Megatest

PREFIX=$(PWD)
CSCOPTS= 
INSTALL=install
SRCFILES = common.scm items.scm launch.scm \
           ods.scm runconfig.scm server.scm configf.scm \
           db.scm keys.scm margs.scm megatest-version.scm \
           process.scm runs.scm tasks.scm tests.scm genexample.scm \
	   fs-transport.scm zmq-transport.scm http-transport.scm \
	   fs-transport.scm http-transport.scm \
           client.scm gutils.scm synchash.scm daemon.scm

GUISRCF  = dashboard-tests.scm dashboard-guimonitor.scm 

OFILES   = $(SRCFILES:%.scm=%.o)
GOFILES  = $(GUISRCF:%.scm=%.o)

tests.o db.o launch.o runs.o dashboard-tests.o dashboard-guimonitor.o dashboard-main.o monitor.o dashboard.o megatest.o : db_records.scm
tests.o runs.o dashboard.o dashboard-tests.o dashboard-main.o  : run_records.scm
db.o ezsteps.o keys.o launch.o megatest.o monitor.o runs-for-ref.o runs.o tests.o : key_records.scm
tests.o tasks.o dashboard-tasks.o : task_records.scm
runs.o : test_records.scm
megatest.o : megatest-fossil-hash.scm

# Temporary while transitioning to new routine
runs.o : run-tests-queue-classic.scm  run-tests-queue-new.scm

megatest-fossil-hash.scm : $(SRCFILES) megatest.scm *_records.scm
	echo "(define megatest-fossil-hash \"$(MTESTHASH)\")" > megatest-fossil-hash.new
	if ! diff -q megatest-fossil-hash.new megatest-fossil-hash.scm ; then echo copying .new to .scm;cp -f megatest-fossil-hash.new megatest-fossil-hash.scm;fi

$(OFILES) $(GOFILES) : common_records.scm 

%.o : %.scm

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

install : bin $(PREFIX)/bin/mtest $(PREFIX)/bin/megatest $(PREFIX)/bin/dboard $(PREFIX)/bin/dashboard $(HELPERS) $(PREFIX)/bin/nbfake $(PREFIX)/bin/nbfind $(PREFIX)/bin/newdboard

deploytarg/apropos.so : Makefile
	for i in apropos base64 canvas-draw csv-xml directory-utils dot-locking extras fmt format hostinfo http-client intarweb json md5 message-digest posix posix-extras readline regex regex-case s11n spiffy spiffy-request-vars sqlite3 srfi-1 srfi-18 srfi-69 tcp test uri-common zmq check-errors synch matchable sql-null tcp-server rpc blob-utils string-utils variable-item defstruct uri-generic sendfile opensll openssl lookup-table list-utils stack; do \
	for i in apropos base64 canvas-draw csv-xml directory-utils dot-locking extras fmt format hostinfo http-client intarweb json md5 message-digest posix posix-extras readline regex regex-case s11n spiffy spiffy-request-vars sqlite3 srfi-1 srfi-18 srfi-69 tcp test uri-common check-errors synch matchable sql-null tcp-server rpc blob-utils string-utils variable-item defstruct uri-generic sendfile opensll openssl lookup-table list-utils stack; do \
	chicken-install -prefix deploytarg -deploy $$i;done

deploytarg/libsqlite3.so : 
	CSC_OPTIONS="-Ideploytarg -Ldeploytarg" $CHICKEN_INSTALL -prefix deploytarg -deploy sqlite3

1. All run control access to db is direct.
2. All test machines must have megatest available
3. Tests may or may not have file system access to the originating
   run area. rsync is used to pull the test area to the home host
   if and only if the originating area can not be seen via file 
   system. NO LONGER TRUE. Rsync is used but file system must be visible.
4. All db access is done via the home host. NOT IMPLEMENTED YET.


fdktestqa on Apr 29, 2013: 1812 tests

INFO: (0) Max cached queries was    10
INFO: (0) Number of cached writes   41335
INFO: (0) Average cached write time 206.081553163179 ms
INFO: (0) Number non-cached queries 74289
INFO: (0) Average non-cached time   1055.09826488444 ms
INFO: (0) Server shutdown complete. Exiting

Start: 0 at Sun Apr 28 22:18:25 MST 2013
Max:  52 at Sun Apr 28 23:06:59 MST 2013
End:   6 at Sun Apr 28 23:47:51 MST 2013

;;======================================================================
;; C L I E N T S
;;======================================================================

(require-extension (srfi 18) extras tcp s11n)

(use sqlite3 srfi-1 posix regex regex-case srfi-69 hostinfo md5 message-digest zmq)
(use sqlite3 srfi-1 posix regex regex-case srfi-69 hostinfo md5 message-digest)
;; (use zmq)

(import (prefix sqlite3 sqlite3:))

(use spiffy uri-common intarweb http-client spiffy-request-vars)
(use spiffy uri-common intarweb http-client spiffy-request-vars uri-common intarweb)

(declare (unit client))

(declare (uses common))
(declare (uses db))
(declare (uses tasks)) ;; tasks are where stuff is maintained about what is running.

    			       (string->symbol (tasks:hostinfo-get-transport hostinfo))
			       'fs))
    ;; ;; DEBUG STUFF
    ;; (if (eq? *transport-type* 'fs)(begin (print "ERROR!!!!!!! refusing to run with transport " *transport-type*)(exit 99)))
    
    (debug:print-info 11 "Using transport type of " *transport-type* (if hostinfo (conc " to connect to " hostinfo) ""))
    (case *transport-type* 
      ((fs)(if (not *megatest-db*)(set! *megatest-db* (open-db))))
      ((fs) ;; (if (not *megatest-db*)(set! *megatest-db* (open-db))))
       ;; we are not doing fs any longer. let's cheat and start up a server
       (set! *transport-type* #f)
       (system "megatest -list-servers | grep alive || megatest -server - -daemonize && sleep 3")
       (thread-sleep! 1)
       (if (> numtries 0)
	   (client:setup numtries: (- numtries 1))))
      ((http)
       (http-transport:client-connect (tasks:hostinfo-get-interface hostinfo)
				      (tasks:hostinfo-get-port hostinfo)))
      ((zmq)
       (zmq-transport:client-connect (tasks:hostinfo-get-interface hostinfo)
				     (tasks:hostinfo-get-port      hostinfo)
				     (tasks:hostinfo-get-pubport   hostinfo)))

(define configf:key-sys-pr (regexp "^(\\S+)\\s+\\[system\\s+(\\S+.*)\\]\\s*$"))
(define configf:key-val-pr (regexp "^(\\S+)(\\s+(.*)|())$"))
(define configf:comment-rx (regexp "^\\s*#.*"))
(define configf:cont-ln-rx (regexp "^(\\s+)(\\S+.*)$"))

;; read a line and process any #{ ... } constructs

(define configf:var-expand-regex (regexp "^(.*)#\\{(scheme|system|shell|getenv|get|runconfigs-get)\\s+([^\\}\\{]*)\\}(.*)"))
(define configf:var-expand-regex (regexp "^(.*)#\\{(scheme|system|shell|getenv|get|runconfigs-get|rget)\\s+([^\\}\\{]*)\\}(.*)"))
(define (configf:process-line l ht)
  (let loop ((res l))
    (if (string? res)
	(let ((matchdat (string-search configf:var-expand-regex res)))
	  (if matchdat
	      (let* ((prestr  (list-ref matchdat 1))
		     (cmdtype (list-ref matchdat 2)) ;; eval, system, shell, getenv

				((getenv)(conc "(lambda (ht)(get-environment-variable \"" cmd "\"))"))
				((get)   
				 (let* ((parts (string-split cmd))
					(sect  (car parts))
					(var   (cadr parts)))
				   (conc "(lambda (ht)(config-lookup ht \"" sect "\" \"" var "\"))")))
				((runconfigs-get) (conc "(lambda (ht)(runconfigs-get ht \"" cmd "\"))"))
				((rget)           (conc "(lambda (ht)(runconfigs-get ht \"" cmd "\"))"))
				(else "(lambda (ht)(print \"ERROR\") \"ERROR\")"))))
		;; (print "fullcmd=" fullcmd)
		(with-input-from-string fullcmd
		  (lambda ()
		    (set! result ((eval (read)) ht))))
		(loop (conc prestr result poststr)))
	      res))

	    (print "ERROR: " cmd " returned bad exit code " status))
	  ""))))

;; Lookup a value in runconfigs based on -reqtarg or -target
(define (runconfigs-get config var)
  (let ((targ (or (args:get-arg "-reqtarg")(args:get-arg "-target"))))
    (if targ
	(config-lookup config targ var)
	#f)))
	(or (configf:lookup config targ var)
	    (configf:lookup config "default" var))
	(configf:lookup config "default" var))))

(define-inline (configf:read-line p ht allow-processing)
  (let loop ((inl (read-line p)))
    (if (and (string? inl)
	     (not (string-null? inl))
	     (equal? "\\" (string-take-right inl 1))) ;; last character is \ 
	(let ((nextl (read-line p)))
	  (if (not (eof-object? nextl))
	      (loop (string-append inl nextl))))
  (if (and allow-processing 
	   (not (eq? allow-processing 'return-string)))
      (configf:process-line (read-line p) ht)
      (read-line p)))
	(if (and allow-processing 
		 (not (eq? allow-processing 'return-string)))
	    (configf:process-line inl ht)
	    inl))))

;; read a config file, returns hash table of alists

;; read a config file, returns hash table of alists
;; adds to ht if given (must be #f otherwise)
;; envion-patt is a regex spec that identifies sections that will be eval'd
;; in the environment on the fly

	  (exit 1))
	(let* ((run-id        (if testdat (db:test-get-run_id testdat) #f))
	       (keydat        (if testdat (open-run-close db:get-key-val-pairs #f run-id) #f))
	       (rundat        (if testdat (open-run-close db:get-run-info #f run-id) #f))
	       (runname       (if testdat (db:get-value-by-header (db:get-row rundat)
								  (db:get-header rundat)
								  "runname") #f))
	       (teststeps     (if testdat (db:get-compressed-steps test-id) '()))
	       (logfile       "/this/dir/better/not/exist")
	       (rundir        logfile)
	       (teststeps     (if testdat (db:get-compressed-steps test-id work-area: rundir) '()))
	       (testfullname  (if testdat (db:test-get-fullname testdat) "Gathering data ..."))
	       (testname      (if testdat (db:test-get-testname testdat) "n/a"))
	       (testmeta      (if testdat 
				  (let ((tm (open-run-close db:testmeta-get-record #f testname)))
				    (if tm tm (make-db:testmeta)))
				  (make-db:testmeta)))

						 ";xterm -T \"" (string-translate testfullname "()" "  ") "\" " shell "&")))
				 (message-window  (conc "Directory " rundir " not found")))))
	       (refreshdat (lambda ()
			     (let* ((curr-mod-time (file-modification-time db-path))
				    (need-update   (or (and (> curr-mod-time db-mod-time)
							    (> (current-seconds) (+ last-update 2))) ;; every two seconds if db touched
						       request-update))
				    (newtestdat (if need-update (open-run-close db:get-test-info-by-id #f test-id))))
				    (newtestdat (if need-update 
						    (handle-exceptions
						     exn 
						     (debug:print-info 2 "test db access issue: " ((condition-property-accessor 'exn 'message) exn))
						     (open-run-close db:get-test-info-by-id #f test-id )))))
			       (cond
				((and need-update newtestdat)
				 (set! testdat newtestdat)
				 (set! teststeps    (db:get-compressed-steps test-id))
				 (set! teststeps    (db:get-compressed-steps test-id work-area: rundir))
				 (set! logfile      (conc (db:test-get-rundir testdat) "/" (db:test-get-final_logf testdat)))
				 (set! rundir       (db:test-get-rundir testdat))
				 (set! testfullname (db:test-get-fullname testdat))
				 ;; (debug:print 0 "INFO: teststeps=" (intersperse teststeps "\n    "))
				 )
				(need-update ;; if this was true and yet there is no data ....
				 (db:test-set-testname! testdat "DEAD OR DELETED TEST"))))))

;; (import (prefix rpc rpc:))

(use sqlite3 srfi-1 posix regex regex-case srfi-69 csv-xml s11n md5 message-digest base64)
(import (prefix sqlite3 sqlite3:))
(import (prefix base64 base64:))

;; Note, try to remove this dependency 
(use zmq)
;; (use zmq)

(declare (unit db))
(declare (uses common))
(declare (uses keys))
(declare (uses ods))
(declare (uses fs-transport))
(declare (uses client))

    ))

;;======================================================================
;; T E S T   S P E C I F I C   D B 
;;======================================================================

;; Create the sqlite db for the individual test(s)
(define (open-test-db testpath) 
  (debug:print-info 11 "open-test-db " testpath)
  (if (and testpath 
	   (directory? testpath)
	   (file-read-access? testpath))
      (let* ((dbpath    (conc testpath "/testdat.db"))
(define (open-test-db work-area) 
  (debug:print-info 11 "open-test-db " work-area)
  (if (and work-area 
	   (directory? work-area)
	   (file-read-access? work-area))
      (let* ((dbpath    (conc work-area "/testdat.db"))
	     (dbexists  (file-exists? dbpath))
	     (handler   (make-busy-timeout (if (args:get-arg "-override-timeout")
					       (string->number (args:get-arg "-override-timeout"))
					       136000))))
	(handle-exceptions
	 exn
	 (begin
	   (debug:print 0 "ERROR: problem accessing test db " testpath ", you probably should clean and re-run this test"
	   (debug:print 0 "ERROR: problem accessing test db " work-area ", you probably should clean and re-run this test"
			((condition-property-accessor 'exn 'message) exn))
	   #f)
	 (set! db (sqlite3:open-database dbpath)))
	(sqlite3:set-busy-handler! db handler)
	(if (not dbexists)
	    (begin
	      (sqlite3:execute db "PRAGMA synchronous = FULL;")
	      (debug:print-info 11 "Initialized test database " dbpath)
	      (db:testdb-initialize db)))
	;; (sqlite3:execute db "PRAGMA synchronous = 0;")
	(debug:print-info 11 "open-test-db END (sucessful)" testpath)
	(debug:print-info 11 "open-test-db END (sucessful)" work-area)
	;; now let's test that everything is correct
	(handle-exceptions
	 exn
	 (begin
	   (debug:print 0 "ERROR: problem accessing test db " testpath ", you probably should clean and re-run this test"
	   (debug:print 0 "ERROR: problem accessing test db " work-area ", you probably should clean and re-run this test"
			((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)
      (begin
	(debug:print-info 11 "open-test-db END (unsucessful)" testpath)
	(debug:print-info 11 "open-test-db END (unsucessful)" work-area)
	#f)))

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

(define (db:testdb-initialize db)
  (debug:print 11 "db:testdb-initialize START")
  (for-each
   (lambda (sqlcmd)

	(if (equal? hed field)
	    (vector-ref row n)
	    (if (null? tal) #f (loop (car tal)(cdr tal)(+ n 1)))))))

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

(define (db:get-run-name-from-id db run-id)
  (let ((res #f))
    (sqlite3:for-each-row
     (lambda (runname)
       (set! res runname))
     db
     "SELECT runname FROM runs WHERE id=?;"
     run-id)
    res))

(define (db:get-run-key-val db run-id key)
  (let ((res #f))
    (sqlite3:for-each-row
     (lambda (val)
       (set! res val))
     db 
     (conc "SELECT " (key:get-fieldname key) " FROM runs WHERE id=?;")
     run-id)
    res))

;; keys list to key1,key2,key3 ...
(define (runs:get-std-run-fields keys remfields)
  (let* ((header    (append (map key:get-fieldname keys)
			    remfields))
	 (keystr    (conc (keys->keystr keys) ","
			  (string-intersperse remfields ","))))

    (string-intersperse (map (lambda (patt)
			       (let ((wildtype (if (substring-index "%" patt) "LIKE" "GLOB")))
				 (conc fieldname " " wildtype " '" patt "'")))
			     (if (null? patts)
				 '("")
				 patts))
			comparator)))


;; register a test run with the db
(define (db:register-run db keys keyvallst runname state status user)
  (debug:print 3 "runs:register-run, keys: " keys " keyvallst: " keyvallst " runname: " runname " state: " state " status: " status " user: " user)
  (let* ((keystr    (keys->keystr keys))
	 (comma     (if (> (length keys) 0) "," ""))
	 (andstr    (if (> (length keys) 0) " AND " ""))
	 (valslots  (keys->valslots keys)) ;; ?,?,? ...
	 (keyvals   (map cadr keyvallst))
	 (allvals   (append (list runname state status user) keyvals))
	 (qryvals   (append (list runname) keyvals))
	 (key=?str  (string-intersperse (map (lambda (k)(conc (key:get-fieldname k) "=?")) keys) " AND ")))
    (debug:print 3 "keys: " keys " allvals: " allvals " keyvals: " keyvals)
    (debug:print 2 "NOTE: using target " (string-intersperse keyvals "/") " for this run")
    (if (and runname (null? (filter (lambda (x)(not x)) keyvals))) ;; there must be a better way to "apply and"
	(let ((res #f))
	  (apply sqlite3:execute db (conc "INSERT OR IGNORE INTO runs (runname,state,status,owner,event_time" comma keystr ") VALUES (?,?,?,?,strftime('%s','now')" comma valslots ");")
		 allvals)
	  (apply sqlite3:for-each-row 
	   (lambda (id)
	     (set! res id))
	   db
	   (let ((qry (conc "SELECT id FROM runs WHERE (runname=? " andstr key=?str ");")))
	     ;(debug:print 4 "qry: " qry) 
	     qry)
	   qryvals)
	  (sqlite3:execute db "UPDATE runs SET state=?,status=? WHERE id=?;" state status res)
	  res) 
	(begin
	  (debug:print 0 "ERROR: 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 ...
;;
(define (db:get-runs db runpatt count offset keypatts)

	  (hash-table-set! *target* run-id thekey)
	  thekey))))

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

(define (db:tests-register-test db run-id test-name item-path)
  (debug:print-info 11 "db:tests-register-test START db=" db ", run-id=" run-id ", test-name=" test-name ", item-path=\"" item-path "\"")
  (let ((item-paths (if (equal? item-path "")
			(list item-path)
			(list item-path ""))))
    (for-each 
     (lambda (pth)
       (sqlite3:execute db "INSERT OR IGNORE INTO tests (run_id,testname,event_time,item_path,state,status) VALUES (?,?,strftime('%s','now'),?,'NOT_STARTED','n/a');" 
			run-id 
			test-name
			pth))
     item-paths)
  (debug:print-info 11 "db:tests-register-test END db=" db ", run-id=" run-id ", test-name=" test-name ", item-path=\"" item-path "\"")
    #f))

;; states and statuses are lists, turn them into ("PASS","FAIL"...) and use NOT IN
;; i.e. these lists define what to NOT show.
;; states and statuses are required to be lists, empty is ok
;; not-in #t = above behaviour, #f = must match
(define (db:get-tests-for-run db run-id testpatt states statuses 
			      #!key (not-in #t)
			      (sort-by #f) ;; 'rundir 'event_time

     db 
     qry
     )
    (debug:print-info 11 "db:get-tests-for-run START run-ids=" run-ids ", testpatt=" testpatt ", states=" states ", statuses=" statuses ", not-in=" not-in ", sort-by=" sort-by)
    res))

;; this one is a bit broken BUG FIXME
(define (db:delete-test-step-records db test-id)
(define (db:delete-test-step-records db test-id #!key (work-area #f))
  ;; Breaking it into two queries for better file access interleaving
  (let* ((tdb (db:open-test-db-by-test-id db test-id)))
  (let* ((tdb (db:open-test-db-by-test-id db test-id work-area: work-area)))
    ;; test db's can go away - must check every time
    (if tdb
	(begin
	  (sqlite3:execute tdb "DELETE FROM test_steps;")
	  (sqlite3:execute tdb "DELETE FROM test_data;")
	  (sqlite3:finalize! tdb)))))

(define (db:delete-old-deleted-test-records db)
  (let ((targtime (- (current-seconds)(* 30 24 60 60)))) ;; one month in the past
    (sqlite3:execute db "DELETE FROM tests WHERE state='DELETED' AND event_time<?;" targtime)))

;; set tests with state currstate and status currstatus to newstate and newstatus
;; use currstate = #f and or currstatus = #f to apply to any state or status respectively
;; WARNING: SQL injection risk
;; WARNING: SQL injection risk. NB// See new but not yet used "faster" version below
;;
(define (db:set-tests-state-status db run-id testnames currstate currstatus newstate newstatus)
  (for-each (lambda (testname)
	      (let ((qry (conc "UPDATE tests SET state=?,status=? WHERE "
			       (if currstate  (conc "state='" currstate "' AND ") "")
			       (if currstatus (conc "status='" currstatus "' AND ") "")
			       " run_id=? AND testname=? AND NOT (item_path='' AND testname in (SELECT DISTINCT testname FROM tests WHERE testname=? AND item_path != ''));")))
		;;(debug:print 0 "QRY: " qry)
		(sqlite3:execute db qry run-id newstate newstatus testname testname)))
	    testnames))


(define (cdb:set-tests-state-status-faster serverdat run-id testnames currstate currstatus newstate newstatus)
  ;; Convert #f to wildcard %
  (if (null? testnames)
      #t
      (let ((currstate  (if currstate currstate "%"))
	    (currstatus (if currstatus currstatus "%")))
	(let loop ((hed (car testnames))
		   (tal (cdr testnames))
		   (thr '()))
	  (let ((th1 (if newstate  (create-thread (cbd:client-call serverdat 'update-test-state  #t *default-numtries* newstate  currstate  run-id testname testname)) #f))
		(th2 (if newstatus (create-thread (cbd:client-call serverdat 'update-test-status #t *default-numtries* newstatus currstatus run-id testname testname)) #f)))
	    (thread-start! th1)
	    (thread-start! th2)
	    (if (null? tal)
		(loop (car tal)(cdr tal)(cons th1 (cons th2 thr)))
		(for-each
		 (lambda (th)
		   (if th (thread-join! th)))
		 thr)))))))

(define (cdb:delete-tests-in-state serverdat run-id state)
  (cdb:client-call serverdat 'delete-tests-in-state #t *default-numtries* run-id state))

(define (cdb:tests-update-cpuload-diskfree serverdat test-id cpuload diskfree)
  (cdb:client-call serverdat 'update-cpuload-diskfree #t *default-numtries* cpuload diskfree test-id))

(define (cdb:tests-update-run-duration serverdat test-id minutes)
  (cdb:client-call serverdat 'update-run-duration #t *default-numtries* minutes test-id))

(define (cdb:tests-update-uname-host serverdat test-id uname hostname)
  (cdb:client-call serverdat 'update-uname-host #t *default-numtries* test-id uname hostname))

;; speed up for common cases with a little logic
(define (db:test-set-state-status-by-id db test-id newstate newstatus newcomment)
  (cond
   ((and newstate newstatus newcomment)
    (sqlite3:exectute db "UPDATE tests SET state=?,status=?,comment=? WHERE id=?;" newstate newstatus test-id))
   ((and newstate newstatus)
    (sqlite3:execute db "UPDATE tests SET state=?,status=? WHERE id=?;" newstate newstatus test-id))

     run-id testname item-path)
    res))

(define db:get-test-id db:get-test-id-not-cached)

;; given a test-info record, patch in the latest data from the testdat.db file
;; found in the test run directory
;;
;; NOT USED
;;
(define (db:patch-tdb-data-into-test-info db test-id res)
  (let ((tdb (db:open-test-db-by-test-id db test-id)))
(define (db:patch-tdb-data-into-test-info db test-id res #!key (work-area #f))
  (let ((tdb (db:open-test-db-by-test-id db test-id work-area: work-area)))
    ;; get state and status from megatest.db in real time
    ;; other fields that perhaps should be updated:
    ;;   fail_count
    ;;   pass_count
    ;;   final_logf
    (sqlite3:for-each-row
     (lambda (state status final_logf)

;; QUEUE UP META, TEST STATUS AND STEPS REMOTE ACCESS
;;======================================================================

;; NOTE: Can remove the regex and base64 encoding for zmq
(define (db:obj->string obj)
  (case *transport-type*
    ((fs) obj)
	((http)
    ((http)
     (string-substitute
       (regexp "=") "_"
         (base64:base64-encode (with-output-to-string (lambda ()(serialize obj))))
        #t))
      (regexp "=") "_"
      (base64:base64-encode (with-output-to-string (lambda ()(serialize obj))))
      #t))
    ((zmq)(with-output-to-string (lambda ()(serialize obj))))
    (else obj)))

(define (db:string->obj msg)
  (case *transport-type*
   ((fs) msg)
   ((http)

	    (query-sig   (message-digest-string (md5-primitive) (conc qtype immediate params)))
	    (zdat        (db:obj->string (vector client-sig qtype immediate query-sig params (current-seconds))))) ;; (with-output-to-string (lambda ()(serialize params))))
       (debug:print-info 11 "zdat=" zdat)
       (let* ((res  #f)
	      (rawdat      (http-transport:client-send-receive serverdat zdat))
	      (tmp         #f))
	 (debug:print-info 11 "Sent " zdat ", received " rawdat)
	 (if rawdat
	     (begin
	 (set! tmp (db:string->obj rawdat))
	 (vector-ref tmp 2))))
	       (set! tmp (db:string->obj rawdat))
	       (vector-ref tmp 2))
	     (begin
	       (debug:print 0 "ERROR: Communication with the server failed. Exiting if possible")
	       (exit 1))))))
    ((zmq)
     (handle-exceptions
      exn
      (begin
	(debug:print-info 0 "cdb:client-call timeout or error. Trying again in 5 seconds")
	(thread-sleep! 5) 
	(if (> numretries 0)(apply cdb:client-call serverdat qtype immediate (- numretries 1) params)))
      (let* ((push-socket (vector-ref serverdat 0))
	     (sub-socket  (vector-ref serverdat 1))
	     (client-sig  (client:get-signature))
	     (query-sig   (message-digest-string (md5-primitive) (conc qtype immediate params)))
	     (zdat        (db:obj->string (vector client-sig qtype immediate query-sig params (current-seconds)))) ;; (with-output-to-string (lambda ()(serialize params))))

			       ;; this should match (client:get-signature)
			       ;; we will need to process "all" messages here some day
			       (receive-message* sub-socket)
			       ;; now get the actual message
			       (let ((myres (db:string->obj (receive-message* sub-socket))))
				 (if (equal? query-sig (vector-ref myres 1))
				     (set! res (vector-ref myres 2))
				     (loop))))))
	     (timeout (lambda ()
			(let loop ((n numretries))
			  (thread-sleep! 15)
			  (if (not res)
			      (if (> numretries 0)
				  (begin
				    (debug:print 2 "WARNING: no reply to query " params ", trying resend")
				    (debug:print-info 11 "re-sending message")
				    (send-message push-socket zdat)
				    (debug:print-info 11 "message re-sent")
				    (loop (- n 1)))
				  ;; (apply cdb:client-call *runremote* qtype immediate (- numretries 1) params))
				  (begin
				    (debug:print 0 "ERROR: cdb:client-call timed out " params ", exiting.")
				    (exit 5))))))))
				     (loop)))))))
	    ;; (timeout (lambda ()
	    ;;     	(let loop ((n numretries))
	    ;;     	  (thread-sleep! 15)
	    ;;     	  (if (not res)
	    ;;     	      (if (> numretries 0)
	    ;;     		  (begin
	    ;;     		    (debug:print 2 "WARNING: no reply to query " params ", trying resend")
	    ;;     		    (debug:print-info 11 "re-sending message")
	    ;;     		    (send-message push-socket zdat)
	    ;;     		    (debug:print-info 11 "message re-sent")
	    ;;     		    (loop (- n 1)))
	    ;;     		  ;; (apply cdb:client-call *runremote* qtype immediate (- numretries 1) params))
	    ;;     		  (begin
	    ;;     		    (debug:print 0 "ERROR: cdb:client-call timed out " params ", exiting.")
	    ;;     		    (exit 5))))))))
	(debug:print-info 11 "Starting threads")
	(let ((th1 (make-thread send-receive "send receive"))
	      (th2 (make-thread timeout      "timeout")))
	      ;; (th2 (make-thread timeout      "timeout"))
	      )
	  (thread-start! th1)
	  (thread-start! th2)
	  ;; (thread-start! th2)
	  (thread-join!  th1)
	  (debug:print-info 11 "cdb:client-call returning res=" res)
	  res))))))
  
(define (cdb:set-verbosity serverdat val)
  (cdb:client-call serverdat 'set-verbosity #f *default-numtries* val))

(define (cdb:test-rollup-test_data-pass-fail serverdat test-id)
  (cdb:client-call serverdat 'test_data-pf-rollup #t *default-numtries* test-id test-id test-id test-id))

(define (cdb:pass-fail-counts serverdat test-id fail-count pass-count)
  (cdb:client-call serverdat 'pass-fail-counts #t *default-numtries* fail-count pass-count test-id))

(define (cdb:tests-register-test serverdat run-id test-name item-path)
  (let ((item-paths (if (equal? item-path "")
			(list item-path)
			(list item-path ""))))
    (cdb:client-call serverdat 'register-test #t *default-numtries* run-id test-name item-path)))
  (cdb:client-call serverdat 'register-test #t *default-numtries* run-id test-name item-path))

(define (cdb:flush-queue serverdat)
  (cdb:client-call serverdat 'flush #f *default-numtries*))

(define (cdb:kill-server serverdat)
  (cdb:client-call serverdat 'killserver #t *default-numtries*))

	   (debug:print 2 "Found path: " path)
	   (debug:print 2 "No such path: " path)))
     db
     "SELECT rundir,final_logf FROM tests WHERE run_id=? AND testname=? AND item_path='';"
     run-id test-name)
    res))

;;======================================================================
;; A G R E G A T E D   T R A N S A C T I O N   D B   W R I T E S 
;;======================================================================

(define db:queries 
  (list '(register-test          "INSERT OR IGNORE INTO tests (run_id,testname,event_time,item_path,state,status) VALUES (?,?,strftime('%s','now'),?,'NOT_STARTED','n/a');")
	'(state-status           "UPDATE tests SET state=?,status=? WHERE id=?;")
	'(state-status-msg       "UPDATE tests SET state=?,status=?,comment=? WHERE id=?;")
	'(pass-fail-counts       "UPDATE tests SET fail_count=?,pass_count=? WHERE id=?;")
	;; test_data-pf-rollup is used to set a tests PASS/FAIL based on the pass/fail info from the steps
	'(test_data-pf-rollup    "UPDATE tests
                                    SET status=CASE WHEN (SELECT fail_count FROM tests WHERE id=?) > 0 
                                      THEN 'FAIL'
                                    WHEN (SELECT pass_count FROM tests WHERE id=?) > 0 AND 
                                      (SELECT status FROM tests WHERE id=?) NOT IN ('WARN','FAIL')
                                    THEN 'PASS'
                                    ELSE status
                                    END WHERE id=?;")
	'(test-set-log            "UPDATE tests SET final_logf=? WHERE id=?;")
	'(test-set-rundir-by-test-id "UPDATE tests SET rundir=? WHERE id=?")
	'(test-set-rundir         "UPDATE tests SET rundir=? WHERE run_id=? AND testname=? AND item_path=?;")
	'(delete-tests-in-state   "DELETE FROM tests WHERE state=? AND run_id=?;")
	'(tests:test-set-toplog   "UPDATE tests SET final_logf=? WHERE run_id=? AND testname=? AND item_path='';")
	'(update-cpuload-diskfree "UPDATE tests SET cpuload=?,diskfree=? WHERE id=?;")
	'(update-run-duration     "UPDATE tests SET run_duration=? WHERE id=?;")
	'(update-uname-host       "UPDATE tests SET uname=?,host=? WHERE id=?;")
	'(update-test-state       "UPDATE tests SET state=? WHERE state=? AND run_id=? AND testname=? AND NOT (item_path='' AND testname IN (SELECT DISTINCT testname FROM tests WHERE testname=? AND item_path != ''));")
	'(update-test-status      "UPDATE tests SET status=? WHERE status like ? AND run_id=? AND testname=? AND NOT (item_path='' AND testname IN (SELECT DISTINCT testname FROM tests WHERE testname=? AND item_path != ''));")
    ))

;; do not run these as part of the transaction
(define db:special-queries   '(rollup-tests-pass-fail
			       db:roll-up-pass-fail-counts
                               login
                               immediate

      (mutex-unlock! *completed-mutex*)
      (if (and (not got-it)
	       (< (current-seconds) timeout))
	  (begin
	    (thread-sleep! 0.01)
	    (loop))))
    (set! *number-of-writes*   (+ *number-of-writes*   1))
    (set! *writes-total-delay* (+ *writes-total-delay* 1))
    (set! *writes-total-delay* (+ *writes-total-delay* (- (current-milliseconds) start-time)))
    got-it))
	  
(define (db:process-queue-item db item)
  (let* ((stmt-key       (cdb:packet-get-qtype item))
	 (qry-sig        (cdb:packet-get-query-sig item))
	 (return-address (cdb:packet-get-client-sig item))
	 (params         (cdb:packet-get-params item))

       (set! res (cons (vector id itempath state status run_duration logf comment) res)))
     db
     "SELECT id,item_path,state,status,run_duration,final_logf,comment FROM tests WHERE run_id=? AND testname=? AND item_path != '';"
     run-id test-name)
    res))

;; Rollup the pass/fail counts from itemized tests into fail_count and pass_count
;; NOTE: Is this duplicating (db:test-data-rollup db test-id status) ????
(define (db:roll-up-pass-fail-counts db run-id test-name item-path status)
  ;; (cdb:flush-queue *runremote*)
  (if (and (not (equal? item-path ""))
	   (member status '("PASS" "WARN" "FAIL" "WAIVED" "RUNNING" "CHECK" "SKIP")))
      (begin
	(sqlite3:execute 
	 db
	 "UPDATE tests 
             SET fail_count=(SELECT count(id) FROM tests WHERE run_id=? AND testname=? AND item_path != '' AND status='FAIL'),
             SET fail_count=(SELECT count(id) FROM tests WHERE run_id=? AND testname=? AND item_path != '' AND status IN ('FAIL','CHECK')),
                 pass_count=(SELECT count(id) FROM tests WHERE run_id=? AND testname=? AND item_path != '' AND status IN ('PASS','WARN','WAIVED'))
             WHERE run_id=? AND testname=? AND item_path='';"
	 run-id test-name run-id test-name run-id test-name)
        ;; (thread-sleep! 0.1) ;; give other processes a chance here, no, better to be done ASAP?
	(if (equal? status "RUNNING") ;; running takes priority over all other states, force the test state to RUNNING
	    (sqlite3:execute db "UPDATE tests SET state=? WHERE run_id=? AND testname=? AND item_path='';" "RUNNING" run-id test-name)
	    (sqlite3:execute
	     db
	     "UPDATE tests
                       SET state=CASE 
                                   WHEN (SELECT count(id) FROM tests 
                                                WHERE run_id=? AND testname=?
                                                     AND item_path != '' 
                                                     AND state in ('RUNNING','NOT_STARTED')) > 0 THEN 'RUNNING'
                                   ELSE 'COMPLETED' END,
                                      status=CASE 
                                            WHEN fail_count > 0 THEN 'FAIL' 
                                            WHEN pass_count > 0 AND fail_count=0 THEN 'PASS' 
                                            ELSE 'UNKNOWN' END
                            status=CASE 
                                  WHEN fail_count > 0 THEN 'FAIL' 
                                  WHEN pass_count > 0 AND fail_count=0 THEN 'PASS' 
                                  WHEN (SELECT count(id) FROM tests
                                         WHERE run_id=? AND testname=?
                                              AND item_path != ''
                                              AND status = 'SKIP') > 0 THEN 'SKIP'
                                  ELSE 'UNKNOWN' END
                       WHERE run_id=? AND testname=? AND item_path='';"
	     run-id test-name run-id test-name))
	     run-id test-name run-id test-name run-id test-name))
	#f)
      #f))

;;======================================================================
;; Tests meta data
;;======================================================================

(define (db:testmeta-update-field db testname field value)
  (sqlite3:execute db (conc "UPDATE test_meta SET " field "=? WHERE testname=?;") value testname))

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

(define (db:csv->test-data db test-id csvdata)
(define (db:csv->test-data db test-id csvdata #!key (work-area #f))
  (debug:print 4 "test-id " test-id ", csvdata: " csvdata)
  (let ((tdb     (db:open-test-db-by-test-id db test-id)))
  (let ((tdb     (db:open-test-db-by-test-id db test-id work-area: work-area)))
    (if tdb
	(let ((csvlist (csv->list (make-csv-reader
				   (open-input-string csvdata)
				   '((strip-leading-whitespace? #t)
				     (strip-trailing-whitespace? #t)) )))) ;; (csv->list csvdata)))
	  (for-each 
	   (lambda (csvrow)

			       ((<)  (if (<  value expected) "pass" "fail"))
			       ((>=) (if (>= value expected) "pass" "fail"))
			       ((<=) (if (<= value expected) "pass" "fail"))
			       (else (conc "ERROR: bad tol comparator " tol))))))
	       (debug:print 4 "AFTER2: category: " category " variable: " variable " value: " value 
			    ", expected: " expected " tol: " tol " units: " units " status: " status " comment: " comment)
	       (sqlite3:execute tdb "INSERT OR REPLACE INTO test_data (test_id,category,variable,value,expected,tol,units,comment,status,type) VALUES (?,?,?,?,?,?,?,?,?,?);"
				test-id category variable value expected tol units (if comment comment "") status type)
	       (sqlite3:finalize! tdb)))
				test-id category variable value expected tol units (if comment comment "") status type)))
	   csvlist)
	  (sqlite3:finalize! tdb)))))
	   csvlist)))))

;; get a list of test_data records matching categorypatt
(define (db:read-test-data db test-id categorypatt)
  (let ((tdb  (db:open-test-db-by-test-id db test-id)))
(define (db:read-test-data db test-id categorypatt #!key (work-area #f))
  (let ((tdb  (db:open-test-db-by-test-id db test-id work-area: work-area)))
    (if tdb
	(let ((res '()))
	  (sqlite3:for-each-row 
	   (lambda (id test_id category variable value expected tol units comment status type)
	     (set! res (cons (vector id test_id category variable value expected tol units comment status type) res)))
	   tdb
	   "SELECT id,test_id,category,variable,value,expected,tol,units,comment,status,type FROM test_data WHERE test_id=? AND category LIKE ? ORDER BY category,variable;" test-id categorypatt)
	  (sqlite3:finalize! tdb)
	  (reverse res))
	'())))

;; NOTE: Run this local with #f for db !!!
(define (db:load-test-data db test-id)
(define (db:load-test-data db test-id #!key (work-area #f))
  (let loop ((lin (read-line)))
    (if (not (eof-object? lin))
	(begin
	  (debug:print 4 lin)
	  (db:csv->test-data db test-id lin)
	  (db:csv->test-data db test-id lin work-area: work-area)
	  (loop (read-line)))))
  ;; roll up the current results.
  ;; FIXME: Add the status to 
  (db:test-data-rollup db test-id #f))
  (db:test-data-rollup db test-id #f work-area: work-area))

;; WARNING: Do NOT call this for the parent test on an iterated test
;; Roll up test_data pass/fail results
;; look at the test_data status field, 
;;    if all are pass (any case) and the test status is PASS or NULL or '' then set test status to PASS.
;;    if one or more are fail (any case) then set test status to PASS, non "pass" or "fail" are ignored
(define (db:test-data-rollup db test-id status)
  (let ((tdb (db:open-test-db-by-test-id db test-id))
(define (db:test-data-rollup db test-id status #!key (work-area #f))
  (let ((tdb (db:open-test-db-by-test-id db test-id work-area: work-area))
	(fail-count 0)
	(pass-count 0))
    (if tdb
	(begin
	  (sqlite3:for-each-row
	   (lambda (fcount pcount)
	     (set! fail-count fcount)
	     (set! pass-count pcount))
	   tdb 
	   "SELECT (SELECT count(id) FROM test_data WHERE test_id=? AND status like 'fail') AS fail_count,
                   (SELECT count(id) FROM test_data WHERE test_id=? AND status like 'pass') AS pass_count;"
	   test-id test-id)
	  (sqlite3:finalize! tdb)

	  ;; Now rollup the counts to the central megatest.db
	  (cdb:pass-fail-counts *runremote* test-id fail-count pass-count)
	  ;; (sqlite3:execute db "UPDATE tests SET fail_count=?,pass_count=? WHERE id=?;" 
	  ;;                     fail-count pass-count test-id)

	  ;; The flush is not needed with the transaction based write agregation enabled. Remove these commented lines
	  ;; next time you read this!
	  ;;
	  (cdb:flush-queue *runremote*)
	  ;; (cdb:flush-queue *runremote*)
	  ;; (thread-sleep! 1) ;; play nice with the queue by ensuring the rollup is at least 10ms later than the set
	  
	  ;; if the test is not FAIL then set status based on the fail and pass counts.
	  (cdb:test-rollup-test_data-pass-fail *runremote* test-id)
	  ;; (sqlite3:execute
	  ;;  db   ;;; NOTE: Should this be WARN,FAIL? A WARN is not a FAIL????? BUG FIXME
	  ;;  "UPDATE tests

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

(define (db:step-get-time-as-string vec)
  (seconds->time-string (db:step-get-event_time vec)))

;; db-get-test-steps-for-run
(define (db:get-steps-for-test db test-id)
  (let* ((tdb (db:open-test-db-by-test-id db test-id))
(define (db:get-steps-for-test db test-id #!key (work-area #f))
  (let* ((tdb (db:open-test-db-by-test-id db test-id work-area: work-area))
	 (res '()))
    (if tdb
	(begin
	  (sqlite3:for-each-row 
	   (lambda (id test-id stepname state status event-time logfile)
	     (set! res (cons (vector id test-id stepname state status event-time (if (string? logfile) logfile "")) res)))
	   tdb
	   "SELECT id,test_id,stepname,state,status,event_time,logfile FROM test_steps WHERE test_id=? ORDER BY id ASC;" ;; event_time DESC,id ASC;
	   test-id)
	  (sqlite3:finalize! tdb)
	  (reverse res))
	'())))

;; get a pretty table to summarize steps
;;
(define (db:get-steps-table db test-id)
  (let ((steps   (db:get-steps-for-test db test-id)))
(define (db:get-steps-table db test-id #!key (work-area #f))
  (let ((steps   (db:get-steps-for-test db test-id work-area: work-area)))
    ;; organise the steps for better readability
    (let ((res (make-hash-table)))
      (for-each 
       (lambda (step)
	 (debug:print 6 "step=" step)
	 (let ((record (hash-table-ref/default 
			res 

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

;; get a pretty table to summarize steps
;;
(define (db:get-steps-table-list db test-id)
  (let ((steps   (db:get-steps-for-test db test-id)))
(define (db:get-steps-table-list db test-id #!key (work-area #f))
  (let ((steps   (db:get-steps-for-test db test-id work-area: work-area)))
    ;; organise the steps for better readability
    (let ((res (make-hash-table)))
      (for-each 
       (lambda (step)
	 (debug:print 6 "step=" step)
	 (let ((record (hash-table-ref/default 
			res 

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

(define (db:get-compressed-steps test-id)
  (let* ((comprsteps (open-run-close db:get-steps-table #f test-id)))
    (map (lambda (x)
	   ;; take advantage of the \n on time->string
	   (vector
	    (vector-ref x 0)
	    (let ((s (vector-ref x 1)))
	      (if (number? s)(seconds->time-string s) s))
	    (let ((s (vector-ref x 2)))
	      (if (number? s)(seconds->time-string s) s))
	    (vector-ref x 3)    ;; status
	    (vector-ref x 4)
	    (vector-ref x 5)))  ;; time delta
	 (sort (hash-table-values comprsteps)
	       (lambda (a b)
		 (let ((time-a (vector-ref a 1))
		       (time-b (vector-ref b 1)))
		   (if (and (number? time-a)(number? time-b))
		       (if (< time-a time-b)
			   #t
			   (if (eq? time-a time-b)
			       (string<? (conc (vector-ref a 2))
					 (conc (vector-ref b 2)))
			       #f))
		       (string<? (conc time-a)(conc time-b)))))))))
(define (db:get-compressed-steps test-id #!key (work-area #f))
  (if (or (not work-area)
	  (file-exists? (conc work-area "/testdat.db")))
      (let* ((comprsteps (open-run-close db:get-steps-table #f test-id work-area: work-area)))
	(map (lambda (x)
	       ;; take advantage of the \n on time->string
	       (vector
		(vector-ref x 0)
		(let ((s (vector-ref x 1)))
		  (if (number? s)(seconds->time-string s) s))
		(let ((s (vector-ref x 2)))
		  (if (number? s)(seconds->time-string s) s))
		(vector-ref x 3)    ;; status
		(vector-ref x 4)
		(vector-ref x 5)))  ;; time delta
	     (sort (hash-table-values comprsteps)
		   (lambda (a b)
		     (let ((time-a (vector-ref a 1))
			   (time-b (vector-ref b 1)))
		       (if (and (number? time-a)(number? time-b))
			   (if (< time-a time-b)
			       #t
			       (if (eq? time-a time-b)
				   (string<? (conc (vector-ref a 2))
					     (conc (vector-ref b 2)))
				   #f))
			   (string<? (conc time-a)(conc time-b))))))))
      '()))

;;======================================================================
;; M I S C   M A N A G E M E N T   I T E M S 
;;======================================================================

;; the new prereqs calculation, looks also at itempath if specified
;; all prereqs must be met:
;;    if prereq test with itempath='' is COMPLETED and PASS, WARN, CHECK, or WAIVED then prereq is met
;;    if prereq test with itempath=ref-item-path and COMPLETED with PASS, WARN, CHECK, or WAIVED then prereq is met
;;
;; Note: do not convert to remote as it calls remote under the hood
;; Note: mode 'normal means that tests must be COMPLETED and ok (i.e. PASS, WARN, CHECK, SKIP or WAIVED)
;;       mode 'toplevel means that tests must be COMPLETED only
;;       mode 'itemmatch means that tests items must be COMPLETED and (PASS|WARN|WAIVED|CHECK) [[ NB// NOT IMPLEMENTED YET ]]
;; 
(define (db:get-prereqs-not-met db run-id waitons ref-item-path #!key (mode 'normal))
(define (db:get-prereqs-not-met run-id waitons ref-item-path #!key (mode 'normal))
  (if (or (not waitons)
	  (null? waitons))
      '()
      (let* ((unmet-pre-reqs '())
	     (result         '()))
	(for-each 
	 (lambda (waitontest-name)
	   ;; by getting the tests with matching name we are looking only at the matching test 
	   ;; and related sub items
	   (let ((tests             (db:get-tests-for-run db run-id waitontest-name '() '()))
	   (let ((tests             (cdb:remote-run db:get-tests-for-run #f run-id waitontest-name '() '()))
		 (ever-seen         #f)
		 (parent-waiton-met #f)
		 (item-waiton-met   #f))
	     (for-each 
	      (lambda (test)
		;; (if (equal? waitontest-name (db:test-get-testname test)) ;; by defintion this had better be true ...
		(let* ((state             (db:test-get-state test))

	     ;; if the test is not found then clearly the waiton is not met...
	     ;; (if (not ever-seen)(set! result (cons waitontest-name result)))))
	     (if (not ever-seen)
		 (set! result (append (if (null? tests)(list waitontest-name) tests) result)))))
	 waitons)
	(delete-duplicates result))))

(define (db:teststep-set-status! db test-id teststep-name state-in status-in comment logfile)
(define (db:teststep-set-status! db test-id teststep-name state-in status-in comment logfile #!key (work-area #f))
  (debug:print 4 "test-id: " test-id " teststep-name: " teststep-name)
  ;;                 db:open-test-db-by-test-id does cdb:remote-run
  (let* ((tdb       (db:open-test-db-by-test-id db test-id))
  (let* ((tdb       (db:open-test-db-by-test-id db test-id work-area: work-area))
	 (state     (items:check-valid-items "state" state-in))
	 (status    (items:check-valid-items "status" status-in)))
    (if (or (not state)(not status))
	(debug:print 3 "WARNING: Invalid " (if status "status" "state")
		     " value \"" (if status state-in status-in) "\", update your validvalues section in megatest.config"))
    (if tdb
	(begin

;;  PURPOSE.

(require-extension (srfi 18) extras tcp s11n)

(use sqlite3 srfi-1 posix regex regex-case srfi-69 hostinfo md5 message-digest)
(import (prefix sqlite3 sqlite3:))

(use spiffy uri-common intarweb http-client spiffy-request-vars)
(use spiffy uri-common intarweb http-client spiffy-request-vars  uri-common intarweb)

;; Configurations for server
(tcp-buffer-size 2048)
(max-connections 2048) 

(declare (unit http-transport))

(declare (uses common))
(declare (uses db))
(declare (uses tests))
(declare (uses tasks)) ;; tasks are where stuff is maintained about what is running.
(declare (uses server))
(declare (uses daemon))

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

(define (http-transport:make-server-url hostport)
  (if (not hostport)
      #f

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

;; Call this to start the actual server
;;

(define *db:process-queue-mutex* (make-mutex))

(define (server:get-best-guess-address hostname)
  (let ((res #f))
    (for-each 
     (lambda (adr)
       (if (not (eq? (u8vector-ref adr 0) 127))
	   (set! res adr)))
     (vector->list (hostinfo-addresses (hostname->hostinfo hostname))))
    (string-intersperse 
     (map number->string
	  (u8vector->list
	   (if res res (hostname->ip hostname)))) ".")))

(define (http-transport:run hostn)
  (debug:print 2 "Attempting to start the server ...")
  (if (not *toppath*)
      (if (not (setup-for-run))
	  (begin
	    (debug:print 0 "ERROR: cannot find megatest.config, cannot start server, exiting")
	    (exit))))
  (let* (;; (iface           (if (string=? "-" hostn)
	 ;;        	      #f ;; (get-host-name) 
	 ;;        	      hostn))
	 (db              #f) ;;        (open-db)) ;; we don't want the server to be opening and closing the db unnecesarily
	 (hostname        (get-host-name))
	 (ipaddrstr       (let ((ipstr (if (string=? "-" hostn)
					   (string-intersperse (map number->string (u8vector->list (hostname->ip hostname))) ".")
					   ;; (string-intersperse (map number->string (u8vector->list (hostname->ip hostname))) ".")
					   (server:get-best-guess-address hostname)
					   #f)))
			    (if ipstr ipstr hostn))) ;; hostname)))
	 (start-port    (if (and (args:get-arg "-port")
				 (string->number (args:get-arg "-port")))
			    (string->number (args:get-arg "-port"))
			    (if (and (config-lookup  *configdat* "server" "port")
				     (string->number (config-lookup  *configdat* "server" "port")))

(define (http-transport:try-start-server ipaddrstr portnum)
  (handle-exceptions
   exn
   (begin
     (print-error-message exn)
     (if (< portnum 9000)
	 (begin 
	   (print "WARNING: failed to start on portnum: " portnum ", trying next port")
	   (debug:print 0 "WARNING: failed to start on portnum: " portnum ", trying next port")
	   (thread-sleep! 0.1)
	   ;; (open-run-close tasks:remove-server-records tasks:open-db)
	   (open-run-close tasks:server-delete tasks:open-db ipaddrstr portnum)
	   (http-transport:try-start-server ipaddrstr (+ portnum 1)))
	 (print "ERROR: Tried and tried but could not start the server")))
   ;; any error in following steps will result in a retry
   (set! *runremote* (list ipaddrstr portnum))
   ;; (open-run-close tasks:remove-server-records tasks:open-db)
   (open-run-close tasks:server-register 
		   tasks:open-db 
		   (current-process-id)
		   ipaddrstr portnum 0 'live 'http)
   (print "INFO: Trying to start server on " ipaddrstr ":" portnum)
		   ipaddrstr portnum 0 'startup 'http)
   (debug:print 1 "INFO: Trying to start server on " ipaddrstr ":" portnum)
   ;; This starts the spiffy server
   ;; NEED WAY TO SET IP TO #f TO BIND ALL
   (start-server port: portnum)
   (start-server bind-address: ipaddrstr port: portnum)
   (open-run-close tasks:server-delete tasks:open-db ipaddrstr portnum)
   (print "INFO: server has been stopped")))
   (debug:print 1 "INFO: server has been stopped")))

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

;;======================================================================
;; C L I E N T S
;;======================================================================

(define *http-mutex* (make-mutex))

;; (system "megatest -list-servers | grep alive || megatest -server - -daemonize && sleep 4")

;; <html>
;; <head></head>
;; <body>1 Hello, world! Goodbye Dolly</body></html>
;; Send msg to serverdat and receive result
(define (http-transport:client-send-receive serverdat msg)
  (let* ((url        (http-transport:make-server-url serverdat))
	 (fullurl    (conc url "/ctrl")) ;; (conc url "/?dat=" msg)))
	 (numretries 0))     
(define (http-transport:client-send-receive serverdat msg #!key (numretries 10))
  (let* (;; (url        (http-transport:make-server-url serverdat))
	 (fullurl    (caddr serverdat)) ;; (conc url "/ctrl")) ;; (conc url "/?dat=" msg)))
	 (res        #f))
    (handle-exceptions
     exn
     (begin
       (print "ERROR IN http-transport:client-send-receive " ((condition-property-accessor 'exn 'message) exn))
       (thread-sleep! 2)
     (if (< numretries 200)
	 (http-transport:client-send-receive serverdat msg))
       (if (> numretries 0)
	   (http-transport:client-send-receive serverdat msg numretries: (- numretries 1))))
     (begin
       (debug:print-info 11 "fullurl=" fullurl "\n")
       ;; set up the http-client here
       (max-retry-attempts 100)
       (max-retry-attempts 5)
       ;; consider all requests indempotent
       (retry-request? (lambda (request)
			 (thread-sleep! (/ (if (> numretries 100) 100 numretries) 10))
			 (set! numretries (+ numretries 1))
			 #t))
			 #t))   ;;  		 (thread-sleep! (/ (if (> numretries 100) 100 numretries) 10))
       ;; (set! numretries (- numretries 1))
       ;;  		 #t))
       ;; send the data and get the response
       ;; extract the needed info from the http data and 
       ;; process and return it.
       (let* ((send-recieve (lambda ()
			      (mutex-lock! *http-mutex*)
       (let* ((res   (with-input-from-request fullurl 
					      ;; #f
			      (set! res (with-input-from-request 
					 fullurl 
					      ;; msg 
					      (list (cons 'dat msg)) 
					      read-string)))
					 (list (cons 'dat msg)) 
					 read-string))
			      (close-all-connections!) 
			      (mutex-unlock! *http-mutex*)))
	      (time-out     (lambda ()
			      (thread-sleep! 5)
			      (if (not res)
				  (begin
				    (debug:print 0 "WARNING: communication with the server timed out.")
				    (mutex-unlock! *http-mutex*)
				    (http-transport:client-send-receive serverdat msg numretries: (- numretries 1))
				    (if (< numtries 3) ;; on last try just exit
					(begin
					  (debug:print 0 "ERROR: communication with the server timed out. Giving up.")
					  (exit 1)))))))
	      (th1 (make-thread send-recieve "with-input-from-request"))
	      (th2 (make-thread time-out     "time out")))
	 (thread-start! th1)
	 (thread-start! th2)
	 (thread-join! th1)
	 (thread-terminate! th2)
	 (debug:print-info 11 "got res=" res)
	 (let ((match (string-search (regexp "<body>(.*)<.body>") res)))
	   (debug:print-info 11 "match=" match)
	   (let ((final (cadr match)))
	     (debug:print-info 11 "final=" final)
	     final)))))))

(define (http-transport:client-connect iface port)
  (let* ((login-res   #f)
	 (uri-dat     (make-request method: 'POST uri: (uri-reference (conc "http://" iface ":" port "/ctrl"))))
	 (serverdat   (list iface port)))
	 (serverdat   (list iface port uri-dat)))
    (set! login-res (client:login serverdat))
    (if (and (not (null? login-res))
	     (car login-res))
	(begin
	  (debug:print-info 2 "Logged in and connected to " iface ":" port)
	  (set! *runremote* serverdat)
	  serverdat)
	(begin
	  (debug:print-info 2 "Failed to login or connect to " iface ":" port)
	  (set! *runremote* #f)
	  #f))))
	  (debug:print-info 0 "ERROR: Failed to login or connect to " iface ":" port)
	  (exit 1)))))
;; 	  (set! *runremote* #f)
;; 	  (set! *transport-type* 'fs)
;; 	  #f))))


;; run http-transport:keep-running in a parallel thread to monitor that the db is being 
;; used and to shutdown after sometime if it is not.
;;
(define (http-transport:keep-running)
  ;; if none running or if > 20 seconds since 
  ;; server last used then start shutdown
  ;; This thread waits for the server to come alive
  (let* ((server-info (let loop ()
                        (let ((sdat #f))
                          (mutex-lock! *heartbeat-mutex*)
                          (set! sdat *runremote*)
                          (mutex-unlock! *heartbeat-mutex*)
                          (if sdat sdat
                          (if sdat
			      sdat
                              (begin
                                (sleep 4)
                                (loop))))))
         (iface       (car server-info))
         (port        (cadr server-info))
         (last-access 0)
	 (tdb         (tasks:open-db))
	 (spid        ;;(open-run-close tasks:server-get-server-id tasks:open-db #f iface port #f))
	 (spid        (tasks:server-get-server-id tdb #f iface port #f))
	   (tasks:server-get-server-id tdb #f iface port #f))
	 (server-timeout (let ((tmo (config-lookup  *configdat* "server" "timeout")))
			   (if (and (string? tmo)
				    (string->number tmo))
			       (* 60 60 (string->number tmo))
			       ;; default to three days
			       (* 3 24 60)))))
			       (* 3 24 60 60)))))
    (debug:print-info 2 "server-timeout: " server-timeout ", server pid: " spid " on " iface ":" port)
    (let loop ((count 0))
      (thread-sleep! 4) ;; no need to do this very often
      ;; NB// sync currently does NOT return queue-length
      (let () ;; (queue-len (cdb:client-call server-info 'sync #t 1)))
      ;; (print "Server running, count is " count)
        (if (< count 1) ;; 3x3 = 9 secs aprox
            (loop (+ count 1)))
        
	;; Check that iface and port have not changed (can happen if server port collides)
	(mutex-lock! *heartbeat-mutex*)
	(set! sdat *runremote*)
	(mutex-unlock! *heartbeat-mutex*)

	(if (not (equal? sdat (list iface port)))
	(if (or (not (equal? sdat (list iface port)))
		(not spid))
	    (begin 
	      (debug:print-info 1 "interface changed, refreshing iface and port info")
	      (debug:print-info 0 "interface changed, refreshing iface and port info")
	      (set! iface (car sdat))
	      (set! port  (cadr sdat))
	      (set! spid  (tasks:server-get-server-id tdb #f iface port #f))))

        ;; NOTE: Get rid of this mechanism! It really is not needed...
        ;; (open-run-close tasks:server-update-heartbeat tasks:open-db spid)
        (tasks:server-update-heartbeat tdb spid)
      
        ;; (if ;; (or (> numrunning 0) ;; stay alive for two days after last access
        (mutex-lock! *heartbeat-mutex*)
        (set! last-access *last-db-access*)
        (mutex-unlock! *heartbeat-mutex*)
	;; (debug:print 11 "last-access=" last-access ", server-timeout=" server-timeout)
        (if (> (+ last-access server-timeout)
               (current-seconds))
            (begin
              (debug:print-info 2 "Server continuing, seconds since last db access: " (- (current-seconds) last-access))
              (loop 0))
            (begin
              (debug:print-info 0 "Starting to shutdown the server.")
              ;; need to delete only *my* server entry (future use)
              (set! *time-to-exit* #t)
              (tasks:server-deregister-self tdb (get-host-name))
              (open-run-close tasks:server-deregister-self tasks:open-db (get-host-name))
              (thread-sleep! 1)
              (debug:print-info 0 "Max cached queries was    " *max-cache-size*)
	      (debug:print-info 0 "Number of cached writes   " *number-of-writes*)
	      (debug:print-info 0 "Average cached write time "
				(if (eq? *number-of-writes* 0)
				    "n/a (no writes)"
				    (/ *writes-total-delay*

	      (thread-start! th3)
	      (thread-start! th1)
	      (set! *didsomething* #t)
	      (thread-join! th2))
	    (debug:print 0 "ERROR: Failed to setup for megatest")))
    (exit)))

;; (use trace)
;; (trace http-transport:keep-running 
;;        tasks:server-update-heartbeat
;;        tasks:server-get-server-id)
;;        tasks:get-best-server
;;        http-transport:run
;;        http-transport:launch
;;        http-transport:try-start-server
;;        http-transport:client-send-receive
;;        http-transport:make-server-url
;;        tasks:server-register
;;        tasks:server-delete
;;        start-server
;;        hostname->ip
;;        with-input-from-request
;;        tasks:server-deregister-self)

(define (http-transport:server-signal-handler signum)
  (handle-exceptions
   exn
   (debug:print " ... exiting ...")
   (let ((th1 (make-thread (lambda ()
			     (thread-sleep! 1))
			     ;; (if (not *received-response*)

	'())))

(define (launch:execute encoded-cmd)
  (let* ((cmdinfo   (read (open-input-string (base64:base64-decode encoded-cmd)))))
    (setenv "MT_CMDINFO" encoded-cmd)
    (if (list? cmdinfo) ;; ((testpath /tmp/mrwellan/jazzmind/src/example_run/tests/sqlitespeed)
	;; (test-name sqlitespeed) (runscript runscript.rb) (db-host localhost) (run-id 1))
	(let* ((testpath  (assoc/default 'testpath  cmdinfo))  ;; How is testpath different from work-area ??
	(let* ((testpath  (assoc/default 'testpath  cmdinfo))  ;; testpath is the test spec area
	       (top-path  (assoc/default 'toppath   cmdinfo))
	       (work-area (assoc/default 'work-area cmdinfo))
	       (work-area (assoc/default 'work-area cmdinfo))  ;; work-area is the test run area
	       (test-name (assoc/default 'test-name cmdinfo))
	       (runscript (assoc/default 'runscript cmdinfo))
	       (ezsteps   (assoc/default 'ezsteps   cmdinfo))
	       ;; (runremote (assoc/default 'runremote cmdinfo))
	       (transport (assoc/default 'transport cmdinfo))
	       (serverinf (assoc/default 'serverinf cmdinfo))
	       (port      (assoc/default 'port      cmdinfo))

	  ;; Can setup as client for server mode now
	  ;; (client:setup)

	  (change-directory *toppath*) 
	  (set-megatest-env-vars run-id) ;; these may be needed by the launching process
	  (change-directory work-area) 

	  (open-run-close set-run-config-vars #f run-id keys keyvals)
	  (set-run-config-vars run-id keys keyvals target) ;; (db:get-target db run-id))
	  ;; environment overrides are done *before* the remaining critical envars.
	  (alist->env-vars env-ovrd)
	  (set-megatest-env-vars run-id)
	  (set-item-env-vars itemdat)
	  (save-environment-as-files "megatest")
	  ;; open-run-close not needed for test-set-meta-info
	  (test-set-meta-info #f test-id run-id test-name itemdat 0)
	  (tests:set-meta-info #f test-id run-id test-name itemdat 0 work-area)
	  (tests:test-set-status! test-id "REMOTEHOSTSTART" "n/a" (args:get-arg "-m") #f)
	  (if (args:get-arg "-xterm")
	      (set! fullrunscript "xterm")
	      (if (and fullrunscript (not (file-execute-access? fullrunscript)))
		  (system (conc "chmod ug+x " fullrunscript))))
	  ;; We are about to actually kick off the test
	  ;; so this is a good place to remove the records for 

						   ;;       (set! script (conc script "source " prev-env))))
						   
						   ;; call the command using mt_ezstep
						   (set! script (conc "mt_ezstep " stepname " " (if prevstep prevstep "-") " " stepcmd))

						   (debug:print 4 "script: " script)
						   ;; DO NOT remote
						   (db:teststep-set-status! #f test-id stepname "start" "-" #f #f)
						   (db:teststep-set-status! #f test-id stepname "start" "-" #f #f work-area: work-area)
						   ;; now launch
						   (let ((pid (process-run script)))
						     (let processloop ((i 0))
						       (let-values (((pid-val exit-status exit-code)(process-wait pid #t)))
								   (mutex-lock! m)
								   (vector-set! exit-info 0 pid)
								   (vector-set! exit-info 1 exit-status)
								   (vector-set! exit-info 2 exit-code)
								   (mutex-unlock! m)
								   (if (eq? pid-val 0)
								       (begin
									 (thread-sleep! 2)
									 (processloop (+ i 1))))
								   ))
                                                     (let ((exinfo (vector-ref exit-info 2))
                                                           (logfna (if logpro-used (conc stepname ".html") "")))
						       ;; testing if procedures called in a remote call cause problems (ans: no or so I suspect)
						       (db:teststep-set-status! #f test-id stepname "end" exinfo #f logfna))
						       (db:teststep-set-status! #f test-id stepname "end" exinfo #f logfna work-area: work-area))
						     (if logpro-used
							 (cdb:test-set-log! *runremote*  test-id (conc stepname ".html")))
						     ;; set the test final status
						     (let* ((this-step-status (cond
									       ((and (eq? (vector-ref exit-info 2) 2) logpro-used) 'warn)
									       ((eq? (vector-ref exit-info 2) 0)                   'pass)
									       (else 'fail)))

							    (current-seconds) 
							    start-seconds)))))
					(kill-tries 0))
				   (let loop ((minutes   (calc-minutes)))
				     (begin
				       (set! kill-job? (test-get-kill-request test-id)) ;; run-id test-name itemdat))
				       ;; open-run-close not needed for test-set-meta-info
				       (test-set-meta-info #f test-id run-id test-name itemdat minutes)
				       (tests:set-meta-info #f test-id run-id test-name itemdat minutes work-area)
				       (if kill-job? 
					   (begin
					     (mutex-lock! m)
					     (let* ((pid (vector-ref exit-info 0)))
					       (if (number? pid)
						   (begin
						     (debug:print 0 "WARNING: Request received to kill job (attempt # " kill-tries ")")

				     ((eq? rollup-status 2)
				      ;; if the current status is AUTO the defer to the calculated value but qualify (i.e. make this AUTO-WARN)
				      (if (equal? (db:test-get-status testinfo) "AUTO") "AUTO-WARN" "WARN"))
				     (else "FAIL"))
				    (args:get-arg "-m") #f)))
	      ;; for automated creation of the rollup html file this is a good place...
	      (if (not (equal? item-path ""))
		  (open-run-close tests:summarize-items #f run-id test-name #f)) ;; don't force - just update if no
		  (tests:summarize-items #f run-id test-name #f)) ;; don't force - just update if no
	      )
	    (mutex-unlock! m)
	    ;; (exec-results (cmd-run->list fullrunscript)) ;;  (list ">" (conc test-name "-run.log"))))
	    ;; (success      exec-results)) ;; (eq? (cadr exec-results) 0)))
	    (debug:print 2 "Output from running " fullrunscript ", pid " (vector-ref exit-info 0) " in work area " 
			 work-area ":\n====\n exit code " (vector-ref exit-info 2) "\n" "====\n")
	    ;; (sqlite3:finalize! db)

;; 
;;  <linkdir> - <target> - <testname> [ - <itempath> ]
;; 
;; All log file links should be stored relative to the top of link path
;;  
;; <target> - <testname> [ - <itempath> ] 
;;
(define (create-work-area db run-id test-id test-src-path disk-path testname itemdat)
(define (create-work-area run-id run-info key-vals test-id test-src-path disk-path testname itemdat)
  (let* ((run-info (cdb:remote-run db:get-run-info #f run-id))
	 (item-path (item-list->path itemdat))
  (let* ((item-path (item-list->path itemdat))
	 (runname  (db:get-value-by-header (db:get-row run-info)
					   (db:get-header run-info)
					   "runname"))
	 ;; convert back to db: from rdb: - this is always run at server end
	 (key-vals (cdb:remote-run db:get-key-vals #f run-id))
	 (target   (string-intersperse key-vals "/"))

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

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

    ;;   (system  (conc "ln -sf " test-path " " testlink)))
    (if (directory? test-path)
	(begin
	  (let* ((ovrcmd (let ((cmd (config-lookup *configdat* "setup" "testcopycmd")))
			   (if cmd
			       ;; substitute the TEST_SRC_PATH and TEST_TARG_PATH
			       (string-substitute "TEST_TARG_PATH" test-path
						  (string-substitute "TEST_SRC_PATH" test-src-path cmd))
						  (string-substitute "TEST_SRC_PATH" test-src-path cmd #t) #t)
			       #f)))
		 (cmd    (if ovrcmd 
			     ovrcmd
			     (conc "rsync -av" (if (debug:debug-mode 1) "" "q") " " test-src-path "/ " test-path "/")))
			     (conc "rsync -av" (if (debug:debug-mode 1) "" "q") " " test-src-path "/ " test-path "/"
				   " >> " test-path "/mt_launch.log 2>> " test-path "/mt_launch.log")))
		 (status (system cmd)))
	    (if (not (eq? status 0))
		(debug:print 2 "ERROR: problem with running \"" cmd "\"")))
	  (list lnkpathf lnkpath ))
	(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 
;; 4. remotely run the test on allocated host
;;    - could be ssh to host from hosts table (update regularly with load)
;;    - could be netbatch
;;      (launch-test db (cadr status) test-conf))
(define (launch-test db run-id runname test-conf keyvallst test-name test-path itemdat params)
(define (launch-test test-id run-id run-info key-vals runname test-conf keyvallst test-name test-path itemdat params)
  (change-directory *toppath*)
  (alist->env-vars ;; consolidate this code with the code in megatest.scm for "-execute"
   (list ;; (list "MT_TEST_RUN_DIR" work-area)
    (list "MT_RUN_AREA_HOME" *toppath*)
    (list "MT_TEST_NAME" test-name)
    ;; (list "MT_ITEM_INFO" (conc itemdat)) 
    (list "MT_RUNNAME"   runname)

	 (work-area  #f)
	 (toptest-work-area #f) ;; for iterated tests the top test contains data relevant for all
	 (diskpath   #f)
	 (cmdparms   #f)
	 (fullcmd    #f) ;; (define a (with-output-to-string (lambda ()(write x))))
	 (mt-bindir-path #f)
	 (item-path (item-list->path itemdat))
	 (test-id    (cdb:remote-run db:get-test-id #f run-id test-name item-path))
	 ;; (test-id    (cdb:remote-run db:get-test-id #f run-id test-name item-path))
	 (testinfo   (cdb:get-test-info-by-id *runremote* test-id))
	 (mt_target  (string-intersperse (map cadr keyvallst) "/"))
	 (debug-param (append (if (args:get-arg "-debug")  (list "-debug" (args:get-arg "-debug")) '())
			      (if (args:get-arg "-logging")(list "-logging") '()))))
    (if hosts (set! hosts (string-split hosts)))
    ;; set the megatest to be called on the remote host
    (if (not remote-megatest)(set! remote-megatest local-megatest)) ;; "megatest"))
    (set! mt-bindir-path (pathname-directory remote-megatest))
    (if launcher (set! launcher (string-split launcher)))
    ;; set up the run work area for this test
    (set! diskpath (get-best-disk *configdat*))
    (if diskpath
	(let ((dat  (open-run-close create-work-area db run-id test-id test-path diskpath test-name itemdat)))
	(let ((dat  (create-work-area run-id run-info key-vals test-id test-path diskpath test-name itemdat)))
	  (set! work-area (car dat))
	  (set! toptest-work-area (cadr dat))
	  (debug:print-info 2 "Using work area " work-area))
	(begin
	  (set! work-area (conc test-path "/tmp_run"))
	  (create-directory work-area #t)
	  (debug:print 0 "WARNING: No disk work area specified - running in the test directory under tmp_run")))

			    (append (list (list "MT_TEST_RUN_DIR" work-area)
					  (list "MT_TEST_NAME" test-name)
					  (list "MT_ITEM_INFO" (conc itemdat)) 
					  (list "MT_RUNNAME"   runname)
					  (list "MT_TARGET"    mt_target)
					  )
				    itemdat)))
	   (launch-results (apply (if (equal? (configf:lookup *configdat* "setup" "launchwait") "yes")
	   (launch-results (apply cmd-run-with-stderr->list ;; cmd-run-proc-each-line
				      cmd-run-with-stderr->list
				      process-run)
				  (if useshell
				      (string-intersperse fullcmd " ")
				      (car fullcmd))
				  ;; conc
				  (if useshell
				      '()
				      (cdr fullcmd))))) ;;  launcher fullcmd)));; (apply cmd-run-proc-each-line launcher print fullcmd))) ;; (cmd-run->list fullcmd))
      (with-output-to-file "mt_launch.log"
	(lambda ()
	  (apply print launch-results)))
				      (cdr fullcmd)))))
      (if (list? launch-results)
	  (with-output-to-file "mt_launch.log"
	    (lambda ()
	      (apply print launch-results))
	    #:append))
      (debug:print 2 "Launching completed, updating db")
      (debug:print 2 "Launch results: " launch-results)
      (if (not launch-results)
	  (begin
	    (print "ERROR: Failed to run " (string-intersperse fullcmd " ") ", exiting now")
	    ;; (sqlite3:finalize! db)
	    ;; good ole "exit" seems not to work
	    ;; (_exit 9)
	    ;; but this hack will work! Thanks go to Alan Post of the Chicken email list
	    ;; NB// Is this still needed? Should be safe to go back to "exit" now?
	    (process-signal (current-process-id) signal/kill)
	    ))
          (begin
            (print "ERROR: Failed to run " (string-intersperse fullcmd " ") ", exiting now")
            ;; (sqlite3:finalize! db)
            ;; good ole "exit" seems not to work
            ;; (_exit 9)
            ;; but this hack will work! Thanks go to Alan Post of the Chicken email list
            ;; NB// Is this still needed? Should be safe to go back to "exit" now?
            (process-signal (current-process-id) signal/kill)
            ))
      (alist->env-vars miscprevvals)
      (alist->env-vars testprevvals)
      (alist->env-vars commonprevvals)
      launch-results))
  (change-directory *toppath*))

;; Always use two digit decimal
;; 1.01, 1.02...1.10,1.11 ... 1.99,2.00..

(declare (unit megatest-version))

(define megatest-version 1.5415)
(define megatest-version 1.5423)

;; Copyright 2006-2012, Matthew Welland.
;; 
;;  This program is made available under the GNU GPL version 2.0 or
;;  greater. See the accompanying file COPYING for details.
;; 
;;  This program is distributed WITHOUT ANY WARRANTY; without even the
;;  implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
;;  PURPOSE.

;; (include "common.scm")
;; (include "megatest-version.scm")

(use sqlite3 srfi-1 posix regex regex-case srfi-69 base64 format readline apropos json) ;; (srfi 18) extras)
(use sqlite3 srfi-1 posix regex regex-case srfi-69 base64 format readline apropos json http-client) ;; (srfi 18) extras)
(import (prefix sqlite3 sqlite3:))
(import (prefix base64 base64:))

(use zmq)
;; (use zmq)

(declare (uses common))
(declare (uses megatest-version))
(declare (uses margs))
(declare (uses runs))
(declare (uses launch))
(declare (uses server))
(declare (uses client))
(declare (uses tests))
(declare (uses genexample))
(declare (uses daemon))

(define *db* #f) ;; this is only for the repl, do not use in general!!!!

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

;; (use trace dot-locking)
;; (trace
(let ((debugcontrolf (conc (get-environment-variable "HOME") "/.megatestrc")))
  (if (file-exists? debugcontrolf)
;;  cdb:client-call
;;  cdb:remote-run
;;  cdb:test-set-status-state
;;  change-directory
;;  db:process-queue-item
;;  db:test-get-logfile-info
;;  db:teststep-set-status!
;;  nice-path
;;  obtain-dot-lock
;;  open-run-close
;;  read-config
      (load debugcontrolf)))
;;  runs:can-run-more-tests
;;  sqlite3:execute
;;  sqlite3:for-each-row
;;  tests:check-waiver-eligibility
;;  tests:summarize-items
;;  tests:test-set-status!
;;  thread-sleep!
;;)
       


(define help (conc "
Megatest, documentation at http://www.kiatoa.com/fossils/megatest
  version " megatest-version "
  license GPL, Copyright Matt Welland 2006-2012

Usage: megatest [options]

                            patterns.
  -list-disks             : list the disks available for storing runs
  -list-targets           : list the targets in runconfigs.config
  -list-db-targets        : list the target combinations used in the db
  -show-config            : dump the internal representation of the megatest.config file
  -show-runconfig         : dump the internal representation of the runconfigs.config file
  -dumpmode json          : dump in json format instead of sexpr
  -show-cmdinfo           : dump the command info for a test (run in test environment)

Misc 
  -rebuild-db             : bring the database schema up to date
  -update-meta            : update the tests metadata for all tests
  -env2file fname         : write the environment to fname.csh and fname.sh
  -setvars VAR1=val1,VAR2=val2 : Add environment variables to a run NB// these are
                                 overwritten by values set in config files.
  -server -|hostname      : start the server (reduces contention on megatest.db), use
                            - to automatically figure out hostname
  -transport http|zmq     : use http or zmq for transport (default is http) 
  -transport http|fs      : use http or direct access for transport (default is http) 
  -daemonize              : fork into background and disconnect from stdin/out
  -list-servers           : list the servers 
  -stop-server id         : stop server specified by id (see output of -list-servers)
  -repl                   : start a repl (useful for extending megatest)
  -load file.scm          : load and run file.scm

Spreadsheet generation

			"-list-servers"
			;; mist queries
			"-list-disks"
			"-list-targets"
			"-list-db-targets"
			"-show-runconfig"
			"-show-config"
			"-show-cmdinfo"
			;; queries
			"-test-paths" ;; get path(s) to a test, ordered by youngest first

			"-runall"    ;; run all tests
			"-remove-runs"
			"-rebuild-db"
			"-rollup"

    (if (not (null? (lset-intersection 
		     equal?
		     (hash-table-keys args:arg-hash)
		     '("-runtests"    "-list-runs"   "-rollup"
		       "-remove-runs" "-lock"        "-unlock"
		       "-update-meta" "-extract-ods"))))
	(if (setup-for-run)
	    (let ((servers (open-run-close tasks:get-best-server tasks:open-db)))
	    (let loop ((servers  (open-run-close tasks:get-best-server tasks:open-db))
		       (trycount 0))
	      (if (or (not servers)
		      (null? servers))
		  (begin
		    (if (eq? trycount 0) ;; just do the server start once
			(begin
		    (debug:print 0 "INFO: Starting server as none running ...")
		    ;; (server:launch (string->symbol (args:get-arg "-transport" "http"))))
		    (system (conc (car (argv)) " -server - -daemonize -transport " (args:get-arg "-transport" "http")))
		    (thread-sleep! 3)) ;; give the server a few seconds to start
		  (debug:print 0 "INFO: Servers already running " servers)
			  (debug:print 0 "INFO: Starting server as none running ...")
			  ;; (server:launch (string->symbol (args:get-arg "-transport" "http"))))
			  ;; (process-run (car (argv)) (list "-server" "-" "-daemonize" "-transport" (args:get-arg "-transport" "http")))
			  (system "megatest -list-servers | grep alive || megatest -server - -daemonize && sleep 3")
			  ;; (process-fork (lambda ()
			  ;;       	  (daemon:ize)
			  ;;       	  (server:launch (string->symbol (args:get-arg "-transport" "http")))))
			  ;; (thread-sleep! 3)
			  )
			(debug:print-info 0 "Waiting for server to start"))
		    (loop (open-run-close tasks:get-best-server tasks:open-db) 
			  (+ trycount 1)))
		  (debug:print 0 "INFO: Server(s) running " servers)
		  )))))
	

(if (or (args:get-arg "-list-servers")
	(args:get-arg "-stop-server"))
    (let ((tl (setup-for-run)))
      (if tl 
	  (let* ((servers (open-run-close tasks:get-all-servers tasks:open-db))
		 (fmtstr  "~5a~8a~8a~20a~20a~10a~10a~10a~10a~10a\n")

		 (if (equal? id sid)
		     (begin
		       (debug:print-info 0 "Attempting to stop server with pid " pid)
		       (tasks:kill-server status hostname pullport pid transport)))))
	     servers)
	    (debug:print-info 1 "Done with listservers")
	    (set! *didsomething* #t)
	    (exit) ;; must do, would have to add checks to many/all calls below
	    (exit)) ;; must do, would have to add checks to many/all calls below
	    )
	  (exit)))
    ;; if not list or kill then start a client (if appropriate)
    (if (or (args-defined? "-h" "-version" "-gen-megatest-area" "-gen-megatest-test")
	    (eq? (length (hash-table-keys args:arg-hash)) 0))
	(debug:print-info 1 "Server connection not needed")
	;; ok, so lets connect to the server
	(client:launch)))

;;======================================================================
;; Weird special calls that need to run *after* the server has started?
;;======================================================================

(if (args:get-arg "-list-targets")
    (let ((targets (common:get-runconfig-targets)))
      (print "Found "(length targets) " targets")
      (for-each (lambda (x)
		  ;; (print "[" x "]"))
		  (print x))
		targets)
      (set! *didsomething* #t)))

(if (args:get-arg "-show-runconfig")
    (let* ((target (if (args:get-arg "-reqtarg")
		       (args:get-arg "-reqtarg")
		       (if (args:get-arg "-target")
			   (args:get-arg "-target")
			   #f)))
	   (sections (if target (list "default" target) #f))
	   (data     (read-config "runconfigs.config" #f #t sections: sections)))
(define (full-runconfigs-read)
  (let* ((keys   (cdb:remote-run get-keys #f))
	 (target (if (args:get-arg "-reqtarg")
		     (args:get-arg "-reqtarg")
		     (if (args:get-arg "-target")
			 (args:get-arg "-target")
			 #f)))
	 (key-vals (if target (keys:target->keyval keys target) #f))
	 (sections (if target (list "default" target) #f))
	 (data     (begin
		     (setenv "MT_RUN_AREA_HOME" *toppath*)
		     (if key-vals
			 (for-each (lambda (kt)
				     (setenv (car kt) (cadr kt)))
				   key-vals))
		     (read-config "runconfigs.config" #f #t sections: sections))))
    data))


(if (args:get-arg "-show-runconfig")
    (let ((data (full-runconfigs-read)))
      ;; keep this one local
      (cond
       ((not (args:get-arg "-dumpmode"))
	(pp (hash-table->alist data)))
       ((string=? (args:get-arg "-dumpmode") "json")
	(json-write data))
       (else

       ((not (args:get-arg "-dumpmode"))
	(pp (hash-table->alist data)))
       ((string=? (args:get-arg "-dumpmode") "json")
	(json-write data))
       (else
	(debug:print 0 "ERROR: -dumpmode of " (args:get-arg "-dumpmode") " not recognised")))
      (set! *didsomething* #t)))

(if (args:get-arg "-show-cmdinfo")
    (let ((data (read (open-input-string (base64:base64-decode (getenv "MT_CMDINFO"))))))
      (if (equal? (args:get-arg "-dumpmode") "json")
	  (json-write data)
	  (pp data))
      (set! *didsomething* #t)))

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

;; since several actions can be specified on the command line the removal
;; is done first

		       (begin
			 (hash-table-set! seen targetstr #t)
			 ;; (print "[" targetstr "]"))))
			 (print targetstr))))
	       (if (not db-targets)
		   (let* ((run-id (db:get-value-by-header run header "id"))
			  (tests  (cdb:remote-run db:get-tests-for-run #f run-id testpatt '() '())))
		     (print "Run: " targetstr "/" (db:get-value-by-header run header "runname") 
		     (debug:print 1 "Run: " targetstr " status: " (db:get-value-by-header run header "state")
				  " run-id: " run-id ", number tests: " (length tests))
			    " status: " (db:get-value-by-header run header "state")
			    " run-id: " run-id ", number tests: " (length tests))
		     (for-each 
		      (lambda (test)
			(format #t
				"  Test: ~25a State: ~15a Status: ~15a Runtime: ~5@as Time: ~22a Host: ~10a\n"
				(conc (db:test-get-testname test)
				      (if (equal? (db:test-get-item-path test) "")
					  "" 

  (general-run-call 
   "-runtests" 
   "run a test" 
   (lambda (target runname keys keynames keyvallst)
     (runs:run-tests target
		     runname
		     (args:get-arg "-runtests")
		     (args:get-arg "-testpatt")
		     (args:get-arg "-runtests")
		     user
		     args:arg-hash))))

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

	     (testpath  (assoc/default 'testpath  cmdinfo))
	     (test-name (assoc/default 'test-name cmdinfo))
	     (runscript (assoc/default 'runscript cmdinfo))
	     (db-host   (assoc/default 'db-host   cmdinfo))
	     (run-id    (assoc/default 'run-id    cmdinfo))
	     (test-id   (assoc/default 'test-id   cmdinfo))
	     (itemdat   (assoc/default 'itemdat   cmdinfo))
	     (work-area (assoc/default 'work-area cmdinfo))
	     (db        #f))
	(change-directory testpath)
	;; (set! *runremote* runremote)
	(set! *transport-type* (string->symbol transport))
	(if (not (setup-for-run))
	    (begin
	      (debug:print 0 "Failed to setup, exiting")
	      (exit 1)))
	(if (and state status)
	    ;; DO NOT remote run
	    (db:teststep-set-status! db test-id step state status msg logfile)
	    ;; DO NOT remote run, makes calls to the testdat.db test db.
	    (db:teststep-set-status! db test-id step state status msg logfile work-area: work-area)
	    (begin
	      (debug:print 0 "ERROR: 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")
       (args:get-arg ":state")
       (args:get-arg ":status")
       (args:get-arg "-setlog")
       (args:get-arg "-m"))
      ;; (if db (sqlite3:finalize! db))
      (set! *didsomething* #t)))
    
(if (or (args:get-arg "-setlog")       ;; since setting up is so costly lets piggyback on -test-status
(if (or (and (args:get-arg "-setlog")       ;; since setting up is so costly lets piggyback on -test-status
	     (not (args:get-arg "-step")))  ;; -setlog may have been processed already in the "-step" previous
	(args:get-arg "-set-toplog")
	(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"))

	       (testpath  (assoc/default 'testpath  cmdinfo))
	       (test-name (assoc/default 'test-name cmdinfo))
	       (runscript (assoc/default 'runscript cmdinfo))
	       (db-host   (assoc/default 'db-host   cmdinfo))
	       (run-id    (assoc/default 'run-id    cmdinfo))
	       (test-id   (assoc/default 'test-id   cmdinfo))
	       (itemdat   (assoc/default 'itemdat   cmdinfo))
	       (work-area (assoc/default 'work-area cmdinfo))
	       (db        #f) ;; (open-db))
	       (state     (args:get-arg ":state"))
	       (status    (args:get-arg ":status")))
	  (change-directory testpath)
	  ;; (set! *runremote* runremote)
	  (set! *transport-type* (string->symbol transport))
	  (if (not (setup-for-run))
	      (begin
		(debug:print 0 "Failed to setup, exiting")
		(exit 1)))

	  ;; can setup as client for server mode now
	  ;; (client:setup)

	  (if (args:get-arg "-load-test-data")
	      ;; has sub commands that are rdb:
	      ;; DO NOT put this one into either cdb:remote-run or open-run-close
	      (db:load-test-data db test-id))
	      (db:load-test-data db test-id work-area: work-area))
	  (if (args:get-arg "-setlog")
	      (let ((logfname (args:get-arg "-setlog")))
		(cdb:test-set-log! *runremote* test-id logfname)))
	  (if (args:get-arg "-set-toplog")
	      ;; DO NOT run remote
	      (tests:test-set-toplog! db run-id test-name (args:get-arg "-set-toplog")))
	  (if (args:get-arg "-summarize-items")

				       ((zsh bash sh ash) "2>&1 >")
				       (else ">&")))
			 (fullcmd    (conc "(" (string-intersperse 
						(cons cmd params) " ")
					   ") " redir " " logfile)))
		    ;; mark the start of the test
		    ;; DO NOT run remote
		    (db:teststep-set-status! db test-id stepname "start" "n/a" (args:get-arg "-m") logfile)
		    (db:teststep-set-status! db test-id stepname "start" "n/a" (args:get-arg "-m") logfile work-area: work-area)
		    ;; run the test step
		    (debug:print-info 2 "Running \"" fullcmd "\"")
		    (change-directory startingdir)
		    (set! exitstat (system fullcmd)) ;; cmd params))
		    (set! *globalexitstatus* exitstat)
		    (change-directory testpath)
		    ;; run logpro if applicable ;; (process-run "ls" (list "/foo" "2>&1" "blah.log"))
		    (if logprofile
			(let* ((htmllogfile (conc stepname ".html"))
			       (oldexitstat exitstat)
			       (cmd         (string-intersperse (list "logpro" logprofile htmllogfile "<" logfile ">" (conc stepname "_logpro.log")) " ")))
			  (debug:print-info 2 "running \"" cmd "\"")
			  (change-directory startingdir)
			  (set! exitstat (system cmd))
			  (set! *globalexitstatus* exitstat) ;; no necessary
			  (change-directory testpath)
			  (cdb:test-set-log! *runremote* test-id htmllogfile)))
		    (let ((msg (args:get-arg "-m")))
		      ;; DO NOT run remote
		      (db:teststep-set-status! db test-id stepname "end" exitstat msg logfile))
		      (db:teststep-set-status! db test-id stepname "end" exitstat msg logfile work-area: work-area))
		    )))
	  (if (or (args:get-arg "-test-status")
		  (args:get-arg "-set-values"))
	      (let ((newstatus (cond
				((number? status)       (if (equal? status 0) "PASS" "FAIL"))
				((and (string? status)
				      (string->number status))(if (equal? (string->number status) 0) "PASS" "FAIL"))

		    (begin
		      (debug:print 0 "ERROR: You must specify :state and :status with every call to -test-status\n" help)
		      ;; (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! test-id state newstatus msg otherdata))))
		  (tests:test-set-status! test-id state newstatus msg otherdata work-area: work-area))))
	  (if db (sqlite3:finalize! db))
	  (set! *didsomething* #t))))

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

		(load (args:get-arg "-load"))))
	  (exit))
      (set! *didsomething* #t)))

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

(if *runremote* (close-all-connections!))

;; this is the socket if we are a client
;; (if (and *runremote*
;; 	 (socket? *runremote*))
;;     (close-socket *runremote*))

(if (not *didsomething*)

	  (for-each (lambda (fullkey)
		      (debug:print 2 (format #f "~20a ~a\n" fullkey (hash-table-ref/default whatfound fullkey 0))))
		    sections)
	  (debug:print 2 "---")
	  (set! *already-seen-runconfig-info* #t)))
    finaldat))

(define (set-run-config-vars db run-id keys keyvals)
(define (set-run-config-vars run-id keys keyvals targ-from-db)
  (push-directory *toppath*)
  (let ((runconfigf (conc  *toppath* "/runconfigs.config"))
	(targ       (or (args:get-arg "-target")
			(args:get-arg "-reqtarg")
			(db:get-target db run-id))))
			targ-from-db)))
    (pop-directory)
    (if (file-exists? runconfigf)
	(setup-env-defaults runconfigf run-id #t keys keyvals
			    environ-patt: (conc "(default"
						(if targ
						    (conc "|" targ ")")
						    ")")))

    (vector header res)))

(define (runs:test-get-full-path test)
  (let* ((testname (db:test-get-testname   test))
	 (itempath (db:test-get-item-path test)))
    (conc testname (if (equal? itempath "") "" (conc "(" itempath ")")))))

(define (db:get-run-key-val db run-id key)
  (let ((res #f))
    (sqlite3:for-each-row
     (lambda (val)
       (set! res val))
     db 
     (conc "SELECT " (key:get-fieldname key) " FROM runs WHERE id=?;")
     run-id)
    res))

(define (db:get-run-name-from-id db run-id)
  (let ((res #f))
    (sqlite3:for-each-row
     (lambda (runname)
       (set! res runname))
     db
     "SELECT runname FROM runs WHERE id=?;"
     run-id)
    res))

(define (set-megatest-env-vars run-id)
  (let ((keys (cdb:remote-run db:get-keys #f))
(define (set-megatest-env-vars run-id #!key (inkeys #f)(inrunname #f))
  (let ((keys (if inkeys inkeys (cdb:remote-run db:get-keys #f)))
	(vals (hash-table-ref/default *env-vars-by-run-id* run-id #f)))
    ;; get the info from the db and put it in the cache
    (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 key (cdb:remote-run db:get-run-key-val #f run-id key)))
	   keys)))
    ;; from the cached data set the vars
    (hash-table-for-each
     vals
     (lambda (key val)
       (debug:print 2 "setenv " (key:get-fieldname key) " " val)
       (setenv (key:get-fieldname key) val)))
    (alist->env-vars (hash-table-ref/default *configdat* "env-override" '()))
    ;; Lets use this as an opportunity to put MT_RUNNAME in the environment
    (setenv "MT_RUNNAME" (cdb:remote-run db:get-run-name-from-id #f run-id))
    (setenv "MT_RUNNAME" (if inrunname inrunname (cdb:remote-run db:get-run-name-from-id #f run-id)))
    (setenv "MT_RUN_AREA_HOME" *toppath*)
    ))

(define (set-item-env-vars itemdat)
  (for-each (lambda (item)
	      (debug:print 2 "setenv " (car item) " " (cadr item))
	      (setenv (car item) (cadr item)))
	    itemdat))

(define *last-num-running-tests* 0)

;; Every time can-run-more-tests is called increment the delay
;; if the cou
(define *runs:can-run-more-tests-delay* 0)
(define (runs:shrink-can-run-more-tests-delay)
  (set! *runs:can-run-more-tests-delay* 0)) ;; (/ *runs:can-run-more-tests-delay* 2)))
(define *runs:can-run-more-tests-count* 0)
(define (runs:shrink-can-run-more-tests-count)
  (set! *runs:can-run-more-tests-count* 0)) ;; (/ *runs:can-run-more-tests-count* 2)))

(define (runs:can-run-more-tests test-record)
  (thread-sleep! *runs:can-run-more-tests-delay*)
(define (runs:can-run-more-tests test-record max-concurrent-jobs)
  (thread-sleep! (cond
		  ((> *runs:can-run-more-tests-count* 20) 2);; obviously haven't had any work to do for a while
		  (else 0)))
  (let* ((tconfig                 (tests:testqueue-get-testconfig test-record))
	 (jobgroup                (config-lookup tconfig "requirements" "jobgroup"))
	 ;; Heuristic fix. These are getting called too rapidly when jobs are running or stuck
	 ;; so we are going to increment a global delay by 0.1 seconds up to 10 seconds
	 ;; every time runs:can-run-more-tests is called.
	 ;; when a test is launched or other activity occurs divide the delay by 2
	 (num-running             (cdb:remote-run db:get-count-tests-running #f))
	 (num-running-in-jobgroup (cdb:remote-run db:get-count-tests-running-in-jobgroup #f jobgroup))
	 (max-concurrent-jobs     (let ((mcj (config-lookup *configdat* "setup"     "max_concurrent_jobs")))
				    (if (and mcj (string->number mcj))
					(string->number mcj)
					1)))
	 (job-group-limit         (config-lookup *configdat* "jobgroups" jobgroup)))
    (if (and (> (+ num-running num-running-in-jobgroup) 0)
	     (< *runs:can-run-more-tests-delay* 1))
    (if (> (+ num-running num-running-in-jobgroup) 0)
	(set! *runs:can-run-more-tests-count* (+ *runs:can-run-more-tests-count* 1)))
	(begin
	  (set! *runs:can-run-more-tests-delay* (+ *runs:can-run-more-tests-delay* 0.009))
	  (debug:print-info 14 "can-run-more-tests-delay: " *runs:can-run-more-tests-delay*)))
    (if (not (eq? *last-num-running-tests* num-running))
	(begin
	  (debug:print 2 "max-concurrent-jobs: " max-concurrent-jobs ", num-running: " num-running)
	  (set! *last-num-running-tests* num-running)))
    (if (not (eq? 0 *globalexitstatus*))
	(list #f num-running num-running-in-jobgroup max-concurrent-jobs job-group-limit)
	(let ((can-not-run-more (cond

;;======================================================================
;; New methodology. These routines will replace the above in time. For
;; now the code is duplicated. This stuff is initially used in the monitor
;; based code.
;;======================================================================

;; register a test run with the db
(define (runs:register-run db keys keyvallst runname state status user)
  (debug:print 3 "runs:register-run, keys: " keys " keyvallst: " keyvallst " runname: " runname " state: " state " status: " status " user: " user)
  (let* ((keystr    (keys->keystr keys))
	 (comma     (if (> (length keys) 0) "," ""))
	 (andstr    (if (> (length keys) 0) " AND " ""))
	 (valslots  (keys->valslots keys)) ;; ?,?,? ...
	 (keyvals   (map cadr keyvallst))
	 (allvals   (append (list runname state status user) keyvals))
	 (qryvals   (append (list runname) keyvals))
	 (key=?str  (string-intersperse (map (lambda (k)(conc (key:get-fieldname k) "=?")) keys) " AND ")))
    (debug:print 3 "keys: " keys " allvals: " allvals " keyvals: " keyvals)
    (debug:print 2 "NOTE: using target " (string-intersperse keyvals "/") " for this run")
    (if (and runname (null? (filter (lambda (x)(not x)) keyvals))) ;; there must be a better way to "apply and"
	(let ((res #f))
	  (apply sqlite3:execute db (conc "INSERT OR IGNORE INTO runs (runname,state,status,owner,event_time" comma keystr ") VALUES (?,?,?,?,strftime('%s','now')" comma valslots ");")
		 allvals)
	  (apply sqlite3:for-each-row 
	   (lambda (id)
	     (set! res id))
	   db
	   (let ((qry (conc "SELECT id FROM runs WHERE (runname=? " andstr key=?str ");")))
	     ;(debug:print 4 "qry: " qry) 
	     qry)
	   qryvals)
	  (sqlite3:execute db "UPDATE runs SET state=?,status=? WHERE id=?;" state status res)
	  res) 
	(begin
	  (debug:print 0 "ERROR: Called without all necessary keys")
	  #f))))

;; This is a duplicate of run-tests (which has been deprecated). Use this one instead of run tests.
;; keyvals.
;;
;;  test-names: Comma separated patterns same as test-patts but used in selection 
;;              of tests to run. The item portions are not respected.
;;              FIXME: error out if /patt specified
;;            
(define (runs:run-tests target runname test-names test-patts user flags)
  (common:clear-caches) ;; clear all caches
  (let* ((db          #f)
	 (keys        (cdb:remote-run db:get-keys #f))
	 (keyvallst   (keys:target->keyval keys target))
	 (run-id      (cdb:remote-run runs:register-run #f keys keyvallst runname "new" "n/a" user))  ;;  test-name)))
	 (run-id      (cdb:remote-run db:register-run #f keys keyvallst runname "new" "n/a" user))  ;;  test-name)))
	 (keyvals     (if run-id (cdb:remote-run db:get-key-vals #f run-id) #f))
	 (deferred    '()) ;; delay running these since they have a waiton clause
	 ;; keepgoing is the defacto modality now, will add hit-n-run a bit later
	 ;; (keepgoing   (hash-table-ref/default flags "-keepgoing" #f))
	 (runconfigf   (conc  *toppath* "/runconfigs.config"))
	 (required-tests '())
	 (test-records (make-hash-table)))
	 (test-records (make-hash-table))
     (all-test-names (tests:get-valid-tests *toppath* "%"))) ;; we need a list of all valid tests to check waiton names)

    (set-megatest-env-vars run-id) ;; these may be needed by the launching process
    (set-megatest-env-vars run-id inkeys: keys) ;; these may be needed by the launching process

    (if (file-exists? runconfigf)
	(setup-env-defaults runconfigf run-id *already-seen-runconfig-info* keys keyvals "pre-launch-env-vars")
	(debug:print 0 "WARNING: You do not have a run config file: " runconfigf))
    
    ;; look up all tests matching the comma separated list of globs in
    ;; test-patts (using % as wildcard)

		 (waitons (let ((instr (if config 
					   (config-lookup config "requirements" "waiton")
					   (begin ;; No config means this is a non-existant test
					     (debug:print 0 "ERROR: non-existent required test \"" hed "\"")
					     (if db (sqlite3:finalize! db))
					     (exit 1)))))
			    (debug:print-info 8 "waitons string is " instr)
			    (let ((newwaitons
			    (string-split (cond
					   ((procedure? instr)
					    (let ((res (instr)))
					      (debug:print-info 8 "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 0 "ERROR: something went wrong in processing waitons for test " hed)
					    ""))))))
				   (string-split (cond
						  ((procedure? instr)
						   (let ((res (instr)))
						     (debug:print-info 8 "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 0 "ERROR: something went wrong in processing waitons for test " hed)
						   "")))))
			      (filter (lambda (x)
					(if (member x all-test-names)
					    #t
					    (begin
					      (debug:print 0 "ERROR: test " hed " has unrecognised waiton testname " x)
					      #f)))
				      newwaitons)))))
	    (debug:print-info 8 "waitons: " waitons)
	    ;; check for hed in waitons => this would be circular, remove it and issue an
	    ;; error
	    (if (member hed waitons)
		(begin
		  (debug:print 0 "ERROR: test " hed " has listed itself as a waiton, please correct this!")
		  (set! waitons (filter (lambda (x)(not (equal? x hed))) waitons))))

	      (if (not (null? remtests))
		  (loop (car remtests)(cdr remtests)))))))

    (if (not (null? required-tests))
	(debug:print-info 1 "Adding " required-tests " to the run queue"))
    ;; NOTE: these are all parent tests, items are not expanded yet.
    (debug:print-info 4 "test-records=" (hash-table->alist test-records))
    (let ((reglen (any->number  (configf:lookup *configdat* "setup" "runqueue"))))
      (if reglen
	  (runs:run-tests-queue-new     run-id runname test-records keyvallst flags test-patts required-tests reglen)
    (runs:run-tests-queue run-id runname test-records keyvallst flags test-patts)
	  (runs:run-tests-queue-classic run-id runname test-records keyvallst flags test-patts required-tests)))
    (debug:print-info 4 "All done by here")))

(define (runs:calc-fails prereqs-not-met)
  (filter (lambda (test)
	    (and (vector? test) ;; not (string? test))
		 (equal? (db:test-get-state test) "COMPLETED")
		 (not (member (db:test-get-status test)

	     (conc t)
	     (conc (db:test-get-testname t) ":" (db:test-get-state t) "/" (db:test-get-status t))))
       lst))

(define (runs:make-full-test-name testname itempath)
  (if (equal? itempath "") testname (conc testname "/" itempath)))

;; test-records is a hash table testname:item_path => vector < testname testconfig waitons priority items-info ... >
(define (runs:run-tests-queue run-id runname test-records keyvallst flags test-patts)
(define (runs:queue-next-hed tal reg n regful)
    ;; At this point the list of parent tests is expanded 
    ;; NB// Should expand items here and then insert into the run queue.
  (debug:print 5 "test-records: " test-records ", keyvallst: " keyvallst " flags: " (hash-table->alist flags))
  (let ((sorted-test-names (tests:sort-by-priority-and-waiton test-records))
	(test-registery    (make-hash-table))
	(num-retries        0)
	(max-retries       (config-lookup *configdat* "setup" "maxretries")))
    (set! max-retries (if (and max-retries (string->number max-retries))(string->number max-retries) 100))
    (if (not (null? sorted-test-names))
  (if regful
	(let loop ((hed         (car sorted-test-names))
		   (tal         (cdr sorted-test-names))
		   (reruns      '()))
	  (if (not (null? reruns))(debug:print-info 4 "reruns=" reruns))
	  ;; (print "Top of loop, hed=" hed ", tal=" tal " ,reruns=" reruns)
	  (let* ((test-record (hash-table-ref test-records hed))
		 (test-name   (tests:testqueue-get-testname test-record))
		 (tconfig     (tests:testqueue-get-testconfig test-record))
		 (testmode    (let ((m (config-lookup tconfig "requirements" "mode")))
				(if m (string->symbol m) 'normal)))
		 (waitons     (tests:testqueue-get-waitons    test-record))
		 (priority    (tests:testqueue-get-priority   test-record))
		 (itemdat     (tests:testqueue-get-itemdat    test-record)) ;; itemdat can be a string, list or #f
		 (items       (tests:testqueue-get-items      test-record))
		 (item-path   (item-list->path itemdat))
		 (newtal      (append tal (list hed))))
	    
      (if (null? reg) ;; doesn't make sense, this is probably NOT the problem of the car
	    (debug:print 6
			 "test-name: " test-name
			 "\n  hed:         " hed
			 "\n  itemdat:     " itemdat
			 "\n  items:       " items
			 "\n  item-path:   " item-path
			 "\n  waitons:     " waitons
			 "\n  num-retries: " num-retries
			 "\n  tal:         " tal
	  (car tal)
			 "\n  reruns:      " reruns)

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

      (car tal)))
	    (cond ;; OUTER COND
	     ((not items) ;; when false the test is ok to be handed off to launch (but not before)
	      (let* ((run-limits-info         (open-run-close runs:can-run-more-tests test-record)) ;; look at the test jobgroup and tot jobs running
		     (have-resources          (car run-limits-info))
		     (num-running             (list-ref run-limits-info 1))
		     (num-running-in-jobgroup (list-ref run-limits-info 2))
		     (max-concurrent-jobs     (list-ref run-limits-info 3))
		     (job-group-limit         (list-ref run-limits-info 4))
		     (prereqs-not-met         (open-run-close db:get-prereqs-not-met #f run-id waitons item-path mode: testmode))
		     (fails                   (runs:calc-fails prereqs-not-met))
		     (non-completed           (runs:calc-not-completed prereqs-not-met)))
		(debug:print-info 8 "have-resources: " have-resources " prereqs-not-met: " 
			     (string-intersperse 
			      (map (lambda (t)
				     (if (vector? t)
					 (conc (db:test-get-state t) "/" (db:test-get-status t))
					 (conc " WARNING: t is not a vector=" t )))
				   prereqs-not-met) ", ") " fails: " fails)
		(debug:print-info 4 "hed=" hed "\n  test-record=" test-record "\n  test-name: " test-name "\n  item-path: " item-path "\n  test-patts: " test-patts)

		;; Don't know at this time if the test have been launched at some time in the past
		;; i.e. is this a re-launch?
		(debug:print-info 4 "run-limits-info = " run-limits-info)
		(cond ;; INNER COND #1 for a launchable test
		 ;; Check item path against item-patts
		 ((not (tests:match test-patts (tests:testqueue-get-testname test-record) item-path)) ;; This test/itempath is not to be run
		  ;; else the run is stuck, temporarily or permanently
		  ;; but should check if it is due to lack of resources vs. prerequisites
		  (debug:print-info 1 "Skipping " (tests:testqueue-get-testname test-record) " " item-path " as it doesn't match " test-patts)
		  ;; (thread-sleep! *global-delta*)
		  (if (not (null? tal))
		      (loop (car tal)(cdr tal) reruns)))
		 ( ;; (and
		  (not (hash-table-ref/default test-registery (runs:make-full-test-name test-name item-path) #f))
		      ;; (and max-concurrent-jobs (> (- max-concurrent-jobs num-running) 5)))
		  (debug:print-info 4 "Pre-registering test " test-name "/" item-path " to create placeholder" )
		  (open-run-close db:tests-register-test #f run-id test-name item-path)
		  (hash-table-set! test-registery (runs:make-full-test-name test-name item-path) #t)
		  ;; (thread-sleep! *global-delta*)
(runs:shrink-can-run-more-tests-delay)
(define (runs:queue-next-tal tal reg n regful)
		  (loop (car newtal)(cdr newtal) reruns))
		 ((not have-resources) ;; simply try again after waiting a second
		  (debug:print-info 1 "no resources to run new tests, waiting ...")
		  ;; (thread-sleep! (+ 2 *global-delta*))
		  ;; could have done hed tal here but doing car/cdr of newtal to rotate tests
		  (loop (car newtal)(cdr newtal) reruns))
		 ((and have-resources
  (if regful
		       (or (null? prereqs-not-met)
			   (and (eq? testmode 'toplevel)
				(null? non-completed))))
		  (run:test run-id runname keyvallst test-record flags #f)
(runs:shrink-can-run-more-tests-delay)
		  ;; (thread-sleep! *global-delta*)
		  (if (not (null? tal))
		      (loop (car tal)(cdr tal) reruns)))
      tal
      (let ((newtal (cdr tal)))
		 (else ;; must be we have unmet prerequisites
		    (debug:print 4 "FAILS: " fails)
		    ;; If one or more of the prereqs-not-met are FAIL then we can issue
		    ;; a message and drop hed from the items to be processed.
		    (if (null? fails)
			(begin
	(if (null? newtal)
	    reg
			  ;; couldn't run, take a breather
			  (debug:print-info 4 "Shouldn't really get here, race condition? Unable to launch more tests at this moment, killing time ...")
			  ;; (thread-sleep! (+ 0.01 *global-delta*)) ;; long sleep here - no resources, may as well be patient
			  ;; we made new tal by sticking hed at the back of the list
			  (loop (car newtal)(cdr newtal) reruns))
	    newtal
			;; the waiton is FAIL so no point in trying to run hed ever again
			(if (not (null? tal))
			    (if (vector? hed)
				(begin (debug:print 1 "WARN: Dropping test " (db:test-get-testname hed) "/" (db:test-get-item-path hed)
						    " from the launch list as it has prerequistes that are FAIL")
(runs:shrink-can-run-more-tests-delay)
				       ;; (thread-sleep! *global-delta*)
				       (loop (car tal)(cdr tal) (cons hed reruns)))
				(begin
				  (debug:print 1 "WARN: Test not processed correctly. Could be a race condition in your test implementation? " hed) ;;  " as it has prerequistes that are FAIL. (NOTE: hed is not a vector)")
(runs:shrink-can-run-more-tests-delay)
				  ;; (thread-sleep! (+ 0.01 *global-delta*))
				  (loop hed tal reruns))))))))) ;; END OF INNER COND
	     
	    ))))
	     ;; case where an items came in as a list been processed
	     ((and (list? items)     ;; thus we know our items are already calculated
		   (not   itemdat)) ;; and not yet expanded into the list of things to be done
	      (if (and (debug:debug-mode 1) ;; (>= *verbosity* 1)
		       (> (length items) 0)
		       (> (length (car items)) 0))
		  (pp items))
	      (for-each
	       (lambda (my-itemdat)
		 (let* ((new-test-record (let ((newrec (make-tests:testqueue)))
					   (vector-copy! test-record newrec)
					   newrec))
			(my-item-path (item-list->path my-itemdat)))
		   (if (tests:match test-patts hed my-item-path) ;; (patt-list-match my-item-path item-patts)           ;; yes, we want to process this item, NOTE: Should not need this check here!
		       (let ((newtestname (runs:make-full-test-name hed my-item-path)))    ;; test names are unique on testname/item-path
			 (tests:testqueue-set-items!     new-test-record #f)
			 (tests:testqueue-set-itemdat!   new-test-record my-itemdat)
			 (tests:testqueue-set-item_path! new-test-record my-item-path)
			 (hash-table-set! test-records newtestname new-test-record)
			 (set! tal (cons newtestname tal)))))) ;; since these are itemized create new test names testname/itempath
	       items)
	      (if (not (null? tal))
		  (begin
		    (debug:print-info 4 "End of items list, looping with next after short delay")
                    ;; (thread-sleep! (+ 0.01 *global-delta*))
		    (loop (car tal)(cdr tal) reruns))))

	     ;; if items is a proc then need to run items:get-items-from-config, get the list and loop 
	     ;;    - but only do that if resources exist to kick off the job
	     ((or (procedure? items)(eq? items 'have-procedure))
	      (let ((can-run-more    (runs:can-run-more-tests test-record)))
		(if (and (list? can-run-more)
(define (runs:queue-next-reg tal reg n regful)
  (if regful
			 (car can-run-more))
		    (let* ((prereqs-not-met (open-run-close db:get-prereqs-not-met #f run-id waitons item-path mode: testmode))
			   (fails           (runs:calc-fails prereqs-not-met))
			   (non-completed   (runs:calc-not-completed prereqs-not-met)))
		      (debug:print-info 8 "can-run-more: " can-run-more
				   "\n testname:        " hed
				   "\n prereqs-not-met: " (runs:pretty-string prereqs-not-met)
				   "\n non-completed:   " (runs:pretty-string non-completed) 
				   "\n fails:           " (runs:pretty-string fails)
				   "\n testmode:        " testmode
				   "\n num-retries:     " num-retries
				   "\n (eq? testmode 'toplevel): " (eq? testmode 'toplevel)
				   "\n (null? non-completed):    " (null? non-completed)
				   "\n reruns:          " reruns
      (cdr reg)
				   "\n items:           " items
				   "\n can-run-more:    " can-run-more)
		      ;; (thread-sleep! (+ 0.01 *global-delta*))
		      (cond ;; INNER COND #2
		       ((or (null? prereqs-not-met) ;; all prereqs met, fire off the test
			    ;; or, if it is a 'toplevel test and all prereqs not met are COMPLETED then launch
			    (and (eq? testmode 'toplevel)
      (if (eq? (length tal) 1)
				 (null? non-completed)))
			(let ((test-name (tests:testqueue-get-testname test-record)))
			  (setenv "MT_TEST_NAME" test-name) ;; 
			  (setenv "MT_RUNNAME"   runname)
			  (set-megatest-env-vars run-id) ;; these may be needed by the launching process
			  (let ((items-list (items:get-items-from-config tconfig)))
			    (if (list? items-list)
				(begin
	  '()
				  (tests:testqueue-set-items! test-record items-list)
				  ;; (thread-sleep! *global-delta*)
				  (loop hed tal reruns))
				(begin
				  (debug:print 0 "ERROR: The proc from reading the setup did not yield a list - please report this")
				  (exit 1))))))
	  reg)))
		       ((null? fails)
			(debug:print-info 4 "fails is null, moving on in the queue but keeping " hed " for now")
			;; only increment num-retries when there are no tests runing
			(if (eq? 0 (list-ref can-run-more 1))
			    (begin
			      (if (> num-retries 100) ;; first 100 retries are low time cost
				  (thread-sleep! (+ 2 *global-delta*))
				  (thread-sleep! (+ 0.01 *global-delta*)))
			      (set! num-retries (+ num-retries 1))))
			(if (> num-retries  max-retries)
			    (if (not (null? tal))
				(loop (car tal)(cdr tal) reruns))
			    (loop (car newtal)(cdr newtal) reruns))) ;; an issue with prereqs not yet met?
		       ((and (not (null? fails))(eq? testmode 'normal))
			(debug:print-info 1 "test "  hed " (mode=" testmode ") has failed prerequisite(s); "
				     (string-intersperse (map (lambda (t)(conc (db:test-get-testname t) ":" (db:test-get-state t)"/"(db:test-get-status t))) fails) ", ")
				     ", removing it from to-do list")
			(if (not (null? tal))
			    (begin
                              ;; (thread-sleep! *global-delta*)
			      (loop (car tal)(cdr tal)(cons hed reruns)))))
		       (else
			(debug:print 8 "ERROR: No handler for this condition.")
			(thread-sleep! (+ 1 *global-delta*))
			(loop (car newtal)(cdr newtal) reruns)))) ;; END OF IF CAN RUN MORE

		    ;; if can't run more just loop with next possible test
		    (begin
		      (debug:print-info 4 "processing the case with a lambda for items or 'have-procedure. Moving through the queue without dropping " hed)
		      ;; (thread-sleep! (+ 2 *global-delta*))
		      (loop (car newtal)(cdr newtal) reruns))))) ;; END OF (or (procedure? items)(eq? items 'have-procedure))
	     
(include "run-tests-queue-classic.scm")
	     ;; this case should not happen, added to help catch any bugs
	     ((and (list? items) itemdat)
	      (debug:print 0 "ERROR: 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 "full drop through, if reruns is less than 100 we will force retry them, reruns=" reruns ", tal=" tal)
		(if (< num-retries max-retries)
		    (set! newlst (append reruns newlst)))
		(set! num-retries (+ num-retries 1))
		;; (thread-sleep! (+ 1 *global-delta*))
		(if (not (null? newlst))
		    ;; since reruns have been tacked on to newlst create new reruns from junked
		    (loop (car newlst)(cdr newlst)(delete-duplicates junked)))))
	     ((not (null? tal))
	      (debug:print-info 4 "I'm pretty sure I shouldn't get here."))
	     (else
	      (debug:print-info 4 "Exiting loop with...\n  hed=" hed "\n  tal=" tal "\n  reruns=" reruns))
	     )))) ;; LET* ((test-record

(include "run-tests-queue-new.scm")
    ;; we get here on "drop through" - loop for next test in queue
    ;; FIXME!!!! THIS SHOULD NOT REQUIRE AN EXIT!!!!!!!
    
    (debug:print-info 1 "All tests launched")
    (thread-sleep! 0.5)
    ;; FIXME! This harsh exit should not be necessary....
    ;; (if (not *runremote*)(exit)) ;; 
    #f)) ;; return a #f as a hint that we are done
  ;; Here we need to check that all the tests remaining to be run are eligible to run
  ;; and are not blocked by failed
  

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

		 )
    ;; setting itemdat to a list if it is #f
    (if (not itemdat)(set! itemdat '()))
    (set! item-path (item-list->path itemdat))
    (debug:print 2 "Attempting to launch test " test-name (if (equal? item-path "/") "/" item-path))
    (setenv "MT_TEST_NAME" test-name) ;; 
    (setenv "MT_RUNNAME"   runname)
    (set-megatest-env-vars run-id) ;; these may be needed by the launching process
    (set-megatest-env-vars run-id inrunname: runname) ;; these may be needed by the launching process
    (change-directory *toppath*)

    ;; Here is where the test_meta table is best updated
    ;; Yes, another use of a global for caching. Need a better way?
    (if (not (hash-table-ref/default *test-meta-updated* test-name #f))
        (begin
	   (hash-table-set! *test-meta-updated* test-name #t)
           (open-run-close runs:update-test_meta db test-name test-conf)))
           (runs:update-test_meta test-name test-conf)))
    
    ;; (lambda (itemdat) ;;; ((ripeness "overripe") (temperature "cool") (season "summer"))
    (let* ((new-test-path (string-intersperse (cons test-path (map cadr itemdat)) "/"))
	   (new-test-name (if (equal? item-path "") test-name (conc test-name "/" item-path))) ;; just need it to be unique
	   (test-id       (cdb:remote-run db:get-test-id #f  run-id test-name item-path))
	   (testdat       (cdb:get-test-info-by-id *runremote* test-id)))
      (if (not testdat)
	  (begin
	    ;; ensure that the path exists before registering the test
	    ;; NOPE: Cannot! Don't know yet which disk area will be assigned....
	    ;; (system (conc "mkdir -p " new-test-path))
	    ;;
	    ;; (open-run-close tests:register-test db run-id test-name item-path)
	    ;;
	    ;; NB// for the above line. I want the test to be registered long before this routine gets called!
	    ;;
	    (set! test-id (open-run-close db:get-test-id db run-id test-name item-path))
	    (if (not test-id)
		(begin
		  (debug:print 2 "WARN: Test not pre-created? test-name=" test-name ", item-path=" item-path ", run-id=" run-id)
		  (open-run-close db:tests-register-test #f run-id test-name item-path)
		  (cdb:tests-register-test *runremote* run-id test-name item-path)
		  (set! test-id (open-run-close db:get-test-id db run-id test-name item-path))))
	    (debug:print-info 4 "test-id=" test-id ", run-id=" run-id ", test-name=" test-name ", item-path=\"" item-path "\"")
	    (set! testdat (cdb:get-test-info-by-id *runremote* test-id))))
      (set! test-id (db:test-get-id testdat))
      (change-directory test-path)
      (case (if force ;; (args:get-arg "-force")
		'NOT_STARTED

	   (if (not runflag)
	       (if (not parent-test)
		   (debug:print 1 "NOTE: Not starting test " new-test-name " as it is state \"" (test:get-state testdat) 
				"\" and status \"" (test:get-status testdat) "\", use -rerun \"" (test:get-status testdat)
                                "\" or -force to override"))
	       ;; NOTE: No longer be checking prerequisites here! Will never get here unless prereqs are
	       ;;       already met.
	       ;; This would be a great place to do the process-fork
	       (if (not (launch-test #f run-id runname test-conf keyvallst test-name test-path itemdat flags))
	       (if (not (launch-test test-id run-id run-info key-vals runname test-conf keyvallst test-name test-path itemdat flags))
		   (begin
		     (print "ERROR: Failed to launch the test. Exiting as soon as possible")
		     (set! *globalexitstatus* 1) ;; 
		     (process-signal (current-process-id) signal/kill))))))
	((KILLED) 
	 (debug:print 1 "NOTE: " new-test-name " is already running or was explictly killed, use -force to launch it."))
	((LAUNCHED REMOTEHOSTSTART RUNNING)  

 	    ;;     	 (args:get-arg "-runtests")))
	    ;;     (client:setup) ;; This is a duplicate startup!!!??? BUG?
	    ;;     ))
	(set! keys (open-run-close db:get-keys db))
	;; 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))) 
		   (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 0 "ERROR: [" (args:get-arg "-reqtarg") "] not found in " runconfigf)
		    (if db (sqlite3:finalize! db))
		    (exit 1))))
	    (if (args:get-arg "-target")

		      (debug:print-info 0 "Skipping lock/unlock on " run-id))))
	      runs)))
;;======================================================================
;; Rollup runs
;;======================================================================

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

;; Update test_meta for all tests
(define (runs:update-all-test_meta db)
  (let ((test-names (get-all-legal-tests)))
    (for-each 
     (lambda (test-name)
       (let* ((test-path    (conc *toppath* "/tests/" test-name))
	      (test-configf (conc test-path "/testconfig"))
	      (testexists   (and (file-exists? test-configf)(file-read-access? test-configf)))
	      ;; read configs with tricks turned off (i.e. no system)
	      (test-conf    (if testexists (read-config test-configf #f #f)(make-hash-table))))
	 ;; use the open-run-close instead of passing in db
	 (runs:update-test_meta #f test-name test-conf)))
	 (runs:update-test_meta test-name test-conf)))
     test-names)))

;; This could probably be refactored into one complex query ...
(define (runs:rollup-run keys keyvallst runname user) ;; was target, now keyvallst
  (debug:print 4 "runs:rollup-run, keys: " keys " keyvallst: " keyvallst " :runname " runname " user: " user)
  (let* ((db              #f) ;; (keyvalllst      (keys:target->keyval keys target))
	 (new-run-id      (open-run-close runs:register-run db keys keyvallst runname "new" "n/a" user))
	 (new-run-id      (cdb:remote-run db:register-run #f keys keyvallst runname "new" "n/a" user))
	 (prev-tests      (open-run-close test:get-matching-previous-test-run-records db new-run-id "%" "%"))
	 (curr-tests      (open-run-close db:get-tests-for-run db new-run-id "%/%" '() '()))
	 (curr-tests-hash (make-hash-table)))
    (open-run-close db:update-run-event_time db new-run-id)
    ;; index the already saved tests by testname and itemdat in curr-tests-hash
    (for-each
     (lambda (testdat)

;; Copyright 2006-2012, Matthew Welland.
;; 
;;  This program is made available under the GNU GPL version 2.0 or
;;  greater. See the accompanying file COPYING for details.
;; 
;;  This program is distributed WITHOUT ANY WARRANTY; without even the
;;  implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
;;  PURPOSE.

(require-extension (srfi 18) extras tcp s11n)

(use srfi-1 posix regex regex-case srfi-69 hostinfo md5 message-digest zmq)
(use srfi-1 posix regex regex-case srfi-69 hostinfo md5 message-digest)
;; (use zmq)

(use spiffy uri-common intarweb http-client spiffy-request-vars)

(declare (unit server))

(declare (uses common))
(declare (uses db))
(declare (uses tasks)) ;; tasks are where stuff is maintained about what is running.
(declare (uses synchash))
(declare (uses http-transport))
(declare (uses zmq-transport))
;; (declare (uses zmq-transport))
(declare (uses daemon))

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

(define (server:make-server-url hostport)
  (if (not hostport)

       (begin
	 (debug:print 0 "ERROR: tasks:server-get-server-id needs (hostname and pid) OR (iface and port) OR (hostname and port)")
	 "SELECT id FROM servers WHERE pid=-999;")))
     (if hostname hostname iface)(if pid pid port))
    res))

(define (tasks:server-update-heartbeat mdb server-id)
  (debug:print-info 0 "Heart beat update of server id=" server-id)
  (handle-exceptions
   exn
   (begin
     (debug:print 0 "WARNING: probable timeout on monitor.db access")
     (thread-sleep! 1)
     (tasks:server-update-heartbeat mdb server-id))
  (sqlite3:execute mdb "UPDATE servers SET heartbeat=strftime('%s','now') WHERE id=?;" server-id))
   (sqlite3:execute mdb "UPDATE servers SET heartbeat=strftime('%s','now') WHERE id=?;" server-id)))

;; alive servers keep the heartbeat field upto date with seconds every 6 or so seconds
(define (tasks:server-alive? mdb server-id #!key (iface #f)(hostname #f)(port #f)(pid #f))
  (let* ((server-id  (if server-id 
			 server-id
			 (tasks:server-get-server-id mdb hostname iface port pid)))
	 (heartbeat-delta 99e9))

	   (res      #f))
      ;; (print "tests:glob-like-match => notpatt: " notpatt ", newpatt: " newpatt ", finpatt: " finpatt)
      (set! res (string-match (regexp finpatt (if like #t #f)) str))
      (if notpatt (not res) res))))

;; if itempath is #f then look only at the testname part
;;
(define (tests:match patterns testname itempath)
(define (tests:match patterns testname itempath #!key (required '()))
  (if (string? patterns)
      (let ((patts (string-split patterns ",")))
      (let ((patts (append (string-split patterns ",") required)))
	(if (null? patts) ;;; no pattern(s) means no match
	    #f
	    (let loop ((patt (car patts))
		       (tal  (cdr patts)))
	      ;; (print "loop: patt: " patt ", tal " tal)
	      (if (string=? patt "")
		  #f ;; nothing ever matches empty string - policy

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


;; Do not rpc this one, do the underlying calls!!!
(define (tests:test-set-status! test-id state status comment dat)
(define (tests:test-set-status! test-id state status comment dat #!key (work-area #f))
  (debug:print-info 4 "tests:test-set-status! test-id=" test-id ", state=" state ", status=" status ", dat=" dat)
  (let* ((db          #f)
	 (real-status status)
	 (otherdat    (if dat dat (make-hash-table)))
	 (testdat     (cdb:get-test-info-by-id *runremote* test-id))
	 (run-id      (db:test-get-run_id testdat))
	 (test-name   (db:test-get-testname   testdat))

    ;; update the primary record IF state AND status are defined
    (if (and state status)
	(cdb:test-set-status-state *runremote* test-id real-status state (if waived waived comment)))
    
    ;; if status is "AUTO" then call rollup (note, this one modifies data in test
    ;; run area, it does remote calls under the hood.
    (if (and test-id state status (equal? status "AUTO")) 
	(db:test-data-rollup #f test-id status))
	(db:test-data-rollup #f test-id status work-area: work-area))

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

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

			   variable ","
			   value    ","
			   expected ","
			   tol      ","
			   units    ","
			   dcomment ",," ;; extra comma for status
			   type     )))
	    ;; This was run remote, don't think that makes sense.
	    (cdb:remote-run db:csv->test-data #f test-id
	    (db:csv->test-data #f test-id
				dat))))
      
    ;; need to update the top test record if PASS or FAIL and this is a subtest
    (if (not (equal? item-path ""))
	(cdb:roll-up-pass-fail-counts *runremote* run-id test-name item-path status))

    (if (or (and (string? comment)

	 (lambda (count)
	   (set! res count))
	 tdb
	 "SELECT count(id) FROM test_rundat;")
	res))
  0)

(define (db:update-central-meta-info db test-id cpuload diskfree minutes num-records uname hostname)
  (sqlite3:execute db "UPDATE tests SET cpuload=?,diskfree=? WHERE id=?;"
		   cpuload
		   diskfree
		   test-id)
  (if minutes (sqlite3:execute db "UPDATE tests SET run_duration=? WHERE id=?;" minutes test-id))
  (if (eq? num-records 0)
      (sqlite3:execute db "UPDATE tests SET uname=?,host=? WHERE id=?;"
(define (tests:update-central-meta-info test-id cpuload diskfree minutes num-records uname hostname)
  ;; This is a good candidate for threading the requests to enable
  ;; transactionized write at the server
  (cdb:tests-update-cpuload-diskfree *runremote* test-id cpuload diskfree)
  ;; (let ((db (open-db)))
    ;; (sqlite3:execute db "UPDATE tests SET cpuload=?,diskfree=? WHERE id=?;"
    ;;     	     cpuload
    ;;     	     diskfree
    ;;     	     test-id)
    (if minutes 
	(cdb:tests-update-run-duration *runremote* test-id minutes))
	;; (sqlite3:execute db "UPDATE tests SET run_duration=? WHERE id=?;" minutes test-id))
    (if (eq? num-records 0)
	(cdb:tests-update-uname-host *runremote* test-id uname hostname))
	;;(sqlite3:execute db "UPDATE tests SET uname=?,host=? WHERE id=?;" uname hostname test-id))
		       uname hostname test-id)))

(define (test-set-meta-info db test-id run-id testname itemdat minutes)
    ;;(sqlite3:finalize! db))
    )
  
(define (tests:set-meta-info db test-id run-id testname itemdat minutes work-area)
  ;; DOES cdb:remote-run under the hood!
  (let* ((tdb         (db:open-test-db-by-test-id db test-id))
  (let* ((tdb         (db:open-test-db-by-test-id db test-id work-area: work-area))
	 (num-records (test:tdb-get-rundat-count tdb))
	 (cpuload  (get-cpu-load))
	 (diskfree (get-df (current-directory))))
    (if (eq? (modulo num-records 10) 0) ;; every ten records update central
	(let ((uname    (get-uname "-srvpio"))
	      (hostname (get-host-name)))
	  (cdb:remote-run db:update-central-meta-info db test-id cpuload diskfree minutes num-records uname hostname)))
	  (tests:update-central-meta-info test-id cpuload diskfree minutes num-records uname hostname)))
    (sqlite3:execute tdb "INSERT INTO test_rundat (update_time,cpuload,diskfree,run_duration) VALUES (strftime('%s','now'),?,?,?);"
		     cpuload diskfree minutes)))
		     cpuload diskfree minutes)
    (sqlite3:finalize! tdb)))
	  
;;======================================================================
;; A R C H I V I N G
;;======================================================================

(define (test:archive db test-id)
  #f)

(define (test:archive-tests db keynames target)
  #f)

	cd fullrun;megatest -runtests %/,%/winter -target ubuntu/nfs/none :runname $(RUNNAME)_03  -debug $(DEBUG)
	sleep 40;cd fullrun;megatest -target ubuntu/nfs/none :runname $(RUNNAME) -set-state-status COMPLETED,FORCED :state COMPLETED :status PASS -testpatt ez_p%s,runfirst/ -debug $(DEBUG) $(LOGGING)


test3 : fullprep
	cd fullrun;$(MEGATEST) -runtests runfirst -reqtarg ubuntu/nfs/none :runname $(RUNNAME)_b -debug 10

test4 : fullprep
	cd fullrun;$(MEGATEST) -debug $(DEBUG) -runall -reqtarg ubuntu/nfs/none :runname $(RUNNAME)_b -m "This is a comment specific to a run" -v $(LOGGING)
test4 : cleanprep
	@echo "WARNING: No longer running fullprep, test converage may be lessened"
	cd fullrun;time $(MEGATEST) -debug $(DEBUG) -runtests % -reqtarg ubuntu/nfs/none :runname $(RUNNAME)_b -m "This is a comment specific to a run" -v $(LOGGING)

# NOTE: Only one instance can be a server
test5 : fullprep
	cd fullrun;sleep  0;$(MEGATEST) -runtests % -target $(TARGET) :runname $(RUNNAME)_aa -debug $(DEBUG) $(LOGGING) > aa.log 2> aa.log &
	cd fullrun;sleep 10;$(MEGATEST) -runtests % -target $(TARGET) :runname $(RUNNAME)_ab -debug $(DEBUG) $(LOGGING) > ab.log 2> ab.log &
	cd fullrun;sleep 10;$(MEGATEST) -runtests % -target $(TARGET) :runname $(RUNNAME)_ac -debug $(DEBUG) $(LOGGING) > ac.log 2> ac.log &
	cd fullrun;sleep 10;$(MEGATEST) -runtests % -target $(TARGET) :runname $(RUNNAME)_ad -debug $(DEBUG) $(LOGGING) > ad.log 2> ad.log &	
#	cd fullrun;sleep 10;$(MEGATEST) -runtests % -target $(TARGET) :runname $(RUNNAME)_ae -debug $(DEBUG) $(LOGGING) > ae.log 2> ae.log &	
#	cd fullrun;sleep 10;$(MEGATEST) -runtests % -target $(TARGET) :runname $(RUNNAME)_af -debug $(DEBUG) $(LOGGING) > af.log 2> af.log &	
test5 : cleanprep
	@echo "WARNING: No longer running fullprep, test converage may be lessened"
	cd fullrun;sleep 0;$(MEGATEST) -runtests % -target $(TARGET) :runname $(RUNNAME)_aa -debug $(DEBUG) $(LOGGING) > aa.log 2> aa.log &
	cd fullrun;sleep 0;$(MEGATEST) -runtests % -target $(TARGET) :runname $(RUNNAME)_ab -debug $(DEBUG) $(LOGGING) > ab.log 2> ab.log &
	cd fullrun;sleep 0;$(MEGATEST) -runtests % -target $(TARGET) :runname $(RUNNAME)_ac -debug $(DEBUG) $(LOGGING) > ac.log 2> ac.log &
	cd fullrun;sleep 0;$(MEGATEST) -runtests % -target $(TARGET) :runname $(RUNNAME)_ad -debug $(DEBUG) $(LOGGING) > ad.log 2> ad.log &	
#	cd fullrun;sleep 0;$(MEGATEST) -runtests % -target $(TARGET) :runname $(RUNNAME)_ae -debug $(DEBUG) $(LOGGING) > ae.log 2> ae.log &	
#	cd fullrun;sleep 0;$(MEGATEST) -runtests % -target $(TARGET) :runname $(RUNNAME)_af -debug $(DEBUG) $(LOGGING) > af.log 2> af.log &	

test6: fullprep
	cd fullrun;$(MEGATEST) -runtests runfirst -testpatt %/1 -reqtarg ubuntu/nfs/none :runname $(RUNNAME)_itempatt -v
	cd fullrun;$(MEGATEST) -runtests runfirst -testpatt %blahha% -reqtarg ubuntu/nfs/none :runname $(RUNNAME)_itempatt -debug 10
	cd fullrun;$(MEGATEST) -rollup :runname newrun -target ubuntu/nfs/none -debug 10


cleanprep : ../*.scm Makefile */*.config
	mkdir -p /tmp/mt_runs /tmp/mt_links
	mkdir -p fullrun/tmp/mt_runs fullrun/tmp/mt_links
	cd ..;make;make install
	rm -f */logging.db
	touch cleanprep

fullprep : cleanprep
	cd fullrun;$(MEGATEST) -remove-runs :runname $(RUNNAME)% -target %/%/% -testpatt %/%
	cd fullrun;$(BINPATH)/dashboard -rows 15 &

[setup]
testcopycmd cp --remove-destination -rlv TEST_SRC_PATH/. TEST_TARG_PATH/.
testcopycmd cp --remove-destination -rlv TEST_SRC_PATH/. TEST_TARG_PATH/. >> TEST_TARG_PATH/mt_launch.log 2>> TEST_TARG_PATH/mt_launch.log
runqueue 2

[include ../fdk.config]

[server]
timeout 0.01

#!/bin/sh
if [ $NUMBER -lt 200 ];then 
sleep 10
   sleep $NUMBER
else
   sleep 200
fi

exit 0

# Add additional steps here. Format is "stepname script"
[ezsteps]
step1 step1.sh

# Test requirements are specified here
[requirements]
# waiton setup
priority 0

# Iteration for your tests are controlled by the items section
[items]
NUMBER #{scheme (string-intersperse (map number->string (sort (let loop ((a 0)(res '()))(if (< a 120)(loop (+ a 1)(cons a res)) res)) >)) " ")}
NUMBER #{scheme (string-intersperse (map number->string (sort (let loop ((a 0)(res '()))(if (< a (or (any->number (get-environment-variable "NUMTESTS")) 1100))(loop (+ a 1)(cons a res)) res)) >)) " ")}

# test_meta is a section for storing additional data on your test
[test_meta]
author matt
owner  matt
description An example test
tags tagone,tagtwo

[setup]
# exectutable /path/to/megatest
max_concurrent_jobs 200
max_concurrent_jobs 150

linktree #{getenv MT_RUN_AREA_HOME}/tmp/mt_links

[jobtools]
useshell yes
# ## launcher launches jobs, the job is managed on the target host
## by megatest, comment out launcher to run local

[fields]
sysname TEXT
fsname TEXT
datapath TEXT

# refareas can be searched to find previous runs
# the path points to where megatest.db exists
[refareas]
area1 /tmp/oldarea/megatest

[include config/mt_include_1.config]

[setup]
# Set launchwait to yes to use the old launch run code that waits for the launch process to return before 
# proceeding.
# launchwait yes

# If defined the runs:run-tests-queue-new queue code is used with the register test depth
# given. Otherwise the old code is used. The old code will be removed in the future and 
# a default of 10 used.
# runqueue 2

# It is possible (but not recommended) to override the rsync command used
# to populate the test directories. For test development the following 
# example can be useful
#
testcopycmd cp --remove-destination -rsv TEST_SRC_PATH/. TEST_TARG_PATH/.
# testcopycmd cp --remove-destination -rsv TEST_SRC_PATH/. TEST_TARG_PATH/. >> TEST_TARG_PATH/mt_launch.log 2>> TEST_TARG_PATH/mt_launch.log

# or for hard links

# testcopycmd cp --remove-destination -rlv TEST_SRC_PATH/. TEST_TARG_PATH/.

# FULL or 2, NORMAL or 1, OFF or 0
synchronous OFF

WACKYVAR4 #{scheme (+ 5 6 7)}
WACKYVAR5 #{getenv sysname}/#{getenv fsname}/#{getenv datapath}
WACKYVAR6 #{scheme (args:get-arg "-target")}
PREDICTABLE the_ans
MRAH MT_RUN_AREA_HOME=#{getenv MT_RUN_AREA_HOME}
# The empty var should have a definition with null string
EMPTY_VAR

WRAPPEDVAR This var should have the work blah thrice: \
blah \
blah

# XTERM   [system xterm]
# RUNDEAD [system exit 56]

[server]

# If the server can't be started on this port it will try the next port until
# it succeeds
port 8080

# This server will keep running this number of hours after last access. 
# Three minutes is 0.05 hours
timeout 0.05
timeout 0.025

## disks are:
## name host:/path/to/area
## -or-
## name /path/to/area
[disks]
disk0 /foobarbazz
[include config/mt_include_2.config]

[include #{getenv USER}_testing.config]

[default]
SOMEVAR This should show up in SOMEVAR3

[include #{getenv MT_RUN_AREA_HOME}/common_runconfigs.config]

# #{system echo 'VACKYVAR #{shell pwd}' > $MT_RUN_AREA_HOME/config/$USER.config}
[include ./config/#{getenv USER}.config]

WACKYVAR0 #{get ubuntu/nfs/none CURRENT}
WACKYVAR1 #{scheme (args:get-arg "-target")}

[default/ubuntu/nfs]
WACKYVAR2 #{runconfigs-get CURRENT}

[ubuntu/nfs/none]
WACKYVAR2 #{runconfigs-get CURRENT}
SOMEVAR2  This should show up in SOMEVAR4 if the target is ubuntu/nfs/none

[default]
SOMEVAR3 #{rget SOMEVAR}
SOMEVAR4 #{rget SOMEVAR2}
SOMEVAR5 #{runconfigs-get SOMEVAR2}