Megatest

    (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 (set-megatest-env-vars db run-id)
  (let ((keys (db:get-keys db)))
    (for-each (lambda (key)
		(sqlite3:for-each-row
		 (lambda (val)
		   (debug:print 2 "setenv " (key:get-fieldname key) " " val)
		   (setenv (key:get-fieldname key) val))
		 db 
		 (conc "SELECT " (key:get-fieldname key) " FROM runs WHERE id=?;")

(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)
(define (runs:can-run-more-tests db test-record)
  (let* ((tconfig                 (tests:testqueue-get-testconfig test-record))
	 (jobgroup                (config-lookup tconfig "requirements" "jobgroup"))
	 (num-running             (db:get-count-tests-running db))
	 (num-running-in-jobgroup (db:get-count-tests-running-in-jobgroup db jobgroup))
	 (max-concurrent-jobs     (config-lookup *configdat* "setup"     "max_concurrent_jobs"))
	 (job-group-limit         (config-lookup *configdat* "jobgroups" jobgroup)))
    (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*))
	#f
	(let ((can-not-run-more (cond
				 ;; if max-concurrent-jobs is set and the number running is greater 
				 ;; than it than cannot run more jobs
				 ((and max-concurrent-jobs
				       (string->number max-concurrent-jobs)

	  ;; have to delete test records where NOT_STARTED since they can cause -keepgoing to 
	  ;; get stuck due to becoming inaccessible from a failed test. I.e. if test B depends 
	  ;; on test A but test B reached the point on being registered as NOT_STARTED and test
	  ;; A failed for some reason then on re-run using -keepgoing the run can never complete.
	  (db:delete-tests-in-state db run-id "NOT_STARTED")
	  (rdb:set-tests-state-status db run-id test-names #f "FAIL" "NOT_STARTED" "FAIL")))

    ;; from here on out the db will be opened and closed on every call runs:run-tests-queue
    (sqlite3:finalize! db) 
    ;; now add non-directly referenced dependencies (i.e. waiton)
    (if (not (null? test-names))
	(let loop ((hed (car test-names))
		   (tal (cdr test-names)))         ;; 'return-procs tells the config reader to prep running system but return a proc
	  (debug:print 4 "INFO: hed=" hed " at top of loop")
	  (let* ((config  (tests:get-testconfig hed 'return-procs))
		 (waitons (if config (string-split (let ((w (config-lookup config "requirements" "waiton")))
						     (if w w "")))
			      (begin
				(debug:print 0 "ERROR: non-existent required test \"" hed "\"")
                                (sqlite3:finalize! db)
				(exit 1)))))

	    (let ((remtests (delete-duplicates (append waitons tal))))
	      (if (not (null? remtests))
		  (loop (car remtests)(cdr remtests)))))))

    (if (not (null? required-tests))
	(debug:print 1 "INFO: Adding " required-tests " to the run queue"))
    ;; NOTE: these are all parent tests, items are not expanded yet.
    (debug:print 4 "INFO: test-records=" (hash-table->alist test-records))
    (runs:run-tests-queue run-id runname test-records keyvallst flags)

    (debug:print 4 "INFO: All done by here")))

;; 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)
    ;; 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))
	(item-patts        (hash-table-ref/default flags "-itempatt" #f))
	(test-registery    (make-hash-table)))
    (if (not (null? sorted-test-names))
	(let loop ((hed         (car sorted-test-names))
		   (tal         (cdr sorted-test-names)))

	  (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))

				       (filter
					(lambda (t)
					  (or (not (vector? t))
					      (not (equal? "COMPLETED" (db:test-get-state t)))))
					prereqs-not-met)))
		 (pretty-string (lambda (lst)
				  (map (lambda (t)
					 (if (not (vector? t))
					     (conc t)
					     (conc (db:test-get-testname t) ":" (db:test-get-state t) "/" (db:test-get-status t))))
				       lst))))
	    
	    (debug:print 6
			 "test-name: " test-name
			 "\n  hed: " hed
			 "\n  itemdat:     " itemdat
			 "\n  items:     " items
			 "\n  item-path: " item-path
			 "\n  waitons:   " waitons)

	    ;; 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))))

	    (cond
	     ((not items) ;; when false the test is ok to be handed off to launch (but not before)
	      (let* ((have-resources  (open-run-close runs:can-run-more-tests #f test-record)) ;; look at the test jobgroup and tot jobs running
		     (prereqs-not-met (open-run-close db:get-prereqs-not-met #f run-id waitons item-path mode: testmode))
		     (fails           (calc-fails prereqs-not-met))
		     (non-completed   (calc-not-completed prereqs-not-met)))
		(debug:print 8 "INFO: 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 4 "INFO: hed=" hed)
		;; 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?
		(cond
		 ((not (patt-list-match item-path item-patts))
		  ;; else the run is stuck, temporarily or permanently
		  ;; but should check if it is due to lack of resources vs. prerequisites
		  (debug:print 1 "INFO: Skipping " (tests:testqueue-get-testname test-record) " " item-path " as it doesn't match " item-patts)
		  (if (not (null? tal))
		      (loop (car tal)(cdr tal))))
		 ((not (hash-table-ref/default test-registery (conc test-name "/" item-path) #f))
		  (open-run-close tests:register-test #f run-id test-name item-path)
		  (hash-table-set! test-registery (conc test-name "/" item-path) #t)
		  (loop (car newtal)(cdr newtal)))
		 ((not have-resources) ;; simply try again after waiting a second
		  (thread-sleep! (+ 1 *global-delta*))
		  (debug:print 1 "INFO: no resources to run new tests, waiting ...")
		  ;; could have done hed tal here but doing car/cdr of newtal to rotate tests
		  (loop (car newtal)(cdr newtal)))
		 ((and have-resources
		       (or (null? prereqs-not-met)
			   (and (eq? testmode 'toplevel)
				(null? non-completed))))
		  ;; no loop here, just drop though and use the loop at the bottom 

		  (run:test run-id runname keyvallst test-record flags #f))









		 (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 4 "INFO: Shouldn't really get here, race condition? Unable to launch more tests at this moment, killing time ...")
			  (thread-sleep! (+ 1 *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)))
			;; 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")
				       (loop (car tal)(cdr tal)))
				(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)")
				  (loop hed tal)))))))))
	     
	     ;; 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 (>= *verbosity* 1)
		       (> (length items) 0)
		       (> (length (car items)) 0))

	       items)
	      (if (not (null? tal))
		  (loop (car tal)(cdr tal))))

	     ;; 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    (open-run-close runs:can-run-more-tests #f test-record)))
		(if can-run-more
		    (let* ((prereqs-not-met (open-run-close db:get-prereqs-not-met #f run-id waitons item-path mode: testmode))
			   (fails           (calc-fails prereqs-not-met))
			   (non-completed   (calc-not-completed prereqs-not-met)))
		      (debug:print 8 "INFO: can-run-more: " can-run-more
				   "\n prereqs-not-met: " (pretty-string prereqs-not-met)
				   "\n non-completed:   " (pretty-string non-completed) 
				   "\n fails:           " (pretty-string fails)
				   "\n testmode:        " testmode
				   "\n (eq? testmode 'toplevel) " (eq? testmode 'toplevel)
				   "\n (null? non-completed)    " (null? non-completed))
		      (cond 
		       ((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)
				 (null? non-completed)))
			(let ((test-name (tests:testqueue-get-testname test-record)))
			  (setenv "MT_TEST_NAME" test-name) ;; 
			  (setenv "MT_RUNNAME"   runname)
			  (open-run-close-measure set-megatest-env-vars #f 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)
				  (loop hed tal))
				(begin
				  (debug:print 0 "ERROR: The proc from reading the setup did not yield a list - please report this")

		(debug:print 1 "INFO: 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
	      (let ((newlst (open-run-close tests:filter-non-runnable #f run-id tal test-records))) ;; i.e. not FAIL, WAIVED, INCOMPLETE, PASS, KILLED,
		(thread-sleep! *global-delta*)
		(if (not (null? newlst))
		    (loop (car newlst)(cdr newlst)))))))))

;; 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)
  ;; 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))
	 (rerun        (hash-table-ref/default flags "-rerun" #f))
	 (keepgoing    (hash-table-ref/default flags "-keepgoing" #f))
	 (item-path     "")
	 (db           #f))
    (debug:print 4
		 "test-config: " (hash-table->alist test-conf)
		 "\n   itemdat: " itemdat
		 )
    ;; 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)
    (open-run-close-measure set-megatest-env-vars db run-id) ;; 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)))
    
    ;; (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       (open-run-close db:get-test-id db  run-id test-name item-path))
	   (testdat       (open-run-close db:get-test-info-by-id db 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))
	    (debug:print 4 "test-id=" test-id ", run-id=" run-id ", test-name=" test-name ", item-path=" item-path)
	    (set! testdat (open-run-close db:get-test-info-by-id db test-id))))
      (set! test-id (db:test-get-id testdat))
      (change-directory test-path)
      (case (if force ;; (args:get-arg "-force")
		'NOT_STARTED
		(if testdat
		    (string->symbol (test:get-state testdat))
		    'failed-to-insert))

	   (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.
	       (if (not (launch-test #f run-id 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)  
	 (if (> (- (current-seconds)(+ (db:test-get-event_time testdat)
				       (db:test-get-run_duration testdat)))
		600) ;; i.e. no update for more than 600 seconds
	     (begin
	       (debug:print 0 "WARNING: Test " test-name " appears to be dead. Forcing it to state INCOMPLETE and status STUCK/DEAD")
	       (open-run-close test-set-status! db test-id "INCOMPLETE" "STUCK/DEAD" "Test is stuck or dead" #f))
	     (debug:print 2 "NOTE: " test-name " is already running")))
	(else       (debug:print 0 "ERROR: Failed to launch test " new-test-name ". Unrecognised state " (test:get-state testdat)))))))

;;======================================================================
;; END OF NEW STUFF
;;======================================================================