Megatest

HELPERS=$(addprefix $(PREFIX)/bin/,$(ADTLSCR))
DEPLOYHELPERS=$(addprefix deploytarg/,$(ADTLSCR))
MTESTHASH=$(shell fossil info|grep checkout:| awk '{print $$2}')

CSIPATH=$(shell which csi)
CKPATH=$(shell dirname $(shell dirname $(CSIPATH)))

all : mtest dboard newdboard

mtest: $(OFILES) megatest.o
	csc $(CSCOPTS) $(OFILES) megatest.o -o mtest

dboard : $(OFILES) $(GOFILES) dashboard.scm
	csc $(OFILES) dashboard.scm $(GOFILES) -o dboard

newdboard : newdashboard.scm $(OFILES) $(GOFILES)
	csc $(OFILES) $(GOFILES) newdashboard.scm -o newdboard

deploytarg/libiupcd.so : $(CKPATH)/lib/libiupcd.so
	for i in iup im cd av call sqlite; do \
	  cp $(CKPATH)/lib/lib$$i* deploytarg/ ; \
	done
	cp $(CKPATH)/include/*.h deploytarg

$(PREFIX)/bin/mtest : mtest
	@echo Installing to PREFIX=$(PREFIX)
	$(INSTALL) mtest $(PREFIX)/bin/mtest
	utils/mk_wrapper $(PREFIX) mtest > $(PREFIX)/bin/megatest
	chmod a+x $(PREFIX)/bin/megatest

$(PREFIX)/bin/newdboard : newdboard
	$(INSTALL) newdboard $(PREFIX)/bin/newdboard
	utils/mk_wrapper $(PREFIX) newdboard > $(PREFIX)/bin/newdashboard
	chmod a+x $(PREFIX)/bin/newdashboard

$(HELPERS) : utils/mt_* 
	$(INSTALL) $< $@
	chmod a+x $@

$(DEPLOYHELPERS) : utils/mt_*
	$(INSTALL) $< $@

# install dashboard as dboard so wrapper script can be called dashboard
$(PREFIX)/bin/dboard : dboard $(FILES)
	$(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 \
	chicken-install -prefix deploytarg -deploy $$i;done

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

   (read-config "megatest.config" #f #t)
   "disks" '("none" "")))

;;======================================================================
;; System stuff
;;======================================================================

;; return a nice clean pathname made absolute
(define (nice-path dir)
  (normalize-pathname (if (absolute-pathname? dir)
			  dir
			  (conc (current-directory) "/" dir))))

(define (get-df path)
  (let* ((df-results (cmd-run->list (conc "df " path)))
	 (space-rx   (regexp "([0-9]+)\\s+([0-9]+)%"))
	 (freespc    #f))
    ;; (write df-results)
    (for-each (lambda (l)
		(let ((match (string-search space-rx l)))

(define (common:name->iup-color name)
  (case (string->symbol (string-downcase name))
    ((red)    "223 33 49")
    ((grey)   "192 192 192")
    ((orange) "255 172 13")
    ((purple) "This is unfinished ...")))

(define (common:get-color-for-state-status state status)
  (case (string->symbol state)
    ((COMPLETED)
     (case (string->symbol status)
       ((PASS)        "70  249 73")


       ((WARN WAIVED) "255 172 13")
       ((SKIP)        "230 230 0")
       (else "223 33 49")))
    ((LAUNCHED)         "101 123 142")
    ((CHECK)            "255 100 50")
    ((REMOTEHOSTSTART)  "50  130 195")
    ((RUNNING)          "9   131 232")
    ((KILLREQ)          "39  82  206")
    ((KILLED)           "234 101 17")
    ((NOT_STARTED)      "240 240 240")
    (else               "192 192 192")))

(define (common:get-color-from-status status)
  (cond
   ((equal? status "PASS")    "green")
   ((equal? status "FAIL")    "red")
   ((equal? status "WARN")    "orange")
   ((equal? status "KILLED")  "orange")
   ((equal? status "KILLREQ") "purple")
   ((equal? status "RUNNING") "blue")
   (else "black")))

;;  PURPOSE.
;;======================================================================

;;======================================================================
;; Config file handling
;;======================================================================

(use regex regex-case directory-utils)
(declare (unit configf))
(declare (uses common))
(declare (uses process))

(include "common_records.scm")

;; return list (path fullpath configname)

;; envion-patt is a regex spec that identifies sections that will be eval'd
;; in the environment on the fly
;; sections: #f => get all, else list of sections to gather
(define (read-config path ht allow-system #!key (environ-patt #f)(curr-section #f)(sections #f))
  (debug:print-info 5 "read-config " path " allow-system " allow-system " environ-patt " environ-patt " curr-section: " curr-section " sections: " sections " pwd: " (current-directory))
  (if (not (file-exists? path))
      (begin 
	(debug:print-info 1 "read-config - file not found " path " current path: " (current-directory))
	(if (not ht)(make-hash-table) ht))
      (let ((inp        (open-input-file path))
	    (res        (if (not ht)(make-hash-table) ht)))
	(let loop ((inl               (configf:read-line inp res allow-system)) ;; (read-line inp))
		   (curr-section-name (if curr-section curr-section "default"))
		   (var-flag #f);; turn on for key-var-pr and cont-ln-rx, turn off elsewhere
		   (lead     #f))
	  (debug:print-info 8 "curr-section-name: " curr-section-name " var-flag: " var-flag "\n   inl: \"" inl "\"")
	  (if (eof-object? inl) 
	      (begin
		(close-input-port inp)
		(hash-table-delete! res "") ;; we are using "" as a dumping ground and must remove it before returning the ht
		res)
	      (regex-case 
	       inl 
	       (configf:comment-rx _                  (loop (configf:read-line inp res allow-system) curr-section-name #f #f))
	       (configf:blank-l-rx _                  (loop (configf:read-line inp res allow-system) curr-section-name #f #f))
	       (configf:include-rx ( x include-file ) (let* ((curr-conf-dir (pathname-directory path))
							     (full-conf     (if (absolute-pathname? include-file)
										include-file
										(nice-path 
										 (conc (if curr-conf-dir
											   curr-conf-dir
											   ".")
										       "/" include-file)))))
							(if (file-exists? full-conf)
							    (begin
							      ;; (push-directory conf-dir)
							      (read-config full-conf res allow-system environ-patt: environ-patt curr-section: curr-section-name sections: sections)
							      ;; (pop-directory)
							      (loop (configf:read-line inp res allow-system) curr-section-name #f #f))
							    (begin
							      (debug:print 2 "INFO: include file " include-file " not found (called from " path ")")
							      (debug:print 2 "        " full-conf)
							      (loop (configf:read-line inp res allow-system) curr-section-name #f #f)))))
	       (configf:section-rx ( x section-name ) (loop (configf:read-line inp res allow-system)
							    ;; if we have the sections list then force all settings into "" and delete it later?
							    (if (or (not sections) 
								    (member section-name sections))
								section-name "") ;; stick everything into ""
							    #f #f))
	       (configf:key-sys-pr ( x key cmd      ) (if allow-system

	    (let ((lbl   (iup:label (db:test-get-status testdat) #:expand "HORIZONTAL")))
	      (hash-table-set! widgets "teststatus"
			       (lambda (testdat)
				 (let ((newstatus (db:test-get-status testdat))
				       (oldstatus (iup:attribute lbl "TITLE")))
				   (if (not (equal? oldstatus newstatus))
				       (begin
					 (iup:attribute-set! lbl "FGCOLOR" (common:get-color-for-state-status (db:test-get-state testdat)
												       (db:test-get-status testdat)))
					 (iup:attribute-set! lbl "TITLE" (db:test-get-status testdat)))))))
	      lbl)
	    (store-label "testcomment"
			 (iup:label "TestComment                             "
				    #:expand "HORIZONTAL")
			 (lambda (testdat)

;; use a global for setting the buttons colors
;;                           state status teststeps
(define *state-status* (vector #f #f #f))
(define (update-state-status-buttons testdat)
  (let* ((state  (db:test-get-state  testdat))
	 (status (db:test-get-status testdat))
	 (color  (common:get-color-for-state-status state status)))
    ((vector-ref *state-status* 0) state color)
    ((vector-ref *state-status* 1) status color)))

;;======================================================================
;; Set fields 
;;======================================================================
(define (set-fields-panel test-id testdat)

	     (let* ((btns  (map (lambda (status)
				  (let ((btn (iup:button status
							 #:expand "HORIZONTAL" #:size "50x" #:font "Courier New, -10"
							 #:action (lambda (x)
								    (open-run-close db:test-set-state-status-by-id #f test-id #f status #f)
								    (db:test-set-status! testdat status)))))
				    btn))
				(list  "PASS" "WARN" "FAIL" "CHECK" "n/a" "WAIVED" "SKIP"))))
	       (vector-set! *state-status* 1
			    (lambda (status color)
			      (for-each 
			       (lambda (btn)
				 (let* ((name     (iup:attribute btn "TITLE"))
					(newcolor (if (equal? name status) color "192 192 192")))
				   (if (not (colors-similar? newcolor (iup:attribute btn "BGCOLOR")))

	  (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)
	       (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)))

				    (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))))
			       (cond
				((and need-update newtestdat)
				 (set! testdat newtestdat)
				 (set! teststeps    (db:get-compressed-steps test-id))
				 (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"))))))
	       (widgets      (make-hash-table))
	       (meta-widgets (make-hash-table))
	       (self         #f)
	       (store-label  (lambda (name lbl cmd)
			       (hash-table-set! widgets name 

					    (iup:button "Run Test"    #:action run-test    #:size "80x")
					    (iup:button "Clean Test"  #:action remove-test #:size "80x")
					    (iup:button "Close"       #:action (lambda (x)(exit)) #:size "80x"))
					   (apply 
					    iup:hbox
					    (list command-text-box command-launch-button))))
			       (set-fields-panel test-id testdat)
			       (let ((tabs 
				      (iup:tabs
				       ;; Replace here with matrix
				       (let ((steps-matrix (iup:matrix




							    #:font   "Courier New, -8"
							    #:expand "YES"
							    #:scrollbar "YES"
							    #:numcol 6
							    #:numlin 30
							    #:numcol-visible 6
							    #:numlin-visible 5
							    #:click-cb (lambda (obj lin col status)
									 (if (equal? col 6)
									     (let ((fname (iup:attribute obj (conc lin ":" col))))
									       (viewlog fname)
									       (print "obj: " obj " lin: " lin " col: " col " status: " status)))))
							   ))
					 ;; (let loop ((count 0))
					 ;;   (iup:attribute-set! steps-matrix "FITTOTEXT" (conc "L" count))
					 ;;   (if (< count 30)
					 ;;       (loop (+ count 1))))
					 (iup:attribute-set! steps-matrix "0:1" "Step Name")
					 (iup:attribute-set! steps-matrix "0:2" "Start")
					 (iup:attribute-set! steps-matrix "0:3" "End")

					 (iup:attribute-set! steps-matrix "WIDTH3" "50")
					 (iup:attribute-set! steps-matrix "0:4" "Status")
					 (iup:attribute-set! steps-matrix "WIDTH4" "50")
					 (iup:attribute-set! steps-matrix "0:5" "Duration")
					 (iup:attribute-set! steps-matrix "0:6" "Log File")
					 (iup:attribute-set! steps-matrix "ALIGNMENT1" "ALEFT")
					 ;; (iup:attribute-set! steps-matrix "FIXTOTEXT" "C1")
					 (iup:attribute-set! steps-matrix "RESIZEMATRIX" "YES")
					 (let ((proc





						(lambda (testdat)
						  (if (not (null? teststeps))
						      (let loop ((hed    (car teststeps))
								 (tal    (cdr teststeps))
								 (rownum 1)
								 (colnum 1))
							(let ((val (vector-ref hed (- colnum 1))))

							  (iup:attribute-set! steps-matrix  (conc rownum ":" colnum)(if val (conc val) ""))
							  (if (< colnum 6)




							      (loop hed tal rownum (+ colnum 1))


							      (if (not (null? tal))
								  (loop (car tal)(cdr tal)(+ rownum 1) 1)))))))))
					   (hash-table-set! widgets "StepsMatrix" proc)
					   (proc testdat))
					 steps-matrix)
				       ;; populate the Test Data panel
				       (iup:frame
					#:title "Test Data"
					(let ((test-data
					       (iup:textbox  ;; #:action (lambda (obj char val)
						;;   	#f)
						#:expand "YES"
						#:multiline "YES"
						#:font "Courier New, -10"
						#:size "100x100")))
					  (hash-table-set! widgets "Test Data"
							   (lambda (testdat) ;; 
							     (let* ((currval (iup:attribute test-data "VALUE")) ;; "TITLE"))
								    (fmtstr  "~10a~10a~10a~10a~7a~7a~6a~6a~a") ;; category,variable,value,expected,tol,units,type,comment
								    (newval  (string-intersperse 
									      (append
									       (list 
										(format #f fmtstr "Category" "Variable" "Value" "Expected" "Tol" "Status" "Units" "Type" "Comment")
										(format #f fmtstr "========" "========" "=====" "========" "===" "======" "=====" "====" "======="))
									       (map (lambda (x)
										      (format #f fmtstr
											      (db:test-data-get-category x)
											      (db:test-data-get-variable x)
											      (db:test-data-get-value    x)
											      (db:test-data-get-expected x)
											      (db:test-data-get-tol      x)
											      (db:test-data-get-status   x)
											      (db:test-data-get-units    x)
											      (db:test-data-get-type     x)
											      (db:test-data-get-comment  x)))
										    (open-run-close db:read-test-data #f test-id "%")))
									      "\n")))
							       (if (not (equal? currval newval))
								   (iup:attribute-set! test-data "VALUE" newval ))))) ;; "TITLE" newval)))))
					  test-data)))))
				 (iup:attribute-set! tabs "TABTITLE0" "Steps")
				 (iup:attribute-set! tabs "TABTITLE1" "Test Data")
				 tabs))))
	    (iup:show self)
	    (iup:callback-set! *tim* "ACTION_CB"
			       (lambda (x)
				 ;; Now start keeping the gui updated from the db
				 (refreshdat) ;; update from the db here
					;(thread-suspend! other-thread)
				 ;; update the gui elements here

				(if (> (length parts) 1)(conc "  " (car (string-split (cadr parts) ")"))) newval))))
	      (vector-set! keycol i newval)
	      (iup:attribute-set! lbl "TITLE" munged-val)))
	(iup:attribute-set! lbl "FGCOLOR" (if (hash-table-ref/default *collapsed* newval #f) "0 112 112" "0 0 0"))
	(if (< i maxn)
	    (loop (+ i 1)))))))



















(define (update-buttons uidat numruns numtests)
  (if *please-update-buttons*
      (let* ((runs        (if (> (length *allruns*) numruns)
			      (take-right *allruns* numruns)
			      (pad-list *allruns* numruns)))
	     (lftcol      (dboard:uidat-get-lftcol uidat))
	     (tableheader (dboard:uidat-get-header uidat))

			       (testfullname (test:test-get-fullname test))
			       (teststatus (db:test-get-status   test))
			       (teststate  (db:test-get-state    test))
			       (teststart  (db:test-get-event_time test))
			       (runtime    (db:test-get-run_duration test))
			       (buttontxt  (if (equal? teststate "COMPLETED") teststatus teststate))
			       (button     (vector-ref columndat rown))
			       (color      (common:get-color-for-state-status teststate teststatus))
			       (curr-color (vector-ref buttondat 1)) ;; (iup:attribute button "BGCOLOR"))
			       (curr-title (vector-ref buttondat 2))) ;; (iup:attribute button "TITLE")))
			  (if (not (equal? curr-color color))
			      (iup:attribute-set! button "BGCOLOR" color))
			  (if (not (equal? curr-title buttontxt))
			      (iup:attribute-set! button "TITLE"   buttontxt))
			  (vector-set! buttondat 0 run-id)

	      iup:hbox
	      (map (lambda (status)
		     (iup:toggle status  #:action   (lambda (obj val)
						      (mark-for-update)
						      (if (eq? val 1)
							  (hash-table-set! *status-ignore-hash* status #t)
							  (hash-table-delete! *status-ignore-hash* status)))))
	      '("PASS" "FAIL" "WARN" "CHECK" "WAIVED" "STUCK/DEAD" "n/a" "SKIP")))
	     (apply 
	      iup:hbox
	      (map (lambda (state)
		     (iup:toggle state   #:action   (lambda (obj val)
						      (mark-for-update)
						      (if (eq? val 1)
							  (hash-table-set! *state-ignore-hash* state #t)

(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"))
	     (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"
			((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;")

			(list 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*))

(define (cdb:roll-up-pass-fail-counts serverdat run-id test-name item-path status)
  (cdb:client-call serverdat 'immediate #f *default-numtries* open-run-close db:roll-up-pass-fail-counts #f run-id test-name item-path status))

(define (cdb:get-test-info serverdat run-id test-name item-path)
  (cdb:client-call serverdat 'immediate #f *default-numtries* open-run-close db:get-test-info #f run-id test-name item-path))

(define db:special-queries   '(rollup-tests-pass-fail
			       db:roll-up-pass-fail-counts
                               login
                               immediate
			       flush
			       sync
			       set-verbosity
			       killserver
			       ))

;; not used, intended to indicate to run in calling process
(define db:run-local-queries '()) ;; rollup-tests-pass-fail))

(define (db:process-cached-writes db)
  (let ((queries    (make-hash-table))
	(data       #f))

    (debug:print-info 7 "Current write queue length is " queue-len)

    ;; poll for the write to complete, timeout after 10 seconds
    ;; periodic flushing of the queue is taken care of by 
    ;; db:flush-queue
    (let loop ()
      (thread-sleep! 0.001)
      (mutex-lock! *completed-mutex*)
      (if (hash-table-ref/default *completed-writes* qry-sig #f)
	  (begin
	    (hash-table-delete! *completed-writes* qry-sig)
	    (set! got-it #t)))
      (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))
    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))

	  (debug:print-info 7 "Received " response " from wrapped write")
	  (server:reply return-address qry-sig response response))
	;; otherwise if appropriate flush the queue (this is a read or complex query)
	(begin
	  (cond
	   ((member stmt-key db:special-queries)
	    (let ((starttime (current-milliseconds)))
	      (debug:print-info 9 "Handling special statement " stmt-key)
	      (case stmt-key
		((immediate)
		 ;; This is a read or mixed read-write query, must clear the cache
		 (case *transport-type*
		   ((http)
		    (mutex-lock! *db:process-queue-mutex*)
		    (db:process-cached-writes db)

     run-id test-name)
    res))

;; Rollup the pass/fail counts from itemized tests into fail_count and pass_count
(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'),
                 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

    ;; 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 
			(db:step-get-stepname step) 
			;;        stepname                start end status Duration  Logfile 
			(vector (db:step-get-stepname step) ""   "" ""     ""        ""))))
	   (debug:print 6 "record(before) = " record 
			"\nid:       " (db:step-get-id step)
			"\nstepname: " (db:step-get-stepname step)
			"\nstate:    " (db:step-get-state step)
			"\nstatus:   " (db:step-get-status step)
			"\ntime:     " (db:step-get-event_time step))
	   (case (string->symbol (db:step-get-state step))
	     ((start)(vector-set! record 1 (db:step-get-event_time step))
	      (vector-set! record 3 (if (equal? (vector-ref record 3) "")
					(db:step-get-status step)))
	      (if (> (string-length (db:step-get-logfile step))
		     0)
		  (vector-set! record 5 (db:step-get-logfile step))))
	     ((end)  
	      (vector-set! record 2 (any->number (db:step-get-event_time step)))
	      (vector-set! record 3 (db:step-get-status step))
	      (vector-set! record 4 (let ((startt (any->number (vector-ref record 1)))
					  (endt   (any->number (vector-ref record 2))))
				      (debug:print 4 "record[1]=" (vector-ref record 1) 
						   ", startt=" startt ", endt=" endt
						   ", get-status: " (db:step-get-status step))
				      (if (and (number? startt)(number? endt))
					  (seconds->hr-min-sec (- endt startt)) "-1")))
	      (if (> (string-length (db:step-get-logfile step))
		     0)
		  (vector-set! record 5 (db:step-get-logfile step))))
	     (else
	      (vector-set! record 2 (db:step-get-state step))
	      (vector-set! record 3 (db:step-get-status step))
	      (vector-set! record 4 (db:step-get-event_time step))))
	   (hash-table-set! res (db:step-get-stepname step) record)
	   (debug:print 6 "record(after)  = " record 
			"\nid:       " (db:step-get-id step)
			"\nstepname: " (db:step-get-stepname step)
			"\nstate:    " (db:step-get-state step)
			"\nstatus:   " (db:step-get-status step)
			"\ntime:     " (db:step-get-event_time step))))
       ;; (else   (vector-set! record 1 (db:step-get-event_time step)))
       (sort steps (lambda (a b)
		     (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)))

;; 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)))
    ;; 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 
			(db:step-get-stepname step) 
			;;        stepname                start end status    
			(vector (db:step-get-stepname step) ""   "" ""     "" ""))))
	   (debug:print 6 "record(before) = " record 
			"\nid:       " (db:step-get-id step)
			"\nstepname: " (db:step-get-stepname step)
			"\nstate:    " (db:step-get-state step)

		     (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)))))))))

;;======================================================================
;; 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))
  (if (or (not waitons)
	  (null? waitons))
      '()

	     (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))
		       (status            (db:test-get-status test))
		       (item-path         (db:test-get-item-path test))
		       (is-completed      (equal? state "COMPLETED"))
		       (is-ok             (member status '("PASS" "WARN" "CHECK" "WAIVED" "SKIP")))
		       (same-itempath     (equal? ref-item-path item-path)))
		  (set! ever-seen #t)
		  (cond
		   ;; case 1, non-item (parent test) is 
		   ((and (equal? item-path "") ;; this is the parent test
			 is-completed
			 (or is-ok (eq? mode 'toplevel)))

megatest uses the network filesystem to call home to your master
sqlite3 database.

include::getting_started.txt[] include::writing_tests.txt[]
include::reference.txt[]


Controlled waiver propagation
=============================
If test is FAIL and previous test in run with same MT_TARGET is WAIVED then apply the following rules from the testconfig:
If a waiver check is specified in the testconfig apply the check and if it passes then set this FAIL to WAIVED

Waiver check has two parts, 1) a list of waiver, rulename, filepatterns and 2) the rulename script spec (note that "diff" and "logpro" are predefined)

=========================================================
###### EXAMPLE FROM testconfig #########
# matching file(s) will be diff'd with previous run and logpro applied
# if PASS or WARN result from logpro then WAIVER state is set
#
[waivers]
# logpro_file    rulename      input_glob
waiver_1         logpro        lookittmp.log

[waiver_rules]

# This builtin rule is the default if there is no <waivername>.logpro file
# diff   diff %file1% %file2%

# This builtin rule is applied if a <waivername>.logpro file exists

   (begin
     (print-error-message exn)
     (if (< portnum 9000)
	 (begin 
	   (print "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)
   ;; This starts the spiffy server
   (start-server port: portnum)
   (open-run-close tasks:server-delete tasks:open-db ipaddrstr portnum)
   (print "INFO: server has been stopped")))

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

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

      (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)))
	    (begin 
	      (debug:print-info 1 "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...
        (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*)

(define (http-transport:launch)
  (if (not *toppath*)
      (if (not (setup-for-run))
	  (begin
	    (debug:print 0 "ERROR: cannot find megatest.config, exiting")
	    (exit))))
  (debug:print-info 2 "Starting the standalone server")
  (if (args:get-arg "-daemonize")
      (daemon:ize))
  (let ((hostinfo (open-run-close tasks:get-best-server tasks:open-db)))
    (debug:print 11 "http-transport:launch hostinfo=" hostinfo)
    ;; #(1 "143.182.207.24" 5736 -1 "http" 22771 "hostname")
    (if hostinfo
	(debug:print-info 2 "NOT starting new server, one is already running on " (vector-ref hostinfo 1) ":" (vector-ref hostinfo 2))
	(if *toppath* 
	    (let* ((th2 (make-thread (lambda ()

(define (steprun-good? logpro exitcode)
  (or (eq? exitcode 0)
      (and logpro (eq? exitcode 2))))

;; if handed a string, process it, else look for MT_CMDINFO
(define (launch:get-cmdinfo-assoc-list #!key (encoded-cmd #f))
  (let ((enccmd (if encoded-cmd encoded-cmd (getenv "MT_CMDINFO"))))
    (if enccmd
	(read (open-input-string (base64:base64-decode enccmd)))
	'())))

(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)

;; 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 *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
;;  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
;;  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]
  -h                      : this help
  -version                : print megatest version (currently " megatest-version ")

  -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) 
  -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
  -extract-ods fname.ods  : extract an open document spreadsheet from the database
  -pathmod path           : insert path, i.e. path/runame/itempath/logfile.html
                            will clear the field if no rundir/testname/itempath/logfile

megatest -test-files 'logs/*.log' -target ubuntu/n%/no% :runname w49% -testpatt test_mt%

Called as " (string-intersperse (argv) " ") "
Version " megatest-version ", built from " megatest-fossil-hash ))

;;  -gui                    : start a gui interface
;;  -config fname           : override the runconfig file with fname


;; process args
(define remargs (args:get-args 
		 (argv)
		 (list  "-runtests"  ;; run a specific test
			"-config"    ;; override the config file name
			"-execute"   ;; run the command encoded in the base64 parameter

			":value"
			":expected"
			":tol"
			":units"
			;; misc
			"-server"
			"-transport"
			"-stop-server"
			"-port"
			"-extract-ods"
			"-pathmod"
			"-env2file"
			"-setvars"
			"-set-state-status"
			"-debug" ;; for *verbosity* > 2

		        "-xterm"
		        "-showkeys"
		        "-test-status"
			"-set-values"
			"-load-test-data"
			"-summarize-items"
		        "-gui"
			"-daemonize"
			;; misc
			"-archive"
			"-repl"
			"-lock"
			"-unlock"
			"-list-servers"
			;; mist queries

	(if (setup-for-run)
	    (let ((servers (open-run-close tasks:get-best-server tasks:open-db)))
	      (if (or (not servers)
		      (null? servers))
		  (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)
		  )))))
	

(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")
		 (servers-to-kill '())
		 (killinfo   (args:get-arg "-stop-server"))
		 (khost-port (if killinfo (if (substring-index ":" killinfo)(string-split ":") #f) #f))
		 (sid        (if killinfo (if (substring-index ":" killinfo) #f (string->number killinfo)) #f)))
	    (format #t fmtstr "Id" "MTver" "Pid" "Host" "Interface" "OutPort" "InPort" "LastBeat" "State" "Transport")
	    (format #t fmtstr "==" "=====" "===" "====" "=========" "=======" "======" "========" "=====" "=========")
	    (for-each 
	     (lambda (server)



	       (let* ((id         (vector-ref server 0))
		      (pid        (vector-ref server 1))
		      (hostname   (vector-ref server 2))
		      (interface  (vector-ref server 3))
		      (pullport   (vector-ref server 4))
		      (pubport    (vector-ref server 5))
		      (start-time (vector-ref server 6))
		      (priority   (vector-ref server 7))

		 ;; no need to login as status of #t indicates we are connecting to correct 
		 ;; server
		 (if (equal? state "dead")
		     (if (> last-update (* 25 60 60)) ;; keep records around for slighly over a day.
			 (open-run-close tasks:server-deregister tasks:open-db hostname pullport: pullport pid: pid action: 'delete))
		     (if (> last-update 20)        ;; Mark as dead if not updated in last 20 seconds
			 (open-run-close tasks:server-deregister tasks:open-db hostname pullport: pullport pid: pid)))

		 (format #t fmtstr id mt-ver pid hostname interface pullport pubport last-update
			 (if status "alive" "dead") transport)
		 (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)))
    ;; if not list or kill then start a client (if appropriate)

       ((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-config")
    (let ((data *configdat*)) ;; (read-config "megatest.config" #f #t)))
      ;; 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

;;======================================================================
;; read a config file, loading only the section pertinent
;; to this run field1val/field2val/field3val ...
;;======================================================================

(use format directory-utils)

(declare (unit runconfig))
(declare (uses common))

(include "common_records.scm")

		      (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)
  (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))))
    (pop-directory)
    (if (file-exists? runconfigf)
	(setup-env-defaults runconfigf run-id #t keys keyvals
			    environ-patt: (conc "(default"
						(if targ
						    (conc "|" targ ")")
						    ")")))
	(debug:print 0 "WARNING: You do not have a run config file: " runconfigf))))
 

	 (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))
	(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)

    (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)
			      '("PASS" "WARN" "CHECK" "WAIVED" "SKIP")))))
	  prereqs-not-met))

(define (runs:calc-not-completed prereqs-not-met)
  (filter
   (lambda (t)
     (or (not (vector? t))
	 (not (equal? "COMPLETED" (db:test-get-state t)))))

		(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)
		  (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
		       (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)))
		 (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
			  ;; 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))
			;; 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)

			 (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)

			  (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))))))
		       ((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

			    (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))
	     
	     ;; 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))

	    (force (set! runflag #t))
	    ;; NOT_STARTED, run no matter what
	    ((member (test:get-state testdat) '("DELETED" "NOT_STARTED"))(set! runflag #t))
	    ;; not -rerun and PASS, WARN or CHECK, do no run
	    ((and (or (not rerun)
		      keepgoing)
		  ;; Require to force re-run for COMPLETED or *anything* + PASS,WARN or CHECK
		  (or (member (test:get-status testdat) '("PASS" "WARN" "CHECK" "SKIP"))
		      (member (test:get-state  testdat) '("COMPLETED")))) 
	     (debug:print-info 2 "running test " test-name "/" item-path " suppressed as it is " (test:get-state testdat) " and " (test:get-status testdat))
	     (set! runflag #f))
	    ;; -rerun and status is one of the specifed, run it
	    ((and rerun
		  (let* ((rerunlst   (string-split rerun ","))
			 (must-rerun (member (test:get-status testdat) rerunlst)))

  (if (setup-for-run)
      (let ((db (open-db)))
	(let loop ()
	  (let ((last-write-flush-time #f))
	    (mutex-lock! *incoming-mutex*)
	    (set! last-write-flush-time *server:last-write-flush*)
	    (mutex-unlock! *incoming-mutex*)
	    (if (> (- (current-milliseconds) last-write-flush-time) 10)
		(begin
		  (mutex-lock! *db:process-queue-mutex*)
		  (db:process-cached-writes db)
		  (mutex-unlock! *db:process-queue-mutex*)
		  (thread-sleep! 0.005))))
	  (loop)))
      (begin
	(debug:print 0 "ERROR: failed to setup for Megatest in server:write-queue-handler")
	(exit 1))))
    
;;======================================================================
;; S E R V E R   U T I L I T I E S 

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

(define (tasks:open-db)
  (let* ((dbpath  (conc *toppath* "/monitor.db"))
	 (exists  (file-exists? dbpath))








	 (mdb     (sqlite3:open-database dbpath)) ;; (never-give-up-open-db dbpath))
	 (handler (make-busy-timeout 36000)))
    (sqlite3:set-busy-handler! mdb handler)
    (sqlite3:execute mdb (conc "PRAGMA synchronous = 0;"))
    (if (not exists)
	(begin
	  (sqlite3:execute mdb "CREATE TABLE IF NOT EXISTS tasks_queue (id INTEGER PRIMARY KEY,

	  (case action
	    ((delete)(sqlite3:execute mdb "DELETE FROM servers WHERE  hostname=? AND port=?;" hostname port))
	    (else    (sqlite3:execute mdb "UPDATE servers SET state='dead' WHERE hostname=? AND port=?;" hostname port)))
	  (debug:print 0 "ERROR: tasks:server-deregister called with neither pid nor port specified"))))

(define (tasks:server-deregister-self mdb hostname)
  (tasks:server-deregister mdb hostname pid: (current-process-id)))

;; need a simple call for robustly removing records given host and port
(define (tasks:server-delete mdb hostname port)
  (tasks:server-deregister mdb hostname port: port action: 'delete))

(define (tasks:server-get-server-id mdb hostname iface port pid)
  (debug:print-info 12 "tasks:server-get-server-id " mdb " " hostname " " iface " " port " " pid)
  (let ((res #f))
    (sqlite3:for-each-row
     (lambda (id)
       (set! res id))

;; 		  (if (null? tal)
;; 		      #f
;; 		      (loop (car tal)(cdr tal))))))))))

(define (tasks:remove-server-records mdb)
  (sqlite3:execute mdb "DELETE FROM servers;"))

(define (tasks:mark-server hostname port pid state transport)
  (if port
      (open-run-close tasks:server-deregister tasks:open-db hostname port: port)
      (open-run-close tasks:server-deregister tasks:open-db hostname pid:  pid)))


(define (tasks:kill-server status hostname port pid transport)
  (debug:print-info 1 "Removing defunct server record for " hostname ":" port)
  (if port
      (open-run-close tasks:server-deregister tasks:open-db hostname port: port)
      (open-run-close tasks:server-deregister tasks:open-db hostname pid:  pid))
  (if status ;; #t means alive
      (begin
	(if (equal? hostname (get-host-name))
	    (handle-exceptions
	     exn
	     (debug:print-info 0 "server may or may not be dead, check for megatest -server running as pid " pid "\n"
			       "  EXCEPTION: " ((condition-property-accessor 'exn 'message) exn))
	     (debug:print 1 "Sending signal/term to " pid " on " hostname)
	     (process-signal pid signal/term)
	     (thread-sleep! 5) ;; give it five seconds to die peacefully then do a brutal kill
	     ;;(process-signal pid signal/kill)
	     ) ;; local machine, send sig term
	    (begin
		(debug:print-info 1 "Stopping remote servers not yet supported."))))
	;;      (debug:print-info 1 "Telling alive server on " hostname ":" port " to commit servercide")
	;;      (let ((serverdat (list hostname port)))
	;;	(case (string->symbol transport)
	;;	  ((http)(http-transport:client-connect hostname port))
	;;	  (else  (debug:print "ERROR: remote stopping servers of type " transport " not supported yet")))
	;;	(cdb:kill-server serverdat)))))    ;; remote machine, try telling server to commit suicide
      (begin
	(if status 
	    (if (equal? hostname (get-host-name))
		(begin
		  (debug:print-info 1 "Sending signal/term to " pid " on " hostname)
		  (process-signal pid signal/term)  ;; local machine, send sig term
		  (thread-sleep! 5)                 ;; give it five seconds to die peacefully then do a brutal kill

  ;;   2. logf is same as outputfilename
  (let* ((outputfilename (conc "megatest-rollup-" test-name ".html"))
	 (orig-dir       (current-directory))
	 (logf-info      (cdb:remote-run db:test-get-logfile-info #f run-id test-name))
	 (logf           (if logf-info (cadr logf-info) #f))
	 (path           (if logf-info (car  logf-info) #f)))
    ;; This query finds the path and changes the directory to it for the test

    (if (directory? path)
	(begin
	  (debug:print 4 "Found path: " path)
	  (change-directory path))
	;; (set! outputfilename (conc path "/" outputfilename)))
	(print "No such path: " path))
    (debug:print 4 "summarize-items with logf " logf ", outputfilename " outputfilename " and force " force)
    (if (or (equal? logf "logs/final.log")
	    (equal? logf outputfilename)
	    force)
	(begin
	  (if (obtain-dot-lock outputfilename 1 20 30) ;; retry every second for 20 seconds, call it dead after 30 seconds and steal the lock
	      (print "Obtained lock for " outputfilename)
	      (print "Failed to obtain lock for " outputfilename))

	      (keep-test   #t)
	      (test-id     (cdb:remote-run db:get-test-id #f run-id test-name item-path))
	      (tdat        (cdb:get-test-info-by-id *runremote* test-id)))
	 (if tdat
	     (begin
	       ;; Look at the test state and status
	       (if (or (member (db:test-get-status tdat) 
			       '("PASS" "WARN" "WAIVED" "CHECK" "SKIP"))
		       (member (db:test-get-state tdat)
			       '("INCOMPLETE" "KILLED")))
		   (set! keep-test #f))

	       ;; examine waitons for any fails. If it is FAIL or INCOMPLETE then eliminate this test
	       ;; from the runnable list
	       (if keep-test

LOGGING = 

OS  = $(shell grep ID /etc/*-release|cut -d= -f2)
FS  = $(shell df -T .|tail -1|awk '{print $$2}')
VER = $(shell fsl info|grep checkout|awk '{print $$2}'|cut -c 1-5)

# The NEWTARGET causes some tests to fail. Do not use until this is fixed.
NEWTARGET  = "$(OS)/$(FS)/$(VER)"
TARGET     = "ubuntu/nfs/none"

all : test1 test2 test3 test4 test5

server :
	(cd ..;make;make install) && \
	(cd fullrun;../../bin/megatest -server - -debug 22) 

test0 : cleanprep
	cd simplerun ; $(MEGATEST) -server - -debug $(DEBUG)

test1 : cleanprep
	rm -f simplerun/megatest.db
	rm -rf simplelinks/ simpleruns/
	mkdir -p simplelinks simpleruns
	cd simplerun;cp ../../*_records.scm .;perl -pi.bak -e 's/define-inline/define/' *_records.scm
	cd simplerun;echo '(load "../tests.scm")' | $(MEGATEST) -repl -debug $(DEBUG)

test2 : fullprep
	cd fullrun;$(MEGATEST) -runtests ez_pass,runfirst/a/% -reqtarg ubuntu/nfs/none :runname $(RUNNAME) -debug $(DEBUG) $(LOGGING)
	cd fullrun;megatest -runtests % -target ubuntu/nfs/none :runname $(RUNNAME)_01 -testpatt %/,%/ai -debug $(DEBUG)
	cd fullrun;megatest -runtests %/,%/ai -target ubuntu/nfs/none :runname $(RUNAME)_02 -debug $(DEBUG)
	cd fullrun;megatest -runtests runfirst/%,%/ai -target ubuntu/nfs/none :runname $(RUNNAME)_02 -debug $(DEBUG)
	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)

# 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 &	

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

[fields]
SYSTEM TEXT
RELEASE TEXT

[setup]
# Adjust max_concurrent_jobs to limit how much you load your machines
max_concurrent_jobs 500

# This is your link path, you can move it but it is generally better to keep it stable
linktree #{shell readlink -f #{getenv PWD}/../simplelinks}

[include testqa/configs/megatest.abc.config]

# Valid values for state and status for steps, NB// It is not recommended you use this
[validvalues]
state start end completed

# Job tools are more advanced ways to control how your jobs are launched
[jobtools]
useshell yes
launcher nbfake

[include megatest.def.config]

# You can override environment variables for all your tests here
[env-override]
EXAMPLE_VAR example value

# As you run more tests you may need to add additional disks, the names are arbitrary but must be unique
[disks]
disk0 #{scheme (nice-path "#{getenv PWD}/../simpleruns")}

[setup]
testcopycmd cp --remove-destination -rlv TEST_SRC_PATH/. TEST_TARG_PATH/.

[include ../fdk.config]

[default]
ALLTESTS see this variable

# Your variables here are grouped by targets [SYSTEM/RELEASE]
[SYSTEM_val/RELEASE_val]
ANOTHERVAR only defined if target is SYSTEM_val/RELEASE_val

#!/bin/sh
sleep 10
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)) >)) " ")}

# 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
reviewed never

[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]

# 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/.

# 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
# Throttle roughly scales the db access milliseconds to seconds delay
throttle 0.2
# Max retries allows megatest to re-check that a tests status has changed
# as tests can have transient FAIL status occasionally

# 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

## disks are:
## name host:/path/to/area