Megatest

;; 
;;     You should have received a copy of the GNU General Public License
;;     along with Megatest.  If not, see <http://www.gnu.org/licenses/>.

;;  strftime('%m/%d/%Y %H:%M:%S','now','localtime')

(use (prefix sqlite3 sqlite3:) srfi-1 posix regex regex-case srfi-69 (srfi 18) 
     posix-extras directory-utils pathname-expand typed-records format  sxml-serializer sxml-modifications)

(declare (unit runs))
(declare (uses db))
(declare (uses common))
(declare (uses items))
(declare (uses runconfig))
(declare (uses tests))

(define (runs:can-run-more-tests runsdat run-id jobgroup max-concurrent-jobs)

  ;; Take advantage of a good place to exit if running the one-pass methodology
  (if (and (> (runs:dat-can-run-more-tests-count runsdat) 20)
	   (args:get-arg "-one-pass"))
      (exit 0))

  (thread-sleep! (cond ;; BB: check with Matt.  Should this sleep move
		       ;; to cond clauses below where we determine we
		       ;; have too many jobs running rather than each
		       ;; time the and condition above is true (which
		       ;; seems like always)?
        	  ((> (runs:dat-can-run-more-tests-count runsdat) 20) ;; original intent was - save cycles, wait a long time
		   (if (runs:lownoise "waiting on tasks" 60)(debug:print-info 2 *default-log-port* "waiting for tasks to complete, sleeping briefly ..."))

		   10)  ;; obviously haven't had any work to do for a while
		  (else
		   ;; if have a number for inter-test-delay, use it, else don't delay much, maybe even zero?
		   (configf:lookup-number *configdat* "setup" "inter-test-delay" default: 0.01))))
  
  (let* ((num-running             (rmt:get-count-tests-running run-id))
	 (num-running-in-jobgroup (rmt:get-count-tests-running-in-jobgroup run-id jobgroup))
	 (job-group-limit         (let ((jobg-count (configf:lookup *configdat* "jobgroups" jobgroup)))
				    (if (string? jobg-count)
					(string->number jobg-count)
					jobg-count))))

    (runs:update-all-test_meta #f)

    ;; run the run prehook if there are no tests yet run for this run:
    ;;
    (runs:run-pre-hook run-id)
    ;; mark all test launced flag as false in the meta table 
    (rmt:set-var (conc "lunch-complete-" run-id) "no")
    (debug:print-info 1 *default-log-port* "Setting end-of-run to no")
    (let* ((config-reruns      (let ((x (configf:lookup *configdat* "setup" "reruns")))
			       (if x (string->number x) #f)))
	  (config-rerun-cnt (if config-reruns
			config-reruns
			1)))
    (if (eq? config-rerun-cnt run-count)
      (rmt:set-var (conc "end-of-run-" run-id) "no")))

    (rmt:set-run-state-status run-id "new" "n/a")
    ;; now add non-directly referenced dependencies (i.e. waiton)
    ;;======================================================================
    ;; refactoring this block into tests:get-full-data
    ;;
    ;; What happended, this code is now duplicated in tests!?
    ;;

	    (if (> run-count 0) ;; handle reruns
		(begin
		  (if (not (hash-table-ref/default flags "-preclean" #f))
		      (hash-table-set! flags "-preclean" #t))
		  (if (not (hash-table-ref/default flags "-rerun" #f))
		      (hash-table-set! flags "-rerun" "STUCK/DEAD,n/a,ZERO_ITEMS"))
		  ;; recursive call to self
      (runs:run-tests target runname test-patts user flags run-count: (- run-count 1)))
                  (launch:end-of-run-check run-id)))
	  (debug:print-info 0 *default-log-port* "No tests to run")))
    (debug:print-info 4 *default-log-port* "All done by here")
    ;; TODO: try putting post hook call here
      


    ;  (debug:print-info 2 *default-log-port* " run-count " run-count)
    ;  (runs:run-post-hook run-id))
    ;  (debug:print-info 2 *default-log-port* "Not calling post hook runcount = " run-count ))   
    (rmt:tasks-set-state-given-param-key task-key "done")
     
    ;; (sqlite3:finalize! tasks-db)
    ))


;; loop logic. These are used in runs:run-tests-queue to make it a bit more readable.
;;
;; If reg not full and have items in tal then loop with (car tal)(cdr tal) reg reruns

  (let ((curr-sec (current-seconds)))
    (if (> (- curr-sec (runs:gendat-inc-results-last-update *runs:general-data*)) 5) ;; at least five seconds since last update
	(let* ((run-dat  (or (runs:gendat-run-info *runs:general-data*)(rmt:get-run-info run-id)))
	       (runname  (or (runs:gendat-runname *runs:general-data*)
			     (db:get-value-by-header (db:get-rows run-dat)
						     (db:get-header run-dat) "runname")))
	       (target   (or (runs:gendat-target *runs:general-data*)(rmt:get-target run-id)))
	       (testsdat (let ((res (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)
				     (runs:gendat-inc-results-last-update *runs:general-data*) ;; last update time
				     'dashboard)))
			   (if (list? res)
			       res
			       (begin
				 (debug:print-error
				  0 *default-log-port*
				  "FAILED TO GET DATA using rmt:get-tests-for-run. Notify developers if you see this. result: " res)
				 '())))))
	  (if (not (runs:gendat-run-info *runs:general-data*))
	      (runs:gendat-run-info-set! *runs:general-data* run-dat))
	  (if (not (runs:gendat-runname  *runs:general-data*))
	      (runs:gendat-runname-set! *runs:general-data* runname))
	  (if (not (runs:gendat-target *runs:general-data*))
	      (runs:gendat-target-set! *runs:general-data* target))
	  (for-each

	  (debug:print-info 0 *default-log-port* "Have leftovers!")
	  (loop (car reg)(cdr reg) '() reruns))
	 (else
          (debug:print-info 4 *default-log-port* "cond branch - "  "rtq-9")
	  (debug:print-info 4 *default-log-port* "Exiting loop with...\n  hed=" hed "\n  tal=" tal "\n  reruns=" reruns))
	 ))) ;; end loop on sorted test names
    ;; this is the point where everything is launched and now you can mark the run in metadata table as all launched 
    (rmt:set-var (conc "lunch-complete-" run-id) "yes")  
        
    ;; now *if* -run-wait we wait for all tests to be done
    ;; Now wait for any RUNNING tests to complete (if in run-wait mode)
    (thread-sleep! 10) ;; I think there is a race condition here. Let states/statuses settle
    (let wait-loop ((num-running      (rmt:get-count-tests-running-for-run-id run-id))
		    (prev-num-running 0))
      ;; (debug:print-info 13 *default-log-port* "num-running=" num-running ", prev-num-running=" prev-num-running)
      (if (and (or (args:get-arg "-run-wait")

  ;; 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 ...
;;

			))
		   actions))))
          sorted)))
     ;; (print "Sorted: " (map simple-run-event_time sorted))
     ;; (print "Remove: " (map simple-run-event_time to-remove))))
     (hash-table-keys runs-ht))
    runs-ht))

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


;; (define (runs:remove-all-but-last-n-runs-per-target target-patts runpatt num-to-keep)
;;   (let ((data (runs:get-all-but-most-recent-n-per-target target-patts runpatt num-to-keep)))
;;     (for-each
;;      (lambda (target)
;;        (let ((runs-to-remove (hash-table-ref data target )))
;;          (for-each

	 (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))
	 (bup-mutex    (make-mutex))
         (keep-records (args:get-arg "-keep-records")) ;; used in conjunction with -remove-runs to keep the records, TODO: consolidate this with "mode".
	 (test-records '())) ;; for tasks that we wish to operate on all tests in one fell swoop

    (let* ((write-access-actions '(remove-runs set-state-status archive run-wait kill-runs))
           (dbfile             (conc  *toppath* "/megatest.db"))
           (readonly-mode      (not (file-write-access? dbfile))))
      (when (and readonly-mode
                 (member action write-access-actions))
        (debug:print-error 0 *default-log-port* "megatest.db is readonly.  Cannot proceed with action ["action"] in which write-access isrequired .")

	 (let* ((run-id    (db:get-value-by-header run header "id"))
		(run-state (db:get-value-by-header run header "state"))
		(run-name  (db:get-value-by-header run header "runname"))
		(tests     (if (not (equal? run-state "locked"))
			       (proc-get-tests run-id)
			       '()))
		(lasttpath "/does/not/exist/I/hope")
		(lastrealpath "/does/not/exist/I/hope")
                ;; there may be a number of different disks used in the same run.
                (run-paths-hash (make-hash-table))
		(worker-thread #f))
	   (debug:print-info 4 *default-log-port* "runs:operate-on run=" run ", header=" header)
	   (if (not (null? tests))
	       (begin
		 (case action
                   ((kill-runs)
                    (tasks:kill-runner target run-name "%")

		   ((print-run)
		    (debug:print 1 *default-log-port* "Printing info for run " runkey ", run=" run ", tests=" tests ", header=" header)
		    action)
		   ((run-wait)
		    (debug:print 1 *default-log-port* "Waiting for run " runkey ", run=" runnamepatt " to complete"))
		   ((archive)
		    (debug:print 1 *default-log-port* "Archiving/restoring (" (args:get-arg "-archive") ") data for run: " runkey " " (db:get-value-by-header run header "runname"))
		    (let ((op (string->symbol (args:get-arg "-archive"))))
		      (set! worker-thread
			    (make-thread
			     (lambda ()
			       (case op
				 ((save save-remove keep-html)
				  (archive:run-bup op run-id run-name tests rp-mutex bup-mutex))
				 ((restore)
				  (archive:bup-restore op run-id run-name tests rp-mutex bup-mutex))
				 ((get) ;;; NOTE: This is a special case. We wish to operate on ALL tests in one go
				  (set! test-records (append tests test-records)))
				 (else 
				  (debug:print-error 0 *default-log-port* "unrecognised sub command " op " for -archive. Run \"megatest\" to see help")
				  (exit))))
			     "archive-bup-thread"))
		      (thread-start! worker-thread)
		      (if (eq? op 'get)
			  (thread-join! worker-thread)) ;; we need the test-records set to not overlap
		      ))
		   (else
		    (debug:print-info 0 *default-log-port* "action not recognised " action)))
		 
		 ;; actions that operate on one test at a time can be handled below
		 ;;
		 (let ((sorted-tests     (filter 
					  vector?

                                              (mt:test-set-state-status-by-id run-id (db:test-get-id test) "KILLREQ" "n/a" #f)
                                              (thread-sleep! 1)))
                                        ;; NOTE: This is suboptimal as the testdata will be used later and the state/status may have changed ...
                                        (if (null? tal)
                                            (loop new-test-dat tal)
                                            (loop (car tal)(append tal (list new-test-dat)))))
                                      (begin
                                       (let ((rundir (db:test-get-rundir new-test-dat)))
                                        (if (and (not (string=  rundir "/tmp/badname")) 
                                             (file-exists? rundir)
                                             (substring-index run-name rundir)
                                             (substring-index target rundir)
                                             )
                                          (begin
                                            (set! lasttpath (db:test-get-rundir new-test-dat)) ;; remember this path for run removal
                                            (set! lastrealpath (remove-last-path-directory (resolve-pathname lasttpath)))
                                            (hash-table-set! run-paths-hash lastrealpath 1)
                                            (runs:remove-test-directory new-test-dat mode) ;; 'remove-all)
                                          )
                                          (begin
                                            (debug:print 2 *default-log-port* "Not removing directory " rundir " because either it doesn't exist or has a bad name")
                                            (debug:print 2 *default-log-port* "Is /tmp/badname: " (string=  rundir "/tmp/badname"))
                                            (debug:print 2 *default-log-port* "Exists: " (file-exists? rundir))
                                            (debug:print 2 *default-log-port* "Has run-name: " (substring-index run-name rundir))
                                            (debug:print 2 *default-log-port* "Has target: " (substring-index target rundir))
                                            ;;PJH remove record from db no need to cleanup directory
                                            (case 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))))

                                          )
                                        )
                                      )

                                      (if (not (null? tal))
                                         (loop (car tal)(cdr tal)))))))
				(rmt:update-run-stats run-id (rmt:get-raw-run-stats run-id)))
                               ((kill-runs)
                                ;; RUNNING -> KILLREQ
                                ;; LAUNCHED,RUNNING,REMOTEHOSTSTART -> NOT STARTED
                                (cond
                                 ((and has-subrun (member test-state (list "RUNNING" "LAUNCHED" "REMOTEHOSTSTART" "KILLREQ")))
                                  (common:send-thunk-to-background-thread

					     (debug:print-info 0 *default-log-port* "Estimating disk space usage for " test-fulln ": " (common:get-disk-space-used ddir)))))))
				(if (not (null? tal))
				    (loop (car tal)(cdr tal))))
			       )))
		       )
		     (if worker-thread (thread-join! worker-thread)))
                   (common:join-backgrounded-threads))))
	   
	   ;; remove the run if zero tests remain
	   (if (eq? action 'remove-runs)
	       (let* ((run-id   (db:get-value-by-header run header "id")) ;; NB// masks run-id from above?
                      (remtests (mt:get-tests-for-run run-id #f '("DELETED") '("n/a") not-in: #t)))
		 (if (null? remtests) ;; no more tests remaining
	           (let* ((linkspath (remove-last-path-directory lasttpath))
                          (runpaths (hash-table-keys run-paths-hash))
                    )

                    (debug:print 2 *default-log-port* "run-paths-hash: " (hash-table-keys run-paths-hash))

                    (debug:print 1 *default-log-port* "Removing target " target "run: " run-name)
                    (if (not keep-records)
                      (begin
                        (debug:print 1 *default-log-port* "Removing DB records for the run.")
                        (rmt:delete-run run-id)
                        (rmt:delete-old-deleted-test-records))


                    )
                    (if (not (equal?  linkspath "/does/not/exist/I"))
	               (begin 
                         (debug:print 1 *default-log-port* "Recursively removing links dir " linkspath)
                         (runs:recursive-delete-with-error-msg linkspath)))

                   (for-each (lambda(runpath)
                       (debug:print 1 *default-log-port* "Recursively removing runs dir " runpath)
                       (runs:recursive-delete-with-error-msg runpath)
                     )
                     runpaths
                   )
		       )))))
	 ))
     runs)

    ;; special case - archive get
    (if (equal? (args:get-arg "-archive") "get")
	(archive:bup-get-data "get" #f #f test-records rp-mutex bup-mutex))
    )
  #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))

      ((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 (resolve-pathname 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

;;======================================================================
;; 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)
  (let ((runname (common:args-get-runname))
	(target  (common:args-get-target)))
    (cond
     ((not target)
      (debug:print-error 0 *default-log-port* "Missing required parameter for " switchname ", you must specify the target with -target")
      (exit 3))
     ((not runname)
      (debug:print-error 0 *default-log-port* "Missing required parameter for " switchname ", you must specify the run name with -runname runname")

	    (sqlite3:execute 
	     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)))
		     		   
		     (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")))