Megatest

(use sqlite3 srfi-1 posix regex regex-case srfi-69 dot-locking tcp directory-utils)
(import (prefix sqlite3 sqlite3:))

(declare (unit tests))
(declare (uses lock-queue))
(declare (uses db))

(declare (uses common))
(declare (uses items))
(declare (uses runconfig))


(include "common_records.scm")
(include "key_records.scm")
(include "db_records.scm")
(include "run_records.scm")
(include "test_records.scm")

		     (qry        (conc "(" test-qry " AND " item-qry ")")))
		;; (print "tests:match => patt-parts: " patt-parts ", test-patt: " test-patt ", item-patt: " item-patt)
		(if (null? tal)
		    (string-intersperse (append (reverse res)(list qry)) " OR ")
		    (loop (car tal)(cdr tal)(cons qry res)))))))
      #f))

;; get the previous record for when this test was run where all keys match but runname
;; returns #f if no such test found, returns a single test record if found
;; 
;; Run this server-side
;;
(define (test:get-previous-test-run-record db run-id test-name item-path)
  (let* ((keys    (db:get-keys db))
	 (selstr  (string-intersperse  keys ","))
	 (qrystr  (string-intersperse (map (lambda (x)(conc x "=?")) keys) " AND "))
	 (keyvals #f))
    ;; first look up the key values from the run selected by run-id
    (sqlite3:for-each-row 
     (lambda (a . b)
       (set! keyvals (cons a b)))
     db
     (conc "SELECT " selstr " FROM runs WHERE id=? ORDER BY event_time DESC;") run-id)
    (if (not keyvals)
	#f
	(let ((prev-run-ids '()))
	  (apply sqlite3:for-each-row
		 (lambda (id)
		   (set! prev-run-ids (cons id prev-run-ids)))
		 db
		 (conc "SELECT id FROM runs WHERE " qrystr " AND id != ?;") (append keyvals (list run-id)))
	  ;; for each run starting with the most recent look to see if there is a matching test
	  ;; if found then return that matching test record
	  (debug:print 4 "selstr: " selstr ", qrystr: " qrystr ", keyvals: " keyvals ", previous run ids found: " prev-run-ids)
	  (if (null? prev-run-ids) #f
	      (let loop ((hed (car prev-run-ids))
			 (tal (cdr prev-run-ids)))
		(let ((results (db:get-tests-for-run db hed (conc test-name "/" item-path)'() '() #f #f #f #f #f)))
		  (debug:print 4 "Got tests for run-id " run-id ", test-name " test-name ", item-path " item-path ": " results)
		  (if (and (null? results)
			   (not (null? tal)))
		      (loop (car tal)(cdr tal))
		      (if (null? results) #f
			  (car results))))))))))
    
;; get the previous records for when these tests were run where all keys match but runname
;; NB// Merge this with test:get-previous-test-run-records? This one looks for all matching tests
;; can use wildcards. Also can likely be factored in with get test paths?
;;
;; Run this remotely!!
;;
(define (test:get-matching-previous-test-run-records db run-id test-name item-path)
  (let* ((keys    (db:get-keys db))
	 (selstr  (string-intersperse (map (lambda (x)(vector-ref x 0)) keys) ","))
	 (qrystr  (string-intersperse (map (lambda (x)(conc (vector-ref x 0) "=?")) keys) " AND "))
	 (keyvals #f)
	 (tests-hash (make-hash-table)))
    ;; first look up the key values from the run selected by run-id
    (sqlite3:for-each-row 
     (lambda (a . b)
       (set! keyvals (cons a b)))
     db
     (conc "SELECT " selstr " FROM runs WHERE id=? ORDER BY event_time DESC;") run-id)
    (if (not keyvals)
	'()
	(let ((prev-run-ids '()))
	  (apply sqlite3:for-each-row
		 (lambda (id)
		   (set! prev-run-ids (cons id prev-run-ids)))
		 db
		 (conc "SELECT id FROM runs WHERE " qrystr " AND id != ?;") (append keyvals (list run-id)))
	  ;; collect all matching tests for the runs then
	  ;; extract the most recent test and return that.
	  (debug:print 4 "selstr: " selstr ", qrystr: " qrystr ", keyvals: " keyvals 
		       ", previous run ids found: " prev-run-ids)
	  (if (null? prev-run-ids) '()  ;; no previous runs? return null
	      (let loop ((hed (car prev-run-ids))
			 (tal (cdr prev-run-ids)))
		(let ((results (db:get-tests-for-run db hed (conc test-name "/" item-path) '() '() #f #f #f #f #f)))
		  (debug:print 4 "Got tests for run-id " run-id ", test-name " test-name 
			       ", item-path " item-path " results: " (intersperse results "\n"))
		  ;; Keep only the youngest of any test/item combination
		  (for-each 
		   (lambda (testdat)
		     (let* ((full-testname (conc (db:test-get-testname testdat) "/" (db:test-get-item-path testdat)))
			    (stored-test   (hash-table-ref/default tests-hash full-testname #f)))
		       (if (or (not stored-test)
			       (and stored-test
				    (> (db:test-get-event_time testdat)(db:test-get-event_time stored-test))))
			   ;; this test is younger, store it in the hash
			   (hash-table-set! tests-hash full-testname testdat))))
		   results)
		  (if (null? tal)
		      (map cdr (hash-table->alist tests-hash)) ;; return a list of the most recent tests
		      (loop (car tal)(cdr tal))))))))))

;; Check for waiver eligibility
;;
(define (tests:check-waiver-eligibility testdat prev-testdat)
  (let* ((test-registry (make-hash-table))
	 (testconfig  (tests:get-testconfig (db:test-get-testname testdat) test-registry #f))

	 (test-rundir (db:test-get-rundir testdat))

	 (prev-rundir (db:test-get-rundir prev-testdat))
	 (waivers     (configf:section-vars testconfig "waivers"))
	 (waiver-rx   (regexp "^(\\S+)\\s+(.*)$"))
	 (diff-rule   "diff %file1% %file2%")
	 (logpro-rule "diff %file1% %file2% | logpro %waivername%.logpro %waivername%.html"))
    (if (not (file-exists? test-rundir))
	(begin
	  (debug:print 0 "ERROR: test run directory is gone, cannot propagate waiver")

				    (if (null? tal)
					#t
					(loop (car tal)(cdr tal)))
				    #f))))))
	    (pop-directory)
	    result)))))

(define (tests:test-force-state-status! test-id state status)
  (cdb:test-set-status-state *runremote* test-id status state #f)
  (mt:process-triggers test-id state status))

;; Do not rpc this one, do the underlying calls!!!
(define (tests:test-set-status! test-id state status comment dat #!key (work-area #f))
  (debug:print-info 4 "tests:test-set-status! test-id=" test-id ", state=" state ", status=" status ", dat=" dat)
  (let* ((db          #f)
	 (real-status status)
	 (otherdat    (if dat dat (make-hash-table)))
	 (testdat     (cdb:get-test-info-by-id *runremote* test-id))
	 (run-id      (db:test-get-run_id testdat))
	 (test-name   (db:test-get-testname   testdat))
	 (item-path   (db:test-get-item-path testdat))
	 ;; before proceeding we must find out if the previous test (where all keys matched except runname)
	 ;; was WAIVED if this test is FAIL

	 ;; NOTES:
	 ;;  1. Is the call to test:get-previous-run-record remotified?
	 ;;  2. Add test for testconfig waiver propagation control here
	 ;;
	 (prev-test   (if (equal? status "FAIL")
			  (cdb:remote-run test:get-previous-test-run-record #f run-id test-name item-path)
			  #f))
	 (waived   (if prev-test
		       (if prev-test ;; true if we found a previous test in this run series
			   (let ((prev-status  (db:test-get-status  prev-test))
				 (prev-state   (db:test-get-state   prev-test))
				 (prev-comment (db:test-get-comment prev-test)))
			     (debug:print 4 "prev-status " prev-status ", prev-state " prev-state ", prev-comment " prev-comment)

	(set! real-status "WAIVED"))

    (debug:print 4 "real-status " real-status ", waived " waived ", status " status)

    ;; update the primary record IF state AND status are defined
    (if (and state status)
	(begin
	  (cdb:test-set-status-state *runremote* test-id real-status state (if waived waived comment))
	  (mt:process-triggers test-id state real-status)))
    
    ;; if status is "AUTO" then call rollup (note, this one modifies data in test
    ;; run area, it does remote calls under the hood.
    (if (and test-id state status (equal? status "AUTO")) 
	(db:test-data-rollup #f test-id status work-area: work-area))

    ;; add metadata (need to do this way to avoid SQL injection issues)

    ;; :first_err
    ;; (let ((val (hash-table-ref/default otherdat ":first_err" #f)))
    ;;   (if val
    ;;       (sqlite3:execute db "UPDATE tests SET first_err=? WHERE run_id=? AND testname=? AND item_path=?;" val run-id test-name item-path)))

			   variable ","
			   value    ","
			   expected ","
			   tol      ","
			   units    ","
			   dcomment ",," ;; extra comma for status
			   type     )))
	    ;; This was run remote, don't think that makes sense.
	    (db:csv->test-data #f test-id
				dat))))
      
    ;; need to update the top test record if PASS or FAIL and this is a subtest
    (if (not (equal? item-path ""))
	(mt:roll-up-pass-fail-counts run-id test-name item-path status))

    (if (or (and (string? comment)
		 (string-match (regexp "\\S+") comment))
	    waived)
	(let ((cmt  (if waived waived comment)))
	  (cdb:remote-run db:test-set-comment #f test-id cmt)))
    ))


(define (tests:test-set-toplog! db run-id test-name logf) 
  (cdb:client-call *runremote* 'tests:test-set-toplog #t 2 logf run-id test-name))

(define (tests:summarize-items db run-id test-id test-name force)
  ;; if not force then only update the record if one of these is true:
  ;;   1. logf is "log/final.log
  ;;   2. logf is same as outputfilename
  (let* ((outputfilename (conc "megatest-rollup-" test-name ".html"))
	 (orig-dir       (current-directory))
	 (logf-info      (cdb:remote-run db:test-get-logfile-info #f run-id test-name))
	 (logf           (if logf-info (cadr logf-info) #f))
	 (path           (if logf-info (car  logf-info) #f)))
    ;; This query finds the path and changes the directory to it for the test
    (if (and (string? path)
	     (directory? path)) ;; can get #f here under some wierd conditions. why, unknown ...
	(begin
	  (debug:print 4 "Found path: " path)
	  (change-directory path))
	;; (set! outputfilename (conc path "/" outputfilename)))
	(print "No such path: " path))
    (debug:print 4 "summarize-items with logf " logf ", outputfilename " outputfilename " and force " force)
    (if (or (equal? logf "logs/final.log")
	    (equal? logf outputfilename)
	    force)
	(begin
	  (if ;; (not (obtain-dot-lock outputfilename 1 5 7)) ;; retry every second for 20 seconds, call it dead after 30 seconds and steal the lock
              (not (lock-queue:wait-turn outputfilename test-id))
	      (print "Not updating " outputfilename " as another test item has signed up for the job")
	      (begin
		(print "Obtained lock for " outputfilename)
		(let ((oup    (open-output-file outputfilename))
		      (counts (make-hash-table))
		      (statecounts (make-hash-table))
		      (outtxt "")
		      (tot    0)
		      (testdat (cdb:remote-run db:test-get-records-for-index-file #f run-id test-name)))
		  (with-output-to-port
		      oup
		    (lambda ()
		      (set! outtxt (conc outtxt "<html><title>Summary: " test-name 
					 "</title><body><h2>Summary for " test-name "</h2>"))
		      (for-each
		       (lambda (testrecord)

					      "<td>" state    "</td>" 
					      "<td><font color=" (common:get-color-from-status status)
					      ">"   status   "</font></td>"
					      "<td>" (if (equal? comment "")
							 "&nbsp;"
							 comment) "</td>"
							 "</tr>"))))

		       testdat)




		      (print "<table><tr><td valign=\"top\">")
		      ;; Print out stats for status
		      (set! tot 0)
		      (print "<table cellspacing=\"0\" border=\"1\"><tr><td colspan=\"2\"><h2>State stats</h2></td></tr>")
		      (for-each (lambda (state)
				  (set! tot (+ tot (hash-table-ref statecounts state)))
				  (print "<tr><td>" state "</td><td>" (hash-table-ref statecounts state) "</td></tr>"))

				(hash-table-keys counts))
		      (print "<tr><td>Total</td><td>" tot "</td></tr></table>")
		      (print "</td></td></tr></table>")
		      
		      (print "<table cellspacing=\"0\" border=\"1\">" 
			     "<tr><td>Item</td><td>State</td><td>Status</td><td>Comment</td>"
			     outtxt "</table></body></html>")
		      (release-dot-lock outputfilename)))

		  (close-output-port oup)
		  (lock-queue:release-lock outputfilename test-id)
		  (change-directory orig-dir)
		  ;; NB// tests:test-set-toplog! is remote internal...
		  (tests:test-set-toplog! db run-id test-name outputfilename)
		  )))))))
























;;======================================================================
;; Gather data from test/task specifications
;;======================================================================

;; (define (tests:get-valid-tests testsdir test-patts) ;;  #!key (test-names '()))
;;   (let ((tests (glob (conc testsdir "/tests/*")))) ;; " (string-translate patt "%" "*")))))
;;     (set! tests (filter (lambda (test)(file-exists? (conc test "/testconfig"))) tests))

     (lambda (testkeyname)
       (let* ((test-record (hash-table-ref testrecordshash testkeyname))
	      (test-name   (tests:testqueue-get-testname  test-record))
	      (itemdat     (tests:testqueue-get-itemdat   test-record))
	      (item-path   (tests:testqueue-get-item_path test-record))
	      (waitons     (tests:testqueue-get-waitons   test-record))
	      (keep-test   #t)
	      (test-id     (cdb:remote-run db:get-test-id-cached #f run-id test-name item-path))
	      (tdat        (cdb:remote-run db:get-testinfo-state-status #f test-id))) ;; (cdb:get-test-info-by-id *runremote* test-id)))
	 (if tdat
	     (begin
	       ;; Look at the test state and status
	       (if (or (and (member (db:test-get-status tdat) 
				    '("PASS" "WARN" "WAIVED" "CHECK" "SKIP"))
			    (equal? (db:test-get-state tdat) "COMPLETED"))
		       (member (db:test-get-state tdat)
				    '("INCOMPLETE" "KILLED")))
		   (set! keep-test #f))

	       ;; examine waitons for any fails. If it is FAIL or INCOMPLETE then eliminate this test
	       ;; from the runnable list
	       (if keep-test
		   (for-each (lambda (waiton)
			       ;; for now we are waiting only on the parent test
			       (let* ((parent-test-id (cdb:remote-run db:get-test-id-cached #f run-id waiton ""))
				      (wtdat          (cdb:remote-run db:get-testinfo-state-status #f test-id))) ;; (cdb:get-test-info-by-id *runremote* test-id)))
				 (if (or (and (equal? (db:test-get-state wtdat) "COMPLETED")
					      (member (db:test-get-status wtdat) '("FAIL")))
					 (member (db:test-get-status wtdat)  '("KILLED"))
					 (member (db:test-get-state wtdat)   '("INCOMPETE")))
				 ;; (if (or (member (db:test-get-status wtdat)
				 ;;        	 '("FAIL" "KILLED"))
				 ;;         (member (db:test-get-state wtdat)

;;======================================================================
;; test steps
;;======================================================================

;; teststep-set-status! used to be here

(define (test-get-kill-request test-id) ;; run-id test-name itemdat)
  (let* ((testdat   (cdb:remote-run db:get-testinfo-state-status #f test-id))) ;; (cdb:get-test-info-by-id *runremote* test-id))) ;; run-id test-name item-path)))
    (and testdat
	 (equal? (test:get-state testdat) "KILLREQ"))))

(define (test:tdb-get-rundat-count tdb)
  (if tdb
      (let ((res 0))
	(sqlite3:for-each-row
	 (lambda (count)
	   (set! res count))
	 tdb
	 "SELECT count(id) FROM test_rundat;")
	res))
  0)

(define (tests:update-central-meta-info test-id cpuload diskfree minutes uname hostname)
  ;; This is a good candidate for threading the requests to enable
  ;; transactionized write at the server
  (cdb:tests-update-cpuload-diskfree *runremote* test-id cpuload diskfree)
  (if minutes 
      (cdb:tests-update-run-duration *runremote* test-id minutes))
  (if (and uname hostname)
      (cdb:tests-update-uname-host *runremote* test-id uname hostname)))
  

(define (tests:set-full-meta-info db test-id run-id minutes work-area remtries)
  ;; DOES cdb:remote-run under the hood!
  (let ((remtries 10))




    (handle-exceptions
     exn
     (if (> remtries 0)
	 (begin
	   (set! remtries (- remtries 1))
	   (thread-sleep! 10)
	   (tests:set-full-meta-info db test-id run-id minutes work-area (- remtries 1)))
	 (let ((err-status ((condition-property-accessor 'sqlite3 'status #f) exn)))
	   (debug:print 0 "ERROR: tried for over a minute to update meta info and failed. Giving up")
	   (debug:print 0 "EXCEPTION: database probably overloaded or unreadable.")
	   (debug:print 0 " message: " ((condition-property-accessor 'exn 'message) exn))
	   (print "exn=" (condition->list exn))
	   (debug:print 0 " status:  " ((condition-property-accessor 'sqlite3 'status) exn))
	   (print-call-chain)))
     (let* ((num-records 0) ;; (test:tdb-get-rundat-count tdb))
	    (cpuload  (get-cpu-load))
	    (diskfree (get-df (current-directory)))
	    (uname    (get-uname "-srvpio"))
	    (hostname (get-host-name)))
       (tests:update-testdat-meta-info db test-id work-area cpuload diskfree minutes)

       (tests:update-central-meta-info test-id cpuload diskfree minutes uname hostname)))))
    
(define (tests:set-partial-meta-info db test-id run-id minutes work-area remtries)
  ;; DOES cdb:remote-run under the hood!

  (let* ((cpuload  (get-cpu-load))
	 (diskfree (get-df (current-directory)))
	 (remtries 10))
    (handle-exceptions
     exn
     (if (> remtries 0)
	 (begin
	   (set! remtries (- remtries 1))
	   (thread-sleep! 10)
	   (tests:set-full-meta-info db test-id run-id minutes work-area (- remtries 1)))
	 (let ((err-status ((condition-property-accessor 'sqlite3 'status #f) exn)))
	   (debug:print 0 "ERROR: tried for over a minute to update meta info and failed. Giving up")
	   (debug:print 0 "EXCEPTION: database probably overloaded or unreadable.")
	   (debug:print 0 " message: " ((condition-property-accessor 'exn 'message) exn))
	   (print "exn=" (condition->list exn))
	   (debug:print 0 " status:  " ((condition-property-accessor 'sqlite3 'status) exn))
	   (print-call-chain)))
     (tests:update-testdat-meta-info db test-id work-area cpuload diskfree minutes)
    ;; Update central with uname and hostname = #f
    ;; Is this one of the performance problems? This info should come from testdat-meta anyway
    ;; (tests:update-central-meta-info test-id cpuload diskfree minutes #f #f)
  )))
	 
(define (tests:update-testdat-meta-info db test-id work-area cpuload diskfree minutes)
  (let ((tdb         (db:open-test-db-by-test-id db test-id work-area: work-area)))
    (if (sqlite3:database? tdb)
	(begin
	  (sqlite3:execute tdb "INSERT INTO test_rundat (update_time,cpuload,diskfree,run_duration) VALUES (strftime('%s','now'),?,?,?);"
			   cpuload diskfree minutes)
	  (sqlite3:finalize! tdb))
	(debug:print 2 "Can't update testdat.db for test " test-id " read-only or non-existant"))))
    
(define (tests:testdat-get-testinfo db test-id work-area)
   (let ((tdb         (db:open-test-db-by-test-id db test-id work-area: work-area))
	 (res         '()))
     (if (sqlite3:database? tdb)
	 (begin
	   (sqlite3:for-each-row
	    (lambda (update-time cpuload diskfree run-duration)
	      (set! res (cons (vector update-time cpuload diskfree run-duration) res)))
	    tdb
	    "SELECT update_time,cpuload,diskfree,run_duration FROM test_rundat ORDER BY update_time ASC;")
	   (sqlite3:finalize! tdb)))
     res))

;;======================================================================
;; A R C H I V I N G
;;======================================================================

(define (test:archive db test-id)
  #f)

(define (test:archive-tests db keynames target)
  #f)