Megatest

	stack
	md5
	message-digest
	(prefix mtconfigf configf:)
	;; stml2
	;; (prefix margs args:)
	z3 (prefix base64 base64:)
	(prefix mtargs args:)
	directory-utils)

(include "common_records.scm")
(include "megatest-fossil-hash.scm")
(include "megatest-version.scm")

 ;; no need to export this
(define *verbosity-cache* (make-hash-table))

;;         res))))
        
;; 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 *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)

;; 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 (runconfigs-get config var)
  (let ((targ (common:args-get-target))) ;; (or (args:get-arg "-reqtarg")(args:get-arg "-target")(getenv "MT_TARGET"))))
    (if targ
	(or (configf:lookup config targ var)
	    (configf:lookup config "default" var))
	(configf:lookup config "default" var))))

;;======================================================================
;; client stuff
;;======================================================================

;; client:get-signature
(define (common:client-signature)
  ;; (if *my-client-signature* *my-client-signature*
  ;; (let ((sig
  (conc (get-host-name) " " (current-process-id)))
	;; )(set! *my-client-signature* sig)
	;; *my-client-signature*)))

(define client:get-signature common:client-signature)

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

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

  (string-split
   (with-input-from-pipe
       (conc "/bin/bash -c \"echo " instr "\"")
     read-line)))
  
(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:false-on-exception thunk #!key (message #f))
  (handle-exceptions exn
                     (begin
                       (if message
                           (debug:print-info 0 *default-log-port* message))
                       #f) (thunk) ))

  (handle-exceptions
   exn
   #f
   (if (and (directory-exists? path-string)
            (file-write-access? path-string))
       path-string
       #f)))
;;======================================================================
;; 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))))
    (conc (if (> hrs 0)(conc hrs "hr ") "")
	  (if (> min 0)(conc min "m ")  "")
	  sec "s")))

(define (seconds->time-string sec)
  (time->string 
   (seconds->local-time sec) "%H:%M:%S"))

(define (seconds->work-week/day-time sec)
  (time->string
   (seconds->local-time sec) "ww%V.%u %H:%M"))

(define (seconds->work-week/day sec)
  (time->string
   (seconds->local-time sec) "ww%V.%u"))

(define (seconds->year-work-week/day sec)
  (time->string
   (seconds->local-time sec) "%yww%V.%w"))

(define (seconds->year-work-week/day-time sec)
  (time->string
   (seconds->local-time sec) "%Yww%V.%w %H:%M"))

(define (seconds->year-week/day-time sec)
  (time->string
   (seconds->local-time sec) "%Yw%V.%w %H:%M"))

(define (seconds->quarter sec)
  (case (string->number
	 (time->string 
	  (seconds->local-time sec)
	  "%m"))
    ((1 2 3) 1)
    ((4 5 6) 2)
    ((7 8 9) 3)
    ((10 11 12) 4)
    (else #f)))

;; basic ISO8601 format (e.g. "2017-02-28 06:02:54") date time => Unix epoch
;;
(define (common:date-time->seconds datetime)
  (local-time->seconds (string->time datetime "%Y-%m-%d %H:%M:%S")))

;; given span of seconds tstart to tend
;; find start time to mark and mark delta
;;
(define (common:find-start-mark-and-mark-delta tstart tend)
  (let* ((deltat   (- (max tend (+ tend 10)) tstart)) ;; can't handle runs of less than 4 seconds. Pad it to 10 seconds ...
	 (result   #f)
	 (min      60)
	 (hr       (* 60 60))
	 (day      (* 24 hr))
	 (yr       (* 365 day)) ;; year
	 (mo       (/ yr 12))
	 (wk       (* day 7)))
    (for-each
     (lambda (max-blks)
       (for-each
	(lambda (span) ;; 5 2 1
	  (if (not result)
	      (for-each 
	       (lambda (timeunit timesym) ;; year month day hr min sec
		 (if (not result)
		     (let* ((time-blk (* span timeunit))
			    (num-blks (quotient deltat time-blk)))
		       (if (and (> num-blks 4)(< num-blks max-blks))
			   (let ((first (* (quotient tstart time-blk) time-blk)))
			     (set! result (list span timeunit time-blk first timesym))
			     )))))
	       (list yr mo wk day hr min 1)
	       '(     y  mo w  d   h  m   s))))
	(list 8 6 5 2 1)))
     '(5 10 15 20 30 40 50 500))
    (if values
	(apply values result)
	(values 0 day 1 0 'd))))

;; given x y lim return the cron expansion
;;
(define (common:expand-cron-slash x y lim)
  (let loop ((curr x)
	     (res  `()))
    (if (< curr lim)
	(loop (+ curr y) (cons curr res))
	(reverse res))))

;; expand a complex cron string to a list of cron strings
;;
;;  x/y   => x, x+y, x+2y, x+3y while x+Ny<max_for_field
;;  a,b,c => a, b ,c
;;
;;   NOTE: with flatten a lot of the crud below can be factored down.
;;
(define (common:cron-expand cron-str)
  (if (list? cron-str)
      (flatten
       (fold (lambda (x res)
	       (if (list? x)
		   (let ((newres (map common:cron-expand x)))
		     (append x newres))
		   (cons x res)))
	     '()
	     cron-str)) ;; (map common:cron-expand cron-str))
      (let ((cron-items (string-split cron-str))
	    (slash-rx   (regexp "(\\d+)/(\\d+)"))
	    (comma-rx   (regexp ".*,.*"))
	    (max-vals   '((min        . 60)
			  (hour       . 24)
			  (dayofmonth . 28) ;;; BUG!!!! This will be a bug for some combinations
			  (month      . 12)
			  (dayofweek  . 7))))
	(if (< (length cron-items) 5) ;; bad spec
	    cron-str ;; `(,cron-str)              ;; just return the string, something downstream will fix it
	    (let loop ((hed  (car cron-items))
		       (tal  (cdr cron-items))
		       (type 'min)
		       (type-tal '(hour dayofmonth month dayofweek))
		       (res  '()))
	      (regex-case
		  hed
		(slash-rx ( _ base incr ) (let* ((basen          (string->number base))
						 (incrn          (string->number incr))
						 (expanded-vals  (common:expand-cron-slash basen incrn (alist-ref type max-vals)))
						 (new-list-crons (fold (lambda (x myres)
									 (cons (conc (if (null? res)
											 ""
											 (conc (string-intersperse res " ") " "))
										     x " " (string-intersperse tal " "))
									       myres))
								       '() expanded-vals)))
					    ;; (print "new-list-crons: " new-list-crons)
					    ;; (fold (lambda (x res)
					    ;; 	    (if (list? x)
					    ;; 		(let ((newres (map common:cron-expand x)))
					    ;; 		  (append x newres))
					    ;; 		(cons x res)))
					    ;; 	  '()
					    (flatten (map common:cron-expand new-list-crons))))
		;;					    (map common:cron-expand (map common:cron-expand new-list-crons))))
		(else (if (null? tal)
			  cron-str
			  (loop (car tal)(cdr tal)(car type-tal)(cdr type-tal)(append res (list hed)))))))))))
		      
	    
;; given a cron string and the last time event was processed return #t to run or #f to not run
;;
;;  min    hour   dayofmonth month  dayofweek
;; 0-59    0-23   1-31       1-12   0-6          ### NOTE: dayofweek does not include 7
;;
;;  #t => yes, run the job
;;  #f => no, do not run the job
;;
(define (common:cron-event cron-str now-seconds-in last-done) ;; ref-seconds = #f is NOW.
  (let* ((cron-items     (map string->number (string-split cron-str)))
	 (now-seconds    (or now-seconds-in (current-seconds)))
	 (now-time       (seconds->local-time now-seconds))
	 (last-done-time (seconds->local-time last-done))
	 (all-times      (make-hash-table)))
    ;; (print "cron-items: " cron-items "(length cron-items): " (length cron-items))
    (if (not (eq? (length cron-items) 5)) ;; don't even try to figure out junk strings
	#f
	(match-let (((     cmin chour cdayofmonth cmonth    cdayofweek)
		     cron-items)
		    ;; 0     1    2        3         4    5      6
		    ((nsec nmin nhour ndayofmonth nmonth nyr ndayofweek n7 n8 n9)
		     (vector->list now-time))
		    ((lsec lmin lhour ldayofmonth lmonth lyr ldayofweek l7 l8 l9)
		     (vector->list last-done-time)))
	  ;; create all possible time slots
	  ;; remove invalid slots due to (for example) day of week
	  ;; get the start and end entries for the ref-seconds (current) time
	  ;; if last-done > ref-seconds => this is an ERROR!
	  ;; does the last-done time fall in the legit region?
	  ;;    yes => #f  do not run again this command
	  ;;    no  => #t  ok to run the command
	  (for-each ;; month
	   (lambda (month)
	     (for-each ;; dayofmonth
	      (lambda (dom)
		(for-each
		 (lambda (hr) ;; hour
		   (for-each
		    (lambda (minute) ;; minute
		      (let ((copy-now (apply vector (vector->list now-time))))
			(vector-set! copy-now 0 0) ;; force seconds to zero
			(vector-set! copy-now 1 minute)
			(vector-set! copy-now 2 hr)
			(vector-set! copy-now 3 dom)  ;; dom is already corrected for zero referenced
			(vector-set! copy-now 4 month)
			(let* ((copy-now-secs (local-time->seconds copy-now))
			       (new-copy      (seconds->local-time copy-now-secs))) ;; remake the time vector
			  (if (or (not cdayofweek)
				  (equal? (vector-ref new-copy 6)
					  cdayofweek)) ;; if the day is specified and a match OR if the day is NOT specified
			      (if (or (not cdayofmonth)
				      (equal? (vector-ref new-copy 3)
					      (+ 1 cdayofmonth))) ;; if the month is specified and a match OR if the month is NOT specified
				  (hash-table-set! all-times copy-now-secs new-copy))))))
		    (if cmin
			`(,cmin)  ;; if given cmin, have to use it
			(list (- nmin 1) nmin (+ nmin 1))))) ;; minute
		 (if chour
		     `(,chour)
		     (list (- nhour 1) nhour (+ nhour 1))))) ;; hour
	      (if cdayofmonth
		  `(,cdayofmonth)
		  (list (- ndayofmonth 1) ndayofmonth (+ ndayofmonth 1)))))
	   (if cmonth
	       `(,cmonth)
	       (list (- nmonth 1) nmonth (+ nmonth 1))))
	  (let ((before #f)
		(is-in  #f))
	    (for-each
	     (lambda (moment)
	       (if (and before
			(<= before now-seconds)
			(>= moment now-seconds))
		   (begin
		     ;; (print)
		     ;; (print "Before: " (time->string (seconds->local-time before)))
		     ;; (print "Now:    " (time->string (seconds->local-time now-seconds)))
		     ;; (print "After:  " (time->string (seconds->local-time moment)))
		     ;; (print "Last:   " (time->string (seconds->local-time last-done)))
		     (if (<  last-done before)
			 (set! is-in before))
		     ))
	       (set! before moment))
	     (sort (hash-table-keys all-times) <))
	    is-in)))))

(define (common:extended-cron  cron-str now-seconds-in last-done)
  (let ((expanded-cron (common:cron-expand cron-str)))
    (if (string? expanded-cron)
	(common:cron-event expanded-cron now-seconds-in last-done)
	(let loop ((hed (car expanded-cron))
		   (tal (cdr expanded-cron)))
	  (if (common:cron-event hed now-seconds-in last-done)
	      #t
	      (if (null? tal)
		  #f
		  (loop (car tal)(cdr tal))))))))


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

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

(define (common:clear-caches)
  (set! *target*             (make-hash-table))
  (set! *keys*               (make-hash-table))
  ;; (set! *keyvals*            (make-hash-table))
  (set! *toptest-paths*      (make-hash-table))
  (set! *test-paths*         (make-hash-table))
  (set! *test-ids*           (make-hash-table))
  (set! *test-info*          (make-hash-table))
  (set! *run-info-cache*     (make-hash-table))
  (set! *env-vars-by-run-id* (make-hash-table))
  (set! *test-id-cache*      (make-hash-table)))

  (if (or (number? val)(string? val)) val ""))

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

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

;; NB// #f => return dbdir only
;;      (was planned to be;  zeroth db with name=main.db)
;; 
;; If run-id is #f return to create and retrieve the path where the db will live.
;;
(define db:dbfile-path common:get-db-tmp-area)

;;





















(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*))

;; 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 0 *default-log-port* "unable to rotate log " file ", probably handled by another process.")
	  (debug:print 0 *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)))
    (sparse-vector-set! a 0 (make-sparse-vector))
    a))

(define (sparse-array? a)
  (and (sparse-vector? a)
       (sparse-vector? (sparse-vector-ref a 0))))

(define (sparse-array-ref a x y)
  (let ((row (sparse-vector-ref a x)))
    (if row
	(sparse-vector-ref row y)
	#f)))

(define (sparse-array-set! a x y val)
  (let ((row (sparse-vector-ref a x)))
    (if row
	(sparse-vector-set! row y val)
	(let ((new-row (make-sparse-vector)))
	  (sparse-vector-set! a x new-row)
	  (sparse-vector-set! new-row y val)))))

;;======================================================================
;; L O C K E R S   A N D   B L O C K E R S 
;;======================================================================

;; block further accesses to databases. Call this before shutting db down
(define (common:db-block-further-queries)
  (mutex-lock! *db-access-mutex*)
  (set! *db-access-allowed* #f)
  (mutex-unlock! *db-access-mutex*))

(define (common:db-access-allowed?)
  (let ((val (begin
	       (mutex-lock! *db-access-mutex*)
	       *db-access-allowed*
	       (mutex-unlock! *db-access-mutex*))))
    val))

;;======================================================================
;; U S E F U L   S T U F F
;;======================================================================

;; 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))
    (if (common:file-exists? fname)
	(if (> (- (current-seconds)(file-modification-time fname)) 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
	  (if (> end-time (current-seconds))
	      (begin
		(thread-sleep! 3)
		(loop (common:simple-file-lock fname expire-time: expire-time)))
	      #f)))))

(define (common:simple-file-release-lock fname)
  (handle-exceptions
      exn
      #f ;; I don't really care why this failed (at least for now)
    (delete-file* fname)))

;;======================================================================
;; D E B U G G I N G   S T U F F 
;;======================================================================

;; (define *verbosity*         1)
;; (define *logging*           #f)

(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))))

;; from metadat lookup MEGATEST_VERSION
;;
(define (common:get-last-run-version) ;; RADT => How does this work in send-receive function??; assume it is the value saved in some DB
  (rmt:get-var "MEGATEST_VERSION"))

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

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


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

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

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

(define *watchdog* (make-thread
		    (lambda ()
		      (handle-exceptions
			  exn
			  (begin
			    (print-call-chain)
			    (print " message: " ((condition-property-accessor 'exn 'message) exn)))
			(common:watchdog)))
		    "Watchdog thread"))

;; 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
                                    (thread-sleep! 5)) ;; give the clean up few seconds to do it's stuff
                                  (begin
      				  (thread-sleep! 2)))
      			      (debug:print 4 *default-log-port* " ... done")
      			      )
			    "clean exit")))
      (thread-start! th1)
      (thread-start! th2)
      (thread-join! th1)
      )
    )

  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!

;; '(print (string-intersperse (map cadr (hash-table-ref/default (read-config "megatest.config" \#f \#t) "disks" '"'"'("none" ""))) "\n"))'
(define (common:get-disks #!key (configf #f))
  (hash-table-ref/default 
   (or configf (read-config "megatest.config" #f #t))
   "disks" '("none" "")))

;;======================================================================
;; 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:get-runconfig-targets #!key (configf #f))
  (let ((targs       (sort (map car (hash-table->alist
				     (or configf ;; NOTE: There is no value in using runconfig:read here.
					 (read-config (conc *toppath* "/runconfigs.config")
						      #f #t)
					 (make-hash-table))))
			   string<?))
	(target-patt (args:get-arg "-target")))
    (if target-patt
	(filter (lambda (x)
		  (patt-list-match x target-patt))
		targs)
	targs)))

;; 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
  (mutex-lock! *homehost-mutex*)
  (cond
   (*home-host*
    (mutex-unlock! *homehost-mutex*)
    *home-host*)
   ((not *toppath*)
    (mutex-unlock! *homehost-mutex*)
    (launch:setup) ;; safely mutexed now
    (if (> trynum 0)
	(begin
	  (thread-sleep! 2)
	  (common:get-homehost trynum: (- trynum 1)))
	#f))
   (else
    (let* ((currhost (get-host-name))
	   (bestadrs (server:get-best-guess-address currhost))
	   ;; 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
					 (mutex-unlock! *homehost-mutex*)
					 (car (common:get-homehost))))
				     #f))))))
	   (at-home  (or (equal? homehost currhost)
			 (equal? homehost bestadrs))))
      (set! *home-host* (cons homehost at-home))
      (mutex-unlock! *homehost-mutex*)
      *home-host*))))

;; am I on the homehost?
;;
(define (common:on-homehost?)
  (let ((hh (common:get-homehost)))
    (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)))

;; moving common:htree->html to testsmod.scm to minimize deps on stml2

;; hash-table tree to alist tree
;;
(define (common:htree->atree ht)
  (map (lambda (x)
	 (cons (car x)
	       (let ((y (cdr x)))
		 (if (hash-table? y)
		     (common:htree->atree y)
		     y))))
       (hash-table->alist ht)))

;;======================================================================
;; M U N G E   D A T A   I N T O   N I C E   F O R M S
;;======================================================================

;; Generate an index for a sparse list of key values
;;   ( (rowname1 colname1 val1)(rowname2 colname2 val2) )
;;
;; => 
;;
;;   ( (rowname1 0)(rowname2 1))    ;; rownames -> num
;;     (colname1 0)(colname2 1)) )  ;; colnames -> num
;; 
;; optional apply proc to rownum colnum value
(define (common:sparse-list-generate-index data #!key (proc #f))
  (if (null? data)
      (list '() '())
      (let loop ((hed (car data))
		 (tal (cdr data))
		 (rownames '())
		 (colnames '())
		 (rownum   0)
		 (colnum   0))
	(let* ((rowkey          (car   hed))
	       (colkey          (cadr  hed))
	       (value           (caddr hed))
	       (existing-rowdat (assoc rowkey rownames))
	       (existing-coldat (assoc colkey colnames))
	       (curr-rownum     (if existing-rowdat rownum (+ rownum 1)))
	       (curr-colnum     (if existing-coldat colnum (+ colnum 1)))
	       (new-rownames    (if existing-rowdat rownames (cons (list rowkey curr-rownum) rownames)))
	       (new-colnames    (if existing-coldat colnames (cons (list colkey curr-colnum) colnames))))
	  ;; (debug:print-info 0 *default-log-port* "Processing record: " hed )
	  (if proc (proc curr-rownum curr-colnum rowkey colkey value))
	  (if (null? tal)
	      (list new-rownames new-colnames)
	      (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
;;
(define (common:lazy-modification-time fpath)
  (handle-exceptions
      exn
      0
    (file-modification-time fpath)))

;; find timestamp of newest file associated with a sqlite db file
(define (common:lazy-sqlite-db-modification-time fpath)
  (let* ((glob-list (handle-exceptions
			exn
			`(,(conc "/no/such/file, message: " ((condition-property-accessor 'exn 'message) exn)))
		      (glob (conc fpath "*"))))
         (file-list (if (eq? 0 (length glob-list))
			'("/no/such/file")
			glob-list)))
  (apply max
   (map
    common:lazy-modification-time 
    file-list))))

;; return a nice clean pathname made absolute
(define (common:nice-path dir)
  (let ((match (string-match "^(~[^\\/]*)(\\/.*|)$" dir)))
    (if match ;; using ~ for home?
	(common:nice-path (conc (common:read-link-f (cadr match)) "/" (caddr match)))
	(normalize-pathname (if (absolute-pathname? dir)
				dir
				(conc (current-directory) "/" dir))))))

;; make "nice-path" available in config files and the repl
(define nice-path common:nice-path)

(define (common:read-link-f path)
  (handle-exceptions
      exn
      (begin
	(debug:print-error 0 *default-log-port* "command \"/bin/readlink -f " path "\" failed.")
	path) ;; just give up
    (with-input-from-pipe
	(conc "/bin/readlink -f " path)
      (lambda ()
	(read-line)))))

(define (get-cpu-load #!key (remote-host #f))
  (car (common:get-cpu-load remote-host)))
;;   (let* ((load-res (process:cmd-run->list "uptime"))
;; 	 (load-rx  (regexp "load average:\\s+(\\d+)"))
;; 	 (cpu-load #f))
;;     (for-each (lambda (l)
;; 		(let ((match (string-search load-rx l)))
;; 		  (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 5))
  (let* ((fullpath (conc *toppath* "/logs/" 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)
	       (with-input-from-file fullpath read)
	       (begin
		 (debug:print 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))))
 
(define (common:write-cached-info key dtype dat)
  (let* ((fullpath (conc *toppath* "/logs/" key "-" dtype ".log")))
    (handle-exceptions
     exn
     #f
     (with-output-to-file fullpath (lambda ()(pp dat))))))

;; 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)))
                          (if h
                              h
                              (let ((h (make-host)))
                                (hash-table-set! *host-loads* hostname h)
                                h))))
              (host-info         (common:get-host-info hostname))
              (is-reachable      (car host-info))
              (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-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
		      (max maxload-in 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
    (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 (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))
         (numcpus (common:get-num-cpus hh)))
    (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: 86400) ;; 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)
			       (begin
				 (common:write-cached-info remote-host "num-cpus" numcpu)
				 numcpu)
			       (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))))
	  (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)
  (let ((num-cpus (common:get-num-cpus remote-host)))
    (common:wait-for-cpuload maxload num-cpus 15 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))))

;; for reasons I don't understand multiple calls to real-path in parallel threads
;; must be protected by mutexes
;;
(define (common:real-path inpath)
  ;; (process:cmd-run-with-stderr->list "readlink" "-f" inpath)) ;; cmd . params)
  ;; (let-values 
  ;;  (((inp oup pid) (process "readlink" (list "-f" inpath))))
  ;;  (with-input-from-port inp
  ;;    (let loop ((inl (read-line))
  ;;       	(res #f))
  ;;      (print "inl=" inl)
  ;;      (if (eof-object? inl)
  ;;          (begin
  ;;            (close-input-port inp)
  ;;            (close-output-port oup)
  ;;            ;; (process-wait pid)
  ;;            res)
  ;;          (loop (read-line) inl))))))
  (with-input-from-pipe (conc "readlink -f " inpath) read-line))

;;======================================================================
;; D I S K   S P A C E 
;;======================================================================

(define (common:get-disk-space-used fpath)
  (with-input-from-pipe (conc "/usr/bin/du -s " fpath) read))

;; given path get free space, allows override in [setup]
;; with free-space-script /path/to/some/script.sh
;;
(define (get-df path)
  (if (configf:lookup *configdat* "setup" "free-space-script")
      (with-input-from-pipe 
       (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 
		    (or (configf:lookup *configdat* "setup" "dbdir-space-required")
			"100000")))
	 (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
;;
(define (common:check-db-dir-and-exit-if-insufficient)
  (let* ((spacedat (car (common:check-db-dir-space))) ;; look only at worst for now
	 (is-ok    (car spacedat))
	 (dbspace  (cadr spacedat))
	 (required (caddr spacedat))
	 (dbdir    (cadddr spacedat)))
    (if (not is-ok)
	(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)
        (min-inodes (or (string->number (if (configf:lookup *configdat* "setup" "min_inodes") (configf:lookup *configdat* "setup" "min_inodes") "0")) 0)))
    (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))
             )
	 (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))
	      (let ((var (car vardat))
		    (val (cdr vardat)))
		(hash-table-set! vars var val)
		(unsetenv var))))
	var-patts))
     (get-environment-variables))
    (cond
     ((string? proc)(system proc))
     (proc          (proc)))
    (hash-table-for-each
     vars
     (lambda (var val)
       (setenv var val)))
    vars))


;;======================================================================
;; C O L O R S
;;======================================================================
      
(define (common:name->iup-color name)
  (case (string->symbol (string-downcase name))
    ((red)    "223 33 49")
    ((grey)   "192 192 192")
    ((orange) "255 172 13")
    ((purple) "This is unfinished ...")))

;; (define (common:get-color-for-state-status state status)
;;   (case (string->symbol state)
;;     ((COMPLETED)
;;      (case (string->symbol status)
;;        ((PASS)        "70  249 73")
;;        ((WARN WAIVED) "255 172 13")
;;        ((SKIP)        "230 230 0")
;;        (else "223 33 49")))
;;     ((LAUNCHED)         "101 123 142")
;;     ((CHECK)            "255 100 50")
;;     ((REMOTEHOSTSTART)  "50  130 195")
;;     ((RUNNING)          "9   131 232")
;;     ((KILLREQ)          "39  82  206")
;;     ((KILLED)           "234 101 17")
;;     ((NOT_STARTED)      "240 240 240")
;;     (else               "192 192 192")))

(define (common:iup-color->rgb-hex instr)
  (string-intersperse 
   (map (lambda (x)
          (number->string x 16))
        (map string->number
             (string-split instr)))
   "/"))


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

(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;"))))
    
;;======================================================================
;; 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

(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)))))

;;======================================================================
;; process related stuff
;;======================================================================

(define (process:alive? pid)
  (handle-exceptions
   exn
   ;; possibly pid is a process not a child, look in /proc to see if it is running still
   (common:file-exists? (conc "/proc/" pid))
   (let-values (((rpid exit-type exit-signal)(process-wait pid #t)))
       (and (number? rpid)
	    (equal? rpid pid)))))

(define (process:alive-on-host? host pid)
  (let ((cmd (conc "ssh " host " ps -o pid= -p " pid)))
    (handle-exceptions
     exn
     #f ;; anything goes wrong - assume the process in NOT running.
     (with-input-from-pipe 
      cmd
      (lambda ()
	(let loop ((inl (read-line)))
	  (if (eof-object? inl)
	      #f
	      (let* ((clean-str (string-substitute "^[^\\d]*([0-9]+)[^\\d]*$" "\\1" inl))
		     (innum     (string->number clean-str)))
		(and innum
		     (eq? pid innum))))))))))

(define (process:get-sub-pids pid)
  (with-input-from-pipe
   (conc "pstree -A -p " pid) ;; | tr 'a-z\\-+`()\\.' ' ' " pid)
   (lambda ()
     (let loop ((inl (read-line))
		(res '()))
       (if (eof-object? inl)
	   (reverse res)
	   (let ((nums (map string->number
			    (string-split-fields "\\d+" inl))))
	     (loop (read-line)
		   (append res nums))))))))

;; here is an example line where the shell is sh or bash
;; "find / -print 2&>1 > findall.log"
(define (run-n-wait cmdline #!key (params #f)(print-cmd #f)(run-dir #f))
  (if print-cmd 
      (debug:print 0 *default-log-port* 
		   (if (string? print-cmd)
		       print-cmd
		       "")
		   (if run-dir (conc "Run in " run-dir ";") "")
		   cmdline
		   (if params
		       (conc " " (string-intersperse params " "))
		       "")))
  (if (and run-dir
	   (directory-exists? run-dir))
      (push-directory run-dir))
  (let ((pid (if params
		 (process-run cmdline params)
		 (process-run cmdline))))
    (let loop ((i 0))
      (let-values (((pid-val exit-status exit-code) (process-wait pid #t)))
         (if (eq? pid-val 0)
	     (begin
	       (thread-sleep! 2)
	       (loop (+ i 1)))
	     (begin
	       (if (and run-dir
			(directory-exists? run-dir))
		   (pop-directory))
	       (values pid-val exit-status exit-code)))))))
  


;; 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)))

;;======================================================================
;; Process convience utils
;;======================================================================

(define (process:conservative-read port)
  (let loop ((res ""))
    (if (not (eof-object? (peek-char port)))
	(loop (conc res (read-char port)))
	res)))

(define (process:cmd-run-with-stderr->list cmd . params)
  ;; (print "Called with cmd=" cmd ", proc=" proc ", params=" params)
;;  (handle-exceptions
;;   exn
;;   (begin
;;     (print "ERROR:  Failed to run command: " cmd " " (string-intersperse params " "))
;;     (print "       " ((condition-property-accessor 'exn 'message) exn))
;;     #f)
   (let-values (((fh fho pid fhe) (if (null? params)
				      (process* cmd)
				      (process* cmd params))))
       (let loop ((curr (read-line fh))
		  (result  '()))
	 (let ((errstr (process:conservative-read fhe)))
	   (if (not (string=? errstr ""))
	       (set! result (append result (list errstr)))))
       (if (not (eof-object? curr))
	   (loop (read-line fh)
		 (append result (list curr)))
	   (begin
	     (close-input-port fh)
	     (close-input-port fhe)
	     (close-output-port fho)
	     result))))) ;; )

(define (process:cmd-run-with-stderr-and-exitcode->list cmd . params)
  ;; (print "Called with cmd=" cmd ", proc=" proc ", params=" params)
;;  (handle-exceptions
;;   exn
;;   (begin
;;     (print "ERROR:  Failed to run command: " cmd " " (string-intersperse params " "))
;;     (print "       " ((condition-property-accessor 'exn 'message) exn))
;;     #f)
   (let-values (((fh fho pid fhe) (if (null? params)
				      (process* cmd)
				      (process* cmd params))))
       (let loop ((curr (read-line fh))
		  (result  '()))
	 (let ((errstr (process:conservative-read fhe)))
	   (if (not (string=? errstr ""))
	       (set! result (append result (list errstr)))))
       (if (not (eof-object? curr))
	   (loop (read-line fh)
		 (append result (list curr)))
	   (begin
             (let-values (((anotherpid normalexit? exitstatus)  (process-wait pid)))
	     (close-input-port fh)
	     (close-input-port fhe)
	     (close-output-port fho)
               (list result (if normalexit? exitstatus -1))))))))

#;(define (process:cmd-run-proc-each-line cmd proc . params)
  ;; (print "Called with cmd=" cmd ", proc=" proc ", params=" params)
  (handle-exceptions
   exn
   (begin
     (print "ERROR:  Failed to run command: " cmd " " (string-intersperse params " "))
     (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
     (debug:print 5 *default-log-port* "exn=" (condition->list exn))
     #f)
   (let-values (((fh fho pid) (if (null? params)
				  (process cmd)
				  (process cmd params))))
       (let loop ((curr (read-line fh))
		(result  '()))
       (if (not (eof-object? curr))
	   (loop (read-line fh)
		 (append result (list (proc curr))))
	   (begin
	     (close-input-port fh)
	     ;;(close-input-port fhe)
	     (close-output-port fho)
	     result))))))

#;(define (process:cmd-run-proc-each-line-alt cmd proc)
  (let* ((fh (open-input-pipe cmd))
         (res (port-proc->list fh proc))
         (status (close-input-pipe fh)))
    (if (eq? status 0) res #f)))

(define (process:cmd-run->list cmd #!key (delta-env-alist-or-hash-table '()))
  (common:with-env-vars
   delta-env-alist-or-hash-table
   (lambda ()
     (let* ((fh (open-input-pipe cmd))
            (res (port->list fh))
            (status (close-input-pipe fh)))
       (list res status)))))
   
(define (port->list fh)
  (if (eof-object? fh) #f
      (let loop ((curr (read-line fh))
                 (result '()))
        (if (not (eof-object? curr))
            (loop (read-line fh)
                  (append result (list curr)))
            result))))

(define (port-proc->list fh proc)
  (if (eof-object? fh) #f
      (let loop ((curr (proc (read-line fh)))
                 (result '()))
        (if (not (eof-object? curr))
            (loop (let ((l (read-line fh)))
                    (if (eof-object? l) l (proc l)))
                  (append result (list curr)))
            result))))

;;======================================================================
;; MISC PROCESS RELATED STUFF
;;======================================================================

(define (process:children proc)
  (with-input-from-pipe
   (conc "ps h --ppid " (current-process-id) " -o pid")
   (lambda ()
     (let loop ((inl (read-line))
		(res '()))
       (if (eof-object? inl)
	   (reverse res)
	   (let ((pid (string->number inl)))
	     (if proc (proc pid))
	     (loop (read-line) (cons pid res))))))))


)