Megatest

	 ;;
	 (if (and (not toplevel/children)  ;; special handling needed for toplevel with children
		  prev-test-physical-path
		  (file-exists? prev-test-physical-path)) ;; what to do? abort or clean up or link it in?
	     (let* ((base (pathname-directory prev-test-physical-path))
		    (dirn (pathname-file      prev-test-physical-path))
		    (newn (conc base "/." dirn)))
	       (debug:print-error 0 *default-log-port* "the old directory " prev-test-physical-path ", still exists! Moving it to " newn)
	       (rename-file prev-test-physical-path newn)))

	 (if (and archive-path ;; no point in proceeding if there is no actual archive
		  (not toplevel/children))
	     (begin
	       ;; CREATE WORK AREA
	       ;; test-src-path == #f     ==> don't copy in data from tests directory

	       ;; bup -d /tmp/matt/adisk1/2015_q1/fullrun_e1a40/ restore -C /tmp/seeme fullrun-30/latest/ubuntu/nfs/none/w02.1.20.54_b/

	       ;; DO BUP RESTORE
	       (let* ((new-test-dat        (rmt:get-test-info-by-id run-id test-id))
		      (new-test-path       (if (vector? new-test-dat )
					       (db:test-get-rundir new-test-dat)
					       (begin
						 (debug:print-error 0 *default-log-port* "unable to get data for run-id=" run-id ", test-id=" test-id)
						 (exit 1))))
		      ;; new-test-path won't work - must use best-disk instead? Nope, new-test-path but tack on /..
		      (bup-restore-params  (list "-d" archive-path "restore" "-C" (conc new-test-path "/..") archive-internal-path)))
		 (debug:print-info 0 *default-log-port* "Restoring archived data to " new-test-physical-path " from archive in " archive-path " ... " archive-internal-path)
		 ;; (mutex-lock! bup-mutex)
		 (run-n-wait bup-exe params: bup-restore-params print-cmd: #f)
		 ;; (mutex-unlock! bup-mutex)
		 (mt:test-set-state-status-by-id run-id test-id "COMPLETED" #f #f)))
	     (debug:print-error 0 *default-log-port* "No archive path in the record for run-id=" run-id " test-id=" test-id))))
     (filter vector? tests))))
	 

;;     ((http) (rmt:login-no-auto-client-setup server-info run-id))
;;     (else   (rpc:login-no-auto-client-setup server-info run-id))))
;; 
;; (define (client:setup-rpc run-id)
;;   (debug:print 0 *default-log-port* "INFO: client:setup remaining-tries=" remaining-tries)
;;   (if (<= remaining-tries 0)
;;       (begin
;; 	(debug:print-error 0 *default-log-port* "failed to start or connect to server for run-id " run-id)
;; 	(exit 1))
;;       (let ((host-info (hash-table-ref/default *runremote* run-id #f)))
;; 	(debug:print-info 0 *default-log-port* "client:setup host-info=" host-info ", remaining-tries=" remaining-tries)
;; 	(if host-info
;; 	    (let* ((iface     (car  host-info))
;; 		   (port      (cadr host-info))
;; 		   (start-res (client:connect iface port))

;; lookup_server, need to remove *runremote* stuff
;;
(define (client:setup-http run-id #!key (remaining-tries 10) (failed-connects 0))
  (debug:print-info 2 *default-log-port* "client:setup remaining-tries=" remaining-tries)
  (let* ((tdbdat (tasks:open-db)))
    (if (<= remaining-tries 0)
	(begin
	  (debug:print-error 0 *default-log-port* "failed to start or connect to server for run-id " run-id)
	  (exit 1))
	(let* ((server-dat (tasks:get-server (db:delay-if-busy tdbdat) run-id)))
	  (debug:print-info 4 *default-log-port* "client:setup server-dat=" server-dat ", remaining-tries=" remaining-tries)
	  (if server-dat
	      (let* ((iface     (tasks:hostinfo-get-interface server-dat))
		     (hostname  (tasks:hostinfo-get-hostname  server-dat))
		     (port      (tasks:hostinfo-get-port      server-dat))

;;   (handle-exceptions
;;    exn
;;    (debug:print 0 *default-log-port* " ... exiting ...")
;;    (let ((th1 (make-thread (lambda ()
;; 			     "") ;; do nothing for now (was flush out last call if applicable)
;; 			   "eat response"))
;; 	 (th2 (make-thread (lambda ()
;; 			     (debug:print-error 0 *default-log-port* "Received ^C, attempting clean exit. Please be patient and wait a few seconds before hitting ^C again.")
;; 			     (thread-sleep! 1) ;; give the flush one second to do it's stuff
;; 			     (debug:print 0 *default-log-port* "       Done.")
;; 			     (exit 4))
;; 			   "exit on ^C timer")))
;;      (thread-start! th2)
;;      (thread-start! th1)
;;      (thread-join! th2))))
;; 
;; ;; client:launch
;; ;; Need to set the signal handler somewhere other than here as this
;; ;; routine will go away.
;; ;;
;; (define (client:launch run-id)
;;   (set-signal-handler! signal/int  client:signal-handler)
;;   (set-signal-handler! signal/term client:signal-handler)
;;   (if (client:setup run-id)
;;       (debug:print-info 2 *default-log-port* "connected as client")
;;       (begin
;; 	(debug:print-error 0 *default-log-port* "Failed to connect as client")
;; 	(exit))))
;; 

;;       (old-exit code)))

(define getenv get-environment-variable)
(define (safe-setenv key val)
  (if (and (string? val)(string? key))
      (handle-exceptions
       exn
       (debug:print-error 0 *default-log-port* "bad value for setenv, key=" key ", value=" val)
       (setenv key val))
      (debug:print-error 0 *default-log-port* "bad value for setenv, key=" key ", value=" val)))

(define home (getenv "HOME"))
(define user (getenv "USER"))

;; GLOBAL GLETCHES
(define *db-keys* #f)

(define (common:read-encoded-string instr)
  (handle-exceptions
   exn
   (handle-exceptions
    exn
    (begin
      (debug:print-error 0 *default-log-port* "received bad encoded string \"" instr "\", message: " ((condition-property-accessor 'exn 'message) exn))
      (print-call-chain (current-error-port))
      #f)
    (read (open-input-string (base64:base64-decode instr))))
   (read (open-input-string (z3:decode-buffer (base64:base64-decode instr))))))

;; dot-locking egg seems not to work, using this for now
;; if lock is older than expire-time then remove it and try again

      (thread-start! th1)
      (thread-start! th2)
      (thread-join! th1))))

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

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

		      #f)))
    (if valid
	(if split
	    tlist
	    target)
	(if target
	    (begin
	      (debug:print-error 0 *default-log-port* "Invalid target, spaces or blanks not allowed \"" target "\", target should be: " (string-intersperse keys "/") ", have " tlist " for elements")
	      #f)
	    #f))))

;;======================================================================
;; M I S C   L I S T S
;;======================================================================

;; make "nice-path" available in config files and the repl
(define nice-path common:nice-path)

(define (common:read-link-f path)
  (handle-exceptions
      exn
      (begin
	(debug:print-error 0 *default-log-port* "command \"/bin/readlink -f " path "\" failed.")
	path) ;; just give up
    (with-input-from-pipe
	(conc "/bin/readlink -f " path)
      (lambda ()
	(read-line)))))

(define (get-cpu-load)

  (let* ((spacedat (common:check-db-dir-space))
	 (is-ok    (car spacedat))
	 (dbspace  (cadr spacedat))
	 (required (caddr spacedat))
	 (dbdir    (cadddr spacedat)))
    (if (not is-ok)
	(begin
	  (debug:print-error 0 *default-log-port* "Insufficient space in " dbdir ", require " required ", have " dbspace  ", exiting now.")
	  (exit 1)))))
  
;; paths is list of lists ((name path) ... )
;;
(define (common:get-disk-with-most-free-space disks minsize)
  (let ((best     #f)
	(bestsize 0))

  
(define (debug:print n e . params)
  (if (debug:debug-mode n)
      (with-output-to-port (or e (current-error-port))
	(lambda ()
	  (if *logging*
	      (db:log-event (apply conc params))
	      (apply print params)
	      )))))

(define (debug:print-error n e . params)
  ;; normal print
  (if (debug:debug-mode n)
      (with-output-to-port (or e (current-error-port))
	(lambda ()
	  (if *logging*
	      (db:log-event (apply conc params))
	      ;; (apply print "pid:" (current-process-id) " " params)
	      (apply print "ERROR: " params)
	      ))))
  ;; pass important messages to stderr
  (if (and (eq? n 0)(not (eq? e (current-error-port)))) 
      (with-output-to-port (current-error-port)
	(lambda ()
	  (apply print "ERROR: " params)
	  ))))

(define (debug:print-info n e . params)
  (if (debug:debug-mode n)
      (with-output-to-port (or e (current-error-port))
	(lambda ()
	  (if *logging*
	      (let ((res (format#format #f "INFO: (~a) ~a" n (apply conc params))))

			       (list key val metadata)
			       (list key val))))))

(define (config:eval-string-in-environment str)
  (handle-exceptions
   exn
   (begin
     (debug:print-error 0 *default-log-port* "problem evaluating \"" str "\" in the shell environment")
     #f)
   (let ((cmdres (process:cmd-run->list (conc "echo " str))))
     (if (null? cmdres) ""
	 (caar cmdres)))))

;;======================================================================
;; Make the regexp's needed globally available

										   (cmdres     (process:cmd-run->list cmd))
										   (delta      (- (current-seconds) start-time))
										   (status     (cadr cmdres))
										   (res        (car  cmdres)))
									      (debug:print-info 4 *default-log-port* "" inl "\n => " (string-intersperse res "\n"))
									      (if (not (eq? status 0))
										  (begin
										    (debug:print-error 0 *default-log-port* "problem with " inl ", return code " status
												 " output: " cmdres)))
									      (if (> delta 2)
										  (debug:print-info 0 *default-log-port* "for line \"" inl "\"\n  command: " cmd " took " delta " seconds to run with output:\n   " res)
										  (debug:print-info 9 *default-log-port* "for line \"" inl "\"\n  command: " cmd " took " delta " seconds to run with output:\n   " res))
									      (if (null? res)
										  ""
										  (string-intersperse res " "))))))

								       "")
								   val)))
						      ;; (print "val: " val "\nnewval: \"" newval "\"\nvarflag: " var-flag)
						      (hash-table-set! res curr-section-name 
								       (config:assoc-safe-add alist var-flag newval metadata: metapath))
						      (loop (configf:read-line inp res (calc-allow-system allow-system curr-section-name sections) settings) curr-section-name var-flag (if lead lead whsp)))
						    (loop (configf:read-line inp res (calc-allow-system allow-system curr-section-name sections) settings) curr-section-name #f #f))))
	       (else (debug:print-error 0 *default-log-port* "problem parsing " path ",\n   \"" inl "\"")
		     (set! var-flag #f)
		     (loop (configf:read-line inp res (calc-allow-system allow-system curr-section-name sections) settings) curr-section-name #f #f))))))))
  
;; pathenvvar will set the named var to the path of the config
(define (find-and-read-config fname #!key (environ-patt #f)(given-toppath #f)(pathenvvar #f))
  (let* ((curr-dir   (current-directory))
         (configinfo (find-config fname toppath: given-toppath))

			   (set! res (append res (list hed))))
			  ((not newval) ;; key has been removed
			   (set! new #f))
			  ((not (equal? newval val))
			     (hash-table-set! sechash key newval)
			     (set! new (conc key " " newval)))
			  (else
			   (debug:print-error 0 *default-log-port* "problem parsing line number " lnum "\"" hed "\"")))))
	   (else
	    (debug:print-error 0 *default-log-port* "Problem parsing line num " lnum " :\n   " hed )))
	  (if (not (null? tal))
	      (loop (car tal)(cdr tal)(if new (append res (list new)) res)(+ lnum 1)))
	  ;; drop to here when done processing, res contains modified list of lines
	  (set! fdat res)))

    ;; step 4: Append new values to the section
    (for-each 

		     ;; (print "obj: " obj ", id: " id ", state: " state)
		     (let* ((run-path (tree:node->path obj id))
			    (run-id   (tree-path->run-id ddata (cdr run-path))))
		       (if (number? run-id)
			   (begin
			     (d:data-curr-run-id-set! ddata run-id)
			     (dashboard:update-run-summary-tab))
			   (debug:print-error 0 *default-log-port* "tree-path->run-id returned non-number " run-id)))
		     ;; (print "path: " (tree:node->path obj id) " run-id: " run-id)
		     )))
	 (cell-lookup (make-hash-table))
	 (run-matrix (iup:matrix
		      #:expand "YES"
		      #:click-cb
		      (lambda (obj lin col status)

		     ;; (print "obj: " obj ", id: " id ", state: " state)
		     (let* ((run-path (tree:node->path obj id))
			    (run-id   (tree-path->run-id ddata (cdr run-path))))
		       (if (number? run-id)
			   (begin
			     (d:data-curr-run-id-set! ddata run-id)
			     (dashboard:update-new-view-tab))
			   (debug:print-error 0 *default-log-port* "tree-path->run-id returned non-number " run-id)))
		     ;; (print "path: " (tree:node->path obj id) " run-id: " run-id)
		     )))
	 (cell-lookup (make-hash-table))
	 (run-matrix (iup:matrix
		      #:expand "YES"
		      #:click-cb
		      (lambda (obj lin col status)

;; SQLITE3 HELPERS
;;======================================================================

(define (db:general-sqlite-error-dump exn stmt run-id params)
  (let ((err-status ((condition-property-accessor 'sqlite3 'status #f) exn)))
    ;; check for (exn sqlite3) ((condition-property-accessor 'exn 'message) exn)
    (print "err-status: " err-status)
    (debug:print-error 0 *default-log-port* " query " stmt " failed, params: " params ", error: " ((condition-property-accessor 'exn 'message) exn))
    (print-call-chain (current-error-port))))

;; convert to -inline
(define (db:first-result-default db stmt default . params)
  (handle-exceptions
   exn
   (let ((err-status ((condition-property-accessor 'sqlite3 'status #f) exn)))
     ;; check for (exn sqlite3) ((condition-property-accessor 'exn 'message) exn)
     (if (eq? err-status 'done)
	 default
	 (begin
	   (debug:print-error 0 *default-log-port* " query " stmt " failed, params: " params ", error: " ((condition-property-accessor 'exn 'message) exn))
	   (print-call-chain (current-error-port))
	   default)))
   (apply sqlite3:first-result db stmt params)))

;; Get/open a database
;;    if run-id => get run specific db
;;    if #f     => get main db

		    (db:get-db dbstruct run-id)
		    dbstruct)) ;; cheat, allow for passing in a dbdat
	 (db    (db:dbdat-get-db dbdat)))
    (db:delay-if-busy dbdat)
    (handle-exceptions
     exn
     (begin
       (debug:print-error 0 *default-log-port* "sqlite3 issue in db:with-db, dbstruct=" dbstruct ", run-id=" run-id ", proc=" proc ", params=" params " error: " ((condition-property-accessor 'exn 'message) exn))
       (print-call-chain (current-error-port)))
     (let ((res (apply proc db params)))
       (if (vector? dbstruct)(db:done-with dbstruct run-id r/w))
       res))))

;;======================================================================
;; K E E P   F I L E D B   I N   dbstruct

  (let* ((dbdir           (db:get-dbdir))
	 (fname           (if run-id
			      (if (eq? run-id 0) "main.db" (conc run-id ".db"))
			      #f)))
    (handle-exceptions
     exn
     (begin
       (debug:print-error 0 *default-log-port* "Couldn't create path to " dbdir)
       (exit 1))
     (if (not (directory? dbdir))(create-directory dbdir #t)))
    (if fname
	(conc dbdir "/" fname)
	dbdir)))

(define (db:get-dbdir)

		 (db           (db:lock-create-open dbpath ;; this is the database physically on disk
						    (lambda (db)
						      (handle-exceptions
						       exn
						       (begin
							 ;; (release-dot-lock dbpath)
							 (if (> attemptnum 2)
							     (debug:print-error 0 *default-log-port* "tried twice, cannot create/initialize db for run-id " run-id ", at path " dbpath)
							     (db:open-rundb dbstruct run-id attemptnum (+ attemptnum 1))))
						       (db:initialize-run-id-db db)
						       (sqlite3:execute 
							db
							"INSERT OR IGNORE INTO tests (id,run_id,testname,event_time,item_path,state,status) VALUES (?,?,'bogustest',strftime('%s','now'),'nowherepath','DELETED','n/a');"
							(* run-id 30000) ;; allow for up to 30k tests per run
							run-id)

(define (db:move-and-recreate-db dbdat)
  (let* ((dbpath   (db:dbdat-get-path        dbdat))
	 (dbdir    (pathname-directory       dbpath))
	 (fname    (pathname-strip-directory dbpath))
	 (fnamejnl (conc fname "-journal"))
	 (tmpname  (conc fname "." (current-process-id)))
	 (tmpjnl   (conc fnamejnl "." (current-process-id))))
    (debug:print-error 0 *default-log-port* "" fname " appears corrupted. Making backup \"old/" fname "\"")
    (system (conc "cd " dbdir ";mkdir -p old;cat " fname " > old/" tmpname))
    (system (conc "rm -f " dbpath))
    (if (file-exists? fnamejnl)
	(begin
	  (debug:print-error 0 *default-log-port* "" fnamejnl " found, moving it to old dir as " tmpjnl)
	  (system (conc "cd " dbdir ";mkdir -p old;cat " fnamejnl " > old/" tmpjnl))
	  (system (conc "rm -f " dbdir "/" fnamejnl))))
    ;; attempt to recreate database
    (system (conc "cd " dbdir ";sqlite3 old/" tmpname " .dump | sqlite3 " fname))))
    
;; return #f to indicate the dbdat should be closed/reopened
;; else return dbdat

     (debug:print 0 *default-log-port* " status:  " ((condition-property-accessor 'sqlite3 'status) exn))
     (debug:print 0 *default-log-port* " src db:  " (db:dbdat-get-path fromdb))
     (for-each (lambda (dbdat)
		 (let ((dbpath (db:dbdat-get-path dbdat)))
		   (debug:print 0 *default-log-port* " dbpath:  " dbpath)
		   (if (not (db:repair-db dbdat))
		       (begin
			 (debug:print-error 0 *default-log-port* "Failed to rebuild " dbpath ", exiting now.")
			 (exit)))))
	       (cons todb slave-dbs))
     
     0)
;;      (if *server-run* ;; we are inside a server, throw a sync-failed error
;; 	 (signal (make-composite-condition
;; 		 (make-property-condition 'sync-failed 'message "db:sync-tables failed in a server context.")))
;; 	 0)) ;; return zero for num synced

	 ;; (set! *time-to-exit* #t) ;; let watch dog know that it is time to die.
	 ;; (tasks:server-set-state! (db:delay-if-busy tdbdat) server-id "shutting-down")
	 ;; (portlogger:open-run-close portlogger:set-port port "released")
	 ;; (exit 1)))
   (cond
    ((not fromdb) (debug:print 3 *default-log-port* "WARNING: db:sync-tables called with fromdb missing") -1)
    ((not todb)   (debug:print 3 *default-log-port* "WARNING: db:sync-tables called with todb missing") -2)
    ((not (sqlite3:database? (db:dbdat-get-db fromdb)))
     (debug:print-error 0 *default-log-port* "db:sync-tables called with fromdb not a database " fromdb) -3)
    ((not (sqlite3:database? (db:dbdat-get-db todb)))
     (debug:print-error 0 *default-log-port* "db:sync-tables called with todb not a database " todb) -4)
    (else
     (let ((stmts       (make-hash-table)) ;; table-field => stmt
	   (all-stmts   '())              ;; ( ( stmt1 value1 ) ( stml2 value2 ))
	   (numrecs     (make-hash-table))
	   (start-time  (current-milliseconds))
	   (tot-count   0))
       (for-each ;; table

(define (open-run-close-no-exception-handling  proc idb . params)
  (debug:print-info 11 *default-log-port* "open-run-close-no-exception-handling START given a db=" (if idb "yes " "no ") ", params=" params)
  (if (or *db-write-access*
	  (not (member proc *db:all-write-procs*)))
      (let* ((db (cond
		  ((pair? idb)                 (db:dbdat-get-db idb))
		  ((sqlite3:database? idb)     idb)
		  ((not idb)                   (debug:print-error 0 *default-log-port* "cannot open-run-close with #f anymore"))
		  ((procedure? idb)            (idb))
		  (else   	               (debug:print-error 0 *default-log-port* "cannot open-run-close with #f anymore"))))
	     (res #f))
	(set! res (apply proc db params))
	(if (not idb)(sqlite3:finalize! dbstruct))
	(debug:print-info 11 *default-log-port* "open-run-close-no-exception-handling END" )
	res)
      #f))

					;(debug:print 4 *default-log-port* "qry: " qry) 
		   qry)
		 qryvals)
	  (db:delay-if-busy dbdat)
	  (sqlite3:execute db "UPDATE runs SET state=?,status=?,event_time=strftime('%s','now') WHERE id=? AND state='deleted';" state status res)
	  res) 
	(begin
	  (debug:print-error 0 *default-log-port* "Called without all necessary keys")
	  #f))))

;; replace header and keystr with a call to runs:get-std-run-fields
;;
;; keypatts: ( (KEY1 "abc%def")(KEY2 "%") )
;; runpatts: patt1,patt2 ...
;;

		(let* ((key    (car keyval))
		       (patt   (cadr keyval))
		       (fulkey (conc ":" key))
		       (wildtype (if (substring-index "%" patt) "like" "glob")))
		  (if patt
		      (set! key-patt (conc key-patt " AND " key " " wildtype " '" patt "'"))
		      (begin
			(debug:print-error 0 *default-log-port* "searching for runs with no pattern set for " fulkey)
			(exit 6)))))
	      keyvals)
    (set! qry-str (conc "SELECT " keystr " FROM runs WHERE state != 'deleted' AND runname " runwildtype " ? " key-patt " ORDER BY event_time "
			(if limit  (conc " LIMIT " limit)   "")
			(if offset (conc " OFFSET " offset) "")
			";"))
    (debug:print-info 4 *default-log-port* "runs:get-runs-by-patt qry=" qry-str " " runnamepatt)

	 (with-input-from-string 
	     (z3:decode-buffer
	      (base64:base64-decode
	       (string-substitute 
		(regexp "_") "=" msg #t)))
	   (lambda ()(deserialize)))
	 (begin
	   (debug:print-error 0 *default-log-port* "reception failed. Received " msg " but cannot translate it.")
	   msg))) ;; crude reply for when things go awry
    ((zmq nmsg)(with-input-from-string msg (lambda ()(deserialize))))
    (else msg)))

(define (db:test-set-status-state dbstruct run-id test-id status state msg)
  (let ((dbdat  (db:get-db dbstruct run-id)))
    (if (member state '("LAUNCHED" "REMOTEHOSTSTART"))

		  (hash-table->alist removed))))
       (if c
	   (begin
	     (print "# Changed vars")
	     (map (lambda (dat)(print (car dat) " " (cdr dat)))
		  (hash-table->alist changed)))))
      (else
       (debug:print-error 0 *default-log-port* "No dumpmode specified, use -dumpmode [bash|csh|config]")))))

  (set! *http-requests-in-progress* (- *http-requests-in-progress* 1))
  (let loop ((etime (+ (current-seconds) 5))) ;; give up in five seconds
    (if (> *http-requests-in-progress* 0)
	(if (> etime (current-seconds))
	    (begin
	      (thread-sleep! 0.05)
	      (loop etime))
	    (debug:print-error 0 *default-log-port* "requests still in progress after 5 seconds of waiting. I'm going to pass on cleaning up http connections"))
	(close-all-connections!)))
  (set! *http-connections-next-cleanup* (+ (current-seconds) 10))
  (mutex-unlock! *http-mutex*))

(define (http-transport:inc-requests-and-prep-to-close-all-connections)
  (mutex-lock! *http-mutex*)
  (set! *http-requests-in-progress* (+ 1 *http-requests-in-progress*)))

	 (thread-join! th1)
	 (thread-terminate! th2)
	 (debug:print-info 11 *default-log-port* "got res=" res)
	 (if (vector? res)
	     (if (vector-ref res 0)
		 res
		 (begin ;; note: this code also called in nmsg-transport - consider consolidating it
		   (debug:print-error 0 *default-log-port* "error occured at server, info=" (vector-ref res 2))
		   (debug:print 0 *default-log-port* " client call chain:")
		   (print-call-chain (current-error-port))
		   (debug:print 0 *default-log-port* " server call chain:")
		   (pp (vector-ref res 1) (current-error-port))
		   (signal (vector-ref result 0))))
	     (signal (make-composite-condition
		      (make-property-condition 

      #f))

(define (http-transport:server-dat-update-last-access vec)
  (if (vector? vec)
      (vector-set! vec 5 (current-seconds))
      (begin
	(print-call-chain (current-error-port))
	(debug:print-error 0 *default-log-port* "call to http-transport:server-dat-update-last-access with non-vector!!"))))

;;
;; connect
;;
(define (http-transport:client-connect iface port)
  (let* ((api-url      (conc "http://" iface ":" port "/api"))
	 (api-uri      (uri-reference (conc "http://" iface ":" port "/api")))

				   (> (- (current-seconds) start-time) 2))
			      sdat
                              (begin
				(debug:print-info 0 *default-log-port* "Still waiting, last-sdat=" last-sdat)
                                (sleep 4)
				(if (> (- (current-seconds) start-time) 120) ;; been waiting for two minutes
				    (begin
				      (debug:print-error 0 *default-log-port* "transport appears to have died, exiting server " server-id " for run " run-id)
				      (tasks:server-delete-record (db:delay-if-busy tdbdat) server-id "failed to start, never received server alive signature")
				      (exit))
				    (loop start-time
					  (equal? sdat last-sdat)
					  sdat)))))))
         (iface       (car server-info))
         (port        (cadr server-info))

	     ((sync-failed)(cond
			    ((> bad-sync-count 10) ;; time to give up
			     (http-transport:server-shutdown server-id port))
			    (else ;; (> bad-sync-count 0)  ;; we've had a fail or two, delay and loop
			     (thread-sleep! 5)
			     (loop count server-state (+ bad-sync-count 1)))))
	     ((exn)
	      (debug:print-error 0 *default-log-port* "error from sync code other than 'sync-failed. Attempting to gracefully shutdown the server")
	      (tasks:server-delete-record (db:delay-if-busy tdbdat) server-id " http-transport:keep-running crashed")
	      (exit)))
	    (set! sync-time  (- (current-milliseconds) start-time))
	    (set! rem-time (quotient (- 4000 sync-time) 1000))
	    (debug:print 4 *default-log-port* "SYNC: time= " sync-time ", rem-time=" rem-time)
	    
	    (if (and (<= rem-time 4)

  (handle-exceptions
   exn
   (debug:print 0 *default-log-port* " ... exiting ...")
   (let ((th1 (make-thread (lambda ()
			     (thread-sleep! 1))
			   "eat response"))
	 (th2 (make-thread (lambda ()
			     (debug:print-error 0 *default-log-port* "Received ^C, attempting clean exit. Please be patient and wait a few seconds before hitting ^C again.")
			     (thread-sleep! 3) ;; give the flush three seconds to do it's stuff
			     (debug:print 0 *default-log-port* "       Done.")
			     (exit 4))
			   "exit on ^C timer")))
     (thread-start! th2)
     (thread-start! th1)
     (thread-join! th2))))

  (if (and itemsdat (not (null? itemsdat)))
      (let ((itemlst (filter (lambda (x)
			       (list? x))
			     (map (lambda (x)
				    (debug:print 6 *default-log-port* "item-assoc->item-list x: " x)
				    (if (< (length x) 2)
					(begin
					  (debug:print-error 0 *default-log-port* "malformed items spec " (string-intersperse x " "))
					  (list (car x)'()))
					(let* ((name (car x))
					       (items (cadr x))
					       (ilist (list name (if (string? items)
								     (string-split items)
								     '()))))
					  (if (null? ilist)
					      (debug:print-error 0 *default-log-port* "No items specified for " name))
					  ilist)))
				  itemsdat))))
	(let ((debuglevel 5))
	  (debug:print 5 *default-log-port* "item-assoc->item-list: itemsdat => itemlst ")
	  (if (debug:debug-mode 5)
	      (begin
		(pp itemsdat)

			   item))
		     items))
    (set! itemstable (map (lambda (item)
			    (if (procedure? (cadr item))
				(list (car item)((cadr item)))  ;; evaluate the proc
				item))
			  itemstable))
    (if (and have-items  (null? items))     (debug:print-error 0 *default-log-port* "[items] section in testconfig but no entries defined"))
    (if (and have-itable (null? itemstable))(debug:print-error 0 *default-log-port* "[itemstable] section in testconfig but no entries defined"))
    (if (or (not (null? items))(not (null? itemstable)))
	(append (item-assoc->item-list items)
		(item-table->item-list itemstable))
	'(()))))

;; (pp (item-assoc->item-list itemdat))


	

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

;; given the keys (a list of vectors <key field> or a list of keys) and a target return a keyval list
;; keyval list ( (key1 val1) (key2 val2) ...)
(define (keys:target->keyval keys target)
  (let* ((targlist (string-split target "/"))

	    (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)))
	(if (not (file-exists? ".ezsteps"))(create-directory ".ezsteps"))
	;; if ezsteps was defined then we are sure to have at least one step but check anyway
	(if (not (> (length ezstepslst) 0))
	    (debug:print-error 0 *default-log-port* "ezsteps defined but ezstepslst is zero length")
	    (let loop ((ezstep (car ezstepslst))
		       (tal    (cdr ezstepslst))
		       (prevstep #f))
	      ;; check exit-info (vector-ref exit-info 1)
	      (if (launch:einf-exit-status exit-info) ;; (vector-ref exit-info 1)
		  (let ((logpro-used (launch:runstep ezstep run-id test-id exit-info m tal testconfig))
			(stepname    (car ezstep)))

				  #f
				  (process-signal pid-num signal/kill)))
			       (process:get-sub-pids pid))))
		       ;;    (debug:print-info 0 *default-log-port* "not killing process " pid " as it is not alive"))))
		       pids)
		      (tests:test-set-status! run-id test-id "KILLED"  "KILLED" (args:get-arg "-m") #f))
		    (begin
		      (debug:print-error 0 *default-log-port* "Nothing to kill, pid1=" pid1 ", pid2=" pid2)
		      (tests:test-set-status! run-id test-id "KILLED"  "FAILED TO KILL" (args:get-arg "-m") #f)
		      )))
	      (mutex-unlock! m)
	      ;; no point in sticking around. Exit now.
	      (exit)))
	(if (hash-table-ref/default misc-flags 'keep-going #f)
	    (begin

		  (debug:print 0 *default-log-port* "INFO: Not starting job yet - directory " top-path " not found")
		  (thread-sleep! 10)
		  (loop (+ count 1)))))

	  (let ((sighand (lambda (signum)
			   ;; (signal-mask! signum) ;; to mask or not? seems to cause issues in exiting
			   (if (eq? signum signal/stop)
			       (debug:print-error 0 *default-log-port* "attempt to STOP process. Exiting."))
			   (set! *time-to-exit* #t)
			   (print "Received signal " signum ", cleaning up before exit. Please wait...")
			   (let ((th1 (make-thread (lambda ()
						     (tests:test-force-state-status! run-id test-id "INCOMPLETE" "KILLED")
						     (print "Killed by signal " signum ". Exiting")
						     (thread-sleep! 1)
						     (exit 1))))

		 (test-pid  (db:test-get-process_id  test-info)))
	    (cond
	     ((member (db:test-get-state test-info) '("INCOMPLETE" "KILLED" "UNKNOWN" "KILLREQ" "STUCK")) ;; prior run of this test didn't complete, go ahead and try to rerun
	      (debug:print 0 *default-log-port* "INFO: test is INCOMPLETE or KILLED, treat this execute call as a rerun request")
	      (tests:test-force-state-status! run-id test-id "REMOTEHOSTSTART" "n/a")) ;; prime it for running
	     ((member (db:test-get-state test-info) '("RUNNING" "REMOTEHOSTSTART"))
	      (if (process:alive-on-host? test-host test-pid)
		  (debug:print-error 0 *default-log-port* "test state is "  (db:test-get-state test-info) " and process " test-pid " is still running on host " test-host ", cannot proceed")
		  (tests:test-force-state-status! run-id test-id "REMOTEHOSTSTART" "n/a")))
	     ((not (member (db:test-get-state test-info) '("REMOVING" "REMOTEHOSTSTART" "RUNNING" "KILLREQ")))
	      (tests:test-force-state-status! run-id test-id "REMOTEHOSTSTART" "n/a"))
	     (else ;; (member (db:test-get-state test-info) '("REMOVING" "REMOTEHOSTSTART" "RUNNING" "KILLREQ"))
	      (debug:print-error 0 *default-log-port* "test state is " (db:test-get-state test-info) ", cannot proceed")
	      (exit))))
	  
	  (debug:print 2 *default-log-port* "Exectuing " test-name " (id: " test-id ") on " (get-host-name))
	  (set! keys       (rmt:get-keys))
	  ;; (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: #t))

	    (for-each (lambda (section)
			(for-each (lambda (varval)
				    (let ((var (car varval))
					  (val (cadr varval)))
				      (if (and (string? var)(string? val))
					  (begin
					    (setenv var (config: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)))

	  ;; NFS might not have propagated the directory meta data to the run host - give it time if needed
	  (let loop ((count 0))
	    (if (or (file-exists? work-area)
		    (> count 10))

	  (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)

	    (begin
	      (set! *configdat*  (car first-pass))
	      (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     (or (getenv "MT_LINKTREE")
				       (if *configdat* (configf:lookup *configdat* "setup" "linktree") #f)))

						"/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))))))
    ;; additional house keeping
    (let* ((linktree (or (getenv "MT_LINKTREE")
			 (if *configdat* (configf:lookup *configdat* "setup" "linktree") #f))))
      (if linktree
	  (if (not (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))))
	  (begin
	    (debug:print-error 0 *default-log-port* "linktree not defined in [setup] section of megatest.config")
	    ;; (exit 1)
	    )))
    (if (and *toppath*
	     (directory-exists? *toppath*))
	(setenv "MT_RUN_AREA_HOME" *toppath*)
	(begin
	  (debug:print-error 0 *default-log-port* "failed to find the top path to your Megatest area.")))
    *toppath*))

(define launch:setup launch:setup-new)

(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)))))))

;; Desired directory structure:
;;
;;  <linkdir> - <target> - <testname> -.
;;                                     |
;;                                     v

	  (create-directory linktree #t))) ;; (system (conc "mkdir -p " linktree))))
    ;; create the directory for the tests dir links, this is needed no matter what...
    (if (and (not (directory-exists? lnkbase))
	     (not (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)))
	 (create-directory lnkbase #t)))
    
    ;; 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 (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

    (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 (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

				   " >> " 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 "\"")))
	  (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)))))

;; 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 

   (if (> remtries 0)
       (begin
	 (debug:print 0 *default-log-port* "WARNING: exception on lock-queue:set-state. Trying again in 30 seconds.")
	 (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
	 (thread-sleep! 30)
	 (lock-queue:set-state dbdat test-id newstate remtries: (- remtries 1)))
       (begin
	 (debug:print-error 0 *default-log-port* " Failed to set lock state for test with id " test-id ", error: " ((condition-property-accessor 'exn 'message) exn) ", giving up.")
	 #f))
   (sqlite3:execute (lock-queue:db-dat-get-db dbdat) "UPDATE queue SET state=? WHERE test_id=?;"
		    newstate
		    test-id)))

(define (lock-queue:any-younger? dbdat mystart test-id #!key (remtries 10))
  ;; no need to wait on journal on read only queries
  ;; (tasks:wait-on-journal (lock-queue:db-dat-get-path dbdat) 1200)
  (handle-exceptions
   exn
   (if (> remtries 0)
       (begin
	 (debug:print 0 *default-log-port* "WARNING: exception on lock-queue:any-younger. Removing lockdb and trying again in 5 seconds.")
	 (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
	 (thread-sleep! 5)
         (lock-queue:delete-lock-db dbdat)
	 (lock-queue:any-younger? dbdat mystart test-id remtries: (- remtries 1)))
       (begin
	 (debug:print-error 0 *default-log-port* " Failed to find younger locks for test with id " test-id ", error: " ((condition-property-accessor 'exn 'message) exn) ", giving up.")
	 #f))
   (let ((res #f))
     (sqlite3:for-each-row
      (lambda (tid)
	;; Actually this should not be needed as mystart cannot be simultaneously less than and test-id same as 
	(if (not (equal? tid test-id)) 
	    (set! res tid)))

(if (and (not (null? remargs))
	 (not (or
	       (args:get-arg "-runstep")
	       (args:get-arg "-envcap")
	       (args:get-arg "-envdelta")
	       )
	      ))
    (debug:print-error 0 *default-log-port* "Unrecognised arguments: " (string-intersperse (if (list? remargs) remargs (argv))  " ")))

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

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

      (print help)
      (exit)))

(if (args:get-arg "-start-dir")
    (if (file-exists? (args:get-arg "-start-dir"))
	(change-directory (args:get-arg "-start-dir"))
	(begin
	  (debug:print-error 0 *default-log-port* "non-existant start dir " (args:get-arg "-start-dir") " specified, exiting.")
	  (exit 1))))

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

		      (for-each 
		       (lambda (f)
			 (handle-exceptions
			     exn
			     (debug:print 0 *default-log-port* "WARNING: Failed to remove file " f)
			   (delete-file f)))
		       files))))
	      (debug:print-error 0 *default-log-port* "-clean-cache requires -runname."))
	  (debug:print-error 0 *default-log-port* "-clean-cache requires -target or -reqtarg"))))
	    
	  
(if (args:get-arg "-env2file")
    (begin
      (save-environment-as-files (args:get-arg "-env2file"))
      (set! *didsomething* #t)))

		  (with-output-to-file
		      (args:get-arg "-o")
		    (lambda ()
		      (env:print added removed changed)))
		  (env:print added removed changed))
	      (env:close-database db)
	      (set! *didsomething* #t))
	    (debug:print-error 0 *default-log-port* "Parameter to -envdelta should be new=star-end")))))

;;======================================================================
;; Start the server - can be done in conjunction with -runall or -runtests (one day...)
;;   we start the server if not running else start the client thread
;;======================================================================

(if (args:get-arg "-server")

    ;; Server? Start up here.
    ;;
    (let ((tl        (launch:setup))
	  (run-id    (and (args:get-arg "-run-id")
			  (string->number (args:get-arg "-run-id")))))
      (if run-id
	  (begin
	    (server:launch run-id)
	    (set! *didsomething* #t))
	  (debug:print-error 0 *default-log-port* "server requires run-id be specified with -run-id")))

    ;; Not a server? This section will decide how to communicate
    ;;
    ;;  Setup client for all expect listed here
    (if (null? (lset-intersection 
		equal?
		(hash-table-keys args:arg-hash)

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

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

	 ((not (args:get-arg "-dumpmode"))
	  (pp (hash-table->alist data)))
	 ((string=? (args:get-arg "-dumpmode") "json")
	  (json-write data))
	 ((string=? (args:get-arg "-dumpmode") "ini")
	  (configf:config->ini data))
	 (else
	  (debug:print-error 0 *default-log-port* "-dumpmode of " (args:get-arg "-dumpmode") " not recognised")))
	(set! *didsomething* #t))
      (pop-directory)))

(if (args:get-arg "-show-config")
    (let ((tl   (launch:setup))
	  (data *configdat*)) ;; (read-config "megatest.config" #f #t)))
      (push-directory *toppath*)

       ((not (args:get-arg "-dumpmode"))
	(pp (hash-table->alist data)))
       ((string=? (args:get-arg "-dumpmode") "json")
	(json-write data))
       ((string=? (args:get-arg "-dumpmode") "ini")
	(configf:config->ini data))
       (else
	(debug:print-error 0 *default-log-port* "-dumpmode of " (args:get-arg "-dumpmode") " not recognised")))
      (set! *didsomething* #t)
      (pop-directory)))

(if (args:get-arg "-show-cmdinfo")
    (if (or (args:get-arg ":value")(getenv "MT_CMDINFO"))
	(let ((data (common:read-encoded-string (or (args:get-arg ":value")(getenv "MT_CMDINFO")))))
	  (if (equal? (args:get-arg "-dumpmode") "json")

;; since several actions can be specified on the command line the removal
;; is done first
(define (operate-on action)
  (let* ((runrec (runs:runrec-make-record))
	 (target (common:args-get-target)))
    (cond
     ((not target)
      (debug:print-error 0 *default-log-port* "Missing required parameter for " action ", you must specify -target or -reqtarg")
      (exit 1))
     ((not (or (args:get-arg ":runname")
	       (args:get-arg "-runname")))
      (debug:print-error 0 *default-log-port* "Missing required parameter for " action ", you must specify the run name pattern with -runname patt")
      (exit 2))
     ((not (args:get-arg "-testpatt"))
      (debug:print-error 0 *default-log-port* "Missing required parameter for " action ", you must specify the test pattern with -testpatt")
      (exit 3))
     (else
      (if (not (car *configinfo*))
	  (begin
	    (debug:print-error 0 *default-log-port* "Attempted " action "on test(s) but run area config file not found")
	    (exit 1))
	  ;; put test parameters into convenient variables
	  (begin
	    ;; check for correct version, exit with message if not correct
	    (common:exit-on-version-changed)
	    (runs:operate-on  action
			      target

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

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

			      (newline)))))
		       
		     (for-each 
		      (lambda (test)
		      	(handle-exceptions
			 exn
			 (begin
			   (debug:print-error 0 *default-log-port* "Bad data in test record? " test)
			   (print "exn=" (condition->list exn))
			   (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
			   (print-call-chain (current-error-port)))
			 (let* ((test-id      (if (member "id"           tests-spec)(get-value-by-fieldname test test-field-index "id"          ) #f)) ;; (db:test-get-id         test))
				(testname     (if (member "testname"     tests-spec)(get-value-by-fieldname test test-field-index "testname"    ) #f)) ;; (db:test-get-testname   test))
				(itempath     (if (member "item_path"    tests-spec)(get-value-by-fieldname test test-field-index "item_path"   ) #f)) ;; (db:test-get-item-path  test))
				(comment      (if (member "comment"      tests-spec)(get-value-by-fieldname test test-field-index "comment"     ) #f)) ;; (db:test-get-comment    test))

	       (state     (args:get-arg ":state"))
	       (status    (args:get-arg ":status"))
	       (target    (args:get-arg "-target"))
	       (toppath   (assoc/default 'toppath   cmdinfo)))
	  (change-directory toppath)
	  (if (not target)
	      (begin
		(debug:print-error 0 *default-log-port* "-target is required.")
		(exit 1)))
	  (if (not (launch:setup))
	      (begin
		(debug:print 0 *default-log-port* "Failed to setup, giving up on -test-paths or -test-files, exiting")
		(exit 1)))
	  (let* ((keys     (rmt:get-keys))
		 ;; db:test-get-paths must not be run remote

	  (let ((run-id (string->number (car params)))
		(test-id (string->number (cadr params))))
	    (if (and run-id test-id)
		(begin
		  (launch:recover-test run-id test-id)
		  (set! *didsomething* #t))
		(begin
		  (debug:print-error 0 *default-log-port* "bad run-id or test-id, must be integers")
		  (exit 1)))))))

;;======================================================================
;; Test commands (i.e. for use inside tests)
;;======================================================================

(define (megatest:step step state status logfile msg)
  (if (not (getenv "MT_CMDINFO"))
      (begin
	(debug:print-error 0 *default-log-port* "MT_CMDINFO env var not set, -step must be called *inside* a megatest invoked environment!")
	(exit 5))
      (let* ((cmdinfo   (common:read-encoded-string (getenv "MT_CMDINFO")))
	     (transport (assoc/default 'transport cmdinfo))
	     (testpath  (assoc/default 'testpath  cmdinfo))
	     (test-name (assoc/default 'test-name cmdinfo))
	     (runscript (assoc/default 'runscript cmdinfo))
	     (db-host   (assoc/default 'db-host   cmdinfo))

	      (debug:print 0 *default-log-port* "Failed to setup, exiting")
	      (exit 1)))
	(if (and state status)
	    (let ((comment (launch:load-logpro-dat run-id test-id step)))
	      ;; (rmt:test-set-log! run-id test-id (conc stepname ".html"))))
	      (rmt:teststep-set-status! run-id test-id step state status (or comment msg) logfile))
	    (begin
	      (debug:print-error 0 *default-log-port* "You must specify :state and :status with every call to -step")
	      (exit 6))))))

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

	(args:get-arg "-test-status")
	(args:get-arg "-set-values")
	(args:get-arg "-load-test-data")
	(args:get-arg "-runstep")
	(args:get-arg "-summarize-items"))
    (if (not (getenv "MT_CMDINFO"))
	(begin
	  (debug:print-error 0 *default-log-port* "MT_CMDINFO env var not set, commands -test-status, -runstep and -setlog must be called *inside* a megatest environment!")
	  (exit 5))
	(let* ((startingdir (current-directory))
	       (cmdinfo   (common:read-encoded-string (getenv "MT_CMDINFO")))
	       (transport (assoc/default 'transport cmdinfo))
	       (testpath  (assoc/default 'testpath  cmdinfo))
	       (test-name (assoc/default 'test-name cmdinfo))
	       (runscript (assoc/default 'runscript cmdinfo))

	      (tests:test-set-toplog! run-id test-name (args:get-arg "-set-toplog")))
	  (if (args:get-arg "-summarize-items")
	      ;; DO NOT run remote
	      (tests:summarize-items run-id test-id test-name #t)) ;; do force here
	  (if (args:get-arg "-runstep")
	      (if (null? remargs)
		  (begin
		    (debug:print-error 0 *default-log-port* "nothing specified to run!")
		    (if db (sqlite3:finalize! db))
		    (exit 6))
		  (let* ((stepname   (args:get-arg "-runstep"))
			 (logprofile (args:get-arg "-logpro"))
			 (logfile    (conc stepname ".log"))
			 (cmd        (if (null? remargs) #f (car remargs)))
			 (params     (if cmd (cdr remargs) '()))

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

;;  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)

		      (hash-table-set! *testconfigs* test-name newtcfg)
		      (if old-link-tree 
			  (setenv "MT_LINKTREE" old-link-tree)
			  (unsetenv "MT_LINKTREE"))
		      newtcfg))
		  (if (null? tal)
		      (begin
			(debug:print-error 0 *default-log-port* "No readable testconfig found for " test-name)
			#f)
		      (loop (car tal)(cdr tal))))))))))

			((-1) "monitor.db")
			((0) "main.db")
			(else (conc run-id ".db")))
		      #f)))
    (handle-exceptions
     exn
     (begin
       (debug:print-error 0 *default-log-port* "Couldn't create path to " dbdir)
       (exit 1))
     (if (not (directory? dbdir))(create-directory dbdir #t)))
    (if fname
	(conc dbdir "/" fname)
	dbdir)))

;; -1 => monitor.db

			     #f)))))
    (if db
	db ;; merely return the already opened db
	(let* ((dbfile (areadb:dbfile-path areadat run-id)) ;; not already opened, so open it
	       (db     (if (file-exists? dbfile)
			   (open-database dbfile)
			   (begin
			     (debug:print-error 0 *default-log-port* "I was asked to open " dbfile ", but file does not exist or is not readable.")
			     #f))))
	  (case run-id
	    ((-1)(areadat-monitordb-set! areadat db))
	    ((0) (areadat-maindb-set!    areadat db))
	    (else (rundat-db-set!        rundat  db)))
	  db))))

			       (let ((id  (list-ref row 0))
				     (dat (apply make-rundat (append row (list #f #f))))) ;; add placeholders for tests and db
				 (print row)
				 (hash-table-set! runs id dat))))
	       (sql maindb (conc "SELECT id,"
				 (string-intersperse keys "||'/'||")
				 ",runname,state,status,event_time FROM runs WHERE state != 'deleted';")))
	(debug:print-error 0 *default-log-port* "no main.db found at "  (areadb:dbfile-path areadat 0)))
    areadat))

;; given an areadat and target/runname patt fill up runs data
;;
;; ?????/

;; given a list of run-ids refresh/retrieve runs data into areadat

	(if (> retrynum 0)
	    (begin
	      (debug:print 0 *default-log-port* "WARNING: Failed to connect to server (self) on host " hostn ":" start-port ", trying again.")
	      (tasks:server-delete-record (db:delay-if-busy tdbdat) server-id "failed to start, never received server alive signature")
	      (portlogger:open-run-close portlogger:set-failed start-port)
	      (nmsg-transport:run dbstruct hostn run-id server-id))
	    (begin
	      (debug:print-error 0 *default-log-port* "could not find an open port to start server on. Giving up")
	      (exit 1))))))

(define (nmsg-transport:try-start-server dbstruct run-id portnum server-id)
  (let ((repsoc (nn-socket 'rep)))
    (nn-bind repsoc (conc "tcp://*:" portnum))
    (let loop ((msg-in (nn-recv repsoc)))
      (let* ((dat    (db:string->obj msg-in transport: 'nmsg)))

     (if success (thread-terminate! timeout)))
    ;; raise timeout error if timed out
    (if success
	(if (and (vector? result)
		 (vector-ref result 0)) ;; did it fail at the server?
	    result                ;; nope, all good
	    (begin
	      (debug:print-error 0 *default-log-port* "error occured at server, info=" (vector-ref result 2))
	      (debug:print 0 *default-log-port* " client call chain:")
	      (print-call-chain (current-error-port))
	      (debug:print 0 *default-log-port* " server call chain:")
	      (pp (vector-ref result 1) (current-error-port))
	      (signal (vector-ref result 0))))
	(signal (make-composite-condition
		 (make-property-condition 'timeout 'message "nmsg-transport:client-api-send-receive-raw timed out talking to server"))))))

   exn
   (debug:print 0 *default-log-port* " ... exiting ...")
   (let ((th1 (make-thread (lambda ()
			     (if (not *received-response*)
				 (receive-message* *runremote*))) ;; flush out last call if applicable
			   "eat response"))
	 (th2 (make-thread (lambda ()
			     (debug:print-error 0 *default-log-port* "Received ^C, attempting clean exit. Please be patient and wait a few seconds before hitting ^C again.")
			     (thread-sleep! 3) ;; give the flush three seconds to do it's stuff
			     (debug:print 0 *default-log-port* "       Done.")
			     (exit 4))
			   "exit on ^C timer")))
     (thread-start! th2)
     (thread-start! th1)
     (thread-join! th2))))

(define (portlogger:open-run-close proc . params)
  (let* ((fname  (conc "/tmp/." (current-user-name) "-portlogger.db"))
	 (avail  (tasks:wait-on-journal fname 10))) ;; wait up to about 10 seconds for the journal to go away
    (handle-exceptions
     exn
     (begin
       ;; (release-dot-lock fname)
       (debug:print-error 0 *default-log-port* "portlogger:open-run-close failed. " proc " " params)
       (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
       (debug:print 0 *default-log-port* "exn=" (condition->list exn))
       (if (file-exists? fname)(delete-file fname)) ;; brutally get rid of it
       (print-call-chain (current-error-port)))
     (let* (;; (lock   (obtain-dot-lock fname 2 9 10))
	    (db     (portlogger:open-db fname))
	    (res    (apply proc db params)))

			  (begin
			    (debug:print-info 0 *default-log-port* "Starting server as query time " delta " is over the limit of " max-query)
			    (server:kind-run run-id)))
		      ;; return the result!
		      newres)
		    )))
	    (begin
	      ;; (debug:print-error 0 *default-log-port* "Communication failed!")
	      ;; (mutex-unlock! *send-receive-mutex*)
	      ;; (exit)
	      (rmt:open-qry-close-locally cmd run-id params)
	      )))))

(define (rmt:update-db-stats run-id rawcmd params duration)
  (mutex-lock! *db-stats-mutex*)

	 (resdat         (api:execute-requests dbstruct-local (vector (symbol->string cmd) params)))
	 (success        (vector-ref resdat 0))
	 (res            (vector-ref resdat 1))
	 (duration       (- (current-milliseconds) start)))
    (if (not success)
	(if (> remretries 0)
	    (begin
	      (debug:print-error 0 *default-log-port* "local query failed. Trying again.")
	      (thread-sleep! (/ (random 5000) 1000)) ;; some random delay 
	      (rmt:open-qry-close-locally cmd run-id params remretries: (- remretries 1)))
	    (begin
	      (debug:print-error 0 *default-log-port* "too many retries in rmt:open-qry-close-locally, giving up")
	      #f))
	(begin
	  ;; (rmt:update-db-stats run-id cmd params duration)
	  ;; mark this run as dirty if this was a write
	  (if (not (member cmd api:read-only-queries))
	      (let ((start-time (current-seconds)))
		(mutex-lock! *db-multi-sync-mutex*)

		    (http-transport:client-api-send-receive run-id connection-info cmd params))))
;;		    ((commfail) (vector #f "communications fail")))))
    (if (and res (vector-ref res 0))
	(vector-ref res 1) ;;; YES!! THIS IS CORRECT!! CHANGE IT HERE, THEN CHANGE rmt:send-receive ALSO!!!
	#f)))
;; 	(db:string->obj (vector-ref dat 1))
;; 	(begin
;; 	  (debug:print-error 0 *default-log-port* "rmt:send-receive-no-auto-client-setup failed, attempting to continue. Got " dat)
;; 	  dat))))

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

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

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

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

				 (lambda ()
				   (let ((res (rmt:send-receive 'get-tests-for-run-mindata hed (list hed testpatt states status not-in))))
				     (if (list? res)
					 (begin
					   (mutex-lock! multi-run-mutex)
					   (set! result (append result res))
					   (mutex-unlock! multi-run-mutex))
					 (debug:print-error 0 *default-log-port* "get-tests-for-run-mindata failed for run-id " hed ", testpatt " testpatt ", states " states ", status " status ", not-in " not-in))))
				 (conc "multi-run-thread for run-id " hed)))
		     (newthreads (cons newthread threads)))
		(thread-start! newthread)
		(thread-sleep! 0.05) ;; give that thread some time to start
		(if (null? tal)
		    newthreads
		    (loop (car tal)(cdr tal) newthreads))))))

	 (begin
	   (print "LOGIN_FAILED")
	   (exit 1))))))

(define (rpc-transport:client-setup run-id #!key (remtries 10))
  (if *runremote*
      (begin
	(debug:print-error 0 *default-log-port* "Attempt to connect to server but already connected")
	#f)
      (let* ((host-info (hash-table-ref/default *runremote* run-id #f))) ;; (open-run-close db:get-var #f "SERVER"))
	(if host-info
	    (let ((iface    (car host-info))
		  (port     (cadr host-info))
		  (ping-res ((rpc:procedure 'server:login host port) *toppath*)))
	      (if ping-res

;; 	(if (and port
;; 		 (string->number port))
;; 	    (let ((portn (string->number port)))
;; 	      (debug:print-info 2 *default-log-port* "Setting up to connect to host " host ":" port)
;; 	      (handle-exceptions
;; 	       exn
;; 	       (begin
;; 		 (debug:print-error 0 *default-log-port* "Failed to open a connection to the server at host: " host " port: " port)
;; 		 (debug:print 0 *default-log-port* "   EXCEPTION: " ((condition-property-accessor 'exn 'message) exn))
;; 		 ;; (open-run-close 
;; 		 ;;  (lambda (db . param) 
;; 		 ;;    (sqlite3:execute db "DELETE FROM metadat WHERE var='SERVER'"))
;; 		 ;;  #f)
;; 		 (set! *runremote* #f))
;; 	       (if (and (not (args:get-arg "-server")) ;; no point in the server using the server using the server

(define (setup-env-defaults fname run-id already-seen keyvals #!key (environ-patt #f)(change-env #t))
  (let* ((keys    (map car keyvals))
	 (thekey  (if keyvals 
		      (string-intersperse (map (lambda (x)(if x x "-na-")) (map cadr keyvals)) "/")
		      (or (common:args-get-target)
			  (get-environment-variable "MT_TARGET")
			  (begin
			    (debug:print-error 0 *default-log-port* "setup-env-defaults called with no run-id or -target or -reqtarg")
			    "nothing matches this I hope"))))
	 ;; Why was system disallowed in the reading of the runconfigs file?
	 ;; NOTE: Should be setting env vars based on (target|default)
	 (confdat (read-config fname #f #t environ-patt: environ-patt sections: (list "default" thekey)))
	 (whatfound (make-hash-table))
	 (finaldat  (make-hash-table))
	 (sections (list "default" thekey)))

	 (link-tree (configf:lookup *configdat* "setup" "linktree")))
    (if testname (setenv "MT_TEST_NAME" testname))
    (if itempath (setenv "MT_ITEMPATH"  itempath))

    ;; get the info from the db and put it in the cache
    (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)))
    (if (not (get-environment-variable "MT_TARGET"))(setenv "MT_TARGET" target))
    (alist->env-vars (hash-table-ref/default *configdat* "env-override" '()))
    ;; Lets use this as an opportunity to put MT_RUNNAME in the environment
    (let ((runname  (if inrunname inrunname (rmt:get-run-name-from-id 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))
    (if (and testname link-tree)
	(setenv "MT_TEST_RUN_DIR" (conc (getenv "MT_LINKTREE")  "/"
					(getenv "MT_TARGET")    "/"

	  (let*-values (((waitons waitors config)(tests:get-waitons hed all-tests-registry)))
	    (debug:print-info 8 *default-log-port* "waitons: " waitons)
	    ;; check for hed in waitons => this would be circular, remove it and issue an
	    ;; error
	    (if (or (member hed waitons)
		    (member hed waitors))
		(begin
		  (debug:print-error 0 *default-log-port* "test " hed " has listed itself as a waiton or waitor, please correct this!")
		  (set! waitons (filter (lambda (x)(not (equal? x hed))) waitons))
		  (set! waitors (filter (lambda (x)(not (equal? x hed))) waitors))))
	    
	    ;; (items   (items:get-items-from-config config)))
	    (if (not (hash-table-ref/default test-records hed #f))
		(hash-table-set! test-records
				 hed (vector hed     ;; 0

		 (run-queue-retries 5)
		 (th1        (make-thread (lambda ()
					    (runs:run-tests-queue run-id runname test-records keyvals flags test-patts required-tests (any->number reglen) all-tests-registry))
					    ;; (handle-exceptions
					    ;;  exn
					    ;;  (begin
					    ;;    (print-call-chain (current-error-port))
					    ;;    (debug:print-error 0 *default-log-port* "failure in runs:run-tests-queue thread, error: " ((condition-property-accessor 'exn 'message) exn))
					    ;;    (if (> run-queue-retries 0)
					    ;; 	   (begin
					    ;; 	     (set! run-queue-retries (- run-queue-retries 1))
					    ;; 	     (runs:run-tests-queue run-id runname test-records keyvals flags test-patts required-tests (any->number reglen) all-tests-registry))))
					    ;;  (runs:run-tests-queue run-id runname test-records keyvals flags test-patts required-tests (any->number reglen) all-tests-registry)))
					  "runs:run-tests-queue"))
		 (th2        (make-thread (lambda ()				    

;;   (cond
;;    ((and regfull (null? reg)(not (null? tal)))      (car tal))
;;    ((and regfull (not (null? reg)))                 (car reg))
;;    ((and (not regfull)(null? tal)(not (null? reg))) (car reg))
;;    ((and (not regfull)(not (null? tal)))            (car tal))
;;    (else
;;     (debug:print-error 0 *default-log-port* "runs:queue-next-hed, tal=" tal ", reg=" reg ", n=" n ", regfull=" regfull)
;;     #f)))

(define (runs:queue-next-tal tal reg n regfull)
  (if regfull
      tal
      (if (null? tal) ;; must transfer from reg
	  (cdr reg)

			  (num-items (rmt:test-toplevel-num-items run-id test-name)))
		      (if (and test-id
			       (not (> num-items 0)))
			  (mt:test-set-state-status-by-id run-id test-id "NOT_STARTED" "ZERO_ITEMS" "Failed to run due to failed prerequisites"))))
		(tests:testqueue-set-items! test-record items-list)
		(list hed tal reg reruns))
	      (begin
		(debug:print-error 0 *default-log-port* "The proc from reading the items table did not yield a list - please report this")
		(exit 1))))))

     ((and (null? fails)
	   (null? prereq-fails)
	   (not (null? non-completed)))
      (let* ((allinqueue (map (lambda (x)(if (string? x) x (db:test-get-testname x)))
        		      (append newtal reruns)))

	 (max-concurrent-jobs     (list-ref run-limits-info 3))
	 (job-group-limit         (list-ref run-limits-info 4))
	 (prereqs-not-met         (rmt:get-prereqs-not-met run-id waitons hed item-path mode: testmode itemmaps: itemmaps))
	 ;; (prereqs-not-met         (mt:lazy-get-prereqs-not-met run-id waitons item-path mode: testmode itemmap: itemmap))
	 (fails                   (if (list? prereqs-not-met)
				      (runs:calc-fails prereqs-not-met)
				      (begin
					(debug:print-error 0 *default-log-port* "prereqs-not-met is not a list! " prereqs-not-met)
					'())))
	 (non-completed           (filter (lambda (x)             ;; remove hed from not completed list, duh, of course it is not completed!
					    (not (equal? x hed)))
					  (runs:calc-not-completed prereqs-not-met)))
	 (loop-list               (list hed tal reg reruns))
	 ;; configure the load runner
	 (numcpus                 (common:get-num-cpus))

	(rmt:register-test run-id test-name item-path)
	(if (rmt:get-test-id run-id test-name item-path)
	    (hash-table-set! test-registry (db:test-make-full-name test-name item-path) 'done)
	    (if (> numtries 0)
		(begin
		  (thread-sleep! 0.5)
		  (register-loop (- numtries 1)))
		(debug:print-error 0 *default-log-port* "failed to register test " (db:test-make-full-name test-name item-path)))))
      (if (not (eq? (hash-table-ref/default test-registry (db:test-make-full-name test-name "") #f) 'done))
	  (begin
	    (rmt:register-test run-id test-name "")
	    (if (rmt:get-test-id run-id test-name "")
		(hash-table-set! test-registry (db:test-make-full-name test-name "") 'done))))
      (runs:shrink-can-run-more-tests-count)   ;; DELAY TWEAKER (still needed?)
      (if (and (null? tal)(null? reg))

;;	(debugger-pauser)


	;; check for hed in waitons => this would be circular, remove it and issue an
	;; error
	(if (member test-name waitons)
	    (begin
	      (debug:print-error 0 *default-log-port* "test " test-name " has listed itself as a waiton, please correct this!")
	      (set! waiton (filter (lambda (x)(not (equal? x hed))) waitons))))

	(cond 
	 
	 ;; We want to catch tests that have waitons that are NOT in the queue and discard them IFF 
	 ;; they have been through the wringer 10 or more times
	 ((and (list? waitons)

		  (if loop-list
		      (apply loop loop-list)))
		;; if can't run more just loop with next possible test
		(loop (car newtal)(cdr newtal) reg reruns))))
	    
	 ;; this case should not happen, added to help catch any bugs
	 ((and (list? items) itemdat)
	  (debug:print-error 0 *default-log-port* "Should not have a list of items in a test and the itemspath set - please report this")
	  (exit 1))
	 ((not (null? reruns))
	  (let* ((newlst (tests:filter-non-runnable run-id tal test-records)) ;; i.e. not FAIL, WAIVED, INCOMPLETE, PASS, KILLED,
		 (junked (lset-difference equal? tal newlst)))
	    (debug:print-info 4 *default-log-port* "full drop through, if reruns is less than 100 we will force retry them, reruns=" reruns ", tal=" tal)
	    (if (< num-retries max-retries)
		(set! newlst (append reruns newlst)))

	    (set! testdat (rmt:get-test-info-by-id run-id test-id))
	    (if (not testdat)
		(begin
		  (debug:print-info 0 *default-log-port* "WARNING: server is overloaded, trying again in one second")
		  (thread-sleep! 1)
		  (loop)))))
      (if (not testdat) ;; should NOT happen
	  (debug:print-error 0 *default-log-port* "failed to get test record for test-id " test-id))
      (set! test-id (db:test-get-id testdat))
      (if (file-exists? test-path)
	  (change-directory test-path)
	  (begin
	    (debug:print-error 0 *default-log-port* "test run path not created before attempting to run the test. Perhaps you are running -remove-runs at the same time?")
	    (change-directory *toppath*)))
      (case (if force ;; (args:get-arg "-force")
		'NOT_STARTED
		(if testdat
		    (string->symbol (test:get-state testdat))
		    'failed-to-insert))
	((failed-to-insert)
	 (debug:print-error 0 *default-log-port* "Failed to insert the record into the db"))
	((NOT_STARTED COMPLETED DELETED INCOMPLETE)
	 (let ((runflag #f))
	   (cond
	    ;; -force, run no matter what
	    (force (set! runflag #t))
	    ;; NOT_STARTED, run no matter what
	    ((member (test:get-state testdat) '("DELETED" "NOT_STARTED" "INCOMPLETE"))(set! runflag #t))

	;; 	    6000)) ;; i.e. no update for more than 6000 seconds
	;;      (begin
	;;        (debug:print 0 *default-log-port* "WARNING: Test " test-name " appears to be dead. Forcing it to state INCOMPLETE and status STUCK/DEAD")
	;;        (tests:test-set-status! run-id test-id "INCOMPLETE" "STUCK/DEAD" "" #f))
	;;        ;; (tests:test-set-status! test-id "INCOMPLETE" "STUCK/DEAD" "" #f))
	;;      (debug:print 2 *default-log-port* "NOTE: " test-name " is already running")))
	(else      
	 (debug:print-error 0 *default-log-port* "Failed to launch test " full-test-name ". Unrecognised state " (test:get-state testdat))
	 (case (string->symbol (test:get-state testdat)) 
	   ((COMPLETED INCOMPLETE)
	    (hash-table-set! test-registry (db:test-make-full-name test-name test-path) 'DONOTRUN))
	   (else
	    (hash-table-set! test-registry (db:test-make-full-name test-name test-path) 'DONOTRUN))))))))

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

(define (runs:recursive-delete-with-error-msg real-dir)
  (if (> (system (conc "rm -rf " real-dir)) 0)
      (begin
	;; FAILED, possibly due to permissions, do chmod a+rwx then try one more time
	(system (conc "chmod -R a+rwx " real-dir))
	(if (> (system (conc "rm -rf " real-dir)) 0)
	    (debug:print-error 0 *default-log-port* "There was a problem removing " real-dir " with rm -f")))))

(define (runs:safe-delete-test-dir real-dir)
  ;; first delete all sub-directories
  (directory-fold 
   (lambda (f x)
     (let ((fullname (conc real-dir "/" f)))
       (if (directory? fullname)(runs:recursive-delete-with-error-msg fullname)))

	 (statuses     (if status (string-split status ",") '()))
	 (state-status (if (string? new-state-status) (string-split new-state-status ",") '(#f #f)))
	 (rp-mutex     (make-mutex))
	 (bup-mutex    (make-mutex)))
    (debug:print-info 4 *default-log-port* "runs:operate-on => Header: " header " action: " action " new-state-status: " new-state-status)
    (if (> 2 (length state-status))
	(begin
	  (debug:print-error 0 *default-log-port* "the parameter to -set-state-status is a comma delimited string. E.g. COMPLETED,FAIL")
	  (exit)))
    (for-each
     (lambda (run)
       (let ((runkey (string-intersperse (map (lambda (k)
						(db:get-value-by-header run header k)) keys) "/"))
	     (dirs-to-remove (make-hash-table))
	     (proc-get-tests (lambda (run-id)

		   ((archive)
		    (debug:print 1 *default-log-port* "Archiving/restoring (" (args:get-arg "-archive") ") data for run: " runkey " " (db:get-value-by-header run header "runname"))
		    (set! worker-thread (make-thread (lambda ()
						       (case (string->symbol (args:get-arg "-archive"))
							 ((save save-remove keep-html)(archive:run-bup (args:get-arg "-archive") run-id run-name tests rp-mutex bup-mutex))
							 ((restore)(archive:bup-restore (args:get-arg "-archive") run-id run-name tests rp-mutex bup-mutex))
							 (else 
							  (debug:print-error 0 *default-log-port* "unrecognised sub command to -archive. Run \"megatest\" to see help")
							  (exit))))
						     "archive-bup-thread"))
		    (thread-start! worker-thread))
		   (else
		    (debug:print-info 0 *default-log-port* "action not recognised " action)))
		 
		 ;; actions that operate on one test at a time can be handled below

		       (allow-run-time   10)) ;; seconds to allow for killing tests before just brutally killing 'em
		   (let loop ((test (car sorted-tests))
			      (tal  (cdr sorted-tests)))
		     (let* ((test-id       (db:test-get-id test))
			    (new-test-dat  (rmt:get-test-info-by-id run-id test-id)))
		       (if (not new-test-dat)
			   (begin
			     (debug:print-error 0 *default-log-port* "We have a test-id of " test-id " but no record was found. NOTE: No locking of records is done between processes, do not simultaneously remove the same run from two processes!")
			     (if (not (null? tal))
				 (loop (car tal)(cdr tal))))
			   (let* ((item-path     (db:test-get-item-path new-test-dat))
				  (test-name     (db:test-get-testname new-test-dat))
				  (run-dir       ;;(filedb:get-path *fdb*
				   ;; (rmt:sdb-qry 'getid 
				   (db:test-get-rundir new-test-dat)) ;; )    ;; run dir is from the link tree

	    (debug:print 0 *default-log-port* "WARNING: directory " real-dir " does not exist")
	    (debug:print 0 *default-log-port* "WARNING: no real directory corrosponding to link " run-dir ", nothing done")))
    (if (symbolic-link? run-dir)
	(begin
	  (debug:print-info 1 *default-log-port* "Removing symlink " run-dir)
	  (handle-exceptions
	   exn
	   (debug:print-error 0 *default-log-port* " Failed to remove symlink " run-dir ((condition-property-accessor 'exn 'message) exn) ", attempting to continue")
	   (delete-file run-dir)))
	(if (directory? run-dir)
	    (if (> (directory-fold (lambda (f x)(+ 1 x)) 0 run-dir) 0)
		(debug:print 0 *default-log-port* "WARNING: refusing to remove " run-dir " as it is not empty")
		(handle-exceptions
		 exn
		 (debug:print-error 0 *default-log-port* " Failed to remove directory " run-dir ((condition-property-accessor 'exn 'message) exn) ", attempting to continue")
		 (delete-directory run-dir)))
	    (if (and run-dir
		     (not (member run-dir (list "n/a" "/tmp/badname"))))
		(debug:print 0 *default-log-port* "WARNING: not removing " run-dir " as it either doesn't exist or is not a symlink")
		(debug:print 0 *default-log-port* "NOTE: the run dir for this test is undefined. Test may have already been deleted."))
	    ))
    ;; Only delete the records *after* removing the directory. If things fail we have a record 

;; Since many calls to a run require pretty much the same setup 
;; this wrapper is used to reduce the replication of code
(define (general-run-call switchname action-desc proc)
  (let ((runname (or (args:get-arg "-runname")(args:get-arg ":runname")))
	(target  (common:args-get-target)))
    (cond
     ((not target)
      (debug:print-error 0 *default-log-port* "Missing required parameter for " switchname ", you must specify the target with -target")
      (exit 3))
     ((not runname)
      (debug:print-error 0 *default-log-port* "Missing required parameter for " switchname ", you must specify the run name with -runname runname")
      (exit 3))
     (else
      (let (;; (db   #f)
	    (keys #f))
	(if (launch:setup)
	    (begin
	      (full-runconfigs-read) ;; cache the run config
	      (launch:cache-config)) ;; do not cache here - need to be sure runconfigs is processed
	    (begin 
	      (debug:print 0 *default-log-port* "Failed to setup, exiting")
	      (exit 1)))
	(set! keys (keys:config-get-fields *configdat*))
	;; have enough to process -target or -reqtarg here
	(if (args:get-arg "-reqtarg")
	    (let* ((runconfigf (conc  *toppath* "/runconfigs.config")) ;; DO NOT EVALUATE ALL 
		   (runconfig  (read-config runconfigf #f #t environ-patt: #f)))
	      (if (hash-table-ref/default runconfig (args:get-arg "-reqtarg") #f)
		  (keys:target-set-args keys (args:get-arg "-reqtarg") args:arg-hash)
		    
		  (begin
		    (debug:print-error 0 *default-log-port* "[" (args:get-arg "-reqtarg") "] not found in " runconfigf)
		    ;; (if db (sqlite3:finalize! db))
		    (exit 1)
		    )))
	    (if (args:get-arg "-target")
		(keys:target-set-args keys (args:get-arg "-target" args:arg-hash) args:arg-hash)))
	(if (not (car *configinfo*))
	    (begin
	      (debug:print-error 0 *default-log-port* "Attempted to " action-desc " but run area config file not found")
	      (exit 1))
	    ;; Extract out stuff needed in most or many calls
	    ;; here then call proc
	    (let* ((keyvals    (keys:target->keyval keys target)))
	      (proc target runname keys keyvals)))
	;; (if db (sqlite3:finalize! db))
	(set! *didsomething* #t))))))

;; start_server
;;
(define (server:launch run-id)
  (case *transport-type*
    ((http)(http-transport:launch run-id))
    ((nmsg)(nmsg-transport:launch run-id))
    ((rpc)  (rpc-transport:launch run-id))
    (else (debug:print-error 0 *default-log-port* "unknown server type " *transport-type*))))
;;       (else   (debug:print-error 0 *default-log-port* "No known transport set, transport=" transport ", using rpc")
;; 	      (rpc-transport:launch run-id)))))

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

;; Get the transport

    ((http) (db:obj->string (vector success/fail query-sig result)))
    ((zmq)
     (let ((pub-socket (vector-ref *runremote* 1)))
       (send-message pub-socket return-addr send-more: #t)
       (send-message pub-socket (db:obj->string (vector success/fail query-sig result)))))
    ((fs)   result)
    (else 
     (debug:print-error 0 *default-log-port* "unrecognised transport type: " *transport-type*)
     result)))

;; 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
;;
(define  (server:run run-id)

			(if (eq? (length slst) 2)
			    (list (car slst)(string->number (cadr slst)))
			    #f)))
	   (toppath       (launch:setup))
	   (server-db-dat (if (not host-port)(tasks:get-server (db:delay-if-busy tdbdat) run-id) #f)))
      (if (not run-id)
	  (begin
	    (debug:print-error 0 *default-log-port* "must specify run-id when doing ping, -run-id n")
	    (print "ERROR: No run-id")
	    (exit 1))
	  (if (and (not host-port)
		   (not server-db-dat))
	      (begin
		(print "ERROR: bad host:port")
		(exit 1))

	   (call-with-database
            dbpath
	    (lambda (db)
	       ;;(debug:print 0 *default-log-port* "calling proc " proc " on db " db)
	      (set-busy-handler! db (busy-timeout 10000)) ;; 10 sec timeout
	      (if (not dbexists)(sretrieve:initialize-db db))
	      (proc db)))))
	(debug:print-error 0 *default-log-port* "invalid path for storing database: " path))))

;; copy in directory to dest, validation is done BEFORE calling this
;;
(define (sretrieve:get configdat retriever version comment)
  (let* ((base-dir  (configf:lookup configdat "settings" "base-dir"))
	 (datadir   (conc base-dir "/" version)))
    (if (or (not base-dir)
	    (not (file-exists? base-dir)))
	(begin
	  (debug:print-error 0 *default-log-port* "Bad configuration! base-dir " base-dir " not found")
	  (exit 1)))
    (print datadir)
    (if (not (file-exists? datadir))
	(begin
	  (debug:print-error 0 *default-log-port* "Bad version (" version "), no data found at " datadir "." )
	  (exit 1)))
    
    (sretrieve:db-do
     configdat
     (lambda (db)
       (sretrieve:register-action db "get" retriever datadir comment)))
      (sretrieve:do-as-calling-user

  (let* ((base-dir  (configf:lookup configdat "settings" "base-dir"))
         (allowed-sub-paths (configf:lookup configdat "settings" "allowed-sub-paths"))    
	 (datadir   (conc base-dir "/" file))
         (filename  (conc(pathname-file datadir) "." (pathname-extension datadir))))
    (if (or (not base-dir)
	    (not (file-exists? base-dir)))
	(begin
	  (debug:print-error 0 *default-log-port* "Bad configuration! base-dir " base-dir " not found")
	  (exit 1)))
    (print datadir)
    (if (not (file-exists? datadir))
	(begin
	  (debug:print-error 0 *default-log-port* "File  (" file "), not found at " base-dir "." )
	  (exit 1)))
    (if (directory? datadir)
	(begin
	  (debug:print-error 0 *default-log-port* "(" file ") is a dirctory!! cp cmd works only on files ." )
	  (exit 1)))
    (if(not (string-match (regexp  allowed-sub-paths) file))
        (begin
	  (debug:print-error 0 *default-log-port* "Access denied to file (" file ")!! " )
	  (exit 1)))
     
     (sretrieve:db-do
     configdat
     (lambda (db)
       (sretrieve:register-action db "cp" retriever datadir comment)))
      (sretrieve:do-as-calling-user

  (let* ((base-dir  (configf:lookup configdat "settings" "base-dir"))
         (allowed-sub-paths (configf:lookup configdat "settings" "allowed-sub-paths"))    
	 (datadir   (conc base-dir "/" file))
         (filename  (conc(pathname-file datadir) "." (pathname-extension datadir))))
    (if (or (not base-dir)
	    (not (file-exists? base-dir)))
	(begin
	  (debug:print-error 0 *default-log-port* "Bad configuration! base-dir " base-dir " not found")
	  (exit 1)))
    (print datadir)
    (if (not (file-exists? datadir))
	(begin
	  (debug:print-error 0 *default-log-port* "File  (" file "), not found at " base-dir "." )
	  (exit 1)))
      (if(not (string-match (regexp  allowed-sub-paths) file))
        (begin
	  (debug:print-error 0 *default-log-port* "Access denied to file (" file ")!! " )
	  (exit 1)))
   
        (sretrieve:do-as-calling-user
        (lambda ()
	 ;;(change-directory datadir)
         ;; (debug:print 0 *default-log-port*  "/usr/bin/find" (list datadir "-ls" "|" "grep" "-E" "'"allowed-file-patt"'"))
         ;; (status (with-input-from-pipe "find " datadir " -ls | grep -E '" allowed-file-patt "'" (lambda () (read-line))))

;;							   (glob "*" ".*"))))))))

(define (sretrieve:validate target-dir targ-mk)
  (let* ((normal-path (normalize-pathname targ-mk))
        (targ-path (conc target-dir "/" normal-path)))
    (if (string-contains   normal-path "..")
    (begin
      (debug:print-error 0 *default-log-port* "Path  " targ-mk " resolved outside target area "  target-dir )
      (exit 1)))

    (if (not (string-contains targ-path target-dir))
    (begin
      (debug:print-error 0 *default-log-port* "You cannot update data outside " target-dir ".")
      (exit 1)))
    (debug:print 0 *default-log-port* "Path " targ-mk " is valid.")   
 ))
;; make directory in dest
;;

(define (sretrieve:mkdir configdat submitter target-dir targ-mk comment)
  (let ((targ-path (conc target-dir "/" targ-mk)))
    
    (if (file-exists? targ-path)
	(begin
	  (debug:print-error 0 *default-log-port* "target Directory " targ-path " already exist!!")
	  (exit 1)))
    (sretrieve:db-do
     configdat
     (lambda (db)
       (sretrieve:register-action db "mkdir" submitter targ-mk comment)))
    (let* ((th1         (make-thread
			 (lambda ()

;; create a symlink in dest
;;
(define (sretrieve:ln configdat submitter target-dir targ-link link-name comment)
  (let ((targ-path (conc target-dir "/" link-name)))
    (if (file-exists? targ-path)
	(begin
	  (debug:print-error 0 *default-log-port* "target file " targ-path " already exist!!")
	  (exit 1)))
     (if (not (file-exists? targ-link ))
	(begin
	  (debug:print-error 0 *default-log-port* "target file " targ-link " does not exist!!")
	  (exit 1)))
 
    (sretrieve:db-do
     configdat
     (lambda (db)
       (sretrieve:register-action db "ln" submitter link-name comment)))
    (let* ((th1         (make-thread

;; remove copy of file in dest
;;
(define (sretrieve:rm configdat submitter target-dir targ-file comment)
  (let ((targ-path (conc target-dir "/" targ-file)))
    (if (not (file-exists? targ-path))
	(begin
	  (debug:print-error 0 *default-log-port* "target file " targ-path " not found, nothing to remove.")
	  (exit 1)))
    (sretrieve:db-do
     configdat
     (lambda (db)
       (sretrieve:register-action db "rm" submitter targ-file comment)))
    (let* ((th1         (make-thread
			 (lambda ()

    ;;   <packages-metadir>/<package-type>.config file using
    ;;   <upstream-file> as an input
    (if (file-exists? upstream-file)
	(if (or (not (file-exists? package-config)) ;; if not created call the updater, otherwise call only if upstream newer
		(> (file-modification-time upstream-file)(file-modification-time package-config)))
	    (handle-exceptions
	     exn
	     (debug:print-error 0 *default-log-port* "failed to run script " conversion-script " with params " upstream-file " " package-config)
	     (let ((pid (process-run conversion-script (list upstream-file package-config))))
	       (process-wait pid)))
	    (debug:print 0 *default-log-port* "Skipping update of " package-config " from " upstream-file))
	(debug:print 0 *default-log-port* "Skipping update of " package-config " as " upstream-file " not found"))
    ;; (ini:property-separator-patt " *  *")
    ;; (ini:property-separator #\space)
    (let ((res (if (file-exists? package-config)

	(begin
	  (debug:print 0 *default-log-port* "User \"" (current-user-name) "\" does not have access. Exiting")
	  (exit 1)))
    (case (string->symbol action)
      ((get)
       (if (< (length args) 1)
	   (begin 
	     (debug:print-error 0 *default-log-port* "Missing arguments; " (string-intersperse args ", "))
	     (exit 1)))
       (let* ((remargs     (args:get-args args '("-m" "-i" "-package") '() args:arg-hash 0))
              (version     (car args))
	      (msg         (or (args:get-arg "-m") ""))
	      (package-type (or (args:get-arg "-package")
				default-area))
	      (exe-dir     (configf:lookup configdat "exe-info" "exe-dir")))
;;	      (relconfig   (sretrieve:load-packages configdat exe-dir package-type)))

	 (debug:print 0 *default-log-port* "retrieving " version " of " package-type " as tar data on stdout")
	 (sretrieve:get configdat user version msg)))
         ((cp)
            (if (< (length args) 1)
             (begin 
	     (debug:print-error 0 *default-log-port* "Missing arguments; " (string-intersperse args ", "))
	     (exit 1)))
          (let* ((remargs     (args:get-args args '("-m" "-i" "-package") '() args:arg-hash 0))
              (file     (car args))
	      (msg         (or (args:get-arg "-m") "")) )

	 (debug:print 0 *default-log-port* "copinging " file " to current directory " )
	 (sretrieve:cp configdat user file msg)))
      ((ls)
            (if (< (length args) 1)
             (begin 
	     (debug:print-error 0 *default-log-port* "Missing arguments; " (string-intersperse args ", "))
	     (exit 1)))
          (let* ((remargs     (args:get-args args '("-m" "-i" "-package") '() args:arg-hash 0))
              (dir     (car args))
	      (msg         (or (args:get-arg "-m") "")) )

	 (debug:print 0 *default-log-port* "Listing files in " )
	 (sretrieve:ls configdat user dir msg)))

	 (sretrieve:shell))
	(else
	 (print "ERROR: Unrecognised command. Try \"sretrieve help\""))))
     ;; multi-word commands
     ((null? rema)(print sretrieve:help))
     ((>= (length rema) 2)
      (apply sretrieve:process-action configdat (car rema)(cdr rema)))
     (else (debug:print-error 0 *default-log-port* "Unrecognised command. Try \"sretrieve help\"")))))

(main)

;; Tasks db
;;======================================================================

;; wait up to aprox n seconds for a journal to go away
;;
(define (tasks:wait-on-journal path n #!key (remove #f)(waiting-msg #f))
  (if (not (string? path))
      (debug:print-error 0 *default-log-port* "Called tasks:wait-on-journal with path=" path " (not a string)")
      (let ((fullpath (conc path "-journal")))
	(handle-exceptions
	 exn
	 (begin
	   (print-call-chain (current-error-port))
	   (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
	   (debug:print 0 *default-log-port* " exn=" (condition->list exn))

(define (tasks:get-task-db-path)
  (let ((dbdir  (or (configf:lookup *configdat* "setup" "monitordir")
		    (configf:lookup *configdat* "setup" "dbdir")
		    (conc (configf:lookup *configdat* "setup" "linktree") "/.db"))))
    (handle-exceptions
     exn
     (begin
       (debug:print-error 0 *default-log-port* "Couldn't create path to " dbdir)
       (exit 1))
     (if (not (directory? dbdir))(create-directory dbdir #t)))
    dbdir))

;; If file exists AND
;;    file readable
;;         ==> open it

		(loop (get-rand-port)(- remtries 1))
		(get-rand-port))
	    port))))))

(define (tasks:server-am-i-the-server? mdb run-id)
  (let* ((all    (tasks:server-get-servers-vying-for-run-id mdb run-id))
	 (first  (if (null? all)
		     #f;; (begin (debug:print-error 0 *default-log-port* "no servers listed, should be at least one by now.") 
		       ;;      (sqlite3:finalize! mdb)
		       ;;      (exit 1))
		     (car (db:get-rows all)))))
    (if first
	(let* ((header   (db:get-header all))
	       (id       (db:get-value-by-header first header "id"))
	       (hostname (db:get-value-by-header first header "hostname"))

		   (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 (tasks:start-run dbstruct mdb task)
;;   (let ((flags (make-hash-table)))
;;     (hash-table-set! flags "-rerun" "NOT_STARTED")
;;     (if (not (string=? (tasks:task-get-params task) ""))
;; 	(hash-table-set! flags "-setvars" (tasks:task-get-params task)))

	;; (sqlite3:execute db "PRAGMA synchronous = 0;")
	(debug:print-info 11 *default-log-port* "open-test-db END (sucessful)" work-area)
	;; now let's test that everything is correct
	(handle-exceptions
	 exn
	 (begin
	   (print-call-chain (current-error-port))
	   (debug:print-error 0 *default-log-port* "problem accessing test db " work-area ", you probably should clean and re-run this test or remove the file " 
			dbpath ".\n  "
			((condition-property-accessor 'exn 'message) exn))
	   #f)
	 ;; Is there a cheaper single line operation that will check for existance of a table
	 ;; and raise an exception ?
	 (sqlite3:execute db "SELECT id FROM test_data LIMIT 1;"))
	db)

;; returns waitons waitors tconfigdat
;;
(define (tests:get-waitons test-name all-tests-registry)
   (let* ((config  (tests:get-testconfig test-name all-tests-registry 'return-procs)))
     (let ((instr (if config 
		      (config-lookup config "requirements" "waiton")
		      (begin ;; No config means this is a non-existant test
			(debug:print-error 0 *default-log-port* "non-existent required test \"" test-name "\"")
			(exit 1))))
	   (instr2 (if config
		       (config-lookup config "requirements" "waitor")
		       "")))
       (debug:print-info 8 *default-log-port* "waitons string is " instr ", waitors string is " instr2)
       (let ((newwaitons
	      (string-split (cond
			     ((procedure? instr) ;; here 
			      (let ((res (instr)))
				(debug:print-info 8 *default-log-port* "waiton procedure results in string " res " for test " test-name)
				res))
			     ((string? instr)     instr)
			     (else 
			      ;; NOTE: This is actually the case of *no* waitons! ;; (debug:print-error 0 *default-log-port* "something went wrong in processing waitons for test " test-name)
			      ""))))
	     (newwaitors
	      (string-split (cond
			     ((procedure? instr2)
			      (let ((res (instr2)))
				(debug:print-info 8 *default-log-port* "waitor procedure results in string " res " for test " test-name)
				res))
			     ((string? instr2)     instr2)
			     (else 
			      ;; NOTE: This is actually the case of *no* waitons! ;; (debug:print-error 0 *default-log-port* "something went wrong in processing waitons for test " test-name)
			      "")))))
	 (values
	  ;; the waitons
	  (filter (lambda (x)
		    (if (hash-table-ref/default all-tests-registry x #f)
			#t
			(begin
			  (debug:print-error 0 *default-log-port* "test " test-name " has unrecognised waiton testname " x)
			  #f)))
		  newwaitons)
	  (filter (lambda (x)
		    (if (hash-table-ref/default all-tests-registry x #f)
			#t
			(begin
			  (debug:print-error 0 *default-log-port* "test " test-name " has unrecognised waiton testname " x)
			  #f)))
		  newwaitors)
	  config)))))
					     
;; given waiting-test that is waiting on waiton-test extend test-patt appropriately
;;
;;  genlib/testconfig               sim/testconfig

	  (db:test-get-rundir prev-testdat)) ;; )
	 (waivers     (if testconfig (configf:section-vars testconfig "waivers") '()))
	 (waiver-rx   (regexp "^(\\S+)\\s+(.*)$"))
	 (diff-rule   "diff %file1% %file2%")
	 (logpro-rule "diff %file1% %file2% | logpro %waivername%.logpro %waivername%.html"))
    (if (not (file-exists? test-rundir))
	(begin
	  (debug:print-error 0 *default-log-port* "test run directory is gone, cannot propagate waiver")
	  #f)
	(begin
	  (push-directory test-rundir)
	  (let ((result (if (null? waivers)
			    #f
			    (let loop ((hed (car waivers))
				       (tal (cdr waivers)))

    ;; 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)))

  
;; sort tests by priority and waiton
;; Move test specific stuff to a test unit FIXME one of these days
(define (tests:sort-by-priority-and-waiton test-records)
  (let* ((mungepriority (lambda (priority)
			  (if priority
			      (let ((tmp (any->number priority)))
				(if tmp tmp (begin (debug:print-error 0 *default-log-port* "bad priority value " priority ", using 0") 0)))
			      0)))
	 (all-tests      (hash-table-keys test-records))
	 (all-waited-on  (let loop ((hed (car all-tests))
				    (tal (cdr all-tests))
				    (res '()))
			   (let* ((trec    (hash-table-ref test-records hed))
				  (waitons (or (tests:testqueue-get-waitons trec) '())))

      (let loop ((hed (car test-names))
		 (tal (cdr test-names)))         ;; 'return-procs tells the config reader to prep running system but return a proc
	(debug:print-info 4 *default-log-port* "hed=" hed " at top of loop")
	(let* ((config  (tests:get-testconfig hed all-tests-registry 'return-procs))
	       (waitons (let ((instr (if config 
					 (config-lookup config "requirements" "waiton")
					 (begin ;; No config means this is a non-existant test
					   (debug:print-error 0 *default-log-port* "non-existent required test \"" hed "\", grep through your testconfigs to find and remove or create the test. Discarding and continuing.")
					     ""))))
			  (debug:print-info 8 *default-log-port* "waitons string is " instr)
			  (string-split (cond
					 ((procedure? instr)
					  (let ((res (instr)))
					    (debug:print-info 8 *default-log-port* "waiton procedure results in string " res " for test " hed)
					    res))
					 ((string? instr)     instr)
					 (else 
					  ;; NOTE: This is actually the case of *no* waitons! ;; (debug:print-error 0 *default-log-port* "something went wrong in processing waitons for test " hed)
					  ""))))))
	  (if (not config) ;; this is a non-existant test called in a waiton. 
	      (if (null? tal)
		  test-records
		  (loop (car tal)(cdr tal)))
	      (begin
		(debug:print-info 8 *default-log-port* "waitons: " waitons)
		;; check for hed in waitons => this would be circular, remove it and issue an
		;; error
		(if (member hed waitons)
		    (begin
		      (debug:print-error 0 *default-log-port* "test " hed " has listed itself as a waiton, please correct this!")
		      (set! waitons (filter (lambda (x)(not (equal? x hed))) waitons))))
		
		;; (items   (items:get-items-from-config config)))
		(if (not (hash-table-ref/default test-records hed #f))
		    (hash-table-set! test-records
				     hed (vector hed     ;; 0
						 config  ;; 1

	 (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))
	   (print "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)
  )))