Megatest

		  (if patt
		      (set! key-patt (conc key-patt " AND " key " " wildtype " '" patt "'"))
		      (begin
			(debug:print 0 "ERROR: searching for runs with no pattern set for " fulkey)
			(exit 6)))))
	      keys)
    (set! qry-str (conc "SELECT " keystr " FROM runs WHERE runname " runwildtype " ? " key-patt ";"))
    (debug:print 4 "INFO: runs:get-runs-by-patt qry=" qry-str " " runnamepatt)
    (sqlite3:for-each-row 
     (lambda (a . r)
       (set! res (cons (list->vector (cons a r)) res)))
     db 
     qry-str
     runnamepatt)
    (vector header res)))

    
    ;; look up all tests matching the comma separated list of globs in
    ;; test-patts (using % as wildcard)

    (set! test-names (tests:get-valid-tests *toppath* test-patts test-names: test-names))
    (set! test-names (delete-duplicates test-names))

    (debug:print 0 "INFO: test names " test-names)

    ;; on the first pass or call to run-tests set FAILS to NOT_STARTED if
    ;; -keepgoing is specified
    (if (eq? *passnum* 0)
	(begin
	  ;; 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.
	  (open-run-close db:delete-tests-in-state db run-id "NOT_STARTED")
	  (open-run-close db: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 "\"")
                                (if db (sqlite3:finalize! db))
				(exit 1)))))

						   (itemstable (hash-table-ref/default config "itemstable" #f))) 
					       ;; if either items or items table is a proc return it so test running
					       ;; process can know to call items:get-items-from-config
					       ;; if either is a list and none is a proc go ahead and call get-items
					       ;; otherwise return #f - this is not an iterated test
					       (cond
						((procedure? items)      
						 (debug:print 4 "INFO: items is a procedure, will calc later")
						 items)            ;; calc later
						((procedure? itemstable)
						 (debug:print 4 "INFO: itemstable is a procedure, will calc later")
						 itemstable)       ;; calc later
						((filter (lambda (x)
							   (let ((val (car x)))
							     (if (procedure? val) val #f)))
							 (append (if (list? items) items '())
								 (if (list? itemstable) itemstable '())))
						 'have-procedure)
						((or (list? items)(list? itemstable)) ;; calc now
						 (debug:print 4 "INFO: items and itemstable are lists, calc now\n"
							      "    items: " items " itemstable: " itemstable)
						 (items:get-items-from-config config))
						(else #f)))                           ;; not iterated
					     #f      ;; itemsdat 5
					     #f      ;; spare - used for item-path
					     )))
	    (for-each 
	     (lambda (waiton)
	       (if (and waiton (not (member waiton test-names)))
		   (begin
		     (set! required-tests (cons waiton required-tests))
		     (set! test-names (cons waiton test-names))))) ;; was an append, now a cons
	     waitons)
	    (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 test-patts)
    (debug:print 4 "INFO: 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")))))

	(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))
	(let loop ((hed         (car sorted-test-names))
		   (tal         (cdr sorted-test-names))
		   (reruns      '()))
	  (if (not (null? reruns))(debug:print 4 "INFO: 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))

		     (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 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 "\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 4 "INFO: 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 1 "INFO: 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 4 "INFO: 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*)
		  (loop (car newtal)(cdr newtal) reruns))
		 ((not have-resources) ;; simply try again after waiting a second
		  (debug:print 1 "INFO: no resources to run new tests, waiting ...")
		  (thread-sleep! (+ 0.01 *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)
		  (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 4 "INFO: 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")
				       (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)")
				  (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 (>= *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 4 "INFO: 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    (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           (runs:calc-fails prereqs-not-met))
			   (non-completed   (runs:calc-not-completed prereqs-not-met)))
		      (debug:print 8 "INFO: 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)

				  (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 4 "INFO: 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 1 "INFO: 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 4 "INFO: processing the case with a lambda for items or 'have-procedure. Moving through the queue without dropping " hed)
		      (thread-sleep! (+ 1 *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 (open-run-close tests:filter-non-runnable #f run-id tal test-records)) ;; i.e. not FAIL, WAIVED, INCOMPLETE, PASS, KILLED,
		     (junked (lset-difference equal? tal newlst)))
		(debug:print 4 "INFO: 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 4 "INFO: I'm pretty sure I shouldn't get here."))
	     (else
	      (debug:print 4 "INFO: Exiting loop with...\n  hed=" hed "\n  tal=" tal "\n  reruns=" reruns))
	     )))) ;; LET* ((test-record

    ;; we get here on "drop through" - loop for next test in queue
    ;; FIXME!!!! THIS SHOULD NOT REQUIRE AN EXIT!!!!!!!
    
    (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
  

	    ;;
	    (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)
		  (set! test-id (open-run-close db:get-test-id db run-id test-name item-path))))
	    (debug:print 4 "INFO: 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))

	    ((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"))
		      (member (test:get-state  testdat) '("COMPLETED")))) 
	     (debug:print 2 "INFO: 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)))
		    (debug:print 3 "INFO: -rerun list: " rerun ", test-status: " (test:get-status testdat)", must-rerun: " must-rerun)
		    must-rerun))
	     (debug:print 2 "INFO: Rerun forced for test " test-name "/" item-path)
	     (set! runflag #t))
	    ;; -keepgoing, do not rerun FAIL
	    ((and keepgoing
		  (member (test:get-status testdat) '("FAIL")))
	     (set! runflag #f))
	    ((and (not rerun)
		  (member (test:get-status testdat) '("FAIL" "n/a")))

	 (keys         (open-run-close db:get-keys db))
	 (rundat       (open-run-close runs:get-runs-by-patt db keys runnamepatt))
	 (header       (vector-ref rundat 0))
	 (runs         (vector-ref rundat 1))
	 (states       (if state  (string-split state  ",") '()))
	 (statuses     (if status (string-split status ",") '()))
	 (state-status (if (string? new-state-status) (string-split new-state-status ",") '(#f #f))))
    (debug:print 4 "INFO: runs:operate-on => Header: " header " action: " action " new-state-status: " new-state-status)
    (if (> 2 (length state-status))
	(begin
	  (debug:print 0 "ERROR: the parameter to -set-state-status is a comma delimited string. E.g. COMPLETED,FAIL")
	  (exit)))
    (for-each
     (lambda (run)
       (let ((runkey (string-intersperse (map (lambda (k)
						(db:get-value-by-header run header (vector-ref k 0))) keys) "/"))
	     (dirs-to-remove (make-hash-table)))
	 (let* ((run-id    (db:get-value-by-header run header "id"))
		(run-state (db:get-value-by-header run header "state"))
		(tests     (if (not (equal? run-state "locked"))
			       (open-run-close db:get-tests-for-run db run-id
						      testpatt states statuses
						      not-in:  #f
						      sort-by: (case action
								 ((remove-runs) 'rundir)
								 (else          'event_time)))
			       '()))
		(lasttpath "/does/not/exist/I/hope"))
	   (debug:print 4 "INFO: runs:operate-on run=" run ", header=" header)
	   (if (not (null? tests))
	       (begin
		 (case action
		   ((remove-runs)
		    (debug:print 1 "Removing tests for run: " runkey " " (db:get-value-by-header run header "runname")))
		   ((set-state-status)
		    (debug:print 1 "Modifying state and staus for tests for run: " runkey " " (db:get-value-by-header run header "runname")))
		   ((print-run)
		    (debug:print 1 "Printing info for run " runkey ", run=" run ", tests=" tests ", header=" header)
		    action)
		   (else
		    (print "INFO: action not recognised " action)))
		 (for-each
		  (lambda (test)
		    (let* ((item-path (db:test-get-item-path test))
			   (test-name (db:test-get-testname test))
			   (run-dir   (db:test-get-rundir test))    ;; run dir is from the link tree
			   (real-dir  (if (file-exists? run-dir)
					  (resolve-pathname run-dir)
					  #f))
			   (test-id   (db:test-get-id test)))
		      ;;   (tdb       (db:open-test-db run-dir)))
		      (debug:print 4 "INFO: test=" test) ;;   " (db:test-get-testname test) " id: " (db:test-get-id test) " " item-path " action: " action)
		      (case action
			((remove-runs) ;; the tdb is for future possible. 
			 (open-run-close db:delete-test-records db #f (db:test-get-id test))
			 (debug:print 1 "INFO: Attempting to remove dir " real-dir " and link " run-dir)
			 (if (and real-dir 
				  (> (string-length real-dir) 5)
				  (file-exists? real-dir)) ;; bad heuristic but should prevent /tmp /home etc.
			     (begin ;; let* ((realpath (resolve-pathname run-dir)))
			       (debug:print 1 "INFO: Recursively removing " real-dir)
			       (if (file-exists? real-dir)
				   (if (> (system (conc "rm -rf " real-dir)) 0)
				       (debug:print 0 "ERROR: There was a problem removing " real-dir " with rm -f"))
				   (debug:print 0 "WARNING: test run dir " real-dir " appears to not exist")))
			     (debug:print 0 "WARNING: directory " real-dir " does not exist"))
			 (if (symbolic-link? run-dir)
			     (begin
			       (debug:print 1 "INFO: Removing symlink " run-dir)
			       (delete-file run-dir))
			     (if (directory? run-dir)
				 (if (> (directory-fold (lambda (f x)(+ 1 x)) 0 run-dir) 0)
				     (debug:print 0 "WARNING: refusing to remove " run-dir " as it is not empty")
				     (delete-directory run-dir)) ;; it should be empty by here BUG BUG, add error catch
				 (debug:print 0 "ERROR: refusing to remove " run-dir " as it either doesn't exist or is not a symlink or directory")
				 )))
			((set-state-status)
			 (debug:print 2 "INFO: new state " (car state-status) ", new status " (cadr state-status))
			 (open-run-close db:test-set-state-status-by-id db (db:test-get-id test) (car state-status)(cadr state-status) #f)))))
		  (sort tests (lambda (a b)(let ((dira (db:test-get-rundir a))
						 (dirb (db:test-get-rundir b)))
					     (if (and (string? dira)(string? dirb))
						 (> (string-length dira)(string-length dirb))
						 #f)))))))
	   ;; remove the run if zero tests remain

		(let ((run-id (db:get-value-by-header run header "id")))
		  (if (or lock
			  (and unlock
			       (begin
				 (print "Do you really wish to unlock run " run-id "?\n   y/n: ")
				 (equal? "y" (read-line)))))
		      (open-run-close db:lock/unlock-run db run-id lock unlock user)
		      (debug:print 0 "INFO: 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)