Megatest

# for the modularized stuff

mofiles/commonmod.o : megatest-fossil-hash.scm mofiles/stml2.o \
                      mofiles/mtargs.o mofiles/pkts.o mofiles/mtconfigf.o \
                      mofiles/processmod.o
mofiles/pgdbmod.o   : mofiles/commonmod.o
mofiles/dbmod.o     : mofiles/commonmod.o mofiles/keysmod.o \
                      mofiles/tasksmod.o mofiles/odsmod.o mofiles/testsmod.o
mofiles/tasksmod.o  : mofiles/commonmod.o mofiles/pgdbmod.o
mofiles/rmtmod.o    : mofiles/commonmod.o \
                      mofiles/apimod.o mofiles/ulex.o mofiles/itemsmod.o
mofiles/apimod.o    : mofiles/dbmod.o mofiles/commonmod.o mofiles/servermod.o
mofiles/runsmod.o   : mofiles/testsmod.o
mofiles/mtmod.o     : mofiles/mtconfigf.o mofiles/rmtmod.o

mofiles/servermod.o : mofiles/commonmod.o mofiles/dbmod.o



# Removed from megamod.o dep:   mofiles/ftail.o
mofiles/megamod.o   : \
   mofiles/rmtmod.o \
   mofiles/commonmod.o \
   mofiles/apimod.o \
   mofiles/archivemod.o \

                   ;;===============================================
                   ;; READ/WRITE QUERIES
                   ;;===============================================

                   ((get-keys-write)                        (db:get-keys dbstruct)) ;; force a dummy "write" query to force server; for debug in -repl
                   
                   ;; SERVERS
                   ((start-server)                    (apply server:kind-run params))
                   ((kill-server)                     (set! *server-run* #f))

                   ;; TESTS

                   ;;((test-set-state-status-by-id)     (apply mt:test-set-state-status-by-id dbstruct params))
                   ;;BB - commented out above because it was calling below, eventually, incorrectly (dbstruct passed to mt:test-set-state-status-by-id, which previosly did more, but now only passes thru to db:set-state-status-and-roll-up-items.
                   ((test-set-state-status-by-id)

;; launching and hosts
(defstruct host
  (reachable    #f)
  (last-update  0)
  (last-used    0)
  (last-cpuload 1))







;; GLOBALS

;; CONTEXTS
#;(defstruct cxt
  (taskdb #f)

;;======================================================================

(declare (unit dbmod))
(declare (uses commonmod))
(declare (uses keysmod))
(declare (uses tasksmod))
(declare (uses odsmod))
(declare (uses testsmod))
(declare (uses mtargs))
(declare (uses mtconfigf))
;; (declare (uses servermod))

(module dbmod
	*
	

	call-with-environment-variables)

(import commonmod)
(import keysmod)
(import files)
(import tasksmod)
(import odsmod)
(import testsmod)
(import (prefix mtargs args:))
(import (prefix mtconfigf configf:))
;; (import servermod)

;; (use (prefix ulex ulex:))

(include "common_records.scm")

;; record for keeping state,status and count for doing roll-ups in
;; iterated tests
;;
(defstruct dbr:counts
  (state #f)
  (status #f)
  (count  0)) 












































































































































































































































;;======================================================================
;; SQLITE3 HELPERS
;;======================================================================

(define (db:general-sqlite-error-dump exn stmt . params)
  (let ((err-status ((condition-property-accessor 'sqlite3 'status #f) exn))) ;; RADT ... how does this work?

  node [shape=box,style=filled,fontname="clear",fontsize="10"];

  subgraph cluster_megatest {
    label="megatest";

    rmtmod [label="rmt mod"];

    // commonmod
    mtargs     -> commonmod;
    pkts       -> commonmod;
    mtconfigf  -> commonmod;
    processmod -> commonmod;
    stml2      -> commonmod;
    // apimod

    ulex      -> rmtmod;
    itemsmod  -> rmtmod;
    // dbmod
    commonmod -> dbmod;
    keysmod   -> dbmod;
    tasksmod  -> dbmod;
    odsmod    -> dbmod;
    testsmod  -> dbmod;
    mtargs    -> dbmod;
    mtconfigf -> dbmod;
    // tasksmod
    commonmod -> tasksmod;
    pgdbmod   -> tasksmod;
    mtconfigf -> tasksmod;
    // odsmod
    commonmod -> odsmod;
    // testsmod
    commonmod -> testsmod;
    mtargs    -> testsmod;




    // mtmod
    mtconfigf -> mtmod;
    commonmod -> mtmod;
    rmtmod    -> mtmod;




    // runsmod
    mtmod     -> runsmod;
    commonmod -> runsmod;
    testsmod  -> runsmod;
    // pgdbmod
    commonmod -> pgdbmod;
    mtconfigf -> pgdbmod;
    //servermod
    commonmod -> servermod;
    dbmod     -> servermod;



    // processmod








}
  
}

;; Copyright 2006-2012, Matthew Welland.
;; 
;; This file is part of Megatest.
;; 
;;     Megatest is free software: you can redistribute it and/or modify
;;     it under the terms of the GNU General Public License as published by
;;     the Free Software Foundation, either version 3 of the License, or
;;     (at your option) any later version.
;; 
;;     Megatest is distributed in the hope that it will be useful,
;;     but WITHOUT ANY WARRANTY; without even the implied warranty of
;;     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
;;     GNU General Public License for more details.
;; 
;;     You should have received a copy of the GNU General Public License
;;     along with Megatest.  If not, see <http://www.gnu.org/licenses/>.

(define (http-transport:make-server-url hostport)
  (if (not hostport)
      #f
      (conc "http://" (car hostport) ":" (cadr hostport))))

(define *server-loop-heart-beat* (current-seconds))

;;======================================================================
;; S E R V E R
;; ======================================================================

;; Call this to start the actual server
;;

(define *db:process-queue-mutex* (make-mutex))

(define (http-transport:run hostn)
  ;; Configurations for server
  (tcp-buffer-size 2048)
  (max-connections 2048) 
  (debug:print 2 *default-log-port* "Attempting to start the server ...")
  (let* ((db              #f) ;;        (open-db)) ;; we don't want the server to be opening and closing the db unnecesarily
	 (hostname        (get-host-name))
	 (ipaddrstr       (let ((ipstr (if (string=? "-" hostn)
					   ;; (string-intersperse (map number->string (u8vector->list (hostname->ip hostname))) ".")
					   (server:get-best-guess-address hostname)
					   #f)))
			    (if ipstr ipstr hostn))) ;; hostname))) 
	 (start-port      (portlogger:open-run-close portlogger:find-port))
	 (link-tree-path  (common:get-linktree))
	 (tmp-area        (common:get-db-tmp-area))
	 (start-file      (conc tmp-area "/.server-start")))
    (debug:print-info 0 *default-log-port* "portlogger recommended port: " start-port)
    ;; set some parameters for the server
    (root-path     (if link-tree-path 
		       link-tree-path
		       (current-directory))) ;; WARNING: SECURITY HOLE. FIX ASAP!
    (handle-directory spiffy-directory-listing)
    (handle-exception (lambda (exn chain)
			(signal (make-composite-condition
				 (make-property-condition 
				  'server
				  'message "server error")))))

    ;; http-transport:handle-directory) ;; simple-directory-handler)
    ;; Setup the web server and a /ctrl interface
    ;;
    (vhost-map `(((* any) . ,(lambda (continue)
			       ;; open the db on the first call 
				 ;; This is were we set up the database connections
			       (let* (($   (request-vars source: 'both))
				      (dat ($ 'dat))
				      (res #f))
				 (cond
				  ((equal? (uri-path (request-uri (current-request)))
					   '(/ "api"))
				   (send-response body:    (api:process-request *dbstruct-db* $) ;; the $ is the request vars proc
						  headers: '((content-type text/plain)))
				   (mutex-lock! *heartbeat-mutex*)
				   (set! *db-last-access* (current-seconds))
				   (mutex-unlock! *heartbeat-mutex*))
				  ((equal? (uri-path (request-uri (current-request))) 
					   '(/ ""))
				   (send-response body: (http-transport:main-page)))
				  ((equal? (uri-path (request-uri (current-request))) 
					   '(/ "json_api"))
				   (send-response body: (http-transport:main-page)))
				  ((equal? (uri-path (request-uri (current-request))) 
					   '(/ "runs"))
				   (send-response body: (http-transport:main-page)))
				  ((equal? (uri-path (request-uri (current-request))) 
					   '(/ any))
				   (send-response body: "hey there!\n"
						  headers: '((content-type text/plain))))
				  ((equal? (uri-path (request-uri (current-request))) 
					   '(/ "hey"))
				   (send-response body: "hey there!\n" 
						  headers: '((content-type text/plain))))
                                  ((equal? (uri-path (request-uri (current-request))) 
					   '(/ "jquery3.1.0.js"))
				   (send-response body: (http-transport:show-jquery) 
						  headers: '((content-type application/javascript))))
                                  ((equal? (uri-path (request-uri (current-request))) 
					   '(/ "test_log"))
				   (send-response body: (http-transport:html-test-log $) 
						  headers: '((content-type text/HTML))))    
                                  ((equal? (uri-path (request-uri (current-request))) 
					   '(/ "dashboard"))
				   (send-response body: (http-transport:html-dboard $) 
						  headers: '((content-type text/HTML)))) 
				  (else (continue))))))))
    (with-output-to-file start-file (lambda ()(print (current-process-id))))
    (http-transport:try-start-server ipaddrstr start-port)))

;; This is recursively run by http-transport:run until sucessful
;;
(define (http-transport:try-start-server ipaddrstr portnum)
  (let ((config-hostname (configf:lookup *configdat* "server" "hostname"))
	(config-use-proxy (equal? (configf:lookup *configdat* "client" "use-http_proxy") "yes")))
    (if (not config-use-proxy)
	(determine-proxy (constantly #f)))
    (debug:print-info 0 *default-log-port* "http-transport:try-start-server time=" (seconds->time-string (current-seconds)) " ipaddrsstr=" ipaddrstr " portnum=" portnum " config-hostname=" config-hostname)
    (handle-exceptions
	exn
	(begin
	  (print-error-message exn)
	  (if (< portnum 64000)
	      (begin 
		(debug:print 0 *default-log-port* "WARNING: attempt to start server failed. Trying again ...")
		(debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
		(debug:print 5 *default-log-port* "exn=" (condition->list exn))
		(portlogger:open-run-close portlogger:set-failed portnum)
		(debug:print 0 *default-log-port* "WARNING: failed to start on portnum: " portnum ", trying next port")
		(thread-sleep! 0.1)
		
		;; get_next_port goes here
		(http-transport:try-start-server ipaddrstr
						 (portlogger:open-run-close portlogger:find-port)))
	      (begin
		(print "ERROR: Tried and tried but could not start the server"))))
      ;; any error in following steps will result in a retry
      (set! *server-info* (list ipaddrstr portnum))
      (debug:print 0 *default-log-port* "INFO: Trying to start server on " ipaddrstr ":" portnum)
      ;; This starts the spiffy server
      ;; NEED WAY TO SET IP TO #f TO BIND ALL
      ;; (start-server bind-address: ipaddrstr port: portnum)
      (if config-hostname ;; this is a hint to bind directly
	  (start-server port: portnum bind-address: (if (equal? config-hostname "-")
							ipaddrstr
							config-hostname))
	  (start-server port: portnum))
      (portlogger:open-run-close portlogger:set-port portnum "released")
      (debug:print 1 *default-log-port* "INFO: server has been stopped"))))

;;======================================================================
;; S E R V E R   U T I L I T I E S 
;;======================================================================

;;======================================================================
;; C L I E N T S
;;======================================================================

(define *http-mutex* (make-mutex))

;; NOTE: Large block of code from 32436b426188080f72fceb6894af541fbad9921e removed here
;;       I'm pretty sure it is defunct.

;; This next block all imported en-mass from the api branch
(define *http-requests-in-progress* 0)
(define *http-connections-next-cleanup* (current-seconds))

(define (http-transport:get-time-to-cleanup)
  (let ((res #f))
    (mutex-lock! *http-mutex*)
    (set! res (> (current-seconds) *http-connections-next-cleanup*))
    (mutex-unlock! *http-mutex*)
    res))

(define (http-transport:inc-requests-count)
  (mutex-lock! *http-mutex*)
  (set! *http-requests-in-progress* (+ 1 *http-requests-in-progress*))
  ;; Use this opportunity to slow things down iff there are too many requests in flight
  (if (> *http-requests-in-progress* 5)
      (begin
	(debug:print-info 0 *default-log-port* "Whoa there buddy, ease up...")
	(thread-sleep! 1)))
  (mutex-unlock! *http-mutex*))

(define (http-transport:dec-requests-count proc) 
  (mutex-lock! *http-mutex*)
  (proc)
  (set! *http-requests-in-progress* (- *http-requests-in-progress* 1))
  (mutex-unlock! *http-mutex*))

(define (http-transport:dec-requests-count-and-close-all-connections)
  (set! *http-requests-in-progress* (- *http-requests-in-progress* 1))
  (let loop ((etime (+ (current-seconds) 5))) ;; give up in five seconds
    (if (> *http-requests-in-progress* 0)
	(if (> etime (current-seconds))
	    (begin
	      (thread-sleep! 0.05)
	      (loop etime))
	    (debug:print-error 0 *default-log-port* "requests still in progress after 5 seconds of waiting. I'm going to pass on cleaning up http connections"))
	(close-all-connections!)))
  (set! *http-connections-next-cleanup* (+ (current-seconds) 10))
  (mutex-unlock! *http-mutex*))

(define (http-transport:inc-requests-and-prep-to-close-all-connections)
  (mutex-lock! *http-mutex*)
  (set! *http-requests-in-progress* (+ 1 *http-requests-in-progress*)))

;; Send "cmd" with json payload "params" to serverdat and receive result
;;
(define (http-transport:client-api-send-receive run-id serverdat cmd params #!key (numretries 3)(area-dat #f))
  (let* ((fullurl    (if (vector? serverdat)
			 (http-transport:server-dat-get-api-req serverdat)
			 (begin
			   (debug:print 0 *default-log-port* "FATAL ERROR: http-transport:client-api-send-receive called with no server info")
			   (exit 1))))
	 (res        (vector #f "uninitialized"))
	 (success    #t)
	 (sparams    (db:obj->string params transport: 'http))
	 (runremote  (or area-dat *runremote*)))
       (debug:print-info 11 *default-log-port* "fullurl=" fullurl ", cmd=" cmd ", params=" params ", run-id=" run-id "\n")
       ;; set up the http-client here
       (max-retry-attempts 1)
       ;; consider all requests indempotent
       (retry-request? (lambda (request)
			 #f))
       ;; send the data and get the response
       ;; extract the needed info from the http data and 
       ;; process and return it.
       (let* ((send-recieve (lambda ()
			      (mutex-lock! *http-mutex*)
			      ;; (condition-case (with-input-from-request "http://localhost"; #f read-lines)
			      ;;					       ((exn http client-error) e (print e)))
			      (set! res (vector                ;;; DON'T FORGET - THIS IS THE CLIENT SIDE! NOTE: consider moving this to client.scm since we are only supporting http transport at this time.
					 success
					 (db:string->obj 
					  (handle-exceptions
					      exn
					      (let ((call-chain (get-call-chain))
						    (msg        ((condition-property-accessor 'exn 'message) exn)))
						(set! success #f)
                                                (if (debug:debug-mode 1)
                                                    (debug:print-info 0 *default-log-port* "couldn't talk to server, trying again ...")
                                                    (begin
                                                      (debug:print 0 *default-log-port* "WARNING: failure in with-input-from-request to " fullurl ".")
                                                      (debug:print 0 *default-log-port* " message: " msg)
                                                      (debug:print 0 *default-log-port* " cmd: " cmd " params: " params)
                                                      (debug:print 0 *default-log-port* " call-chain: " call-chain)))
                                                (if runremote
						    (remote-conndat-set! runremote #f))
						;; Killing associated server to allow clean retry.")
						;; (tasks:kill-server-run-id run-id)  ;; better to kill the server in the logic that called this routine?
						(mutex-unlock! *http-mutex*)
					     ;;; (signal (make-composite-condition
					     ;;;          (make-property-condition 'commfail 'message "failed to connect to server")))
					     ;;; "communications failed"
						(db:obj->string #f))
					    (with-input-from-request ;; was dat
					     fullurl 
					     (list (cons 'key (or *server-id* "thekey"))
						   (cons 'cmd cmd)
						   (cons 'params sparams))
					     read-string))
					  transport: 'http)
                                         0)) ;; added this speculatively
			      ;; Shouldn't this be a call to the managed call-all-connections stuff above?
			      (close-all-connections!)
			      (mutex-unlock! *http-mutex*)
			      ))
	      (time-out     (lambda ()
			      (thread-sleep! 45)
			      #f))
	      (th1 (make-thread send-recieve "with-input-from-request"))
	      (th2 (make-thread time-out     "time out")))
	 (thread-start! th1)
	 (thread-start! th2)
	 (thread-join! th1)
	 (thread-terminate! th2)
	 (debug:print-info 11 *default-log-port* "got res=" res)
	 (if (vector? res)
	     (if (vector-ref res 0) ;; this is the first flag or the second flag?
		 res ;; this is the *inner* vector? seriously? why?
                 (if (debug:debug-mode 11)
                     (let ((call-chain (get-call-chain))) ;; note: this code also called in nmsg-transport - consider consolidating it
                       (print-call-chain (current-error-port))
                       (debug:print-error 11 *default-log-port* "error above occured at server, res=" res) ;; " message: " ((condition-property-accessor 'exn 'message) exn))
                       (debug:print 11 *default-log-port* " server call chain:")
                       (pp (vector-ref res 1) (current-error-port))
                       (signal (vector-ref res 0)))
                     res))
	     (signal (make-composite-condition
		      (make-property-condition 
		       'timeout
		       'message "nmsg-transport:client-api-send-receive-raw timed out talking to server")))))))

;; careful closing of connections stored in *runremote*
;;
(define (http-transport:close-connections #!key (area-dat #f))
  (let* ((runremote  (or area-dat *runremote*))
	 (server-dat (if runremote
                         (remote-conndat runremote)
                         #f))) ;; (hash-table-ref/default *runremote* run-id #f)))
    (if (vector? server-dat)
	(let ((api-dat (http-transport:server-dat-get-api-uri server-dat)))
	  (handle-exceptions
	    exn
	    (begin
	      (print-call-chain *default-log-port*)
	      (debug:print-error 0 *default-log-port* " closing connection failed with error: " ((condition-property-accessor 'exn 'message) exn)))
	    (close-connection! api-dat)
            ;;(close-idle-connections!)
	    #t))
	#f)))

;; Moved to commonmod
;;
;; (define (make-http-transport:server-dat)(make-vector 6))
;; (define (http-transport:server-dat-get-iface         vec)    (vector-ref  vec 0))
;; (define (http-transport:server-dat-get-port          vec)    (vector-ref  vec 1))
;; (define (http-transport:server-dat-get-api-uri       vec)    (vector-ref  vec 2))
;; (define (http-transport:server-dat-get-api-url       vec)    (vector-ref  vec 3))
;; (define (http-transport:server-dat-get-api-req       vec)    (vector-ref  vec 4))
;; (define (http-transport:server-dat-get-last-access   vec)    (vector-ref  vec 5))
;; (define (http-transport:server-dat-get-socket        vec)    (vector-ref  vec 6))
;; 
;; (define (http-transport:server-dat-make-url vec)
;;   (if (and (http-transport:server-dat-get-iface vec)
;; 	   (http-transport:server-dat-get-port  vec))
;;       (conc "http://" 
;; 	    (http-transport:server-dat-get-iface vec)
;; 	    ":"
;; 	    (http-transport:server-dat-get-port  vec))
;;       #f))
;; 
;; (define (http-transport:server-dat-update-last-access vec)
;;   (if (vector? vec)
;;       (vector-set! vec 5 (current-seconds))
;;       (begin
;; 	(print-call-chain (current-error-port))
;; 	(debug:print-error 0 *default-log-port* "call to http-transport:server-dat-update-last-access with non-vector!!"))))

;;
;; connect
;;
(define (http-transport:client-connect iface port)
  (let* ((api-url      (conc "http://" iface ":" port "/api"))
	 (api-uri      (uri-reference (conc "http://" iface ":" port "/api")))
	 (api-req      (make-request method: 'POST uri: api-uri))
	 (server-dat   (vector iface port api-uri api-url api-req (current-seconds))))
    server-dat))

;; run http-transport:keep-running in a parallel thread to monitor that the db is being 
;; used and to shutdown after sometime if it is not.
;;
(define (http-transport:keep-running) 
  ;; if none running or if > 20 seconds since 
  ;; server last used then start shutdown
  ;; This thread waits for the server to come alive
  (debug:print-info 0 *default-log-port* "Starting the sync-back, keep alive thread in server")
  (let* ((sdat              #f)
	 (tmp-area          (common:get-db-tmp-area))
	 (started-file      (conc tmp-area "/.server-started"))
	 (server-start-time (current-seconds))
	 (server-info (let loop ((start-time (current-seconds))
				 (changed    #t)
				 (last-sdat  "not this"))
                        (begin ;; let ((sdat #f))
			  (thread-sleep! 0.01)
			  (debug:print-info 0 *default-log-port* "Waiting for server alive signature")
                          (mutex-lock! *heartbeat-mutex*)
                          (set! sdat *server-info*)
                          (mutex-unlock! *heartbeat-mutex*)
                          (if (and sdat
				   (not changed)
				   (> (- (current-seconds) start-time) 2))
			      (begin
				(debug:print-info 0 *default-log-port* "Received server alive signature")
                                (common:save-pkt `((action . alive)
                                                   (T      . server)
                                                   (pid    . ,(current-process-id))
                                                   (ipaddr . ,(car sdat))
                                                   (port   . ,(cadr sdat)))
                                                 *configdat* #t)
				sdat)
                              (begin
				(debug:print-info 0 *default-log-port* "Still waiting, last-sdat=" last-sdat)
                                (sleep 4)
				(if (> (- (current-seconds) start-time) 120) ;; been waiting for two minutes
				    (begin
				      (debug:print-error 0 *default-log-port* "transport appears to have died, exiting server")
                                      (common:save-pkt `((action . died)
                                                         (T      . server)
                                                         (pid    . ,(current-process-id))
                                                         (ipaddr . ,(car sdat))
                                                         (port   . ,(cadr sdat))
                                                         (msg    . "Transport died?"))
						       *configdat* #t)
				      (exit))
				    (loop start-time
					  (equal? sdat last-sdat)
					  sdat)))))))
	 (iface       (car server-info))
         (port        (cadr server-info))
         (last-access 0)
	 (server-timeout (server:expiration-timeout))
	 (server-going  #f)
	 (server-log-file (args:get-arg "-log"))) ;; always set when we are a server

    (with-output-to-file started-file (lambda ()(print (current-process-id))))

    (let loop ((count         0)
	       (server-state 'available)
	       (bad-sync-count 0)
	       (start-time     (current-milliseconds)))
      ;; Use this opportunity to sync the tmp db to megatest.db
      (if (not server-going) ;; *dbstruct-db* 
	  (begin
	    (debug:print 0 *default-log-port* "SERVER: dbprep")
	    (set! *dbstruct-db*  (db:setup #t)) ;;  run-id))
	    (set! server-going #t)
	    (debug:print 0 *default-log-port* "SERVER: running, megatest version: " (common:get-full-version)) ;; NOTE: the server is NOT yet marked as running in the log. We do that in the keep-running routine.
	    (thread-start! *watchdog*)))
      
      ;; when things go wrong we don't want to be doing the various queries too often
      ;; so we strive to run this stuff only every four seconds or so.
      (let* ((sync-time (- (current-milliseconds) start-time))
	    (rem-time  (quotient (- 4000 sync-time) 1000)))
	(if (and (<= rem-time 4)
		 (>  rem-time 0))
	    (thread-sleep! rem-time)))
      
      (if (< count 1) ;; 3x3 = 9 secs aprox
	  (loop (+ count 1) 'running bad-sync-count (current-milliseconds)))
      
      ;; Check that iface and port have not changed (can happen if server port collides)
      (mutex-lock! *heartbeat-mutex*)
      (set! sdat *server-info*)
      (mutex-unlock! *heartbeat-mutex*)
      
      (if (not (equal? sdat (list iface port)))
	  (let ((new-iface (car sdat))
		(new-port  (cadr sdat)))
	    (debug:print-info 0 *default-log-port* "WARNING: interface changed, refreshing iface and port info")
	    (set! iface new-iface)
	    (set! port  new-port)
	    (debug:print 0 *default-log-port* "SERVER STARTED: " iface ":" port " AT " (current-seconds))
	    (flush-output *default-log-port*)))
      
      ;; Transfer *db-last-access* to last-access to use in checking that we are still alive
      (mutex-lock! *heartbeat-mutex*)
      (set! last-access *db-last-access*)
      (mutex-unlock! *heartbeat-mutex*)
      
      (if (common:low-noise-print 120 (conc "server running on " iface ":" port))
	  (begin
	    (debug:print 0 *default-log-port* "SERVER STARTED: " iface ":" port " AT " (current-seconds))
	    (flush-output *default-log-port*)))
      (if (common:low-noise-print 60 "dbstats")
	  (begin
	    (debug:print 0 *default-log-port* "Server stats:")
	    (db:print-current-query-stats)))
      (let* ((hrs-since-start  (/ (- (current-seconds) server-start-time) 3600)))
	(cond
         ((and *server-run*
	       (> (+ last-access server-timeout)
		  (current-seconds)))
          (if (common:low-noise-print 120 "server continuing")
              (debug:print-info 0 *default-log-port* "Server continuing, seconds since last db access: " (- (current-seconds) last-access))
	      (let ((curr-time (current-seconds)))
		(handle-exceptions
		    exn
		    (debug:print 0 *default-log-port* "ERROR: Failed to change timestamp on log file " server-log-file ". Are you out of space on that disk?")
		  (if (not *server-overloaded*)
		      (change-file-times server-log-file curr-time curr-time)))))
          (loop 0 server-state bad-sync-count (current-milliseconds)))
         (else
          (debug:print-info 0 *default-log-port* "Server timed out. seconds since last db access: " (- (current-seconds) last-access))
          (http-transport:server-shutdown port)))))))

(define (http-transport:server-shutdown port)
  (begin
    ;;(BB> "http-transport:server-shutdown called")
    (debug:print-info 0 *default-log-port* "Starting to shutdown the server. pid="(current-process-id))
    ;;
    ;; start_shutdown
    ;;
    (set! *time-to-exit* #t) ;; tell on-exit to be fast as we've already cleaned up
    (portlogger:open-run-close portlogger:set-port port "released")
    (thread-sleep! 1)

    ;; (debug:print-info 0 *default-log-port* "Max cached queries was    " *max-cache-size*)
    ;; (debug:print-info 0 *default-log-port* "Number of cached writes   " *number-of-writes*)
    ;; (debug:print-info 0 *default-log-port* "Average cached write time "
    ;; 		      (if (eq? *number-of-writes* 0)
    ;; 			  "n/a (no writes)"
    ;; 			  (/ *writes-total-delay*
    ;; 			     *number-of-writes*))
    ;; 		      " ms")
    ;; (debug:print-info 0 *default-log-port* "Number non-cached queries "  *number-non-write-queries*)
    ;; (debug:print-info 0 *default-log-port* "Average non-cached time   "
    ;; 		      (if (eq? *number-non-write-queries* 0)
    ;; 			  "n/a (no queries)"
    ;; 			  (/ *total-non-write-delay* 
    ;; 			     *number-non-write-queries*))
    ;; 		      " ms")
    
    (db:print-current-query-stats)
    (common:save-pkt `((action . exit)
                       (T      . server)
                       (pid    . ,(current-process-id)))
                     *configdat* #t)
    (debug:print-info 0 *default-log-port* "Server shutdown complete. Exiting")
    (exit)))

;; all routes though here end in exit ...
;;
;; start_server? 
;;
(define (http-transport:launch)
  ;; check that a server start is in progress, pause or exit if so
  (let* ((tmp-area            (common:get-db-tmp-area))
	 (server-start        (conc tmp-area "/.server-start"))
	 (server-started      (conc tmp-area "/.server-started"))
	 (start-time          (common:lazy-modification-time server-start))
	 (started-time        (common:lazy-modification-time server-started))
	 (server-starting     (< start-time started-time)) ;; if start-time is less than started-time then a server is still starting
	 (start-time-old      (> (- (current-seconds) start-time) 5))
         (cleanup-proc        (lambda (msg)
                                (let* ((serv-fname      (conc "server-" (current-process-id) "-" (get-host-name) ".log"))
                                       (full-serv-fname (conc *toppath* "/logs/" serv-fname))
                                       (new-serv-fname  (conc *toppath* "/logs/" "defunct-" serv-fname)))
                                  (debug:print 0 *default-log-port* msg)
                                  (if (common:file-exists? full-serv-fname)
                                      (system (conc "sleep 1;mv -f " full-serv-fname " " new-serv-fname))
                                      (debug:print 0 *default-log-port* "INFO: cannot move " full-serv-fname " to " new-serv-fname))
                                  (exit)))))
    (if (and (not start-time-old) ;; last server start try was less than five seconds ago
	     (not server-starting))
	(begin
	  (cleanup-proc "NOT starting server, there is either a recently started server or a server in process of starting")
	  (exit)))
    ;; lets not even bother to start if there are already three or more server files ready to go
    (let* ((num-alive   (server:get-num-alive (server:get-list *toppath*))))
      (if (> num-alive 3)
          (begin
            (cleanup-proc (conc "ERROR: Aborting server start because there are already " num-alive " possible servers either running or starting up"))
            (exit))))
  (common:save-pkt `((action . start)
		     (T      . server)
		     (pid    . ,(current-process-id)))
		   *configdat* #t)
    (let* ((th2 (make-thread (lambda ()
                               (debug:print-info 0 *default-log-port* "Server run thread started")
                               (http-transport:run 
                                (if (args:get-arg "-server")
                                    (args:get-arg "-server")
                                    "-")
                                )) "Server run"))
           (th3 (make-thread (lambda ()
                               (debug:print-info 0 *default-log-port* "Server monitor thread started")
                               (http-transport:keep-running)
                               "Keep running"))))
      (thread-start! th2)
      (thread-sleep! 0.25) ;; give the server time to settle before starting the keep-running monitor.
      (thread-start! th3)
      (set! *didsomething* #t)
      (thread-join! th2)
      (exit))))

;; (define (http-transport:server-signal-handler signum)
;;   (signal-mask! signum)
;;   (handle-exceptions
;;    exn
;;    (debug:print 0 *default-log-port* " ... exiting ...")
;;    (let ((th1 (make-thread (lambda ()
;; 			     (thread-sleep! 1))
;; 			   "eat response"))
;; 	 (th2 (make-thread (lambda ()
;; 			     (debug:print-error 0 *default-log-port* "Received ^C, attempting clean exit. Please be patient and wait a few seconds before hitting ^C again.")
;; 			     (thread-sleep! 3) ;; give the flush three seconds to do it's stuff
;; 			     (debug:print 0 *default-log-port* "       Done.")
;; 			     (exit 4))
;; 			   "exit on ^C timer")))
;;      (thread-start! th2)
;;      (thread-start! th1)
;;      (thread-join! th2))))

;;===============================================
;; Java script
;;===============================================
(define (http-transport:show-jquery)
  (let* ((data  (tests:readlines *java-script-lib*)))
(string-join data "\n")))



;;======================================================================
;; web pages
;;======================================================================

(define (http-transport:html-test-log $)
   (let* ((run-id ($ 'runid))
         (test-item ($ 'testname))
         (parts (string-split test-item ":"))
         (test-name (car parts))
             
         (item-name (if (equal? (length parts) 1)
             ""
             (cadr parts))))
  ;(print $) 
(tests:get-test-log run-id test-name item-name)))


(define (http-transport:html-dboard $)
  (let* ((page ($ 'page))
         (oup       (open-output-string)) 
         (bdy "--------------------------")

         (ret  (tests:dynamic-dboard page)))
    (s:output-new  oup  ret)
   (close-output-port oup)

  (set! bdy   (get-output-string oup))
     (conc "<h1>Dashboard</h1>" bdy "<br/> <br/> "  )))

(define (http-transport:main-page)
  (let ((linkpath (root-path)))
    (conc "<head><h1>" (pathname-strip-directory *toppath*) "</h1></head>"
	  "<body>"
	  "Run area: " *toppath*
	  "<h2>Server Stats</h2>"
	  (http-transport:stats-table) 
	  "<hr>"
	  (http-transport:runs linkpath)
	  "<hr>"
	  ;; (http-transport:run-stats)
	  "</body>"
	  )))

(define (http-transport:stats-table)
  (mutex-lock! *heartbeat-mutex*)
  (let ((res 
	 (conc "<table>"
	       ;; "<tr><td>Max cached queries</td>        <td>" *max-cache-size* "</td></tr>"
	       "<tr><td>Number of cached writes</td>   <td>" *number-of-writes* "</td></tr>"
	       "<tr><td>Average cached write time</td> <td>" (if (eq? *number-of-writes* 0)
								 "n/a (no writes)"
								 (/ *writes-total-delay*
								    *number-of-writes*))
	       " ms</td></tr>"
	       "<tr><td>Number non-cached queries</td> <td>"  *number-non-write-queries* "</td></tr>"
	       ;; "<tr><td>Average non-cached time</td>   <td>" (if (eq? *number-non-write-queries* 0)
	       ;; 							 "n/a (no queries)"
	       ;; 							 (/ *total-non-write-delay* 
	       ;; 							    *number-non-write-queries*))
	       " ms</td></tr>"
	       "<tr><td>Last access</td><td>"              (seconds->time-string *db-last-access*) "</td></tr>"
	       "</table>")))
    (mutex-unlock! *heartbeat-mutex*)
    res))

(define (http-transport:runs linkpath)
  (conc "<h3>Runs</h3>"
	(string-intersperse
	 (let ((files (map pathname-strip-directory (glob (conc linkpath "/*")))))
	   (map (lambda (p)
		  (conc "<a href=\"" p "\">" p "</a><br>"))
		files))
	 " ")))

#;(define (http-transport:run-stats)
  (let ((stats (open-run-close db:get-running-stats #f)))
    (conc "<table>"
	  (string-intersperse
	   (map (lambda (stat)
		  (conc "<tr><td>" (car stat) "</td><td>" (cadr stat) "</td></tr>"))
		stats)
	   " ")
	  "</table>")))

;;     but WITHOUT ANY WARRANTY; without even the implied warranty of
;;     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
;;     GNU General Public License for more details.
;; 
;;     You should have received a copy of the GNU General Public License
;;     along with Megatest.  If not, see <http://www.gnu.org/licenses/>.

;;======================================================================
;; launch a task - this runs on the originating host, tests themselves
;;
;;======================================================================


;;======================================================================
;; ezsteps
;;======================================================================

;; ezsteps were going to be coded as
;; stepname[,predstep1,predstep2 ...] [{VAR1=first,second,third}] command to execute
;;   BUT
;; now are
;; stepname {VAR=first,second,third ...} command ...
;; where the {VAR=first,second,third ...} is optional.

;; given an exit code and whether or not logpro was used calculate OK/BAD
;; return #t if we are ok, #f otherwise
(define (steprun-good? logpro exitcode)
  (or (eq? exitcode 0)
      (and logpro (eq? exitcode 2))))

;; if handed a string, process it, else look for MT_CMDINFO
(define (launch:get-cmdinfo-assoc-list #!key (encoded-cmd #f))
  (let ((enccmd (if encoded-cmd encoded-cmd (getenv "MT_CMDINFO"))))
    (if enccmd
	(common:read-encoded-string enccmd)
	'())))

;;                       0           1              2              3
(defstruct launch:einf (pid #t)(exit-status #t)(exit-code #t)(rollup-status 0))

;; return (conc status ": " comment) from the final section so that
;;   the comment can be set in the step record in launch.scm
;;
(define (launch:load-logpro-dat run-id test-id stepname)
  (let ((cname (conc stepname ".dat")))
    (if (common:file-exists? cname)
	(let* ((dat  (read-config cname #f #f))
	       (csvr (db:logpro-dat->csv dat stepname))
	       (csvt (let-values (((fmt-cell fmt-record fmt-csv) (make-format ",")))
		       (fmt-csv (map list->csv-record csvr))))
	       (status (configf:lookup dat "final" "exit-status"))
	       (msg     (configf:lookup dat "final" "message")))
          (if csvt  ;; this if blocked stack dump caused by .dat file from logpro being 0-byte.  fixed by upgrading logpro
              (rmt:csv->test-data run-id test-id csvt)
	      (debug:print 0 *default-log-port* "ERROR: no csvdat exists for run-id: " run-id " test-id: " test-id " stepname: " stepname ", check that logpro version is 1.15 or newer"))
	  ;;  (debug:print-info 13 *default-log-port* "Error: run-id/test-id/stepname="run-id"/"test-id"/"stepname" => bad csvr="csvr)
	  ;;  )
	  (cond
	   ((equal? status "PASS") "PASS") ;; skip the message part if status is pass
	   (status (conc (configf:lookup dat "final" "exit-status") ": " (if msg msg "no message")))
	   (else #f)))
	#f)))

(define (launch:runstep ezstep run-id test-id exit-info m tal testconfig) ;;; TODO: deprecate me in favor of ezsteps.scm
  (let* ((stepname       (car ezstep))  ;; do stuff to run the step
	 (stepinfo       (cadr ezstep))
	;; (let ((info (cadr ezstep)))
	;; 		   (if (proc? info) "" info)))
	;; (stepproc       (let ((info (cadr ezstep)))
	;; 		   (if (proc? info) info #f)))
	 (stepparts      (string-match (regexp "^(\\{([^\\}\\{]*)\\}\\s*|)(.*)$") stepinfo))
	 (stepparams     (list-ref stepparts 2)) ;; for future use, {VAR=1,2,3}, run step for each
	 (paramparts     (if (string? stepparams)
			     (map (lambda (x)(string-split x "=")) (string-split-fields "[^;]*=[^;]*" stepparams))
			     '()))
	 (subrun         (alist-ref "subrun" paramparts equal?))
	 (stepcmd        (list-ref stepparts 3))
	 (script         "") ; "#!/bin/bash\n") ;; yep, we depend on bin/bash FIXME!!!\
	 (logpro-file    (conc stepname ".logpro"))
	 (html-file      (conc stepname ".html"))
	 (dat-file       (conc stepname ".dat"))
	 (tconfig-logpro (configf:lookup testconfig "logpro" stepname))
	 (logpro-used    (common:file-exists? logpro-file)))

    (debug:print 0 *default-log-port* "stepparts: " stepparts ", stepparams: " stepparams
                 ", paramparts: " paramparts ", subrun: " subrun ", stepcmd: " stepcmd)
    
    (if (and tconfig-logpro
	     (not logpro-used)) ;; no logpro file found but have a defn in the testconfig
	(begin
	  (with-output-to-file logpro-file
	    (lambda ()
	      (print ";; logpro file extracted from testconfig\n"
		     ";;")
	      (print tconfig-logpro)))
	  (set! logpro-used #t)))
    
    ;; NB// can safely assume we are in test-area directory
    (debug:print 4 *default-log-port* "ezsteps:\n stepname: " stepname " stepinfo: " stepinfo " stepparts: " stepparts
		 " stepparams: " stepparams " stepcmd: " stepcmd)
    
    ;; ;; first source the previous environment
    ;; (let ((prev-env (conc ".ezsteps/" prevstep (if (string-search (regexp "csh") 
    ;;      							 (get-environment-variable "SHELL")) ".csh" ".sh"))))
    ;;   (if (and prevstep (common:file-exists? prev-env))
    ;;       (set! script (conc script "source " prev-env))))
    
    ;; call the command using mt_ezstep
    ;; (set! script (conc "mt_ezstep " stepname " " (if prevstep prevstep "x") " " stepcmd))
    
    (debug:print 4 *default-log-port* "script: " script)
    (rmt:teststep-set-status! run-id test-id stepname "start" "-" #f #f)
    ;; now launch the actual process
    (call-with-environment-variables 
     (list (cons "PATH" (conc (get-environment-variable "PATH") ":.")))
     (lambda () ;; (process-run "/bin/bash" "-c" "exec ls -l /tmp/foobar > /tmp/delme-more.log 2>&1")
       (let* ((cmd (conc stepcmd " > " stepname ".log 2>&1")) ;; >outfile 2>&1 
	      (pid #f))
	 (let ((proc (lambda ()
		       (set! pid (process-run "/bin/bash" (list "-c" cmd))))))
	   (if subrun
               (begin
                 (debug:print-info 0 *default-log-port* "Running without MT_.* environment variables.")
                 (common:without-vars proc "^MT_.*"))
	       (proc)))
	 
         (with-output-to-file "Makefile.ezsteps"
           (lambda ()
             (print stepname ".log :")
             (print "\t" cmd)
             (if (common:file-exists? (conc stepname ".logpro"))
                 (print "\tlogpro " stepname ".logpro " stepname ".html < " stepname ".log"))
             (print)
             (print stepname " : " stepname ".log")
             (print))
           #:append)

	 (rmt:test-set-top-process-pid run-id test-id pid)
	 (let processloop ((i 0))
	   (let-values (((pid-val exit-status exit-code)(process-wait pid #t)))
		       (mutex-lock! m)
		       (launch:einf-pid-set!         exit-info pid)         ;; (vector-set! exit-info 0 pid)
		       (launch:einf-exit-status-set! exit-info exit-status) ;; (vector-set! exit-info 1 exit-status)
		       (launch:einf-exit-code-set!   exit-info exit-code)   ;; (vector-set! exit-info 2 exit-code)
		       (mutex-unlock! m)
		       (if (eq? pid-val 0)
			   (begin
			     (thread-sleep! 2)
			     (processloop (+ i 1))))
		       )))))
    (debug:print-info 0 *default-log-port* "step " stepname " completed with exit code " (launch:einf-exit-code exit-info)) ;; (vector-ref exit-info 2))
    ;; now run logpro if needed
    (if logpro-used
	(let* ((logpro-exe (or (getenv "LOGPRO_EXE") "logpro"))
               (pid        (process-run (conc "/bin/sh -c '"logpro-exe" "logpro-file " " (conc stepname ".html") " < " stepname ".log > /dev/null'"))))
	  (let processloop ((i 0))
	    (let-values (((pid-val exit-status exit-code)(process-wait pid #t)))
			(mutex-lock! m)
			;; (make-launch:einf pid: pid exit-status: exit-status exit-code: exit-code)
			(launch:einf-pid-set!         exit-info pid)         ;; (vector-set! exit-info 0 pid)
			(launch:einf-exit-status-set! exit-info exit-status) ;; (vector-set! exit-info 1 exit-status)
			(launch:einf-exit-code-set!   exit-info exit-code)   ;; (vector-set! exit-info 2 exit-code)
			(mutex-unlock! m)
			(if (eq? pid-val 0)
			    (begin
			      (thread-sleep! 2)
			      (processloop (+ i 1)))))
	    (debug:print-info 0 *default-log-port* "logpro for step " stepname " exited with code " (launch:einf-exit-code exit-info))))) ;; (vector-ref exit-info 2)))))
    
    (let ((exinfo (launch:einf-exit-code exit-info)) ;; (vector-ref exit-info 2))
	  (logfna (if logpro-used (conc stepname ".html") ""))
	  (comment #f))
      (if logpro-used
	  (let ((datfile (conc stepname ".dat")))
	    ;; load the .dat file into the test_data table if it exists
	    (if (common:file-exists? datfile)
		(set! comment (launch:load-logpro-dat run-id test-id stepname)))
	    (rmt:test-set-log! run-id test-id (conc stepname ".html"))))
      (rmt:teststep-set-status! run-id test-id stepname "end" exinfo comment logfna))
    ;; set the test final status
    (let* ((process-exit-status (launch:einf-exit-code exit-info)) ;; (vector-ref exit-info 2))
	   (this-step-status (cond
			      ((and (eq? process-exit-status 2) logpro-used) 'warn)   ;; logpro 2 = warnings
			      ((and (eq? process-exit-status 3) logpro-used) 'check)  ;; logpro 3 = check
			      ((and (eq? process-exit-status 4) logpro-used) 'waived) ;; logpro 4 = waived
			      ((and (eq? process-exit-status 5) logpro-used) 'abort)  ;; logpro 5 = abort
			      ((and (eq? process-exit-status 6) logpro-used) 'skip)   ;; logpro 6 = skip
			      ((eq? process-exit-status 0)                   'pass)   ;; logpro 0 = pass
			      (else 'fail)))
	   (overall-status   (cond
			      ((eq? (launch:einf-rollup-status exit-info) 2) 'warn) ;; rollup-status (vector-ref exit-info 3)
			      ((eq? (launch:einf-rollup-status exit-info) 0) 'pass) ;; (vector-ref exit-info 3)
			      (else 'fail)))
	   (next-status      (cond 
			      ((eq? overall-status 'pass) this-step-status)
			      ((eq? overall-status 'warn)
			       (if (eq? this-step-status 'fail) 'fail 'warn))
			      ((eq? overall-status 'abort) 'abort)
			      (else 'fail)))
	   (next-state       ;; "RUNNING") ;; WHY WAS THIS CHANGED TO NOT USE (null? tal) ??
	    (cond
	     ((null? tal) ;; more to run?
	      "COMPLETED")
	     (else "RUNNING"))))
      (debug:print 4 *default-log-port* "Exit value received: " (launch:einf-exit-code exit-info) " logpro-used: " logpro-used 
		   " this-step-status: " this-step-status " overall-status: " overall-status 
		   " next-status: " next-status " rollup-status: "  (launch:einf-rollup-status exit-info)) ;; (vector-ref exit-info 3))
      (case next-status
	((warn)
	 (launch:einf-rollup-status-set! exit-info 2) ;; (vector-set! exit-info 3 2) ;; rollup-status
	 ;; NB// test-set-status! does rdb calls under the hood
	 (tests:test-set-status! run-id test-id next-state "WARN" 
				 (if (eq? this-step-status 'warn) "Logpro warning found" #f)
				 #f))
	((check)
	 (launch:einf-rollup-status-set! exit-info 3) ;; (vector-set! exit-info 3 3) ;; rollup-status
	 ;; NB// test-set-status! does rdb calls under the hood
	 (tests:test-set-status! run-id test-id next-state "CHECK" 
				 (if (eq? this-step-status 'check) "Logpro check found" #f)
				 #f))
	((waived)
	 (launch:einf-rollup-status-set! exit-info 4) ;; (vector-set! exit-info 3 3) ;; rollup-status
	 ;; NB// test-set-status! does rdb calls under the hood
	 (tests:test-set-status! run-id test-id next-state "WAIVED" 
				 (if (eq? this-step-status 'check) "Logpro waived found" #f)
				 #f))
	((abort)
	 (launch:einf-rollup-status-set! exit-info 5) ;; (vector-set! exit-info 3 4) ;; rollup-status
	 ;; NB// test-set-status! does rdb calls under the hood
	 (tests:test-set-status! run-id test-id next-state "ABORT" 
				 (if (eq? this-step-status 'abort) "Logpro abort found" #f)
				 #f))
	((skip)
	 (launch:einf-rollup-status-set! exit-info 6) ;; (vector-set! exit-info 3 4) ;; rollup-status
	 ;; NB// test-set-status! does rdb calls under the hood
	 (tests:test-set-status! run-id test-id next-state "SKIP" 
				 (if (eq? this-step-status 'skip) "Logpro skip found" #f)
				 #f))
	((pass)
	 (tests:test-set-status! run-id test-id next-state "PASS" #f #f))
	(else ;; 'fail
	 (launch:einf-rollup-status-set! exit-info 1) ;; (vector-set! exit-info 3 1) ;; force fail, this used to be next-state but that doesn't make sense. should always be "COMPLETED" 
	 (tests:test-set-status! run-id test-id "COMPLETED" "FAIL" (conc "Failed at step " stepname) #f)
	 )))
    logpro-used))

(define (launch:manage-steps run-id test-id item-path fullrunscript ezsteps subrun test-name tconfigreg exit-info m)
  ;; (let-values
  ;;  (((pid exit-status exit-code)
  ;;    (run-n-wait fullrunscript)))
  ;; (tests:test-set-status! test-id "RUNNING" "n/a" #f #f)
  ;; Since we should have a clean slate at this time there is no need to do 
  ;; any of the other stuff that tests:test-set-status! does. Let's just 
  ;; force RUNNING/n/a

  ;; (thread-sleep! 0.3)
  ;; (tests:test-force-state-status! run-id test-id "RUNNING" "n/a")
  (rmt:set-state-status-and-roll-up-items run-id test-name item-path "RUNNING" #f #f) 
  ;; (thread-sleep! 0.3) ;; NFS slowness has caused grief here

  ;; if there is a runscript do it first
  (if fullrunscript
      (let ((pid (process-run fullrunscript)))
	(rmt:test-set-top-process-pid run-id test-id pid)
	(let loop ((i 0))
	  (let-values
	   (((pid-val exit-status exit-code) (process-wait pid #t)))
	   (mutex-lock! m)
	   (launch:einf-pid-set!           exit-info  pid)         ;; (vector-set! exit-info 0 pid)
	   (launch:einf-exit-status-set!   exit-info  exit-status) ;; (vector-set! exit-info 1 exit-status)
	   (launch:einf-exit-code-set!     exit-info  exit-code)   ;; (vector-set! exit-info 2 exit-code)
	   (launch:einf-rollup-status-set! exit-info  exit-code)   ;; (vector-set! exit-info 3 exit-code)  ;; rollup status
	   (mutex-unlock! m)
	   (if (eq? pid-val 0)
	       (begin
		 (thread-sleep! 2)
		 (loop (+ i 1)))
	       )))))
  ;; then, if runscript ran ok (or did not get called)
  ;; do all the ezsteps (if any)
  (if (or ezsteps subrun)
      (let* ((test-run-dir (tests:get-test-path-from-environment))
             (testconfig ;; (read-config (conc work-area "/testconfig") #f #t environ-patt: "pre-launch-env-vars")) ;; FIXME??? is allow-system ok here?
	      ;; NOTE: it is tempting to turn off force-create of testconfig but dynamic
	      ;;       ezstep names need a full re-eval here.
	      (tests:get-testconfig test-name item-path tconfigreg #t force-create: #t)) ;; 'return-procs)))
	     (ezstepslst (if (hash-table? testconfig)
			     (hash-table-ref/default testconfig "ezsteps" '())
			     #f)))
	(if testconfig
	    (hash-table-set! *testconfigs* test-name testconfig) ;; cached for lazy reads later ...
	    (begin
	      (launch:setup)
	      (debug:print 0 *default-log-port* "WARNING: no testconfig found for " test-name " in search path:\n  "
			   (string-intersperse (tests:get-tests-search-path *configdat*) "\n  "))))
	;; after all that, still no testconfig? Time to abort
	(if (not testconfig)
	    (begin
	      (debug:print-error 0 *default-log-port* "Failed to resolve megatest.config, runconfigs.config and testconfig issues. Giving up now")
	      (exit 1)))

	;; create a proc for the subrun if requested, save that proc in the ezsteps table as the last entry
	;; 1. get section [runarun]
	;; 2. unset MT_* vars
	;; 3. fix target
	;; 4. fix runname
	;; 5. fix testpatt or calculate it from contour
	;; 6. launch the run
	;; 7. roll up the run result and or roll up the logpro processed result
	(when (configf:lookup testconfig "subrun" "runwait") ;; we use runwait as the flag that a subrun is requested
            (subrun:initialize-toprun-test testconfig test-run-dir)
	    (let* ((mt-cmd (subrun:launch-cmd test-run-dir)))
              (debug:print-info 0 *default-log-port* "Subrun command is \"" mt-cmd "\"")
              (set! ezsteps #t) ;; set the needed flag
	      (set! ezstepslst
                    (append (or ezstepslst '())
                            (list (list "subrun" (conc "{subrun=true} " mt-cmd)))))))

	;; process the ezsteps
	(if ezsteps
	    (begin
	      (if (not (common:file-exists? ".ezsteps"))(create-directory ".ezsteps"))
	      ;; if ezsteps was defined then we are sure to have at least one step but check anyway
	      (if (not (> (length ezstepslst) 0))
		  (debug:print-error 0 *default-log-port* "ezsteps defined but ezstepslst is zero length")
		  (let loop ((ezstep (car ezstepslst))
			     (tal    (cdr ezstepslst))
			     (prevstep #f))
                    (debug:print-info 0 *default-log-port* "Processing ezstep \"" (string-intersperse ezstep " ") "\"")
		    ;; check exit-info (vector-ref exit-info 1)
		    (if (launch:einf-exit-status exit-info) ;; (vector-ref exit-info 1)
			(let ((logpro-used (launch:runstep ezstep run-id test-id exit-info m tal testconfig))
			      (stepname    (car ezstep)))
			  ;; if logpro-used read in the stepname.dat file
			  (if (and logpro-used (common:file-exists? (conc stepname ".dat")))
			      (launch:load-logpro-dat run-id test-id stepname))
			  (if (steprun-good? logpro-used (launch:einf-exit-code exit-info))
			      (if (not (null? tal))
				  (loop (car tal) (cdr tal) stepname))
			      (debug:print 0 *default-log-port* "WARNING: step " (car ezstep) " failed. Stopping")))
			(debug:print 0 *default-log-port* "WARNING: a prior step failed, stopping at " ezstep)))))))))

(define (launch:monitor-job run-id test-id item-path fullrunscript ezsteps test-name tconfigreg exit-info m work-area runtlim misc-flags)
  (let* ((update-period (string->number (or (configf:lookup *configdat* "setup" "test-stats-update-period") "30")))
         (start-seconds (current-seconds))
	 (calc-minutes  (lambda ()
			  (inexact->exact 
			   (round 
			    (- 
			     (current-seconds) 
			     start-seconds)))))
	 (kill-tries 0))
    ;; (tests:set-full-meta-info #f test-id run-id (calc-minutes) work-area)
    ;; (tests:set-full-meta-info test-id run-id (calc-minutes) work-area)
    (tests:set-full-meta-info #f test-id run-id (calc-minutes) work-area 10)

    (let loop ((minutes   (calc-minutes))
	       (cpu-load  (alist-ref 'adj-core-load (common:get-normalized-cpu-load #f)))
	       (disk-free (get-df (current-directory)))
               (last-sync (current-seconds)))
      #;(common:telemetry-log "zombie" (conc "launch:monitor-job - top of loop encountered at "(current-seconds)" with last-sync="last-sync))
      (let* ((over-time     (> (current-seconds) (+ last-sync update-period)))
             (new-cpu-load  (let* ((load  (alist-ref 'adj-core-load (common:get-normalized-cpu-load #f)))
                                   (delta (abs (- load cpu-load))))
                              (if (> delta 0.1) ;; don't bother updating with small changes
                                  load
                                  #f)))
             (new-disk-free (let* ((df    (if over-time ;; only get df every 30 seconds
                                              (get-df (current-directory))
                                              disk-free))
                                   (delta (abs (- df disk-free))))
                              (if (and (> df 0)
                                       (> (/ delta df) 0.1)) ;; (> delta 200) ;; ignore changes under 200 Meg
                                  df
                                  #f)))
             (do-sync       (or new-cpu-load new-disk-free over-time))

             (test-info   (rmt:get-test-info-by-id run-id test-id))
             (state       (db:test-get-state test-info))
             (status      (db:test-get-status test-info))
             (kill-reason  "no kill reason specified")
             (kill-job?    #f))
        #;(common:telemetry-log "zombie" (conc "launch:monitor-job - decision time encountered at "(current-seconds)" with last-sync="last-sync" do-sync="do-sync" over-time="over-time" update-period="update-period))
        (cond
         ((test-get-kill-request run-id test-id)
          (set! kill-reason "KILLING TEST since received kill request (KILLREQ)")
          (set! kill-job? #t))
         ((and runtlim (> (- (current-seconds) start-seconds) runtlim))
          (set! kill-reason (conc "KILLING TEST DUE TO TIME LIMIT EXCEEDED! Runtime=" (- (current-seconds) start-seconds) " seconds, limit=" runtlim))
          (set! kill-job? #t))
         ((equal? status "DEAD")
          (tests:update-central-meta-info run-id test-id new-cpu-load new-disk-free (calc-minutes) #f #f)
          (rmt:set-state-status-and-roll-up-items run-id test-id 'foo "RUNNING" "n/a" "was marked dead; really still running.")
          ;;(set! kill-reason "KILLING TEST because it was marked as DEAD by launch:handle-zombie-tests (might indicate really overloaded server or else overzealous setup.deadtime)") ;; MARK RUNNING
          (set! kill-job? #f)))

        (debug:print 4 *default-log-port* "cpu: " new-cpu-load " disk: " new-disk-free " last-sync: " last-sync " do-sync: " do-sync)
        (launch:handle-zombie-tests run-id)
        (when do-sync
          ;;(with-output-to-file (conc (getenv "MT_TEST_RUN_DIR") "/last-loadinfo.log" #:append)
          ;;  (lambda () (pp (list (current-seconds) new-cpu-load new-disk-free (calc-minutes)))))
          #;(common:telemetry-log "zombie" (conc  "launch:monitor-job - dosync started at "(current-seconds)))
          (tests:update-central-meta-info run-id test-id new-cpu-load new-disk-free (calc-minutes) #f #f)
          #;(common:telemetry-log "zombie" (conc "launch:monitor-job - dosync finished at "(current-seconds))))
        
	(if kill-job? 
	    (begin
              (debug:print-info 0 *default-log-port* "proceeding to kill test: "kill-reason)
	      (mutex-lock! m)
	      ;; NOTE: The pid can change as different steps are run. Do we need handshaking between this
	      ;;       section and the runit section? Or add a loop that tries three times with a 1/4 second
	      ;;       between tries?
	      (let* ((pid1 (launch:einf-pid exit-info)) ;; (vector-ref exit-info 0))
		     (pid2 (rmt:test-get-top-process-pid run-id test-id))
		     (pids (delete-duplicates (filter number? (list pid1 pid2)))))
		(if (not (null? pids))
		    (begin
		      (for-each
		       (lambda (pid)
			 (handle-exceptions
			  exn
			  (begin
			    (debug:print-info 0 *default-log-port* "Unable to kill process with pid " pid ", possibly already killed.")
			    (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn)))
			  (debug:print 0 *default-log-port* "WARNING: Request received to kill job " pid) ;;  " (attempt # " kill-tries ")")
			  (debug:print-info 0 *default-log-port* "Signal mask=" (signal-mask))
			  ;; (if (process:alive? pid)
			  ;;     (begin
			  (map (lambda (pid-num)
				 (process-signal pid-num signal/term))
			       (process:get-sub-pids pid))
			  (thread-sleep! 5)
			  ;; (if (process:process-alive? pid)
			  (map (lambda (pid-num)
				 (handle-exceptions
				  exn
				  #f
				  (process-signal pid-num signal/kill)))
			       (process:get-sub-pids pid))))
		       ;;    (debug:print-info 0 *default-log-port* "not killing process " pid " as it is not alive"))))
		       pids)
                      ;; BB: question to Matt -- does the tests:test-state-status! encompass rollup to toplevel?  If not, should it?
		      (tests:test-set-status! run-id test-id "KILLED"  "KILLED" (conc (args:get-arg "-m")" "kill-reason) #f)) ;; BB ADDED kill-reason -- confirm OK with Matt
		    (begin
		      (debug:print-error 0 *default-log-port* "Nothing to kill, pid1=" pid1 ", pid2=" pid2)
		      (tests:test-set-status! run-id test-id "KILLED"  "FAILED TO KILL" (conc (args:get-arg "-m")" "kill-reason) #f) ;; BB ADDED kill-reason -- confirm OK with Matt
		      )))
	      (mutex-unlock! m)
	      ;; no point in sticking around. Exit now. But run end of run before exiting?
        (launch:end-of-run-check run-id)
	      (exit)))
	(if (hash-table-ref/default misc-flags 'keep-going #f)
	    (begin
	      (thread-sleep! 3) ;; (+ 3 (random 6))) ;; add some jitter to the call home time to spread out the db accesses
	      (if (hash-table-ref/default misc-flags 'keep-going #f)  ;; keep originals for cpu-load and disk-free unless they change more than the allowed delta
		  (loop (calc-minutes)
                        (or new-cpu-load cpu-load)
                        (or new-disk-free disk-free)
                        (if do-sync (current-seconds) last-sync)))))))
    (tests:update-central-meta-info run-id test-id (get-cpu-load) (get-df (current-directory))(calc-minutes) #f #f))) ;; NOTE: Checking twice for keep-going is intentional


(define (launch:execute encoded-cmd)
  (let* ((cmdinfo    (common:read-encoded-string encoded-cmd))
	 (tconfigreg #f))
    (setenv "MT_CMDINFO" encoded-cmd)
    ;;(bb-check-path msg: "launch:execute incoming")
    (if (list? cmdinfo) ;; ((testpath /tmp/mrwellan/jazzmind/src/example_run/tests/sqlitespeed)
	;; (test-name sqlitespeed) (runscript runscript.rb) (db-host localhost) (run-id 1))
	(let* ((testpath  (assoc/default 'testpath  cmdinfo))  ;; testpath is the test spec area
	       (top-path  (assoc/default 'toppath   cmdinfo))
	       (work-area (assoc/default 'work-area cmdinfo))  ;; work-area is the test run area
	       (test-name (assoc/default 'test-name cmdinfo))
	       (runscript (assoc/default 'runscript cmdinfo))
	       (ezsteps   (assoc/default 'ezsteps   cmdinfo))
	       (subrun    (assoc/default 'subrun    cmdinfo))
	       ;; (runremote (assoc/default 'runremote cmdinfo))
	       ;; (transport (assoc/default 'transport cmdinfo))  ;; not used
	       ;; (serverinf (assoc/default 'serverinf cmdinfo))
	       ;; (port      (assoc/default 'port      cmdinfo))
	       (serverurl (assoc/default 'serverurl cmdinfo))
	       (homehost  (assoc/default 'homehost  cmdinfo))
	       (run-id    (assoc/default 'run-id    cmdinfo))
	       (test-id   (assoc/default 'test-id   cmdinfo))
	       (target    (assoc/default 'target    cmdinfo))
	       (areaname  (assoc/default 'areaname  cmdinfo))
	       (itemdat   (assoc/default 'itemdat   cmdinfo))
	       (env-ovrd  (assoc/default 'env-ovrd  cmdinfo))
	       (set-vars  (assoc/default 'set-vars  cmdinfo)) ;; pre-overrides from -setvar
	       (runname   (assoc/default 'runname   cmdinfo))
	       (megatest  (assoc/default 'megatest  cmdinfo))
	       (runtlim   (assoc/default 'runtlim   cmdinfo))
	       (contour   (assoc/default 'contour   cmdinfo))
	       (item-path (item-list->path itemdat))
	       (mt-bindir-path (assoc/default 'mt-bindir-path cmdinfo))
	       (keys      #f)
	       (keyvals   #f)
	       (fullrunscript (if (not runscript)
                                  #f
                                  (if (substring-index "/" runscript)
                                      runscript ;; use unadultered if contains slashes
                                      (let ((fulln (conc work-area "/" runscript)))
	                                  (if (and (common:file-exists? fulln)
                                                   (file-execute-access? fulln))
                                              fulln
                                              runscript))))) ;; assume it is on the path
               (check-work-area           (lambda ()
                                            ;; NFS might not have propagated the directory meta data to the run host - give it time if needed
                                            (let loop ((count 0))
                                              (if (or (common:directory-exists? work-area)
                                                      (> count 10))
                                                  (change-directory work-area)
                                                  (begin
                                                    (debug:print 0 *default-log-port* "INFO: Not starting job yet - directory " work-area " not found")
                                                    (thread-sleep! 10)
                                                    (loop (+ count 1)))))

                                            (if (not (string=?  (common:real-path work-area)(common:real-path (current-directory))))
                                                (begin
                                                  (debug:print 0 *default-log-port*
                                                               "INFO: we are expecting to be in directory " work-area "\n"
                                                               "     but we are actually in the directory " (current-directory) "\n"
                                                               "     doing another change dir.")
                                                  (change-directory work-area)))
                                            
                                            ;; spot check that the files in testpath are available. Too often NFS delays cause problems here.
                                            (let ((files      (glob (conc testpath "/*")))
                                                  (bad-files '()))
                                              (for-each
                                               (lambda (fullname)
                                                 (let* ((fname (pathname-strip-directory fullname))
                                                        (targn (conc work-area "/" fname)))
                                                   (if (not (file-exists? targn))
                                                       (set! bad-files (cons fname bad-files)))))
                                               files)
                                              (if (not (null? bad-files))
                                                  (begin
                                                    (debug:print 0 *default-log-port* "INFO: test data from " testpath " not copied properly or filesystem problems causing data to not be found. Re-running the copy command.")
                                                    (debug:print 0 *default-log-port* "INFO: missing files from " work-area ": " (string-intersperse bad-files ", "))
                                                    (launch:test-copy testpath work-area))))
                                            ;; one more time, change to the work-area directory
                                            (change-directory work-area)))
	       ) ;; let*

	  (if contour (setenv "MT_CONTOUR" contour))
	  
	  ;; immediated set some key variables from CMDINFO data, yes, these will be set again below ...
	  ;;
	  (setenv "MT_TESTSUITENAME" areaname)
	  (setenv "MT_RUN_AREA_HOME" top-path)
	  (set! *toppath* top-path)
          (change-directory *toppath*) ;; temporarily switch to the run area home
	  (setenv "MT_TEST_RUN_DIR"  work-area)

	  (launch:setup) ;; should be properly in the run area home now

	  (if contour (setenv "MT_CONTOUR" contour))
	  
	  ;; immediated set some key variables from CMDINFO data, yes, these will be set again below ...
	  ;;
	  (setenv "MT_TESTSUITENAME" areaname)
	  (setenv "MT_RUN_AREA_HOME" top-path)
	  (set! *toppath* top-path)
          (change-directory *toppath*) ;; temporarily switch to the run area home
	  (setenv "MT_TEST_RUN_DIR"  work-area)

	  (launch:setup) ;; should be properly in the run area home now
          
	  (set! tconfigreg (tests:get-all)) ;; mapping of testname => test source path
	  (let ((sighand (lambda (signum)
			   ;; (signal-mask! signum) ;; to mask or not? seems to cause issues in exiting
			   (if (eq? signum signal/stop)
			       (debug:print-error 0 *default-log-port* "attempt to STOP process. Exiting."))
			   (set! *time-to-exit* #t)
			   (print "Received signal " signum ", cleaning up before exit (set this test to COMPLETED/ABORT) . Please wait...")
			   (let ((th1 (make-thread (lambda ()
                                                     (print "set test to COMPLETED/ABORT begin.")
						     (rmt:test-set-state-status run-id test-id "COMPLETED" "ABORT" "received kill signal")
                                                     (print "set test to COMPLETED/ABORT complete.")
						     (print "Killed by signal " signum ". Exiting")
						     (exit 1))))
				 (th2 (make-thread (lambda ()
						     (thread-sleep! 20)
						     (debug:print 0 *default-log-port* "Done")
						     (exit 4)))))
			     (thread-start! th2)
			     (thread-start! th1)
			     (thread-join! th2)))))
	    (set-signal-handler! signal/int sighand)
	    (set-signal-handler! signal/term sighand)
	    ) ;; (set-signal-handler! signal/stop sighand)
	  
	  ;; Do not run the test if it is REMOVING, RUNNING, KILLREQ or REMOTEHOSTSTART,
	  ;; Mark the test as REMOTEHOSTSTART *IMMEDIATELY*
	  ;;
	  (let* ((test-info (rmt:get-test-info-by-id run-id test-id))
		 (test-host (if test-info
				(db:test-get-host        test-info)
				(begin
				  (debug:print 0 *default-log-port* "ERROR: failed to find a record for test-id " test-id ", exiting.")
				  (exit))))
		 (test-pid  (db:test-get-process_id  test-info)))
	    (cond
             ;; -mrw- I'm removing KILLREQ from this list so that a test in KILLREQ state is treated as a "do not run" flag.
	     ((member (db:test-get-state test-info) '("INCOMPLETE" "KILLED" "UNKNOWN" "STUCK")) ;; prior run of this test didn't complete, go ahead and try to rerun
	      (debug:print 0 *default-log-port* "INFO: test is INCOMPLETE or KILLED, treat this execute call as a rerun request")
	      ;; (tests:test-force-state-status! run-id test-id "REMOTEHOSTSTART" "n/a")

              (rmt:general-call 'set-test-start-time #f test-id)
              (rmt:test-set-state-status run-id test-id "REMOTEHOSTSTART" "n/a" #f)
	      ) ;; prime it for running
	     ((member (db:test-get-state test-info) '("RUNNING" "REMOTEHOSTSTART"))
	      (if (process:alive-on-host? test-host test-pid)
		  (debug:print-error 0 *default-log-port* "test state is "  (db:test-get-state test-info) " and process " test-pid " is still running on host " test-host ", cannot proceed")
		  (exit)))
	     ((not (member (db:test-get-state test-info) '("REMOVING" "REMOTEHOSTSTART" "RUNNING" "KILLREQ")))
	      ;; (tests:test-force-state-status! run-id test-id "REMOTEHOSTSTART" "n/a")
              (rmt:general-call 'set-test-start-time #f test-id)
	      (rmt:test-set-state-status run-id test-id "REMOTEHOSTSTART" "n/a" #f)
	      )
	     (else ;; (member (db:test-get-state test-info) '("REMOVING" "REMOTEHOSTSTART" "RUNNING" "KILLREQ"))
	      (debug:print-error 0 *default-log-port* "test state is " (db:test-get-state test-info) ", cannot proceed")
	      (exit))))

          ;; cleanup prior execution's steps
          (rmt:delete-steps-for-test! run-id test-id)
          
	  (debug:print 2 *default-log-port* "Executing " test-name " (id: " test-id ") on " (get-host-name))
	  (set! keys       (rmt:get-keys))
	  ;; (runs:set-megatest-env-vars run-id inkeys: keys inkeyvals: keyvals) ;; these may be needed by the launching process
	  ;; one of these is defunct/redundant ...
	  (if (not (launch:setup force-reread: #t))
	      (begin
		(debug:print 0 *default-log-port* "Failed to setup, exiting") 
		;; (sqlite3:finalize! db)
		;; (sqlite3:finalize! tdb)
		(exit 1)))
          ;; validate that the test run area is available
          (check-work-area)
          
          ;; still need to go back to run area home for next couple steps
	  (change-directory *toppath*) 

	  ;; NOTE: Current order is to process runconfigs *before* setting the MT_ vars. This 
	  ;;       seems non-ideal but could well break stuff
	  ;;    BUG? BUG? BUG?
	  
	  (let ((rconfig (full-runconfigs-read)) ;; (read-config (conc  *toppath* "/runconfigs.config") #f #t sections: (list "default" target))))
		(wconfig (read-config "waivers.config" #f #t sections: `( "default" ,target )))) ;; read the waivers config if it exists
	    ;; (setup-env-defaults (conc *toppath* "/runconfigs.config") run-id (make-hash-table) keyvals target)
	    ;; (set-run-config-vars run-id keyvals target) ;; (db:get-target db run-id))
	    ;; Now have runconfigs data loaded, set environment vars
	    (for-each
	     (lambda (section)
	       (for-each
		(lambda (varval)
		  (let ((var (car varval))
			(val (cadr varval)))
		    (if (and (string? var)(string? val))
			(begin
			  (safe-setenv var (configf:eval-string-in-environment val))) ;; val)
			(debug:print-error 0 *default-log-port* "bad variable spec, " var "=" val))))
		(configf:get-section rconfig section)))
	     (list "default" target)))
          ;;(bb-check-path msg: "launch:execute post block 1")

	  ;; NFS might not have propagated the directory meta data to the run host - give it time if needed
	  (let loop ((count 0))
	    (if (or (common:file-exists? work-area)
		    (> count 10))
		(change-directory work-area)
		(begin
		  (debug:print 0 *default-log-port* "INFO: Not starting job yet - directory " work-area " not found")
		  (thread-sleep! 10)
		  (loop (+ count 1)))))

          ;; now we can switch to the work-area?
          (change-directory work-area)
          ;;(bb-check-path msg: "launch:execute post block 1.5")
	  ;; (change-directory work-area) 
	  (set! keyvals    (keys:target->keyval keys target))
	  ;; apply pre-overrides before other variables. The pre-override vars must not
	  ;; clobbers things from the official sources such as megatest.config and runconfigs.config
	  (if (string? set-vars)
	      (let ((varpairs (string-split set-vars ",")))
		(debug:print 4 *default-log-port* "varpairs: " varpairs)
		(map (lambda (varpair)
		       (let ((varval (string-split varpair "=")))
			 (if (eq? (length varval) 2)
			     (let ((var (car varval))
				   (val (cadr varval)))
			       (debug:print 1 *default-log-port* "Adding pre-var/val " var " = " val " to the environment")
			       (setenv var val)))))
		     varpairs)))
          ;;(bb-check-path msg: "launch:execute post block 2")
	  (for-each
	   (lambda (varval)
	     (let ((var (car varval))
		   (val (cadr varval)))
	       (if val
		   (setenv var val)
		   (begin
		     (debug:print-error 0 *default-log-port* "required variable " var " does not have a valid value. Exiting")
		     (exit)))))
	     (list 
	      (list  "MT_TEST_RUN_DIR" work-area)
	      (list  "MT_TEST_NAME" test-name)
	      (list  "MT_ITEM_INFO" (conc itemdat))
	      (list  "MT_ITEMPATH"  item-path)
	      (list  "MT_RUNNAME"   runname)
	      (list  "MT_MEGATEST"  megatest)
	      (list  "MT_TARGET"    target)
	      (list  "MT_LINKTREE"  (common:get-linktree)) ;; (configf:lookup *configdat* "setup" "linktree"))
	      (list  "MT_TESTSUITENAME" (common:get-testsuite-name))))
          ;;(bb-check-path msg: "launch:execute post block 3")

	  (if mt-bindir-path (setenv "PATH" (conc (getenv "PATH") ":" mt-bindir-path)))
          ;;(bb-check-path msg: "launch:execute post block 4")
	  ;; (change-directory top-path)
	  ;; Can setup as client for server mode now
	  ;; (client:setup)

	  
	  ;; environment overrides are done *before* the remaining critical envars.
	  (alist->env-vars env-ovrd)
          ;;(bb-check-path msg: "launch:execute post block 41")
	  (runs:set-megatest-env-vars run-id inkeys: keys inkeyvals: keyvals)
          ;;(bb-check-path msg: "launch:execute post block 42")
	  (set-item-env-vars itemdat)
          ;;(bb-check-path msg: "launch:execute post block 43")
          (let ((blacklist (configf:lookup *configdat* "setup" "blacklistvars")))
            (if blacklist
		(let ((vars (string-split blacklist)))
		  (save-environment-as-files "megatest" ignorevars: vars)
		  (for-each (lambda (var)
			      (unsetenv var))
			    vars))
                (save-environment-as-files "megatest")))
          ;;(bb-check-path msg: "launch:execute post block 44")
	  ;; open-run-close not needed for test-set-meta-info
	  ;; (tests:set-full-meta-info #f test-id run-id 0 work-area)
	  ;; (tests:set-full-meta-info test-id run-id 0 work-area)
	  (tests:set-full-meta-info #f test-id run-id 0 work-area 10)

	  ;; (thread-sleep! 0.3) ;; NFS slowness has caused grief here

	  (if (args:get-arg "-xterm")
	      (set! fullrunscript "xterm")
	      (if (and fullrunscript 
		       (common:file-exists? fullrunscript)
		       (not (file-execute-access? fullrunscript)))
		  (system (conc "chmod ug+x " fullrunscript))))

	  ;; We are about to actually kick off the test
	  ;; so this is a good place to remove the records for 
	  ;; any previous runs
	  ;; (db:test-remove-steps db run-id testname itemdat)
	  ;; now is also a good time to write the .testconfig file
	  (let* ((tconfig-fname   (conc work-area "/.testconfig"))
		 (tconfig-tmpfile (conc tconfig-fname ".tmp"))
		 (tconfig         (tests:get-testconfig test-name item-path tconfigreg #t force-create: #t))) ;; 'return-procs)))
	    (configf:write-alist tconfig tconfig-tmpfile)
	    (file-move tconfig-tmpfile tconfig-fname #t))
	  ;; 
	  (let* ((m            (make-mutex))
		 (kill-job?    #f)
		 (exit-info    (make-launch:einf pid: #t exit-status: #t exit-code: #t rollup-status: 0)) ;; pid exit-status exit-code (i.e. process was successfully run) rollup-status
		 (job-thread   #f)
		 ;; (keep-going   #t)
		 (misc-flags   (let ((ht (make-hash-table)))
				 (hash-table-set! ht 'keep-going #t)
				 ht))
		 (runit        (lambda ()
				 (launch:manage-steps run-id test-id item-path fullrunscript ezsteps subrun test-name tconfigreg exit-info m)))
		 (monitorjob   (lambda ()
				 (launch:monitor-job  run-id test-id item-path fullrunscript ezsteps test-name tconfigreg exit-info m work-area runtlim misc-flags)))
		 (th1          (make-thread monitorjob "monitor job"))
		 (th2          (make-thread runit "run job")))
	    (set! job-thread th2)
	    (thread-start! th1)
	    (thread-start! th2)
	    (thread-join! th2)
	    (debug:print-info 0 *default-log-port* "Megatest exectute of test " test-name ", item path " item-path " complete. Notifying the db ...")
	    (hash-table-set! misc-flags 'keep-going #f)
	    (thread-join! th1)
	    (thread-sleep! 1)       ;; givbe thread th1 a chance to be done TODO: Verify this is needed. At 0.1 I was getting fail to stop, increased to total of 1.1 sec.
	    (mutex-lock! m)
	    (let* ((item-path (item-list->path itemdat))
		   ;; only state and status needed - use lazy routine
		   (testinfo  (rmt:get-testinfo-state-status run-id test-id)))
	      ;; Am I completed?
	      (if (member (db:test-get-state testinfo) '("REMOTEHOSTSTART" "RUNNING")) ;; NOTE: It should *not* be REMOTEHOSTSTART but for reasons I don't yet understand it sometimes gets stuck in that state ;; (not (equal? (db:test-get-state testinfo) "COMPLETED"))
		  (let ((new-state  (if kill-job? "KILLED" "COMPLETED") ;; (if (eq? (vector-ref exit-info 2) 0) ;; exited with "good" status
				                                        ;; "COMPLETED"							                ;; (db:test-get-state testinfo)))   ;; else preseve the state as set within the test
				    )
			(new-status (cond
				     ((not (launch:einf-exit-status exit-info)) "FAIL") ;; job failed to run ... (vector-ref exit-info 1)
				     ((eq? (launch:einf-rollup-status exit-info) 0)     ;; (vector-ref exit-info 3)
				      ;; if the current status is AUTO then defer to the calculated value (i.e. leave this AUTO)
				      (if (equal? (db:test-get-status testinfo) "AUTO") "AUTO" "PASS"))
				     ((eq? (launch:einf-rollup-status exit-info) 1) "FAIL")  ;; (vector-ref exit-info 3)
				     ((eq? (launch:einf-rollup-status exit-info) 2)	     ;;	(vector-ref exit-info 3)
				      ;; if the current status is AUTO the defer to the calculated value but qualify (i.e. make this AUTO-WARN)
				      (if (equal? (db:test-get-status testinfo) "AUTO") "AUTO-WARN" "WARN"))
				     ((eq? (launch:einf-rollup-status exit-info) 3) "CHECK")
				     ((eq? (launch:einf-rollup-status exit-info) 4) "WAIVED")
				     ((eq? (launch:einf-rollup-status exit-info) 5) "ABORT")
				     ((eq? (launch:einf-rollup-status exit-info) 6) "SKIP")
				     (else "FAIL")))) ;; (db:test-get-status testinfo)))
		    (debug:print-info 1 *default-log-port* "Test exited in state=" (db:test-get-state testinfo) ", setting state/status based on exit code of " (launch:einf-exit-status exit-info) " and rollup-status of " (launch:einf-rollup-status exit-info))
		    (tests:test-set-status! run-id 
					    test-id 
					    new-state
					    new-status
					    (args:get-arg "-m") #f)
		    ;; need to update the top test record if PASS or FAIL and this is a subtest
		    ;; NO NEED TO CALL set-state-status-and-roll-up-items HERE, THIS IS DONE IN set-state-status-and-roll-up-items called by tests:test-set-status!
		    ))
	      ;; for automated creation of the rollup html file this is a good place...
	      (if (not (equal? item-path ""))
		  (tests:summarize-items run-id test-id test-name #f))
	      (tests:summarize-test run-id test-id)  ;; don't force - just update if no
	      (rmt:update-run-stats run-id (rmt:get-raw-run-stats run-id)))
	    (mutex-unlock! m)
            (launch:end-of-run-check run-id )
	    (debug:print 2 *default-log-port* "Output from running " fullrunscript ", pid " (launch:einf-pid exit-info) " in work area " 
			 work-area ":\n====\n exit code " (launch:einf-exit-code exit-info) "\n" "====\n")
	    (if (not (launch:einf-exit-status exit-info))
		(exit 4))))
        )))

;; 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 (launch: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* "rollup run state/status")                      
     (rmt:set-state-status-and-roll-up-run  run-id current-state current-status)
 
     (cond 
       ((and all-test-launched (eq? not-completed-cnt 0) (equal? all-test-launched "yes" ))
           	(debug:print 0 *default-log-port* "look for  post hook.")
          	(runs:run-post-hook 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))
           					(launch: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 (launch:is-test-alive host pid)
(if (and host pid (not (equal? host "n/a")))
(let* ((cmd (conc "ssh " host " pstree -A " pid))
      (output (with-input-from-pipe cmd read-lines)))
  (print "cmd: " cmd "\n op: " output )
  (if(eq? (length output) 0)
     #f
     #t))
#t))
 
(define (launch:kill-tests-if-dead run-id)
  (let* ((running-tests (rmt:get-tests-for-run run-id "%" `("RUNNING" "LAUNCHED" "REMOTEHOSTSTART") `() #f #f #f #f #f #f #f #f)))
       (let loop ((running-test (car running-tests))
			     (tal    (cdr running-tests))
			     (kill-cnt 0))
		       (let* ((test-name (vector-ref running-test 2))
                 (item-path (vector-ref running-test 11))
								 (test-id (vector-ref running-test 0))
                 (host (vector-ref running-test 6))
                 (pid  (rmt:test-get-top-process-pid run-id test-id))   
                 (event-time (vector-ref running-test 5))
                 (duration (vector-ref running-test 12))
                 (flag 0)   
                 (curr-time (current-seconds)))
       (if (and (< (+ event-time duration 600) curr-time) (not (launch:is-test-alive host pid))) ;;test has not updated duration in last 10 min then likely its not running but confirm before marking it as killed
           (begin    
			       	(debug:print 0 *default-log-port* "test " test-name "/" item-path " needs to be killed")
              (set! flag 1) 
              (rmt:set-state-status-and-roll-up-items run-id test-name item-path "KILLREQ" "n/a" #f)))
               (if (not (null? tal))
				  (loop (car tal) (cdr tal) (+ kill-cnt flag))
                 (+ kill-cnt flag))))))

;; DO NOT USE - caching of configs is handled in launch:setup now.
;;
(define (launch:cache-config)
  ;; if we have a linktree and -runtests and -target and the directory exists dump the config
  ;; to megatest-(current-seconds).cfg and symlink it to megatest.cfg
  (if (and *configdat* 
	   (or (args:get-arg "-run")
	       (args:get-arg "-runtests")
	       (args:get-arg "-execute")))
      (let* ((linktree (common:get-linktree)) ;; (get-environment-variable "MT_LINKTREE"))
	     (target   (common:args-get-target exit-if-bad: #t))
	     (runname  (or (args:get-arg "-runname")
			   (args:get-arg ":runname")
			   (getenv "MT_RUNNAME")))
	     (fulldir  (conc linktree "/"
			     target "/"
			     runname)))
	(if (and linktree (common:file-exists? linktree)) ;; can't proceed without linktree
	    (begin
	      (debug:print-info 0 *default-log-port* "Have -run with target=" target ", runname=" runname ", fulldir=" fulldir ", testpatt=" (or (args:get-arg "-testpatt") "%"))
	      (if (not (common:file-exists? fulldir))
		  (create-directory fulldir #t)) ;; need to protect with exception handler 
	      (if (and target
		       runname
		       (common:file-exists? fulldir))
		  (let ((tmpfile  (conc fulldir "/.megatest.cfg." (current-seconds)))
			(targfile (conc fulldir "/.megatest.cfg-"  megatest-version "-" megatest-fossil-hash))
			(rconfig  (conc fulldir "/.runconfig." megatest-version "-" megatest-fossil-hash)))
		    (if (common:file-exists? rconfig) ;; only cache megatest.config AFTER runconfigs has been cached
			(begin
			  (debug:print-info 0 *default-log-port* "Caching megatest.config in " tmpfile)
                          (if (not (common:in-running-test?))
                              (configf:write-alist *configdat* tmpfile))
			  (system (conc "ln -sf " tmpfile " " targfile))))
		    )))
	    (debug:print-info 1 *default-log-port* "No linktree yet, no caching configs.")))))


;; gather available information, if legit read configs in this order:
;;
;;   if have cache;
;;      read it a return it
;;   else
;;     megatest.config     (do not cache)
;;     runconfigs.config   (cache if all vars avail)
;;     megatest.config     (cache if all vars avail)
;;   returns:
;;     *toppath*
;;   side effects:
;;     sets; *configdat*    (megatest.config info)
;;           *runconfigdat* (runconfigs.config info)
;;           *configstatus* (status of the read data)
;;
(define (launch:setup #!key (force-reread #f) (areapath #f))
  (mutex-lock! *launch-setup-mutex*)
  (if (and *toppath*
	   (eq? *configstatus* 'fulldata) (not force-reread)) ;; got it all
      (begin
	(debug:print 2 *default-log-port* "NOTE: skipping launch:setup-body call since we have fulldata")
	(mutex-unlock! *launch-setup-mutex*)
	*toppath*)
      (let ((res (launch:setup-body force-reread: force-reread areapath: areapath)))
	(mutex-unlock! *launch-setup-mutex*)
	res)))

;; return paths depending on what info is available.
;;
(define (launch:get-cache-file-paths areapath toppath target mtconfig)
  (let* ((use-cache (common:use-cache?))
         (runname  (common:args-get-runname))
         (linktree (common:get-linktree))
         (testname (common:get-full-test-name))
         (rundir   (if (and runname target linktree)
                       (common:directory-writable? (conc linktree "/" target "/" runname))
                       #f))
         (testdir  (if (and rundir testname)
                       (common:directory-writable? (conc rundir "/" testname))
                       #f))
         (cachedir (or testdir rundir))
         (mtcachef (and cachedir (conc cachedir "/" ".megatest.cfg-"  megatest-version "-" megatest-fossil-hash)))
         (rccachef (and cachedir (conc cachedir "/" ".runconfigs.cfg-"  megatest-version "-" megatest-fossil-hash))))
    (debug:print-info 6 *default-log-port* 
                      "runname=" runname 
                      "\n  linktree=" linktree
                      "\n  testname=" testname
                      "\n  rundir=" rundir 
                      "\n  testdir=" testdir 
                      "\n  cachedir=" cachedir
                      "\n  mtcachef=" mtcachef
                      "\n  rccachef=" rccachef)
    (cons mtcachef rccachef)))

(define (launch:setup-body #!key (force-reread #f) (areapath #f))
  (if (and (eq? *configstatus* 'fulldata)
	   *toppath*
	   (not force-reread)) ;; no need to reprocess
      *toppath*   ;; return toppath
      (let* ((use-cache (common:use-cache?)) ;; BB- use-cache checks *configdat* for use-cache setting.  We do not have *configdat*.  Bootstrapping problem here.
	     (toppath  (or *toppath* areapath (getenv "MT_RUN_AREA_HOME"))) ;; preserve toppath
	     (target   (common:args-get-target))
	     (sections (if target (list "default" target) #f)) ;; for runconfigs
	     (mtconfig (or (args:get-arg "-config") "megatest.config")) ;; allow overriding megatest.config 
             (cachefiles (launch:get-cache-file-paths areapath toppath target mtconfig))
	     ;; checking for null cachefiles should not be necessary, I was seeing error car of '(), might be a chicken bug or a red herring ...
	     (mtcachef   (if (null? cachefiles)
			     #f
			     (car cachefiles))) ;; (and cachedir (conc cachedir "/" ".megatest.cfg-"  megatest-version "-" megatest-fossil-hash)))
	     (rccachef   (if (null? cachefiles)
			     #f
			     (cdr cachefiles)))) ;; (and cachedir (conc cachedir "/" ".runconfigs.cfg-"  megatest-version "-" megatest-fossil-hash)))
	      ;; (cancreate (and cachedir (common:file-exists? cachedir)(file-write-access? cachedir) (not (common:in-running-test?)))))
	(set! *toppath* toppath) ;; This is needed when we are running as a test using CMDINFO as a datasource
        ;;(BB> "launch:setup-body -- cachefiles="cachefiles)
	(cond
	 ;; if mtcachef exists just read it, however we need to assume toppath is available in $MT_RUN_AREA_HOME
	 ((and (not force-reread)
	       mtcachef  rccachef
	       use-cache
	       (get-environment-variable "MT_RUN_AREA_HOME")
	       (common:file-exists? mtcachef)
	       (common:file-exists? rccachef))
          ;;(BB> "launch:setup-body -- cond branch 1 - use-cache")
          (set! *configdat*    (configf:read-alist mtcachef))
          ;;(BB> "launch:setup-body -- 1 set! *configdat*="*configdat*)
	  (set! *runconfigdat* (configf:read-alist rccachef))
	  (set! *configinfo*   (list *configdat*  (get-environment-variable "MT_RUN_AREA_HOME")))
	  (set! *configstatus* 'fulldata)
	  (set! *toppath*      (get-environment-variable "MT_RUN_AREA_HOME"))
	  *toppath*)
	 ;; there are no existing cached configs, do full reads of the configs and cache them
	 ;; we have all the info needed to fully process runconfigs and megatest.config
	 ((and ;; (not force-reread) ;; force-reread is irrelevant in the AND, could however OR it?
	       mtcachef
	       rccachef) ;; BB- why are we doing this without asking if caching is desired?
          ;;(BB> "launch:setup-body -- cond branch 2")
	  (let* ((first-pass    (find-and-read-config        ;; NB// sets MT_RUN_AREA_HOME as side effect
				 mtconfig
				 environ-patt: "env-override"
				 given-toppath: toppath
				 pathenvvar: "MT_RUN_AREA_HOME"))
		 (first-rundat  (let ((toppath (if toppath 
						   toppath
						   (car first-pass))))
				  (read-config ;; (conc toppath "/runconfigs.config") ;; this should be converted to runconfig:read but it is non-trivial, leaving it for now.
				   (conc (if (string? toppath)
					     toppath
					     (get-environment-variable "MT_RUN_AREA_HOME"))
					 "/runconfigs.config")
				   *runconfigdat* #t 
				   sections: sections))))
	    (set! *runconfigdat* first-rundat)
	    (if first-pass  ;; 
		(begin
                  ;;(BB> "launch:setup-body -- \"first-pass\"=first-pass")
		  (set! *configdat*  (car first-pass))
                  ;;(BB> "launch:setup-body -- 2 set! *configdat*="*configdat*)
		  (set! *configinfo* first-pass)
		  (set! *toppath*    (or toppath (cadr first-pass))) ;; use the gathered data unless already have it
		  (set! toppath      *toppath*)
		  (if (not *toppath*)
		      (begin
			(debug:print-error 0 *default-log-port* "you are not in a megatest area!")
			(exit 1)))
		  (setenv "MT_RUN_AREA_HOME" *toppath*)
		  ;; the seed read is done, now read runconfigs, cache it then read megatest.config one more time and cache it
		  (let* ((keys         (rmt:get-keys))
			 (key-vals     (keys:target->keyval keys target))
			 (linktree     (common:get-linktree)) ;; (or (getenv "MT_LINKTREE")(if *configdat* (configf:lookup *configdat* "setup" "linktree") #f)))
					;     (if *configdat*
					; 	   (configf:lookup *configdat* "setup" "linktree")
					; 	   (conc *toppath* "/lt"))))
			 (second-pass  (find-and-read-config
					mtconfig
					environ-patt: "env-override"
					given-toppath: toppath
					pathenvvar: "MT_RUN_AREA_HOME"))
			 (runconfigdat (begin     ;; this read of the runconfigs will see any adjustments made by re-reading megatest.config
					 (for-each (lambda (kt)
						     (setenv (car kt) (cadr kt)))
						   key-vals)
					 (read-config (conc toppath "/runconfigs.config") *runconfigdat* #t ;; consider using runconfig:read some day ...
						      sections: sections)))
                         (cachefiles   (launch:get-cache-file-paths areapath toppath target mtconfig))
                         (mtcachef     (car cachefiles))
                         (rccachef     (cdr cachefiles)))
                    ;;  trap exception due to stale NFS handle -- Error: (open-output-file) cannot open file - Stale NFS file handle: "/p/fdk/gwa/lefkowit/mtTesting/qa/primbeqa/links/p1222/11/PDK_r1.1.1/prim/clean/pcell_testgen/.runconfigs.cfg-1.6427-7d1e789cb3f62f9cde719a4865bb51b3c17ea853" - ticket 220546342
                    ;; TODO - consider 1) using simple-lock to bracket cache write
                    ;;                 2) cache in hash on server, since need to do rmt: anyway to lock.

		    (if rccachef
                        (common:fail-safe
                         (lambda ()
                           (configf:write-alist runconfigdat rccachef))
                         (conc "Could not write cache file - "rccachef)))
                    (if mtcachef
                        (common:fail-safe
                         (lambda ()
                           (configf:write-alist *configdat* mtcachef))
                         (conc "Could not write cache file - "mtcachef)))
		    (set! *runconfigdat* runconfigdat)
		    (if (and rccachef mtcachef) (set! *configstatus* 'fulldata))))
		;; no configs found? should not happen but let's try to recover gracefully, return an empty hash-table
		(set! *configdat* (make-hash-table))
		)))

	 ;; else read what you can and set the flag accordingly
	 ;; here we don't have either mtconfig or rccachef
	 (else
          ;;(BB> "launch:setup-body -- cond branch 3 - else")
	  (let* ((cfgdat   (find-and-read-config 
			    (or (args:get-arg "-config") "megatest.config")
			    environ-patt: "env-override"
			    given-toppath: (get-environment-variable "MT_RUN_AREA_HOME")
			    pathenvvar: "MT_RUN_AREA_HOME")))

            (if (and cfgdat (list? cfgdat) (> (length cfgdat) 0) (hash-table? (car cfgdat)))
		(let* ((toppath  (or (get-environment-variable "MT_RUN_AREA_HOME")(cadr cfgdat)))
		       (rdat     (read-config (conc toppath  ;; convert this to use runconfig:read!
						    "/runconfigs.config") *runconfigdat* #t sections: sections)))
		  (set! *configinfo*   cfgdat)
		  (set! *configdat*    (car cfgdat))
		  (set! *runconfigdat* rdat)
		  (set! *toppath*      toppath)
		  (set! *configstatus* 'partial))
		(begin
		  (debug:print-error 0 *default-log-port* "No " mtconfig " file found. Giving up.")
		  (exit 2))))))
	;; COND ends here.
	
	;; additional house keeping
	(let* ((linktree (or (common:get-linktree)
			     (conc *toppath* "/lt"))))
	  (if linktree
	      (begin
		(if (not (common:file-exists? linktree))
		    (begin
		      (handle-exceptions
			  exn
			  (begin
			    (debug:print-error 0 *default-log-port* "Something went wrong when trying to create linktree dir at " linktree)
			    (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
			    (exit 1))
			(create-directory linktree #t))))
		(handle-exceptions
		    exn
		    (begin
		      (debug:print-error 0 *default-log-port* "Something went wrong when trying to create link to linktree at " *toppath*)
		      (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn)))
		  (let ((tlink (conc *toppath* "/lt")))
		    (if (not (common:file-exists? tlink))
			(create-symbolic-link linktree tlink)))))
	      (begin
		(debug:print-error 0 *default-log-port* "linktree not defined in [setup] section of megatest.config")
		)))
	(if (and *toppath*
		 (directory-exists? *toppath*))
	    (begin
	      (setenv "MT_RUN_AREA_HOME" *toppath*)
	      (setenv "MT_TESTSUITENAME" (common:get-testsuite-name)))
	    (begin
	      (debug:print-error 0 *default-log-port* "failed to find the top path to your Megatest area.")
	      (set! *toppath* #f) ;; force it to be false so we return #f
	      #f))
	
        ;; one more attempt to cache the configs for future reading
        (let* ((cachefiles   (launch:get-cache-file-paths areapath toppath target mtconfig))
               (mtcachef     (car cachefiles))
               (rccachef     (cdr cachefiles)))

          ;; trap exception due to stale NFS handle -- Error: (open-output-file) cannot open file - Stale NFS file handle: "...somepath.../.runconfigs.cfg-1.6427-7d1e789cb3f62f9cde719a4865bb51b3c17ea853" - ticket 220546342
          ;; TODO - consider 1) using simple-lock to bracket cache write
          ;;                 2) cache in hash on server, since need to do rmt: anyway to lock.
          (if (and rccachef *runconfigdat* (not (common:file-exists? rccachef)))
              (common:fail-safe
               (lambda ()
                 (configf:write-alist *runconfigdat* rccachef))
               (conc "Could not write cache file - "rccachef))
              )
          (if (and mtcachef *configdat*    (not (common:file-exists? mtcachef)))
              (common:fail-safe
               (lambda ()
                 (configf:write-alist *configdat* mtcachef))
               (conc "Could not write cache file - "mtcachef))
              )
          (if (and rccachef mtcachef *runconfigdat* *configdat*)
              (set! *configstatus* 'fulldata)))

	;; if have -append-config then read and append here
	(let ((cfname (args:get-arg "-append-config")))
	  (if (and cfname
		   (file-read-access? cfname))
	      (read-config cfname *configdat* #t))) ;; values are added to the hash, no need to do anything special.
	*toppath*)))

(define (get-best-disk confdat testconfig)
  (let* ((disks   (or (and testconfig (hash-table-ref/default testconfig "disks" #f))
		      (hash-table-ref/default confdat "disks" #f)))
	 (minspace (let ((m (configf:lookup confdat "setup" "minspace")))
		     (string->number (or m "10000")))))
    (if disks 
	(let ((res (common:get-disk-with-most-free-space disks minspace))) ;; min size of 1000, seems tad dumb
	  (if res
	      (cdr res)
	      (begin
;;		(if (common:low-noise-print 20 "No valid disks or no disk with enough space")
;;		    (debug:print-error 0 *default-log-port* "No valid disks found in megatest.config. Please add some to your [disks] section and ensure the directory exists and has enough space!\n    You can change minspace in the [setup] section of megatest.config. Current setting is: " minspace))
		;;(exit 1)
                 (if (null? disks)
                     (cons 1 (conc *toppath* "/runs"))
                     (let ((paths (sort disks (lambda (x y) (> (string-length (cadr x)) (string-length (cadr y)))))))
                       (let loop ((head (car paths)) (tail (cdr paths)))
                         (let ((result (handle-exceptions exn #f (create-directory (cadr head) #t))))
                           (if result
                               result
                               (if (null? tail)
                                   (cons 1 (conc *toppath* "/runs"))
                                   (loop (car tail) (cdr tail)))))))))))))) ;; the code creates the necessary directories if it does not exist and returns the path.


(define (launch:test-copy test-src-path test-path)
  (let* ((ovrcmd (let ((cmd (configf:lookup *configdat* "setup" "testcopycmd")))
		   (if cmd
		       ;; substitute the TEST_SRC_PATH and TEST_TARG_PATH
		       (string-substitute "TEST_TARG_PATH" test-path
					  (string-substitute "TEST_SRC_PATH" test-src-path cmd #t) #t)
		       #f)))
	 (cmd    (if ovrcmd 
		     ovrcmd
		     (conc "rsync -av" (if (debug:debug-mode 1) "" "q") " " test-src-path "/ " test-path "/"
			   " >> " test-path "/mt_launch.log 2>> " test-path "/mt_launch.log")))
	 (status (system cmd)))
    (if (not (eq? status 0))
	(debug:print 2 *default-log-port* "ERROR: problem with running \"" cmd "\""))))


;; Desired directory structure:
;;
;;  <linkdir> - <target> - <testname> -.
;;                                     |
;;                                     v
;;  <rundir>  -  <target>  -    <testname> -|- <itempath(s)>
;;
;;  dir stored in test is:
;; 
;;  <linkdir> - <target> - <testname> [ - <itempath> ]
;; 
;; All log file links should be stored relative to the top of link path
;;  
;; <target> - <testname> [ - <itempath> ] 
;;
(define (create-work-area run-id run-info keyvals test-id test-src-path disk-path testname itemdat #!key (remtries 2))
  (let* ((item-path (if (string? itemdat) itemdat (item-list->path itemdat))) ;; if pass in string - just use it
	 (runname   (if (string? run-info) ;; if we pass in a string as run-info use it as run-name.
			run-info
			(db:get-value-by-header (db:get-rows run-info)
						(db:get-header run-info)
						"runname")))
	 (contour   #f) ;; NOT READY FOR THIS (args:get-arg "-contour"))
	 ;; convert back to db: from rdb: - this is always run at server end
	 (target   (string-intersperse (map cadr keyvals) "/"))

	 (not-iterated  (equal? "" item-path))

	 ;; all tests are found at <rundir>/test-base or <linkdir>/test-base
	 (testtop-base (conc target "/" runname "/" testname))
	 (test-base    (conc testtop-base (if not-iterated "" "/") item-path))

	 ;; nb// if itempath is not "" then it is prefixed with "/"
	 (toptest-path (conc disk-path (if contour (conc "/" contour) "") "/" testtop-base))
	 (test-path    (conc disk-path (if contour (conc "/" contour) "") "/" test-base))

	 ;; ensure this exists first as links to subtests must be created there
	 (linktree  (common:get-linktree))
	 ;; WAS: (let ((rd (configf:lookup *configdat* "setup" "linktree")))
	 ;;         (if rd rd (conc *toppath* "/runs"))))
	 ;; which seems wrong ...

	 (lnkbase   (conc linktree (if contour (conc "/" contour) "") "/" target "/" runname))
	 (lnkpath   (conc lnkbase "/" testname))
	 (lnkpathf  (conc lnkpath (if not-iterated "" "/") item-path))
	 (lnktarget (conc lnkpath "/" item-path)))

    ;; Update the rundir path in the test record for all, rundir=physical, shortdir=logical
    ;;                                                 rundir   shortdir
    (rmt:general-call 'test-set-rundir-shortdir run-id lnkpathf test-path testname item-path run-id)

    (debug:print 2 *default-log-port* "INFO:\n       lnkbase=" lnkbase "\n       lnkpath=" lnkpath "\n  toptest-path=" toptest-path "\n     test-path=" test-path)
    (if (not (common:file-exists? linktree))
	(begin
	  (debug:print 0 *default-log-port* "WARNING: linktree did not exist! Creating it now at " linktree)
	  (create-directory linktree #t))) ;; (system (conc "mkdir -p " linktree))))
    ;; create the directory for the tests dir links, this is needed no matter what... try up to three times
    (let loop ((done 3)) 
      (let ((success (if (and (not (common:directory-exists? lnkbase))
			      (not (common:file-exists? lnkbase)))
			 (handle-exceptions
			  exn
			  (begin
			    (debug:print-error 0 *default-log-port* "Problem creating linktree base at " lnkbase)
			    (print-error-message exn (current-error-port))
			    #t)
			  (create-directory lnkbase #t)
			  #f))))
	(if (and (not success)(> done 0))
	    (loop (- done 1)))))
    
    ;; update the toptest record with its location rundir, cache the path
    ;; This wass highly inefficient, one db write for every subtest, potentially
    ;; thousands of unnecessary updates, cache the fact it was set and don't set it 
    ;; again. 

    ;; Now create the link from the test path to the link tree, however
    ;; if the test is iterated it is necessary to create the parent path
    ;; to the iteration. use pathname-directory to trim the path by one
    ;; level
    (if (not not-iterated) ;; i.e. iterated
	(let ((iterated-parent  (pathname-directory (conc lnkpath "/" item-path))))
	  (debug:print-info 2 *default-log-port* "Creating iterated parent " iterated-parent)
	  (handle-exceptions
	   exn
	   (begin
	     (debug:print-error 0 *default-log-port* " Failed to create directory " iterated-parent ((condition-property-accessor 'exn 'message) exn) ", exiting")
	     (exit 1))
	   (create-directory iterated-parent #t))))

    (if (symbolic-link? lnkpath) 
	(handle-exceptions
	 exn
	 (begin
	   (debug:print-error 0 *default-log-port* " Failed to remove symlink " lnkpath ((condition-property-accessor 'exn 'message) exn) ", exiting")
	   (exit 1))
	 (delete-file lnkpath)))

    (if (not (or (common:file-exists? lnkpath)
		 (symbolic-link? lnkpath)))
	(handle-exceptions
	 exn
	 (begin
	   (debug:print-error 0 *default-log-port* " Failed to create symlink " lnkpath ((condition-property-accessor 'exn 'message) exn) ", exiting")
	   (exit 1))
	 (create-symbolic-link toptest-path lnkpath)))
    
    ;; NB - This was not working right - some top tests are not getting the path set!!!
    ;;
    ;; Do the setting of this record after the paths are created so that the shortdir can 
    ;; be set to the real directory location. This is safer for future clean up if the link
    ;; tree is damaged or lost.
    ;; 
    (if (not (hash-table-ref/default *toptest-paths* testname #f))
	(let* ((testinfo       (rmt:get-test-info-by-id run-id test-id)) ;;  run-id testname item-path))
	       (curr-test-path (if testinfo ;; (filedb:get-path *fdb*
							     ;; (db:get-path dbstruct
				   ;; (rmt:sdb-qry 'getstr 
				   (db:test-get-rundir testinfo) ;; ) ;; )
				   #f)))
	  (hash-table-set! *toptest-paths* testname curr-test-path)
	  ;; NB// Was this for the test or for the parent in an iterated test?
	  (rmt:general-call 'test-set-rundir-shortdir run-id lnkpath 
			    (if (common:file-exists? lnkpath)
				;; (resolve-pathname lnkpath)
				(common:nice-path lnkpath)
				lnkpath)
			    testname "" run-id)
	  ;; (rmt:general-call 'test-set-rundir run-id lnkpath testname "") ;; toptest-path)
	  (if (or (not curr-test-path)
		  (not (directory-exists? toptest-path)))
	      (begin
		(debug:print-info 2 *default-log-port* "Creating " toptest-path " and link " lnkpath)
		(handle-exceptions
		 exn
		 #f ;; don't care to catch and deal with errors here for now.
		 (create-directory toptest-path #t))
		(hash-table-set! *toptest-paths* testname toptest-path)))))

    ;; The toptest path has been created, the link to the test in the linktree has
    ;; been created. Now, if this is an iterated test the real test dir must be created
    (if (not not-iterated) ;; this is an iterated test
	(begin ;; (let ((lnktarget (conc lnkpath "/" item-path)))
	  (debug:print 2 *default-log-port* "Setting up sub test run area")
	  (debug:print 2 *default-log-port* " - creating run area in " test-path)
	  (handle-exceptions
	   exn
	   (begin
	     (debug:print-error 0 *default-log-port* " Failed to create directory " test-path ((condition-property-accessor 'exn 'message) exn) ", exiting")
	     (exit 1))
	   (create-directory test-path #t))
	  (debug:print 2 *default-log-port* 
		       " - creating link from: " test-path "\n"
		       "                   to: " lnktarget)

	  ;; If there is already a symlink delete it and recreate it.
	  (handle-exceptions
	   exn
	   (begin
	     (debug:print-error 0 *default-log-port* " Failed to re-create link " lnktarget ((condition-property-accessor 'exn 'message) exn) ", exiting")
	     (exit))
	   (if (symbolic-link? lnktarget)     (delete-file lnktarget))
	   (if (not (common:file-exists? lnktarget)) (create-symbolic-link test-path lnktarget)))))

    (if (not (directory? test-path))
	(create-directory test-path #t)) ;; this is a hack, I don't know why out of the blue this path does not exist sometimes

    (if (and test-src-path (directory? test-path))
	(begin
	  (launch:test-copy test-src-path test-path)
	  (list lnkpathf lnkpath ))
	(if (and test-src-path (> remtries 0))
	    (begin
	      (debug:print-error 0 *default-log-port* "Failed to create work area at " test-path " with link at " lnktarget ", remaining attempts " remtries)
	      ;; 
	      (create-work-area run-id run-info keyvals test-id test-src-path disk-path testname itemdat remtries: (- remtries 1)))
	    (list #f #f)))))


(define (launch:handle-zombie-tests run-id)
  (let* ((key (conc "zombiescan-runid-"run-id))
         (now (current-seconds))
         (threshold (- (current-seconds)  (* 2 (or (configf:lookup-number *configdat* "setup" "deadtime") 120))))
         (val (rmt:get-var key))
         (do-scan?
          (cond
           ((not val)
            #t)
           ((< val threshold)
            #t)
           (else #f))))
    (when do-scan?
      (debug:print 1 *default-log-port* "INFO: search and mark zombie tests")
      (rmt:set-var key (current-seconds))
      (rmt:find-and-mark-incomplete run-id #f))))





;; 1. look though disks list for disk with most space
;; 2. create run dir on disk, path name is meaningful
;; 3. create link from run dir to megatest runs area 
;; 4. remotely run the test on allocated host
;;    - could be ssh to host from hosts table (update regularly with load)
;;    - could be netbatch
;;      (launch-test db (cadr status) test-conf))
(define (launch-test test-id run-id run-info keyvals runname test-conf test-name test-path itemdat params)
  (mutex-lock! *launch-setup-mutex*) ;; setting variables and processing the testconfig is NOT thread-safe, reuse the launch-setup mutex
  (let* ( ;; (lock-key        (conc "test-" test-id))
	;; (got-lock        (let loop ((lock        (rmt:no-sync-get-lock lock-key))
	;; 			     (expire-time (+ (current-seconds) 15))) ;; 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 launch test " keyvals " " runname " " test-name " " test-path)
	;; 			      (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))))))
	 (item-path       (item-list->path itemdat))
	 (contour         #f)) ;; NOT READY FOR THIS (args:get-arg "-contour")))
    (let loop ((delta        (- (current-seconds) *last-launch*))
	       (launch-delay (configf:lookup-number *configdat* "setup" "launch-delay" default: 1)))
      (if (> launch-delay delta)
	  (begin
	    (if (common:low-noise-print 1200 "test launch delay") ;; every two hours or so remind the user about launch delay.
		(debug:print-info 0 *default-log-port* "NOTE: test launches are delayed by " launch-delay " seconds. See megatest.config launch-delay setting to adjust.")) ;; launch of " test-name " for " (- launch-delay delta) " seconds"))
	    (thread-sleep! (- launch-delay delta))
	    (loop (- (current-seconds) *last-launch*) launch-delay))))
    (change-directory *toppath*)
    (alist->env-vars ;; consolidate this code with the code in megatest.scm for "-execute", *maybe* - the longer they are set the longer each launch takes (must be non-overlapping with the vars)
     (append
      (list
       (list "MT_RUN_AREA_HOME" *toppath*)
       (list "MT_TEST_NAME" test-name)
       (list "MT_RUNNAME"   runname)
       (list "MT_ITEMPATH"  item-path)
       (list "MT_CONTOUR"   contour)
       )
      itemdat))
    (let* ((tregistry       (tests:get-all)) ;; third param (below) is system-allowed
           ;; for tconfig, why do we allow fallback to test-conf?
	   (tconfig         (or (tests:get-testconfig test-name item-path tregistry #t force-create: #t)
				(begin
                                  (debug:print 0 *default-log-port* "WARNING: falling back to pre-calculated testconfig. This is likely not desired.")
                                  test-conf))) ;; force re-read now that all vars are set
	   (useshell        (let ((ush (configf:lookup *configdat* "jobtools"     "useshell")))
			      (if ush 
				  (if (equal? ush "no") ;; must use "no" to NOT use shell
				      #f
				      ush)
				  #t)))     ;; default is yes
	   (runscript       (configf:lookup tconfig   "setup"        "runscript"))
	   (ezsteps         (> (length (hash-table-ref/default tconfig "ezsteps" '())) 0)) ;; don't send all the steps, could be big, just send a flag
	   (subrun          (> (length (hash-table-ref/default tconfig "subrun"  '())) 0)) ;; send a flag to process a subrun
	   ;; (diskspace       (configf:lookup tconfig   "requirements" "diskspace"))
	   ;; (memory          (configf:lookup tconfig   "requirements" "memory"))
	   ;; (hosts           (configf:lookup *configdat* "jobtools"     "workhosts")) ;; I'm pretty sure this was never completed
	   (remote-megatest (configf:lookup *configdat* "setup" "executable"))
	   (run-time-limit  (or (configf:lookup  tconfig   "requirements" "runtimelim")
				(configf:lookup  *configdat* "setup" "runtimelim")))
	   ;; FIXME SOMEDAY: not good how this is so obtuse, this hack is to 
	   ;;                allow running from dashboard. Extract the path
	   ;;                from the called megatest and convert dashboard
	   ;;             	  or dboard to megatest
	   (local-megatest  (let* ((lm  (car (argv)))
				   (dir (pathname-directory lm))
				   (exe (pathname-strip-directory lm)))
			      (conc (if dir (conc dir "/") "")
				    (case (string->symbol exe)
				      ((dboard)    "../megatest")
				      ((mtest)     "../megatest")
				      ((dashboard) "megatest")
				      (else exe)))))
	   (launcher        (common:get-launcher *configdat* test-name item-path)) ;; (configf:lookup *configdat* "jobtools"     "launcher"))
	   (test-sig        (conc (common:get-testsuite-name) ":" test-name ":" item-path)) ;; (item-list->path itemdat))) ;; test-path is the full path including the item-path
	   (work-area       #f)
	   (toptest-work-area #f) ;; for iterated tests the top test contains data relevant for all
	   (diskpath   #f)
	   (cmdparms   #f)
	   (fullcmd    #f) ;; (define a (with-output-to-string (lambda ()(write x))))
	   (mt-bindir-path #f)
	   (testinfo   (rmt:get-test-info-by-id run-id test-id))
	   (mt_target  (string-intersperse (map cadr keyvals) "/"))
	   (debug-param (append (if (args:get-arg "-debug")  (list "-debug" (args:get-arg "-debug")) '())
				(if (args:get-arg "-logging")(list "-logging") '()))))
      ;; (if hosts (set! hosts (string-split hosts)))
      ;; set the megatest to be called on the remote host
      (if (not remote-megatest)(set! remote-megatest local-megatest)) ;; "megatest"))
      (set! mt-bindir-path (pathname-directory remote-megatest))
      (if launcher (set! launcher (string-split launcher)))
      ;; set up the run work area for this test
      (if (and (args:get-arg "-preclean") ;; user has requested to preclean for this run
	       (not (member (db:test-get-rundir testinfo)(list "n/a" "/tmp/badname")))) ;; n/a is a placeholder and thus not a read dir
	  (begin
	    (debug:print-info 0 *default-log-port* "attempting to preclean directory " (db:test-get-rundir testinfo) " for test " test-name "/" item-path)
	    (runs:remove-test-directory testinfo 'remove-data-only))) ;; remove data only, do not perturb the record
      
      ;; prevent overlapping actions - set to LAUNCHED as early as possible
      ;;
      ;; the following call handles waiver propogation. cannot yet condense into roll-up-pass-fail
      (tests:test-set-status! run-id test-id "LAUNCHED" "n/a" #f #f) ;; (if launch-results launch-results "FAILED"))
      (rmt:set-state-status-and-roll-up-items run-id test-name item-path #f "LAUNCHED" #f)
      ;; (pp (hash-table->alist tconfig))
      (set! diskpath (get-best-disk *configdat* tconfig))
      (if diskpath
	  (let ((dat  (create-work-area run-id run-info keyvals test-id test-path diskpath test-name itemdat)))
	    (set! work-area (car dat))
	    (set! toptest-work-area (cadr dat))
	    (debug:print-info 2 *default-log-port* "Using work area " work-area))
	  (begin
	    (set! work-area (conc test-path "/tmp_run"))
	    (create-directory work-area #t)
	    (debug:print 0 *default-log-port* "WARNING: No disk work area specified - running in the test directory under tmp_run")))
      (set! cmdparms (base64:base64-encode 
		      (z3:encode-buffer 
		       (with-output-to-string
			 (lambda () ;; (list 'hosts     hosts)
			   (write (list (list 'testpath  test-path)
					;; (list 'transport (conc *transport-type*))
					;; (list 'serverinf *server-info*)
					(list 'homehost  (let* ((hhdat (common:get-homehost)))
							   (if hhdat
							       (car hhdat)
							       #f)))
					(list 'serverurl (if *runremote*
							     (remote-server-url *runremote*)
							     #f)) ;;
					(list 'areaname  (common:get-testsuite-name))
					(list 'toppath   *toppath*)
					(list 'work-area work-area)
					(list 'test-name test-name) 
					(list 'runscript runscript) 
					(list 'run-id    run-id   )
					(list 'test-id   test-id  )
					;; (list 'item-path item-path )
					(list 'itemdat   itemdat  )
					(list 'megatest  remote-megatest)
					(list 'ezsteps   ezsteps)
					(list 'subrun    subrun)
					(list 'target    mt_target)
					(list 'contour   contour)
					(list 'runtlim   (if run-time-limit (common:hms-string->seconds run-time-limit) #f))
					(list 'env-ovrd  (hash-table-ref/default *configdat* "env-override" '())) 
					(list 'set-vars  (if params (hash-table-ref/default params "-setvars" #f)))
					(list 'runname   runname)
					(list 'mt-bindir-path mt-bindir-path))))))))
      
      ;; clean out step records from previous run if they exist
      ;; (rmt:delete-test-step-records run-id test-id)
      ;; if the dir does not exist we may have a itempath where individual variables are a path, launch anyway
      (if (common:file-exists? work-area)
	  (change-directory work-area)) ;; so that log files from the launch process don't clutter the test dir
      (cond
       ;; ((and launcher hosts) ;; must be using ssh hostname
       ;;    (set! fullcmd (append launcher (car hosts)(list remote-megatest "-m" test-sig "-execute" cmdparms) debug-param)))
       ;; (set! fullcmd (append launcher (car hosts)(list remote-megatest test-sig "-execute" cmdparms))))
       (launcher
	(set! fullcmd (append launcher (list remote-megatest "-m" test-sig "-execute" cmdparms) debug-param)))
       ;; (set! fullcmd (append launcher (list remote-megatest test-sig "-execute" cmdparms))))
       (else
	(if (not useshell)(debug:print 0 *default-log-port* "WARNING: internal launching will not work well without \"useshell yes\" in your [jobtools] section"))
	(set! fullcmd (append (list remote-megatest "-m" test-sig "-execute" cmdparms) debug-param (list (if useshell "&" ""))))))
      ;; (set! fullcmd (list remote-megatest test-sig "-execute" cmdparms (if useshell "&" "")))))
      (if (args:get-arg "-xterm")(set! fullcmd (append fullcmd (list "-xterm"))))
      (debug:print 1 *default-log-port* "Launching " work-area)
      ;; set pre-launch-env-vars before launching, keep the vars in prevvals and put the envionment back when done
      (debug:print 4 *default-log-port* "fullcmd: " fullcmd)
      (set! *last-launch* (current-seconds)) ;; all that junk above takes time, set this as late as possible.
      (let* ((commonprevvals (alist->env-vars
			      (hash-table-ref/default *configdat* "env-override" '())))
	     (miscprevvals   (alist->env-vars ;; consolidate this code with the code in megatest.scm for "-execute"
			      (append (list (list "MT_TEST_RUN_DIR" work-area)
					    (list "MT_TEST_NAME" test-name)
					    (list "MT_ITEM_INFO" (conc itemdat)) 
					    (list "MT_RUNNAME"   runname)
					    (list "MT_TARGET"    mt_target)
					    (list "MT_ITEMPATH"  item-path)
					    )
				      itemdat)))
	     (testprevvals   (alist->env-vars
			      (hash-table-ref/default tconfig "pre-launch-env-overrides" '())))
	     ;; Launchwait defaults to true, must override it to turn off wait
	     (launchwait     (if (equal? (configf:lookup *configdat* "setup" "launchwait") "no") #f #t))
	     (launch-results-prev (apply (if launchwait ;; BB: TODO: refactor this to examine return code of launcher, if nonzero, set state to launch failed.
					process:cmd-run-with-stderr-and-exitcode->list
					process-run)
				    (if useshell
					(let ((cmdstr (string-intersperse fullcmd " ")))
					  (if launchwait
					      cmdstr
					      (conc cmdstr " >> mt_launch.log 2>&1 &")))
					(car fullcmd))
				    (if useshell
					'()
					(cdr fullcmd))))
             (success        (if launchwait (equal? 0 (cadr launch-results-prev)) #t))
             (launch-results (if launchwait (car launch-results-prev) launch-results-prev)))
        (if (not success)
            (tests:test-set-status! run-id test-id "COMPLETED" "DEAD" "launcher failed; exited non-zero; check mt_launch.log" #f)) ;; (if launch-results launch-results "FAILED"))
        (mutex-unlock! *launch-setup-mutex*) ;; yes, really should mutex all the way to here. Need to put this entire process into a fork.
	;; (rmt:no-sync-del! lock-key)         ;; release the lock for starting this test
	(if (not launchwait) ;; give the OS a little time to allow the process to start
	    (thread-sleep! 0.01))
	(with-output-to-file "mt_launch.log"
	  (lambda ()
	    (print "LAUNCHCMD: " (string-intersperse fullcmd " "))
	    (if (list? launch-results)
		(apply print launch-results)
		(print "NOTE: launched \"" fullcmd "\"\n  but did not wait for it to proceed. Add the following to megatest.config \n[setup]\nlaunchwait yes\n  if you have problems with this"))
	    #:append))
	(debug:print 2 *default-log-port* "Launching completed, updating db")
	(debug:print 2 *default-log-port* "Launch results: " launch-results)
	(if (not launch-results)
	    (begin
	      (print "ERROR: Failed to run " (string-intersperse fullcmd " ") ", exiting now")
	      ;; (sqlite3:finalize! db)
	      ;; good ole "exit" seems not to work
	      ;; (_exit 9)
	      ;; but this hack will work! Thanks go to Alan Post of the Chicken email list
	      ;; NB// Is this still needed? Should be safe to go back to "exit" now?
	      (process-signal (current-process-id) signal/kill)
	      ))
	(alist->env-vars miscprevvals)
	(alist->env-vars testprevvals)
	(alist->env-vars commonprevvals)
	launch-results))
    (change-directory *toppath*)))

;; recover a test where the top controlling mtest may have died
;;
(define (launch:recover-test run-id test-id)
  ;; this function is called on the test run host via ssh
  ;;
  ;; 1. look at the process from pid
  ;;    - is it owned by calling user
  ;;    - it it's run directory correct for the test
  ;;    - is there a controlling mtest (maybe stuck)
  ;; 2. if recovery is needed watch pid
  ;;    - when it exits take the exit code and do the needful
  ;;
  (let* ((pid (rmt:test-get-top-process-pid run-id test-id))
	 (psres (with-input-from-pipe
		 (conc "ps -F -u " (current-user-name) " | grep -E '" pid " ' | grep -v 'grep -E " pid "'")
		 (lambda ()
		   (read-line))))
	 (rundir (if (string? psres) ;; real process owned by user
		     (read-symbolic-link (conc "/proc/" pid "/cwd"))
		     #f)))
    ;; now wait on that process if all is correct
    ;; periodically update the db with runtime
    ;; when the process exits look at the db, if still RUNNING after 10 seconds set
    ;; state/status appropriately
    (process-wait pid)))

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

;;======================================================================

(declare (unit launchmod))
(declare (uses commonmod))





(module launchmod
	*
	
(import scheme chicken data-structures extras)
(import (prefix sqlite3 sqlite3:) posix typed-records srfi-18 srfi-69 format ports srfi-1 matchable)



(import commonmod)





;; (use (prefix ulex ulex:))

(include "common_records.scm")









































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































)

;; (include "vg_records.scm")

;;======================================================================
;; L O C K I N G   M E C H A N I S M S 
;;======================================================================

;; globals
(define *writes-total-delay* 0)
(define *exit-started* #f)
(define *last-monitor-update-time* 0)
(define *default-log-port* (current-error-port))

;; The watchdog is to keep an eye on things like db sync etc.
;;

;; (include "api-inc.scm")       ;; L3 
(include "archive-inc.scm")  
(include "client-inc.scm")
(include "common-inc.scm")    ;; L5
(include "db-inc.scm")        ;; L4
(include "env-inc.scm")
(include "http-transport-inc.scm")
;; (include "items-inc.scm")
;; (include "keys-inc.scm")
(include "launch-inc.scm")     ;; L1
;; (include "margs-inc.scm")
(include "mt-inc.scm")
(include "ods-inc.scm")        ;; L1
(include "pgdb-inc.scm")

;;======================================================================

(declare (unit mtmod))
(declare (uses mtconfigf))
(declare (uses commonmod))
(declare (uses rmtmod))
(declare (uses testsmod))







(module mtmod
	*
	
(import scheme chicken data-structures extras posix)

(import (prefix mtconfigf configf:)


	commonmod
	rmtmod
	testsmod)










;; This is the Megatest API. All generally "useful" routines will be wrapped or extended
;; here.

;;======================================================================
;;  R U N S

;; watchdog and exit procedures
;;======================================================================

;;======================================================================
;; E X I T   H A N D L I N G
;;======================================================================

(define (common:run-sync?)
    (and (common:on-homehost?)
	 (args:get-arg "-server")))

;;   (let ((ohh (common:on-homehost?))
;; 	(srv (args:get-arg "-server")))
;;     (and ohh srv)))
    ;; (debug:print-info 0 *default-log-port* "common:run-sync? ohh=" ohh ", srv=" srv)

(define *watchdog* (make-thread
		    (lambda ()

   ;; (if full
       '(dejunk)
       ;; '())
       )
  (if (common:api-changed?)
      (common:set-last-run-version)))
























































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































)
	

;;======================================================================
;;  S U P P O R T   F U N C T I O N S
;;======================================================================

;; if a server is either running or in the process of starting call client:setup
;; else return #f to let the calling proc know that there is no server available
;;
(define (rmt:get-connection-info areapath #!key (area-dat #f)) ;; TODO: push areapath down.
  (let* ((runremote (or area-dat *runremote*))
	 (cinfo     (if (remote? runremote)
			(remote-conndat runremote)
			#f)))
	  (if cinfo
	      cinfo
	      (if (server:check-if-running areapath)

;; 
;;      ;; not on homehost, do server query
;;      (else (extras-case-11 *default-log-port* runremote cmd params attemptnum rid)))))

;; bunch of small functions factored out of send-receive to make debug easier
;;

(define (extras-case-11 *default-log-port* runremote cmd params attemptnum rid)
  ;; (mutex-unlock! *rmt-mutex*)
  (debug:print-info 12 *default-log-port* "rmt:send-receive, case  9")
  ;; (mutex-lock! *rmt-mutex*)
  (let* ((conninfo (remote-conndat runremote))
	 (dat      (case (remote-transport runremote)
		     ((http) (condition-case ;; handling here has
					     ;; caused a lot of
					     ;; problems. However it
					     ;; is needed to deal with
					     ;; attemtped
					     ;; communication to
					     ;; servers that have gone
					     ;; away
			      (http-transport:client-api-send-receive 0 conninfo cmd params)
			      ((commfail)(vector #f "communications fail"))
			      ((exn)(vector #f "other fail" (print-call-chain)))))
		     (else
		      (debug:print 0 *default-log-port* "ERROR: transport " (remote-transport runremote) " not supported")
		      (exit))))
	 (success  (if (vector? dat) (vector-ref dat 0) #f))
	 (res      (if (vector? dat) (vector-ref dat 1) #f)))
    (if (and (vector? conninfo) (< 5 (vector-length conninfo)))
	(http-transport:server-dat-update-last-access conninfo) ;; refresh access time
	(begin
	  (debug:print 0 *default-log-port* "INFO: Should not get here! conninfo=" conninfo)
	  (set! conninfo #f)
	  (remote-conndat-set! *runremote* #f) ;; NOTE: *runremote* is global copy of runremote. Purpose: factor out global.
	  (http-transport:close-connections  area-dat: runremote)))
    (debug:print-info 13 *default-log-port* "rmt:send-receive, case  9. conninfo=" conninfo " dat=" dat " runremote = " runremote)
    (mutex-unlock! *rmt-mutex*)
    (if success ;; success only tells us that the transport was
	;; successful, have to examine the data to see if
	;; there was a detected issue at the other end
	(extras-transport-succeded *default-log-port* *rmt-mutex* attemptnum runremote res params rid cmd)
	(extras-transport-failed *default-log-port* *rmt-mutex* attemptnum runremote cmd rid params)

    res))

(define (rmt:send-receive-no-auto-client-setup connection-info cmd run-id params)
  (let* ((run-id   (if run-id run-id 0))
	 (res  	   (handle-exceptions
		    exn
		    #f
		    (http-transport:client-api-send-receive run-id connection-info cmd params))))
    (if (and res (vector-ref res 0))
	(vector-ref res 1) ;;; YES!! THIS IS CORRECT!! CHANGE IT HERE, THEN CHANGE rmt:send-receive ALSO!!!
	#f)))

;; ;; Wrap json library for strings (why the ports crap in the first place?)
;; (define (rmt:dat->json-str dat)
;;   (with-output-to-string 

  (debug:print 0 log-port "WARNING: write transaction requested on a readonly area.  cmd="cmd" params="params)
  #f)

(define (extras-transport-failed *default-log-port* *rmt-mutex* attemptnum runremote cmd rid params)
  (debug:print 0 *default-log-port* "WARNING: communication failed. Trying again, try num: " attemptnum)
  (mutex-lock! *rmt-mutex*)
  (remote-conndat-set!    runremote #f)
  (http-transport:close-connections area-dat: runremote)
  (remote-server-url-set! runremote #f)
  (mutex-unlock! *rmt-mutex*)
  (debug:print-info 12 *default-log-port* "rmt:send-receive, case  9.1")
  (rmt:send-receive cmd rid params attemptnum: (+ attemptnum 1)))
  
(define (extras-transport-succeded *default-log-port* *rmt-mutex* attemptnum runremote res params rid cmd)
  (if (and (vector? res)

						 ;; server is
						 ;; overloaded and we
						 ;; want to ease off
						 ;; the queries
      (let ((wait-delay (+ attemptnum (* attemptnum 10))))
	(debug:print 0 *default-log-port* "WARNING: server is overloaded. Delaying " wait-delay " seconds and trying call again.")
	(mutex-lock! *rmt-mutex*)
	(http-transport:close-connections area-dat: runremote)
	(set! *runremote* #f) ;; force starting over
	(mutex-unlock! *rmt-mutex*)
	(thread-sleep! wait-delay)
	(rmt:send-receive cmd rid params attemptnum: (+ attemptnum 1)))
      res)) ;; All good, return res

;; (include "common_records.scm")
)

;;     along with Megatest.  If not, see <http://www.gnu.org/licenses/>.

;;======================================================================

(declare (unit servermod))
(declare (uses commonmod))
(declare (uses dbmod))




(module servermod
	*
	
(import scheme chicken data-structures extras)
(import (prefix sqlite3 sqlite3:) posix typed-records srfi-18 srfi-69 format ports srfi-1 matchable)



(import commonmod)
(import dbmod)




;; (use (prefix ulex ulex:))

(include "common_records.scm")

(define (server:make-server-url hostport)
  (if (not hostport)

(define (server:record->url servr)
  (match-let (((mod-time host port start-time pid)
	       servr))
    (if (and host port)
	(conc host ":" port)
	#f)))


;; timeout is hms string: 1h 5m 3s, default is 1 minute
;;
(define (server:expiration-timeout)
  (let ((tmo (configf:lookup *configdat* "server" "timeout")))
    (if (and (string? tmo)
	     (common:hms-string->seconds tmo)) ;; BUG: hms-string->seconds is broken, if given "10" returns 0. Also, it doesn't belong in this logic unless the string->number is changed below
        (* 3600 (string->number tmo))
	60)))

;; ping the given server
;;
(define (server:check-server server-record)
  (let* ((server-url (server:record->url server-record))
         (res        (case *transport-type*
                       ((http)(server:ping server-url))
                       ;; ((nmsg)(nmsg-transport:ping (tasks:hostinfo-get-interface server)
                       )))
    (if res
        server-url
	#f)))

;; no longer care if multiple servers are started by accident. older servers will drop off in time.
;;
(define (server:check-if-running areapath) ;;  #!key (numservers "2"))
  (let* ((ns            (server:get-num-servers))
	 (servers       (server:get-best (server:get-list areapath))))
    ;; (print "servers: " servers " ns: " ns)
    (if (or (and servers
		 (null? servers))
	    (not servers)
	    (and (list? servers)
		 (< (length servers) (random ns)))) ;; somewhere between 0 and numservers
        #f
        (let loop ((hed (car servers))
                   (tal (cdr servers)))
          (let ((res (server:check-server hed)))
            (if res
                res
                (if (null? tal)
                    #f
                    (loop (car tal)(cdr tal)))))))))

;;======================================================================
;; P K T S   S T U F F 
;;======================================================================

;; ???

;;======================================================================
;; P K T S   S T U F F 
;;======================================================================

;; ???

;;======================================================================
;; S E R V E R
;;======================================================================

;; Call this to start the actual server
;;

;; all routes though here end in exit ...
;;
;; start_server
;;
(define (server:launch run-id transport-type)
  (case transport-type
    ((http)(http-transport:launch))
    ;;((nmsg)(nmsg-transport:launch run-id))
    ;;((rpc)  (rpc-transport:launch run-id))
    (else (debug:print-error 0 *default-log-port* "unknown server type " transport-type))))

;;======================================================================
;; S E R V E R   U T I L I T I E S 
;;======================================================================

;; Get the transport
(define (server:get-transport)
  (if *transport-type*

    ((rpc)  (db:obj->string (vector success/fail query-sig result)))
    ((http) (db:obj->string (vector success/fail query-sig result)))
    ((fs)   result)
    (else 
     (debug:print-error 0 *default-log-port* "unrecognised transport type: " *transport-type*)
     result)))

;; Given a run id start a server process    ### NOTE ### > file 2>&1 
;; if the run-id is zero and the target-host is set 
;; try running on that host
;;   incidental: rotate logs in logs/ dir.
;;
(define  (server:run areapath) ;; areapath is *toppath* for a given testsuite area
  (let* ((curr-host   (get-host-name))
         ;; (attempt-in-progress (server:start-attempted? areapath))
         ;; (dot-server-url (server:check-if-running areapath))
	 (curr-ip     (server:get-best-guess-address curr-host))
	 (curr-pid    (current-process-id))
	 (homehost    (common:get-homehost)) ;; configf:lookup *configdat* "server" "homehost" ))
	 (target-host (car homehost))
	 (testsuite   (common:get-testsuite-name))
	 (logfile     (conc areapath "/logs/server.log")) ;; -" curr-pid "-" target-host ".log"))
	 (cmdln (conc (common:get-megatest-exe)
		      " -server " (or target-host "-") (if (equal? (configf:lookup *configdat* "server" "daemonize") "yes")
							   " -daemonize "
							   "")
		      ;; " -log " logfile
		      " -m testsuite:" testsuite)) ;; (conc " >> " logfile " 2>&1 &")))))
	 (log-rotate  (make-thread common:rotate-logs  "server run, rotate logs thread"))
         (load-limit  (configf:lookup-number *configdat* "jobtools" "max-server-start-load" default: 3.0)))
    ;; we want the remote server to start in *toppath* so push there
    (push-directory areapath)
    (debug:print 0 *default-log-port* "INFO: Trying to start server (" cmdln ") ...")
    (thread-start! log-rotate)
    
    ;; host.domain.tld match host?
    (if (and target-host 
	     ;; look at target host, is it host.domain.tld or ip address and does it 
	     ;; match current ip or hostname
	     (not (string-match (conc "("curr-host "|" curr-host"\\..*)") target-host))
	     (not (equal? curr-ip target-host)))
	(begin
	  (debug:print-info 0 *default-log-port* "Starting server on " target-host ", logfile is " logfile)
	  (setenv "TARGETHOST" target-host)))
      
    (setenv "TARGETHOST_LOGF" logfile)
    (thread-sleep! (/ (random 5000) 1000)) ;; add about a random (up to 5 seconds) initial delay. It seems pretty common that many running tests request a server at the same time
    (common:wait-for-normalized-load load-limit " delaying server start due to load" target-host) ;; do not try starting servers on an already overloaded machine, just wait forever
    (system (conc "nbfake " cmdln))
    (unsetenv "TARGETHOST_LOGF")
    (if (get-environment-variable "TARGETHOST")(unsetenv "TARGETHOST"))
    (thread-join! log-rotate)
    (pop-directory)))

(define (server:get-client-signature) 
  ;; (if *my-client-signature* *my-client-signature*
  ;; (let ((sig
  (server:mk-signature)) ;; )
;; (set! *my-client-signature* sig)
;;        *my-client-signature*)))

;; kind start up of servers, wait 40 seconds before allowing another server for a given
;; run-id to be launched
(define (server:kind-run areapath)
  (if (not (server:check-if-running areapath)) ;; why try if there is already a server running?
      (let* ((last-run-dat (hash-table-ref/default *server-kind-run* areapath '(0 0))) ;; callnum, whenrun
	     (call-num     (car last-run-dat))
	     (when-run     (cadr last-run-dat))
	     (run-delay    (+ (case call-num
				((0)    0)
				((1)   20)
				((2)  300)
				(else 600))
			      (random 5)))   ;; add a small random number just in case a lot of jobs hit the work hosts simultaneously
	     (lock-file    (conc areapath "/logs/server-start.lock")))
	(if	(> (- (current-seconds) when-run) run-delay)
		(begin
		  (common:simple-file-lock-and-wait lock-file expire-time: 15)
		  (server:run areapath)
		  (thread-sleep! 2) ;; don't release the lock for at least a few seconds
		  (common:simple-file-release-lock lock-file)))
	(hash-table-set! *server-kind-run* areapath (list (+ call-num 1)(current-seconds))))))

(define (server:start-and-wait areapath #!key (timeout 60))
  (let ((give-up-time (+ (current-seconds) timeout)))
    (let loop ((server-url (server:check-if-running areapath))
	       (try-num    0))
      (if (or server-url
	      (> (current-seconds) give-up-time)) ;; server-url will be #f if no server available.
	  server-url
	  (let ((num-ok (length (server:get-best (server:get-list areapath)))))
	    (if (and (> try-num 0)  ;; first time through simply wait a little while then try again
		     (< num-ok 1))  ;; if there are no decent candidates for servers then try starting a new one
		(server:kind-run areapath))
	    (thread-sleep! 5)
	    (loop (server:check-if-running areapath)
		  (+ try-num 1)))))))

(define server:try-running server:run) ;; there is no more per-run servers ;; REMOVE ME. BUG.

(define (server:kill servr)
  (match-let (((mod-time hostname port start-time pid)
	       servr))
    (tasks:kill-server hostname pid)))

;; called in megatest.scm, host-port is string hostname:port
;;
;; NOTE: This is NOT called directly from clients as not all transports support a client running
;;       in the same process as the server.
;;
(define (server:ping host-port-in #!key (do-exit #f))
  (let ((host:port (if (not host-port-in) ;; use read-dotserver to find
		       #f ;; (server:check-if-running *toppath*)
		;; (if (number? host-port-in) ;; we were handed a server-id
		;; 	   (let ((srec (tasks:get-server-by-id (db:delay-if-busy (tasks:open-db)) host-port-in)))
		;; 	     ;; (print "srec: " srec " host-port-in: " host-port-in)
		;; 	     (if srec
		;; 		 (conc (vector-ref srec 3) ":" (vector-ref srec 4))
		;; 		 (conc "no such server-id " host-port-in)))
		       host-port-in))) ;; )
    (let* ((host-port (if host:port
			  (let ((slst (string-split   host:port ":")))
			    (if (eq? (length slst) 2)
				(list (car slst)(string->number (cadr slst)))
				#f))
			  #f)))
;;	   (toppath       (launch:setup)))
      ;; (print "host-port=" host-port)
      (if (not host-port)
	  (begin
	    (if host-port-in
		(debug:print 0 *default-log-port*  "ERROR: bad host:port"))
	    (if do-exit (exit 1))
	    #f)
	  (let* ((iface      (car host-port))
		 (port       (cadr host-port))
		 (server-dat (http-transport:client-connect iface port))
		 (login-res  (rmt:login-no-auto-client-setup server-dat)))
	    (if (and (list? login-res)
		     (car login-res))
		(begin
		  ;; (print "LOGIN_OK")
		  (if do-exit (exit 0))
		  #t)
		(begin
		  ;; (print "LOGIN_FAILED")
		  (if do-exit (exit 1))
		  #f)))))))

;; run ping in separate process, safest way in some cases
;;
(define (server:ping-server ifaceport)
  (with-input-from-pipe 
   (conc (common:get-megatest-exe) " -ping " ifaceport)
   (lambda ()
     (let loop ((inl (read-line))
		(res "NOREPLY"))
       (if (eof-object? inl)
	   (case (string->symbol res)
	     ((NOREPLY)  #f)
	     ((LOGIN_OK) #t)
	     (else       #f))
	   (loop (read-line) inl))))))

;; NOT USED (well, ok, reference in rpc-transport but otherwise not used).
;;
(define (server:login toppath)
  (lambda (toppath)
    (set! *db-last-access* (current-seconds)) ;; might not be needed.
    (if (equal? *toppath* toppath)
	#t
	#f)))

;; (define server:sync-lock-token "SERVER_SYNC_LOCK")
;; (define (server:release-sync-lock)
;;   (db:no-sync-del! *no-sync-db* server:sync-lock-token))
;; (define (server:have-sync-lock?)
;;   (let* ((have-lock-pair (db:no-sync-get-lock *no-sync-db* server:sync-lock-token))
;;          (have-lock?     (car have-lock-pair))
;;          (lock-time      (cdr have-lock-pair))

;;     but WITHOUT ANY WARRANTY; without even the implied warranty of
;;     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
;;     GNU General Public License for more details.
;; 
;;     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')

(define (subrun:subrun-test-initialized? test-run-dir)
  (if (and (common:file-exists? (conc test-run-dir "/subrun-area") )
           (common:file-exists? (conc test-run-dir "/testconfig.subrun") ))
      #t
      #f))

(define (subrun:launch-dashboard test-run-dir)
  (if (subrun:subrun-test-initialized? test-run-dir)
      (let* ((subarea (subrun:get-runarea test-run-dir)))
        (if (and subarea (common:file-exists? subarea))
            (system (conc "cd " subarea ";env -i PATH=$PATH DISPLAY=$DISPLAY HOME=$HOME USER=$USER dashboard &"))))))

(define (subrun:subrun-removed? test-run-dir)
  (if (subrun:subrun-test-initialized? test-run-dir)
      (let ((flagfile (conc test-run-dir "/subrun.removed")))
        (if (common:file-exists? flagfile)
            #t
            #f))
      #t))

(define (subrun:set-subrun-removed test-run-dir)
  (let ((flagfile (conc test-run-dir "/subrun.removed")))
    (if (and (subrun:subrun-test-initialized? test-run-dir) (not (common:file-exists? flagfile)))
        (with-output-to-file flagfile
          (lambda () (print (current-seconds)))))))

(define (subrun:unset-subrun-removed test-run-dir)
  (let ((flagfile (conc test-run-dir "/subrun.removed")))
    (if (and (subrun:subrun-test-initialized? test-run-dir) (common:file-exists? flagfile))
        (delete-file flagfile))))


(define (subrun:testconfig-defines-subrun? testconfig)
  (configf:lookup testconfig "subrun" "runwait")) ;; we use runwait as the flag that a subrun is requested

(define (subrun:initialize-toprun-test  testconfig test-run-dir)
  (let ((ra (configf:lookup testconfig "subrun" "run-area"))
        (logpro (configf:lookup testconfig "subrun" "logpro"))
        (symlink-target (conc test-run-dir "/subrun-area"))
        )
    (if (not ra)      ;; when runarea is not set we default to *toppath*. However 
	(let ((fallback-run-area (or *toppath* (conc test-run-dir "/subrun"))))
	  ;; we need to force the setting in the testconfig so it will
          ;; be preserved in the testconfig.subrun file
	  (configf:set-section-var testconfig "subrun" "run-area" fallback-run-area)
	  (set! ra fallback-run-area)))
    (configf:set-section-var testconfig "logpro" "subrun" logpro) ;; append the logpro rules to the logpro section as stepname subrun
    (if (common:file-exists? symlink-target)
        (delete-file symlink-target))
    (create-symbolic-link ra symlink-target)
    (configf:write-alist testconfig "testconfig.subrun")))

(define (subrun:set-state-status test-run-dir state status new-state-status)
  (if (and (not (subrun:subrun-removed? test-run-dir)) (subrun:subrun-test-initialized? test-run-dir))
      (let* ((action-switches-str
              (conc "-set-state-status "new-state-status
                    (if state (conc " -state "state) "")
                    (if status (conc " -status "status) "")))
             (log-prefix
              (subrun:sanitize-path
               (conc "set-state-status="new-state-status
                     (if state (conc ":state="state) "")
                     (if status (conc "+status="status) ""))))
             (submt-result 
              (subrun:exec-sub-megatest test-run-dir action-switches-str log-prefix)))
        submt-result)))

(define (subrun:remove-subrun test-run-dir keep-records )
  (if (and (not (subrun:subrun-removed? test-run-dir)) (subrun:subrun-test-initialized? test-run-dir))
      (let* ((action-switches-str
              (conc "-remove-runs"
                    (if keep-records "-keep-records " "")
                    ))
             (remove-result
              (subrun:exec-sub-megatest test-run-dir action-switches-str "remove")))
        (if remove-result
            (begin
              (subrun:set-subrun-removed test-run-dir)
              #t)
            #f))
      #t))

(define (subrun:kill-subrun test-run-dir )
  (if (and (not (subrun:subrun-removed? test-run-dir)) (subrun:subrun-test-initialized? test-run-dir))
      (let* ((action-switches-str
              (conc "-kill-runs" ))
             (kill-result
              (subrun:exec-sub-megatest test-run-dir action-switches-str "kill")))
        kill-result)
      #t))

(define (subrun:launch-cmd test-run-dir #!optional (sub-cmd "-run")) ;; BUG: "-run" should be changed to "-rerun-clean" but current doesn't work
  (if (subrun:subrun-removed? test-run-dir)
      (subrun:unset-subrun-removed test-run-dir))      

  (let* ((log-prefix "run")
         (switches (subrun:selector+log-switches test-run-dir log-prefix))
         (run-wait #t)
         (cmd      (conc "megatest " sub-cmd " " switches" "
                         (if run-wait "-run-wait " ""))))
    cmd))


(define (subrun:sanitize-path inpath)
  (let* ((insane-pattern (irregex "[^[a-zA-Z0-9_\\-]")))
    (regex#string-substitute insane-pattern "_" inpath #t)))

(define (subrun:get-runarea test-run-dir)
  (if (subrun:subrun-test-initialized? test-run-dir)
      (let* ((info-alist (subrun:selector+log-alist
                          test-run-dir
                          "foo"))
             (run-area   (if (list? info-alist)
                             (alist-ref "-start-dir" info-alist equal? #f)
                             #f)))
        run-area)
      #f))

(define (subrun:selector+log-alist test-run-dir log-prefix)
  (let* ((switch-def-alist (common:get-param-mapping flavor: 'config))
         (subrunfile   (conc test-run-dir "/testconfig.subrun" ))
         (subrundata   (with-input-from-file subrunfile read))
         (subrunconfig (configf:alist->config subrundata))
         (run-area     (configf:lookup subrunconfig "subrun" "run-area"))
         (defvals      `(("start-dir" . ,(or run-area  ;; default values if not specified in subrun section of tconf
                                             (get-environment-variable "MT_RUN_AREA_HOME")
                                             "/no/rundir/found")) 
                         ("run-name"  . ,(or (get-environment-variable "MT_RUNNAME") "NO-RUNNAME"))
                         ("target"    . ,(or (get-environment-variable "MT_TARGET")  "NO-TARGET"))))
         (switch-alist-pre  (filter-map (lambda (item)
                                          (let* ((config-key (car item))
                                                 (switch     (cdr item))
                                                 (defval     (alist-ref config-key defvals equal? #f))
                                                 (val        (or (configf:lookup subrunconfig "subrun" config-key)
                                                                 defval)))
                                            (if val
                                                (cons switch val)
                                                #f)))
                                        switch-def-alist))

         ;; testpatt may be modified if all three of mode-patt, tag-expr, and testpatt are null
         (mode-patt     (alist-ref "-modepatt" switch-alist-pre equal? #f))
         (tag-expr      (alist-ref "-tagexpr" switch-alist-pre equal? #f))
         (testpatt      (alist-ref "-testpatt" switch-alist-pre equal?
                                   (if (not (or mode-patt tag-expr)) "%" #f))) ;; testpatt is % if not
                                                                               ;; otherwise specified

         ;; define compact-stem for logfile
         (target        (alist-ref "-target" switch-alist-pre equal? #f)) ;; want data-structures alist-ref, not alist-lib alist-ref
         (runname       (alist-ref "-runname" switch-alist-pre equal? #f))


         (compact-stem  (subrun:sanitize-path
                         (conc
                          target
                          "-"
                          runname
                          "-" (or testpatt mode-patt tag-expr "NO-TESTPATT"))))
         (logfile       (conc
                         test-run-dir "/"
                         (if log-prefix
                             (conc (subrun:sanitize-path log-prefix) "-")
                             "")
                         compact-stem
                         ".log"))
         ;; swap out testpatt with modified test-patt and add -log
         (switch-alist  (cons
                         (cons "-log" logfile)
                         (map (lambda (item)
                                (if (equal? (car item) "-testpatt")
                                    (cons "-testpatt" testpatt)
                                    item))
                                switch-alist-pre))))
    switch-alist))
    ;; note - get precmd from subrun section
    ;;   apply to submegatest commands

(define (subrun:get-log-path test-run-dir log-prefix)
  (let* ((alist (subrun:selector+log-alist test-run-dir log-prefix))
         (res   (alist-ref "-log" alist equal? #f)))
    res))

(define (subrun:selector+log-switches test-run-dir log-prefix)
  (let* ((switch-alist (subrun:selector+log-alist test-run-dir log-prefix))
         (res
          (string-intersperse
           (apply
            append
            (map
             (lambda (x)
               (list (car x) (cdr x)))
             switch-alist))
           " ")))
    res))

(define (subrun:exec-sub-megatest test-run-dir action-switches-str log-prefix)
  (let* ((selector-switches  (subrun:selector+log-switches test-run-dir log-prefix))
         (cmd (conc "megatest " selector-switches " " action-switches-str ))
         (pid #f)
         (proc (lambda ()
                 (debug:print-info 0 *default-log-port* "Running sub megatest command: "cmd)
                 ;;(set! pid (process-run "/usr/bin/xterm" (list ))))))
                 (set! pid (process-run "/bin/bash" (list "-c" cmd))))))
    (call-with-environment-variables 
     (list (cons "PATH" (conc (get-environment-variable "PATH") ":.")))
     (lambda  ()
       (common:without-vars proc "^MT_.*")))
    (let processloop ((i 0))
      (let-values (((pid-val exit-status exit-code)(process-wait pid #t)))
        (if (eq? pid-val 0)
            (begin
              (thread-sleep! 2)
              (processloop (+ i 1)))
            (begin
              (debug:print-info 0 *default-log-port* "sub megatest " action-switches-str " completed with exit code " exit-code)
              (if (eq? 0 exit-code)
                  (begin
                    #t)
                  (begin
                    #f))))))))



;; (subrun:exec-sub-megatest "/nfs/pdx/disks/icf_env_disk001/bjbarcla/gwa/issues/mtdev/165/megatest/ext-tests/tests/subrun-usecases/toparea/links/SYSTEM_val/RELEASE_val/go/toptest" "-foo" "foo")

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

;;======================================================================

(declare (unit subrunmod))
(declare (uses commonmod))


(module subrunmod
	*
	
(import scheme chicken data-structures extras)
(import (prefix sqlite3 sqlite3:) posix typed-records srfi-18 srfi-69 format ports srfi-1 matchable)


(import commonmod)


;; (use (prefix ulex ulex:))

(include "common_records.scm")




































































































































































































































)

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

;;======================================================================
;; Tests
;;======================================================================

;; return items given config
;;
(define (tests:get-items tconfig)
  (let ((items      (hash-table-ref/default tconfig "items" #f)) ;; items 4
	(itemstable (hash-table-ref/default tconfig "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-info 4 *default-log-port* "items is a procedure, will calc later")
      items)            ;; calc later
     ((procedure? itemstable)
      (debug:print-info 4 *default-log-port* "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-info 4 *default-log-port* "items and itemstable are lists, calc now\n"
			"    items: " items " itemstable: " itemstable)
      (items:get-items-from-config tconfig))
     (else #f))))                           ;; not iterated


;; returns waitons waitors tconfigdat
;;
(define (tests:get-waitons test-name all-tests-registry)
   (let* ((config  (tests:get-testconfig test-name #f all-tests-registry 'return-procs))) ;; assuming no problems with immediate evaluation, this could be simplified ('return-procs -> #t)
     (let ((instr (if config 
		      (configf:lookup config "requirements" "waiton")
		      (begin ;; No config means this is a non-existant test
			(debug:print-error 0 *default-log-port* "non-existent required test \"" test-name "\"")
			(exit 1))))
	   (instr2 (if config
		       (configf:lookup config "requirements" "waitor")
		       "")))
       (debug:print-info 8 *default-log-port* "waitons string is " instr ", waitors string is " instr2)
       (let ((newwaitons
	      (string-split (cond
			     ((procedure? instr) ;; here 
			      (let ((res (instr)))
				(debug:print-info 8 *default-log-port* "waiton procedure results in string " res " for test " test-name)
				res))
			     ((string? instr)     instr)
			     (else 
			      ;; NOTE: This is actually the case of *no* waitons! ;; (debug:print-error 0 *default-log-port* "something went wrong in processing waitons for test " test-name)
			      ""))))
	     (newwaitors
	      (string-split (cond
			     ((procedure? instr2)
			      (let ((res (instr2)))
				(debug:print-info 8 *default-log-port* "waitor procedure results in string " res " for test " test-name)
				res))
			     ((string? instr2)     instr2)
			     (else 
			      ;; NOTE: This is actually the case of *no* waitons! ;; (debug:print-error 0 *default-log-port* "something went wrong in processing waitons for test " test-name)
			      "")))))
	 (values
	  ;; the waitons
	  (filter (lambda (x)
		    (if (hash-table-ref/default all-tests-registry x #f)
			#t
			(begin
			  (debug:print-error 0 *default-log-port* "test " test-name " has unrecognised waiton testname " x)
			  #f)))
		  newwaitons)
	  (filter (lambda (x)
		    (if (hash-table-ref/default all-tests-registry x #f)
			#t
			(begin
			  (debug:print-error 0 *default-log-port* "test " test-name " has unrecognised waiton testname " x)
			  #f)))
		  newwaitors)
	  config)))))
					     
;; 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) (db:test-get-item-path testdat) test-registry #f))
	 (test-rundir ;; (sdb:qry 'passstr 
	  (db:test-get-rundir testdat)) ;; )
	 (prev-rundir ;; (sdb:qry 'passstr 
	  (db:test-get-rundir prev-testdat)) ;; )
	 (waivers     (if testconfig (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 (common:file-exists? test-rundir))
	(begin
	  (debug:print-error 0 *default-log-port* "test run directory is gone, cannot propagate waiver")
	  #f)
	(begin
	  (push-directory test-rundir)
	  (let ((result (if (null? waivers)
			    #f
			    (let loop ((hed (car waivers))
				       (tal (cdr waivers)))
			      (debug:print 0 *default-log-port* "INFO: Applying waiver rule \"" hed "\"")
			      (let* ((waiver      (configf:lookup testconfig "waivers" hed))
				     (wparts      (if waiver (string-match waiver-rx waiver) #f))
				     (waiver-rule (if wparts (cadr wparts)  #f))
				     (waiver-glob (if wparts (caddr wparts) #f))
				     (logpro-file (if waiver
						      (let ((fname (conc hed ".logpro")))
							(if (common:file-exists? fname)
							    fname 
							    (begin
							      (debug:print 0 *default-log-port* "INFO: No logpro file " fname " falling back to diff")
							      #f)))
						      #f))
				     ;; if rule by name of waiver-rule is found in testconfig - use it
				     ;; else if waivername.logpro exists use logpro-rule
				     ;; else default to diff-rule
				     (rule-string (let ((rule (configf:lookup testconfig "waiver_rules" waiver-rule)))
						    (if rule
							rule
							(if logpro-file
							    logpro-rule
							    (begin
							      (debug:print 0 *default-log-port* "INFO: No logpro file " logpro-file " found, using diff rule")
							      diff-rule)))))
				     ;; (string-substitute "%file1%" "foofoo.txt" "This is %file1% and so is this %file1%." #t)
				     (processed-cmd (string-substitute 
						     "%file1%" (conc test-rundir "/" waiver-glob)
						     (string-substitute
						      "%file2%" (conc prev-rundir "/" waiver-glob)
						      (string-substitute
						       "%waivername%" hed rule-string #t) #t) #t))
				     (res            #f))
				(debug:print 0 *default-log-port* "INFO: waiver command is \"" processed-cmd "\"")
				(if (eq? (system processed-cmd) 0)
				    (if (null? tal)
					#t
					(loop (car tal)(cdr tal)))
				    #f))))))
	    (pop-directory)
	    result)))))

;; Do not rpc this one, do the underlying calls!!!
(define (tests:test-set-status! run-id test-id state status comment dat #!key (work-area #f))
  (let* ((real-status status)
	 (otherdat    (if dat dat (make-hash-table)))
	 (testdat     (rmt:get-test-info-by-id run-id test-id))
	 (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")
			  (rmt:get-previous-test-run-record 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 *default-log-port* "prev-status " prev-status ", prev-state " prev-state ", prev-comment " prev-comment)
			     (if (and (equal? prev-state  "COMPLETED")
				      (equal? prev-status "WAIVED"))
				 (if comment
				     comment
				     prev-comment) ;; waived is either the comment or #f
				 #f))
			   #f)
		       #f)))
    (if (and waived 
	     (tests:check-waiver-eligibility testdat prev-test))
	(set! real-status "WAIVED"))

    (debug:print 4 *default-log-port* "real-status " real-status ", waived " waived ", status " status)

    ;; update the primary record IF state AND status are defined
    (if (and state status)
	(begin
	  (rmt:set-state-status-and-roll-up-items run-id test-id item-path state real-status (if waived waived comment))
	  ;; (mt:process-triggers run-id test-id state real-status) ;; triggers are called in test-set-state-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")) 
    ;; 	(rmt:test-data-rollup run-id test-id status))

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

    (let ((category (hash-table-ref/default otherdat ":category" ""))
	  (variable (hash-table-ref/default otherdat ":variable" ""))
	  (value    (hash-table-ref/default otherdat ":value"    #f))
	  (expected (hash-table-ref/default otherdat ":expected" "n/a"))
	  (tol      (hash-table-ref/default otherdat ":tol"      "n/a"))
	  (units    (hash-table-ref/default otherdat ":units"    ""))
	  (type     (hash-table-ref/default otherdat ":type"     ""))
	  (dcomment (hash-table-ref/default otherdat ":comment"  "")))
      (debug:print 4 *default-log-port* 
		   "category: " category ", variable: " variable ", value: " value
		   ", expected: " expected ", tol: " tol ", units: " units)
      (if (and value) ;; require only value; BB was- all three required
	  (let ((dat (conc category ","
			   variable ","
			   value    ","
			   expected ","
			   tol      ","
			   units    ","
			   dcomment ",," ;; extra comma for status
			   type     )))
	    ;; This was run remote, don't think that makes sense. Perhaps not, but that is the easiest path for the moment.
	    (rmt:csv->test-data run-id test-id
				dat)
	    ;; This was added in check-in a5adfa3f9a. Message was: "...added delay in set-values to allow for delayed write on server start"
	    ;; I'm inserting an arbitrary rmt: call to force/ensure that the server is available to (hopefully) prevent a communication issue.
	    (rmt:get-var "MEGATEST_VERSION") ;; this does NOTHING but ensure the server is reachable. This is almost certainly NOT needed :)
            ;; BB - commentiong out arbitrary 10 second wait (thread-sleep! 10) ;; add 10 second delay before quit incase rmt needs time to start a server.
            )))
      
    ;; need to update the top test record if PASS or FAIL and this is a subtest
    ;;;;;; (if (not (equal? item-path ""))
    ;;;;;;     (rmt:set-state-status-and-roll-up-items run-id test-name item-path state status #f) ;;;;;)

    (if (or (and (string? comment)
		 (string-match (regexp "\\S+") comment))
	    waived)
	(let ((cmt  (if waived waived comment)))
	  (rmt:general-call 'set-test-comment run-id cmt test-id)))))

(define (tests:test-set-toplog! run-id test-name logf) 
  (rmt:general-call 'tests:test-set-toplog run-id logf run-id test-name))

(define (tests:summarize-items 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      (rmt:test-get-logfile-info 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 *default-log-port* "Found path: " path)
	  (change-directory path))
	;; (set! outputfilename (conc path "/" outputfilename)))
	(debug:print-error 0 *default-log-port* "summarize-items for run-id=" run-id ", test-name=" test-name ", no such path: " path))
    (debug:print 4 *default-log-port* "summarize-items with logf " logf ", outputfilename " outputfilename " and force " force)
    (if (or (equal? logf "logs/final.log")
	    (equal? logf outputfilename)
	    force)
	(let ((my-start-time (current-seconds))
	      (lockf         (conc outputfilename ".lock")))
	  (let loop ((have-lock  (common:simple-file-lock lockf)))
	    (if have-lock
		(let ((script (configf:lookup *configdat* "testrollup" test-name)))
		  (print "Obtained lock for " outputfilename)
		  (rmt:set-state-status-and-roll-up-items run-id test-name "" #f #f #f)
		  (if script
		      (system (conc script " > " outputfilename " & "))
		      (tests:generate-html-summary-for-iterated-test run-id test-id test-name outputfilename))
		  (common:simple-file-release-lock lockf)
		  (change-directory orig-dir)
		  ;; NB// tests:test-set-toplog! is remote internal...
		  (tests:test-set-toplog! run-id test-name outputfilename))
		;; didn't get the lock, check to see if current update started later than this 
		;; update, if so we can exit without doing any work
		(if (> my-start-time (handle-exceptions
					 exn
					 0
				       (file-modification-time lockf)))
		    ;; we started since current re-gen in flight, delay a little and try again
		    (begin
		      (debug:print-info 1 *default-log-port* "Waiting to update " outputfilename ", another test currently updating it")
		      (thread-sleep! (+ 5 (random 5))) ;; delay between 5 and 10 seconds
		      (loop (common:simple-file-lock lockf))))))))))

(define (tests:generate-html-summary-for-iterated-test run-id test-id test-name outputfilename)
  (let ((counts              (make-hash-table))
	(statecounts         (make-hash-table))
	(outtxt              "")
	(tot                 0)
	(testdat             (rmt:test-get-records-for-index-file run-id test-name)))
    (with-output-to-file outputfilename
      (lambda ()
	(set! outtxt (conc outtxt "<html><title>Summary: " test-name 
			   "</title><body><h2>Summary for " test-name "</h2>"))
	(for-each
	 (lambda (testrecord)
	   (let ((id             (vector-ref testrecord 0))
		 (itempath       (vector-ref testrecord 1))
		 (state          (vector-ref testrecord 2))
		 (status         (vector-ref testrecord 3))
		 (run_duration   (vector-ref testrecord 4))
		 (logf           (vector-ref testrecord 5))
		 (comment        (vector-ref testrecord 6)))
	     (hash-table-set! counts status (+ 1 (hash-table-ref/default counts status 0)))
	     (hash-table-set! statecounts state (+ 1 (hash-table-ref/default statecounts state 0)))
	     (set! outtxt (conc outtxt "<tr>"
				;; "<td><a href=\"" itempath "/" logf "\"> " itempath "</a></td>" 
				"<td><a href=\"" itempath "/test-summary.html\"> " itempath "</a></td>" 
				"<td>" state    "</td>" 
				"<td><font color=" (common:get-color-from-status status)
				">"   status   "</font></td>"
				"<td>" (if (equal? comment "")
					   "&nbsp;"
					   comment) "</td>"
					   "</tr>"))))
	 (if (list? testdat)
	     testdat
	     (begin
	       (print "ERROR: failed to get records with rmt:test-get-records-for-index-file run-id=" run-id "test-name=" test-name)
	       '())))
	
	(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 statecounts))
	(print "<tr><td>Total</td><td>" tot "</td></tr></table>")
	(print "</td><td valign=\"top\">")
	;; Print out stats for state
	(set! tot 0)
	(print "<table cellspacing=\"0\" border=\"1\"><tr><td colspan=\"2\"><h2>Status stats</h2></td></tr>")
	(for-each (lambda (status)
		    (set! tot (+ tot (hash-table-ref counts status)))
		    (print "<tr><td><font color=\"" (common:get-color-from-status status) "\">" status
			   "</font></td><td>" (hash-table-ref counts status) "</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)
	;;(rmt:update-run-stats 
	;; run-id
	;; (hash-table-map
	;;  state-status-counts
	;;  (lambda (key val)
	;;	(append key (list val)))))
	))))

(define tests:css-jscript-block
#<<EOF
<style type="text/css">
ul.LinkedList { display: block; }
/* ul.LinkedList ul { display: none; } */
.HandCursorStyle { cursor: pointer; cursor: hand; }  /* For IE */
th {background-color: #8c8c8c;}
td.test {background-color: #d9dbdd;}
td.PASS {background-color: #347533;}
td.FAIL {background-color: #cc2812;}
td.SKIP{background-color: #FFD733;}
td.WARN {background-color: #EA8724;}
td.WAIVED {background-color: #838A12;}
td.ABORT{background-color: #EA24B7;}
.PASS .link, .SKIP .link, .WARN .link,.WAIVED .link,.ABORT .link, .FAIL .link{color: #FFFFFF;}


</style>


  <script type="text/JavaScript">

    function filtersome() {
  $("tr").show();
  $(".test").filter(
    function() {
      var names = $('#testname').val().split(',');
      var good=1;
      for (var i=0, len=names.length; i<len; i++) {
        var uname=names[i];
        console.log("Trying to check for " + uname); 
        if($(this).text().indexOf(uname) != -1) {
          good= 0;
          console.log("Found "+uname);
        }
      }
      return good; 
    }
  ).parent().hide();
//  $(".sum").show();
}
  
    // Add this to the onload event of the BODY element
    function addEvents() {
      activateTree(document.getElementById("LinkedList1"));
    }

    // This function traverses the list and add links 
    // to nested list items
    function activateTree(oList) {
      // Collapse the tree
      for (var i=0; i < oList.getElementsByTagName("ul").length; i++) {
        oList.getElementsByTagName("ul")[i].style.display="none";            
      }                                                                  
      // Add the click-event handler to the list items
      if (oList.addEventListener) {
        oList.addEventListener("click", toggleBranch, false);
      } else if (oList.attachEvent) { // For IE
        oList.attachEvent("onclick", toggleBranch);
      }
      // Make the nested items look like links
      addLinksToBranches(oList);
    }

    // This is the click-event handler
    function toggleBranch(event) {
      var oBranch, cSubBranches;
      if (event.target) {
        oBranch = event.target;
      } else if (event.srcElement) { // For IE
        oBranch = event.srcElement;
      }
      cSubBranches = oBranch.getElementsByTagName("ul");
      if (cSubBranches.length > 0) {
        if (cSubBranches[0].style.display == "block") {
          cSubBranches[0].style.display = "none";
        } else {
          cSubBranches[0].style.display = "block";
        }
      }
    }

    // This function makes nested list items look like links
    function addLinksToBranches(oList) {
      var cBranches = oList.getElementsByTagName("li");
      var i, n, cSubBranches;
      if (cBranches.length > 0) {
        for (i=0, n = cBranches.length; i < n; i++) {
          cSubBranches = cBranches[i].getElementsByTagName("ul");
          if (cSubBranches.length > 0) {
            addLinksToBranches(cSubBranches[0]);
            cBranches[i].className = "HandCursorStyle";
            cBranches[i].style.color = "blue";
            cSubBranches[0].style.color = "black";
            cSubBranches[0].style.cursor = "auto";
          }
        }
      }
    }
  </script>
EOF
)

(define tests:css-jscript-block-dynamic 
#<<EOF
           <script src= ./jquery3.1.0.js></script> 
EOF
)

(define  (test:js-block javascript-lib)
   (conc  "<script src=" javascript-lib "></script>" ))


(define tests:css-jscript-block-static (test:js-block *java-script-lib*))

(define (tests:css-jscript-block-cond dynamic) 
      (if (equal? dynamic  #t)
       tests:css-jscript-block-dynamic
       tests:css-jscript-block-static))

       
(define (tests:run-record->test-path run numkeys)
   (append (take (vector->list run) numkeys)
	   (list (vector-ref run (+ 1 numkeys)))))


(define (tests:get-rest-data runs header numkeys)
   (let ((resh (make-hash-table)))
   (for-each
     (lambda (run)
        (let* ((run-id (db:get-value-by-header run header "id"))
               (run-dir      (tests:run-record->test-path run numkeys))
	       (test-data    (rmt:get-tests-for-run
				   run-id
                                   "%"       ;; testnamepatt
				   '()        ;; states
				   '()        ;; statuses
				   #f         ;; offset
				   #f         ;; num-to-get
				   #f         ;; hide/not-hide
				   #f         ;; sort-by
				   #f         ;; sort-order
				   #f         ;; 'shortlist                           ;; qrytype
                                   0         ;; last update
				   #f)))
            
            (map (lambda (test)
                 (let* ((test-name (vector-ref test 2))
                        (test-html-path (conc (vector-ref test 10) "/" (vector-ref test 13)))
                        (test-item (conc test-name ":" (vector-ref test 11)))
                        (test-status (vector-ref test 4)))
                         
                (if (not (hash-table-ref/default resh test-name  #f))
                      (hash-table-set! resh test-name   (make-hash-table)))
                (if (not (hash-table-ref/default (hash-table-ref/default resh test-name  #f)  test-item  #f))
                       (hash-table-set! (hash-table-ref/default resh test-name  #f) test-item   (make-hash-table))) 
               (hash-table-set!  (hash-table-ref/default (hash-table-ref/default resh test-name  #f) test-item #f) run-id (list test-status test-html-path)))) 
        test-data)))
      runs)
   resh))


;; hash-table tree to html list tree
;;
;;   tipfunc takes two parameters: y the tip value and path the path to that point
;;
(define (common:htree->html ht path tipfunc)
  (let ((datlist 	(sort (hash-table->alist ht)
                              (lambda (a b)
                                (string< (car a)(car b))))))
    (if (null? datlist)
    	(tipfunc #f path) ;; really shouldn't get here
	(s:ul
	 (map (lambda (x)
		(let* ((levelname (car x))
		       (y         (cdr x))
		       (newpath   (append path (list levelname)))
		       (leaf      (or (not (hash-table? y))
				      (null? (hash-table-keys y)))))
		  (if leaf
		      (s:li (tipfunc y newpath))
		      (s:li
		       (list 
			levelname
			(common:htree->html y newpath tipfunc))))))
	      datlist)))))


;; tests:genrate dashboard body 
;;

(define (tests:dashboard-body page pg-size keys numkeys  total-runs linktree area-name get-prev-links get-next-links flag run-patt target-patt)
  (let* ((start (* page pg-size)) 
	       ;(runsdat   (rmt:get-runs "%" pg-size start (map (lambda (x)(list x "%")) keys)))
         (runsdat   (rmt:get-runs-by-patt  keys run-patt target-patt start pg-size #f 0 sort-order: "desc"))
                    ; db:get-runs-by-patt   keys runnamepatt targpatt offset limit fields last-update   
	       (header    (vector-ref runsdat 0))
	       (runs      (vector-ref runsdat 1))
         (ctr 0)
         (test-runs-hash (tests:get-rest-data runs header numkeys))
         (test-list (hash-table-keys test-runs-hash))) 
  
  (s:html tests:css-jscript-block (tests:css-jscript-block-cond flag)
		   (s:title "Summary for " area-name)
		   (s:body 'onload "addEvents();"
                          (get-prev-links page linktree)
                          (get-next-links page linktree total-runs)
                           
			   (s:h1 "Summary for " area-name)
                           (s:h3 "Filter" )
                           (s:input 'type "text"  'name "testname" 'id "testname" 'length "30" 'onkeyup "filtersome()")
			   ;; top list
         
			   (s:table 'id "LinkedList1" 'border "1" 'cellspacing 0
                            (map (lambda (key)
				 (let* ((res (s:tr 'class "something" 
				  (s:th key )
                                   (map (lambda (run)
                                   (s:th  (vector-ref run ctr)))
                                  runs))))
                             (set! ctr (+ ctr 1))
                               res))
                               keys)
                               (s:tr
				 (s:th "Run Name")
                                  (map (lambda (run)
                                   (s:th (db:get-value-by-header run header "runname")))
                                  runs))
                              
                               (map (lambda (test-name)
                                 (let* ((item-hash (hash-table-ref/default test-runs-hash test-name  #f))
                                         (item-keys (sort (hash-table-keys item-hash) string<=?))) 
                                          (map (lambda (item-name)  
  		                             (let* ((res (s:tr  'class item-name
				                         (s:td  item-name 'class "test" )
                                                           (map (lambda (run)
                                                               (let* ((run-test (hash-table-ref/default item-hash item-name  #f))
                                                                      (run-id (db:get-value-by-header run header "id"))
                                                                      (result (hash-table-ref/default run-test run-id "n/a"))
                                                                      ;(relative-path (get-relative-path)) 
                                                                      (status (if (string? result)
									                                                            	result
										                                                            (car result)))
                                                                        (link (if (string? result)
										                                                            result
                                                                                (if (equal? flag #t) 
                                                                                (s:a (car result) 'href (conc "./test_log?runid=" run-id "&testname="  item-name ))
  																																						  (s:a (car result) 'href (string-substitute  (conc linktree "/")  "" (cadr result)  "-"))))))
                                                                       (s:td  link 'class status)))
                                                                runs))))
                                                        res))
                                                   item-keys)))
                               test-list)))))) 

;; (tests:create-html-tree "test-index.html")
;;
(define (tests:create-html-tree outf)
   (let* ((lockfile  (conc outf ".lock"))
	 			 (runs-to-process '())
         (linktree  (common:get-linktree))
         (area-name (common:get-testsuite-name))
	  		 (keys      (rmt:get-keys))
	  		 (numkeys   (length keys))
         (run-patt (or (args:get-arg "-run-patt")
                        (args:get-arg "-runname")
                        "%"))
         (target (or  (args:get-arg "-target-patt") 
											(args:get-arg "-target")
                      "%"))
         (targlist (string-split target "/"))
         (numtarg  (length targlist))  
         (targtweaked (if (> numkeys numtarg)
			   								(append targlist (make-list (- numkeys numtarg) "%"))
			  								targlist))
         (target-patt (string-join targtweaked "/"))
         ;(total-runs  (rmt:get-num-runs "%")) ;;this needs to be changed to filter by target
          (total-runs (rmt:get-runs-cnt-by-patt run-patt target-patt keys )) 
         (pg-size 10))
    (if (common:simple-file-lock lockfile)
        (begin
         ;(print total-runs)    
        (let loop ((page 0))
	(let* ((oup       (open-output-file (or outf (conc linktree "/page" page ".html"))))
               (get-prev-links (lambda (page linktree )   
                            (let* ((link  (if (not (eq? page 0))
                                   (s:a "&lt;&lt;prev" 'href (conc  "page" (- page 1) ".html"))
                                   (s:a "" 'href (conc   "page"  page ".html")))))
                               link)))
               (get-next-links (lambda (page linktree total-runs)   
                            (let* ((link  (if (> total-runs (+ 10 (* page pg-size)))
                                   (s:a "next&gt;&gt;" 'href (conc  "page"  (+ page 1) ".html"))
                                   (s:a "" 'href (conc   "page" page  ".html")))))
                               link))) )
          (print "total runs: " total-runs) 
          (s:output-new
	   			 oup
	   					(tests:dashboard-body page pg-size keys numkeys total-runs linktree area-name get-prev-links get-next-links #f run-patt target-patt)) ;; update this function
          (close-output-port oup)
         ; (set! page (+ 1 page))
          (if (> total-runs (* (+ 1 page) pg-size))
           (loop (+ 1  page)))))
	  (common:simple-file-release-lock lockfile))
	            
	#f)))


(define (tests:readlines filename)
  (call-with-input-file filename
    (lambda (p)
      (let loop ((line (read-line p))
                 (result '()))
        (if (eof-object? line)
            (reverse result)
            (loop (read-line p) (cons line result)))))))

(define (tests:get-test-log run-id test-name item-name)
  (let* ((test-data    (rmt:get-tests-for-run
				   (string->number run-id)
                                    test-name      ;; testnamepatt
				   '()        ;; states
				   '()        ;; statuses
				   #f         ;; offset
				   #f         ;; num-to-get
				   #f         ;; hide/not-hide
				   #f         ;; sort-by
				   #f         ;; sort-order
				   #f         ;; 'shortlist                           ;; qrytype
                                   0         ;; last update
				   #f))
         (path "")
         (found 0))
    (debug:print-info 0 *default-log-port* "found: " found )

   (let loop ((hed (car test-data))
		 (tal (cdr test-data)))
          (debug:print-info 0 *default-log-port* "item: " (vector-ref hed 11) (vector-ref hed 10) "/" (vector-ref hed 13))

	(if (equal? (vector-ref hed 11) item-name)
            (begin
              (set! found 1) 
	      (set! path (conc (vector-ref hed 10) "/" (vector-ref hed 13)))))
	    (if (and (not (null? tal)) (equal? found 0))
		(loop (car tal)(cdr tal))))
   (if (equal? path "")
     "<H2>Data not found</H2>"
     (string-join (tests:readlines path) "\n"))))


(define (tests:dynamic-dboard page)
;(define (tests:create-html-tree o)
 (let* (
;(page "1")
          (linktree  (common:get-linktree))
         (area-name (common:get-testsuite-name))
	       (keys      (rmt:get-keys))
	       (numkeys   (length keys))
         (targtweaked (make-list numkeys "%"))
         (target-patt (string-join targtweaked "/"))
         (total-runs  (rmt:get-num-runs "%"))
         (pg-size 10)
         (pg (if (equal? page #f)
                 0
                 (- (string->number page) 1)))
          (get-prev-links  (lambda (pg linktree)
                           (debug:print-info 0 *default-log-port* "val: " (- 1 pg))
                          (let* ((link  (if (not (eq? pg 0))
                               (s:a  "&lt;&lt;prev " 'href (conc  "dashboard?page="  pg  ))
                               (s:a "" 'href (conc  "dashboard?page=" pg)))))
                               link)))
          (get-next-links   (lambda (pg linktree total-runs)  
                            (debug:print-info 0 *default-log-port* "val: " pg)
                             (debug:print-info 0 *default-log-port* "val: " total-runs " size" pg-size)
 
                            (let* ((link  (if (> total-runs (+ 10 (* pg pg-size)))
                              (s:a  "next&gt;&gt; "  'href (conc  "dashboard?page="  (+ pg 2)  ))
                             (s:a "" 'href (conc  "dashboard?page=" pg  )))))
                             link)))
         (html-body (tests:dashboard-body pg pg-size keys numkeys total-runs linktree area-name get-prev-links get-next-links #t "%" target-patt))) ;; update tis function
        html-body))

(define (tests:create-html-summary outf)
 (let* ((lockfile  (conc outf ".lock"))
        (linktree  (common:get-linktree))
				(keys      (rmt:get-keys))
        (area-name (common:get-testsuite-name))
        (run-patt (or (args:get-arg "-run-patt")
                        (args:get-arg "-runname")
                        "%"))
        (target (or (args:get-arg "-target-patt")
                        (args:get-arg "-target")
                        "%"))
         (targlist (string-split target "/"))
         (numkeys  (length keys))
	       (numtarg  (length targlist))  
         (targtweaked (if (> numkeys numtarg)
			   								(append targlist (make-list (- numkeys numtarg) "%"))
			  								targlist))
        (target-patt (string-join targtweaked "/")))
    (if (common:simple-file-lock lockfile)
        (begin
          (let* (;(runsdat1   (rmt:get-runs run-patt #f #f (map (lambda (x)(list x "%")) keys)))
                 (runsdat   (rmt:get-runs-by-patt  keys run-patt target-patt #f #f #f 0))
					       (runs      (vector-ref runsdat 1))
                 (header      (vector-ref runsdat 0))
        	       (oup       (open-output-file (or outf (conc linktree "/targets.html"))))
                 (target-hash (test:create-target-hash runs header (length keys))))
           (test:create-target-html target-hash oup area-name linktree)
          (test:create-run-html  runs area-name linktree (length keys) header))
	  (common:simple-file-release-lock lockfile))
	#f)))

(define (test:get-test-hash test-data)
	(let ((resh (make-hash-table)))
    	(map (lambda (test)
        (let* ((test-name (vector-ref test 2))
               (test-html-path (if (file-exists? (conc (vector-ref test 10) "/test-summary.html"))
																 (conc (vector-ref test 10) "/test-summary.html" )
							 									 (conc (vector-ref test 10) "/" (vector-ref test 13))))
               (test-item  (vector-ref test 11))
               (test-status (vector-ref test 4)))
               (if (not (hash-table-ref/default resh test-item  #f))
                   (hash-table-set! resh test-item   (make-hash-table)))
               (hash-table-set! (hash-table-ref/default resh test-item  #f) test-name (list test-status test-html-path)))) 
        test-data)
resh))

(define (test:get-data->b-keys ordered-data a-keys)
  (delete-duplicates
   (sort (apply
	  append
	  (map (lambda (sub-key)
		 (let ((subdat (hash-table-ref ordered-data sub-key)))
		   (hash-table-keys subdat)))
	       a-keys))
	 string>=?)))


(define (test:create-run-html runs area-name linktree numkeys header)
  (map (lambda (run)
		 (let* ((target (string-join (take (vector->list run) numkeys) "/"))
						(run-name (db:get-value-by-header run header "runname"))
            (run-time (seconds->work-week/day-time (db:get-value-by-header run header "event_time")))
						(oup (if (file-exists? (conc linktree "/" target "/" run-name))
                        (open-output-file (conc linktree "/" target "/" run-name "/run.html"))
                         #f))
            (run-id (db:get-value-by-header run header "id"))
            (test-data    (rmt:get-tests-for-run
				  								 run-id
                           "%"       ;; testnamepatt
				  								 '()        ;; states
				   								 '()        ;; statuses
				  								 	#f         ;; offset
				  						 			#f         ;; num-to-get
				   									#f         ;; hide/not-hide
				  								  #f         ;; sort-by
				   									#f         ;; sort-order
				   									#f         ;; 'shortlist                           ;; qrytype
                            0         ;; last update
				  									#f))
            (item-test-hash (test:get-test-hash test-data))
            (items  (hash-table-keys item-test-hash))
 						(test-names (test:get-data->b-keys item-test-hash items)))
    (if oup
      (begin 
     (s:output-new
	   oup
	   (s:html tests:css-jscript-block (tests:css-jscript-block-cond #f)
		   (s:title "Runs View " run-name)
		   (s:body
		     (s:h1 "Runs View " )
         (s:h3 "Target" target)
				 (s:p 
					(s:b "Run name" ) run-name)
         (s:p 
					(s:b "Run Date" ) run-time)
         (s:table 'border 1 'cellspacing 0
           (s:tr
           (s:th "Items")
           (map (lambda (test)
            (s:th test))
           test-names))  
           (map (lambda (item) 
					  (let* ((test-hash (hash-table-ref/default item-test-hash item  #f)))
								 (if test-hash
                  (begin
									(s:tr
					  			(s:td 'class "test" item)
            			(map (lambda (test)
						  		(let* ((test-details (hash-table-ref/default test-hash test  #f))
												(status (if test-details
																(car test-details)))
                        (link (if test-details 
														(string-substitute  (conc linktree "/" target "/" run-name "/")  "" (cadr test-details) "-"))))
                   (if test-details
											(s:td 'class status
												(s:a 'class "link" 'href link status ))
                      (s:td "")))) 			
									test-names))))))
				  (sort items string<=?))))))
		(close-output-port oup))
    (debug:print-info 0 "Skip: Dirctory structure " linktree "/" target "/" run-name " does not exist. Megatest will not create run.html"))))
runs))

(define (test:create-target-hash runs header numkeys)
  (let ((resh (make-hash-table)))
   (for-each
     (lambda (run)
        (let* ((run-name (db:get-value-by-header run header "runname"))
               (target   (string-join (take (vector->list run) numkeys) "/"))
               (run-list (hash-table-ref/default resh target  #f)))
               
               (if (not run-list)
                   (hash-table-set! resh target   (list run-name))
                   (hash-table-set! resh target   (cons run-name run-list)))))
      runs)
   resh))

(define (test:get-max-run-cnt target-hash targets)
   (let* ((cnt 0 ))
   (map (lambda (target)
        (let* ((runs  (hash-table-ref/default target-hash target  #f))
               (run-length (if runs
																(length runs)
                                 0)))
  
              (if (< cnt run-length)
               (set! cnt  run-length)))) 
		targets) 
cnt))
 
(define (test:pad-runs target-hash targets max-row-length)
 (map (lambda (target)
        (let loop ((run-list  (hash-table-ref/default target-hash target  #f)))
               (if (< (length run-list) max-row-length)
                 (begin  
               		 (hash-table-set! target-hash target   (cons "" run-list))
               		 (loop (hash-table-ref/default target-hash target  #f) ))))) 
		targets)
   target-hash)

(define (test:create-target-html target-hash oup area-name linktree)
  (let* ((targets (hash-table-keys target-hash))
         (max-row-length (test:get-max-run-cnt target-hash targets))
         (pad-runs-hash (test:pad-runs target-hash targets max-row-length)))
   (s:output-new
	   oup
	   (s:html tests:css-jscript-block (tests:css-jscript-block-cond #f)

		   (s:title "Target View " area-name)
		   (s:body
		   (s:h1 "Target View " area-name)
					(s:table 'id "LinkedList1" 'border "1" 'cellspacing 0
             (s:tr 'class "something" 
               (s:th "Target")
								(s:th 'colspan max-row-length "Runs"))                                              
                (let* ((tbl (map (lambda (target)
                      (s:tr
                      (s:td 'class "test" target)
										  (let* ((runs  (hash-table-ref/default target-hash target  #f))
														 (rest-row (map (lambda (run)
																				(if (equal? run "")
																						(s:td run)
                                            (if (file-exists?(conc linktree "/" target "/" run ))
																						(begin 
																							(s:td 
																							(s:a 'href (conc  target "/" run "/run.html") run))))))
																				(reverse runs))))
                              rest-row)))
                                   targets)))
                           tbl)))))
          (close-output-port oup)))


(define (tests:create-html-tree-old outf)
   (let* ((lockfile  (conc outf ".lock"))
	 (runs-to-process '()))
    (if (common:simple-file-lock lockfile)
	(let* ((linktree  (common:get-linktree))
	       (oup       (open-output-file (or outf (conc linktree "/runs-index.html"))))
	       (area-name (common:get-testsuite-name))
	       (keys      (rmt:get-keys))
	       (numkeys   (length keys))
	       (runsdat   (rmt:get-runs "%" #f #f (map (lambda (x)(list x "%")) keys)))
	       (header    (vector-ref runsdat 0))
	       (runs      (vector-ref runsdat 1))
	       (runtreedat (map (lambda (x)
				  (tests:run-record->test-path x numkeys))
				runs))
	       (runs-htree (common:list->htree runtreedat)))
	  (set! runs-to-process runs)
	  (s:output-new
	   oup
	   (s:html tests:css-jscript-block
		   (s:title "Summary for " area-name)
		   (s:body 'onload "addEvents();"
			   (s:h1 "Summary for " area-name)
			   ;; top list
			   (s:ul 'id "LinkedList1" 'class "LinkedList"
				 (s:li
				  "Runs"
				  (common:htree->html runs-htree
						      '()
						      (lambda (x p)
							(let* ((targ-path (string-intersperse p "/"))
                                                               (full-path (conc linktree "/" targ-path))
                                                               (run-name  (car (reverse p))))
                                                          (if (and (common:file-exists? full-path)
                                                                   (directory?   full-path)
                                                                   (file-write-access? full-path))
                                                              (s:a run-name 'href (conc targ-path "/run-summary.html"))
                                                              (begin
                                                                (debug:print 0 *default-log-port* "INFO: Can't create " targ-path "/run-summary.html")
                                                                (conc run-name " (Not able to create summary at " targ-path ")")))))))))))
          (close-output-port oup)
	  (common:simple-file-release-lock lockfile)
               
	  (for-each
	   (lambda (run)
	     (let* ((test-subpath (tests:run-record->test-path run numkeys))
		    (run-id       (db:get-value-by-header run header "id"))
                    (run-dir      (tests:run-record->test-path run numkeys))
		    (test-dats    (rmt:get-tests-for-run
				   run-id
                                   "%/"       ;; testnamepatt
				   '()        ;; states
				   '()        ;; statuses
				   #f         ;; offset
				   #f         ;; num-to-get
				   #f         ;; hide/not-hide
				   #f         ;; sort-by
				   #f         ;; sort-order
				   #f         ;; 'shortlist                           ;; qrytype
                                   0         ;; last update
				   #f))
                    (tests-tree-dat (map (lambda (test-dat)
                                         ;; (tests:run-record->test-path x numkeys))
                                         (let* ((test-name  (db:test-get-testname test-dat))
                                                (item-path  (db:test-get-item-path test-dat))
                                                (full-name  (db:test-make-full-name test-name item-path))
                                                (path-parts (string-split full-name)))
                                           path-parts))
                                       test-dats))
                    (tests-htree (common:list->htree tests-tree-dat))
                    (html-dir    (conc linktree "/" (string-intersperse run-dir "/")))
                    (html-path   (conc html-dir "/run-summary.html"))
                    (oup         (if (and (common:file-exists? html-dir)
                                          (directory?   html-dir)
                                          (file-write-access? html-dir))
                                     (open-output-file  html-path)
                                     #f)))
               ;; (print "run-dir: " run-dir ", tests-tree-dat: " tests-tree-dat)
               (if oup
                   (begin
                     (s:output-new
                      oup
                      (s:html tests:css-jscript-block
                              (s:title "Summary for " area-name)
                              (s:body 'onload "addEvents();"
                                      (s:h1 "Summary for " (string-intersperse run-dir "/"))
                                      ;; top list
                                      (s:ul 'id "LinkedList1" 'class "LinkedList"
                                            (s:li
                                             "Tests"
                                             (common:htree->html tests-htree
                                                                 '()
                                                                 (lambda (x p)
                                                                   (let* ((targ-path (string-intersperse p "/"))
                                                                          (test-name (car p))
                                                                          (item-path ;; (if (> (length p) 2) ;; test-name + run-name
                                                                           (string-intersperse p "/"))
                                                                          (full-targ (conc html-dir "/" targ-path))
                                                                          (std-file  (conc full-targ "/test-summary.html"))
                                                                          (alt-file  (conc full-targ "/megatest-rollup-" test-name ".html"))
                                                                          (html-file (if (common:file-exists? alt-file)
                                                                                         alt-file
                                                                                         std-file))
                                                                          (run-name  (car (reverse p))))
                                                                     (if (and (not (common:file-exists? full-targ))
                                                                              (directory? full-targ)
                                                                              (file-write-access? full-targ))
                                                                         (tests:summarize-test 
                                                                          run-id 
                                                                          (rmt:get-test-id run-id test-name item-path)))
                                                                     (if (common:file-exists? full-targ)
                                                                         (s:a run-name 'href html-file)
                                                                         (begin
                                                                           (debug:print 0 *default-log-port* "ERROR: can't access " full-targ)
                                                                           (conc "No summary for " run-name)))))
                                                                 ))))))
                     (close-output-port oup)))))
           runs)
          #t)
	#f)))







;; CHECK - WAS THIS ADDED OR REMOVED? MANUAL MERGE WITH API STUFF!!!
;;
;; get a pretty table to summarize steps
;;
;; (define (dcommon:process-steps-table steps);; db test-id #!key (work-area #f))
(define (tests:process-steps-table steps);; db test-id #!key (work-area #f))
;;  (let ((steps   (db:get-steps-for-test db test-id work-area: work-area)))
    ;; organise the steps for better readability
    (let ((res (make-hash-table)))
      (for-each 
       (lambda (step)
	 (debug:print 6 *default-log-port* "step=" step)
	 (let ((record (hash-table-ref/default 
			res 
			(tdb:step-get-stepname step)
			;;           0                      1    2    3       4         5       6       7
			;;        stepname                start end status Duration  Logfile Comment  first-id
			(vector (tdb:step-get-stepname step) ""   "" ""     ""        ""     ""       #f))))
	   (debug:print 6 *default-log-port* "record(before) = " record 
			"\nid:       " (tdb:step-get-id step)
			"\nstepname: " (tdb:step-get-stepname step)
			"\nstate:    " (tdb:step-get-state step)
			"\nstatus:   " (tdb:step-get-status step)
			"\ntime:     " (tdb:step-get-event_time step))
	   (if (not (vector-ref record 7))(vector-set! record 7 (tdb:step-get-id step))) ;; do not clobber the id if previously set
	   (case (string->symbol (tdb:step-get-state step))
	     ((start)(vector-set! record 1 (tdb:step-get-event_time step))
	      (vector-set! record 3 (if (equal? (vector-ref record 3) "")
					(tdb:step-get-status step)))
	      (if (> (string-length (tdb:step-get-logfile step))
		     0)
		  (vector-set! record 5 (tdb:step-get-logfile step))))
	     ((end)  
	      (vector-set! record 2 (any->number (tdb:step-get-event_time step)))
	      (vector-set! record 3 (tdb:step-get-status step))
	      (vector-set! record 4 (let ((startt (any->number (vector-ref record 1)))
					  (endt   (any->number (vector-ref record 2))))
				      (debug:print 4 *default-log-port* "record[1]=" (vector-ref record 1) 
						   ", startt=" startt ", endt=" endt
						   ", get-status: " (tdb:step-get-status step))
				      (if (and (number? startt)(number? endt))
					  (seconds->hr-min-sec (- endt startt)) "-1")))
	      (if (> (string-length (tdb:step-get-logfile step))
		     0)
		  (vector-set! record 5 (tdb:step-get-logfile step)))
	      (if (> (string-length (tdb:step-get-comment step))
		     0)
		  (vector-set! record 6 (tdb:step-get-comment step))))
	     (else
	      (vector-set! record 2 (tdb:step-get-state step))
	      (vector-set! record 3 (tdb:step-get-status step))
	      (vector-set! record 4 (tdb:step-get-event_time step))
	      (vector-set! record 6 (tdb:step-get-comment step))))
	   (hash-table-set! res (tdb:step-get-stepname step) record)
	   (debug:print 6 *default-log-port* "record(after)  = " record 
			"\nid:       " (tdb:step-get-id step)
			"\nstepname: " (tdb:step-get-stepname step)
			"\nstate:    " (tdb:step-get-state step)
			"\nstatus:   " (tdb:step-get-status step)
			"\ntime:     " (tdb:step-get-event_time step))))
       ;; (else   (vector-set! record 1 (tdb:step-get-event_time step)))
       (sort steps (lambda (a b)
		     (cond
		      ((<   (tdb:step-get-event_time a)(tdb:step-get-event_time b)) #t)
		      ((eq? (tdb:step-get-event_time a)(tdb:step-get-event_time b)) 
		       (<   (tdb:step-get-id a)        (tdb:step-get-id b)))
		      (else #f)))))
      res))

;; 
;;
(define (tests:get-compressed-steps run-id test-id)
  (let* ((steps-data  (rmt:get-steps-for-test run-id test-id)) ;;      0       1    2    3       4       5       6      7       
	 (comprsteps  (tests:process-steps-table steps-data))) ;; #<stepname start end status Duration Logfile Comment id>
    (map (lambda (x)
	   ;; take advantage of the \n on time->string
	   (vector    ;; we are constructing basically the original vector but collapsing start end records
	    (vector-ref x 0)                              ;; id        0
	    (let ((s (vector-ref x 1)))
	      (if (number? s)(seconds->time-string s) s)) ;; starttime 1
	    (let ((s (vector-ref x 2)))
	      (if (number? s)(seconds->time-string s) s)) ;; endtime   2
	    (vector-ref x 3)                              ;; status    3    
	    (vector-ref x 4)                              ;; duration  4
	    (vector-ref x 5)                              ;; logfile   5
	    (vector-ref x 6)                              ;; comment   6
	    (vector-ref x 7)))                            ;; id        7
	 (sort (hash-table-values comprsteps)
	       (lambda (a b)
		 (let ((time-a (vector-ref a 1))
		       (time-b (vector-ref b 1))
		       (id-a   (vector-ref a 7))
		       (id-b   (vector-ref b 7)))
		   (if (and (number? time-a)(number? time-b))
		       (if (< time-a time-b)
			   #t
			   (if (eq? time-a time-b)
			       (< id-a id-b)
			       ;; (string<? (conc (vector-ref a 2))
			       ;;	    (conc (vector-ref b 2)))
			       #f))
		       (string<? (conc time-a)(conc time-b)))))))))


;; summarize test in to a file test-summary.html in the test directory
;;
(define (tests:summarize-test run-id test-id)
  (let* ((test-dat  (rmt:get-test-info-by-id run-id test-id))
	 (out-dir   (db:test-get-rundir test-dat))
	 (out-file  (conc out-dir "/test-summary.html")))
    ;; first verify we are able to write the output file
    (if (not (file-write-access? out-dir))
	(debug:print 0 *default-log-port* "ERROR: cannot write test-summary.html to " out-dir)
	(let* (;; (steps-dat (rmt:get-steps-for-test run-id test-id))
	       (test-name (db:test-get-testname test-dat))
	       (item-path (db:test-get-item-path test-dat))
	       (full-name (db:test-make-full-name test-name item-path))
	       (oup       (open-output-file out-file))
	       (status    (db:test-get-status   test-dat))
	       (color     (common:get-color-from-status status))
	       (logf      (db:test-get-final_logf test-dat))
	       (steps-dat (tests:get-compressed-steps run-id test-id)))
	  ;; (dcommon:get-compressed-steps #f 1 30045)
	  ;; (#("wasting_time" "23:36:13" "23:36:21" "0" "8.0s" "wasting_time.log"))
	  
	  (s:output-new
	   oup
	   (s:html
	    (s:title "Summary for " full-name)
	    (s:body 
	     (s:h2 "Summary for " full-name)
	     (s:table 'cellspacing "0" 'border "1"
		      (s:tr (s:td "run id")   (s:td (db:test-get-run_id   test-dat))
			    (s:td "test id")  (s:td (db:test-get-id       test-dat)))
		      (s:tr (s:td "testname") (s:td test-name)
			    (s:td "itempath") (s:td item-path))
		      (s:tr (s:td "state")    (s:td (db:test-get-state    test-dat))
			    (s:td "status")   (s:td (s:a 'href logf (s:font 'color color status))))
		      (s:tr (s:td "TestDate") (s:td (seconds->work-week/day-time 
						     (db:test-get-event_time test-dat)))
			    (s:td "Duration") (s:td (seconds->hr-min-sec (db:test-get-run_duration test-dat)))))
	     (s:h3 "Log files")
	     (s:table 
	      'cellspacing "0" 'border "1"
	      (s:tr (s:td "Final log")(s:td (s:a 'href logf logf))))
	     (s:table
	      'cellspacing "0" 'border "1"
	      (s:tr (s:td "Step Name")(s:td "Start")(s:td "End")(s:td "Status")(s:td "Duration")(s:td "Log File"))
	      (map (lambda (step-dat)
		     (s:tr (s:td (tdb:steps-table-get-stepname step-dat))
			   (s:td (tdb:steps-table-get-start    step-dat))
			   (s:td (tdb:steps-table-get-end      step-dat))
			   (s:td (tdb:steps-table-get-status   step-dat))
			   (s:td (tdb:steps-table-get-runtime  step-dat))
			   (s:td (let ((step-log (tdb:steps-table-get-log-file step-dat)))
				   (s:a 'href step-log step-log)))))
		   steps-dat))
	     )))
	  (close-output-port oup)))))
	  
	  
;; MUST BE CALLED local!
;;
(define (tests:test-get-paths-matching keynames target fnamepatt #!key (res '()))
  ;; BUG: Move the values derived from args to parameters and push to megatest.scm
  (let* ((testpatt   (or (args:get-arg "-testpatt")(args:get-arg "-testpatt") "%"))
	 (statepatt  (or (args:get-arg "-state")   (args:get-arg ":state")    "%"))
	 (statuspatt (or (args:get-arg "-status")  (args:get-arg ":status")   "%"))
	 (runname    (or (args:get-arg "-runname") (args:get-arg ":runname")  "%"))
	 (paths-from-db (rmt:test-get-paths-matching-keynames-target-new keynames target res
					testpatt
					statepatt
					statuspatt
					runname)))
    (if fnamepatt
	(apply append 
	       (map (lambda (p)
		      (if (directory-exists? p)
			  (let ((glob-query (conc p "/" fnamepatt)))
			    (handle-exceptions
				exn
				(with-input-from-pipe
				    (conc "echo " glob-query)
				  read-lines)  ;; we aren't going to try too hard. If glob breaks it is likely because someone tried to do */*/*.log or similar
			      (glob glob-query)))
			  '()))
		    paths-from-db))
	paths-from-db)))

			      
;;======================================================================
;; 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)(common:file-exists? (conc test "/testconfig"))) tests))
;;     (delete-duplicates
;;      (filter (lambda (testname)
;; 	       (tests:match test-patts testname #f))
;; 	     (map (lambda (testp)
;; 		    (last (string-split testp "/")))
;; 		  tests)))))

(define (tests:get-test-path-from-environment)
  (if (and (getenv "MT_LINKTREE")
	   (getenv "MT_TARGET")
	   (getenv "MT_RUNNAME")
	   (getenv "MT_TEST_NAME")
	   (getenv "MT_ITEMPATH"))
      (conc (getenv "MT_LINKTREE")  "/"
	    (getenv "MT_TARGET")    "/"
	    (getenv "MT_RUNNAME")   "/"
	    (getenv "MT_TEST_NAME")
	    (if (and (getenv "MT_ITEMPATH")
                     (not (string=? "" (getenv "MT_ITEMPATH"))))
		(conc "/" (getenv "MT_ITEMPATH"))
                ""))
      #f))

;; if .testconfig exists in test directory read and return it
;; else if have cached copy in *testconfigs* return it IFF there is a section "have fulldata"
;; else read the testconfig file
;;   if have path to test directory save the config as .testconfig and return it
;;
(define (tests:get-testconfig test-name item-path test-registry system-allowed
			      #!key (force-create #f)(allow-write-cache #t)(wait-a-minute #f))
  (let* ((use-cache    (common:use-cache?))
	 (cache-path   (tests:get-test-path-from-environment))
	 (cache-file   (and cache-path (conc cache-path "/.testconfig")))
	 (cache-exists (and cache-file
			    (not force-create)  ;; if force-create then pretend there is no cache to read
			    (common:file-exists? cache-file)))
	 (cached-dat   (if (and (not force-create)
				cache-exists
				use-cache)
			   (handle-exceptions
			    exn
			    #f ;; any issues, just give up with the cached version and re-read
			    (configf:read-alist cache-file))
			   #f))
         (test-full-name (if (and item-path (not (string-null? item-path)))
                             (conc test-name "/" item-path)
                             test-name)))
    (if cached-dat
	cached-dat
	(let ((dat (hash-table-ref/default *testconfigs* test-full-name #f)))
	  (if (and  dat ;; have a locally cached version
		    (hash-table-ref/default dat "have fulldata" #f)) ;; marked as good data?
	      dat
	      ;; no cached data available
	      (let* ((treg         (or test-registry
				       (tests:get-all)))
		     (test-path    (or (hash-table-ref/default treg test-name #f)
                                       (let* ((local-tcdir (conc (getenv "MT_LINKTREE") "/"
                                                                 (getenv "MT_TARGET") "/"
                                                                 (getenv "MT_RUNNAME") "/"
                                                                 test-name "/" item-path))
                                              (local-tcfg (conc local-tcdir "/testconfig")))
                                         (if (common:file-exists? local-tcfg)
                                             local-tcdir
                                             #f))
				       (conc *toppath* "/tests/" test-name)))
		     (test-configf (conc test-path "/testconfig"))
		     (testexists   (let loopa ((tries-left 30))
                                     (cond
                                      (
                                       (and (common:file-exists? test-configf)(file-read-access? test-configf))
                                       #t)
                                      (
                                       (common:file-exists? test-configf)
                                       (debug:print 0 *default-log-port* "WARNING: Cannot read testconfig file: "test-configf)
                                       #f)
                                      (
                                       (and wait-a-minute (> tries-left 0))
                                       (thread-sleep! 10)
                                       (debug:print 0 *default-log-port* "WARNING: testconfig file does not exist: "test-configf" will retry in 10 seconds.  Tries left: "tries-left) ;; BB: this fires
                                       (loopa (sub1 tries-left)))
                                      (else
                                       (debug:print 0 *default-log-port* "WARNING: testconfig file does not exist: "test-configf) ;; BB: this fires
                                       #f))))
		     (tcfg         (if testexists
				       (read-config test-configf #f system-allowed
						    environ-patt: (if system-allowed
								      "pre-launch-env-vars"
								      #f))
				       #f)))
		(if (and tcfg cache-file) (hash-table-set! tcfg "have fulldata" #t)) ;; mark this as fully read data
		(if tcfg (hash-table-set! *testconfigs* test-full-name tcfg))
		(if (and testexists
			 cache-file
			 (file-write-access? cache-path)
			 allow-write-cache)
		    (let ((tpath (conc cache-path "/.testconfig")))
		      (debug:print-info 1 *default-log-port* "Caching testconfig for " test-name " in " tpath)
                      (if (and tcfg (not (common:in-running-test?)))
                          (configf:write-alist tcfg tpath))))
		tcfg))))))
  
;; sort tests by priority and waiton
;; Move test specific stuff to a test unit FIXME one of these days
(define (tests:sort-by-priority-and-waiton test-records)
  (if (eq? (hash-table-size test-records) 0)
      '()
      (let* ((mungepriority (lambda (priority)
			      (if priority
				  (let ((tmp (any->number priority)))
				    (if tmp tmp (begin (debug:print-error 0 *default-log-port* "bad priority value " priority ", using 0") 0)))
				  0)))
	     (all-tests      (hash-table-keys test-records))
	     (all-waited-on  (let loop ((hed (car all-tests))
					(tal (cdr all-tests))
					(res '()))
			       (let* ((trec    (hash-table-ref test-records hed))
				      (waitons (or (tests:testqueue-get-waitons trec) '())))
				 (if (null? tal)
				     (append res waitons)
				     (loop (car tal)(cdr tal)(append res waitons))))))
	     (sort-fn1 
	      (lambda (a b)
		(let* ((a-record   (hash-table-ref test-records a))
		       (b-record   (hash-table-ref test-records b))
		       (a-waitons  (or (tests:testqueue-get-waitons a-record) '()))
		       (b-waitons  (or (tests:testqueue-get-waitons b-record) '()))
		       (a-config   (tests:testqueue-get-testconfig  a-record))
		       (b-config   (tests:testqueue-get-testconfig  b-record))
		       (a-raw-pri  (configf:lookup a-config "requirements" "priority"))
		       (b-raw-pri  (configf:lookup b-config "requirements" "priority"))
		       (a-priority (mungepriority a-raw-pri))
		       (b-priority (mungepriority b-raw-pri)))
		  (tests:testqueue-set-priority! a-record a-priority)
		  (tests:testqueue-set-priority! b-record b-priority)
		  ;; (debug:print 0 *default-log-port* "a=" a ", b=" b ", a-waitons=" a-waitons ", b-waitons=" b-waitons)
		  (cond
		   ;; is 
		   ((member a b-waitons)          ;; is b waiting on a?
		    ;; (debug:print 0 *default-log-port* "case1")
		    #t)
		   ((member b a-waitons)          ;; is a waiting on b?
		    ;; (debug:print 0 *default-log-port* "case2")
		    #f)
		   ((and (not (null? a-waitons))  ;; both have waitons - do not disturb
			 (not (null? b-waitons)))
		    ;; (debug:print 0 *default-log-port* "case2.1")
		    #t)
		   ((and (null? a-waitons)        ;; no waitons for a but b has waitons
			 (not (null? b-waitons)))
		    ;; (debug:print 0 *default-log-port* "case3")
		    #f)
		   ((and (not (null? a-waitons))  ;; a has waitons but b does not
			 (null? b-waitons)) 
		    ;; (debug:print 0 *default-log-port* "case4")
		    #t)
		   ((not (eq? a-priority b-priority)) ;; use
		    (> a-priority b-priority))
		   (else
		    ;; (debug:print 0 *default-log-port* "case5")
		    (string>? a b))))))
	     
	     (sort-fn2
	      (lambda (a b)
		(> (mungepriority (tests:testqueue-get-priority (hash-table-ref test-records a)))
		   (mungepriority (tests:testqueue-get-priority (hash-table-ref test-records b)))))))
	;; (let ((dot-res (tests:run-dot (tests:tests->dot test-records) "plain")))
	;;   (debug:print "dot-res=" dot-res))
	;; (let ((data (map cdr (filter
	;;     		  (lambda (x)(equal? "node" (car x)))
	;;     		  (map string-split (tests:easy-dot test-records "plain"))))))
	;;   (map car (sort data (lambda (a b)
	;;     		    (> (string->number (caddr a))(string->number (caddr b)))))))
	;; ))
	(sort all-tests sort-fn1)))) ;; avoid dealing with deleted tests, look at the hash table

(define (tests:easy-dot test-records outtype)
  (let-values (((fd temp-path) (file-mkstemp (conc "/tmp/" (current-user-name) ".XXXXXX"))))
    (let ((all-testnames (hash-table-keys test-records))
	  (temp-port     (open-output-file* fd)))
      ;; (format temp-port "This file is ~A.~%" temp-path)
      (format temp-port "digraph tests {\n")
      (format temp-port "  size=4,8\n")
      ;; (format temp-port "   splines=none\n")
      (for-each
       (lambda (testname)
	 (let* ((testrec (hash-table-ref test-records testname))
		(waitons (or (tests:testqueue-get-waitons testrec) '())))
	   (for-each
	    (lambda (waiton)
	      (format temp-port (conc "   " waiton " -> " testname " [splines=ortho]\n")))
	    waitons)))
       all-testnames)
      (format temp-port "}\n")
      (close-output-port temp-port)
      (with-input-from-pipe
       (conc "env -i PATH=$PATH dot -T" outtype " < " temp-path)
       (lambda ()
	 (let ((res (read-lines)))
	   ;; (delete-file temp-path)
	   res))))))

(define (tests:write-dot-file test-records fname sizex sizey)
  (if (file-write-access? (pathname-directory fname))
      (with-output-to-file fname
	(lambda ()
	  (map print (tests:tests->dot test-records sizex sizey))))))

(define (tests:tests->dot test-records sizex sizey)
  (let ((all-testnames (hash-table-keys test-records)))
    (if (null? all-testnames)
	'()
	(let loop ((hed (car all-testnames))
		   (tal (cdr all-testnames))
		   (res (list "digraph tests {"
			      (conc " size=\"" (or sizex 11) "," (or sizey 11) "\";")
			      " ratio=0.95;"
			      )))
	  (let* ((testrec (hash-table-ref test-records hed))
		 (waitons (or (tests:testqueue-get-waitons testrec) '()))
		 (newres  (append res
				  (if (null? waitons)
				      (list (conc "   \"" hed "\" [shape=box];"))
				      (map (lambda (waiton)
					     (conc "   \"" waiton "\" -> \"" hed "\" [shape=box];"))
					   waitons)
				      ))))
	    (if (null? tal)
		(append newres (list "}"))
		(loop (car tal)(cdr tal) newres)
		))))))

;; (tests:run-dot (list "digraph tests {" "a -> b" "}") "plain")

(define (tests:run-dot indat outtype) ;; outtype is plain, fig, dot, etc. http://www.graphviz.org/content/output-formats
  (let-values (((inp oup pid)(process "env -i PATH=$PATH dot" (list "-T" outtype))))
    (with-output-to-port oup
      (lambda ()
	(map print indat)))
    (close-output-port oup)
    (let ((res (with-input-from-port inp
		 (lambda ()
		   (read-lines)))))
      (close-input-port inp)
      res)))

;; read data from tmp file or create if not exists
;; if exists regen in background
;;
(define (tests:lazy-dot testrecords  outtype sizex sizey)
  (let ((dfile (conc "/tmp/." (current-user-name) "-" (server:mk-signature) ".dot"))
	(fname (conc "/tmp/." (current-user-name) "-" (server:mk-signature) ".dotdat")))
    (tests:write-dot-file testrecords dfile sizex sizey)
    (if (common:file-exists? fname)
	(let ((res (with-input-from-file fname
		     (lambda ()
		       (read-lines)))))
	  (system (conc "env -i PATH=$PATH dot -T " outtype " < " dfile " > " fname "&"))
	  res)
	(begin
	  (system (conc "env -i PATH=$PATH dot -T " outtype " < " dfile " > " fname))
	  (with-input-from-file fname
	    (lambda ()
	      (read-lines)))))))
	  

;; for each test:
;;   
(define (tests:filter-non-runnable run-id testkeynames testrecordshash)
  (let ((runnables '()))
    (for-each
     (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     (rmt:get-test-id run-id test-name item-path))
	      (tdat        (rmt:get-testinfo-state-status run-id 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 (rmt:get-test-id run-id waiton ""))
				      (wtdat          (rmt:get-testinfo-state-status run-id 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" "ABORT")))
					 (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)
				 ;;        	 '("INCOMPETE")))
				     (set! keep-test #f)))) ;; no point in running this one again
			     waitons))))
	 (if keep-test (set! runnables (cons testkeyname runnables)))))
     testkeynames)
    runnables))

;;======================================================================
;; refactoring this block into tests:get-full-data from line 263 of runs.scm
;;======================================================================
;; hed is the test name
;; test-records is a hash of test-name => test record
(define (tests:get-full-data test-names test-records required-tests all-tests-registry)
  (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-info 4 *default-log-port* "hed=" hed " at top of loop")
        ;; don't know item-path at this time, let the testconfig get the top level testconfig
	(let* ((config  (tests:get-testconfig hed #f all-tests-registry 'return-procs))
	       (waitons (let ((instr (if config 
					 (configf:lookup config "requirements" "waiton")
					 (begin ;; No config means this is a non-existant test
					   (debug:print-error 0 *default-log-port* "non-existent required test \"" hed "\", grep through your testconfigs to find and remove or create the test. Discarding and continuing.")
					     ""))))
			  (debug:print-info 8 *default-log-port* "waitons string is " instr)
			  (string-split (cond
					 ((procedure? instr)
					  (let ((res (instr)))
					    (debug:print-info 8 *default-log-port* "waiton procedure results in string " res " for test " hed)
					    res))
					 ((string? instr)     instr)
					 (else 
					  ;; NOTE: This is actually the case of *no* waitons! ;; (debug:print-error 0 *default-log-port* "something went wrong in processing waitons for test " hed)
					  ""))))))
	  (if (not config) ;; this is a non-existant test called in a waiton. 
	      (if (null? tal)
		  test-records
		  (loop (car tal)(cdr tal)))
	      (begin
		(debug:print-info 8 *default-log-port* "waitons: " waitons)
		;; check for hed in waitons => this would be circular, remove it and issue an
		;; error
		(if (member hed waitons)
		    (begin
		      (debug:print-error 0 *default-log-port* "test " hed " has listed itself as a waiton, please correct this!")
		      (set! waitons (filter (lambda (x)(not (equal? x hed))) waitons))))
		
		;; (items   (items:get-items-from-config config)))
		(if (not (hash-table-ref/default test-records hed #f))
		    (hash-table-set! test-records
				     hed (vector hed     ;; 0
						 config  ;; 1
						 waitons ;; 2
						 (configf:lookup config "requirements" "priority")     ;; priority 3
						 (let ((items      (hash-table-ref/default config "items" #f)) ;; items 4
						       (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-info 4 *default-log-port* "items is a procedure, will calc later")
						     items)            ;; calc later
						    ((procedure? itemstable)
						     (debug:print-info 4 *default-log-port* "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-info 4 *default-log-port* "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))
		      test-records))))))))

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

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

(define (test-get-kill-request run-id test-id) ;; run-id test-name itemdat)
  (let* ((testdat   (rmt:get-test-info-by-id run-id test-id)))
    (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 run-id test-id cpuload diskfree minutes uname hostname)
  (rmt:general-call 'update-test-rundat run-id test-id (current-seconds) (or cpuload -1)(or diskfree -1) -1 (or minutes -1))
  (if (and cpuload diskfree)
      (rmt:general-call 'update-cpuload-diskfree run-id cpuload diskfree test-id))
  (if minutes 
      (rmt:general-call 'update-run-duration run-id minutes test-id))
  (if (and uname hostname)
      (rmt:general-call 'update-uname-host run-id uname hostname test-id)))
  
;; This one is for running with no db access (i.e. via rmt: internally)
(define (tests:set-full-meta-info db test-id run-id minutes work-area remtries)
;; (define (tests:set-full-meta-info test-id run-id minutes work-area)
;;  (let ((remtries 10))
  (let* ((cpuload  (get-cpu-load))
	 (diskfree (get-df (current-directory)))
	 (uname    (get-uname "-srvpio"))
	 (hostname (get-host-name)))
    (tests:update-central-meta-info run-id test-id cpuload diskfree minutes uname hostname)))
    
;; (define (tests:set-partial-meta-info test-id run-id minutes work-area)
#;(define (tests:set-partial-meta-info test-id run-id minutes work-area remtries)
  (let* ((cpuload  (get-cpu-load))
	 (diskfree (get-df (current-directory)))
	 (remtries 10))
    (handle-exceptions
     exn
     (if (> remtries 0)
	 (begin
	   (print-call-chain (current-error-port))
	   (debug:print-info 0 *default-log-port* "WARNING: failed to set meta info. Will try " remtries " more times")
	   (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-error 0 *default-log-port* "tried for over a minute to update meta info and failed. Giving up")
	   (debug:print 0 *default-log-port* "EXCEPTION: database probably overloaded or unreadable.")
	   (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
	   (debug:print 5 *default-log-port* "exn=" (condition->list exn))
	   (debug:print 0 *default-log-port* " status:  " ((condition-property-accessor 'sqlite3 'status) exn))
	   (print-call-chain (current-error-port))))
     (tests:update-testdat-meta-info db test-id work-area cpuload diskfree minutes)
  )))
	 
;;======================================================================
;; A R C H I V I N G
;;======================================================================

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

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