Megatest

(declare (uses csv-xml))
(declare (uses keysmod))
(declare (uses mtmod))
(declare (uses processmod))
(declare (uses dbmod))
(declare (uses rmtmod))
(declare (uses testsmod))



(module runsmod
	*
	
(import scheme
	(prefix sqlite3 sqlite3:)
	chicken.base

	chicken.process
	chicken.process-context
	chicken.process-context.posix
	chicken.sort
	chicken.string
	chicken.time
	chicken.time.posix



	(prefix base64 base64:)
	csv-xml
	directory-utils
	matchable
	regex
	s11n
	srfi-1
	srfi-13
	srfi-18
	srfi-69
	stack

	system-information
	typed-records
	z3

	(prefix mtargs args:)
	commonmod
	configfmod
	debugprint
	keysmod
	mtmod
	mtver
	processmod
	dbmod
	rmtmod
	testsmod


	
	)



;; use this struct to facilitate refactoring
;;

(defstruct runs:dat
  reglen regfull
  runname max-concurrent-jobs run-id
  test-patts required-tests test-registry

	      (set! rtime (+ rtime work-time))
	      ((or proc runs:parallel-runners-mgmt) rdat)
	      (loop)))))
    (let* ((done-time (current-seconds)))
      (debug:print-info 0 *default-log-port* "DONE: rtime=" rtime ", elapsed time=" (- done-time startt)
			", ratio=" (/ rtime (- done-time startt))))))

(define (runs:get-mt-env-alist run-id runname target testname itempath)
  ;;(bb-check-path msg: "runs:set-megatest-env-vars entry")
  `(("MT_TEST_NAME"     . ,testname)
    
    ("MT_ITEMPATH"      . ,itempath)

    ("MT_TARGET"        . ,target)
    
    ("MT_RUNNAME"       . ,runname)
    
    ("MT_RUN_AREA_HOME" . ,*toppath*)
    
    ,@(let* ((link-tree (common:get-linktree))) ;; (configf:lookup *configdat* "setup" "linktree")))
        (if link-tree
            (list (cons "MT_LINKTREE" link-tree)
                  
                  (cons "MT_TEST_RUN_DIR"
                        (conc link-tree  "/" target "/" runname "/" testname
                              (if (and (string? itempath) (not (equal? itempath "")))
                                  (conc "/" itempath)
                                  "")))
                  )
            '()))
    
    ,@(map
       (lambda (key)
         (cons  (car key) (cadr key)))
       (keys:target->keyval (rmt:get-keys) target))
    
    ,@(map (lambda (var)
             (let ((val (configf:lookup *configdat* "env-override" var)))
               (cons var val)))
           (configf:section-vars *configdat* "env-override"))))
    
;; set up needed environment variables given a run-id and optionally a target, itempath etc.
;;
(define (runs:set-megatest-env-vars run-id #!key (inkeys #f)(inrunname #f)(inkeyvals #f)(intarget #f)(testname #f)(itempath #f))
  ;;(bb-check-path msg: "runs:set-megatest-env-vars entry")
  (let* ((target    (or intarget 
			(common:args-get-target)
			(get-environment-variable "MT_TARGET")))
	 (keys      (if inkeys    inkeys    (rmt:get-keys)))
	 (keyvals   (if inkeyvals inkeyvals (keys:target->keyval keys target)))
	 (vals      (hash-table-ref/default *env-vars-by-run-id* run-id #f))
	 (link-tree (common:get-linktree))) ;; (configf:lookup *configdat* "setup" "linktree")))
    (if testname (setenv "MT_TEST_NAME" testname))
    (if itempath (setenv "MT_ITEMPATH"  itempath))

    ;; get the info from the db and put it in the cache
    (if link-tree
	(setenv "MT_LINKTREE" link-tree)
	(debug:print-error 0 *default-log-port* "linktree not set, should be set in megatest.config in [setup] section."))
    (if (not vals)
	(let ((ht (make-hash-table)))
	  (hash-table-set! *env-vars-by-run-id* run-id ht)
	  (set! vals ht)
	  (for-each
	   (lambda (key)
	     (hash-table-set! vals (car key) (cadr key)))
	   keyvals)))
    ;; from the cached data set the vars
    
    (hash-table-for-each
     vals
     (lambda (key val)
       (debug:print 2 *default-log-port* "setenv " key " " val)
       (safe-setenv key val)))
    ;;(bb-check-path msg: "runs:set-megatest-env-vars block 1")
    ;;(BB> "*env-vars-by-run-id*/runid("run-id" vals="(hash-table->alist vals))

    (if (not (get-environment-variable "MT_TARGET"))(setenv "MT_TARGET" target))
    ;; we had a case where there was an exception generated by the hash-table-ref
    ;; due to *configdat* being #f Adding a handle and exit
    (let fatal-loop ((count 0)) 
      (handle-exceptions
	  exn
	  (let ((call-chain (get-call-chain))
		(msg        ((condition-property-accessor 'exn 'message) exn)))
	    (if (< count 5)
		(begin ;; this call is colliding, do some crude stuff to fix it.
		  (debug:print 0 *default-log-port* "ERROR: *configdat* was inaccessible! This should never happen. Retry #" count
			       ", exn=" exn)
		  (launch:setup force-reread: #t)
		  (fatal-loop (+ count 1))) 
		(begin
		  (debug:print 0 *default-log-port* "FATAL: *configdat* was inaccessible! This should never happen. Retried " count
			       " times. Message: " msg)
		  (debug:print 0 *default-log-port* "Call chain:")
		  (with-output-to-port *default-log-port*

                    (lambda ()
                      (print "*configdat* is >>"*configdat*"<<")
                      (pp *configdat*)
                      (pp call-chain)))
                  
		  (exit 1))))
          ;;(bb-check-path msg: "runs:set-megatest-env-vars block 1.5")
          (when (or (not *configdat*) (not (hash-table? *configdat*)))
              (debug:print 0 *default-log-port* "WARNING: *configdat* was inaccessible! This should never happen.  Brute force reread.")
              ;;(BB> "ERROR: *configdat* was inaccessible! This should never happen.  Brute force reread.")
              (thread-sleep! 2) ;; assuming nfs lag.
              (launch:setup force-reread: #t))
          (alist->env-vars (hash-table-ref/default *configdat* "env-override" '())))) ;;;; environment is tainted HERE in this let block.
    ;;(bb-check-path msg: "runs:set-megatest-env-vars block 2")
    ;; Lets use this as an opportunity to put MT_RUNNAME in the environment
    (let ((runname  (if inrunname inrunname (rmt:get-run-name-from-id run-id))))
      (if runname
	  (setenv "MT_RUNNAME" runname)
	  (debug:print-error 0 *default-log-port* "no value for runname for id " run-id)))
    (setenv "MT_RUN_AREA_HOME" *toppath*)
    ;; if a testname and itempath are available set the remaining appropriate variables
    (if testname (setenv "MT_TEST_NAME" testname))
    (if itempath (setenv "MT_ITEMPATH"  itempath))
    ;;(bb-check-path msg: "runs:set-megatest-env-vars block 3")
    (if (and testname link-tree)
	(setenv "MT_TEST_RUN_DIR" (conc (getenv "MT_LINKTREE")  "/"
					(getenv "MT_TARGET")    "/"
					(getenv "MT_RUNNAME")   "/"
					(getenv "MT_TEST_NAME")
					(if (and itempath
						 (not (equal? itempath "")))
					    (conc "/" itempath)
					    ""))))))

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

;; Every time can-run-more-tests is called increment the delay
;;
;; NOTE: We run this server-side!! Do not use this global except in the runs:can-run-more-tests routine
;;
(define *last-num-running-tests* 0)
;; (define *runs:can-run-more-tests-count* 0)
(define (runs:shrink-can-run-more-tests-count runsdat)

                   (debug:print 0 *default-log-port* "Message: " ((condition-property-accessor 'exn 'message) exn) ", exn=" exn)
                   (debug:print 0 *default-log-port* "ERROR: failed to run pre-hook " run-pre-hook ", check the log " log-file))
                 (debug:print-info 0 *default-log-port* "running run-pre-hook: \"" run-pre-hook "\", log is " actual-logf)
                 (system (conc run-pre-hook " >> " actual-logf " 2>&1"))
                 (debug:print-info 0 *default-log-port* "pre-hook \"" run-pre-hook "\" took " (- (current-seconds) start-time) " seconds to run.")))
              (debug:print 0 *default-log-port* "Skipping pre-hook call \"" run-pre-hook "\" as there are existing tests for this run.")))))
   


(define (runs:run-post-hook run-id)
    (let* ((run-post-hook   (configf:lookup *configdat* "runs" "post-hook"))
           (existing-tests (if run-post-hook
                               (rmt:get-tests-for-run run-id "%" '() '() ;; run-id testpatt states statuses
                                                      #f #f ;; offset limit
                                                      #f ;; not-in
                                                      #f ;; sort-by
                                                      #f ;; sort-order
                                                      #f ;; get full data (not 'shortlist)
                                                      0 ;; (runs:gendat-inc-results-last-update *runs:general-data*) ;; last update time
                                                      'dashboard)
                               '()))
           (log-dir         (conc *toppath* "/logs"))
           (log-file        (conc "post-hook-" (string-translate (getenv "MT_TARGET") "/" "-") "-" (getenv "MT_RUNNAME") ".log"))
           (full-log-fname  (conc log-dir "/" log-file)))
      (if run-post-hook
          ;; (if (null? existing-tests)
          ;;    (debug:print 0 *default-log-port* "Skipping post-hook call \"" run-post-hook "\" as there are existing tests for this run.")))))
	  (let* ((use-log-dir (if (not (directory-exists? log-dir))
				  (handle-exceptions
				      exn
				      (begin
					(debug:print 0 *default-log-port* "WARNING: Failed to create " log-dir ", exn=" exn)
					#f)
				    (create-directory log-dir #t)
				    #t)
				  #t))
		 (start-time   (current-seconds))
		 (actual-logf  (if use-log-dir full-log-fname log-file)))
	    (handle-exceptions
		exn
		(begin
		  (print-call-chain *default-log-port*)
		  (debug:print 0 *default-log-port* "Message: " ((condition-property-accessor 'exn 'message) exn) ", exn=" exn)
		  (debug:print 0 *default-log-port* "ERROR: failed to run post-hook " run-post-hook ", check the log " log-file))
	      (debug:print-info 0 *default-log-port* "running run-post-hook: \"" run-post-hook "\", log is " actual-logf)
	      (system (conc run-post-hook " >> " actual-logf " 2>&1"))
	      (debug:print-info 0 *default-log-port* "post-hook \"" run-post-hook "\" took " (- (current-seconds) start-time) " seconds to run."))))))


(define (runs:rerun-hook test-id new-test-path testdat rerunlst)
    (let* ((rerun-hook   (configf:lookup *configdat* "runs" "rerun-hook"))
           (log-dir         (conc *toppath* "/reruns/logs"))
           (target (getenv "MT_TARGET"))
           (runname (common:args-get-runname))
           (rundir (db:test-get-rundir testdat))
           (tarfiledir (conc *toppath* "/reruns"))
           (status (db:test-get-status testdat))
           (comment (conc "\"" (db:test-get-comment testdat) "\"" ))
           (testname (db:test-get-testname testdat))
           (itempath (db:test-get-item-path testdat))
           (file-body (conc status "-" (string-translate target "/" "-") "-" runname "." testname (if (not (string=? itempath "")) (conc  "." (string-translate itempath "/" "-")) "")))
           (log-file (conc file-body ".log"))
           ;; (log-file        (conc status "-" (string-translate target "/" "-") "-" runname "." testname (if (not (string=? itempath "")) (conc  "." (string-translate itempath "/" "-")) "") ".log"))
           (full-log-fname  (conc log-dir "/" log-file))
           (tarfilename (conc file-body ".tar"))
           ;; (tarfilename (conc status "." (string-translate target "/" "-") "." runname "." testname (if (not (string=? itempath "")) (conc  "." (string-translate itempath "/" "-")) "") ".tar"))
           )
      (if rerun-hook
	  (let* ((use-log-dir (if (not (directory-exists? log-dir))
				  (handle-exceptions
				      exn
				      (begin
					(debug:print 0 *default-log-port* "WARNING: Failed to create " log-dir ", exn=" exn)
					#f)
				    (create-directory log-dir #t)
				    #t)
				  #t))
		 (start-time   (current-seconds))
		 (actual-logf  (if use-log-dir full-log-fname log-file))
                 (sys-call-text (conc rerun-hook " " tarfilename " " rundir " " actual-logf " " runname " " tarfiledir " " status " " target " " comment " " testname " " itempath " >> " actual-logf " 2>&1"))
                 )
	    (debug:print 2 *default-log-port* "Found rerun-hook in config:" rerun-hook)
	    (handle-exceptions
		exn
		(begin
		  (print-call-chain *default-log-port*)
		  (debug:print 0 *default-log-port* "Message: " ((condition-property-accessor 'exn 'message) exn) ", exn=" exn)
		  (debug:print 0 *default-log-port* "ERROR: failed to run rerun-hook " rerun-hook ", check the log " log-file))
	      (debug:print-info 0 *default-log-port* "running rerun-hook: \"" rerun-hook "\", log is " actual-logf)
              ;; call the hook
              (debug:print-info 0 *default-log-port* "Calling rerun-hook for " test-id new-test-path testdat rerunlst)
              (debug:print-info 0 *default-log-port* "rerun hook: " rerun-hook)
              (debug:print-info 0 *default-log-port* "tarfilename: " tarfilename)
              (debug:print-info 0 *default-log-port* "rundir: " rundir)
              (debug:print-info 0 *default-log-port* "actual-logf: " actual-logf)
              (debug:print-info 0 *default-log-port* "runname: " runname)
              (debug:print-info 0 *default-log-port* "sys-call-text: " sys-call-text)
	      (system sys-call-text)
	      (debug:print-info 0 *default-log-port* "rerun-hook \"" rerun-hook "\" took " (- (current-seconds) start-time) " seconds to run."))))))




;; return #t when all items in waitors-upon list are represented in test-patt, #f otherwise.
(define (runs:testpatts-mention-waitors-upon? test-patt waitors-upon)
  (null? (tests:filter-test-names-not-matched waitors-upon test-patt)))

(define (get-dir-up-n dir . params) 
  (let ((dparts  (string-split dir "/"))
	(count   (if (null? params) 1 (car params))))
    (conc "/" (string-intersperse 
	       (take dparts (- (length dparts) count))
	       "/"))))

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

(define (runs:safe-delete-test-dir real-dir)
  ;; first delete all sub-directories
  (directory-fold 
   (lambda (f x)
     (let ((fullname (conc real-dir "/" f)))
       (if (directory? fullname)(runs:recursive-delete-with-error-msg fullname)))
     (+ 1 x))
   0 real-dir)
  ;; then files other than *testdat.db*
  (directory-fold 
   (lambda (f x)
     (let ((fullname (conc real-dir "/" f)))
       (if (not (string-search (regexp "testdat.db") f))
	   (runs:recursive-delete-with-error-msg fullname)))
     (+ 1 x))
   0 real-dir #t)
  ;; then the entire directory
  (runs:recursive-delete-with-error-msg real-dir))

;; cleanup often needs to remove all but the last N runs per target
;;
;; target-patts a1/b1/c1,a2/b2/c2 ...
;;
;; This will fail if called with empty target or a bad target (i.e. missing or extra fields)
;;
(define (runs:get-hash-by-target target-patts runpatt)
  (let* ((targets (string-split target-patts ","))
	 (keys    (rmt:get-keys))
	 (res-ht  (make-hash-table))) ;; target -> ( runrecord1 runrecord2 ... )
    (for-each
     (lambda (target-patt)
       (let ((runs     (rmt:simple-get-runs runpatt #f #f target-patt #f)))
	 (for-each
	  (lambda (run)
	    (let ((target (simple-run-target run)))

  (let* ((dparts  (string-split path-in "/"))
    (path-out (conc "/" (string-intersperse (take dparts (- (length dparts) 1)) "/")))
    )
    path-out
  )
)


(define (runs:operate-on action target runnamepatt testpatt #!key (state #f)(status #f)(new-state-status #f)(mode #f)(options '()))
  (common:clear-caches) ;; clear all caches
  (let* ((db           #f)
	 ;; (tdbdat       (tasks:open-db))
	 (keys         (rmt:get-keys))
	 (rundat       (mt:get-runs-by-patt keys runnamepatt target))
	 (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)))
	 (rp-mutex     (make-mutex))

             (print "db archive started")  
             (archive:megatest-db target runnamepatt)
             (print "db archived")))
    )
  #t
  )

(define (runs:remove-test-directory test mode) ;; remove-data-only)
  (let* ((run-dir       (db:test-get-rundir test))    ;; run dir is from the link tree
	 (real-dir      (if (common:file-exists? run-dir)
			    ;; (resolve-pathname run-dir)
			    (common:nice-path run-dir)
			    #f))
         (clean-mode    (or mode 'remove-all))
         (test-id       (db:test-get-id test))
        ;; (lock-key      (conc "test-" test-id))
        ;; (got-lock      (let loop ((lock        (rmt:no-sync-get-lock lock-key))
	;; 			     (expire-time (+ (current-seconds) 30))) ;; give up on getting the lock and steal it after 15 seconds
	;; 		    (if (car lock)
	;; 			#t
	;; 			(if (> (current-seconds) expire-time)
	;; 			    (begin
	;; 			      (debug:print-info 0 *default-log-port* "Timed out waiting for a lock to clean test with id " test-id)
	;; 			      (rmt:no-sync-del! lock-key) ;; destroy the lock
	;; 			      (loop (rmt:no-sync-get-lock lock-key) expire-time)) ;; 
	;; 			    (begin
	;; 			      (thread-sleep! 1)
	 ;; 			      (loop (rmt:no-sync-get-lock lock-key) expire-time)))))))
	 )
    (case clean-mode
      ((remove-data-only)(mt:test-set-state-status-by-id (db:test-get-run_id test)(db:test-get-id test) "CLEANING" "LOCKED" #f))
      ((remove-all)      (mt:test-set-state-status-by-id (db:test-get-run_id test)(db:test-get-id test) "REMOVING" "LOCKED" #f))
      ((archive-remove)  (mt:test-set-state-status-by-id (db:test-get-run_id test)(db:test-get-id test) "ARCHIVE_REMOVING" #f #f)))
    (debug:print-info 1 *default-log-port* "Attempting to remove " (if real-dir (conc " dir " real-dir " and ") "") " link " run-dir)
    (if (and real-dir 
	     (> (string-length real-dir) 5)
	     (common:file-exists? real-dir)) ;; bad heuristic but should prevent /tmp /home etc.
	(let* ((realpath (realpath run-dir)))
	  (debug:print-info 1 *default-log-port* "Recursively removing " realpath)
	  (if (common:file-exists? realpath)
	      (runs:safe-delete-test-dir realpath)
	      (debug:print 0 *default-log-port* "WARNING: test dir " realpath " appears to not exist or is not readable")))
	(if real-dir 
	    (debug:print 0 *default-log-port* "WARNING: directory " real-dir " does not exist")
	    (debug:print 0 *default-log-port* "WARNING: no real directory corrosponding to link " run-dir ", nothing done")))
    (if (symbolic-link? run-dir)
	(begin
	  (debug:print-info 1 *default-log-port* "Removing symlink " run-dir)
	  (handle-exceptions
	   exn
	   (debug:print-error 0 *default-log-port* " Failed to remove symlink " run-dir ((condition-property-accessor 'exn 'message) exn) ", attempting to continue, exn=" exn)
	   (delete-file run-dir)))
	(if (directory? run-dir)
	    (if (> (directory-fold (lambda (f x)(+ 1 x)) 0 run-dir) 0)
		(debug:print 0 *default-log-port* "WARNING: refusing to remove " run-dir " as it is not empty")
		(handle-exceptions
		 exn
		 (debug:print-error 0 *default-log-port* " Failed to remove directory " run-dir ((condition-property-accessor 'exn 'message) exn) ", attempting to continue, exn=" exn)
		 (delete-directory run-dir)))
	    (if (and run-dir
		     (not (member run-dir (list "n/a" "/tmp/badname"))))
		(debug:print 0 *default-log-port* "WARNING: not removing " run-dir " as it either doesn't exist or is not a symlink")
		(debug:print 0 *default-log-port* "NOTE: the run dir for this test is undefined. Test may have already been deleted."))
	    ))
    ;; Only delete the records *after* removing the directory. If things fail we have a record 
    (case clean-mode
      ((remove-data-only)(mt:test-set-state-status-by-id (db:test-get-run_id test)(db:test-get-id test) (db:test-get-state test)(db:test-get-status test) #f))
      ((archive-remove)  (mt:test-set-state-status-by-id (db:test-get-run_id test)(db:test-get-id test) "ARCHIVED" #f #f))
      (else (rmt:delete-test-records (db:test-get-run_id test) (db:test-get-id test))))
    ;; (rmt:no-sync-del! lock-key)
    ))

;;======================================================================
;; Routines for manipulating runs
;;======================================================================

;; Since many calls to a run require pretty much the same setup 
;; this wrapper is used to reduce the replication of code
(define (general-run-call switchname action-desc proc)

	     db 
	     (conc "INSERT OR REPLACE INTO test_data (test_id,category,variable,value,expected,tol,units,comment) "
		   "SELECT " (db:test-get-id new-testdat) ",category,variable,value,expected,tol,units,comment FROM test_data WHERE test_id=?;")
	     (db:test-get-id testdat))))
	 ))
     prev-tests)))

(define doc-template 
  '(*TOP*
    (*PI* xml "version='1.0'")
    (testsuite)))

(define (runs:update-junit-test-reporter-xml run-id)
  (let*	((junit-test-reporter	(configf:lookup *configdat* "runs" "junit-test-reporter-xml"))
	 (junit-test-report-dir  (configf:lookup *configdat* "runs" "junit-test-report-dir"))
	 (xml-dir		(if (and junit-test-reporter (equal? junit-test-reporter "yes" ))
				    (if junit-test-report-dir
					junit-test-report-dir
					(conc (getenv "MT_LINKTREE") "/" (getenv "MT_TARGET") "/" (getenv "MT_RUNNAME")))
				    #f))
	 (xml-ts-name		(if xml-dir
				    (conc (getenv "MT_TESTSUITENAME")"."(string-translate (getenv "MT_TARGET") "/" ".") "." (getenv "MT_RUNNAME"))
				    #f))
         (keyname               (if xml-ts-name (common:get-signature xml-ts-name) #f))
	 (xml-path		(if xml-dir
				    (conc xml-dir "/" keyname ".xml")
				    #f))

	 (test-data		(if xml-dir
				    (rmt:get-tests-for-run run-id "%" '() '() ;; run-id testpatt states statuses
							   #f #f ;; offset limit
							   #f ;; not-in
							   #f ;; sort-by
							   #f ;; sort-order
							   #f ;; get full data (not 'shortlist)
							   0 ;; (runs:gendat-inc-results-last-update *runs:general-data*) ;; last update time
							   #f)
				    '()))
	 (tests-count		(if xml-dir (length test-data) #f)))
    (if (and junit-test-reporter (equal? junit-test-reporter "yes" ))
	(begin
					;((sxml-modify! `("testsuite" insert-into (@ (name ,xml-ts-name) (tests ,tests-count)))) doc)

	  (let loop	((test		(car test-data))
			 (tail		(cdr test-data))
			 (doc		doc-template)
			 (fail-cnt	0)
			 (error-cnt	0))
	    (let*	((test-name	(vector-ref test 2))
			 (test-itempath	(vector-ref test 11))
			 (tc-name	(conc test-name (if (and test-itempath (not (equal? test-itempath ""))) (conc "." (string-translate test-itempath "/" "." )) "")))
			 (test-state	(vector-ref test 3))
			 (comment	(vector-ref test 14))   
			 (test-status	(vector-ref test 4))
			 (exc-msg	(conc "No bucket for State " test-state " Status " test-status))
			 (new-doc	(cond 
					 ((member test-state (list "RUNNING" ))
					  ((sxml-modify `("testsuite" insert-into (testcase (@ (name ,tc-name)) (inProgress)))) doc))
					 ((member test-state (list "LAUNCHED" "REMOTEHOSTSTART"  "NOT_STARTED"))
					  ((sxml-modify `("testsuite" insert-into (testcase (@ (name ,tc-name)) (inQueue)))) doc))
					 ((member test-status (list "PASS" "WARN" "WAIVED"))
					  ((sxml-modify `("testsuite" insert-into (testcase (@ (name ,tc-name))))) doc))
					 ((member test-status (list "FAIL" "CHECK"))
					  ((sxml-modify `("testsuite" insert-into (testcase (@ (name ,tc-name)) (failure (@ (message ,comment) (type "failure")))))) doc)) 
					 ((member test-status (list "DEAD" "KILLED" "ABORT" "PREQ_FAIL" "PREQ_DISCARDED"))
					  ((sxml-modify `("testsuite" insert-into (testcase (@ (name ,tc-name)) (failure (@ (message ,comment) (type "error")))))) doc))
					 ((member test-status (list "SKIP"))
					  ((sxml-modify `("testsuite" insert-into (testcase (@ (name ,tc-name)) (skipped (@ (type "skipped")))))) doc))
					 (else 
					  (debug:print 0 *default-log-port* (conc "What do I do with State " test-state " Status " test-status))
					  ((sxml-modify `("testsuite" insert-into (testcase (@ (name ,tc-name)) (failure (@ (message ,exc-msg) (type "error")))))) doc))))
			 (new-error-cnt	(if (member test-status (list "DEAD" "KILLED" "ABORT" "PREQ_FAIL" "PREQ_DISCARDED"))
					    (+ error-cnt 1) 
					    error-cnt))
			 (new-fail-cnt	(if (member test-status (list "FAIL" "CHECK"))
					    (+ fail-cnt 1)
					    fail-cnt)))
 	      (if (null? tail)
		    (let* ((final-doc ((sxml-modify `("testsuite" insert-into (@ (name ,xml-ts-name) (tests ,tests-count) (errors ,error-cnt) (failures ,fail-cnt)))) new-doc)))
		    (debug:print 0 *default-log-port* "modify attrib error=" error-cnt " fail= " fail-cnt)
		    (handle-exceptions
		     exn
		     (let*	((msg	((condition-property-accessor 'exn 'message) exn)))
		       (debug:print 0 *default-log-port* (conc "WARNING: Failed to update file" xml-path". Message:" msg ", exn=" exn)))
		     		   
		     (if (not (file-exists? xml-dir)) 
			 (create-directory xml-dir #t))
                     (if (not (rmt:no-sync-get/default keyname #f)) 
                       (begin
			 (rmt:no-sync-set  keyname "on")
			 (debug:print 0 *default-log-port* "creating xml at " xml-path)
		         (with-output-to-file xml-path
		         (lambda ()
			   (print (sxml-serializer#serialize-sxml final-doc  ns-prefixes: (list (cons 'gnm "http://foo"))))))
                           (rmt:no-sync-del! keyname))
                          (debug:print 0 *default-log-port*  "Could not get the lock. Skip writing the xml file."))))
		  (loop (car tail) (cdr tail) new-doc new-fail-cnt new-error-cnt))))))))
	 
     
;; clean cache files
(define (runs:clean-cache target runname toppath)
  (if target
      (if runname
	  (let* ((linktree (common:get-linktree)) ;; (if toppath (configf:lookup *configdat* "setup" "linktree")))
		 (runtop   (conc linktree "/" target "/" runname))
		 (files    (if (common:file-exists? runtop)

			 exn
			 (debug:print 0 *default-log-port* "WARNING: Failed to remove file " f ", exn=" exn)
		       (delete-file f)))
		   files))))
	  (debug:print-error 0 *default-log-port* "-clean-cache requires -runname."))
      (debug:print-error 0 *default-log-port* "-clean-cache requires -target or -reqtarg")))

;; Spec for End of test
;; At end of each test call, after marking self as COMPLETED do run-state-status-rollup
;; At transition to run COMPLETED/X do hooks
;; Definition: test_dead if event_time + duration + 1 minute? < current_time AND
;; we can prove the process is not alive (ssh host pstree -A pid)
;; if dead safe to mark the test as killed in the db
;; State/status table
;; new
;; 100% COMPLETED/ (PASS,FAIL,ABORT etc.) ==> COMPLETED / X where X is same as itemized rollup
;; > 3 RUNNING with not test_dead do nothing (run should already be RUNNING/ na
;; > 0 RUNNING and test_dead then send KILLREQ ==> COMPLETED
;; 0 RUNNING ==> this is actually the first condition, should not get here

(define (runs:end-of-run-check run-id )
  (let* ((not-completed-cnt (rmt:get-not-completed-cnt run-id))  
	 (running-cnt (rmt:get-count-tests-running-for-run-id run-id))
	 (all-test-launched (rmt:get-var (conc "lunch-complete-" run-id)))
	 (current-state (rmt:get-run-state run-id))
	 (current-status (rmt:get-run-status run-id)))
    ;;get-vars run-id to query metadata table to check if all completed. if all-test-launched = yes then only not-completed-cnt = 0 means everyting is completed if no entry found in the table do nothing 
    (debug:print 0 *default-log-port* "Running test cnt :" running-cnt)                      
    (rmt:set-state-status-and-roll-up-run  run-id current-state current-status)
    (runs:update-junit-test-reporter-xml run-id) 
    (cond 
     ((and all-test-launched (eq? not-completed-cnt 0) (equal? all-test-launched "yes" ))
      (if (and (equal? (rmt:get-var (conc "end-of-run-" run-id)) "no") (common:simple-lock (conc "endOfRun" run-id)))
	  (begin
	    (debug:print 4 *default-log-port* "look for  post hook. currseconds: " (current-seconds) " EOR " (rmt:get-var (conc "end-of-run-" run-id)))
	    (debug:print 0 *default-log-port* "End of Run Detected.")
	    (rmt:set-var (conc "end-of-run-" run-id) "yes")
					;(thread-sleep! 10)
	    (runs:run-post-hook run-id)
	    (debug:print 4 *default-log-port* "currseconds: " (current-seconds)" eor: " (rmt:get-var (conc "end-of-run-" run-id)))
	    (common:simple-unlock (conc "endOfRun" run-id)))
	  (debug:print 0 *default-log-port* "End of Run Detected but not running post hook. This should happen when eor is set to yes. This will happen only when 2 tests exit at smae time. eor= " (rmt:get-var (conc "end-of-run-" run-id)))))
     ((> running-cnt 3) 
      (debug:print 0 *default-log-port* "There are " running-cnt " tests running." ))
     ((> running-cnt 0)
      (debug:print 0 *default-log-port* "running cnt > 0 but <= 3 kill-running-tests-if-dead" )
      (let ((kill-cnt (launch:kill-tests-if-dead run-id)))
	(if (and all-test-launched  (equal? all-test-launched "yes") (eq? kill-cnt running-cnt))
	    (runs:end-of-run-check run-id)))) ;;todo
     (else  (debug:print 0 *default-log-port* "Should it get here?? May be everything is not launched yet. Running test cnt:" running-cnt " Not completed test cnt:" not-completed-cnt)
	    (let* ((not-completed-tests (rmt:get-tests-for-run run-id "%" `("NOT_STARTED" "RUNNING" "LAUNCHED" "REMOTEHOSTSTART") `() #f #f #f #f #f #f #f #f)))
	      (if (> (length not-completed-tests) 0) 
		  (let loop ((running-test (car not-completed-tests))
			     (tal    (cdr not-completed-tests)))
		    (let* ((test-name (vector-ref running-test 2))
			   (item-path (vector-ref running-test 11)))
		      (debug:print 0 *default-log-port* "test " test-name "/" item-path " not completed")
		      (if (not (null? tal))
			  (loop (car tal) (cdr tal)))))))))))

(define (runs:find-and-mark-incomplete-and-check-end-of-run run-id ovr-deadtime)
  (rmt:find-and-mark-incomplete run-id ovr-deadtime)
  (runs:end-of-run-check run-id))

;; kill any runner processes (i.e. processes handling -runtests) that match target/runname
;; 
;; do a remote call to get the task queue info but do the killing as self here.
;;
(define (tasks:kill-runner target run-name testpatt)
  (let ((records    (rmt:tasks-find-task-queue-records target run-name testpatt "running" "run-tests"))
	(hostpid-rx (regexp "\\s+(\\w+)\\s+(\\d+)$"))) ;; host pid is at end of param string