest-time) (< last-update smallest-time))
        								(hash-table-set! smallest-last-update-time "smallest-time" last-update))
                      (set! pgdb-data-id  (pgdb:get-test-data-id dbh pgdb-test-id  category variable)))
		   #f)))
                (hash-table-set! data-ht data-id pgdb-data-id ))
             (begin
                 (debug:print-info 1 *default-log-port*  "Error: Test not in pgdb"))))

      (debug:print-info 1 *default-log-port*  "Error: Could not get test data info for data id " test-data-id ))))	;; this is a wierd senario need to debug      	
   test-data-ids)))



(define (tasks:sync-tests-data dbh cached-info test-ids area-info smallest-last-update-time)
  (let ((test-ht (hash-table-ref cached-info 'tests)))
    (for-each
     (lambda (test-id)
      ; (print test-id)
       (let* ((test-info    (rmt:get-test-info-by-id #f test-id))
	      (run-id       (db:test-get-run_id    test-info)) ;; look these up in db_records.scm
	      (test-id      (db:test-get-id        test-info))
	      (test-name    (db:test-get-testname  test-info))
	      (item-path    (db:test-get-item-path test-info))
	      (state        (db:test-get-state     test-info))
	      (status       (db:test-get-status    test-info))
	      (host         (db:test-get-host      test-info))
        (pid          (db:test-get-process_id test-info)) 
	      (cpuload      (db:test-get-cpuload   test-info))
	      (diskfree     (db:test-get-diskfree  test-info))
	      (uname        (db:test-get-uname     test-info))
	      (run-dir      (db:test-get-rundir    test-info))
	      (log-file     (db:test-get-final_logf test-info))
	      (run-duration (db:test-get-run_duration test-info))
	      (comment      (db:test-get-comment   test-info))
	      (event-time   (db:test-get-event_time test-info))
	      (archived     (db:test-get-archived  test-info))
        (last-update  (db:test-get-last_update  test-info))
	      (pgdb-run-id  (tasks:run-id->mtpg-run-id dbh cached-info run-id area-info smallest-last-update-time))
        (smallest-time (hash-table-ref/default smallest-last-update-time "smallest-time" #f))       
	      (pgdb-test-id (if pgdb-run-id 
				(begin
                                  ;(print pgdb-run-id)    
                                 (pgdb:get-test-id dbh pgdb-run-id test-name item-path))
                                 #f)))
	 ;; "id"           "run_id"        "testname"  "state"      "status"      "event_time"
	 ;; "host"         "cpuload"       "diskfree"  "uname"      "rundir"      "item_path"
	 ;; "run_duration" "final_logf"    "comment"   "shortdir"   "attemptnum"  "archived"
         (if pgdb-run-id
           (begin
	   (if pgdb-test-id ;; have a record
	     (begin ;; let ((key-name (conc run-id "/" test-name "/" item-path)))
	       (debug:print-info 0 *default-log-port*  "Updating existing test with run-id: " run-id " and test-id: " test-id " pgdb run id: " pgdb-run-id "  pgdb-test-id "  pgdb-test-id)
         (let* ((pgdb-last-update (pgdb:get-test-last-update dbh pgdb-test-id)))
         (if (and  (>  last-update pgdb-last-update) (or (not smallest-time) (< last-update smallest-time))) ;;if last-update is same as pgdb-last-update then it is safe to assume the records are identical and we can use a larger last update time.
        (hash-table-set! smallest-last-update-time "smallest-time" last-update))) 
	       (pgdb:update-test dbh pgdb-test-id pgdb-run-id test-name item-path state status host cpuload diskfree uname run-dir log-file run-duration comment event-time archived last-update pid))
	     (begin 
           (debug:print-info 0 *default-log-port*  "Inserting test with run-id: " run-id " and test-id: " test-id  " pgdb run id: " pgdb-run-id)
           (pgdb:insert-test dbh pgdb-run-id test-name item-path state status host cpuload diskfree uname run-dir log-file run-duration comment event-time archived last-update pid)
            (if (or (not smallest-time) (< last-update smallest-time))
        				(hash-table-set! smallest-last-update-time "smallest-time" last-update))
           (set! pgdb-test-id (pgdb:get-test-id dbh pgdb-run-id test-name item-path))))
           (hash-table-set! test-ht test-id pgdb-test-id))
           (debug:print-info 1 *default-log-port*  "WARNING: Skipping run with run-id:" run-id ". This run was created after privious sync and removed before this sync."))))
     test-ids)))

(define (task:add-area-tag dbh area-info tag) 
  (let* ((tag-info (pgdb:get-tag-info-by-name dbh tag)))
   (if (not tag-info)
     (begin   
     (if (handle-exceptions
	   exn
	   (begin 
               (debug:print-info 1 *default-log-port*  ((condition-property-accessor 'exn 'message) exn))     
	   #f)
	   (pgdb:insert-tag  dbh   tag))
                       (set! tag-info (pgdb:get-tag-info-by-name dbh tag))
		  #f)))
     ;;add to area_tags
     (handle-exceptions
	   exn
	   (begin 
               (debug:print-info 1 *default-log-port*  ((condition-property-accessor 'exn 'message) exn))     
	   #f)
           (if (not (pgdb:is-area-taged-with-a-tag dbh (vector-ref tag-info 0)  (vector-ref area-info 0)))  
	   (pgdb:insert-area-tag  dbh   (vector-ref tag-info 0)  (vector-ref area-info 0))))))

(define (tasks:sync-run-data dbh cached-info run-ids area-info smallest-last-update-time) 
  (for-each
     (lambda (run-id)
      (debug:print-info 1 *default-log-port*   "Check if run with " run-id " needs to be synced" )
       (tasks:run-id->mtpg-run-id dbh cached-info run-id area-info smallest-last-update-time))
run-ids))


;; get runs changed since last sync
;; (define (tasks:sync-test-data dbh cached-info area-info)
;;   (let* ((

(define (tasks:sync-to-postgres configdat dest)
  (print "In sync")
  (let* ((dbh         (pgdb:open configdat dbname: dest))
	 (area-info   (pgdb:get-area-by-path dbh *toppath*))
	 (cached-info (make-hash-table))
	 (start       (current-seconds))
   (test-patt   (if (args:get-arg "-testpatt")
											(args:get-arg "-testpatt")
                      "%"))
   (target         (if (args:get-arg "-target")
														 (args:get-arg "-target")
													#f))
    (run-name         (if (args:get-arg "-runname")
														 (args:get-arg "-runname")
													#f)))
     (if (and target  (not run-name))
       (begin
					(print "Error: Provide runname")
          (exit 1)))
     (if (and (not target)  run-name)
       (begin
					(print "Error: Provide target")
          (exit 1)))
    ;(print "123")
    ;(exit 1) 
    (for-each (lambda (dtype)
		(hash-table-set! cached-info dtype (make-hash-table)))
	      '(runs targets tests steps data))
    (hash-table-set! cached-info 'start start) ;; when done we'll set sync times to this
    (if area-info
	(let* ((last-sync-time (vector-ref area-info 3))
	       (smallest-last-update-time  (make-hash-table))
         (changed      (if (and target run-name)
                            (rmt:get-run-record-ids target run-name (rmt:get-keys) test-patt)
                            (rmt:get-changed-record-ids last-sync-time)))
	       (run-ids        (alist-ref 'runs       changed))
	       (test-ids       (alist-ref 'tests      changed))
	       (test-step-ids  (alist-ref 'test_steps changed))
	       (test-data-ids  (alist-ref 'test_data  changed))
	       (run-stat-ids   (alist-ref 'run_stats  changed))
         (area-tag    (if (args:get-arg "-area-tag") 
                                 (args:get-arg "-area-tag")
                                 (if (args:get-arg "-area") 
                                   (args:get-arg "-area") 
                                   ""))))
           (if (and (equal? area-tag "") (not (pgdb:is-area-taged dbh (vector-ref area-info 0))))
            (set! area-tag *default-area-tag*)) 
           (if (not (equal? area-tag "")) 
             (task:add-area-tag dbh area-info area-tag)) 
	  (if (or (not (null? test-ids)) (not (null? run-ids)))
	      (begin
                (debug:print-info 0 *default-log-port*  "syncing runs")   
	              (tasks:sync-run-data dbh cached-info run-ids area-info smallest-last-update-time) 
                (debug:print-info 0 *default-log-port*  "syncing tests")
		            (tasks:sync-tests-data dbh cached-info test-ids area-info smallest-last-update-time)
                (debug:print-info 0 *default-log-port*  "syncing test steps")
                (tasks:sync-test-steps dbh cached-info test-step-ids smallest-last-update-time)
								(debug:print-info 0 *default-log-port*  "syncing test data")
                (tasks:sync-test-gen-data dbh cached-info test-data-ids smallest-last-update-time)
                (print "----------done---------------")))
     (let*  ((smallest-time (hash-table-ref/default smallest-last-update-time "smallest-time" #f)))
     (debug:print-info 0 "smallest-time :" smallest-time  " last-sync-time " last-sync-time)
    (if (not (and target run-name)) 
	  (if (or (and smallest-time (> smallest-time last-sync-time)) (and smallest-time (eq? last-sync-time 0)))
				(pgdb:write-sync-time dbh area-info smallest-time))))) ;;this needs to be changed
	(if (tasks:set-area dbh configdat)
	    (tasks:sync-to-postgres configdat dest)
	    (begin
	      (debug:print 0 *default-log-port* "ERROR: unable to create an area record")
	      #f)))))

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

;; faux-lock is deprecated. Please use simple-lock below
;;
(define (common:faux-lock keyname #!key (wait-time 8)(allow-lock-steal #t))
  (if (rmt:no-sync-get/default keyname #f) ;; do not be tempted to compare to pid. locking is a one-shot action, if already locked for this pid it doesn't actually count
      (if (> wait-time 0)
	  (begin
	    (thread-sleep! 1)
	    (if (eq? wait-time 1) ;; only one second left, steal the lock
		(begin
		  (debug:print-info 0 *default-log-port* "stealing lock for " keyname)
		  (common:faux-unlock keyname force: #t)))
	    (common:faux-lock keyname wait-time: (- wait-time 1)))
	  #f)
      (begin
        (rmt:no-sync-set keyname (conc (current-process-id)))
        (equal? (conc (current-process-id)) (conc (rmt:no-sync-get/default keyname #f))))))

(define (common:faux-unlock keyname #!key (force #f))
  (if (or force (equal? (conc (current-process-id)) (conc (rmt:no-sync-get/default keyname #f))))
      (begin
        (if (rmt:no-sync-get/default keyname #f) (rmt:no-sync-del! keyname))
        #t)
      #f))

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

(define (common:simple-unlock keyname #!key (force #f))
  (rmt:no-sync-del! keyname))

;;======================================================================
;; db based host calls
;;======================================================================

;;======================================================================
;;  T E S T   L A U N C H I N G   P E R   I T E M   W I T H   H O S T   T Y P E S
;;======================================================================
;; 
;; [hosts]
;; arm cubie01 cubie02
;; x86_64 zeus xena myth01
;; allhosts #{g hosts arm} #{g hosts x86_64}
;; 
;; [host-types]
;; general #MTLOWESTLOAD #{g hosts allhosts}
;; arm     #MTLOWESTLOAD #{g hosts arm}
;; nbgeneral nbjob run JOBCOMMAND -log $MT_LINKTREE/$MT_TARGET/$MT_RUNNAME.$MT_TESTNAME-$MT_ITEM_PATH.lgo
;;
;; [host-rules]
;; # maxnload   => max normalized load
;; # maxnjobs   => max jobs per cpu
;; # maxjobrate => max jobs per second
;; general maxnload=1.1; maxnjobs=1.2; maxjobrate=0.1 
;; 
;; [launchers]
;; envsetup general
;; xor/%/n 4C16G
;; % nbgeneral
;; 
;; [jobtools]
;; # if defined and not "no" flexi-launcher will bypass "launcher" unless no match.
;; flexi-launcher yes  
;; launcher nbfake
;;
(define (common:get-launcher configdat testname itempath)
  (let ((fallback-launcher (configf:lookup configdat "jobtools" "launcher")))
    (if (and (configf:lookup configdat "jobtools" "flexi-launcher") ;; overrides launcher
	     (not (equal? (configf:lookup configdat "jobtools" "flexi-launcher") "no")))
	(let* ((launchers         (hash-table-ref/default configdat "launchers" '())))
	  (if (null? launchers)
	      fallback-launcher
	      (let loop ((hed (car launchers))
			 (tal (cdr launchers)))
		(let ((patt      (car hed))
		      (host-type (cadr hed)))
		  (if (tests:match patt testname itempath)
		      (begin
			(debug:print-info 2 *default-log-port* "Have flexi-launcher match for " testname "/" itempath " = " host-type)
			(let ((launcher (configf:lookup configdat "host-types" host-type)))
			  (if launcher
			      (let* ((launcher-parts (string-split launcher))
				     (launcher-exe   (car launcher-parts)))
				(if (equal? launcher-exe "#MTLOWESTLOAD") ;; this is our special case, we will find the lowest load and craft a nbfake commandline
				    (let host-loop ((targ-host (common:get-least-loaded-host (cdr launcher-parts) host-type configdat))
						    (count     100))
				      (if targ-host
					  (conc "remrun " targ-host)
					  (if (> count 0)
					      (begin
						(debug:print 0 *default-log-port* "INFO: Waiting for a host for host-type " host-type)
						(thread-sleep! (- 101 count))
						(host-loop (common:get-least-loaded-host (cdr launcher-parts) host-type configdat)
							   (- count 1)))
					      (begin
						(debug:print 0 *default-log-port* "FATAL: Failed to find a host from #MTLOWESTLOAD for host-type " host-type)
						(exit)))))
				    launcher))
			      (begin
				(debug:print-info 0 *default-log-port* "WARNING: no launcher found for host-type " host-type)
				(if (null? tal)
				    fallback-launcher
				    (loop (car tal)(cdr tal)))))))
		      ;; no match, try again
		      (if (null? tal)
			  fallback-launcher
			  (loop (car tal)(cdr tal))))))))
	fallback-launcher)))

;; 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)
  (let* ((loadinfo                         (rmt:get-latest-host-load hostname)) ;; if this host happens to have been recently used by a test reuse the load data
         (load                             (car loadinfo))
         (load-sample-time                 (cdr loadinfo))
         (load-sample-age                  (- (current-seconds) load-sample-time))
         (loadinfo-timeout-seconds         6) ;; this was 20 seconds, seems way too lax. Switch to 6 seconds
         (host-last-update-timeout-seconds 4)
         (host-rec (hash-table-ref/default *host-loads* hostname #f))
         )
    (cond
     ((< load-sample-age loadinfo-timeout-seconds)
      (list #t
            load-sample-time
            load))
     ((and host-rec
           (< (current-seconds) (+ (host-last-update host-rec) host-last-update-timeout-seconds)))
      (list #t
            (host-last-update host-rec)
            (host-last-cpuload host-rec )))
     ((common:unix-ping hostname)
      (list #t
            (current-seconds)
            (alist-ref 'adj-core-load (common:get-normalized-cpu-load hostname)))) ;; this is cheaper than you might think. get-normalized-cpu-load is cached for up to 5 seconds
     (else
      (list #f 0 -1) ;; bad host, don't use!
      ))))

;; see defstruct host at top of file.
;;    host: reachable last-update last-used last-cpuload
;;
(define (common:update-host-loads-table hosts-raw)
  (let* ((hosts (filter (lambda (x)
                          (string-match (regexp "^\\S+$") x))
                        hosts-raw)))
    (for-each
     (lambda (hostname)
       (let* ((rec       (let ((h (hash-table-ref/default *host-loads* hostname #f)))
                          (if h
                              h
                              (let ((h (make-host)))
                                (hash-table-set! *host-loads* hostname h)
                                h))))
              (host-info         (common:get-host-info hostname))
              (is-reachable      (car host-info))
              (last-reached-time (cadr host-info))
              (load              (caddr host-info)))
         (host-reachable-set!    rec is-reachable)
         (host-last-update-set!  rec last-reached-time)
         (host-last-cpuload-set! rec load)))
     hosts)))

;; go through the hosts from least recently used to most recently used, pick the first that meets the load criteral from the
;; [host-rules] section.
;;
(define (common:get-least-loaded-host hosts-raw host-type configdat)
  (let* ((rdat       (configf:lookup configdat "host-rules" host-type))
	 (rules      (common:val->alist (or rdat "") convert: #t))   ;; maxnload, maxnjobs, maxjobrate
	 (maxnload   (common:alist-ref/default 'maxnload rules 1.5)) ;; max normalized load
	 (maxnjobs   (common:alist-ref/default 'maxnjobs rules 1.5)) ;; max normalized number of jobs
	 (maxjobrate (common:alist-ref/default 'maxjobrate rules (/ 1 6))) ;; max rate of submitting jobs to a given host in jobs/second
	 (hosts      (filter (lambda (x)
			       (string-match (regexp "^\\S+$") x))
			     hosts-raw))
         ;; (best-host #f)
	 (get-rec    (lambda (hostname)
		       ;; (print "get-rec hostname=" hostname)
		       (let ((h (hash-table-ref/default *host-loads* hostname #f)))
			 (if h
			     h
			     (let ((h (make-host)))
			       (hash-table-set! *host-loads* hostname h)
			       h)))))
         (best-load 99999)
         (curr-time (current-seconds))
	 (get-hosts-sorted (lambda (hosts)
			     (sort hosts (lambda (a b)
					   (let ((a-rec (get-rec a))
						 (b-rec (get-rec b)))
					     ;; (print "a=" a " a-rec=" a-rec " host-last-used=" (host-last-used a-rec))
					     ;; (print "b=" b " b-rec=" b-rec " host-last-used=" (host-last-used b-rec))
					     (< (host-last-used a-rec)
						(host-last-used b-rec))))))))
    (debug:print 0 *default-log-port* "INFO: hosts-sorted=" (get-hosts-sorted hosts))
    (if (null? hosts)
	#f ;; no hosts to select from. All done and giving up now.
	(let ((hosts-sorted (get-hosts-sorted hosts)))
	  (common:update-host-loads-table hosts)
	  (let loop ((hostname  (car hosts-sorted))
		     (tal       (cdr hosts-sorted))
		     (best-host #f))
	    (let* ((rec       (get-rec hostname))
		   (reachable (host-reachable     rec))
		   (load      (host-last-cpuload  rec))
		   (last-used (host-last-used     rec))
		   (delta     (- curr-time last-used))
		   (job-rate  (if (> delta 0)
				  (/ 1 delta)
				  999)) ;; jobs per second
		   (new-best  
		    (cond
		     ((not reachable)
		      (debug:print 0 *default-log-port* "Skipping host " hostname " as it cannot be reached.")
		      best-host)
		     ((and (< load maxnload)        ;; load is acceptable
			   (< job-rate maxjobrate)) ;; job rate is acceptable
		      (set! best-load load)
		      hostname)
		     (else best-host))))
	      (debug:print 0 *default-log-port* "INFO: Trying host " hostname " with load " load ", last used " delta " seconds ago, with job-rate " job-rate " for running a test." )
	      (if new-best
		  (begin ;; found a host, return it
		    (debug:print 0 *default-log-port* "INFO: Found host: " new-best " load: " load " last-used: " delta " seconds ago, with job-rate: " job-rate)
		    (host-last-used-set! rec curr-time)
		    new-best)
		  (if (null? tal) #f (loop (car tal)(cdr tal) best-host)))))))))

(define (common:wait-for-cpuload maxload-in numcpus-in waitdelay #!key (count 1000) (msg #f)(remote-host #f)(force-maxload #f))
  (let* ((loadavg (common:get-cpu-load remote-host))
	 (numcpus (if (<= 1 numcpus-in) ;; not possible to have zero.  If we get 1, it's possible that we got the previous default, and we should check again
		      (common:get-num-cpus remote-host)
		      numcpus-in))
	 (maxload (if force-maxload
		      maxload-in
		      (max maxload-in 0.5))) ;; so maxload must be greater than 0.5 for now BUG - FIXME?
	 (first   (car loadavg))
	 (next    (cadr loadavg))
	 (adjload (* maxload (max 1 numcpus))) ;; possible bug where numcpus (or could be maxload) is zero, crude fallback is to at least use 1
	 (loadjmp (- first next))
         (adjwait (min (+ 300 (random 10)) (abs (* (+ (random 10)(/ (- 1000 count) 10) waitdelay) (- first adjload) ))  ))) ;; add some randomness to the time to break any alignment where netbatch dumps many jobs to machines simultaneously
    (debug:print-info 1 *default-log-port* "Checking cpuload on " (or remote-host "localhost") ", maxload: " maxload
		      ", load: " first ", adjload: " adjload ", loadjmp: " loadjmp)
    (cond
     ((and (> first adjload)
	   (> count 0))
      (debug:print-info 0 *default-log-port* "server start delayed " adjwait " seconds due to load " first " exceeding max of " adjload " on server " (or remote-host (get-host-name)) " (normalized load-limit: " maxload ") " (if msg msg ""))
      (thread-sleep! adjwait)
      (common:wait-for-cpuload maxload numcpus waitdelay count: (- count 1) msg: msg remote-host: remote-host))
     ((and (> loadjmp numcpus)
	   (> count 0))
      (debug:print-info 0 *default-log-port* "waiting " adjwait " seconds due to load jump " loadjmp " > numcpus " numcpus (if msg msg ""))
      (thread-sleep! adjwait)
      (common:wait-for-cpuload maxload numcpus waitdelay count: (- count 1) msg: msg remote-host: remote-host)))))

(define (common:wait-for-homehost-load maxload msg)
  (let* ((hh-dat (if (common:on-homehost?) ;; if we are on the homehost then pass in #f so the calls are local.
                     #f
                     (common:get-homehost)))
         (hh     (if hh-dat (car hh-dat) #f))
         (numcpus (common:get-num-cpus hh)))
    (common:wait-for-normalized-load maxload msg hh)))

(define (common:get-num-cpus remote-host)
  (let* ((actual-host (or remote-host (get-host-name))))
    (or (common:get-cached-info actual-host "num-cpus" age: 86400) ;; hosts had better not be changing the number of cpus too often!
	(let* ((proc   (lambda ()
			 (let loop ((numcpu 0)
				    (inl    (read-line)))
			   (if (eof-object? inl)
			       (begin
				 (common:write-cached-info remote-host "num-cpus" numcpu)
				 numcpu)
			       (loop (if (string-match "^processor\\s+:\\s+\\d+$" inl)
					 (+ numcpu 1)
					 numcpu)
				     (read-line))))))
	       (result (if remote-host
			   (with-input-from-pipe 
			       (conc "ssh " remote-host " cat /proc/cpuinfo")
			     proc)
			   (with-input-from-file "/proc/cpuinfo" proc))))
	  (common:write-cached-info actual-host "num-cpus" result)
	  result))))

;; wait for normalized cpu load to drop below maxload
;;
(define (common:wait-for-normalized-load maxload msg remote-host)
  (let ((num-cpus (common:get-num-cpus remote-host)))
    (common:wait-for-cpuload maxload num-cpus 15 msg: msg remote-host: remote-host)))

;;======================================================================
;; D E B U G G I N G   S T U F F 
;;======================================================================

;; (define *verbosity*         1)
;; (define *logging*           #f)

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

;; postive number if megatest version > db version
;; negative number if megatest version < db version
(define (common:version-db-delta)
         (- megatest-version (common:get-last-run-version-number)))

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

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

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

(define (common:api-changed?)
  (not (equal? (substring (->string megatest-version) 0 4)
               (substring (conc (common:get-last-run-version)) 0 4))))

;; '(print (string-intersperse (map cadr (hash-table-ref/default (read-config "megatest.config" \#f \#t) "disks" '"'"'("none" ""))) "\n"))'
(define (common:get-disks #!key (configf #f))
  (hash-table-ref/default 
   (or configf (configf:read-config "megatest.config" #f #t))
   "disks" '("none" "")))

;;======================================================================
;; watchdog and exit procedures
;;======================================================================

;;======================================================================
;; E X I T   H A N D L I N G
;;======================================================================

;;   (let ((ohh (common:on-homehost?))
;; 	(srv (args:get-arg "-server")))
;;     (and ohh srv)))
    ;; (debug:print-info 0 *default-log-port* "common:run-sync? ohh=" ohh ", srv=" srv)

(define *watchdog* (make-thread
		    (lambda ()
		      (handle-exceptions
			  exn
			  (begin
			    (print-call-chain)
			    (print " message: " ((condition-property-accessor 'exn 'message) exn)))
			(common:watchdog)))
		    "Watchdog thread"))

;; currently the primary job of the watchdog is to run the sync back to megatest.db from the db in /tmp
;; if we are on the homehost and we are a server (by definition we are on the homehost if we are a server)
;;
(define (common:readonly-watchdog dbstruct)
  (thread-sleep! 0.05) ;; delay for startup
  (debug:print-info 13 *default-log-port* "common:readonly-watchdog entered.")
  ;; sync megatest.db to /tmp/.../megatst.db
  (let* ((sync-cool-off-duration   3)
        (golden-mtdb     (dbr:dbstruct-mtdb dbstruct))
        (golden-mtpath   (db:dbdat-get-path golden-mtdb))
        (tmp-mtdb        (dbr:dbstruct-tmpdb dbstruct))
        (tmp-mtpath      (db:dbdat-get-path tmp-mtdb)))
    (debug:print-info 0 *default-log-port* "Read-only periodic sync thread started.")
    (let loop ((last-sync-time 0))
      (debug:print-info 13 *default-log-port* "loop top tmp-mtpath="tmp-mtpath" golden-mtpath="golden-mtpath)
      (let* ((duration-since-last-sync (- (current-seconds) last-sync-time)))
        (debug:print-info 13 *default-log-port* "duration-since-last-sync="duration-since-last-sync)
        (if (and (not *time-to-exit*)
                 (< duration-since-last-sync sync-cool-off-duration))
            (thread-sleep! (- sync-cool-off-duration duration-since-last-sync)))
        (if (not *time-to-exit*)
            (let ((golden-mtdb-mtime (file-modification-time golden-mtpath))
                  (tmp-mtdb-mtime    (file-modification-time tmp-mtpath)))
	      (if (> golden-mtdb-mtime tmp-mtdb-mtime)
		  (if (< golden-mtdb-mtime (- (current-seconds) 3)) ;; file has NOT been touched in past three seconds, this way multiple servers won't fight to sync back
		      (let ((res (db:multi-db-sync dbstruct 'old2new)))
			(debug:print-info 13 *default-log-port* "rosync called, " res " records transferred."))))
              (loop (current-seconds)))
            #t)))
    (debug:print-info 0 *default-log-port* "Exiting readonly-watchdog timer, *time-to-exit* = " *time-to-exit*" pid="(current-process-id)" mtpath="golden-mtpath)))


;; TODO: for multiple areas, we will have multiple watchdogs; and multiple threads to manage
(define (common:watchdog)
  (debug:print-info 13 *default-log-port* "common:watchdog entered.")
  (if (launch:setup)
      (if (common:on-homehost?)
	  (let ((dbstruct (db:setup #t)))
	    (debug:print-info 13 *default-log-port* "after db:setup with dbstruct=" dbstruct)
	    (cond
	     ((dbr:dbstruct-read-only dbstruct)
	      (debug:print-info 13 *default-log-port* "loading read-only watchdog")
	      (common:readonly-watchdog dbstruct))
	     (else
	      (debug:print-info 13 *default-log-port* "loading writable-watchdog.")
              (let* ((syncer (or (configf:lookup *configdat* "server" "sync-method") "brute-force-sync")))
                (cond
                 ((equal? syncer "brute-force-sync")
                  (server:writable-watchdog-bruteforce dbstruct))
                 ((equal? syncer "delta-sync")
                  (server:writable-watchdog-deltasync dbstruct))
                 (else
                  (debug:print-error 0 *default-log-port* "Unknown server/sync-method specified ("syncer") - valid values are brute-force-sync and delta-sync.")
                  (exit 1)))
                ;;(debug:print 1 *default-log-port* "INFO: ["(common:human-time)"] Syncer started (method="syncer")")
                )))
	    (debug:print-info 13 *default-log-port* "watchdog done."))
	  (debug:print-info 13 *default-log-port* "no need for watchdog on non-homehost"))))


(define (std-exit-procedure)
  ;;(common:telemetry-log-close)
  (on-exit (lambda () 0))
  ;;(debug:print-info 13 *default-log-port* "std-exit-procedure called; *time-to-exit*="*time-to-exit*)
  (let ((no-hurry  (if *time-to-exit* ;; hurry up
		       #f
		       (begin
			 (set! *time-to-exit* #t)
			 #t))))
    (debug:print-info 4 *default-log-port* "starting exit process, finalizing databases.")
    (if (and no-hurry (debug:debug-mode 18))
	(rmt:print-db-stats))
    (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)
					  (sqlite3:finalize! db #t)
					  ;; (vector-set! *task-db* 0 #f)
					  (set! *task-db* #f)))))
                              #;(http-client#close-all-connections!)
                              ;; (if (and *runremote*
                              ;;          (remote-conndat *runremote*))
                              ;;     (begin
                              ;;       (http-client#close-all-connections!))) ;; for http-client
                              (if (not (eq? *default-log-port* (current-error-port)))
                                  (close-output-port *default-log-port*))
			      (set! *default-log-port* (current-error-port))) "Cleanup db exit thread"))
	  (th2 (make-thread (lambda ()
			      (debug:print 4 *default-log-port* "Attempting clean exit. Please be patient and wait a few seconds...")
			      (if no-hurry
                                  (begin
                                    (thread-sleep! 5)) ;; give the clean up few seconds to do it's stuff
                                  (begin
      				  (thread-sleep! 2)))
      			      (debug:print 4 *default-log-port* " ... done")
      			      )
			    "clean exit")))
      (thread-start! th1)
      (thread-start! th2)
      (thread-join! th1)
      )
    )

  0)

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

(define (special-signal-handler signum)
  ;; (signal-mask! signum)
  (set! *time-to-exit* #t)
  ;;(debug:print-info 13 *default-log-port* "got signal "signum)
  (debug:print-error 0 *default-log-port* "Received signal " signum " sending email befor exiting!!")
  ;;TODO send email to notify admin contact listed in the config that the lisner got killed
  ;; (std-exit-procedure) ;; shouldn't need this since we are exiting and it will be called anyway
  (exit))


(set-signal-handler! signal/int  std-signal-handler)  ;; ^C
(set-signal-handler! signal/term std-signal-handler)

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

;; Force a megatest cleanup-db if version is changed and skip-version-check not specified
;; Do NOT check if not on homehost!
;;
(define (common:exit-on-version-changed)
  (if (common:on-homehost?)
      (if (common:api-changed?)
	  (let* ((mtconf (conc (get-environment-variable "MT_RUN_AREA_HOME") "/megatest.config"))
                (dbfile  (conc (get-environment-variable "MT_RUN_AREA_HOME") "/megatest.db"))
                (read-only (not (file-write-access? dbfile)))
                (dbstruct (db:setup #t)))
	    (debug:print 0 *default-log-port*
			 "WARNING: Version mismatch!\n"
			 "   expected: " (common:version-signature) "\n"
			 "   got:      " (common:get-last-run-version))
            (cond
             ((get-environment-variable "MT_SKIP_DB_MIGRATE") #t)
             ((and (common:file-exists? mtconf) (common:file-exists? dbfile) (not read-only)
                   (eq? (current-user-id)(file-owner mtconf))) ;; safe to run -cleanup-db
              (debug:print 0 *default-log-port* "   I see you are the owner of megatest.config, attempting to cleanup and reset to new version")
              (handle-exceptions
               exn
               (begin
                 (debug:print 0 *default-log-port* "Failed to switch versions.")
                 (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
                 (print-call-chain (current-error-port))
                 (exit 1))
               (common:cleanup-db dbstruct)))
             ((not (common:file-exists? mtconf))
              (debug:print 0 *default-log-port* "   megatest.config does not exist in this area.  Cannot proceed with megatest version migration.")
              (exit 1))
             ((not (common:file-exists? dbfile))
              (debug:print 0 *default-log-port* "   megatest.db does not exist in this area.  Cannot proceed with megatest version migration.")
              (exit 1))
             ((not (eq? (current-user-id)(file-owner mtconf)))
              (debug:print 0 *default-log-port* "   You do not own megatest.db in this area.  Cannot proceed with megatest version migration.")
              (exit 1))
             (read-only
              (debug:print 0 *default-log-port* "   You have read-only access to this area.  Cannot proceed with megatest version migration.")
              (exit 1))
             (else
              (debug:print 0 *default-log-port* " to switch versions you can run: \"megatest -cleanup-db\"")
              (exit 1)))))))
;;      (begin
;;	(debug:print 0 *default-log-port* "ERROR: cannot migrate version unless on homehost. Exiting.")
;;	(exit 1))))

;; Move me elsewhere ...
;; RADT => Why do we meed the version check here, this is called only if version misma
;;
(define (common:cleanup-db dbstruct #!key (full #f))
  (apply db:multi-db-sync 
   dbstruct
   'schema
   ;; 'new2old
   'killservers
   'adj-target
   ;; 'old2new
   'new2old
   ;; (if full
       '(dejunk)
       ;; '())
       )
  (if (common:api-changed?)
      (common:set-last-run-version)))

;; 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 host-port-in #!key (do-exit #f))
  (let ((host:port (if (not host-port-in) ;; use read-dotserver to find
		       #f ;; (server:check-if-running *toppath*)
		;; (if (number? host-port-in) ;; we were handed a server-id
		;; 	   (let ((srec (tasks:get-server-by-id (db:delay-if-busy (tasks:open-db)) host-port-in)))
		;; 	     ;; (print "srec: " srec " host-port-in: " host-port-in)
		;; 	     (if srec
		;; 		 (conc (vector-ref srec 3) ":" (vector-ref srec 4))
		;; 		 (conc "no such server-id " host-port-in)))
		       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)))
;;	   (toppath       (launch:setup)))
      ;; (print "host-port=" host-port)
      (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 server-dat)))
	    (if (and (list? login-res)
		     (car login-res))
		(begin
		  ;; (print "LOGIN_OK")
		  (if do-exit (exit 0))
		  #t)
		(begin
		  ;; (print "LOGIN_FAILED")
		  (if do-exit (exit 1))
		  #f)))))))

;; run ping in separate process, safest way in some cases
;;
(define (server:ping-server ifaceport)
  (with-input-from-pipe 
   (conc (common:get-megatest-exe) " -ping " ifaceport)
   (lambda ()
     (let loop ((inl (read-line))
		(res "NOREPLY"))
       (if (eof-object? inl)
	   (case (string->symbol res)
	     ((NOREPLY)  #f)
	     ((LOGIN_OK) #t)
	     (else       #f))
	   (loop (read-line) inl))))))

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

;; no longer care if multiple servers are started by accident. older servers will drop off in time.
;;
#;(define (server:check-if-running areapath) ;;  #!key (numservers "2"))
  (let* ((ns            (server:get-num-servers))
	 (servers       (server:get-best (server:get-list areapath))))
    ;; (print "servers: " servers " ns: " ns)
    (if (or (and servers
		 (null? servers))
	    (not servers)
	    (and (list? servers)
		 (< (length servers) (random ns)))) ;; somewhere between 0 and numservers
        #f
        (let loop ((hed (car servers))
                   (tal (cdr servers)))
          (let ((res (server:check-server hed)))
            (if res
                res
                (if (null? tal)
                    #f
                    (loop (car tal)(cdr tal)))))))))


;; NOT USED (well, ok, reference in rpc-transport but otherwise not used).
;;
(define (server:login toppath)
  (lambda (toppath)
    (set! *db-last-access* (current-seconds)) ;; might not be needed.
    (if (equal? *toppath* toppath)
	#t
	#f)))
;; Given a run id start a server process    ### NOTE ### > file 2>&1 
;; if the run-id is zero and the target-host is set 
;; try running on that host
;;   incidental: rotate logs in logs/ dir.
;;
#;(define  (server:run areapath) ;; areapath is *toppath* for a given testsuite area
  (let* ((curr-host   (get-host-name))
         ;; (attempt-in-progress (server:start-attempted? areapath))
         ;; (dot-server-url (server:check-if-running areapath))
	 (curr-ip     (server:get-best-guess-address curr-host))
	 (curr-pid    (current-process-id))
	 (homehost    (common:get-homehost)) ;; configf:lookup *configdat* "server" "homehost" ))
	 (target-host (car homehost))
	 (testsuite   (common:get-testsuite-name))
	 (logfile     (conc areapath "/logs/server.log")) ;; -" curr-pid "-" target-host ".log"))
	 (cmdln (conc (common:get-megatest-exe)
		      " -server " (or target-host "-") (if (equal? (configf:lookup *configdat* "server" "daemonize") "yes")
							   " -daemonize "
							   "")
		      ;; " -log " logfile
		      " -m testsuite:" testsuite)) ;; (conc " >> " logfile " 2>&1 &")))))
	 (log-rotate  (make-thread common:rotate-logs  "server run, rotate logs thread"))
         (load-limit  (configf:lookup-number *configdat* "jobtools" "max-server-start-load" default: 3.0)))
    ;; we want the remote server to start in *toppath* so push there
    (push-directory areapath)
    (debug:print 0 *default-log-port* "INFO: Trying to start server (" cmdln ") ...")
    (thread-start! log-rotate)
    
    ;; host.domain.tld match host?
    (if (and target-host 
	     ;; look at target host, is it host.domain.tld or ip address and does it 
	     ;; match current ip or hostname
	     (not (string-match (conc "("curr-host "|" curr-host"\\..*)") target-host))
	     (not (equal? curr-ip target-host)))
	(begin
	  (debug:print-info 0 *default-log-port* "Starting server on " target-host ", logfile is " logfile)
	  (setenv "TARGETHOST" target-host)))
      
    (setenv "TARGETHOST_LOGF" logfile)
    (thread-sleep! (/ (random 5000) 1000)) ;; add about a random (up to 5 seconds) initial delay. It seems pretty common that many running tests request a server at the same time
    (common:wait-for-normalized-load load-limit " delaying server start due to load" target-host) ;; do not try starting servers on an already overloaded machine, just wait forever
    (system (conc "nbfake " cmdln))
    (unsetenv "TARGETHOST_LOGF")
    (if (get-environment-variable "TARGETHOST")(unsetenv "TARGETHOST"))
    (thread-join! log-rotate)
    (pop-directory)))

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

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

#;(define (server:start-and-wait areapath #!key (timeout 60))
  (let ((give-up-time (+ (current-seconds) timeout)))
    (let loop ((server-url (server:check-if-running areapath))
	       (try-num    0))
      (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 (and (> try-num 0)  ;; first time through simply wait a little while then try again
		     (< 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 areapath)
		  (+ try-num 1)))))))

;;======================================================================
;; make html output
;;======================================================================

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

(define (tests:summarize-items 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 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 run-id test-name "" #f #f #f)
		  (if script
		      (system (conc script " > " outputfilename " & "))
		      (tests:generate-html-summary-for-iterated-test 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! 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 (handle-exceptions
					 exn
					 0
				       (file-modification-time lockf)))
		    ;; we started since current re-gen in flight, delay a little and try again
		    (begin
		      (debug:print-info 1 *default-log-port* "Waiting to update " outputfilename ", another test currently updating it")
		      (thread-sleep! (+ 5 (random 5))) ;; delay between 5 and 10 seconds
		      (loop (common:simple-file-lock lockf))))))))))

(define (tests:generate-html-summary-for-iterated-test run-id test-id test-name outputfilename)
  (let ((counts              (make-hash-table))
	(statecounts         (make-hash-table))
	(outtxt              "")
	(tot                 0)
	(testdat             (rmt:test-get-records-for-index-file 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))
		 (itempath       (vector-ref testrecord 1))
		 (state          (vector-ref testrecord 2))
		 (status         (vector-ref testrecord 3))
		 (run_duration   (vector-ref testrecord 4))
		 (logf           (vector-ref testrecord 5))
		 (comment        (vector-ref testrecord 6)))
	     (hash-table-set! counts status (+ 1 (hash-table-ref/default counts status 0)))
	     (hash-table-set! statecounts state (+ 1 (hash-table-ref/default statecounts state 0)))
	     (set! outtxt (conc outtxt "<tr>"
				;; "<td><a href=\"" itempath "/" logf "\"> " itempath "</a></td>" 
				"<td><a href=\"" itempath "/test-summary.html\"> " itempath "</a></td>" 
				"<td>" state    "</td>" 
				"<td><font color=" (common:get-color-from-status status)
				">"   status   "</font></td>"
				"<td>" (if (equal? comment "")
					   "&nbsp;"
					   comment) "</td>"
					   "</tr>"))))
	 (if (list? testdat)
	     testdat
	     (begin
	       (print "ERROR: failed to get records with rmt:test-get-records-for-index-file run-id=" run-id "test-name=" test-name)
	       '())))
	
	(print "<table><tr><td valign=\"top\">")
	;; Print out stats for status
	(set! tot 0)
	(print "<table cellspacing=\"0\" border=\"1\"><tr><td colspan=\"2\"><h2>State stats</h2></td></tr>")
	(for-each (lambda (state)
		    (set! tot (+ tot (hash-table-ref statecounts state)))
		    (print "<tr><td>" state "</td><td>" (hash-table-ref statecounts state) "</td></tr>"))
		  (hash-table-keys statecounts))
	(print "<tr><td>Total</td><td>" tot "</td></tr></table>")
	(print "</td><td valign=\"top\">")
	;; Print out stats for state
	(set! tot 0)
	(print "<table cellspacing=\"0\" border=\"1\"><tr><td colspan=\"2\"><h2>Status stats</h2></td></tr>")
	(for-each (lambda (status)
		    (set! tot (+ tot (hash-table-ref counts status)))
		    (print "<tr><td><font color=\"" (common:get-color-from-status status) "\">" status
			   "</font></td><td>" (hash-table-ref counts status) "</td></tr>"))
		  (hash-table-keys counts))
	(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 
	;; run-id
	;; (hash-table-map
	;;  state-status-counts
	;;  (lambda (key val)
	;;	(append key (list val)))))
	))))

(define tests:css-jscript-block
#<<EOF
<style type="text/css">
ul.LinkedList { display: block; }
/* ul.LinkedList ul { display: none; } */
.HandCursorStyle { cursor: pointer; cursor: hand; }  /* For IE */
th {background-color: #8c8c8c;}
td.test {background-color: #d9dbdd;}
td.PASS {background-color: #347533;}
td.FAIL {background-color: #cc2812;}
td.SKIP{background-color: #FFD733;}
td.WARN {background-color: #EA8724;}
td.WAIVED {background-color: #838A12;}
td.ABORT{background-color: #EA24B7;}
.PASS .link, .SKIP .link, .WARN .link,.WAIVED .link,.ABORT .link, .FAIL .link{color: #FFFFFF;}


</style>


  <script type="text/JavaScript">

    function filtersome() {
  $("tr").show();
  $(".test").filter(
    function() {
      var names = $('#testname').val().split(',');
      var good=1;
      for (var i=0, len=names.length; i<len; i++) {
        var uname=names[i];
        console.log("Trying to check for " + uname); 
        if($(this).text().indexOf(uname) != -1) {
          good= 0;
          console.log("Found "+uname);
        }
      }
      return good; 
    }
  ).parent().hide();
//  $(".sum").show();
}
  
    // Add this to the onload event of the BODY element
    function addEvents() {
      activateTree(document.getElementById("LinkedList1"));
    }

    // This function traverses the list and add links 
    // to nested list items
    function activateTree(oList) {
      // Collapse the tree
      for (var i=0; i < oList.getElementsByTagName("ul").length; i++) {
        oList.getElementsByTagName("ul")[i].style.display="none";            
      }                                                                  
      // Add the click-event handler to the list items
      if (oList.addEventListener) {
        oList.addEventListener("click", toggleBranch, false);
      } else if (oList.attachEvent) { // For IE
        oList.attachEvent("onclick", toggleBranch);
      }
      // Make the nested items look like links
      addLinksToBranches(oList);
    }

    // This is the click-event handler
    function toggleBranch(event) {
      var oBranch, cSubBranches;
      if (event.target) {
        oBranch = event.target;
      } else if (event.srcElement) { // For IE
        oBranch = event.srcElement;
      }
      cSubBranches = oBranch.getElementsByTagName("ul");
      if (cSubBranches.length > 0) {
        if (cSubBranches[0].style.display == "block") {
          cSubBranches[0].style.display = "none";
        } else {
          cSubBranches[0].style.display = "block";
        }
      }
    }

    // This function makes nested list items look like links
    function addLinksToBranches(oList) {
      var cBranches = oList.getElementsByTagName("li");
      var i, n, cSubBranches;
      if (cBranches.length > 0) {
        for (i=0, n = cBranches.length; i < n; i++) {
          cSubBranches = cBranches[i].getElementsByTagName("ul");
          if (cSubBranches.length > 0) {
            addLinksToBranches(cSubBranches[0]);
            cBranches[i].className = "HandCursorStyle";
            cBranches[i].style.color = "blue";
            cSubBranches[0].style.color = "black";
            cSubBranches[0].style.cursor = "auto";
          }
        }
      }
    }
  </script>
EOF
)

(define tests:css-jscript-block-dynamic 
#<<EOF
           <script src= ./jquery3.1.0.js></script> 
EOF
)

(define  (test:js-block javascript-lib)
   (conc  "<script src=" javascript-lib "></script>" ))


(define tests:css-jscript-block-static (test:js-block *java-script-lib*))

(define (tests:css-jscript-block-cond dynamic) 
      (if (equal? dynamic  #t)
       tests:css-jscript-block-dynamic
       tests:css-jscript-block-static))

       
(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 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
				   run-id
                                   "%"       ;; testnamepatt
				   '()        ;; states
				   '()        ;; statuses
				   #f         ;; offset
				   #f         ;; num-to-get
				   #f         ;; hide/not-hide
				   #f         ;; sort-by
				   #f         ;; sort-order
				   #f         ;; 'shortlist                           ;; qrytype
                                   0         ;; last update
				   #f)))
            
            (map (lambda (test)
                 (let* ((test-name (vector-ref test 2))
                        (test-html-path (conc (vector-ref test 10) "/" (vector-ref test 13)))
                        (test-item (conc test-name ":" (vector-ref test 11)))
                        (test-status (vector-ref test 4)))
                         
                (if (not (hash-table-ref/default resh test-name  #f))
                      (hash-table-set! resh test-name   (make-hash-table)))
                (if (not (hash-table-ref/default (hash-table-ref/default resh test-name  #f)  test-item  #f))
                       (hash-table-set! (hash-table-ref/default resh test-name  #f) test-item   (make-hash-table))) 
               (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))


;; hash-table tree to html list tree
;;
;;   tipfunc takes two parameters: y the tip value and path the path to that point
;;
(define (common:htree->html ht path tipfunc)
  (let ((datlist 	(sort (hash-table->alist ht)
                              (lambda (a b)
                                (string< (car a)(car b))))))
    (if (null? datlist)
    	(tipfunc #f path) ;; really shouldn't get here
	(s:ul
	 (map (lambda (x)
		(let* ((levelname (car x))
		       (y         (cdr x))
		       (newpath   (append path (list levelname)))
		       (leaf      (or (not (hash-table? y))
				      (null? (hash-table-keys y)))))
		  (if leaf
		      (s:li (tipfunc y newpath))
		      (s:li
		       (list 
			levelname
			(common:htree->html y newpath tipfunc))))))
	      datlist)))))


;; tests:genrate dashboard body 
;;

(define (tests:dashboard-body page pg-size keys numkeys  total-runs linktree area-name get-prev-links get-next-links flag run-patt target-patt)
  (let* ((start (* page pg-size)) 
	       ;(runsdat   (rmt:get-runs "%" pg-size start (map (lambda (x)(list x "%")) keys)))
         (runsdat   (rmt:get-runs-by-patt  keys run-patt target-patt start pg-size #f 0 sort-order: "desc"))
                    ; db:get-runs-by-patt   keys runnamepatt targpatt offset limit fields last-update   
	       (header    (vector-ref runsdat 0))
	       (runs      (vector-ref runsdat 1))
         (ctr 0)
         (test-runs-hash (tests:get-rest-data runs header numkeys))
         (test-list (hash-table-keys test-runs-hash))) 
  
  (s:html tests:css-jscript-block (tests:css-jscript-block-cond flag)
		   (s:title "Summary for " area-name)
		   (s:body 'onload "addEvents();"
                          (get-prev-links page linktree)
                          (get-next-links page linktree total-runs)
                           
			   (s:h1 "Summary for " area-name)
                           (s:h3 "Filter" )
                           (s:input 'type "text"  'name "testname" 'id "testname" 'length "30" 'onkeyup "filtersome()")
			   ;; top list
         
			   (s:table 'id "LinkedList1" 'border "1" 'cellspacing 0
                            (map (lambda (key)
				 (let* ((res (s:tr 'class "something" 
				  (s:th key )
                                   (map (lambda (run)
                                   (s:th  (vector-ref run ctr)))
                                  runs))))
                             (set! ctr (+ ctr 1))
                               res))
                               keys)
                               (s:tr
				 (s:th "Run Name")
                                  (map (lambda (run)
                                   (s:th (db:get-value-by-header run header "runname")))
                                  runs))
                              
                               (map (lambda (test-name)
                                 (let* ((item-hash (hash-table-ref/default test-runs-hash test-name  #f))
                                         (item-keys (sort (hash-table-keys item-hash) string<=?))) 
                                          (map (lambda (item-name)  
  		                             (let* ((res (s:tr  'class item-name
				                         (s:td  item-name 'class "test" )
                                                           (map (lambda (run)
                                                               (let* ((run-test (hash-table-ref/default item-hash item-name  #f))
                                                                      (run-id (db:get-value-by-header run header "id"))
                                                                      (result (hash-table-ref/default run-test run-id "n/a"))
                                                                      ;(relative-path (get-relative-path)) 
                                                                      (status (if (string? result)
									                                                            	result
										                                                            (car result)))
                                                                        (link (if (string? result)
										                                                            result
                                                                                (if (equal? flag #t) 
                                                                                (s:a (car result) 'href (conc "./test_log?runid=" run-id "&testname="  item-name ))
  																																						  (s:a (car result) 'href (string-substitute  (conc linktree "/")  "" (cadr result)  "-"))))))
                                                                       (s:td  link 'class status)))
                                                                runs))))
                                                        res))
                                                   item-keys)))
                               test-list)))))) 

;; (tests:create-html-tree "test-index.html")
;;
(define (tests:create-html-tree outf)
   (let* ((lockfile  (conc outf ".lock"))
	 			 (runs-to-process '())
         (linktree  (common:get-linktree))
         (area-name (common:get-testsuite-name))
	  		 (keys      (rmt:get-keys))
	  		 (numkeys   (length keys))
         (run-patt (or (args:get-arg "-run-patt")
                        (args:get-arg "-runname")
                        "%"))
         (target (or  (args:get-arg "-target-patt") 
											(args:get-arg "-target")
                      "%"))
         (targlist (string-split target "/"))
         (numtarg  (length targlist))  
         (targtweaked (if (> numkeys numtarg)
			   								(append targlist (make-list (- numkeys numtarg) "%"))
			  								targlist))
         (target-patt (string-join targtweaked "/"))
         ;(total-runs  (rmt:get-num-runs "%")) ;;this needs to be changed to filter by target
          (total-runs (rmt:get-runs-cnt-by-patt run-patt target-patt keys )) 
         (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"))))
               (get-prev-links (lambda (page linktree )   
                            (let* ((link  (if (not (eq? page 0))
                                   (s:a "&lt;&lt;prev" 'href (conc  "page" (- page 1) ".html"))
                                   (s:a "" 'href (conc   "page"  page ".html")))))
                               link)))
               (get-next-links (lambda (page linktree total-runs)   
                            (let* ((link  (if (> total-runs (+ 10 (* page pg-size)))
                                   (s:a "next&gt;&gt;" 'href (conc  "page"  (+ page 1) ".html"))
                                   (s:a "" 'href (conc   "page" page  ".html")))))
                               link))) )
          (print "total runs: " total-runs) 
          (s:output-new
	   			 oup
	   					(tests:dashboard-body page pg-size keys numkeys total-runs linktree area-name get-prev-links get-next-links #f run-patt target-patt)) ;; update this function
          (close-output-port oup)
         ; (set! page (+ 1 page))
          (if (> total-runs (* (+ 1 page) pg-size))
           (loop (+ 1  page)))))
	  (common:simple-file-release-lock lockfile))
	            
	#f)))


(define (tests:readlines filename)
  (call-with-input-file filename
    (lambda (p)
      (let loop ((line (read-line p))
                 (result '()))
        (if (eof-object? line)
            (reverse result)
            (loop (read-line p) (cons line result)))))))

(define (tests:get-test-log run-id test-name item-name)
  (let* ((test-data    (rmt:get-tests-for-run
				   (string->number run-id)
                                    test-name      ;; testnamepatt
				   '()        ;; states
				   '()        ;; statuses
				   #f         ;; offset
				   #f         ;; num-to-get
				   #f         ;; hide/not-hide
				   #f         ;; sort-by
				   #f         ;; sort-order
				   #f         ;; 'shortlist                           ;; qrytype
                                   0         ;; last update
				   #f))
         (path "")
         (found 0))
    (debug:print-info 0 *default-log-port* "found: " found )

   (let loop ((hed (car test-data))
		 (tal (cdr test-data)))
          (debug:print-info 0 *default-log-port* "item: " (vector-ref hed 11) (vector-ref hed 10) "/" (vector-ref hed 13))

	(if (equal? (vector-ref hed 11) item-name)
            (begin
              (set! found 1) 
	      (set! path (conc (vector-ref hed 10) "/" (vector-ref hed 13)))))
	    (if (and (not (null? tal)) (equal? found 0))
		(loop (car tal)(cdr tal))))
   (if (equal? path "")
     "<H2>Data not found</H2>"
     (string-join (tests:readlines path) "\n"))))


(define (tests:dynamic-dboard page)
;(define (tests:create-html-tree o)
 (let* (
;(page "1")
          (linktree  (common:get-linktree))
         (area-name (common:get-testsuite-name))
	       (keys      (rmt:get-keys))
	       (numkeys   (length keys))
         (targtweaked (make-list numkeys "%"))
         (target-patt (string-join targtweaked "/"))
         (total-runs  (rmt:get-num-runs "%"))
         (pg-size 10)
         (pg (if (equal? page #f)
                 0
                 (- (string->number page) 1)))
          (get-prev-links  (lambda (pg linktree)
                           (debug:print-info 0 *default-log-port* "val: " (- 1 pg))
                          (let* ((link  (if (not (eq? pg 0))
                               (s:a  "&lt;&lt;prev " 'href (conc  "dashboard?page="  pg  ))
                               (s:a "" 'href (conc  "dashboard?page=" pg)))))
                               link)))
          (get-next-links   (lambda (pg linktree total-runs)  
                            (debug:print-info 0 *default-log-port* "val: " pg)
                             (debug:print-info 0 *default-log-port* "val: " total-runs " size" pg-size)
 
                            (let* ((link  (if (> total-runs (+ 10 (* pg pg-size)))
                              (s:a  "next&gt;&gt; "  'href (conc  "dashboard?page="  (+ pg 2)  ))
                             (s:a "" 'href (conc  "dashboard?page=" pg  )))))
                             link)))
         (html-body (tests:dashboard-body pg pg-size keys numkeys total-runs linktree area-name get-prev-links get-next-links #t "%" target-patt))) ;; update tis function
        html-body))

(define (tests:create-html-summary outf)
 (let* ((lockfile  (conc outf ".lock"))
        (linktree  (common:get-linktree))
				(keys      (rmt:get-keys))
        (area-name (common:get-testsuite-name))
        (run-patt (or (args:get-arg "-run-patt")
                        (args:get-arg "-runname")
                        "%"))
        (target (or (args:get-arg "-target-patt")
                        (args:get-arg "-target")
                        "%"))
         (targlist (string-split target "/"))
         (numkeys  (length keys))
	       (numtarg  (length targlist))  
         (targtweaked (if (> numkeys numtarg)
			   								(append targlist (make-list (- numkeys numtarg) "%"))
			  								targlist))
        (target-patt (string-join targtweaked "/")))
    (if (common:simple-file-lock lockfile)
        (begin
          (let* (;(runsdat1   (rmt:get-runs run-patt #f #f (map (lambda (x)(list x "%")) keys)))
                 (runsdat   (rmt:get-runs-by-patt  keys run-patt target-patt #f #f #f 0))
					       (runs      (vector-ref runsdat 1))
                 (header      (vector-ref runsdat 0))
        	       (oup       (open-output-file (or outf (conc linktree "/targets.html"))))
                 (target-hash (test:create-target-hash runs header (length keys))))
           (test:create-target-html target-hash oup area-name linktree)
          (test:create-run-html  runs area-name linktree (length keys) header))
	  (common:simple-file-release-lock lockfile))
	#f)))

(define (test:get-test-hash test-data)
	(let ((resh (make-hash-table)))
    	(map (lambda (test)
        (let* ((test-name (vector-ref test 2))
               (test-html-path (if (file-exists? (conc (vector-ref test 10) "/test-summary.html"))
																 (conc (vector-ref test 10) "/test-summary.html" )
							 									 (conc (vector-ref test 10) "/" (vector-ref test 13))))
               (test-item  (vector-ref test 11))
               (test-status (vector-ref test 4)))
               (if (not (hash-table-ref/default resh test-item  #f))
                   (hash-table-set! resh test-item   (make-hash-table)))
               (hash-table-set! (hash-table-ref/default resh test-item  #f) test-name (list test-status test-html-path)))) 
        test-data)
resh))

(define (test:get-data->b-keys ordered-data a-keys)
  (delete-duplicates
   (sort (apply
	  append
	  (map (lambda (sub-key)
		 (let ((subdat (hash-table-ref ordered-data sub-key)))
		   (hash-table-keys subdat)))
	       a-keys))
	 string>=?)))


(define (test:create-run-html runs area-name linktree numkeys header)
  (map (lambda (run)
		 (let* ((target (string-join (take (vector->list run) numkeys) "/"))
						(run-name (db:get-value-by-header run header "runname"))
            (run-time (seconds->work-week/day-time (db:get-value-by-header run header "event_time")))
						(oup (if (file-exists? (conc linktree "/" target "/" run-name))
                        (open-output-file (conc linktree "/" target "/" run-name "/run.html"))
                         #f))
            (run-id (db:get-value-by-header run header "id"))
            (test-data    (rmt:get-tests-for-run
				  								 run-id
                           "%"       ;; testnamepatt
				  								 '()        ;; states
				   								 '()        ;; statuses
				  								 	#f         ;; offset
				  						 			#f         ;; num-to-get
				   									#f         ;; hide/not-hide
				  								  #f         ;; sort-by
				   									#f         ;; sort-order
				   									#f         ;; 'shortlist                           ;; qrytype
                            0         ;; last update
				  									#f))
            (item-test-hash (test:get-test-hash test-data))
            (items  (hash-table-keys item-test-hash))
 						(test-names (test:get-data->b-keys item-test-hash items)))
    (if oup
      (begin 
     (s:output-new
	   oup
	   (s:html tests:css-jscript-block (tests:css-jscript-block-cond #f)
		   (s:title "Runs View " run-name)
		   (s:body
		     (s:h1 "Runs View " )
         (s:h3 "Target" target)
				 (s:p 
					(s:b "Run name" ) run-name)
         (s:p 
					(s:b "Run Date" ) run-time)
         (s:table 'border 1 'cellspacing 0
           (s:tr
           (s:th "Items")
           (map (lambda (test)
            (s:th test))
           test-names))  
           (map (lambda (item) 
					  (let* ((test-hash (hash-table-ref/default item-test-hash item  #f)))
								 (if test-hash
                  (begin
									(s:tr
					  			(s:td 'class "test" item)
            			(map (lambda (test)
						  		(let* ((test-details (hash-table-ref/default test-hash test  #f))
												(status (if test-details
																(car test-details)))
                        (link (if test-details 
														(string-substitute  (conc linktree "/" target "/" run-name "/")  "" (cadr test-details) "-"))))
                   (if test-details
											(s:td 'class status
												(s:a 'class "link" 'href link status ))
                      (s:td "")))) 			
									test-names))))))
				  (sort items string<=?))))))
		(close-output-port oup))
    (debug:print-info 0 "Skip: Dirctory structure " linktree "/" target "/" run-name " does not exist. Megatest will not create run.html"))))
runs))

(define (test:create-target-hash runs header numkeys)
  (let ((resh (make-hash-table)))
   (for-each
     (lambda (run)
        (let* ((run-name (db:get-value-by-header run header "runname"))
               (target   (string-join (take (vector->list run) numkeys) "/"))
               (run-list (hash-table-ref/default resh target  #f)))
               
               (if (not run-list)
                   (hash-table-set! resh target   (list run-name))
                   (hash-table-set! resh target   (cons run-name run-list)))))
      runs)
   resh))

(define (test:get-max-run-cnt target-hash targets)
   (let* ((cnt 0 ))
   (map (lambda (target)
        (let* ((runs  (hash-table-ref/default target-hash target  #f))
               (run-length (if runs
																(length runs)
                                 0)))
  
              (if (< cnt run-length)
               (set! cnt  run-length)))) 
		targets) 
cnt))
 
(define (test:pad-runs target-hash targets max-row-length)
 (map (lambda (target)
        (let loop ((run-list  (hash-table-ref/default target-hash target  #f)))
               (if (< (length run-list) max-row-length)
                 (begin  
               		 (hash-table-set! target-hash target   (cons "" run-list))
               		 (loop (hash-table-ref/default target-hash target  #f) ))))) 
		targets)
   target-hash)

(define (test:create-target-html target-hash oup area-name linktree)
  (let* ((targets (hash-table-keys target-hash))
         (max-row-length (test:get-max-run-cnt target-hash targets))
         (pad-runs-hash (test:pad-runs target-hash targets max-row-length)))
   (s:output-new
	   oup
	   (s:html tests:css-jscript-block (tests:css-jscript-block-cond #f)

		   (s:title "Target View " area-name)
		   (s:body
		   (s:h1 "Target View " area-name)
					(s:table 'id "LinkedList1" 'border "1" 'cellspacing 0
             (s:tr 'class "something" 
               (s:th "Target")
								(s:th 'colspan max-row-length "Runs"))                                              
                (let* ((tbl (map (lambda (target)
                      (s:tr
                      (s:td 'class "test" target)
										  (let* ((runs  (hash-table-ref/default target-hash target  #f))
														 (rest-row (map (lambda (run)
																				(if (equal? run "")
																						(s:td run)
                                            (if (file-exists?(conc linktree "/" target "/" run ))
																						(begin 
																							(s:td 
																							(s:a 'href (conc  target "/" run "/run.html") run))))))
																				(reverse runs))))
                              rest-row)))
                                   targets)))
                           tbl)))))
          (close-output-port oup)))


(define (tests:create-html-tree-old 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))
	       (numkeys   (length keys))
	       (runsdat   (rmt:get-runs "%" #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)
	  (s:output-new
	   oup
	   (s:html tests:css-jscript-block
		   (s:title "Summary for " area-name)
		   (s:body 'onload "addEvents();"
			   (s:h1 "Summary for " area-name)
			   ;; top list
			   (s:ul 'id "LinkedList1" 'class "LinkedList"
				 (s:li
				  "Runs"
				  (common:htree->html runs-htree
						      '()
						      (lambda (x p)
							(let* ((targ-path (string-intersperse p "/"))
                                                               (full-path (conc linktree "/" targ-path))
                                                               (run-name  (car (reverse p))))
                                                          (if (and (common:file-exists? full-path)
                                                                   (directory?   full-path)
                                                                   (file-write-access? full-path))
                                                              (s:a run-name 'href (conc targ-path "/run-summary.html"))
                                                              (begin
                                                                (debug:print 0 *default-log-port* "INFO: Can't create " targ-path "/run-summary.html")
                                                                (conc run-name " (Not able to create summary at " targ-path ")")))))))))))
          (close-output-port oup)
	  (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
				   run-id
                                   "%/"       ;; testnamepatt
				   '()        ;; states
				   '()        ;; statuses
				   #f         ;; offset
				   #f         ;; num-to-get
				   #f         ;; hide/not-hide
				   #f         ;; sort-by
				   #f         ;; sort-order
				   #f         ;; 'shortlist                           ;; qrytype
                                   0         ;; last update
				   #f))
                    (tests-tree-dat (map (lambda (test-dat)
                                         ;; (tests:run-record->test-path x numkeys))
                                         (let* ((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))
                                                (path-parts (string-split full-name)))
                                           path-parts))
                                       test-dats))
                    (tests-htree (common:list->htree tests-tree-dat))
                    (html-dir    (conc linktree "/" (string-intersperse run-dir "/")))
                    (html-path   (conc html-dir "/run-summary.html"))
                    (oup         (if (and (common:file-exists? html-dir)
                                          (directory?   html-dir)
                                          (file-write-access? html-dir))
                                     (open-output-file  html-path)
                                     #f)))
               ;; (print "run-dir: " run-dir ", tests-tree-dat: " tests-tree-dat)
               (if oup
                   (begin
                     (s:output-new
                      oup
                      (s:html tests:css-jscript-block
                              (s:title "Summary for " area-name)
                              (s:body 'onload "addEvents();"
                                      (s:h1 "Summary for " (string-intersperse run-dir "/"))
                                      ;; top list
                                      (s:ul 'id "LinkedList1" 'class "LinkedList"
                                            (s:li
                                             "Tests"
                                             (common:htree->html tests-htree
                                                                 '()
                                                                 (lambda (x p)
                                                                   (let* ((targ-path (string-intersperse p "/"))
                                                                          (test-name (car p))
                                                                          (item-path ;; (if (> (length p) 2) ;; test-name + run-name
                                                                           (string-intersperse p "/"))
                                                                          (full-targ (conc html-dir "/" targ-path))
                                                                          (std-file  (conc full-targ "/test-summary.html"))
                                                                          (alt-file  (conc full-targ "/megatest-rollup-" test-name ".html"))
                                                                          (html-file (if (common:file-exists? alt-file)
                                                                                         alt-file
                                                                                         std-file))
                                                                          (run-name  (car (reverse p))))
                                                                     (if (and (not (common:file-exists? full-targ))
                                                                              (directory? full-targ)
                                                                              (file-write-access? full-targ))
                                                                         (tests:summarize-test 
                                                                          run-id 
                                                                          (rmt:get-test-id run-id test-name item-path)))
                                                                     (if (common:file-exists? full-targ)
                                                                         (s:a run-name 'href html-file)
                                                                         (begin
                                                                           (debug:print 0 *default-log-port* "ERROR: can't access " full-targ)
                                                                           (conc "No summary for " run-name)))))
                                                                 ))))))
                     (close-output-port oup)))))
           runs)
          #t)
	#f)))




(define (tests:update-central-meta-info run-id test-id cpuload diskfree minutes uname hostname)
  (rmt:general-call '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 'update-cpuload-diskfree run-id cpuload diskfree test-id))
  (if minutes 
      (rmt:general-call 'update-run-duration run-id minutes test-id))
  (if (and uname hostname)
      (rmt:general-call '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 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 run-id test-id cpuload diskfree minutes uname hostname)))
    
;; 
;;
(define (tests:get-compressed-steps run-id test-id)
  (let* ((steps-data  (rmt:get-steps-for-test run-id test-id)) ;;      0       1    2    3       4       5       6      7       
	 (comprsteps  (tests:process-steps-table steps-data))) ;; #<stepname start end status Duration Logfile Comment id>
    (map (lambda (x)
	   ;; take advantage of the \n on time->string
	   (vector    ;; we are constructing basically the original vector but collapsing start end records
	    (vector-ref x 0)                              ;; id        0
	    (let ((s (vector-ref x 1)))
	      (if (number? s)(seconds->time-string s) s)) ;; starttime 1
	    (let ((s (vector-ref x 2)))
	      (if (number? s)(seconds->time-string s) s)) ;; endtime   2
	    (vector-ref x 3)                              ;; status    3    
	    (vector-ref x 4)                              ;; duration  4
	    (vector-ref x 5)                              ;; logfile   5
	    (vector-ref x 6)                              ;; comment   6
	    (vector-ref x 7)))                            ;; id        7
	 (sort (hash-table-values comprsteps)
	       (lambda (a b)
		 (let ((time-a (vector-ref a 1))
		       (time-b (vector-ref b 1))
		       (id-a   (vector-ref a 7))
		       (id-b   (vector-ref b 7)))
		   (if (and (number? time-a)(number? time-b))
		       (if (< time-a time-b)
			   #t
			   (if (eq? time-a time-b)
			       (< id-a id-b)
			       ;; (string<? (conc (vector-ref a 2))
			       ;;	    (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 run-id test-id)
  (let* ((test-dat  (rmt:get-test-info-by-id run-id test-id))
	 (out-dir   (db:test-get-rundir test-dat))
	 (out-file  (conc out-dir "/test-summary.html")))
    ;; first verify we are able to write the output file
    (if (not (file-write-access? out-dir))
	(debug:print 0 *default-log-port* "ERROR: cannot write test-summary.html to " out-dir)
	(let* (;; (steps-dat (rmt:get-steps-for-test run-id test-id))
	       (test-name (db:test-get-testname test-dat))
	       (item-path (db:test-get-item-path test-dat))
	       (full-name (db:test-make-full-name test-name item-path))
	       (oup       (open-output-file out-file))
	       (status    (db:test-get-status   test-dat))
	       (color     (common:get-color-from-status status))
	       (logf      (db:test-get-final_logf test-dat))
	       (steps-dat (tests:get-compressed-steps 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)
	    (s:body 
	     (s:h2 "Summary for " full-name)
	     (s:table 'cellspacing "0" 'border "1"
		      (s:tr (s:td "run id")   (s:td (db:test-get-run_id   test-dat))
			    (s:td "test id")  (s:td (db:test-get-id       test-dat)))
		      (s:tr (s:td "testname") (s:td test-name)
			    (s:td "itempath") (s:td item-path))
		      (s:tr (s:td "state")    (s:td (db:test-get-state    test-dat))
			    (s:td "status")   (s:td (s:a 'href logf (s:font 'color color status))))
		      (s:tr (s:td "TestDate") (s:td (seconds->work-week/day-time 
						     (db:test-get-event_time test-dat)))
			    (s:td "Duration") (s:td (seconds->hr-min-sec (db:test-get-run_duration test-dat)))))
	     (s:h3 "Log files")
	     (s:table 
	      'cellspacing "0" 'border "1"
	      (s:tr (s:td "Final log")(s:td (s:a 'href logf logf))))
	     (s:table
	      'cellspacing "0" 'border "1"
	      (s:tr (s:td "Step Name")(s:td "Start")(s:td "End")(s:td "Status")(s:td "Duration")(s:td "Log File"))
	      (map (lambda (step-dat)
		     (s:tr (s:td (tdb:steps-table-get-stepname step-dat))
			   (s:td (tdb:steps-table-get-start    step-dat))
			   (s:td (tdb:steps-table-get-end      step-dat))
			   (s:td (tdb:steps-table-get-status   step-dat))
			   (s:td (tdb:steps-table-get-runtime  step-dat))
			   (s:td (let ((step-log (tdb:steps-table-get-log-file step-dat)))
				   (s:a 'href step-log step-log)))))
		   steps-dat))
	     )))
	  (close-output-port oup)))))
	  
	  
;; MUST BE CALLED local!
;;
(define (tests:test-get-paths-matching 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 keynames target res
					testpatt
					statepatt
					statuspatt
					runname)))
    (if fnamepatt
	(apply append 
	       (map (lambda (p)
		      (if (directory-exists? p)
			  (let ((glob-query (conc p "/" fnamepatt)))
			    (handle-exceptions
				exn
				(with-input-from-pipe
				    (conc "echo " glob-query)
				  read-lines)  ;; we aren't going to try too hard. If glob breaks it is likely because someone tried to do */*/*.log or similar
			      (glob glob-query)))
			  '()))
		    paths-from-db))
	paths-from-db)))

			      
;; for each test:
;;   
(define (tests:filter-non-runnable 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 run-id test-name item-path))
	      (tdat        (rmt:get-testinfo-state-status 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 run-id waiton ""))
				      (wtdat          (rmt:get-testinfo-state-status 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)
				 ;;        	 '("INCOMPETE")))
				     (set! keep-test #f)))) ;; no point in running this one again
			     waitons))))
	 (if keep-test (set! runnables (cons testkeyname runnables)))))
     testkeynames)
    runnables))

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

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

(define (test-get-kill-request run-id test-id) ;; run-id test-name itemdat)
  (let* ((testdat   (rmt:get-test-info-by-id 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: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 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))
	   (debug:print 5 *default-log-port* "exn=" (condition->list exn))
	   (debug:print 0 *default-log-port* " status:  " ((condition-property-accessor 'sqlite3 'status) exn))
	   (print-call-chain (current-error-port))))
     (tests:update-testdat-meta-info db test-id work-area cpuload diskfree minutes)
  )))
	 

;;======================================================================
;; launch a task - this runs on the originating host, tests themselves
;;
;;======================================================================


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

;; ezsteps were going to be coded as
;; stepname[,predstep1,predstep2 ...] [{VAR1=first,second,third}] command to execute
;;   BUT
;; now are
;; stepname {VAR=first,second,third ...} command ...
;; where the {VAR=first,second,third ...} is optional.

;; given an exit code and whether or not logpro was used calculate OK/BAD
;; return #t if we are ok, #f otherwise
(define (steprun-good? logpro exitcode)
  (or (eq? exitcode 0)
      (and logpro (eq? exitcode 2))))

;; if handed a string, process it, else look for MT_CMDINFO
(define (launch:get-cmdinfo-assoc-list #!key (encoded-cmd #f))
  (let ((enccmd (if encoded-cmd encoded-cmd (getenv "MT_CMDINFO"))))
    (if enccmd
	(common:read-encoded-string enccmd)
	'())))

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

(define (launch:runstep ezstep run-id test-id exit-info m tal testconfig) ;;; TODO: deprecate me in favor of ezsteps.scm
  (let* ((stepname       (car ezstep))  ;; do stuff to run the step
	 (stepinfo       (cadr ezstep))
	;; (let ((info (cadr ezstep)))
	;; 		   (if (proc? info) "" info)))
	;; (stepproc       (let ((info (cadr ezstep)))
	;; 		   (if (proc? info) info #f)))
	 (stepparts      (string-match (regexp "^(\\{([^\\}\\{]*)\\}\\s*|)(.*)$") stepinfo))
	 (stepparams     (list-ref stepparts 2)) ;; for future use, {VAR=1,2,3}, run step for each
	 (paramparts     (if (string? stepparams)
			     (map (lambda (x)(string-split x "=")) (string-split-fields "[^;]*=[^;]*" stepparams))
			     '()))
	 (subrun         (alist-ref "subrun" paramparts equal?))
	 (stepcmd        (list-ref stepparts 3))
	 (script         "") ; "#!/bin/bash\n") ;; yep, we depend on bin/bash FIXME!!!\
	 (logpro-file    (conc stepname ".logpro"))
	 (html-file      (conc stepname ".html"))
	 (dat-file       (conc stepname ".dat"))
	 (tconfig-logpro (configf:lookup testconfig "logpro" stepname))
	 (logpro-used    (common:file-exists? logpro-file)))

    (debug:print 0 *default-log-port* "stepparts: " stepparts ", stepparams: " stepparams
                 ", paramparts: " paramparts ", subrun: " subrun ", stepcmd: " stepcmd)
    
    (if (and tconfig-logpro
	     (not logpro-used)) ;; no logpro file found but have a defn in the testconfig
	(begin
	  (with-output-to-file logpro-file
	    (lambda ()
	      (print ";; logpro file extracted from testconfig\n"
		     ";;")
	      (print tconfig-logpro)))
	  (set! logpro-used #t)))
    
    ;; NB// can safely assume we are in test-area directory
    (debug:print 4 *default-log-port* "ezsteps:\n stepname: " stepname " stepinfo: " stepinfo " stepparts: " stepparts
		 " stepparams: " stepparams " stepcmd: " stepcmd)
    
    ;; ;; first source the previous environment
    ;; (let ((prev-env (conc ".ezsteps/" prevstep (if (string-search (regexp "csh") 
    ;;      							 (get-environment-variable "SHELL")) ".csh" ".sh"))))
    ;;   (if (and prevstep (common:file-exists? prev-env))
    ;;       (set! script (conc script "source " prev-env))))
    
    ;; call the command using mt_ezstep
    ;; (set! script (conc "mt_ezstep " stepname " " (if prevstep prevstep "x") " " stepcmd))
    
    (debug:print 4 *default-log-port* "script: " script)
    (rmt:teststep-set-status! run-id test-id stepname "start" "-" #f #f)
    ;; now launch the actual process
    (call-with-environment-variables 
     (list (cons "PATH" (conc (get-environment-variable "PATH") ":.")))
     (lambda () ;; (process-run "/bin/bash" "-c" "exec ls -l /tmp/foobar > /tmp/delme-more.log 2>&1")
       (let* ((cmd (conc stepcmd " > " stepname ".log 2>&1")) ;; >outfile 2>&1 
	      (pid #f))
	 (let ((proc (lambda ()
		       (set! pid (process-run "/bin/bash" (list "-c" cmd))))))
	   (if subrun
               (begin
                 (debug:print-info 0 *default-log-port* "Running without MT_.* environment variables.")
                 (common:without-vars proc "^MT_.*"))
	       (proc)))
	 
         (with-output-to-file "Makefile.ezsteps"
           (lambda ()
             (print stepname ".log :")
             (print "\t" cmd)
             (if (common:file-exists? (conc stepname ".logpro"))
                 (print "\tlogpro " stepname ".logpro " stepname ".html < " stepname ".log"))
             (print)
             (print stepname " : " stepname ".log")
             (print))
           #:append)

	 (rmt:test-set-top-process-pid run-id test-id pid)
	 (let 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)
		       (if (eq? pid-val 0)
			   (begin
			     (thread-sleep! 2)
			     (processloop (+ i 1))))
		       )))))
    (debug:print-info 0 *default-log-port* "step " stepname " completed with exit code " (launch:einf-exit-code exit-info)) ;; (vector-ref exit-info 2))
    ;; now run logpro if needed
    (if logpro-used
	(let* ((logpro-exe (or (getenv "LOGPRO_EXE") "logpro"))
               (pid        (process-run (conc "/bin/sh -c '"logpro-exe" "logpro-file " " (conc stepname ".html") " < " stepname ".log > /dev/null'"))))
	  (let processloop ((i 0))
	    (let-values (((pid-val exit-status exit-code)(process-wait pid #t)))
			(mutex-lock! m)
			;; (make-launch:einf pid: pid exit-status: exit-status exit-code: exit-code)
			(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)
			(if (eq? pid-val 0)
			    (begin
			      (thread-sleep! 2)
			      (processloop (+ i 1)))))
	    (debug:print-info 0 *default-log-port* "logpro for step " stepname " exited with code " (launch:einf-exit-code exit-info))))) ;; (vector-ref exit-info 2)))))
    
    (let ((exinfo (launch:einf-exit-code exit-info)) ;; (vector-ref exit-info 2))
	  (logfna (if logpro-used (conc stepname ".html") ""))
	  (comment #f))
      (if logpro-used
	  (let ((datfile (conc stepname ".dat")))
	    ;; load the .dat file into the test_data table if it exists
	    (if (common:file-exists? datfile)
		(set! comment (launch:load-logpro-dat run-id test-id stepname)))
	    (rmt:test-set-log! run-id test-id (conc stepname ".html"))))
      (rmt:teststep-set-status! run-id test-id stepname "end" exinfo comment logfna))
    ;; set the test final status
    (let* ((process-exit-status (launch:einf-exit-code exit-info)) ;; (vector-ref exit-info 2))
	   (this-step-status (cond
			      ((and (eq? process-exit-status 2) logpro-used) 'warn)   ;; logpro 2 = warnings
			      ((and (eq? process-exit-status 3) logpro-used) 'check)  ;; logpro 3 = check
			      ((and (eq? process-exit-status 4) logpro-used) 'waived) ;; logpro 4 = waived
			      ((and (eq? process-exit-status 5) logpro-used) 'abort)  ;; logpro 5 = abort
			      ((and (eq? process-exit-status 6) logpro-used) 'skip)   ;; logpro 6 = skip
			      ((eq? process-exit-status 0)                   'pass)   ;; logpro 0 = pass
			      (else 'fail)))
	   (overall-status   (cond
			      ((eq? (launch:einf-rollup-status exit-info) 2) 'warn) ;; rollup-status (vector-ref exit-info 3)
			      ((eq? (launch:einf-rollup-status exit-info) 0) 'pass) ;; (vector-ref exit-info 3)
			      (else 'fail)))
	   (next-status      (cond 
			      ((eq? overall-status 'pass) this-step-status)
			      ((eq? overall-status 'warn)
			       (if (eq? this-step-status 'fail) 'fail 'warn))
			      ((eq? overall-status 'abort) 'abort)
			      (else 'fail)))
	   (next-state       ;; "RUNNING") ;; WHY WAS THIS CHANGED TO NOT USE (null? tal) ??
	    (cond
	     ((null? tal) ;; more to run?
	      "COMPLETED")
	     (else "RUNNING"))))
      (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! 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! 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! 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! 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! run-id test-id next-state "SKIP" 
				 (if (eq? this-step-status 'skip) "Logpro skip found" #f)
				 #f))
	((pass)
	 (tests:test-set-status! 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! run-id test-id "COMPLETED" "FAIL" (conc "Failed at step " stepname) #f)
	 )))
    logpro-used))

(define (launch:manage-steps run-id test-id item-path fullrunscript ezsteps subrun 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 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 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:einf-rollup-status-set! exit-info  exit-code)   ;; (vector-set! exit-info 3 exit-code)  ;; rollup status
	   (mutex-unlock! m)
	   (if (eq? pid-val 0)
	       (begin
		 (thread-sleep! 2)
		 (loop (+ i 1)))
	       )))))
  ;; then, if runscript ran ok (or did not get called)
  ;; do all the ezsteps (if any)
  (if (or ezsteps subrun)
      (let* ((test-run-dir (tests:get-test-path-from-environment))
             (testconfig ;; (read-config (conc work-area "/testconfig") #f #t environ-patt: "pre-launch-env-vars")) ;; FIXME??? is allow-system ok here?
	      ;; NOTE: it is tempting to turn off force-create of testconfig but dynamic
	      ;;       ezstep names need a full re-eval here.
	      (tests:get-testconfig test-name item-path tconfigreg #t force-create: #t)) ;; 'return-procs)))
	     (ezstepslst (if (hash-table? testconfig)
			     (hash-table-ref/default testconfig "ezsteps" '())
			     #f)))
	(if testconfig
	    (hash-table-set! *testconfigs* test-name testconfig) ;; cached for lazy reads later ...
	    (begin
	      (launch:setup)
	      (debug:print 0 *default-log-port* "WARNING: no testconfig found for " test-name " in search path:\n  "
			   (string-intersperse (tests:get-tests-search-path *configdat*) "\n  "))))
	;; after all that, still no testconfig? Time to abort
	(if (not testconfig)
	    (begin
	      (debug:print-error 0 *default-log-port* "Failed to resolve megatest.config, runconfigs.config and testconfig issues. Giving up now")
	      (exit 1)))

	;; create a proc for the subrun if requested, save that proc in the ezsteps table as the last entry
	;; 1. get section [runarun]
	;; 2. unset MT_* vars
	;; 3. fix target
	;; 4. fix runname
	;; 5. fix testpatt or calculate it from contour
	;; 6. launch the run
	;; 7. roll up the run result and or roll up the logpro processed result
	(when (configf:lookup testconfig "subrun" "runwait") ;; we use runwait as the flag that a subrun is requested
            (subrun:initialize-toprun-test testconfig test-run-dir)
	    (let* ((mt-cmd (subrun:launch-cmd test-run-dir)))
              (debug:print-info 0 *default-log-port* "Subrun command is \"" mt-cmd "\"")
              (set! ezsteps #t) ;; set the needed flag
	      (set! ezstepslst
                    (append (or ezstepslst '())
                            (list (list "subrun" (conc "{subrun=true} " mt-cmd)))))))

	;; process the ezsteps
	(if ezsteps
	    (begin
	      (if (not (common:file-exists? ".ezsteps"))(create-directory ".ezsteps"))
	      ;; if ezsteps was defined then we are sure to have at least one step but check anyway
	      (if (not (> (length ezstepslst) 0))
		  (debug:print-error 0 *default-log-port* "ezsteps defined but ezstepslst is zero length")
		  (let loop ((ezstep (car ezstepslst))
			     (tal    (cdr ezstepslst))
			     (prevstep #f))
                    (debug:print-info 0 *default-log-port* "Processing ezstep \"" (string-intersperse ezstep " ") "\"")
		    ;; check exit-info (vector-ref exit-info 1)
		    (if (launch:einf-exit-status exit-info) ;; (vector-ref exit-info 1)
			(let ((logpro-used (launch:runstep ezstep run-id test-id exit-info m tal testconfig))
			      (stepname    (car ezstep)))
			  ;; if logpro-used read in the stepname.dat file
			  (if (and logpro-used (common:file-exists? (conc stepname ".dat")))
			      (launch:load-logpro-dat run-id test-id stepname))
			  (if (steprun-good? logpro-used (launch:einf-exit-code exit-info))
			      (if (not (null? tal))
				  (loop (car tal) (cdr tal) stepname))
			      (debug:print 0 *default-log-port* "WARNING: step " (car ezstep) " failed. Stopping")))
			(debug:print 0 *default-log-port* "WARNING: a prior step failed, stopping at " ezstep)))))))))

(define (launch:monitor-job run-id test-id item-path fullrunscript ezsteps test-name tconfigreg exit-info m work-area runtlim misc-flags)
  (let* ((update-period (string->number (or (configf:lookup *configdat* "setup" "test-stats-update-period") "30")))
         (start-seconds (current-seconds))
	 (calc-minutes  (lambda ()
			  (inexact->exact 
			   (round 
			    (- 
			     (current-seconds) 
			     start-seconds)))))
	 (kill-tries 0))
    ;; (tests:set-full-meta-info #f test-id run-id (calc-minutes) work-area)
    ;; (tests:set-full-meta-info test-id run-id (calc-minutes) work-area)
    (tests:set-full-meta-info #f test-id run-id (calc-minutes) work-area 10)

    (let loop ((minutes   (calc-minutes))
	       (cpu-load  (alist-ref 'adj-core-load (common:get-normalized-cpu-load #f)))
	       (disk-free (get-df (current-directory)))
               (last-sync (current-seconds)))
      #;(common:telemetry-log "zombie" (conc "launch:monitor-job - top of loop encountered at "(current-seconds)" with last-sync="last-sync))
      (let* ((over-time     (> (current-seconds) (+ last-sync update-period)))
             (new-cpu-load  (let* ((load  (alist-ref 'adj-core-load (common:get-normalized-cpu-load #f)))
                                   (delta (abs (- load cpu-load))))
                              (if (> delta 0.1) ;; don't bother updating with small changes
                                  load
                                  #f)))
             (new-disk-free (let* ((df    (if over-time ;; only get df every 30 seconds
                                              (get-df (current-directory))
                                              disk-free))
                                   (delta (abs (- df disk-free))))
                              (if (and (> df 0)
                                       (> (/ delta df) 0.1)) ;; (> delta 200) ;; ignore changes under 200 Meg
                                  df
                                  #f)))
             (do-sync       (or new-cpu-load new-disk-free over-time))

             (test-info   (rmt:get-test-info-by-id run-id test-id))
             (state       (db:test-get-state test-info))
             (status      (db:test-get-status test-info))
             (kill-reason  "no kill reason specified")
             (kill-job?    #f))
        #;(common:telemetry-log "zombie" (conc "launch:monitor-job - decision time encountered at "(current-seconds)" with last-sync="last-sync" do-sync="do-sync" over-time="over-time" update-period="update-period))
        (cond
         ((test-get-kill-request run-id test-id)
          (set! kill-reason "KILLING TEST since received kill request (KILLREQ)")
          (set! kill-job? #t))
         ((and runtlim (> (- (current-seconds) start-seconds) runtlim))
          (set! kill-reason (conc "KILLING TEST DUE TO TIME LIMIT EXCEEDED! Runtime=" (- (current-seconds) start-seconds) " seconds, limit=" runtlim))
          (set! kill-job? #t))
         ((equal? status "DEAD")
          (tests:update-central-meta-info run-id test-id new-cpu-load new-disk-free (calc-minutes) #f #f)
          (rmt:set-state-status-and-roll-up-items run-id test-id 'foo "RUNNING" "n/a" "was marked dead; really still running.")
          ;;(set! kill-reason "KILLING TEST because it was marked as DEAD by launch:handle-zombie-tests (might indicate really overloaded server or else overzealous setup.deadtime)") ;; MARK RUNNING
          (set! kill-job? #f)))

        (debug:print 4 *default-log-port* "cpu: " new-cpu-load " disk: " new-disk-free " last-sync: " last-sync " do-sync: " do-sync)
        (launch:handle-zombie-tests run-id)
        (when do-sync
          ;;(with-output-to-file (conc (getenv "MT_TEST_RUN_DIR") "/last-loadinfo.log" #:append)
          ;;  (lambda () (pp (list (current-seconds) new-cpu-load new-disk-free (calc-minutes)))))
          #;(common:telemetry-log "zombie" (conc  "launch:monitor-job - dosync started at "(current-seconds)))
          (tests:update-central-meta-info run-id test-id new-cpu-load new-disk-free (calc-minutes) #f #f)
          #;(common:telemetry-log "zombie" (conc "launch:monitor-job - dosync finished at "(current-seconds))))
        
	(if kill-job? 
	    (begin
              (debug:print-info 0 *default-log-port* "proceeding to kill test: "kill-reason)
	      (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 run-id test-id))
		     (pids (delete-duplicates (filter number? (list pid1 pid2)))))
		(if (not (null? pids))
		    (begin
		      (for-each
		       (lambda (pid)
			 (handle-exceptions
			  exn
			  (begin
			    (debug:print-info 0 *default-log-port* "Unable to kill process with pid " pid ", possibly already killed.")
			    (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn)))
			  (debug:print 0 *default-log-port* "WARNING: Request received to kill job " pid) ;;  " (attempt # " kill-tries ")")
			  (debug:print-info 0 *default-log-port* "Signal mask=" (signal-mask))
			  ;; (if (process:alive? pid)
			  ;;     (begin
			  (map (lambda (pid-num)
				 (process-signal pid-num signal/term))
			       (process:get-sub-pids pid))
			  (thread-sleep! 5)
			  ;; (if (process:process-alive? pid)
			  (map (lambda (pid-num)
				 (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)
                      ;; BB: question to Matt -- does the tests:test-state-status! encompass rollup to toplevel?  If not, should it?
		      (tests:test-set-status! run-id test-id "KILLED"  "KILLED" (conc (args:get-arg "-m")" "kill-reason) #f)) ;; BB ADDED kill-reason -- confirm OK with Matt
		    (begin
		      (debug:print-error 0 *default-log-port* "Nothing to kill, pid1=" pid1 ", pid2=" pid2)
		      (tests:test-set-status! run-id test-id "KILLED"  "FAILED TO KILL" (conc (args:get-arg "-m")" "kill-reason) #f) ;; BB ADDED kill-reason -- confirm OK with Matt
		      )))
	      (mutex-unlock! m)
	      ;; no point in sticking around. Exit now. But run end of run before exiting?
        (launch:end-of-run-check run-id)
	      (exit)))
	(if (hash-table-ref/default misc-flags 'keep-going #f)
	    (begin
	      (thread-sleep! 3) ;; (+ 3 (random 6))) ;; add some jitter to the call home time to spread out the db accesses
	      (if (hash-table-ref/default misc-flags 'keep-going #f)  ;; keep originals for cpu-load and disk-free unless they change more than the allowed delta
		  (loop (calc-minutes)
                        (or new-cpu-load cpu-load)
                        (or new-disk-free disk-free)
                        (if do-sync (current-seconds) last-sync)))))))
    (tests:update-central-meta-info run-id test-id (get-cpu-load) (get-df (current-directory))(calc-minutes) #f #f))) ;; NOTE: Checking twice for keep-going is intentional

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

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

    (if (not (get-environment-variable "MT_TARGET"))(setenv "MT_TARGET" target))
    ;; we had a case where there was an exception generated by the hash-table-ref
    ;; due to *configdat* being #f Adding a handle and exit
    (let fatal-loop ((count 0)) 
      (handle-exceptions
	  exn
	  (let ((call-chain (get-call-chain))
		(msg        ((condition-property-accessor 'exn 'message) exn)))
	    (if (< count 5)
		(begin ;; this call is colliding, do some crude stuff to fix it.
		  (debug:print 0 *default-log-port* "ERROR: *configdat* was inaccessible! This should never happen. Retry #" count)
		  (launch:setup force-reread: #t)
		  (fatal-loop (+ count 1))) 
		(begin
		  (debug:print 0 *default-log-port* "FATAL: *configdat* was inaccessible! This should never happen. Retried " count " times. Message: " msg)
		  (debug:print 0 *default-log-port* "Call chain:")
		  (with-output-to-port *default-log-port*

                    (lambda ()
                      (print "*configdat* is >>"*configdat*"<<")
                      (pp *configdat*)
                      (pp call-chain)))
                  
		  (exit 1))))
          ;;(bb-check-path msg: "runs:set-megatest-env-vars block 1.5")
          (when (or (not *configdat*) (not (hash-table? *configdat*)))
              (debug:print 0 *default-log-port* "WARNING: *configdat* was inaccessible! This should never happen.  Brute force reread.")
              ;;(BB> "ERROR: *configdat* was inaccessible! This should never happen.  Brute force reread.")
              (thread-sleep! 2) ;; assuming nfs lag.
              (launch:setup force-reread: #t))
          (alist->env-vars (hash-table-ref/default *configdat* "env-override" '())))) ;;;; environment is tainted HERE in this let block.
    ;;(bb-check-path msg: "runs:set-megatest-env-vars block 2")
    ;; Lets use this as an opportunity to put MT_RUNNAME in the environment
    (let ((runname  (if inrunname inrunname (rmt:get-run-name-from-id run-id))))
      (if runname
	  (setenv "MT_RUNNAME" runname)
	  (debug:print-error 0 *default-log-port* "no value for runname for id " run-id)))
    (setenv "MT_RUN_AREA_HOME" *toppath*)
    ;; if a testname and itempath are available set the remaining appropriate variables
    (if testname (setenv "MT_TEST_NAME" testname))
    (if itempath (setenv "MT_ITEMPATH"  itempath))
    ;;(bb-check-path msg: "runs:set-megatest-env-vars block 3")
    (if (and testname link-tree)
	(setenv "MT_TEST_RUN_DIR" (conc (getenv "MT_LINKTREE")  "/"
					(getenv "MT_TARGET")    "/"
					(getenv "MT_RUNNAME")   "/"
					(getenv "MT_TEST_NAME")
					(if (and itempath
						 (not (equal? itempath "")))
					    (conc "/" itempath)
					    ""))))))

(define (launch:execute encoded-cmd)
  (let* ((cmdinfo    (common:read-encoded-string encoded-cmd))
	 (tconfigreg #f))
    (setenv "MT_CMDINFO" encoded-cmd)
    ;;(bb-check-path msg: "launch:execute incoming")
    (if (list? cmdinfo) ;; ((testpath /tmp/mrwellan/jazzmind/src/example_run/tests/sqlitespeed)
	;; (test-name sqlitespeed) (runscript runscript.rb) (db-host localhost) (run-id 1))
	(let* ((testpath  (assoc/default 'testpath  cmdinfo))  ;; testpath is the test spec area
	       (top-path  (assoc/default 'toppath   cmdinfo))
	       (work-area (assoc/default 'work-area cmdinfo))  ;; work-area is the test run area
	       (test-name (assoc/default 'test-name cmdinfo))
	       (runscript (assoc/default 'runscript cmdinfo))
	       (ezsteps   (assoc/default 'ezsteps   cmdinfo))
	       (subrun    (assoc/default 'subrun    cmdinfo))
	       ;; (runremote (assoc/default 'runremote cmdinfo))
	       ;; (transport (assoc/default 'transport cmdinfo))  ;; not used
	       ;; (serverinf (assoc/default 'serverinf cmdinfo))
	       ;; (port      (assoc/default 'port      cmdinfo))
	       (serverurl (assoc/default 'serverurl cmdinfo))
	       (homehost  (assoc/default 'homehost  cmdinfo))
	       (run-id    (assoc/default 'run-id    cmdinfo))
	       (test-id   (assoc/default 'test-id   cmdinfo))
	       (target    (assoc/default 'target    cmdinfo))
	       (areaname  (assoc/default 'areaname  cmdinfo))
	       (itemdat   (assoc/default 'itemdat   cmdinfo))
	       (env-ovrd  (assoc/default 'env-ovrd  cmdinfo))
	       (set-vars  (assoc/default 'set-vars  cmdinfo)) ;; pre-overrides from -setvar
	       (runname   (assoc/default 'runname   cmdinfo))
	       (megatest  (assoc/default 'megatest  cmdinfo))
	       (runtlim   (assoc/default 'runtlim   cmdinfo))
	       (contour   (assoc/default 'contour   cmdinfo))
	       (item-path (item-list->path itemdat))
	       (mt-bindir-path (assoc/default 'mt-bindir-path cmdinfo))
	       (keys      #f)
	       (keyvals   #f)
	       (fullrunscript (if (not runscript)
                                  #f
                                  (if (substring-index "/" runscript)
                                      runscript ;; use unadultered if contains slashes
                                      (let ((fulln (conc work-area "/" runscript)))
	                                  (if (and (common:file-exists? fulln)
                                                   (file-execute-access? fulln))
                                              fulln
                                              runscript))))) ;; assume it is on the path
               (check-work-area           (lambda ()
                                            ;; NFS might not have propagated the directory meta data to the run host - give it time if needed
                                            (let loop ((count 0))
                                              (if (or (common:directory-exists? work-area)
                                                      (> count 10))
                                                  (change-directory work-area)
                                                  (begin
                                                    (debug:print 0 *default-log-port* "INFO: Not starting job yet - directory " work-area " not found")
                                                    (thread-sleep! 10)
                                                    (loop (+ count 1)))))

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

	  (if contour (setenv "MT_CONTOUR" contour))
	  
	  ;; immediated set some key variables from CMDINFO data, yes, these will be set again below ...
	  ;;
	  (setenv "MT_TESTSUITENAME" areaname)
	  (setenv "MT_RUN_AREA_HOME" top-path)
	  (set! *toppath* top-path)
          (change-directory *toppath*) ;; temporarily switch to the run area home
	  (setenv "MT_TEST_RUN_DIR"  work-area)

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

	  (if contour (setenv "MT_CONTOUR" contour))
	  
	  ;; immediated set some key variables from CMDINFO data, yes, these will be set again below ...
	  ;;
	  (setenv "MT_TESTSUITENAME" areaname)
	  (setenv "MT_RUN_AREA_HOME" top-path)
	  (set! *toppath* top-path)
          (change-directory *toppath*) ;; temporarily switch to the run area home
	  (setenv "MT_TEST_RUN_DIR"  work-area)

	  (launch:setup) ;; should be properly in the run area home now
          
	  (set! tconfigreg (tests:get-all)) ;; mapping of testname => test source path
	  (let ((sighand (lambda (signum)
			   ;; (signal-mask! signum) ;; to mask or not? seems to cause issues in exiting
			   (if (eq? signum signal/stop)
			       (debug:print-error 0 *default-log-port* "attempt to STOP process. Exiting."))
			   (set! *time-to-exit* #t)
			   (print "Received signal " signum ", cleaning up before exit (set this test to COMPLETED/ABORT) . Please wait...")
			   (let ((th1 (make-thread (lambda ()
                                                     (print "set test to COMPLETED/ABORT begin.")
						     (rmt:test-set-state-status run-id test-id "COMPLETED" "ABORT" "received kill signal")
                                                     (print "set test to COMPLETED/ABORT complete.")
						     (print "Killed by signal " signum ". Exiting")
						     (exit 1))))
				 (th2 (make-thread (lambda ()
						     (thread-sleep! 20)
						     (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 run-id test-id))
		 (test-host (if test-info
				(db:test-get-host        test-info)
				(begin
				  (debug:print 0 *default-log-port* "ERROR: failed to find a record for test-id " test-id ", exiting.")
				  (exit))))
		 (test-pid  (db:test-get-process_id  test-info)))
	    (cond
             ;; -mrw- I'm removing KILLREQ from this list so that a test in KILLREQ state is treated as a "do not run" flag.
	     ((member (db:test-get-state test-info) '("INCOMPLETE" "KILLED" "UNKNOWN" "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:general-call 'set-test-start-time #f test-id)
              (rmt:test-set-state-status 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")
		  (exit)))
	     ((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:general-call 'set-test-start-time #f test-id)
	      (rmt:test-set-state-status 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))))

          ;; cleanup prior execution's steps
          (rmt:delete-steps-for-test! run-id test-id)
          
	  (debug:print 2 *default-log-port* "Executing " test-name " (id: " test-id ") on " (get-host-name))
	  (set! keys       (rmt:get-keys))
	  ;; (runs:set-megatest-env-vars run-id inkeys: keys inkeyvals: keyvals) ;; these may be needed by the launching process
	  ;; one of these is defunct/redundant ...
	  (if (not (launch:setup force-reread: #t))
	      (begin
		(debug:print 0 *default-log-port* "Failed to setup, exiting") 
		;; (sqlite3:finalize! db)
		;; (sqlite3:finalize! tdb)
		(exit 1)))
          ;; validate that the test run area is available
          (check-work-area)
          
          ;; still need to go back to run area home for next couple steps
	  (change-directory *toppath*) 

	  ;; NOTE: Current order is to process runconfigs *before* setting the MT_ vars. This 
	  ;;       seems non-ideal but could well break stuff
	  ;;    BUG? BUG? BUG?
	  
	  (let ((rconfig (full-runconfigs-read)) ;; (read-config (conc  *toppath* "/runconfigs.config") #f #t sections: (list "default" target))))
		(wconfig (configf:read-config "waivers.config" #f #t sections: `( "default" ,target )))) ;; read the waivers config if it exists
	    ;; (setup-env-defaults (conc *toppath* "/runconfigs.config") run-id (make-hash-table) keyvals target)
	    ;; (set-run-config-vars run-id keyvals target) ;; (db:get-target db run-id))
	    ;; Now have runconfigs data loaded, set environment vars
	    (for-each
	     (lambda (section)
	       (for-each
		(lambda (varval)
		  (let ((var (car varval))
			(val (cadr varval)))
		    (if (and (string? var)(string? val))
			(begin
			  (safe-setenv var (configf:eval-string-in-environment val))) ;; val)
			(debug:print-error 0 *default-log-port* "bad variable spec, " var "=" val))))
		(configf:get-section rconfig section)))
	     (list "default" target)))
          ;;(bb-check-path msg: "launch:execute post block 1")

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

          ;; now we can switch to the work-area?
          (change-directory work-area)
          ;;(bb-check-path msg: "launch:execute post block 1.5")
	  ;; (change-directory work-area) 
	  (set! keyvals    (keys:target->keyval keys target))
	  ;; apply pre-overrides before other variables. The pre-override vars must not
	  ;; clobbers things from the official sources such as megatest.config and runconfigs.config
	  (if (string? set-vars)
	      (let ((varpairs (string-split set-vars ",")))
		(debug:print 4 *default-log-port* "varpairs: " varpairs)
		(map (lambda (varpair)
		       (let ((varval (string-split varpair "=")))
			 (if (eq? (length varval) 2)
			     (let ((var (car varval))
				   (val (cadr varval)))
			       (debug:print 1 *default-log-port* "Adding pre-var/val " var " = " val " to the environment")
			       (setenv var val)))))
		     varpairs)))
          ;;(bb-check-path msg: "launch:execute post block 2")
	  (for-each
	   (lambda (varval)
	     (let ((var (car varval))
		   (val (cadr varval)))
	       (if val
		   (setenv var val)
		   (begin
		     (debug:print-error 0 *default-log-port* "required variable " var " does not have a valid value. Exiting")
		     (exit)))))
	     (list 
	      (list  "MT_TEST_RUN_DIR" work-area)
	      (list  "MT_TEST_NAME" test-name)
	      (list  "MT_ITEM_INFO" (conc itemdat))
	      (list  "MT_ITEMPATH"  item-path)
	      (list  "MT_RUNNAME"   runname)
	      (list  "MT_MEGATEST"  megatest)
	      (list  "MT_TARGET"    target)
	      (list  "MT_LINKTREE"  (common:get-linktree)) ;; (configf:lookup *configdat* "setup" "linktree"))
	      (list  "MT_TESTSUITENAME" (common:get-testsuite-name))))
          ;;(bb-check-path msg: "launch:execute post block 3")

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

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

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

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

	  ;; We are about to actually kick off the test
	  ;; so this is a good place to remove the records for 
	  ;; any previous runs
	  ;; (db:test-remove-steps db run-id testname itemdat)
	  ;; now is also a good time to write the .testconfig file
	  (let* ((tconfig-fname   (conc work-area "/.testconfig"))
		 (tconfig-tmpfile (conc tconfig-fname ".tmp"))
		 (tconfig         (tests:get-testconfig test-name item-path tconfigreg #t force-create: #t))) ;; 'return-procs)))
	    (configf:write-alist tconfig tconfig-tmpfile)
	    (file-move tconfig-tmpfile tconfig-fname #t))
	  ;; 
	  (let* ((m            (make-mutex))
		 (kill-job?    #f)
		 (exit-info    (make-launch:einf pid: #t exit-status: #t exit-code: #t rollup-status: 0)) ;; pid exit-status exit-code (i.e. process was successfully run) rollup-status
		 (job-thread   #f)
		 ;; (keep-going   #t)
		 (misc-flags   (let ((ht (make-hash-table)))
				 (hash-table-set! ht 'keep-going #t)
				 ht))
		 (runit        (lambda ()
				 (launch:manage-steps run-id test-id item-path fullrunscript ezsteps subrun test-name tconfigreg exit-info m)))
		 (monitorjob   (lambda ()
				 (launch:monitor-job  run-id test-id item-path fullrunscript ezsteps test-name tconfigreg exit-info m work-area runtlim misc-flags)))
		 (th1          (make-thread monitorjob "monitor job"))
		 (th2          (make-thread runit "run job")))
	    (set! job-thread th2)
	    (thread-start! th1)
	    (thread-start! th2)
	    (thread-join! th2)
	    (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 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
				     ((not (launch:einf-exit-status exit-info)) "FAIL") ;; job failed to run ... (vector-ref exit-info 1)
				     ((eq? (launch:einf-rollup-status exit-info) 0)     ;; (vector-ref exit-info 3)
				      ;; if the current status is AUTO then defer to the calculated value (i.e. leave this AUTO)
				      (if (equal? (db:test-get-status testinfo) "AUTO") "AUTO" "PASS"))
				     ((eq? (launch:einf-rollup-status exit-info) 1) "FAIL")  ;; (vector-ref exit-info 3)
				     ((eq? (launch:einf-rollup-status exit-info) 2)	     ;;	(vector-ref exit-info 3)
				      ;; 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"))
				     ((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! 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 run-id test-id test-name #f))
	      (tests:summarize-test run-id test-id)  ;; don't force - just update if no
	      (rmt:update-run-stats run-id (rmt:get-raw-run-stats run-id)))
	    (mutex-unlock! m)
            (launch:end-of-run-check run-id )
	    (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:is-test-alive host pid)
(if (and host pid (not (equal? host "n/a")))
(let* ((cmd (conc "ssh " host " pstree -A " pid))
      (output (with-input-from-pipe cmd read-lines)))
  (print "cmd: " cmd "\n op: " output )
  (if(eq? (length output) 0)
     #f
     #t))
#t))
 
(define (launch:kill-tests-if-dead run-id)
  (let* ((running-tests (rmt:get-tests-for-run run-id "%" `("RUNNING" "LAUNCHED" "REMOTEHOSTSTART") `() #f #f #f #f #f #f #f #f)))
       (let loop ((running-test (car running-tests))
			     (tal    (cdr running-tests))
			     (kill-cnt 0))
		       (let* ((test-name (vector-ref running-test 2))
                 (item-path (vector-ref running-test 11))
								 (test-id (vector-ref running-test 0))
                 (host (vector-ref running-test 6))
                 (pid  (rmt:test-get-top-process-pid run-id test-id))   
                 (event-time (vector-ref running-test 5))
                 (duration (vector-ref running-test 12))
                 (flag 0)   
                 (curr-time (current-seconds)))
       (if (and (< (+ event-time duration 600) curr-time) (not (launch:is-test-alive host pid))) ;;test has not updated duration in last 10 min then likely its not running but confirm before marking it as killed
           (begin    
			       	(debug:print 0 *default-log-port* "test " test-name "/" item-path " needs to be killed")
              (set! flag 1) 
              (rmt:set-state-status-and-roll-up-items run-id test-name item-path "KILLREQ" "n/a" #f)))
               (if (not (null? tal))
				  (loop (car tal) (cdr tal) (+ kill-cnt flag))
                 (+ kill-cnt flag))))))

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


;; gather available information, if legit read configs in this order:
;;
;;   if have cache;
;;      read it a return it
;;   else
;;     megatest.config     (do not cache)
;;     runconfigs.config   (cache if all vars avail)
;;     megatest.config     (cache if all vars avail)
;;   returns:
;;     *toppath*
;;   side effects:
;;     sets; *configdat*    (megatest.config info)
;;           *runconfigdat* (runconfigs.config info)
;;           *configstatus* (status of the read data)
;;
(define (launch:setup #!key (force-reread #f) (areapath #f))
  (mutex-lock! *launch-setup-mutex*)
  (if (and *toppath*
	   (eq? *configstatus* 'fulldata) (not force-reread)) ;; got it all
      (begin
	(debug:print 2 *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-reread: force-reread areapath: areapath)))
	(mutex-unlock! *launch-setup-mutex*)
	res)))

;; return paths depending on what info is available.
;;
(define (launch:get-cache-file-paths areapath toppath target mtconfig)
  (let* ((use-cache (common:use-cache?))
         (runname  (common:args-get-runname))
         (linktree (common:get-linktree))
         (testname (common:get-full-test-name))
         (rundir   (if (and runname target linktree)
                       (common:directory-writable? (conc linktree "/" target "/" runname))
                       #f))
         (testdir  (if (and rundir testname)
                       (common:directory-writable? (conc rundir "/" testname))
                       #f))
         (cachedir (or testdir rundir))
         (mtcachef (and cachedir (conc cachedir "/" ".megatest.cfg-"  megatest-version "-" megatest-fossil-hash)))
         (rccachef (and cachedir (conc cachedir "/" ".runconfigs.cfg-"  megatest-version "-" megatest-fossil-hash))))
    (debug:print-info 6 *default-log-port* 
                      "runname=" runname 
                      "\n  linktree=" linktree
                      "\n  testname=" testname
                      "\n  rundir=" rundir 
                      "\n  testdir=" testdir 
                      "\n  cachedir=" cachedir
                      "\n  mtcachef=" mtcachef
                      "\n  rccachef=" rccachef)
    (cons mtcachef rccachef)))

(define (launch:setup-body #!key (force-reread #f) (areapath #f))
  (if (and (eq? *configstatus* 'fulldata)
	   *toppath*
	   (not force-reread)) ;; no need to reprocess
      *toppath*   ;; return toppath
      (let* ((use-cache (common:use-cache?)) ;; BB- use-cache checks *configdat* for use-cache setting.  We do not have *configdat*.  Bootstrapping problem here.
	     (toppath  (or *toppath* areapath (getenv "MT_RUN_AREA_HOME"))) ;; preserve toppath
	     (target   (common:args-get-target))
	     (sections (if target (list "default" target) #f)) ;; for runconfigs
	     (mtconfig (or (args:get-arg "-config") "megatest.config")) ;; allow overriding megatest.config 
             (cachefiles (launch:get-cache-file-paths areapath toppath target mtconfig))
	     ;; checking for null cachefiles should not be necessary, I was seeing error car of '(), might be a chicken bug or a red herring ...
	     (mtcachef   (if (null? cachefiles)
			     #f
			     (car cachefiles))) ;; (and cachedir (conc cachedir "/" ".megatest.cfg-"  megatest-version "-" megatest-fossil-hash)))
	     (rccachef   (if (null? cachefiles)
			     #f
			     (cdr cachefiles)))) ;; (and cachedir (conc cachedir "/" ".runconfigs.cfg-"  megatest-version "-" megatest-fossil-hash)))
	      ;; (cancreate (and cachedir (common:file-exists? cachedir)(file-write-access? cachedir) (not (common:in-running-test?)))))
	(set! *toppath* toppath) ;; This is needed when we are running as a test using CMDINFO as a datasource
        ;;(BB> "launch:setup-body -- cachefiles="cachefiles)
	(cond
	 ;; if mtcachef exists just read it, however we need to assume toppath is available in $MT_RUN_AREA_HOME
	 ((and (not force-reread)
	       mtcachef  rccachef
	       use-cache
	       (get-environment-variable "MT_RUN_AREA_HOME")
	       (common:file-exists? mtcachef)
	       (common:file-exists? rccachef))
          ;;(BB> "launch:setup-body -- cond branch 1 - use-cache")
          (set! *configdat*    (configf:read-alist mtcachef))
          ;;(BB> "launch:setup-body -- 1 set! *configdat*="*configdat*)
	  (set! *runconfigdat* (configf:read-alist rccachef))
	  (set! *configinfo*   (list *configdat*  (get-environment-variable "MT_RUN_AREA_HOME")))
	  (set! *configstatus* 'fulldata)
	  (set! *toppath*      (get-environment-variable "MT_RUN_AREA_HOME"))
	  *toppath*)
	 ;; there are no existing cached configs, do full reads of the configs and cache them
	 ;; we have all the info needed to fully process runconfigs and megatest.config
	 ((and ;; (not force-reread) ;; force-reread is irrelevant in the AND, could however OR it?
	       mtcachef
	       rccachef) ;; BB- why are we doing this without asking if caching is desired?
          ;;(BB> "launch:setup-body -- cond branch 2")
	  (let* ((first-pass    (configf:find-and-read-config        ;; NB// sets MT_RUN_AREA_HOME as side effect
				 mtconfig
				 environ-patt: "env-override"
				 given-toppath: toppath
				 pathenvvar: "MT_RUN_AREA_HOME"))
		 (first-rundat  (let ((toppath (if toppath 
						   toppath
						   (car first-pass))))
				  (configf:read-config ;; (conc toppath "/runconfigs.config") ;; this should be converted to runconfig:read but it is non-trivial, leaving it for now.
				   (conc (if (string? toppath)
					     toppath
					     (get-environment-variable "MT_RUN_AREA_HOME"))
					 "/runconfigs.config")
				   *runconfigdat* #t 
				   sections: sections))))
	    (set! *runconfigdat* first-rundat)
	    (if first-pass  ;; 
		(begin
                  ;;(BB> "launch:setup-body -- \"first-pass\"=first-pass")
		  (set! *configdat*  (car first-pass))
                  ;;(BB> "launch:setup-body -- 2 set! *configdat*="*configdat*)
		  (set! *configinfo* first-pass)
		  (set! *toppath*    (or toppath (cadr first-pass))) ;; use the gathered data unless already have it
		  (set! toppath      *toppath*)
		  (if (not *toppath*)
		      (begin
			(debug:print-error 0 *default-log-port* "you are not in a megatest area!")
			(exit 1)))
		  (setenv "MT_RUN_AREA_HOME" *toppath*)
		  ;; the seed read is done, now read runconfigs, cache it then read megatest.config one more time and cache it
		  (let* ((keys         (rmt:get-keys))
			 (key-vals     (keys:target->keyval keys target))
			 (linktree     (common:get-linktree)) ;; (or (getenv "MT_LINKTREE")(if *configdat* (configf:lookup *configdat* "setup" "linktree") #f)))
					;     (if *configdat*
					; 	   (configf:lookup *configdat* "setup" "linktree")
					; 	   (conc *toppath* "/lt"))))
			 (second-pass  (configf:find-and-read-config
					mtconfig
					environ-patt: "env-override"
					given-toppath: toppath
					pathenvvar: "MT_RUN_AREA_HOME"))
			 (runconfigdat (begin     ;; this read of the runconfigs will see any adjustments made by re-reading megatest.config
					 (for-each (lambda (kt)
						     (setenv (car kt) (cadr kt)))
						   key-vals)
					 (configf:read-config (conc toppath "/runconfigs.config") *runconfigdat* #t ;; consider using runconfig:read some day ...
						      sections: sections)))
                         (cachefiles   (launch:get-cache-file-paths areapath toppath target mtconfig))
                         (mtcachef     (car cachefiles))
                         (rccachef     (cdr cachefiles)))
                    ;;  trap exception due to stale NFS handle -- Error: (open-output-file) cannot open file - Stale NFS file handle: "/p/fdk/gwa/lefkowit/mtTesting/qa/primbeqa/links/p1222/11/PDK_r1.1.1/prim/clean/pcell_testgen/.runconfigs.cfg-1.6427-7d1e789cb3f62f9cde719a4865bb51b3c17ea853" - ticket 220546342
                    ;; TODO - consider 1) using simple-lock to bracket cache write
                    ;;                 2) cache in hash on server, since need to do rmt: anyway to lock.

		    (if rccachef
                        (common:fail-safe
                         (lambda ()
                           (configf:write-alist runconfigdat rccachef))
                         (conc "Could not write cache file - "rccachef)))
                    (if mtcachef
                        (common:fail-safe
                         (lambda ()
                           (configf:write-alist *configdat* mtcachef))
                         (conc "Could not write cache file - "mtcachef)))
		    (set! *runconfigdat* runconfigdat)
		    (if (and rccachef mtcachef) (set! *configstatus* 'fulldata))))
		;; no configs found? should not happen but let's try to recover gracefully, return an empty hash-table
		(set! *configdat* (make-hash-table))
		)))

	 ;; else read what you can and set the flag accordingly
	 ;; here we don't have either mtconfig or rccachef
	 (else
          ;;(BB> "launch:setup-body -- cond branch 3 - else")
	  (let* ((cfgdat   (configf:find-and-read-config 
			    (or (args:get-arg "-config") "megatest.config")
			    environ-patt: "env-override"
			    given-toppath: (get-environment-variable "MT_RUN_AREA_HOME")
			    pathenvvar: "MT_RUN_AREA_HOME")))

            (if (and cfgdat (list? cfgdat) (> (length cfgdat) 0) (hash-table? (car cfgdat)))
		(let* ((toppath  (or (get-environment-variable "MT_RUN_AREA_HOME")(cadr cfgdat)))
		       (rdat     (configf:read-config (conc toppath  ;; convert this to use runconfig:read!
						    "/runconfigs.config") *runconfigdat* #t sections: sections)))
		  (set! *configinfo*   cfgdat)
		  (set! *configdat*    (car cfgdat))
		  (set! *runconfigdat* rdat)
		  (set! *toppath*      toppath)
		  (set! *configstatus* 'partial))
		(begin
		  (debug:print-error 0 *default-log-port* "No " mtconfig " file found. Giving up.")
		  (exit 2))))))
	;; COND ends here.
	
	;; additional house keeping
	(let* ((linktree (or (common:get-linktree)
			     (conc *toppath* "/lt"))))
	  (if linktree
	      (begin
		(if (not (common:file-exists? linktree))
		    (begin
		      (handle-exceptions
			  exn
			  (begin
			    (debug:print-error 0 *default-log-port* "Something went wrong when trying to create linktree dir at " linktree)
			    (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
			    (exit 1))
			(create-directory linktree #t))))
		(handle-exceptions
		    exn
		    (begin
		      (debug:print-error 0 *default-log-port* "Something went wrong when trying to create link to linktree at " *toppath*)
		      (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn)))
		  (let ((tlink (conc *toppath* "/lt")))
		    (if (not (common:file-exists? tlink))
			(create-symbolic-link linktree tlink)))))
	      (begin
		(debug:print-error 0 *default-log-port* "linktree not defined in [setup] section of megatest.config")
		)))
	(if (and *toppath*
		 (directory-exists? *toppath*))
	    (begin
	      (setenv "MT_RUN_AREA_HOME" *toppath*)
	      (setenv "MT_TESTSUITENAME" (common:get-testsuite-name)))
	    (begin
	      (debug:print-error 0 *default-log-port* "failed to find the top path to your Megatest area.")
	      (set! *toppath* #f) ;; force it to be false so we return #f
	      #f))
	
        ;; one more attempt to cache the configs for future reading
        (let* ((cachefiles   (launch:get-cache-file-paths areapath toppath target mtconfig))
               (mtcachef     (car cachefiles))
               (rccachef     (cdr cachefiles)))

          ;; trap exception due to stale NFS handle -- Error: (open-output-file) cannot open file - Stale NFS file handle: "...somepath.../.runconfigs.cfg-1.6427-7d1e789cb3f62f9cde719a4865bb51b3c17ea853" - ticket 220546342
          ;; TODO - consider 1) using simple-lock to bracket cache write
          ;;                 2) cache in hash on server, since need to do rmt: anyway to lock.
          (if (and rccachef *runconfigdat* (not (common:file-exists? rccachef)))
              (common:fail-safe
               (lambda ()
                 (configf:write-alist *runconfigdat* rccachef))
               (conc "Could not write cache file - "rccachef))
              )
          (if (and mtcachef *configdat*    (not (common:file-exists? mtcachef)))
              (common:fail-safe
               (lambda ()
                 (configf:write-alist *configdat* mtcachef))
               (conc "Could not write cache file - "mtcachef))
              )
          (if (and rccachef mtcachef *runconfigdat* *configdat*)
              (set! *configstatus* 'fulldata)))

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

(define (get-best-disk confdat testconfig)
  (let* ((disks   (or (and testconfig (hash-table-ref/default testconfig "disks" #f))
		      (hash-table-ref/default confdat "disks" #f)))
	 (minspace (let ((m (configf:lookup confdat "setup" "minspace")))
		     (string->number (or m "10000")))))
    (if disks 
	(let ((res (common:get-disk-with-most-free-space disks minspace))) ;; min size of 1000, seems tad dumb
	  (if res
	      (cdr res)
	      (begin
;;		(if (common:low-noise-print 20 "No valid disks or no disk with enough space")
;;		    (debug:print-error 0 *default-log-port* "No valid disks found in megatest.config. Please add some to your [disks] section and ensure the directory exists and has enough space!\n    You can change minspace in the [setup] section of megatest.config. Current setting is: " minspace))
		;;(exit 1)
                 (if (null? disks)
                     (cons 1 (conc *toppath* "/runs"))
                     (let ((paths (sort disks (lambda (x y) (> (string-length (cadr x)) (string-length (cadr y)))))))
                       (let loop ((head (car paths)) (tail (cdr paths)))
                         (let ((result (handle-exceptions exn #f (create-directory (cadr head) #t))))
                           (if result
                               result
                               (if (null? tail)
                                   (cons 1 (conc *toppath* "/runs"))
                                   (loop (car tail) (cdr tail)))))))))))))) ;; the code creates the necessary directories if it does not exist and returns the path.


(define (launch:test-copy test-src-path test-path)
  (let* ((ovrcmd (let ((cmd (configf:lookup *configdat* "setup" "testcopycmd")))
		   (if cmd
		       ;; substitute the TEST_SRC_PATH and TEST_TARG_PATH
		       (string-substitute "TEST_TARG_PATH" test-path
					  (string-substitute "TEST_SRC_PATH" test-src-path cmd #t) #t)
		       #f)))
	 (cmd    (if ovrcmd 
		     ovrcmd
		     (conc "rsync -av" (if (debug:debug-mode 1) "" "q") " " test-src-path "/ " test-path "/"
			   " >> " test-path "/mt_launch.log 2>> " test-path "/mt_launch.log")))
	 (status (system cmd)))
    (if (not (eq? status 0))
	(debug:print 2 *default-log-port* "ERROR: problem with running \"" cmd "\""))))


;; Desired directory structure:
;;
;;  <linkdir> - <target> - <testname> -.
;;                                     |
;;                                     v
;;  <rundir>  -  <target>  -    <testname> -|- <itempath(s)>
;;
;;  dir stored in test is:
;; 
;;  <linkdir> - <target> - <testname> [ - <itempath> ]
;; 
;; All log file links should be stored relative to the top of link path
;;  
;; <target> - <testname> [ - <itempath> ] 
;;
(define (create-work-area run-id run-info keyvals test-id test-src-path disk-path testname itemdat #!key (remtries 2))
  (let* ((item-path (if (string? itemdat) itemdat (item-list->path itemdat))) ;; if pass in string - just use it
	 (runname   (if (string? run-info) ;; if we pass in a string as run-info use it as run-name.
			run-info
			(db:get-value-by-header (db:get-rows run-info)
						(db:get-header run-info)
						"runname")))
	 (contour   #f) ;; NOT READY FOR THIS (args:get-arg "-contour"))
	 ;; convert back to db: from rdb: - this is always run at server end
	 (target   (string-intersperse (map cadr keyvals) "/"))

	 (not-iterated  (equal? "" item-path))

	 ;; all tests are found at <rundir>/test-base or <linkdir>/test-base
	 (testtop-base (conc target "/" runname "/" testname))
	 (test-base    (conc testtop-base (if not-iterated "" "/") item-path))

	 ;; nb// if itempath is not "" then it is prefixed with "/"
	 (toptest-path (conc disk-path (if contour (conc "/" contour) "") "/" testtop-base))
	 (test-path    (conc disk-path (if contour (conc "/" contour) "") "/" test-base))

	 ;; ensure this exists first as links to subtests must be created there
	 (linktree  (common:get-linktree))
	 ;; WAS: (let ((rd (configf:lookup *configdat* "setup" "linktree")))
	 ;;         (if rd rd (conc *toppath* "/runs"))))
	 ;; which seems wrong ...

	 (lnkbase   (conc linktree (if contour (conc "/" contour) "") "/" 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 '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 (common: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... try up to three times
    (let loop ((done 3)) 
      (let ((success (if (and (not (common:directory-exists? lnkbase))
			      (not (common:file-exists? lnkbase)))
			 (handle-exceptions
			  exn
			  (begin
			    (debug:print-error 0 *default-log-port* "Problem creating linktree base at " lnkbase)
			    (print-error-message exn (current-error-port))
			    #t)
			  (create-directory lnkbase #t)
			  #f))))
	(if (and (not success)(> done 0))
	    (loop (- done 1)))))
    
    ;; update the toptest record with its location rundir, cache the path
    ;; This wass highly inefficient, one db write for every subtest, potentially
    ;; thousands of unnecessary updates, cache the fact it was set and don't set it 
    ;; again. 

    ;; Now create the link from the test path to the link tree, however
    ;; if the test is iterated it is necessary to create the parent path
    ;; to the iteration. use pathname-directory to trim the path by one
    ;; level
    (if (not not-iterated) ;; i.e. iterated
	(let ((iterated-parent  (pathname-directory (conc lnkpath "/" item-path))))
	  (debug:print-info 2 *default-log-port* "Creating iterated parent " iterated-parent)
	  (handle-exceptions
	   exn
	   (begin
	     (debug:print-error 0 *default-log-port* " Failed to create directory " iterated-parent ((condition-property-accessor 'exn 'message) exn) ", exiting")
	     (exit 1))
	   (create-directory iterated-parent #t))))

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

    (if (not (or (common:file-exists? lnkpath)
		 (symbolic-link? lnkpath)))
	(handle-exceptions
	 exn
	 (begin
	   (debug:print-error 0 *default-log-port* " Failed to create symlink " lnkpath ((condition-property-accessor 'exn 'message) exn) ", exiting")
	   (exit 1))
	 (create-symbolic-link toptest-path lnkpath)))
    
    ;; 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 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 'getstr 
				   (db:test-get-rundir testinfo) ;; ) ;; )
				   #f)))
	  (hash-table-set! *toptest-paths* testname curr-test-path)
	  ;; NB// Was this for the test or for the parent in an iterated test?
	  (rmt:general-call 'test-set-rundir-shortdir run-id lnkpath 
			    (if (common:file-exists? lnkpath)
				;; (resolve-pathname lnkpath)
				(common:nice-path lnkpath)
				lnkpath)
			    testname "" run-id)
	  ;; (rmt:general-call 'test-set-rundir run-id lnkpath testname "") ;; toptest-path)
	  (if (or (not curr-test-path)
		  (not (directory-exists? toptest-path)))
	      (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.
		 (create-directory toptest-path #t))
		(hash-table-set! *toptest-paths* testname toptest-path)))))

    ;; The toptest path has been created, the link to the test in the linktree has
    ;; been created. Now, if this is an iterated test the real test dir must be created
    (if (not not-iterated) ;; this is an iterated test
	(begin ;; (let ((lnktarget (conc lnkpath "/" item-path)))
	  (debug:print 2 *default-log-port* "Setting up sub test run area")
	  (debug:print 2 *default-log-port* " - creating run area in " test-path)
	  (handle-exceptions
	   exn
	   (begin
	     (debug:print-error 0 *default-log-port* " Failed to create directory " test-path ((condition-property-accessor 'exn 'message) exn) ", exiting")
	     (exit 1))
	   (create-directory test-path #t))
	  (debug:print 2 *default-log-port* 
		       " - creating link from: " test-path "\n"
		       "                   to: " lnktarget)

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

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

    (if (and test-src-path (directory? test-path))
	(begin
	  (launch:test-copy test-src-path test-path)
	  (list lnkpathf lnkpath ))
	(if (and test-src-path (> remtries 0))
	    (begin
	      (debug:print-error 0 *default-log-port* "Failed to create work area at " test-path " with link at " lnktarget ", remaining attempts " remtries)
	      ;; 
	      (create-work-area run-id run-info keyvals test-id test-src-path disk-path testname itemdat remtries: (- remtries 1)))
	    (list #f #f)))))


(define (launch:handle-zombie-tests run-id)
  (let* ((key (conc "zombiescan-runid-"run-id))
         (now (current-seconds))
         (threshold (- (current-seconds)  (* 2 (or (configf:lookup-number *configdat* "setup" "deadtime") 120))))
         (val (rmt:get-var key))
         (do-scan?
          (cond
           ((not val)
            #t)
           ((< val threshold)
            #t)
           (else #f))))
    (when do-scan?
      (debug:print 1 *default-log-port* "INFO: search and mark zombie tests")
      (rmt:set-var key (current-seconds))
      (rmt:find-and-mark-incomplete run-id #f))))





;; 1. look though disks list for disk with most space
;; 2. create run dir on disk, path name is meaningful
;; 3. create link from run dir to megatest runs area 
;; 4. remotely run the test on allocated host
;;    - could be ssh to host from hosts table (update regularly with load)
;;    - could be netbatch
;;      (launch-test db (cadr status) test-conf))
(define (launch-test test-id run-id run-info keyvals runname test-conf test-name test-path itemdat params)
  (mutex-lock! *launch-setup-mutex*) ;; setting variables and processing the testconfig is NOT thread-safe, reuse the launch-setup mutex
  (let* ( ;; (lock-key        (conc "test-" test-id))
	;; (got-lock        (let loop ((lock        (rmt:no-sync-get-lock lock-key))
	;; 			     (expire-time (+ (current-seconds) 15))) ;; give up on getting the lock and steal it after 15 seconds
	;; 		    (if (car lock)
	;; 			#t
	;; 			(if (> (current-seconds) expire-time)
	;; 			    (begin
	;; 			      (debug:print-info 0 *default-log-port* "Timed out waiting for a lock to launch test " keyvals " " runname " " test-name " " test-path)
	;; 			      (rmt:no-sync-del! lock-key) ;; destroy the lock
	;; 			      (loop (rmt:no-sync-get-lock lock-key) expire-time)) ;; 
	;; 			    (begin
	;; 			      (thread-sleep! 1)
	;; 			      (loop (rmt:no-sync-get-lock lock-key) expire-time))))))
	 (item-path       (item-list->path itemdat))
	 (contour         #f)) ;; NOT READY FOR THIS (args:get-arg "-contour")))
    (let loop ((delta        (- (current-seconds) *last-launch*))
	       (launch-delay (configf:lookup-number *configdat* "setup" "launch-delay" default: 1)))
      (if (> launch-delay delta)
	  (begin
	    (if (common:low-noise-print 1200 "test launch delay") ;; every two hours or so remind the user about launch delay.
		(debug:print-info 0 *default-log-port* "NOTE: test launches are delayed by " launch-delay " seconds. See megatest.config launch-delay setting to adjust.")) ;; launch of " test-name " for " (- launch-delay delta) " seconds"))
	    (thread-sleep! (- launch-delay delta))
	    (loop (- (current-seconds) *last-launch*) launch-delay))))
    (change-directory *toppath*)
    (alist->env-vars ;; consolidate this code with the code in megatest.scm for "-execute", *maybe* - the longer they are set the longer each launch takes (must be non-overlapping with the vars)
     (append
      (list
       (list "MT_RUN_AREA_HOME" *toppath*)
       (list "MT_TEST_NAME" test-name)
       (list "MT_RUNNAME"   runname)
       (list "MT_ITEMPATH"  item-path)
       (list "MT_CONTOUR"   contour)
       )
      itemdat))
    (let* ((tregistry       (tests:get-all)) ;; third param (below) is system-allowed
           ;; for tconfig, why do we allow fallback to test-conf?
	   (tconfig         (or (tests:get-testconfig test-name item-path tregistry #t force-create: #t)
				(begin
                                  (debug:print 0 *default-log-port* "WARNING: falling back to pre-calculated testconfig. This is likely not desired.")
                                  test-conf))) ;; force re-read now that all vars are set
	   (useshell        (let ((ush (configf:lookup *configdat* "jobtools"     "useshell")))
			      (if ush 
				  (if (equal? ush "no") ;; must use "no" to NOT use shell
				      #f
				      ush)
				  #t)))     ;; default is yes
	   (runscript       (configf:lookup tconfig   "setup"        "runscript"))
	   (ezsteps         (> (length (hash-table-ref/default tconfig "ezsteps" '())) 0)) ;; don't send all the steps, could be big, just send a flag
	   (subrun          (> (length (hash-table-ref/default tconfig "subrun"  '())) 0)) ;; send a flag to process a subrun
	   ;; (diskspace       (configf:lookup tconfig   "requirements" "diskspace"))
	   ;; (memory          (configf:lookup tconfig   "requirements" "memory"))
	   ;; (hosts           (configf:lookup *configdat* "jobtools"     "workhosts")) ;; I'm pretty sure this was never completed
	   (remote-megatest (configf:lookup *configdat* "setup" "executable"))
	   (run-time-limit  (or (configf:lookup  tconfig   "requirements" "runtimelim")
				(configf:lookup  *configdat* "setup" "runtimelim")))
	   ;; FIXME SOMEDAY: not good how this is so obtuse, this hack is to 
	   ;;                allow running from dashboard. Extract the path
	   ;;                from the called megatest and convert dashboard
	   ;;             	  or dboard to megatest
	   (local-megatest  (let* ((lm  (car (argv)))
				   (dir (pathname-directory lm))
				   (exe (pathname-strip-directory lm)))
			      (conc (if dir (conc dir "/") "")
				    (case (string->symbol exe)
				      ((dboard)    "../megatest")
				      ((mtest)     "../megatest")
				      ((dashboard) "megatest")
				      (else exe)))))
	   (launcher        (common:get-launcher *configdat* test-name item-path)) ;; (configf:lookup *configdat* "jobtools"     "launcher"))
	   (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 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 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! run-id test-id "LAUNCHED" "n/a" #f #f) ;; (if launch-results launch-results "FAILED"))
      (rmt:set-state-status-and-roll-up-items 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))
	  (begin
	    (set! work-area (conc test-path "/tmp_run"))
	    (create-directory work-area #t)
	    (debug:print 0 *default-log-port* "WARNING: No disk work area specified - running in the test directory under tmp_run")))
      (set! cmdparms (base64:base64-encode 
		      (z3:encode-buffer 
		       (with-output-to-string
			 (lambda () ;; (list 'hosts     hosts)
			   (write (list (list 'testpath  test-path)
					;; (list 'transport (conc *transport-type*))
					;; (list 'serverinf *server-info*)
					(list 'homehost  (let* ((hhdat (common:get-homehost)))
							   (if hhdat
							       (car hhdat)
							       #f)))
					(list 'serverurl (if *runremote*
							     (remote-server-url *runremote*)
							     #f)) ;;
					(list 'areaname  (common:get-testsuite-name))
					(list 'toppath   *toppath*)
					(list 'work-area work-area)
					(list 'test-name test-name) 
					(list 'runscript runscript) 
					(list 'run-id    run-id   )
					(list 'test-id   test-id  )
					;; (list 'item-path item-path )
					(list 'itemdat   itemdat  )
					(list 'megatest  remote-megatest)
					(list 'ezsteps   ezsteps)
					(list 'subrun    subrun)
					(list 'target    mt_target)
					(list 'contour   contour)
					(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 run-id test-id)
      ;; if the dir does not exist we may have a itempath where individual variables are a path, launch anyway
      (if (common:file-exists? work-area)
	  (change-directory work-area)) ;; so that log files from the launch process don't clutter the test dir
      (cond
       ;; ((and launcher hosts) ;; must be using ssh hostname
       ;;    (set! fullcmd (append launcher (car hosts)(list remote-megatest "-m" test-sig "-execute" cmdparms) debug-param)))
       ;; (set! fullcmd (append launcher (car hosts)(list remote-megatest test-sig "-execute" cmdparms))))
       (launcher
	(set! fullcmd (append launcher (list remote-megatest "-m" test-sig "-execute" cmdparms) debug-param)))
       ;; (set! fullcmd (append launcher (list remote-megatest test-sig "-execute" cmdparms))))
       (else
	(if (not useshell)(debug:print 0 *default-log-port* "WARNING: internal launching will not work well without \"useshell yes\" in your [jobtools] section"))
	(set! fullcmd (append (list remote-megatest "-m" test-sig "-execute" cmdparms) debug-param (list (if useshell "&" ""))))))
      ;; (set! fullcmd (list remote-megatest test-sig "-execute" cmdparms (if useshell "&" "")))))
      (if (args:get-arg "-xterm")(set! fullcmd (append fullcmd (list "-xterm"))))
      (debug:print 1 *default-log-port* "Launching " work-area)
      ;; set pre-launch-env-vars before launching, keep the vars in prevvals and put the envionment back when done
      (debug:print 4 *default-log-port* "fullcmd: " fullcmd)
      (set! *last-launch* (current-seconds)) ;; all that junk above takes time, set this as late as possible.
      (let* ((commonprevvals (alist->env-vars
			      (hash-table-ref/default *configdat* "env-override" '())))
	     (miscprevvals   (alist->env-vars ;; consolidate this code with the code in megatest.scm for "-execute"
			      (append (list (list "MT_TEST_RUN_DIR" work-area)
					    (list "MT_TEST_NAME" test-name)
					    (list "MT_ITEM_INFO" (conc itemdat)) 
					    (list "MT_RUNNAME"   runname)
					    (list "MT_TARGET"    mt_target)
					    (list "MT_ITEMPATH"  item-path)
					    )
				      itemdat)))
	     (testprevvals   (alist->env-vars
			      (hash-table-ref/default tconfig "pre-launch-env-overrides" '())))
	     ;; Launchwait defaults to true, must override it to turn off wait
	     (launchwait     (if (equal? (configf:lookup *configdat* "setup" "launchwait") "no") #f #t))
	     (launch-results-prev (apply (if launchwait ;; BB: TODO: refactor this to examine return code of launcher, if nonzero, set state to launch failed.
					process:cmd-run-with-stderr-and-exitcode->list
					process-run)
				    (if useshell
					(let ((cmdstr (string-intersperse fullcmd " ")))
					  (if launchwait
					      cmdstr
					      (conc cmdstr " >> mt_launch.log 2>&1 &")))
					(car fullcmd))
				    (if useshell
					'()
					(cdr fullcmd))))
             (success        (if launchwait (equal? 0 (cadr launch-results-prev)) #t))
             (launch-results (if launchwait (car launch-results-prev) launch-results-prev)))
        (if (not success)
            (tests:test-set-status! run-id test-id "COMPLETED" "DEAD" "launcher failed; exited non-zero; check mt_launch.log" #f)) ;; (if launch-results launch-results "FAILED"))
        (mutex-unlock! *launch-setup-mutex*) ;; yes, really should mutex all the way to here. Need to put this entire process into a fork.
	;; (rmt:no-sync-del! lock-key)         ;; release the lock for starting this test
	(if (not launchwait) ;; give the OS a little time to allow the process to start
	    (thread-sleep! 0.01))
	(with-output-to-file "mt_launch.log"
	  (lambda ()
	    (print "LAUNCHCMD: " (string-intersperse fullcmd " "))
	    (if (list? launch-results)
		(apply print launch-results)
		(print "NOTE: launched \"" fullcmd "\"\n  but did not wait for it to proceed. Add the following to megatest.config \n[setup]\nlaunchwait yes\n  if you have problems with this"))
	    #:append))
	(debug:print 2 *default-log-port* "Launching completed, updating db")
	(debug:print 2 *default-log-port* "Launch results: " launch-results)
	(if (not launch-results)
	    (begin
	      (print "ERROR: Failed to run " (string-intersperse fullcmd " ") ", exiting now")
	      ;; (sqlite3:finalize! db)
	      ;; good ole "exit" seems not to work
	      ;; (_exit 9)
	      ;; but this hack will work! Thanks go to Alan Post of the Chicken email list
	      ;; NB// Is this still needed? Should be safe to go back to "exit" now?
	      (process-signal (current-process-id) signal/kill)
	      ))
	(alist->env-vars miscprevvals)
	(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 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-pid 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)))


;; Do not rpc this one, do the underlying calls!!!
(define (tests:test-set-status! 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 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 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)
			     (if (and (equal? prev-state  "COMPLETED")
				      (equal? prev-status "WAIVED"))
				 (if comment
				     comment
				     prev-comment) ;; waived is either the comment or #f
				 #f))
			   #f)
		       #f)))
    (if (and waived 
	     (tests:check-waiver-eligibility testdat prev-test))
	(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 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 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)))
    ;; 
    ;; ;; :first_warn
    ;; (let ((val (hash-table-ref/default otherdat ":first_warn" #f)))
    ;;   (if val
    ;;       (sqlite3:execute db "UPDATE tests SET first_warn=? WHERE run_id=? AND testname=? AND item_path=?;" val run-id test-name item-path)))

    (let ((category (hash-table-ref/default otherdat ":category" ""))
	  (variable (hash-table-ref/default otherdat ":variable" ""))
	  (value    (hash-table-ref/default otherdat ":value"    #f))
	  (expected (hash-table-ref/default otherdat ":expected" "n/a"))
	  (tol      (hash-table-ref/default otherdat ":tol"      "n/a"))
	  (units    (hash-table-ref/default otherdat ":units"    ""))
	  (type     (hash-table-ref/default otherdat ":type"     ""))
	  (dcomment (hash-table-ref/default otherdat ":comment"  "")))
      (debug:print 4 *default-log-port* 
		   "category: " category ", variable: " variable ", value: " value
		   ", expected: " expected ", tol: " tol ", units: " units)
      (if (and value) ;; require only value; BB was- all three required
	  (let ((dat (conc category ","
			   variable ","
			   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 run-id test-id
				dat)
	    ;; This was added in check-in a5adfa3f9a. Message was: "...added delay in set-values to allow for delayed write on server start"
	    ;; I'm inserting an arbitrary rmt: call to force/ensure that the server is available to (hopefully) prevent a communication issue.
	    (rmt:get-var "MEGATEST_VERSION") ;; this does NOTHING but ensure the server is reachable. This is almost certainly NOT needed :)
            ;; BB - commentiong out arbitrary 10 second wait (thread-sleep! 10) ;; add 10 second delay before quit incase rmt needs time to start a server.
            )))
      
    ;; 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 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 'set-test-comment run-id cmt test-id)))))



)