Megatest

;;======================================================================
;; 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/>.

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

(use srfi-1 data-structures posix regex-case (prefix base64 base64:)
     format dot-locking csv-xml z3 udp ;; sql-de-lite
     hostinfo md5 message-digest typed-records directory-utils stack
     matchable regex posix (srfi 18) extras ;; tcp 

     (prefix nanomsg nmsg:)


     (prefix sqlite3 sqlite3:)
     pkts (prefix dbi dbi:)
     )

(declare (unit common))
(declare (uses commonmod))
(import commonmod)

(include "common_records.scm")


;; (require-library margs)
;; (include "margs.scm")

;; (define old-exit exit)
;; 
;; (define (exit . code)
;;   (if (null? code)
;;       (old-exit)
;;       (old-exit code)))


;; execute thunk, return value.  If exception thrown, trap exception, return #f, and emit nonfatal condition note to *default-log-port* .
;; arguments - thunk, message
(define (common:fail-safe thunk warning-message-on-exception)
  (handle-exceptions
   exn
   (begin
     (debug:print-info 0 *default-log-port* "notable but nonfatal condition - "warning-message-on-exception)
     (debug:print-info 0 *default-log-port*
                       (string-substitute "\n?Error:" "nonfatal condition:"
                                          (with-output-to-string
                                            (lambda ()
                                              (print-error-message exn) ))))
     (debug:print-info 0 *default-log-port* "    -- continuing after nonfatal condition...")
     #f)
   (thunk)))

(define getenv get-environment-variable)
(define (safe-setenv key val)
  (if (or (substring-index "!" key) (substring-index ":" key)) ;; variables containing : are for internal use and cannot be environment variables.
      (debug:print-error 4 *default-log-port* "skip setting internal use only variables containing \":\" or starting with \"!\"")
      (if (and (string? val)
	       (string? key))
	  (handle-exceptions
	      exn
	      (debug:print-error 0 *default-log-port* "bad value for setenv, key=" key ", value=" val)
	    (setenv key val))
	  (debug:print-error 0 *default-log-port* "bad value for setenv, key=" key ", value=" val))))

(define home (getenv "HOME"))
(define user (getenv "USER"))


;; returns list of fd count, socket count
(define (get-file-descriptor-count #!key  (pid (current-process-id )))
  (list
    (length (glob (conc "/proc/" pid "/fd/*")))
    (length  (filter identity (map socket? (glob (conc "/proc/" pid "/fd/*")))))
  )
)

  

;; GLOBALS

;; CONTEXTS
(defstruct cxt
  (taskdb #f)
  (cmutex (make-mutex)))
;; (define *contexts* (make-hash-table))
;; (define *context-mutex* (make-mutex))

;; A hash table that can be accessed by #{scheme ...} calls in
;; config files. Allows communicating between confgs
;;
(define *user-hash-data* (make-hash-table))

(define *db-keys* #f)

(define *pkts-info*    (make-hash-table)) ;; store stuff like the last parent here
(define *configinfo*   #f)   ;; raw results from setup, includes toppath and table from megatest.config
(define *runconfigdat* #f)   ;; run configs data
(define *configdat*    #f)   ;; megatest.config data
(define *configstatus* #f)   ;; status of data; 'fulldata : all processing done, #f : no data yet, 'partialdata : partial read done
(define *toppath*      #f)
(define *already-seen-runconfig-info* #f)

(define *test-meta-updated* (make-hash-table))
(define *globalexitstatus*  0) ;; attempt to work around possible thread issues
(define *passnum*           0) ;; when running track calls to run-tests or similar
;; (define *alt-log-file* #f)  ;; used by -log
(define *common:denoise*    (make-hash-table)) ;; for low noise printing
(define *default-log-port*  (current-error-port))
(define *time-zero* (current-seconds)) ;; for the watchdog
(define *default-area-tag* "local")

;; DATABASE
(define *dbstruct-db*         #f) ;; used to cache the dbstruct in db:setup. Goal is to remove this.
;; db stats
(define *db-stats*            (make-hash-table)) ;; hash of vectors < count duration-total >
(define *db-stats-mutex*      (make-mutex))
;; db access

(define *task-db*             #f) ;; (vector db path-to-db)
(define *db-access-allowed*   #t) ;; flag to allow access
(define *db-access-mutex*     (make-mutex))
(define *db-transaction-mutex* (make-mutex))
(define *db-cache-path*       #f)
(define *db-with-db-mutex*    (make-mutex))
(define *db-api-call-time*    (make-hash-table)) ;; hash of command => (list of times)
;; no sync db
(define *no-sync-db*          #f)

;; SERVER
(define *my-client-signature* #f)
(define *transport-type*    'http)             ;; override with [server] transport http|rpc|nmsg
(define *runremote*         #f)                ;; if set up for server communication this will hold <host port>
;; (define *max-cache-size*    0)
(define *logged-in-clients* (make-hash-table))
(define *server-id*         #f)
(define *server-info*       #f)  ;; good candidate for easily convert to non-global
(define *time-to-exit*      #f)
(define *server-run*        #t)
(define *run-id*            #f)
(define *server-kind-run*   (make-hash-table))
(define *home-host*         #f)
;; (define *total-non-write-delay* 0)
(define *heartbeat-mutex*   (make-mutex))
(define *api-process-request-count* 0)
(define *max-api-process-requests* 0)
(define *server-overloaded*  #f)

;; client
(define *rmt-mutex*         (make-mutex))     ;; remote access calls mutex 

;; RPC transport
(define *rpc:listener*      #f)

;; KEY info
(define *target*            (make-hash-table)) ;; cache the target here; target is keyval1/keyval2/.../keyvalN
(define *keys*              (make-hash-table)) ;; cache the keys here
(define *keyvals*           (make-hash-table))
(define *toptest-paths*     (make-hash-table)) ;; cache toptest path settings here
(define *test-paths*        (make-hash-table)) ;; cache test-id to test run paths here
(define *test-ids*          (make-hash-table)) ;; cache run-id, testname, and item-path => test-id
(define *test-info*         (make-hash-table)) ;; cache the test info records, update the state, status, run_duration etc. from testdat.db

(define *run-info-cache*     (make-hash-table)) ;; run info is stable, no need to reget
(define *launch-setup-mutex* (make-mutex))     ;; need to be able to call launch:setup often so mutex it and re-call the real deal only if *toppath* not set
(define *homehost-mutex*     (make-mutex))

;; Miscellaneous
(define *triggers-mutex*     (make-mutex))     ;; block overlapping processing of triggers

(use posix-extras pathname-expand files)

;; this plugs a hole in posix-extras in recent chicken versions > 4.9)
(let-values (( (chicken-release-number chicken-major-version)
               (apply values
                      (map string->number
                           (take
                            (string-split (chicken-version) ".")
                            2)))))
  (let ((resolve-pathname-broken?
         (or (> chicken-release-number 4)
             (and (eq? 4 chicken-release-number) (> chicken-major-version 9)))))
    (if resolve-pathname-broken?
        (define ##sys#expand-home-path pathname-expand))))
      
(define (realpath x) (resolve-pathname  (pathname-expand (or x "/dev/null")) ))

(define (common:get-this-exe-fullpath #!key (argv (argv)))
  (let* ((this-script
          (cond
           ((and (> (length argv) 2)
                 (string-match "^(.*/csi|csi)$" (car argv))
                 (string-match "^-(s|ss|sx|script)$" (cadr argv)))
            (caddr argv))
           (else (car argv))))
         (fullpath (realpath this-script)))
    fullpath))
(define *common:this-exe-fullpath* (common:get-this-exe-fullpath))
(define *common:this-exe-dir*      (pathname-directory *common:this-exe-fullpath*))
(define *common:this-exe-name*     (pathname-strip-directory *common:this-exe-fullpath*))

(define *common:logpro-exit-code->status-sym-alist*
  '( ( 0 . pass )
     ( 1 . fail )
     ( 2 . warn )
     ( 3 . check )
     ( 4 . waived )
     ( 5 . abort )
     ( 6 . skip )))

(define (common:logpro-exit-code->status-sym exit-code)
  (or (alist-ref exit-code *common:logpro-exit-code->status-sym-alist*) 'fail))

(define (common:worse-status-sym ss1 ss2)
  (let loop ((status-syms-remaining '(abort fail check skip warn waived pass)))
    (cond
     ((null? status-syms-remaining)
      'fail)
     ((eq? (car status-syms-remaining) ss1)
      ss1)
     ((eq? (car status-syms-remaining) ss2)
      ss2)
     (else
      (loop (cdr status-syms-remaining))))))

(define (common:steps-can-proceed-given-status-sym status-sym)
  (if (member status-sym '(warn waived pass))
      #t
      #f))

(define (status-sym->string status-sym)
  (case status-sym
      ((pass) "PASS")
    ((fail) "FAIL")
    ((warn) "WARN")
    ((check) "CHECK")
    ((waived) "WAIVED")
    ((abort) "ABORT")
    ((skip) "SKIP")
    (else "FAIL")))

(define (common:logpro-exit-code->test-status exit-code)
  (status-sym->string (common:logpro-exit-code->status-sym exit-code)))

(defstruct remote
  (hh-dat            (common:get-homehost)) ;; homehost record ( addr . hhflag )
  (server-url        (if *toppath* (server:check-if-running *toppath*))) ;; (server:check-if-running *toppath*) #f))
  (last-server-check 0)  ;; last time we checked to see if the server was alive
  (conndat           #f)
  (transport         *transport-type*)
  (server-timeout    (server:expiration-timeout))
  (force-server      #f)
  (ro-mode           #f)  
  (ro-mode-checked   #f)) ;; flag that indicates we have checked for ro-mode

;; launching and hosts
(defstruct host
  (reachable    #f)

(define (common:get-last-run-version-number)
  (string->number 
   (substring (common:get-last-run-version) 0 6)))

(define (common:set-last-run-version)
  (rmt:set-var "MEGATEST_VERSION" (common:version-signature)))

;; postive number if megatest version > db version
;; negative number if megatest version < db version
(define (common:version-db-delta)
         (- megatest-version (common:get-last-run-version-number)))

(define (common:version-changed?)
  (not (equal? (common:get-last-run-version)
               (common:version-signature))))

(define (common:api-changed?)
  (not (equal? (substring (->string megatest-version) 0 4)
               (substring (conc (common:get-last-run-version)) 0 4))))


(define (common:get-sync-lock-filepath)
  (let* ((tmp-area     (common:get-db-tmp-area))
         (lockfile     (conc tmp-area "/megatest.db.sync-lock")))
    lockfile))
    
;; Move me elsewhere ...
;; RADT => Why do we meed the version check here, this is called only if version misma
;;
(define (common:cleanup-db dbstruct #!key (full #f))
  (apply db:multi-db-sync 
   dbstruct
   'schema
   ;; 'new2old
   'killservers

   'adj-target
   ;; 'old2new
   'new2old
   ;; (if full
       '(dejunk)
       ;; '())
       )
  (if (common:api-changed?)
      (common:set-last-run-version)))

(define (common:snapshot-file filepath #!key (subdir  ".") )
  (if (file-exists? filepath)
      (let* ((age-sec  (lambda (file)
                         (if (file-exists? file)
                             (- (current-seconds) (file-modification-time file))
                             1000000000))) ;; return really old value if file doesn't exist.  we want to clobber it if old or not exist.
             (ok-flag  #t)
             (age-mins (lambda (file) (/ (age-sec   file) 60)))
             (age-hrs  (lambda (file) (/ (age-mins file) 60)))
             (age-days (lambda (file) (/ (age-hrs   file) 24)))
             (age-wks  (lambda (file) (/ (age-days  file) 7)))
             (docmd    (lambda (cmd)
                         (cond
                          (ok-flag
                           (let ((res (system cmd)))
                             (cond
                              ((eq? 0 res)
                               #t)
                              (else
                               (set! ok-flag #f)
                               (debug:print 0 *default-log-port* "ERROR: ["(common:human-time)"] Command failed with exit code "
                                            (if (< res 0)
                                                res
                                                (/ res 8)) " ["cmd"]" )
                               #f))))
                          (else
                           (debug:print 0 *default-log-port* "ERROR: ["(common:human-time)"] Not runnining command due to prior error. ["cmd"]")
                           #f))))
             (copy     (lambda (src dest) (docmd (conc "/bin/cp '"src"' '"dest"'"))))
             (copy+zip (lambda (src dest) (docmd (conc "gzip -c - < '"src"' > '"dest"'"))))
             (fullpath (realpath filepath))
             (basedir  (pathname-directory fullpath))
             (basefile (pathname-strip-directory fullpath))
             ;;(prevfile (conc filepath ".prev.gz"))
             (minsfile (conc basedir "/" subdir "/" basefile ".mins.gz"))
             (hrsfile  (conc basedir "/" subdir "/" basefile ".hrs.gz"))
             (daysfile (conc basedir "/" subdir "/" basefile ".days.gz"))
             (wksfile  (conc basedir "/" subdir "/" basefile ".weeks.gz")))

        ;; create subdir it not exists
        (if (not (directory-exists? (conc basedir "/" subdir)))
            (docmd (conc "/bin/mkdir -p '"(conc basedir "/" subdir)"'")))
        
        ;; copy&zip <file> to <file>.mins if not exists
        (if (not (file-exists? minsfile))
            (copy+zip filepath minsfile))
        ;; copy <file>.mins to <file>.hrs if not exists
        (if (not (file-exists? hrsfile))
            (copy minsfile hrsfile))
        ;; copy <file>.hrs to <file>.days if not exists
        (if (not (file-exists? daysfile))
            (copy hrsfile daysfile))
        ;; copy <file>.days to <file>.weeks if not exists
        (if (not (file-exists? wksfile))
            (copy daysfile wksfile))


        ;; if age(<file>.mins.gz) >= 1h:
        ;;     copy <file>.mins.gz <file>.hrs.gz
        ;;     copy <prev file> <file>.mins.gz
        (when (>= (age-mins minsfile) 1)
          (copy minsfile hrsfile)
          (copy+zip filepath  minsfile))
        
        ;; if age(<file>.hrs.gz) >= 1d:
        ;;     copy <file>.hrs.gz <file>.days.gz
        ;;     copy <file>.mins.gz <file>.hrs.gz
        (when (>= (age-days hrsfile) 1)
          (copy hrsfile daysfile)
          (copy minsfile hrsfile))
        
        ;; if age(<file>.days.gz) >= 1w:
        ;;     copy <file>.days.gz <file>.weeks.gz
        ;;     copy <file>.hrs.gz <file>.days.gz
        (when (>= (age-wks daysfile) 1)
          (copy daysfile wksfile)
          (copy hrsfile daysfile))
        #t)
      #f))
  
        
        
;; Rotate logs, logic: 
;;                 if > 500k and older than 1 week:
;;                     remove previous compressed log and compress this log
;; WARNING: This proc operates assuming that it is in the directory above the
;;          logs directory you wish to log-rotate.
;;
(define (common:rotate-logs)
  (let* ((all-files (make-hash-table))
	 (stats     (make-hash-table))
	 (inc-stat  (lambda (key)
		      (hash-table-set! stats key (+ (hash-table-ref/default stats key 0) 1))))
	(max-allowed (string->number (or (configf:lookup *configdat* "setup" "max-logfiles") "300")))) ;; name -> age
    (if (not (directory-exists? "logs"))(create-directory "logs"))
    (directory-fold 
     (lambda (file rem)
       (handle-exceptions
	exn
	(begin
	  (debug:print-info 2 *default-log-port* "unable to rotate log " file ", probably handled by another process, this is safe to ignore.")
	  (debug:print 2 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
	  ;; (print-call-chain (current-error-port)) ;; 
	  )
	(let* ((fullname  (conc "logs/" file))
	       (mod-time  (file-modification-time fullname))
	       (file-age  (- (current-seconds) mod-time)))
	  (hash-table-set! all-files file mod-time)
	  (if (or (and (string-match "^.*.log" file)
		       (> (file-size fullname) 200000))
		  (and (string-match "^server-.*.log" file)
		       (> (- (current-seconds) (file-modification-time fullname))
			  (* 8 60 60))))
	      (let ((gzfile (conc fullname ".gz")))
		(if (common:file-exists? gzfile)
		    (begin
		      (debug:print-info 0 *default-log-port* "removing " gzfile)
		      (delete-file* gzfile)
		      (hash-table-delete!  all-files gzfile) ;; needed?
		      ))
		(debug:print-info 0 *default-log-port* "compressing " file)
		(system (conc "gzip " fullname))
		(inc-stat "gzipped")
		(hash-table-set! all-files (conc file ".gz") file-age)  ;; add the .gz file and remove the base file
		(hash-table-delete! all-files file)
		)
	      (if (> file-age (* (string->number (or (configf:lookup *configdat* "setup" "log-expire-days") "30")) 24 3600))
		  (handle-exceptions
		   exn
		   #f
		   (if (directory? fullname)
		       (begin
			 (debug:print-error 0 *default-log-port* fullname " in logs directory is a directory! Cannot rotate it, it is best to not put subdirectories in the logs dir.")
			 (inc-stat "directories"))
		       (begin
			 (delete-file* fullname)
			 (inc-stat "deleted")))
		   (hash-table-delete! all-files file)))))))
     '()
     "logs")
    (for-each
     (lambda (category)
       (let ((quant (hash-table-ref/default stats category 0)))
	 (if (> quant 0)
	     (debug:print-info 0 *default-log-port* category " log files: " quant))))
     `("deleted" "gzipped" "directories"))
    (let ((num-logs (hash-table-size all-files)))
      (if (> num-logs max-allowed) ;; because NFS => don't let number of logs exceed 300
	  (let ((files (take (sort (hash-table-keys all-files)
				   (lambda (a b)
				     (< (hash-table-ref all-files a)(hash-table-ref all-files b))))
			     (- num-logs max-allowed))))
	    (for-each
	     (lambda (file)
	       (let* ((fullname (conc "logs/" file)))
		 (if (directory? fullname)
		     (debug:print-error 0 *default-log-port* fullname " in logs directory is a directory! Cannot rotate it, it is best to not put subdirectories in the logs dir.")
		     (handle-exceptions
		      exn
		      (debug:print-error 0 *default-log-port* "failed to remove " fullname)
		      (delete-file* fullname)))))
	     files)
	    (debug:print-info 0 *default-log-port* "Deleted " (length files) " files from logs, keeping " max-allowed " files."))))))
  
;; Force a megatest cleanup-db if version is changed and skip-version-check not specified
;; Do NOT check if not on homehost!
;;
(define (common:exit-on-version-changed)
  (if (common:on-homehost?)
      (if (common:api-changed?)
	  (let* ((mtconf (conc (get-environment-variable "MT_RUN_AREA_HOME") "/megatest.config"))
                (dbfile  (conc (get-environment-variable "MT_RUN_AREA_HOME") "/megatest.db"))
                (read-only (not (file-write-access? dbfile)))
                (dbstruct (db:setup #t)))
	    (debug:print 0 *default-log-port*
			 "WARNING: Version mismatch!\n"
			 "   expected: " (common:version-signature) "\n"
			 "   got:      " (common:get-last-run-version))
            (cond
             ((get-environment-variable "MT_SKIP_DB_MIGRATE") #t)
             ((and (common:file-exists? mtconf) (common:file-exists? dbfile) (not read-only)
                   (eq? (current-user-id)(file-owner mtconf))) ;; safe to run -cleanup-db
              (debug:print 0 *default-log-port* "   I see you are the owner of megatest.config, attempting to cleanup and reset to new version")
              (handle-exceptions
               exn
               (begin
                 (debug:print 0 *default-log-port* "Failed to switch versions.")
                 (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
                 (print-call-chain (current-error-port))
                 (exit 1))
               (common:cleanup-db dbstruct)))
             ((not (common:file-exists? mtconf))
              (debug:print 0 *default-log-port* "   megatest.config does not exist in this area.  Cannot proceed with megatest version migration.")
              (exit 1))
             ((not (common:file-exists? dbfile))
              (debug:print 0 *default-log-port* "   megatest.db does not exist in this area.  Cannot proceed with megatest version migration.")
              (exit 1))
             ((not (eq? (current-user-id)(file-owner mtconf)))
              (debug:print 0 *default-log-port* "   You do not own megatest.db in this area.  Cannot proceed with megatest version migration.")
              (exit 1))
             (read-only
              (debug:print 0 *default-log-port* "   You have read-only access to this area.  Cannot proceed with megatest version migration.")
              (exit 1))
             (else
              (debug:print 0 *default-log-port* " to switch versions you can run: \"megatest -cleanup-db\"")
              (exit 1)))))))
;;      (begin
;;	(debug:print 0 *default-log-port* "ERROR: cannot migrate version unless on homehost. Exiting.")
;;	(exit 1))))

;;======================================================================
;; S P A R S E   A R R A Y S
;;======================================================================

(define (make-sparse-array)
  (let ((a (make-sparse-vector)))

	  (common:to-alist (cdr dat))))
   ((hash-table? dat)
    (map common:to-alist (hash-table->alist dat)))
   (else
    (if dat
	dat
	""))))

(define (common:alist-ref/default key alist default)
  (or (alist-ref key alist) default))

(define (common:low-noise-print waitval . keys)
  (let* ((key      (string-intersperse (map conc keys) "-" ))
	 (lasttime (hash-table-ref/default *common:denoise* key 0))
	 (currtime (current-seconds)))
    (if (> (- currtime lasttime) waitval)
	(begin

   (read (open-input-string (z3:decode-buffer (base64:base64-decode instr))))))

;; dot-locking egg seems not to work, using this for now
;; if lock is older than expire-time then remove it and try again
;; to get the lock
;;
(define (common:simple-file-lock fname #!key (expire-time 300))
  (let ((fmod-time (handle-exceptions
		       ext
		     (current-seconds)
		     (file-modification-time fname))))
    (if (common:file-exists? fname)
	(if (> (- (current-seconds) fmod-time) expire-time)
	    (begin
	      (handle-exceptions exn #f (delete-file* fname))	
	      (common:simple-file-lock fname expire-time: expire-time))
	    #f)
	(let ((key-string (conc (get-host-name) "-" (current-process-id))))
	  (with-output-to-file fname
	    (lambda ()
	      (print key-string)))
	  (thread-sleep! 0.25)
	  (if (common:file-exists? fname)
	      (handle-exceptions exn
                #f 
                (with-input-from-file fname
	  	  (lambda ()
		    (equal? key-string (read-line)))))
	      #f)))))

(define (common:simple-file-lock-and-wait fname #!key (expire-time 300))
  (let ((end-time (+ expire-time (current-seconds))))
    (let loop ((got-lock (common:simple-file-lock fname expire-time: expire-time)))
      (if got-lock
	  #t

      #f ;; I don't really care why this failed (at least for now)
    (delete-file* fname)))

;;======================================================================
;; S T A T E S   A N D   S T A T U S E S
;;======================================================================

;; BBnote: *common:std-states* - dashboard filter control and test control state buttons defined here; used in set-fields-panel and dboard:make-controls
(define *common:std-states*   ;; for toggle buttons in dashboard
  '(
    (0 "ARCHIVED")
    (1 "STUCK")
    (2 "KILLREQ")
    (3 "KILLED")
    (4 "NOT_STARTED")
    (5 "COMPLETED")
    (6 "LAUNCHED")
    (7 "REMOTEHOSTSTART")
    (8 "RUNNING")
    ))

(define *common:dont-roll-up-states*
  '("DELETED"
    "REMOVING"
    "CLEANING"
    "ARCHIVE_REMOVING"
    ))

;; BBnote: *common:std-statuses* dashboard filter control and test control status buttons defined here; used in set-fields-panel and dboard:make-controls
;; note these statuses are sorted from better to worse.
;; This sort order is important to dcommon:status-compare3 and db:set-state-status-and-roll-up-items
(define *common:std-statuses*
  '(;; (0 "DELETED")  
    (1 "n/a")
    (2 "PASS")

    (3 "SKIP")
    (4 "WARN")
    (5 "WAIVED")
    (6 "CHECK")
    (7 "STUCK/DEAD")
    (8 "DEAD")
    (9 "FAIL")
    (10 "PREQ_FAIL")
    (11 "PREQ_DISCARDED")
    (12 "ABORT")))

(define *common:ended-states*       ;; states which indicate the test is stopped and will not proceed
  '("COMPLETED" "ARCHIVED" "KILLED" "KILLREQ" "STUCK" "INCOMPLETE" ))

(define *common:badly-ended-states* ;; these roll up as CHECK, i.e. results need to be checked
  '("KILLED" "KILLREQ" "STUCK" "INCOMPLETE" "DEAD"))

(define *common:well-ended-states* ;; an item's prereq in this state allows item to proceed
  '("PASS" "WARN" "CHECK" "WAIVED" "SKIP"))

;; BBnote: *common:running-states* used from db:set-state-status-and-roll-up-items
(define *common:running-states*     ;; test is either running or can be run
  '("RUNNING" "REMOTEHOSTSTART" "LAUNCHED" "STARTED"))

(define *common:cant-run-states*    ;; These are stopping conditions that prevent a test from being run
  '("COMPLETED" "KILLED" "UNKNOWN" "INCOMPLETE" "ARCHIVED"))

(define *common:not-started-ok-statuses* ;; if not one of these statuses when in not_started state treat as dead
  '("n/a" "na" "PASS" "FAIL" "WARN" "CHECK" "WAIVED" "DEAD" "SKIP"))

;; group tests into buckets corresponding to rollup
;;; Running, completed-pass,  completed-non-pass + worst status, not started.
;; filter out 
;(define (common:categorize-items-for-rollup in-tests)
;  (

(define (common:special-sort items order comp)
  (let ((items-order (map reverse order))
        (acomp       (or comp >)))
    (sort items
        (lambda (a b)
          (let ((a-num (cadr (or (assoc a items-order) '(0 0))))

(define (common:get-testsuite-name)
  (or (configf:lookup *configdat* "setup" "area-name") ;; megatest is a flexible tool, testsuite is too limiting a description.
      (configf:lookup *configdat* "setup" "testsuite" )
      (getenv "MT_TESTSUITE_NAME")
      (if (string? *toppath* )
          (pathname-file *toppath*)
          #f))) ;; (pathname-file (current-directory)))))

;; safe getting of toppath
(define (common:get-toppath areapath)
  (or *toppath*
      (if areapath
	  (begin
	    (set! *toppath* areapath)
	    (setenv "MT_RUN_AREA_HOME" areapath)
	    areapath)
	  #f)
      (if (getenv "MT_RUN_AREA_HOME")
	  (begin
	    (set! *toppath* (getenv "MT_RUN_AREA_HOME"))
	    *toppath*)
	  #f)
      ;; last resort, look for megatest.config
      (let loop ((thepath (realpath ".")))
	(if (file-exists? (conc thepath "/megatest.config"))
	    thepath
	    (if (equal? thepath "/")
		(begin
		  (debug:print-error 0 *default-log-port* "Unable to find megatest home directory.")
		  #f)
		(loop (pathname-directory thepath)))))
      ))

(define common:get-area-name common:get-testsuite-name)

(define (common:get-db-tmp-area . junk)
  (if *db-cache-path*
      *db-cache-path*
      (if *toppath* ;; common:get-create-writeable-dir
	  (handle-exceptions
	      exn
	      (begin
		(debug:print-error 0 *default-log-port* "Couldn't create path to " *db-cache-path*)
		(exit 1))
	    (let ((dbpath (common:get-create-writeable-dir
			   (list (conc "/tmp/" (current-user-name)
				       "/megatest_localdb/"
				       (common:get-testsuite-name) "/"
				       (string-translate *toppath* "/" ".")))))) ;;  #t))))
	      (set! *db-cache-path* dbpath)
	      dbpath))
	  #f)))

(define (common:get-area-path-signature)
  (message-digest-string (md5-primitive) *toppath*))

(define (common:get-signature str)
  (message-digest-string (md5-primitive) str))

;;======================================================================
;; 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 *wdnum* 0)
(define *wdnum*mutex (make-mutex))


(define (common:human-time)
  (time->string (seconds->local-time (current-seconds)) "%Y-%m-%d %H:%M:%S"))


;; currently the primary job of the watchdog is to run the sync back to megatest.db from the db in /tmp
;; if we are on the homehost and we are a server (by definition we are on the homehost if we are a server)
;;


(define (common:readonly-watchdog dbstruct)
  (thread-sleep! 0.05) ;; delay for startup
  (debug:print-info 13 *default-log-port* "common:readonly-watchdog entered.")
  ;; sync megatest.db to /tmp/.../megatst.db
  (let* ((sync-cool-off-duration   3)
        (golden-mtdb     (dbr:dbstruct-mtdb dbstruct))
        (golden-mtpath   (db:dbdat-get-path golden-mtdb))
        (tmp-mtdb        (dbr:dbstruct-tmpdb dbstruct))
        (tmp-mtpath      (db:dbdat-get-path tmp-mtdb)))
    (debug:print-info 0 *default-log-port* "Read-only periodic sync thread started.")
    (let loop ((last-sync-time 0))
      (debug:print-info 13 *default-log-port* "loop top tmp-mtpath="tmp-mtpath" golden-mtpath="golden-mtpath)
      (let* ((duration-since-last-sync (- (current-seconds) last-sync-time)))
        (debug:print-info 13 *default-log-port* "duration-since-last-sync="duration-since-last-sync)
        (if (and (not *time-to-exit*)
                 (< duration-since-last-sync sync-cool-off-duration))
            (thread-sleep! (- sync-cool-off-duration duration-since-last-sync)))
        (if (not *time-to-exit*)
            (let ((golden-mtdb-mtime (file-modification-time golden-mtpath))
                  (tmp-mtdb-mtime    (file-modification-time tmp-mtpath)))
	      (if (> golden-mtdb-mtime tmp-mtdb-mtime)
		  (if (< golden-mtdb-mtime (- (current-seconds) 3)) ;; file has NOT been touched in past three seconds, this way multiple servers won't fight to sync back
		      (let ((res (db:multi-db-sync dbstruct 'old2new)))
			(debug:print-info 13 *default-log-port* "rosync called, " res " records transferred."))))
              (loop (current-seconds)))
            #t)))
    (debug:print-info 0 *default-log-port* "Exiting readonly-watchdog timer, *time-to-exit* = " *time-to-exit*" pid="(current-process-id)" mtpath="golden-mtpath)))







































































;; TODO: for multiple areas, we will have multiple watchdogs; and multiple threads to manage
(define (common:watchdog)
  (debug:print-info 13 *default-log-port* "common:watchdog entered.")
  (if (launch:setup)
      (if (common:on-homehost?)
	  (let ((dbstruct (db:setup #t)))
	    (debug:print-info 13 *default-log-port* "after db:setup with dbstruct=" dbstruct)
	    (cond
	     ((dbr:dbstruct-read-only dbstruct)
	      (debug:print-info 13 *default-log-port* "loading read-only watchdog")
	      (common:readonly-watchdog dbstruct))
	     (else
	      (debug:print-info 13 *default-log-port* "loading writable-watchdog.")
              (let* ((syncer (or (configf:lookup *configdat* "server" "sync-method") "brute-force-sync")))
                (cond
                 ((equal? syncer "brute-force-sync")
                  (server:writable-watchdog-bruteforce dbstruct))
                 ((equal? syncer "delta-sync")
                  (server:writable-watchdog-deltasync dbstruct))
                 (else
                  (debug:print-error 0 *default-log-port* "Unknown server/sync-method specified ("syncer") - valid values are brute-force-sync and delta-sync.")
                  (exit 1)))
                ;;(debug:print 1 *default-log-port* "INFO: ["(common:human-time)"] Syncer started (method="syncer")")
                )))
	    (debug:print-info 13 *default-log-port* "watchdog done."))
	  (debug:print-info 13 *default-log-port* "no need for watchdog on non-homehost"))))


(define (std-exit-procedure)
  ;;(common:telemetry-log-close)
  (on-exit (lambda () 0))
  ;;(debug:print-info 13 *default-log-port* "std-exit-procedure called; *time-to-exit*="*time-to-exit*)
  (let ((no-hurry  (if *time-to-exit* ;; hurry up
		       #f
		       (begin
			 (set! *time-to-exit* #t)
			 #t))))
    (debug:print-info 4 *default-log-port* "starting exit process, finalizing databases.")
    (if (and no-hurry (debug:debug-mode 18))
	(rmt:print-db-stats))
    (let ((th1 (make-thread (lambda () ;; thread for cleaning up, give it five seconds
                              (if *dbstruct-db* (db:close-all *dbstruct-db*)) ;; one second allocated
			      (if *task-db*    
				  (let ((db (cdr *task-db*)))
				    (if (sqlite3:database? db)
					(begin
					  (sqlite3:interrupt! db)
					  (sqlite3:finalize! db #t)
					  ;; (vector-set! *task-db* 0 #f)
					  (set! *task-db* #f)))))
                              (http-client#close-all-connections!)
                              ;; (if (and *runremote*
                              ;;          (remote-conndat *runremote*))
                              ;;     (begin
                              ;;       (http-client#close-all-connections!))) ;; for http-client
                              (if (not (eq? *default-log-port* (current-error-port)))
                                  (close-output-port *default-log-port*))
			      (set! *default-log-port* (current-error-port))) "Cleanup db exit thread"))
	  (th2 (make-thread (lambda ()
			      (debug:print 4 *default-log-port* "Attempting clean exit. Please be patient and wait a few seconds...")
			      (if no-hurry
                                  (begin

      )
    )

  0)

(define (std-signal-handler signum)
  ;; (signal-mask! signum)
  (set! *time-to-exit* #t) 
  ;;(debug:print-info 13 *default-log-port* "got signal "signum)
  (debug:print-error 0 *default-log-port* "Received signal " signum " aaa exiting promptly")
  ;; (std-exit-procedure) ;; shouldn't need this since we are exiting and it will be called anyway
  (exit))

(define (special-signal-handler signum)
  ;; (signal-mask! signum)
  (set! *time-to-exit* #t)
  ;;(debug:print-info 13 *default-log-port* "got signal "signum)
  (debug:print-error 0 *default-log-port* "Received signal " signum " sending email befor exiting!!")
  ;;TODO send email to notify admin contact listed in the config that the lisner got killed
  ;; (std-exit-procedure) ;; shouldn't need this since we are exiting and it will be called anyway
  (exit))


(set-signal-handler! signal/int  std-signal-handler)  ;; ^C
(set-signal-handler! signal/term std-signal-handler)

;; (set-signal-handler! signal/stop std-signal-handler)  ;; ^Z NO, do NOT handle ^Z!

;;======================================================================
;; M I S C   U T I L S
;;======================================================================

;; convert stuff to a number if possible

(define (common:which cmds)
  (if (null? cmds)
      #f
      (let loop ((hed (car cmds))
		 (tal (cdr cmds)))
	(let ((res (with-input-from-pipe (conc "which " hed) read-line)))
	  (if (and (string? res)
		   (common:file-exists? res))
	      res
	      (if (null? tal)
		  #f
		  (loop (car tal)(cdr tal))))))))
  
(define (common:get-install-area)
  (let ((exe-path (car (argv))))
    (if (common:file-exists? exe-path)
	(handle-exceptions
	 exn
	 #f
	 (pathname-directory
	  (pathname-directory 
	   (pathname-directory exe-path))))
	#f)))

;; return first path that can be created or already exists and is writable
;;
(define (common:get-create-writeable-dir dirs)
  (if (null? dirs)
      #f
      (let loop ((hed (car dirs))
		 (tal (cdr dirs)))
	(let ((res (or (and (directory? hed)
			    (file-write-access? hed)
			    hed)
		       (handle-exceptions
			   exn
			   (begin
			     (debug:print-info 0 *default-log-port* "could not create " hed ", this might cause problems down the road.")
			     #f)
			(create-directory hed #t)))))
	  (if (and (string? res)
		   (directory? res))
	      res
	      (if (null? tal)
		  #f
		  (loop (car tal)(cdr tal))))))))

;; use bash to expand a glob. Does NOT handle paths with spaces!
;;
(define (common:bash-glob instr)
  (string-split
   (with-input-from-pipe
       (conc "/bin/bash -c \"echo " instr "\"")
       read-line)))

;;======================================================================
;; Some safety net stuff
;;======================================================================

;; return input if it is a list or return null
(define (common:list-or-null inlst #!key (ovrd #f)(message #f))
  (if (list? inlst)
      inlst
      (begin
	(if message (debug:print-error 0 *default-log-port* message))
	(or ovrd '()))))
  
;;======================================================================
;; T A R G E T S  ,   S T A T E ,   S T A T U S ,   
;;                    R U N N A M E    A N D   T E S T P A T T
;;======================================================================

;; (map print (map car (hash-table->alist (read-config "runconfigs.config" #f #t))))

(define (common:args-get-status)
  (or (args:get-arg "-status")(args:get-arg ":status")))

(define (common:args-get-testpatt rconf)
  (let* (;; (tagexpr       (args:get-arg "-tagexpr"))
         ;; (tags-testpatt (if tagexpr (string-join (runs:get-tests-matching-tags tagexpr) ",") #f))
         (testpatt-key  (or (args:get-arg "-modepatt") (args:get-arg "--modepatt") "TESTPATT"))
         (args-testpatt (or (args:get-arg "-testpatt") (args:get-arg "-runtests") "%"))
         (rtestpatt     (if rconf (runconfigs-get rconf testpatt-key) #f)))
    (cond
     ((or (args:get-arg "--modepatt") (args:get-arg "-modepatt")) ;; modepatt is a forced setting, when set it MUST refer to an existing PATT in the runconfig
      (if rconf
	  (let* ((patts-from-mode-patt	  (runconfigs-get rconf testpatt-key)))
	    (debug:print-info 0 *default-log-port* "modepatt defined is: "testpatt-key" runconfigs values for  " testpatt-key " "  patts-from-mode-patt)
	    patts-from-mode-patt)
	  (begin
	    (debug:print-info 0 *default-log-port* " modepatt defined is: "testpatt-key" runconfigs values for  " testpatt-key) ;;  " " patts-from-mode-patt)
	    #f)))     ;; We do NOT fall back to "%"
     ;; (tags-testpatt
     ;;  (debug:print-info 0 *default-log-port* "-tagexpr "tagexpr" selects testpatt "tags-testpatt)
     ;;  tags-testpatt)
     ((and (equal? args-testpatt "%") rtestpatt)
      (debug:print-info 0 *default-log-port* "testpatt defined in "testpatt-key" from runconfigs: " rtestpatt)
      rtestpatt)
     (else 
      (debug:print-info 0 *default-log-port* "using testpatt " args-testpatt " rtestpatt:" rtestpatt)
      args-testpatt))))



(define (common:false-on-exception thunk #!key (message #f))
  (handle-exceptions exn
                     (begin
                       (if message
                           (debug:print-info 0 *default-log-port* message))
                       #f) (thunk) ))

(define (common:file-exists? path-string #!key (silent #f))
  ;; this avoids stack dumps in the case where 

  ;;;; TODO: catch permission denied exceptions and emit appropriate warnings, eg:  system error while trying to access file: "/nfs/pdx/disks/icf_env_disk001/bjbarcla/gwa/issues/mtdev/randy-slow/reproduce/q...
  (common:false-on-exception (lambda () (file-exists? path-string))
                             message: (if (not silent)
                                          (conc "Unable to access path: " path-string)
                                          #f)
                             ))

(define (common:directory-exists? path-string)
  ;;;; TODO: catch permission denied exceptions and emit appropriate warnings, eg:  system error while trying to access file: "/nfs/pdx/disks/icf_env_disk001/bjbarcla/gwa/issues/mtdev/randy-slow/reproduce/q...
  (common:false-on-exception (lambda () (directory-exists? path-string))
                             message: (conc "Unable to access path: " path-string)
                             ))

;; does the directory exist and do we have write access?
;;
;;    returns the directory or #f
;;
(define (common:directory-writable? path-string)
  (handle-exceptions
   exn
   #f
   (if (and (directory-exists? path-string)
            (file-write-access? path-string))
       path-string
       #f)))

(define (common:get-linktree)
  (or (getenv "MT_LINKTREE")
      (if *configdat*
	  (configf:lookup *configdat* "setup" "linktree")
	  #f)
      (if (or *toppath* (getenv "MT_RUN_AREA_HOME"))
	  (conc (or *toppath* (getenv "MT_RUN_AREA_HOME")) "/lt")

	  #f)
      (let* ((tp (common:get-toppath #f))
	     (lt (conc tp "/lt")))
	(if (not tp)(debug:print 0 *default-log-port* "WARNING: cannot calculate best path for linktree, using " lt))
	lt)))

(define (common:args-get-runname)
  (let ((res (or (args:get-arg "-runname")
		 (args:get-arg ":runname")
		 (getenv "MT_RUNNAME"))))
    ;; (if res (set-environment-variable "MT_RUNNAME" res)) ;; not sure if this is a good idea. side effect and all ...
    res))

(define (common:get-fields cfgdat)
  (let ((fields (hash-table-ref/default cfgdat "fields" '())))
    (map car fields)))

(define (common:args-get-target #!key (split #f)(exit-if-bad #f))
  (let* ((keys    (if (hash-table? *configdat*) (common:get-fields *configdat*) '()))
	 (numkeys (length keys))
	 (target  (or (args:get-arg "-reqtarg")
		      (args:get-arg "-target")
		      (getenv "MT_TARGET")))
	 (tlist   (if target (string-split target "/" #t) '()))
	 (valid   (if target
		      (or (null? keys) ;; probably don't know our keys yet

	    target)
	(if target
	    (begin
	      (debug:print-error 0 *default-log-port* "Invalid target, spaces or blanks not allowed \"" target "\", target should be: " (string-intersperse keys "/") ", have " tlist " for elements")
	      (if exit-if-bad (exit 1))
	      #f)
	    #f))))

;; looking only (at least for now) at the MT_ variables craft the full testname
;;
(define (common:get-full-test-name)
  (if (getenv "MT_TEST_NAME")
      (if (and (getenv "MT_ITEMPATH")
               (not (equal? (getenv "MT_ITEMPATH") "")))
          (getenv "MT_TEST_NAME")
          (conc (getenv "MT_TEST_NAME") "/" (getenv "MT_ITEMPATH")))
      #f))

;; logic for getting homehost. Returns (host . at-home)
;; IF *toppath* is not set, wait up to five seconds trying every two seconds
;; (this is to accomodate the watchdog)
;;
(define (common:get-homehost #!key (trynum 5))
  ;; called often especially at start up. use mutex to eliminate collisions

	   ;; first look in config, then look in file .homehost, create it if not found
	   (homehost (or (configf:lookup *configdat* "server" "homehost" )
			 (handle-exceptions
			     exn
			     (if (> trynum 0)
				 (let ((delay-time (* (- 5 trynum) 5)))
				   (mutex-unlock! *homehost-mutex*)
				   (debug:print 0 *default-log-port* "ERROR: ["(common:human-time)"] Failed to read .homehost file, delaying " delay-time " seconds and trying again, message: "  ((condition-property-accessor 'exn 'message) exn))
				   (thread-sleep! delay-time)
				   (common:get-homehost trynum: (- trynum 1)))
				 (begin
				   (mutex-unlock! *homehost-mutex*)
				   (debug:print 0 *default-log-port* "ERROR: ["(common:human-time)"] Failed to read .homehost file after trying five times. Giving up and exiting, message: "  ((condition-property-accessor 'exn 'message) exn))
				   (exit 1)))
			   (let ((hhf (conc *toppath* "/.homehost")))
			     (if (common:file-exists? hhf)
				 (with-input-from-file hhf read-line)
				 (if (file-write-access? *toppath*)
				     (begin
				       (with-output-to-file hhf
					 (lambda ()
					   (print bestadrs)))
				       (begin

    (if hh
	(cdr hh)
	#f)))

;; do we honor the caches of the config files?
;;
(define (common:use-cache?)
  (let ((res #t)) ;; priority by order of evaluation
    (if *configdat* ;; sillyness here. can't use setup/use-cache to know if we can use the cached files!
	(if (equal? (configf:lookup *configdat* "setup" "use-cache") "no")
	    (set! res #f)
	    (if (equal? (configf:lookup *configdat* "setup" "use-cache") "yes")
		(set! res #t))))
    (if (args:get-arg "-no-cache")(set! res #f)) ;; overrides setting in "setup"
    (if (getenv "MT_USE_CACHE")
	(if (equal? (getenv "MT_USE_CACHE") "yes")
	    (set! res #t)
	    (if (equal? (getenv "MT_USE_CACHE") "no")
		(set! res #f))))    ;; overrides -no-cache switch
    res))
  
;; force use of server?
;;
(define (common:force-server?)
  (let* ((force-setting (configf:lookup *configdat* "server" "force"))
	 (force-type    (if force-setting (string->symbol force-setting) #f))
	 (force-result  (case force-type
			  ((#f)     #f)
			  ((always) #t)
			  ((test)   (if (args:get-arg "-execute") ;; we are in a test
					#t
					#f))
			  (else
			   (debug:print 0 *default-log-port* "ERROR: Bad server force setting " force-setting ", forcing server.")
			   #t)))) ;; default to requiring server
    (if force-result
	(begin
	  (debug:print-info 0 *default-log-port* "forcing use of server, force setting is \"" force-setting "\".")
	  #t)
	#f)))








;;======================================================================
;; M I S C   L I S T S
;;======================================================================

;; items in lista are matched value and position in listb
;; return the remaining items in listb or #f
;;

	      (loop (car tal)
		    (cdr tal)
		    new-rownames
		    new-colnames
		    (if (> curr-rownum rownum) curr-rownum rownum)
		    (if (> curr-colnum colnum) curr-colnum colnum)
		    ))))))

;; if it looks like a number -> convert it to a number, else return it
;;
(define (common:lazy-convert inval)
  (let* ((as-num (if (string? inval)(string->number inval) #f)))
    (or as-num inval)))

;; convert string a=1; b=2; c=a silly thing; d=
;; to '((a . 1)(b . 2)(c . "a silly thing")(d . ""))
;;
(define (common:val->alist val #!key (convert #f))
  (let ((val-list (string-split-fields ";\\s*" val #:infix)))
    (if val-list
	(map (lambda (x)
	       (let ((f (string-split-fields "\\s*=\\s*" x #:infix)))
		 (case (length f)
		   ((0) `(,#f))  ;; null string case
		   ((1) `(,(string->symbol (car f))))
		   ((2) `(,(string->symbol (car f)) . ,(let ((inval (cadr f)))
							 (if convert (common:lazy-convert inval) inval))))
		   (else f))))
	     val-list)
	'())))

;;======================================================================
;; S Y S T E M   S T U F F
;;======================================================================

;; lazy-safe get file mod time. on any error (file not existing etc.) return 0
;;

;; 		  (if match
;; 		      (let ((newval (string->number (cadr match))))
;; 			(if (number? newval)
;; 			    (set! cpu-load newval))))))
;; 	      (car load-res))
;;     cpu-load))

;; get values from cached info from dropping file in logs dir
;;  e.g. key is host and dtype is normalized-load
;;
(define (common:get-cached-info key dtype #!key (age 10))
  (if *toppath*
      (let* ((fullpath (conc *toppath* "/.sysdata/" key "-" dtype ".log")))
	(if (and (file-exists? fullpath)
		 (file-read-access? fullpath))
	    (handle-exceptions
	     exn
	     #f
	     (debug:print 2 *default-log-port* "reading file " fullpath)
	     (let ((real-age (- (current-seconds)(file-change-time fullpath)))) 
	       (if (< real-age age)
		   (handle-exceptions
		       exn
		     (begin
		       (debug:print-info 1 *default-log-port* " removing bad file " fullpath)
		       (delete-file* fullpath)
		       #f)
		     (with-input-from-file fullpath read))
		   (begin
		     (debug:print-info 2 *default-log-port* "file " fullpath " is too old (" real-age" seconds) to trust, skipping reading it")
		     #f))))
	    (begin
	      (debug:print 2 *default-log-port* "not reading file " fullpath)
	      #f)))
      #f))
 
(define (common:write-cached-info key dtype dat)
  (if *toppath*
      (let* ((fulldir  (conc *toppath* "/.sysdata"))
	     (fullpath (conc fulldir "/" key "-" dtype ".log")))
	(if (not (file-exists? fulldir))(create-directory fulldir #t))
	(handle-exceptions
	 exn
	 #f
	 (with-output-to-file fullpath (lambda ()(pp dat)))))
      #f))
  
;; get cpu load by reading from /proc/loadavg, return all three values
;;
(define (common:get-cpu-load remote-host)
  (handle-exceptions
   exn
   '(99 99 99)
   (let* ((actual-hostname (or remote-host (get-host-name) "localhost")))
     (or (common:get-cached-info actual-hostname "cpu-load")
	 (let ((result (if remote-host
			   (map (lambda (res)
				  (if (eof-object? res) 9e99 res))
			        (with-input-from-pipe 
				 (conc "ssh " remote-host " cat /proc/loadavg")
				 (lambda ()(list (read)(read)(read)))))
			   (with-input-from-file "/proc/loadavg" 
			     (lambda ()(list (read)(read)(read)))))))
	   (common:write-cached-info actual-hostname "cpu-load" result)
	                     result)))))

;; get normalized cpu load by reading from /proc/loadavg and /proc/cpuinfo return all three values and the number of real cpus and the number of threads
;; returns alist '((adj-cpu-load . normalized-proc-load) ... etc.
;;  keys: adj-proc-load, adj-core-load, 1m-load, 5m-load, 15m-load
;;
(define (common:get-normalized-cpu-load remote-host)
  (let ((res (common:get-normalized-cpu-load-raw remote-host))
	(default `((adj-proc-load . 2) ;; there is no right answer
		   (adj-core-load . 2)
		   (1m-load       . 2)
		   (5m-load       . 0) ;; causes a large delta - thus causing default of throttling if stuff goes wrong
		   (15m-load      . 0)
		   (proc          . 1)
		   (core          . 1)
		   (phys          . 1)
		   (error         . #t))))
    (cond
     ((and (list? res)
	   (> (length res) 2))
      res)
     ((eq? res #f)   default) ;; add messages?
     ((eq? res #f) default)   ;; this would be the #eof
     (else default))))

(define (common:get-normalized-cpu-load-raw remote-host)
  (let* ((actual-host (or remote-host (get-host-name)))) ;; #f is localhost
    (or (common:get-cached-info actual-host "normalized-load")
	(let ((data (if remote-host
			(with-input-from-pipe 
			    (conc "ssh " remote-host " \"cat /proc/loadavg;cat /proc/cpuinfo;echo end\"")
			  read-lines)
			(append 
			 (with-input-from-file "/proc/loadavg" 
			   read-lines)
			 (with-input-from-file "/proc/cpuinfo"
			   read-lines)
			 (list "end"))))
	      (load-rx  (regexp "^([\\d\\.]+)\\s+([\\d\\.]+)\\s+([\\d\\.]+)\\s+.*$"))
	      (proc-rx  (regexp "^processor\\s+:\\s+(\\d+)\\s*$"))
	      (core-rx  (regexp "^core id\\s+:\\s+(\\d+)\\s*$"))
	      (phys-rx  (regexp "^physical id\\s+:\\s+(\\d+)\\s*$"))
	      (max-num  (lambda (p n)(max (string->number p) n))))
	  ;; (print "data=" data)
	  (if (null? data) ;; something went wrong
	      #f
	      (let loop ((hed      (car data))
			 (tal      (cdr data))
			 (loads    #f)
			 (proc-num 0)  ;; processor includes threads
			 (phys-num 0)  ;; physical chip on motherboard
			 (core-num 0)) ;; core
		;;; (print hed ", " loads ", " proc-num ", " phys-num ", " core-num)
		(if (null? tal) ;; have all our data, calculate normalized load and return result
		    (let* ((act-proc (+ proc-num 1))
			   (act-phys (+ phys-num 1))
			   (act-core (+ core-num 1))
			   (adj-proc-load (/ (car loads) act-proc))
			   (adj-core-load (/ (car loads) act-core))
			   (result
			    (append (list (cons 'adj-proc-load adj-proc-load)
					  (cons 'adj-core-load adj-core-load))
				    (list (cons '1m-load (car loads))
					  (cons '5m-load (cadr loads))
					  (cons '15m-load (caddr loads)))
				    (list (cons 'proc act-proc)
					  (cons 'core act-core)
					  (cons 'phys act-phys)))))
		      (common:write-cached-info actual-host "normalized-load" result)
		      result)
		  (regex-case
		   hed
		   (load-rx  ( x l1 l5 l15 ) (loop (car tal)(cdr tal)(map string->number (list l1 l5 l15)) proc-num phys-num core-num))
		   (proc-rx  ( x p         ) (loop (car tal)(cdr tal) loads           (max-num p proc-num) phys-num core-num))
		   (phys-rx  ( x p         ) (loop (car tal)(cdr tal) loads           proc-num (max-num p phys-num) core-num))
		   (core-rx  ( x c         ) (loop (car tal)(cdr tal) loads           proc-num phys-num (max-num c core-num)))
		   (else 
		    (begin
		      ;; (print "NO MATCH: " hed)
		      (loop (car tal)(cdr tal) loads proc-num phys-num core-num)))))))))))

(define (common:unix-ping hostname)
  (let ((res (system (conc "ping -c 1 " hostname " > /dev/null"))))
    (eq? res 0)))

;; ideally put all this info into the db, no need to preserve it across moving homehost
;;
;; return list of
;;  ( reachable? cpuload update-time )
(define (common:get-host-info hostname)
  (let* ((loadinfo                         (rmt:get-latest-host-load hostname)) ;; if this host happens to have been recently used by a test reuse the load data
         (load                             (car loadinfo))
         (load-sample-time                 (cdr loadinfo))
         (load-sample-age                  (- (current-seconds) load-sample-time))
         (loadinfo-timeout-seconds         6) ;; this was 20 seconds, seems way too lax. Switch to 6 seconds
         (host-last-update-timeout-seconds 4)
         (host-rec (hash-table-ref/default *host-loads* hostname #f))
         )
    (cond
     ((< load-sample-age loadinfo-timeout-seconds)
      (list #t
            load-sample-time
            load))
     ((and host-rec
           (< (current-seconds) (+ (host-last-update host-rec) host-last-update-timeout-seconds)))
      (list #t
            (host-last-update host-rec)
            (host-last-cpuload host-rec )))
     ((common:unix-ping hostname)
      (list #t
            (current-seconds)
            (alist-ref 'adj-core-load (common:get-normalized-cpu-load hostname)))) ;; this is cheaper than you might think. get-normalized-cpu-load is cached for up to 5 seconds
     (else
      (list #f 0 -1) ;; bad host, don't use!
      ))))

;; see defstruct host at top of file.
;;    host: reachable last-update last-used last-cpuload
;;
(define (common:update-host-loads-table hosts-raw)
  (let* ((hosts (filter (lambda (x)
                          (string-match (regexp "^\\S+$") x))
                        hosts-raw)))
    (for-each
     (lambda (hostname)
       (let* ((rec       (let ((h (hash-table-ref/default *host-loads* hostname #f)))

              (last-reached-time (cadr host-info))
              (load              (caddr host-info)))
         (host-reachable-set!    rec is-reachable)
         (host-last-update-set!  rec last-reached-time)
         (host-last-cpuload-set! rec load)))
     hosts)))

;; go through the hosts from least recently used to most recently used, pick the first that meets the load criteral from the
;; [host-rules] section.
;;
(define (common:get-least-loaded-host hosts-raw host-type configdat)
  (let* ((rdat       (configf:lookup configdat "host-rules" host-type))
	 (rules      (common:val->alist (or rdat "") convert: #t))   ;; maxnload, maxnjobs, maxjobrate
	 (maxnload   (common:alist-ref/default 'maxnload rules 1.5)) ;; max normalized load
	 (maxnjobs   (common:alist-ref/default 'maxnjobs rules 1.5)) ;; max normalized number of jobs
	 (maxjobrate (common:alist-ref/default 'maxjobrate rules (/ 1 6))) ;; max rate of submitting jobs to a given host in jobs/second
	 (hosts      (filter (lambda (x)
			       (string-match (regexp "^\\S+$") x))
			     hosts-raw))
         ;; (best-host #f)
	 (get-rec    (lambda (hostname)
		       ;; (print "get-rec hostname=" hostname)
		       (let ((h (hash-table-ref/default *host-loads* hostname #f)))
			 (if h
			     h
			     (let ((h (make-host)))
			       (hash-table-set! *host-loads* hostname h)
			       h)))))
         (best-load 99999)
         (curr-time (current-seconds))
	 (get-hosts-sorted (lambda (hosts)
			     (sort hosts (lambda (a b)
					   (let ((a-rec (get-rec a))
						 (b-rec (get-rec b)))
					     ;; (print "a=" a " a-rec=" a-rec " host-last-used=" (host-last-used a-rec))
					     ;; (print "b=" b " b-rec=" b-rec " host-last-used=" (host-last-used b-rec))
					     (< (host-last-used a-rec)
						(host-last-used b-rec))))))))
    (debug:print 0 *default-log-port* "INFO: hosts-sorted=" (get-hosts-sorted hosts))
    (if (null? hosts)
	#f ;; no hosts to select from. All done and giving up now.
	(let ((hosts-sorted (get-hosts-sorted hosts)))
	  (common:update-host-loads-table hosts)
	  (let loop ((hostname  (car hosts-sorted))
		     (tal       (cdr hosts-sorted))
		     (best-host #f))
	    (let* ((rec       (get-rec hostname))






		   (reachable (host-reachable     rec))
		   (load      (host-last-cpuload  rec))

		   (last-used (host-last-used     rec))


		   (delta     (- curr-time last-used))

		   (job-rate  (if (> delta 0)
				  (/ 1 delta)

				  999)) ;; jobs per second
		   (new-best  
		    (cond
		     ((not reachable)
		      (debug:print 0 *default-log-port* "Skipping host " hostname " as it cannot be reached.")
		      best-host)
		     ((and (< load maxnload)        ;; load is acceptable

			   (< job-rate maxjobrate)) ;; job rate is acceptable
		      (set! best-load load)
		      hostname)
		     (else best-host))))
	      (debug:print 0 *default-log-port* "INFO: Trying host " hostname " with load " load ", last used " delta " seconds ago, with job-rate " job-rate " for running a test." )
	      (if new-best
		  (begin ;; found a host, return it
		    (debug:print 0 *default-log-port* "INFO: Found host: " new-best " load: " load " last-used: " delta " seconds ago, with job-rate: " job-rate)
		    (host-last-used-set! rec curr-time)
		    new-best)
		  (if (null? tal) #f (loop (car tal)(cdr tal) best-host)))))))))

(define (common:wait-for-homehost-load maxload msg)

  (let* ((hh-dat (if (common:on-homehost?) ;; if we are on the homehost then pass in #f so the calls are local.
                     #f
                     (common:get-homehost)))


         (hh     (if hh-dat (car hh-dat) #f)))
    (common:wait-for-normalized-load maxload msg hh)))

(define (common:get-num-cpus remote-host)
  (let* ((actual-host (or remote-host (get-host-name))))
    (or (common:get-cached-info actual-host "num-cpus" age: (+ 2592000 (random 3600))) ;; hosts had better not be changing the number of cpus too often!
	(let* ((proc   (lambda ()
			 (let loop ((numcpu 0)
				    (inl    (read-line)))
			   (if (eof-object? inl)
			       (if (> numcpu 0)
				   numcpu
				   #f) ;; if zero return #f so caller knows that things are not working
			       (loop (if (string-match "^processor\\s+:\\s+\\d+$" inl)
					 (+ numcpu 1)
					 numcpu)
				     (read-line))))))
	       (result (if remote-host
			   (with-input-from-pipe 
			       (conc "ssh " remote-host " cat /proc/cpuinfo")
			     proc)
			   (with-input-from-file "/proc/cpuinfo" proc))))
	  (if (and (number? result)
		   (> result 0))
	      (common:write-cached-info actual-host "num-cpus" result))
	  result))))

;; wait for normalized cpu load to drop below maxload
;;
(define (common:wait-for-normalized-load maxload msg remote-host #!optional (rem-tries 5))
  (let ((num-cpus (common:get-num-cpus remote-host)))
    (if num-cpus
	(common:wait-for-cpuload maxload num-cpus 15 msg: msg remote-host: remote-host)
	(begin
	  (thread-sleep! (random 60)) ;; we failed to get num cpus. wait a bit and try again
	  (if (> rem-tries 0)
	      (common:wait-for-normalized-load maxload msg remote-host (- rem-tries 1))
	      #f)))))

;; DO NOT CALL THIS DIRECTLY. It is called from common:wait-for-normalized-load
;;
(define (common:wait-for-cpuload maxload-in numcpus-in waitdelay #!key (count 1000) (msg #f)(remote-host #f)(force-maxload #f))
  (let* ((loadavg (common:get-cpu-load remote-host))
	 (numcpus (if (<= 1 numcpus-in) ;; not possible to have zero.  If we get 1, it's possible that we got the previous default, and we should check again
		      (common:get-num-cpus remote-host)
		      numcpus-in))
	 (maxload (if force-maxload
		      maxload-in
		      (if (number? maxload-in)
			  (max maxload-in 0.5)
			  0.5))) ;; so maxload must be greater than 0.5 for now BUG - FIXME?
	 (first   (car loadavg))
	 (next    (cadr loadavg))
	 (adjload (* maxload (max 1 numcpus))) ;; possible bug where numcpus (or could be maxload) is zero, crude fallback is to at least use 1
	 (loadjmp (- first next))
         (adjwait (min (+ 300 (random 10)) (abs (* (+ (random 10)(/ (- 1000 count) 10) waitdelay) (- first adjload) ))  )));; add some randomness to the time to break any alignment where netbatch dumps many jobs to machines simultaneously
    ;; let's let the user know once in a long while that load checking is happening but not constantly report it
    (if (> (random 100) 75) ;; about 25% of the time
	(debug:print-info 1 *default-log-port* "Checking cpuload on " (or remote-host "localhost") ", maxload: " maxload
			  ", load: " first ", adjload: " adjload ", loadjmp: " loadjmp))
    (cond
     ((and (> first adjload)
	   (> count 0))
      (debug:print-info 0 *default-log-port* "server start delayed " adjwait " seconds due to load " first " exceeding max of " adjload " on server " (or remote-host (get-host-name)) " (normalized load-limit: " maxload ") " (if msg msg ""))
      (thread-sleep! adjwait)
      (common:wait-for-cpuload maxload numcpus waitdelay count: (- count 1) msg: msg remote-host: remote-host))
     ((and (> loadjmp numcpus)
	   (> count 0))
      (debug:print-info 0 *default-log-port* "waiting " adjwait " seconds due to load jump " loadjmp " > numcpus " numcpus (if msg msg ""))
      (thread-sleep! adjwait)
      (common:wait-for-cpuload maxload numcpus waitdelay count: (- count 1) msg: msg remote-host: remote-host)))))

(define (get-uname . params)
  (let* ((uname-res (process:cmd-run->list (conc "uname " (if (null? params) "-a" (car params)))))
	 (uname #f))
    (if (null? (car uname-res))
	"unknown"
	(caar uname-res))))

       (conc (configf:lookup *configdat* "setup" "free-space-script") " " path)
       (lambda ()
	 (let ((res (read-line)))
	   (if (string? res)
	       (string->number res)))))
      (get-unix-df path)))

(define (get-free-inodes path)
  (if (configf:lookup *configdat* "setup" "free-inodes-script")
      (with-input-from-pipe 
	  (conc (configf:lookup *configdat* "setup" "free-inodes-script") " " path)
	(lambda ()
	  (let ((res (read-line)))
	    (if (string? res)
		(string->number res)))))
      (get-unix-inodes path)))

(define (get-unix-df path)
  (let* ((df-results (process:cmd-run->list (conc "df " path)))
	 (space-rx   (regexp "([0-9]+)\\s+([0-9]+)%"))
	 (freespc    #f))
    ;; (write df-results)
    (for-each (lambda (l)
		(let ((match (string-search space-rx l)))
		  (if match 
		      (let ((newval (string->number (cadr match))))
			(if (number? newval)
			    (set! freespc newval))))))
	      (car df-results))
    freespc))

(define (get-unix-inodes path)
  (let* ((df-results (process:cmd-run->list (conc "df -i " path)))
	 (space-rx   (regexp "([0-9]+)\\s+([0-9]+)%"))
	 (freenodes    0)) ;; 0 is a better failsafe than #f here.
    ;; (write df-results)
    (for-each (lambda (l)
		(let ((match (string-search space-rx l)))
		  (if match 
		      (let ((newval (string->number (cadr match))))
			(if (number? newval)
			    (set! freenodes newval))))))
	      (car df-results))
    freenodes))

(define (common:check-space-in-dir dirpath required)
  (let* ((dbspace  (if (directory? dirpath)
		       (get-df dirpath)
		       0)))
    (list (> dbspace required)
	  dbspace
	  required
	  dirpath)))

;; check space in dbdir and in megatest dir
;; returns: ok/not dbspace required-space
;;
(define (common:check-db-dir-space)
  (let* ((required (string->number 
                    ;; default is 1GB (or actually a billion bytes) This is the number of 1 kB blocks.
		    (or (configf:lookup *configdat* "setup" "dbdir-space-required")
			"1000000")))
	 (dbdir    (common:get-db-tmp-area)) ;; (db:get-dbdir))
	 (tdbspace (common:check-space-in-dir dbdir required))
	 (mdbspace (common:check-space-in-dir *toppath* required)))
    (sort (list tdbspace mdbspace) (lambda (a b)
				     (< (cadr a)(cadr b))))))
    
;; check available space in dbdir, exit if insufficient

	(begin
	  (debug:print-error 0 *default-log-port* "Insufficient space in " dbdir ", require " required ", have " dbspace  ", exiting now.")
	  (exit 1)))))
  
;; paths is list of lists ((name path) ... )
;;
(define (common:get-disk-with-most-free-space disks minsize)
  (let* ((best     #f)
	(bestsize 0)
        (default-min-inodes-string "1000000")
        (default-min-inodes (string->number default-min-inodes-string))
        (min-inodes (or (string->number (if (configf:lookup *configdat* "setup" "min_inodes") (configf:lookup *configdat* "setup" "min_inodes") default-min-inodes-string)) default-min-inodes)))

    (for-each 
     (lambda (disk-num)
       (let* ((dirpath    (cadr (assoc disk-num disks)))
	      (freespc    (cond
			   ((not (directory? dirpath))
			    (if (common:low-noise-print 300 "disks not a dir " disk-num)
				(debug:print 0 *default-log-port* "WARNING: disk " disk-num " at path \"" dirpath "\" is not a directory - ignoring it."))
			    -1)
			   ((not (file-write-access? dirpath))
			    (if (common:low-noise-print 300 "disks not writeable " disk-num)
				(debug:print 0 *default-log-port* "WARNING: disk " disk-num " at path \"" dirpath "\" is not writeable - ignoring it."))
			    -1)
			   ((not (eq? (string-ref dirpath 0) #\/))
			    (if (common:low-noise-print 300 "disks not a proper path " disk-num)
				(debug:print 0 *default-log-port* "WARNING: disk " disk-num " at path \"" dirpath "\" is not a fully qualified path - ignoring it."))
			    -1)
			   (else
			    (get-df dirpath))))
	      (free-inodes (cond
			   ((not (directory? dirpath))
			    (if (common:low-noise-print 300 "disks not a dir " disk-num)
				(debug:print 0 *default-log-port* "WARNING: disk " disk-num " at path \"" dirpath "\" is not a directory - ignoring it."))
			    -1)
			   ((not (file-write-access? dirpath))
			    (if (common:low-noise-print 300 "disks not writeable " disk-num)
				(debug:print 0 *default-log-port* "WARNING: disk " disk-num " at path \"" dirpath "\" is not writeable - ignoring it."))
			    -1)
			   ((not (eq? (string-ref dirpath 0) #\/))
			    (if (common:low-noise-print 300 "disks not a proper path " disk-num)
				(debug:print 0 *default-log-port* "WARNING: disk " disk-num " at path \"" dirpath "\" is not a fully qualified path - ignoring it."))
			    -1)
			   (else
			    (get-free-inodes dirpath))))
             ;;(free-inodes (get-free-inodes dirpath))
             )
             (debug:print 2 *default-log-port* "INFO: disk " disk-num " path " dirpath " free space " freespc " free inodes " free-inodes)
	 (if (and (> freespc bestsize)(> free-inodes min-inodes ))
	     (begin
	       (set! best     (cons disk-num dirpath))
	       (set! bestsize freespc)))
        ;;(print "Processing: " disk-num " bestsize: " bestsize " best: " best " freespc: " freespc " min-inodes: " min-inodes " free-inodes: " free-inodes)
      ))
     (map car disks))
    (if (and best (> bestsize minsize))
	best
	#f))) ;; #f means no disk candidate found

;; convert a spec string to a list of vectors #( rx  action rx-string )
(define (common:spec-string->list-of-specs spec-string actions)
  (let ((spec-strings  (string-split-fields "\\s*;\\s*" spec-string #:infix))
	(actions-regex (regexp (conc "^(.*)\\s+(" (string-intersperse (map conc actions) "|") ")"))))
    (filter
     (lambda (x) x)
     (map (lambda (s)
	    (let ((m (string-match actions-regex s)))
	      (if m
		  (vector (regexp (cadr m))(string->symbol (caddr m))(cadr m))
		  (begin
		    (debug:print 0 *default-log-port* "WARNING: Unrecognised rule \"" s "\" in clean-up specification.")
		    #f))))
	  spec-strings))))
  
;; given a list of specs rx . rule and a file return the first matching rule
;; 
(define (common:file-find-rule fname rules) ;; rule is vector #( rx action rx-string)
  (let loop ((rule   (car rules))
	     (tail   (cdr rules)))
    (let ((rx (vector-ref rule 0))
	  (rn (vector-ref rule 1))) ;; rule name
      (if (string-match rx fname)
	  rule ;; return the whole rule so regex can be printed etc.
	  (if (null? tail)
	      #f
	      (loop (car tail)(cdr tail)))))))

;; given a spec apply some rules to a directory
;;
;;  WARNING: This function will REMOVE files - be sure your spec and path is correct!
;;
;; spec format:
;;   file-regex1 action; file-regex2 action; ...
;; e.g.
;;   .*\.log$ keep; .* remove
;;   --> keep all .log files, remove everything else
;; limitations:
;;   cannot have a rule with ; as part of the spec
;;   not very flexible, would be nice to return binned file names?
;; supported rules:
;;   keep     - keep this file
;;   remove   - remove this file
;;   compress - compress this file
;;
(define (common:dir-clean-up path spec-string #!key (compress "gzip")(actions '(keep remove compress))(remove-empty #f))
  (let* ((specs         (common:spec-string->list-of-specs spec-string actions))
	 (keepers       (make-hash-table))
	 (directories   (make-hash-table)))
    (find-files 
     path
     action: (lambda (p res)
	       (let ((rule  (common:file-find-rule p specs)))
		 (cond
		  ((directory? p)(hash-table-set! directories p #t))
		  (else
		   (case (vector-ref rule 1)
		     ((keep)(hash-table-set! keepers p rule))
		     ((remove)
		      (print "Removing file " p)
		      (delete-file p))
		     ((compress)
		      (print "Compressing file " p)
		      (system (conc compress " " p)))
		     (else
		      (print "No match for file " p))))))))
    (if remove-empty
	(for-each
	 (lambda (d)
	   (if (null? (glob (conc d "/.*")(conc d "/*")))
	       (begin
		 (print "Removing empty directory " d)
		 (delete-directory d))))
	 (sort (hash-table-keys directories) (lambda (a b)(> (string-length a)(string-length b))))))
    ))

;;======================================================================
;; E N V I R O N M E N T   V A R S
;;======================================================================
(define (bb-check-path #!key (msg "check-path: "))
  (let ((path (or (get-environment-variable "PATH") "none")))
    (debug:print-info 0 *default-log-port* (conc msg" : $PATH="path))
    (if (string-match "^.*/isoenv-core/.*" path)
        (debug:print-error 0 *default-log-port* (conc msg" : !!ISOENV PRESENT!!")) ;; remove for prod
        (debug:print-info 1 *default-log-port* (conc msg" : **no isoenv present**")))))

	      
(define (save-environment-as-files fname #!key (ignorevars (list "USER" "HOME" "DISPLAY" "LS_COLORS" "XKEYSYMDB" "EDITOR" "MAKEFLAGS" "MAKEF" "MAKEOVERRIDES")))
  ;;(bb-check-path msg: "save-environment-as-files entry")
  (let ((envvars (get-environment-variables))
        (whitesp (regexp "[^a-zA-Z0-9_\\-:,\\.\\/%$]"))
	(mungeval (lambda (val)
		    (cond
		     ((eq? val #t) "") ;; convert #t to empty string
		     ((eq? val #f) #f) ;; convert #f to itself (still thinking about this one
		     (else val)))))
    (with-output-to-file (conc fname ".csh")
       (lambda ()
          (for-each (lambda (keyval)
		      (let* ((key   (car keyval))
			     (val   (cdr keyval))
			     (delim (if (string-search whitesp val) 
					"\""
					"")))
			(print (if (or (member key ignorevars)
				       (string-search whitesp key))
				   "# setenv "
				   "setenv ")
			       key " " delim (mungeval val) delim)))
		    envvars)))
     (with-output-to-file (conc fname ".sh")
       (lambda ()
          (for-each (lambda (keyval)
		      (let* ((key (car keyval))
			     (val (cdr keyval))
			     (delim (if (string-search whitesp val) 
					"\""
					"")))
			(print (if (or (member key ignorevars)
				       (string-search whitesp key)
				       (string-search ":" key)) ;; internal only values to be skipped.
				   "# export "
				   "export ")
			       key "=" delim (mungeval val) delim)))
                    envvars)))))


(define (common:get-param-mapping #!key (flavor #f))
  "returns alist mapping string keys in testconfig/subrun to megatest command line switches; if flavor is switch-symbol, maps tcmt symbolic switches to megatest switches"
  (let ((default '(("tag-expr"  . "-tagexpr")
                   ("mode-patt" . "-modepatt")
                   ("run-name"  . "-runname")
                   ("contour"   . "-contour")
                   ("target"    . "-target")
                   ("test-patt" . "-testpatt")
                   ("msg"       . "-m")
                   ("log"       . "-log")
                   ("start-dir" . "-start-dir")
                   ("new"       . "-set-state-status"))))
    (if (eq? flavor 'switch-symbol)
        (map (lambda (x)
               (cons (string->symbol (conc "-" (car x))) (cdr x)))
             default)
        default)))

;; set some env vars from an alist, return an alist with original values
;; (("VAR" "value") ...)
;; a value of #f means "unset this var"
;;
(define (alist->env-vars lst)
  (if (list? lst)
      (let ((res '()))
	(for-each (lambda (p)
		    (let* ((var (car  p))
			   (val (cadr p))
			   (prv (get-environment-variable var)))
		      (set! res (cons (list var prv) res))
		      (if val 
			  (safe-setenv var (->string val))
			  (unsetenv var))))
		  lst)
	res)
      '()))


;; clear vars matching pattern, run proc, set vars back
;; if proc is a string run that string as a command with
;; system.
;;
(define *common:orig-env*
  (let ((envvars (get-environment-variables)))
    (if (get-environment-variable "MT_ORIG_ENV")
        (with-input-from-string
            (z3:decode-buffer (base64:base64-decode (get-environment-variable "MT_ORIG_ENV")))
          read)
        (filter-map (lambda (x)
                      (if (string-match "^MT_.*" (car x))
                          #f
                          x))
                    envvars))))

(define (common:with-orig-env proc)
  (let  ((current-env (get-environment-variables)))
    (for-each (lambda (x) (unsetenv (car x)))             current-env)
    (for-each (lambda (x) (setenv (car x) (cdr x))) *common:orig-env*)
    (let ((rv (cond
               ((string? proc)(system proc))
               (proc          (proc)))))
      (for-each (lambda (x) (unsetenv (car x))) *common:orig-env*)
      (for-each (lambda (x) (setenv (car x) (cdr x))) current-env)
      rv)))

(define (common:without-vars proc . var-patts)
  (let ((vars (make-hash-table)))
    (for-each
     (lambda (vardat) ;; each env var
       (for-each
	(lambda (var-patt)
	  (if (string-match var-patt (car vardat))

     (proc          (proc)))
    (hash-table-for-each
     vars
     (lambda (var val)
       (setenv var val)))
    vars))


(define (common:run-a-command cmd #!key (with-vars #f) (with-orig-env #f))
  (let* ((pre-cmd  (dtests:get-pre-command))
         (post-cmd (dtests:get-post-command))
         (fullcmd  (if (or pre-cmd post-cmd)
                       (conc pre-cmd cmd post-cmd)
                       (conc "viewscreen " cmd))))
    (debug:print-info 02 *default-log-port* "Running command: " fullcmd)
    (cond
     (with-vars     (common:without-vars  fullcmd))
     (with-orig-env (common:with-orig-env fullcmd))
     (else          (common:without-vars  fullcmd "MT_.*")))))
		  
;;======================================================================
;; T I M E   A N D   D A T E
;;======================================================================

;; Convert strings like "5s 2h 3m" => 60x60x2 + 3x60 + 5
(define (common:hms-string->seconds tstr)
  (let ((parts     (string-split-fields "\\w+" tstr))
	(time-secs 0)
	;; s=seconds, m=minutes, h=hours, d=days, M=months, y=years, w=weeks
	(trx       (regexp "(\\d+)([smhdMyw])")))
    (for-each (lambda (part)
		(let ((match  (string-match trx part)))
		  (if match
		      (let ((val (string->number (cadr match)))
			    (unt (caddr match)))
			(if val 
			    (set! time-secs (+ time-secs (* val
							    (case (string->symbol unt)
							      ((s) 1)
							      ((m) 60) ;; minutes
							      ((h) 3600)
							      ((d) 86400)
							      ((w) 604800)
							      ((M) 2628000) ;; aproximately one month
							      ((y) 31536000)
							      (else #f))))))))))
	      parts)
    time-secs))
		       
(define (seconds->hr-min-sec secs)
  (let* ((hrs (quotient secs 3600))
	 (min (quotient (- secs (* hrs 3600)) 60))
	 (sec (- secs (* hrs 3600)(* min 60))))

  (string-intersperse 
   (map (lambda (x)
          (number->string x 16))
        (map string->number
             (string-split instr)))
   "/"))

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

;; faux-lock is deprecated. Please use simple-lock below
;;
(define (common:faux-lock keyname #!key (wait-time 8)(allow-lock-steal #t))
  (if (rmt:no-sync-get/default keyname #f) ;; do not be tempted to compare to pid. locking is a one-shot action, if already locked for this pid it doesn't actually count
      (if (> wait-time 0)
	  (begin
	    (thread-sleep! 1)
	    (if (eq? wait-time 1) ;; only one second left, steal the lock
		(begin
		  (debug:print-info 0 *default-log-port* "stealing lock for " keyname)
		  (common:faux-unlock keyname force: #t)))
	    (common:faux-lock keyname wait-time: (- wait-time 1)))
	  #f)
      (begin
        (rmt:no-sync-set keyname (conc (current-process-id)))
        (equal? (conc (current-process-id)) (conc (rmt:no-sync-get/default keyname #f))))))

(define (common:faux-unlock keyname #!key (force #f))
  (if (or force (equal? (conc (current-process-id)) (conc (rmt:no-sync-get/default keyname #f))))
      (begin
        (if (rmt:no-sync-get/default keyname #f) (rmt:no-sync-del! keyname))
        #t)
      #f))

;; simple lock. improve and converge on this one.
;;
(define (common:simple-lock keyname)
  (rmt:no-sync-get-lock keyname))

(define (common:simple-unlock keyname #!key (force #f))
  (rmt:no-sync-del! keyname))


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

(define (common:in-running-test?)
  (and (args:get-arg "-execute") (get-environment-variable "MT_CMDINFO")))

(define (common:get-color-from-status status)
  (cond
   ((equal? status "PASS")    "green")
   ((equal? status "FAIL")    "red")
   ((equal? status "WARN")    "orange")
   ((equal? status "KILLED")  "orange")
   ((equal? status "KILLREQ") "purple")
   ((equal? status "RUNNING") "blue")
   ((equal? status "ABORT")   "brown")
   (else "black")))

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












;; 
;; 
;; (define (common:send-dboard-main-changed)
;;   (let* ((dashboard-ips (mddb:get-dashboards)))
;;     (for-each
;;      (lambda (ipadr)
;;        (let* ((soc (common:open-nm-req (conc "tcp://" ipadr)))
;; 	      (msg (conc "main " *toppath*))
;; 	      (res (common:nm-send-receive-timeout soc msg)))
;; 	 (if (not res) ;; couldn't reach that dashboard - remove it from db
;; 	     (print "ERROR: couldn't reach dashboard " ipadr))
;; 	 res))
;;      dashboard-ips)))
;;     
;;     
;; ;;======================================================================
;; ;; D A S H B O A R D   D B 
;; ;;======================================================================
;; 
;; (define (mddb:open-db)
;;   (let* ((db (open-database (conc (get-environment-variable "HOME") "/.dashboard.db"))))
;;     (set-busy-handler! db (busy-timeout 10000))
;;     (for-each
;;      (lambda (qry)
;;        (exec (sql db qry)))
;;      (list 
;;       "CREATE TABLE IF NOT EXISTS vars       (id INTEGER PRIMARY KEY,key TEXT, val TEXT, CONSTRAINT varsconstraint UNIQUE (key));"
;;       "CREATE TABLE IF NOT EXISTS dashboards (
;;           id         INTEGER PRIMARY KEY,
;;           pid        INTEGER,
;;           username   TEXT,
;;           hostname   TEXT,
;;           ipaddr     TEXT,
;;           portnum    INTEGER,
;;           start_time TIMESTAMP DEFAULT (strftime('%s','now')),
;;              CONSTRAINT hostport UNIQUE (hostname,portnum)
;;         );"
;;       ))
;;     db))
;; 
;; ;; register a dashboard 
;; ;;
;; (define (mddb:register-dashboard port)
;;   (let* ((pid      (current-process-id))
;; 	 (hostname (get-host-name))
;; 	 (ipaddr   (server:get-best-guess-address hostname))
;; 	 (username (current-user-name)) ;; (car userinfo)))
;; 	 (db      (mddb:open-db)))
;;     (print "Register monitor, pid: " pid ", hostname: " hostname ", port: " port ", username: " username)
;;     (exec (sql db "INSERT OR REPLACE INTO dashboards (pid,username,hostname,ipaddr,portnum) VALUES (?,?,?,?,?);")
;; 	   pid username hostname ipaddr port)
;;     (close-database db)))
;; 
;; ;; unregister a monitor
;; ;;
;; (define (mddb:unregister-dashboard host port)
;;   (let* ((db      (mddb:open-db)))
;;     (print "Register unregister monitor, host:port=" host ":" port)
;;     (exec (sql db "DELETE FROM dashboards WHERE hostname=? AND portnum=?;") host port)
;;     (close-database db)))
;; 
;; ;; get registered dashboards
;; ;;
;; (define (mddb:get-dashboards)
;;   (let ((db (mddb:open-db)))
;;     (query fetch-column
;; 	   (sql db "SELECT ipaddr || ':' || portnum FROM dashboards;"))))
    
;;======================================================================
;;  T E S T   L A U N C H I N G   P E R   I T E M   W I T H   H O S T   T Y P E S
;;======================================================================
;; 
;; [hosts]
;; arm cubie01 cubie02
;; x86_64 zeus xena myth01
;; allhosts #{g hosts arm} #{g hosts x86_64}
;; 
;; [host-types]
;; general #MTLOWESTLOAD #{g hosts allhosts}
;; arm     #MTLOWESTLOAD #{g hosts arm}
;; nbgeneral nbjob run JOBCOMMAND -log $MT_LINKTREE/$MT_TARGET/$MT_RUNNAME.$MT_TESTNAME-$MT_ITEM_PATH.lgo
;;
;; [host-rules]
;; # maxnload   => max normalized load
;; # maxnjobs   => max jobs per cpu
;; # maxjobrate => max jobs per second
;; general maxnload=1.1; maxnjobs=1.2; maxjobrate=0.1 
;; 
;; [launchers]
;; envsetup general
;; xor/%/n 4C16G
;; % nbgeneral
;; 
;; [jobtools]

		      (begin
			(debug:print-info 2 *default-log-port* "Have flexi-launcher match for " testname "/" itempath " = " host-type)
			(let ((launcher (configf:lookup configdat "host-types" host-type)))
			  (if launcher
			      (let* ((launcher-parts (string-split launcher))
				     (launcher-exe   (car launcher-parts)))
				(if (equal? launcher-exe "#MTLOWESTLOAD") ;; this is our special case, we will find the lowest load and craft a nbfake commandline
				    (let host-loop ((targ-host (common:get-least-loaded-host (cdr launcher-parts) host-type configdat))
						    (count     100))
				      (if targ-host
					  (conc "remrun " targ-host)
					  (if (> count 0)
					      (begin
						(debug:print 0 *default-log-port* "INFO: Waiting for a host for host-type " host-type)
						(thread-sleep! (- 101 count))
						(host-loop (common:get-least-loaded-host (cdr launcher-parts) host-type configdat)
							   (- count 1)))
					      (begin
						(debug:print 0 *default-log-port* "FATAL: Failed to find a host from #MTLOWESTLOAD for host-type " host-type)
						(exit)))))
				    launcher))
			      (begin
				(debug:print-info 0 *default-log-port* "WARNING: no launcher found for host-type " host-type)
				(if (null? tal)
				    fallback-launcher
				    (loop (car tal)(cdr tal)))))))
		      ;; no match, try again
		      (if (null? tal)
			  fallback-launcher
			  (loop (car tal)(cdr tal))))))))
	fallback-launcher)))

;;======================================================================
;; NMSG AND NEW API
;;======================================================================

;; nm based server experiment, keep around for now.
;;
(define (nm:start-server dbconn #!key (given-host-name #f))
  (let* ((srvdat    (start-raw-server given-host-name: given-host-name))
	 (host-name (srvdat-host srvdat))
	 (soc       (srvdat-soc srvdat)))
    
    ;; start the queue processor (save for second round of development)
    ;;
    (thread-start! (make-thread! (lambda ()(queue-processor dbconn) "Queue processor")))
    ;; msg is an alist
    ;;  'r host:port  <== where to return the data
    ;;  'p params     <== data to apply the command to
    ;;  'e j|s|l      <== encoding of the params. default is s (sexp), if not specified is assumed to be default
    ;;  'c command    <== look up the function to call using this key
    ;;
    (let loop ((msg-in (nn-recv soc)))
      (if (not (equal? msg-in "quit"))
	  (let* ((dat        (decode msg-in))
		 (host-port  (alist-ref 'r dat)) ;; this is for the reverse req rep where the server is a client of the original client
		 (params     (alist-ref 'p dat))
		 (command    (let ((c (alist-ref 'c dat)))(if c (string->symbol c) #f)))
		 (all-good   (and host-port params command (hash-table-exists? *commands* command))))
	    (if all-good
		(let ((cmddat (make-qitem
			       command:   command
			       host-port: host-port
			       params:    params)))
		  (queue-push cmddat) 		;; put request into the queue
		  (nn-send soc "queued"))         ;; reply with "queued"
		(print "ERROR: ["(common:human-time)"] BAD request " dat))
	    (loop (nn-recv soc)))))
    (nn-close soc)))

;;======================================================================
;; D A S H B O A R D   U S E R   V I E W S
;;======================================================================

;; first read ~/views.config if it exists, then read $MTRAH/views.config if it exists
;;
(define (common:load-views-config)
  (let* ((view-cfgdat    (make-hash-table))
	 (home-cfgfile   (conc (get-environment-variable "HOME") "/.mtviews.config"))
	 (mthome-cfgfile (conc *toppath* "/.mtviews.config")))
    (if (common:file-exists? mthome-cfgfile)
	(read-config mthome-cfgfile view-cfgdat #t))
    ;; we load the home dir file AFTER the MTRAH file so the user can clobber settings when running the dashboard in read-only areas
    (if (common:file-exists? home-cfgfile)
	(read-config home-cfgfile view-cfgdat #t))
    view-cfgdat))

;;======================================================================
;; H I E R A R C H I C A L   H A S H   T A B L E S
;;======================================================================

;; Every element including top element is a vector:
;;   <vector subhash value>

(define (hh:make-hh #!key (ht #f)(value #f))
  (vector (or ht    (make-hash-table)) value))

;; used internally
(define-inline (hh:set-ht! hh ht)       (vector-set! hh 0 ht))
(define-inline (hh:get-ht hh)           (vector-ref  hh 0))
(define-inline (hh:set-value! hh value) (vector-set! hh 1 value))
(define-inline (hh:get-value  hh value) (vector-ref  hh 1))

;; given a hierarchial hash and some keys look up the value ...
;;
(define (hh:get hh . keys)
  (if (null? keys)
      (vector-ref hh 1) ;; we have reached the end of the line, return the value sought
      (let ((sub-ht (hh:get-ht hh)))
	(if sub-ht ;; yes, there is more hierarchy
	    (let ((sub-hh (hash-table-ref/default sub-ht (car keys) #f)))
	      (if sub-hh
		  (apply hh:get sub-hh (cdr keys))
		  #f))
	    #f))))

;; given a hierarchial hash, a value and some keys, add needed hierarcy and insert the value
;;
(define (hh:set! hh value . keys)
  (if (null? keys)
      (hh:set-value! hh value) ;; we have reached the end of the line, store the value
      (let ((sub-ht (hh:get-ht hh)))
	(if sub-ht ;; yes, there is more hierarchy
	    (let ((sub-hh (hash-table-ref/default sub-ht (car keys) #f)))
	      (if (not sub-hh) ;; we'll need to add the next level of hierarchy
		  (let ((new-sub-hh (hh:make-hh)))
		    (hash-table-set! sub-ht (car keys) new-sub-hh)
		    (apply hh:set! new-sub-hh value (cdr keys)))
		  (apply hh:set! sub-hh value (cdr keys))))    ;; call the sub-hierhash with remaining keys
	    (begin
	      (hh:set-ht! hh (make-hash-table))
	      (apply hh:set! hh value keys))))))
  
;; Manage pkts, used in servers, tests and likely other contexts so put
;; in common
;;======================================================================

(define common:pkts-spec
  '((default . ((parent    . P)
                (action    . a)
                (filename  . f)))
    (configf . ((parent    . P)
                (action    . a)
                (filename  . f)))
    (server  . ((action    . a)
		(pid       . d)
		(ipaddr    . i)
		(port      . p)
		(parent    . P)))
    			  
    (test    . ((cpuuse    . c)
		(diskuse   . d)
		(item-path . i)
		(runname   . r)
		(state     . s)
		(target    . t)
		(status    . u)
		(parent    . P)))))

(define (common:get-pkts-dirs mtconf use-lt)
  (let* ((pktsdirs-str (or (configf:lookup mtconf "setup"  "pktsdirs")
			   (and use-lt
				(conc (or *toppath*
					  (current-directory))
				      "/lt/.pkts"))))
	 (pktsdirs  (if pktsdirs-str
			(string-split pktsdirs-str " ")
			#f)))
    pktsdirs))

;; use-lt is use linktree "lt" link to find pkts dir
(define (common:save-pkt pktalist-in mtconf use-lt #!key (add-only #f)) ;; add-only saves the pkt only if there is a parent already
  (if (or add-only
	  (hash-table-exists? *pkts-info* 'last-parent))
      (let* ((parent   (hash-table-ref/default *pkts-info* 'last-parent #f))
	     (pktalist (if parent
			   (cons `(parent . ,parent)
				 pktalist-in)
			   pktalist-in)))
	(let-values (((uuid pkt)
		      (alist->pkt pktalist common:pkts-spec)))
	  (hash-table-set! *pkts-info* 'last-parent uuid)
	  (let ((pktsdir (or (hash-table-ref/default *pkts-info* 'pkts-dir #f)
			     (let* ((pktsdirs (common:get-pkts-dirs mtconf use-lt))
				    (pktsdir   (car pktsdirs))) ;; assume it is there
			       (hash-table-set! *pkts-info* 'pkts-dir pktsdir)
			       pktsdir))))
            (handle-exceptions
             exn
             (debug:print-info 0 "failed to write out packet to " pktsdir) ;; don't care if this failed for now but MUST FIX - BUG!!
             (if (not (file-exists? pktsdir))
                 (create-directory pktsdir #t))
             (with-output-to-file
                 (conc pktsdir "/" uuid ".pkt")
               (lambda ()
                 (print pkt)))))))))
	
(define (common:with-queue-db mtconf proc #!key (use-lt #f)(toppath-in #f))
  (let* ((pktsdirs (common:get-pkts-dirs mtconf use-lt))
	 (pktsdir  (if pktsdirs (car pktsdirs) #f))
	 (toppath  (or (configf:lookup mtconf "scratchdat" "toppath")
		       toppath-in))
	 (pdbpath  (or (configf:lookup mtconf "setup"  "pdbpath") pktsdir)))
    (cond
     ((not (and  pktsdir toppath pdbpath))
      (debug:print 0 *default-log-port* "ERROR: settings are missing in your megatest.config for area management.")
      (debug:print  0 *default-log-port* "  you need to have pktsdirs in the [setup] section."))
     ((not (common:file-exists? pktsdir))
      (debug:print 0 *default-log-port* "ERROR: pkts directory not found " pktsdir))
     ((not (equal? (file-owner pktsdir)(current-effective-user-id)))
      (debug:print 0 *default-log-port* "ERROR: directory " pktsdir " is not owned by " (current-effective-user-name)))
     (else
	(let* ((pdb  (open-queue-db pdbpath "pkts.db"
				    schema: '("CREATE TABLE groups (id INTEGER PRIMARY KEY,groupname TEXT, CONSTRAINT group_constraint UNIQUE (groupname));"))))
	  (proc pktsdirs pktsdir pdb)
	  (dbi:close pdb))))))

(define (common:load-pkts-to-db mtconf #!key (use-lt #f))
  (common:with-queue-db
   mtconf
   (lambda (pktsdirs pktsdir pdb)
     (for-each
      (lambda (pktsdir) ;; look at all
	(cond
	 ((not (common:file-exists? pktsdir))
	  (debug:print 0 *default-log-port* "ERROR: packets directory " pktsdir " does not exist."))
	 ((not (directory? pktsdir))
	  (debug:print 0 *default-log-port* "ERROR: packets directory path " pktsdir " is not a directory."))
	 ((not (file-read-access? pktsdir))
	  (debug:print 0 *default-log-port* "ERROR: packets directory path " pktsdir " is not readable."))
	 (else
	  (debug:print-info 0 *default-log-port* "Loading packets found in " pktsdir)
	  (let ((pkts (glob (conc pktsdir "/*.pkt"))))
	    (for-each
	     (lambda (pkt)
	       (let* ((uuid    (cadr (string-match ".*/([0-9a-f]+).pkt" pkt)))
		      (exists  (lookup-by-uuid pdb uuid #f)))
		 (if (not exists)
		     (let* ((pktdat (string-intersperse
				     (with-input-from-file pkt read-lines)
				     "\n"))
			    (apkt   (pkt->alist pktdat))
			    (ptype  (alist-ref 'T apkt)))
		       (add-to-queue pdb pktdat uuid (or ptype 'cmd) #f 0)
		       (debug:print 4 *default-log-port* "Added " uuid " of type " ptype " to queue"))
		     (debug:print 4 *default-log-port* "pkt: " uuid " exists, skipping...")
		     )))
	     pkts)))))
      pktsdirs))
   use-lt: use-lt))

(define (common:get-pkt-alists pkts)
  (map (lambda (x)
	 (alist-ref 'apkt x)) ;; 'pkta pulls out the alist from the read pkt
       pkts))

;; given list of pkts (alist mode) return list of D cards as Unix epoch, sorted descending
;; also delete duplicates by target i.e. (car pkt)
;;
(define (common:get-pkt-times pkts)
  (delete-duplicates
   (sort 
    (map (lambda (x)
	   `(,(alist-ref 't x) . ,(string->number (alist-ref 'D x))))
	 pkts)
    (lambda (a b)(> (cdr a)(cdr b))))      ;; sort descending
   (lambda (a b)(equal? (car a)(car b))))) ;; remove duplicates by target



;; accept an alist or hash table containing envvar/env value pairs (value of #f causes unset) 
;;   execute thunk in context of environment modified as per this list
;;   restore env to prior state then return value of eval'd thunk.
;;   ** this is not thread safe **
(define (common:with-env-vars delta-env-alist-or-hash-table thunk)
  (let* ((delta-env-alist (if (hash-table? delta-env-alist-or-hash-table)
                              (hash-table->alist delta-env-alist-or-hash-table)
                              delta-env-alist-or-hash-table))
         (restore-thunks
          (filter
           identity
           (map (lambda (env-pair)
                  (let* ((env-var     (car env-pair))
                         (new-val     (let ((tmp (cdr env-pair)))
                                        (if (list? tmp) (car tmp) tmp)))
                         (current-val (get-environment-variable env-var))
                         (restore-thunk
                          (cond
                           ((not current-val) (lambda () (unsetenv env-var)))
                           ((not (string? new-val)) #f)
                           ((eq? current-val new-val) #f)
                           (else 
                            (lambda () (setenv env-var current-val))))))
                    ;;(when (not (string? new-val))
                    ;;    (debug:print 0 *default-log-port* " PROBLEM: not a string: "new-val"\n from env-alist:\n"delta-env-alist)
                    ;;    (pp delta-env-alist)
                    ;;    (exit 1))
                        
                    
                    (cond
                     ((not new-val)  ;; modify env here
                      (unsetenv env-var))
                     ((string? new-val)
                      (setenv env-var new-val)))
                    restore-thunk))
                delta-env-alist))))
    (let ((rv (thunk)))
      (for-each (lambda (x) (x)) restore-thunks) ;; restore env to original state
      rv)))

(define *common:thread-punchlist* (make-hash-table))
(define (common:send-thunk-to-background-thread thunk #!key (name #f))
  ;;(BB> "launched thread " name)

  ;; we need a unique name for the thread.
  (let* ((realname (if name
                       (if (not (hash-table-ref/default *common:thread-punchlist* name #f))
                           name
                           (conc name"-" (symbol->string (gensym))))
                       (conc "anonymous-"(symbol->string (gensym)))))
         (realthunk (lambda ()
                      (let ((res (thunk)))
                        (hash-table-delete! *common:thread-punchlist* realname)
                        res)))
         (thread   (make-thread realthunk realname)))
    (hash-table-set! *common:thread-punchlist* realname thread)
    (thread-start! thread)
    ))

(define (common:join-backgrounded-threads)
  ;; may need to trap and ignore exceptions -- dunno how atomic threads are...
  (for-each
   (lambda (thread-name)
     (let* ((thread (hash-table-ref/default *common:thread-punchlist* thread-name #f)))
       (if thread
           (handle-exceptions
           exn
           #t ;; just ignore it, it might have died in the meantime so joining it will throw an exception
           (thread-join! thread))
           )))
   (hash-table-keys *common:thread-punchlist*)))

(define *common:telemetry-log-state* 'startup)
(define *common:telemetry-log-socket* #f)

(define (common:telemetry-log-open)
  (if (eq? *common:telemetry-log-state* 'startup)
      (let* ((serverhost (configf:lookup *configdat* "telemetry" "host"))
             (serverport (configf:lookup-number *configdat* "telemetry" "port"))
             (user (or (get-environment-variable "USER") "unknown"))
             (host (or (get-environment-variable "HOST") "unknown")))
        (set! *common:telemetry-log-state*
              (handle-exceptions
               exn
               (begin
                 (debug:print-info 0 *default-log-port* "common-telemetry-log open udp port failure")
                 'broken)
               (if (and serverhost serverport user host)
                   (let* ((s (udp-open-socket)))
                     ;;(udp-bind! s #f 0)
                     (udp-connect! s serverhost serverport)
                     (set! *common:telemetry-log-socket* s)
                     'open)
                   'not-needed))))))
  
(define (common:telemetry-log event #!key (payload '()))
  (if (eq? *common:telemetry-log-state* 'startup)
      (common:telemetry-log-open))

  (if (eq? 'open *common:telemetry-log-state*)
      (handle-exceptions
       exn
       (begin
         (debug:print-info 0 *default-log-port* "common-telemetry-log comms failure ; disabled (no server?)")
         ;;(define *common:telemetry-log-state* 'broken-or-no-server-preclose)
         ;;(common:telemetry-log-close)
         (define *common:telemetry-log-state* 'broken-or-no-server)
         (set! *common:telemetry-log-socket* #f)
         )
       (if (and *common:telemetry-log-socket* event) ;; TODO - filter on event against telemetry.want-events
           (let* ((user (or (get-environment-variable "USER") "unknown"))
                  (host (or (get-environment-variable "HOST") "unknown"))
                  (start (conc "[megatest "event"]"))
                  (toppath (or *toppath* "/dev/null"))
                  (payload-serialized
                   (base64:base64-encode
                    (z3:encode-buffer
                     (with-output-to-string (lambda () (pp payload))))))
                  (msg     (conc user":"host":"start":"(current-process-id)":"(car (argv))":"
                                 toppath":"payload-serialized)))
             (udp-send *common:telemetry-log-socket* msg))))))
  
(define (common:telemetry-log-close)
  (when (or (member *common:telemetry-log-state* '(broken-or-no-server-preclose open)) *common:telemetry-log-socket*)
    (handle-exceptions
     exn
     (begin
       (define *common:telemetry-log-state* 'closed-fail)
       (debug:print-info 0 *default-log-port* "common-telemetry-log closure failure")
       )
     (begin
       (define *common:telemetry-log-state* 'closed)
       (udp-close-socket *common:telemetry-log-socket*)
       (set! *common:telemetry-log-socket* #f)))))