Megatest

tests/installall/megatest.db
tests/installall/monitor.db
tests/megatest.db
tests/fdktestqa/simplelinks/*
tests/fdktestqa/testqa/megatest.db
tests/fdktestqa/testqa/monitor.db
megatest-fossil-hash.scm
tests/release/runs/*
tests/release/links/*
tests/release/megatest.db

# make install CSCOPTS='-accumulate-profile -profile-name $(PWD)/profile-ww$(shell date +%V.%u)'
PREFIX=$(PWD)
CSCOPTS= 
INSTALL=install
SRCFILES = common.scm items.scm launch.scm \
           ods.scm runconfig.scm server.scm configf.scm \
           db.scm keys.scm margs.scm megatest-version.scm \
           process.scm runs.scm tasks.scm tests.scm genexample.scm \
	   http-transport.scm nmsg-transport.scm filedb.scm \
           client.scm gutils.scm synchash.scm daemon.scm mt.scm dcommon.scm \
	   tree.scm ezsteps.scm lock-queue.scm sdb.scm \
	   rmt.scm api.scm tdb.scm rpc-transport.scm \
	   portlogger.scm archive.scm

# Eggs to install (straightforward ones)
EGGS=matchable readline apropos base64 regex-literals format regex-case test coops trace csv \
     dot-locking posix-utils posix-extras directory-utils hostinfo tcp-server rpc csv-xml fmt \
     json md5 awful http-client spiffy uri-common intarweb spiffy-request-vars \
     spiffy-directory-listing ssax sxml-serializer sxml-modifications iup canvas-draw sqlite3

CKPATH=$(shell dirname $(shell dirname $(CSIPATH)))
# ARCHSTR=$(shell uname -m)_$(shell uname -r)
# BASH_MACHTYPE=$(shell bash -c "echo \$$MACHTYPE")
# ARCHSTR=$(BASH_MACHTYPE)_$(shell lsb_release -sr)
ARCHSTR=$(shell lsb_release -sr)
# ARCHSTR=$(shell bash -c "echo \$$MACHTYPE")

all : $(PREFIX)/bin/.$(ARCHSTR) mtest dboard 




mtest: $(OFILES) readline-fix.scm megatest.o
	csc $(CSCOPTS) $(OFILES) megatest.o -o mtest

dboard : $(OFILES) $(GOFILES) dashboard.scm
	csc $(OFILES) dashboard.scm $(GOFILES) -o dboard

ndboard : newdashboard.scm $(OFILES) $(GOFILES)
	csc $(OFILES) $(GOFILES) newdashboard.scm -o ndboard

multi-dboard : multi-dboard.scm $(OFILES) $(GOFILES)
	csc $(OFILES) $(GOFILES) multi-dboard.scm -o multi-dboard

# 
# $(PREFIX)/bin/revtagfsl : utils/revtagfsl.scm
#	csc utils/revtagfsl.scm -o $(PREFIX)/bin/revtagfsl

# Special dependencies for the includes
tests.o db.o launch.o runs.o dashboard-tests.o dashboard-guimonitor.o dashboard-main.o monitor.o dashboard.o  \
  archive.o megatest.o : db_records.scm
tests.o runs.o dashboard.o dashboard-tests.o dashboard-main.o  : run_records.scm
db.o ezsteps.o keys.o launch.o megatest.o monitor.o runs-for-ref.o runs.o tests.o : key_records.scm
tests.o tasks.o dashboard-tasks.o : task_records.scm
runs.o : test_records.scm
megatest.o : megatest-fossil-hash.scm
client.scm common.scm configf.scm dashboard-guimonitor.scm dashboard-tests.scm dashboard.scm db.scm dcommon.scm ezsteps.scm fs-transport.scm http-transport.scm index-tree.scm items.scm keys.scm launch.scm megatest.scm monitor.scm mt.scm newdashboard.scm runconfig.scm runs.scm server.scm tdb.scm tests.scm tree.scm zmq-transport.scm : common_records.scm rpc-transport.scm

# Temporary while transitioning to new routine
# runs.o : run-tests-queue-classic.scm  run-tests-queue-new.scm

megatest-fossil-hash.scm : $(SRCFILES) megatest.scm *_records.scm
	echo "(define megatest-fossil-hash \"$(MTESTHASH)\")" > megatest-fossil-hash.new
	if ! diff -q megatest-fossil-hash.new megatest-fossil-hash.scm ; then echo copying .new to .scm;cp -f megatest-fossil-hash.new megatest-fossil-hash.scm;fi

$(OFILES) $(GOFILES) : common_records.scm 

%.o : %.scm
	csc $(CSCOPTS) -c $<

$(PREFIX)/bin/.$(ARCHSTR)/mtest : mtest
	@echo Installing to PREFIX=$(PREFIX)
	$(INSTALL) mtest $(PREFIX)/bin/.$(ARCHSTR)/mtest
	utils/mk_wrapper $(PREFIX) mtest $(PREFIX)/bin/megatest
	chmod a+x $(PREFIX)/bin/megatest

$(PREFIX)/bin/.$(ARCHSTR)/ndboard : ndboard
	$(INSTALL) ndboard $(PREFIX)/bin/.$(ARCHSTR)/ndboard

$(PREFIX)/bin/newdashboard : $(PREFIX)/bin/.$(ARCHSTR)/ndboard
	utils/mk_wrapper $(PREFIX) ndboard $(PREFIX)/bin/newdashboard
	chmod a+x $(PREFIX)/bin/newdashboard

$(PREFIX)/bin/.$(ARCHSTR)/mdboard : multi-dboard
	$(INSTALL) multi-dboard $(PREFIX)/bin/.$(ARCHSTR)/mdboard

$(PREFIX)/bin/mdboard : $(PREFIX)/bin/.$(ARCHSTR)/mdboard
	utils/mk_wrapper $(PREFIX) mdboard $(PREFIX)/bin/mdboard
	chmod a+x $(PREFIX)/bin/mdboard

# $(HELPERS) : utils/%
# 	$(INSTALL) $< $@
# 	chmod a+x $@

$(PREFIX)/bin/mt_laststep : utils/mt_laststep
	$(INSTALL) $< $@
	chmod a+x $@

$(PREFIX)/bin/mt_runstep : utils/mt_runstep
	$(INSTALL) $< $@
	chmod a+x $@

$(PREFIX)/bin/mt_ezstep : utils/mt_ezstep
	$(INSTALL) $< $@
	chmod a+x $@

$(PREFIX)/bin/mt_xterm : utils/mt_xterm
	$(INSTALL) $< $@
	chmod a+x $@

$(PREFIX)/bin/nbfake : utils/nbfake
	$(INSTALL) $< $@
	chmod a+x $@

$(PREFIX)/bin/nbfind : utils/nbfind
	$(INSTALL) $< $@
	chmod a+x $@

$(PREFIX)/bin/loadrunner : utils/loadrunner
	$(INSTALL) $< $@
	chmod a+x $@

# $(PREFIX)/bin/refdb : refdb
# 	$(INSTALL) $< $@
# 	chmod a+x $@

deploytarg/nbfake : utils/nbfake
	$(INSTALL) $< $@
	chmod a+x $@

deploytarg/nbfind : utils/nbfind
	$(INSTALL) $< $@
	chmod a+x $@


# install dashboard as dboard so wrapper script can be called dashboard
$(PREFIX)/bin/.$(ARCHSTR)/dboard : dboard $(FILES)
	utils/mk_wrapper $(PREFIX) dboard $(PREFIX)/bin/dashboard
	chmod a+x $(PREFIX)/bin/dashboard
	$(INSTALL) dboard $(PREFIX)/bin/.$(ARCHSTR)/dboard

install : $(PREFIX)/bin/.$(ARCHSTR) $(PREFIX)/bin/.$(ARCHSTR)/mtest $(PREFIX)/bin/megatest \
          $(PREFIX)/bin/.$(ARCHSTR)/dboard $(PREFIX)/bin/dashboard $(HELPERS) $(PREFIX)/bin/nbfake \
	  $(PREFIX)/bin/nbfind $(PREFIX)/bin/loadrunner $(PREFIX)/bin/mt_xterm \
          $(PREFIX)/bin/newdashboard $(PREFIX)/bin/mdboard

$(PREFIX)/bin/.$(ARCHSTR) : 
	mkdir -p $(PREFIX)/bin/.$(ARCHSTR)

test: tests/tests.scm
	cd tests;csi -I .. -b -n tests.scm

	csc -deploy $(CSCOPTS) $(OFILES) megatest.scm -o deploytarg
	mv deploytarg/deploytarg deploytarg/mtest

deploytarg/dboard :  $(OFILES) $(GOFILES) dashboard.scm deploytarg/apropos.so
	csc -deploy $(OFILES) $(GOFILES) dashboard.scm -o deploytarg
	mv deploytarg/deploytarg deploytarg/dboard

# DATASHAREO=configf.o common.o process.o tree.o dcommon.o margs.o launch.o gutils.o db.o synchash.o server.o \
#            megatest-version.o tdb.o ods.o mt.o keys.o
datashare-testing/sd : datashare.scm $(OFILES)
	csc datashare.scm $(OFILES) -o datashare-testing/sd

sd : datashare-testing/sd
	mkdir -p /tmp/$(USER)/datashare/disk1 /tmp/$(USER)/basepath

xterm : sd
	(export BASEPATH=/tmp/$(USER)/basepath ; export PATH="$(PWD)/datashare-testing:$(PATH)" ; xterm &)

#  "(define (toplevel-command . a) #f)"
readline-fix.scm :
	if egrep 'version.*3.0' $(shell dirname $(shell dirname $(shell which csi)))/lib/chicken/7/readline.setup-info;then \
           echo "(use-legacy-bindings)" > readline-fix.scm; \
	else \
	   echo "" > readline-fix.scm;\
	fi

======================================================================
New way of launching needed to accomodate different target hosttypes
for items
======================================================================

[flavors]
general ssh #{getbgesthost general}
nbgeneral nbjob run JOBCOMMAND -log $MT_LINKTREE/$MT_TARGET/$MT_RUNNAME.$MT_TESTNAME-$MT_ITEM_PATH.lgo

[hosts]
general cubian xena

[launchers]
envsetup general
xor/%/n 4C16G
% nbgeneral

[jobtools]
launcher internal



======================================================================
Try writing to in-memory db and every 2-5 seconds syncing to megatest.db
======================================================================

First, how much time will it take to write back the changes:

1. Get the run table

(define (get-all db)(let ((res '()))(for-each-row (lambda (a . b)(set! res (cons (apply vector a b) res))) db "SELECT * FROM tests;") res))
(define tdata (let ((start (current-milliseconds))(res (get-all *db*)))(print (- (current-milliseconds) start))res))

Result ranges from 34ms to 89ms but mostly around 40ms for 623 records on moosefs

Projecting to 15000 records:

  Slow   2 seconds to read all
  Median 1 second to read all

This seems like it would work with an update period of 2-5 seconds

TODO
----

1. open-db opens in-memory db and megatest.db, put handles in *memdb* and *db*, *memdb* is < run-id dbh >
2. Server is part of runtests 
   a. server start cycle - adapt to per run-id
      i. states; starting, started, stopping, stopped
   b. turn off write coalesing
3. Calls to -runtests, -remove-runs etc. 
   a. Might talk to running server if run specific
   b. Can talk to megatest.db but not a generally good idea
   c. Can start a runserver 
4. Dashboard is fine except for writes?

======================================================================
Routines to convert for runs.scm

cdb:remote-run db:register-run

cdb:delete-tests-in-state *runremote*
cdb:get-test-info-by-id *runremote*
cdb:remote-run db:delete-old-deleted-test-records
cdb:remote-run db:delete-run
cdb:remote-run db:delete-test-records
cdb:remote-run db:delete-tests-for-run
cdb:remote-run db:find-and-mark-incomplete
cdb:remote-run db:get-count-tests-running
cdb:remote-run db:get-count-tests-running-in-jobgroup
cdb:remote-run db:get-keys
cdb:remote-run db:get-run-info
cdb:remote-run db:get-run-key-val
cdb:remote-run db:get-run-name-from-id
cdb:remote-run db:get-steps-for-test
cdb:remote-run db:get-test-id-cached
cdb:remote-run db:get-tests-for-runs-mindata
cdb:remote-run db:lock/unlock-run
cdb:remote-run db:set-sync
cdb:remote-run db:set-tests-state-status
cdb:remote-run db:set-var
cdb:remote-run db:testmeta-add-record
cdb:remote-run db:testmeta-get-record
cdb:remote-run db:testmeta-update-field
cdb:remote-run db:update-run-event_time
cdb:remote-run instead
cdb:remote-run server:start
cdb:remote-run test:get-matching-previous-test-run-records
cdb:tests-register-test *runremote*
(define (runs:shrink-can-run-more-tests-count) ;; the db is a dummy var so we can use cdb:remote-run

======================================================================

[87cbe68f31] 
[be405e8e2e]

# FROM andyjpg on #chicken

(let ((original-exit (exit-handler)))

TODO
====

Migration to inmem db plus per run db
-------------------------------------

. Re-work the dbstruct data structure?
.. Move main.db to global?
.. [ run-id.db inmemdb last-mod last-read last-sync inuse ]
. Re-work all queries to use run-id to dereference server
. Open main.db directly in calls to -runtests etc. No need to talk remote?

;;======================================================================
;; Copyright 2006-2013, Matthew Welland.
;; 
;;  This program is made available under the GNU GPL version 2.0 or
;;  greater. See the accompanying file COPYING for details.
;; 
;;  This program is distributed WITHOUT ANY WARRANTY; without even the
;;  implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
;;  PURPOSE.
;;======================================================================

(declare (unit api))
(declare (uses rmt))
(declare (uses db))
(declare (uses tasks))

;; allow these queries through without starting a server
;;
(define api:read-only-queries
  '(get-key-val-pairs
    get-keys
    test-toplevel-num-items
    get-test-info-by-id
    test-get-rundir-from-test-id
    get-count-tests-running-for-testname
    get-count-tests-running
    get-count-tests-running-in-jobgroup
    get-previous-test-run-record
    get-matching-previous-test-run-records
    test-get-logfile-info
    test-get-records-for-index-file
    get-testinfo-state-status
    test-get-top-process-pid
    test-get-paths-matching-keynames-target-new
    get-prereqs-not-met
    get-count-tests-running-for-run-id
    get-run-info
    get-run-status
    register-run
    get-tests-for-run
    get-test-id
    get-tests-for-runs-mindata
    get-run-name-from-id
    get-runs
    get-num-runs
    get-all-run-ids
    get-prev-run-ids
    get-run-ids-matching-target
    get-runs-by-patt
    get-steps-data
    get-steps-for-test
    login
    testmeta-get-record
    have-incompletes?
    synchash-get
    ))

(define api:write-queries
  '(
    ;; SERVERS
    start-server
    kill-server

    ;; TESTS
    test-set-state-status-by-id
    delete-test-records
    delete-old-deleted-test-records
    test-set-status-state
    test-set-top-process-pid
    roll-up-pass-fail-counts
    update-pass-fail-counts
    top-test-set-per-pf-counts ;; (db:top-test-set-per-pf-counts (db:get-db *db* 5) 5 "runfirst")

    ;; RUNS
    register-run
    set-tests-state-status
    delete-run
    lock/unlock-run
    update-run-event_time
    mark-incomplete

    ;; STEPS
    teststep-set-status!

    ;; TEST DATA
    test-data-rollup
    csv->test-data

    ;; MISC
    sync-inmem->db

    ;; TESTMETA
    testmeta-add-record
    testmeta-update-field

    ;; TASKS
    tasks-add
    tasks-set-state-given-param-key
    ))

;; These are called by the server on recipt of /api calls
;;    - keep it simple, only return the actual result of the call, i.e. no meta info here
;;
;;    - returns #( flag result )
;;
(define (api:execute-requests dbstruct dat)
  (handle-exceptions
   exn
   (let ((call-chain (get-call-chain)))
     (print-call-chain (current-error-port))
     (debug:print 0 " message: " ((condition-property-accessor 'exn 'message) exn))       
     (vector #f (vector exn call-chain dat))) ;; return some stuff for debug if an exception happens
   (if (not (vector? dat))                    ;; it is an error to not receive a vector
       (vector #f #f "remote must be called with a vector")       
       (vector                                   ;; return a vector + the returned data structure
	#t 
	(let ((cmd    (vector-ref dat 0))
	      (params (vector-ref dat 1)))
	  (case (if (symbol? cmd)
		    cmd
		    (string->symbol cmd))

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

	    ;; SERVERS
	    ((start-server)                    (apply server:kind-run params))
	    ((kill-server)                     (set! *server-run* #f))

	    ;; TESTS
	    ((test-set-state-status-by-id)     (apply db:test-set-state-status-by-id dbstruct params))
	    ((delete-test-records)             (apply db:delete-test-records dbstruct params))
	    ((delete-old-deleted-test-records) (apply db:delete-old-deleted-test-records dbstruct params))
	    ((test-set-status-state)           (apply db:test-set-status-state dbstruct params))
	    ((test-set-top-process-pid)        (apply db:test-set-top-process-pid dbstruct params))
	    ((roll-up-pass-fail-counts)        (apply db:roll-up-pass-fail-counts dbstruct params))
	    ((update-pass-fail-counts)         (apply db:general-call dbstruct 'update-pass-fail-counts params))
	    ((top-test-set-per-pf-counts)      (apply db:top-test-set-per-pf-counts dbstruct params))
	    ((test-set-archive-block-id)       (apply db:test-set-archive-block-id dbstruct params))

	    ;; RUNS
	    ((register-run)                 (apply db:register-run dbstruct params))
	    ((set-tests-state-status)       (apply db:set-tests-state-status dbstruct params))
	    ((delete-run)                   (apply db:delete-run dbstruct params))
	    ((lock/unlock-run)              (apply db:lock/unlock-run dbstruct params))
	    ((update-run-event_time)        (apply db:update-run-event_time dbstruct params))

	    ;; STEPS
	    ((teststep-set-status!)         (apply db:teststep-set-status! dbstruct params))

	    ;; TEST DATA
	    ((test-data-rollup)             (apply db:test-data-rollup dbstruct params))
	    ((csv->test-data)               (apply db:csv->test-data dbstruct params))

	    ;; MISC
	    ((sync-inmem->db)               (let ((run-id (car params)))
					      (db:sync-touched dbstruct run-id force-sync: #t)))
	    ((mark-incomplete)              (apply db:find-and-mark-incomplete dbstruct params))

	    ;; TESTMETA
	    ((testmeta-add-record)       (apply db:testmeta-add-record dbstruct params))
	    ((testmeta-update-field)     (apply db:testmeta-update-field dbstruct params))

	    ;; TASKS
	    ((tasks-add)                 (apply tasks:add dbstruct params))   
	    ((tasks-set-state-given-param-key) (apply tasks:set-state-given-param-key dbstruct params))

	    ;; ARCHIVES
	    ;; ((archive-get-allocations)   
	    ((archive-register-disk)     (apply db:archive-register-disk dbstruct params))
	    ((archive-register-block-name)(apply db:archive-register-block-name dbstruct params))
	    ((archive-allocate-testsuite/area-to-block)(apply db:archive-allocate-testsuite/area-to-block dbstruct block-id testsuite-name areakey))

	    ;;======================================================================
	    ;; READ ONLY QUERIES
	    ;;======================================================================

	    ;; KEYS
	    ((get-key-val-pairs)               (apply db:get-key-val-pairs dbstruct params))
	    ((get-keys)                        (db:get-keys dbstruct))
	    ((get-key-vals)                    (apply db:get-key-vals dbstruct params))
	    ((get-targets)                     (db:get-targets  dbstruct))

	    ;; ARCHIVES
	    ((test-get-archive-block-info)     (apply db:test-get-archive-block-info dbstruct params))
	    
	    ;; TESTS
	    ((test-toplevel-num-items)         (apply db:test-toplevel-num-items dbstruct params))
	    ((get-test-info-by-id)	       (apply db:get-test-info-by-id dbstruct params))
	    ((test-get-rundir-from-test-id)    (apply db:test-get-rundir-from-test-id dbstruct params))
	    ((get-count-tests-running-for-testname) (apply db:get-count-tests-running-for-testname dbstruct params))
	    ((get-count-tests-running)         (apply db:get-count-tests-running dbstruct params))
	    ((get-count-tests-running-in-jobgroup) (apply db:get-count-tests-running-in-jobgroup dbstruct params))
	    ;; ((delete-test-step-records)        (apply db:delete-test-step-records dbstruct params))
	    ((get-previous-test-run-record)    (apply db:get-previous-test-run-record dbstruct params))
	    ((get-matching-previous-test-run-records)(apply db:get-matching-previous-test-run-records dbstruct params))
	    ((test-get-logfile-info)           (apply db:test-get-logfile-info dbstruct params))
	    ((test-get-records-for-index-file)  (apply db:test-get-records-for-index-file dbstruct params))
	    ((get-testinfo-state-status)       (apply db:get-testinfo-state-status dbstruct params))
	    ((test-get-top-process-pid)        (apply db:test-get-top-process-pid dbstruct params))
	    ((test-get-paths-matching-keynames-target-new) (apply db:test-get-paths-matching-keynames-target-new dbstruct params))
	    ((get-prereqs-not-met)             (apply db:get-prereqs-not-met dbstruct params))
	    ((get-count-tests-running-for-run-id) (apply db:get-count-tests-running-for-run-id dbstruct params))
	    ((synchash-get)                    (apply synchash:server-get dbstruct params))

	    ;; RUNS
	    ((get-run-info)                 (apply db:get-run-info dbstruct params))
	    ((get-run-status)               (apply db:get-run-status dbstruct params))
	    ((set-run-status)               (apply db:set-run-status dbstruct params))
	    ((get-tests-for-run)            (apply db:get-tests-for-run dbstruct params))
	    ((get-test-id)                  (apply db:get-test-id dbstruct params))
	    ((get-tests-for-run-mindata)    (apply db:get-tests-for-run-mindata dbstruct params))
	    ((get-runs)                     (apply db:get-runs dbstruct params))
	    ((get-num-runs)                 (apply db:get-num-runs dbstruct params))
	    ((get-all-run-ids)              (db:get-all-run-ids dbstruct))
	    ((get-prev-run-ids)             (apply db:get-prev-run-ids dbstruct params))
	    ((get-run-ids-matching-target)  (apply db:get-run-ids-matching-target dbstruct params))
	    ((get-runs-by-patt)             (apply db:get-runs-by-patt dbstruct params))
	    ((get-run-name-from-id)         (apply db:get-run-name-from-id dbstruct params))

	    ;; STEPS
	    ((get-steps-data)               (apply db:get-steps-data dbstruct params))
	    ((get-steps-for-test)           (apply db:get-steps-for-test dbstruct params))

	    ;; MISC
	    ((have-incompletes?)            (apply db:have-incompletes? dbstruct params))
	    ((login)                        (apply db:login dbstruct params))
	    ((general-call)                 (let ((stmtname   (car params))
						  (run-id     (cadr params))
						  (realparams (cddr params)))
					      (db:with-db dbstruct run-id #t ;; these are all for modifying the db
							  (lambda (db)
							    (db:general-call db stmtname realparams)))))
	    ((sdb-qry)                      (apply sdb:qry params))
	    ((ping)                         (current-process-id))

	    ;; TESTMETA
	    ((testmeta-get-record)       (apply db:testmeta-get-record dbstruct params))

	    ;; TASKS 
	    ((find-task-queue-records)   (apply tasks:find-task-queue-records dbstruct params))))))))


;; http-server  send-response
;;                 api:process-request
;;                    db:*
;;
;; NB// Runs on the server as part of the server loop
;;
(define (api:process-request dbstruct $) ;; the $ is the request vars proc
  (let* ((cmd     ($ 'cmd))
	 (paramsj ($ 'params))
	 (params  (db:string->obj paramsj transport: 'http)) ;; (rmt:json-str->dat paramsj))
	 (resdat  (api:execute-requests dbstruct (vector cmd params))) ;; #( flag result )
	 (res     (vector-ref resdat 1)))

    ;; This can be here but needs controls to ensure it doesn't run more than every 4 seconds
    ;; (rmt:dat->json-str
    ;;  (if (or (string? res)
    ;;          (list?   res)
    ;;          (number? res)
    ;;          (boolean? res))
    ;;      res 
    ;;      (list "ERROR, not string, list, number or boolean" 1 cmd params res)))))
    (db:obj->string res transport: 'http)))

;; Copyright 2006-2014, Matthew Welland.
;; 
;;  This program is made available under the GNU GPL version 2.0 or
;;  greater. See the accompanying file COPYING for details.
;; 
;;  This program is distributed WITHOUT ANY WARRANTY; without even the
;;  implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
;;  PURPOSE.

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

(use sqlite3 srfi-1 posix regex regex-case srfi-69 dot-locking format md5 message-digest srfi-18)
(import (prefix sqlite3 sqlite3:))

(declare (unit archive))
(declare (uses db))
(declare (uses common))

(include "common_records.scm")
(include "db_records.scm")

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

;; NOT CURRENTLY USED
;;
(define (archive:main linktree target runname testname itempath options)
  (let ((testdir (conc linktree "/" target "/" runname "/" testname "/" itempatt))
	(flavor  'plain) ;; type of machine to run jobs on
	(maxload 1.5)   ;; max allowed load for this work
	(adisks  (archive:get-archive-disks)))
    ;; get testdir size
    ;;   - hand off du to job mgr
    (if (and (file-exists? testdir)
	     (file-is-writable? testdir))
	(let* ((dused  (jobrunner:run-job 
			flavor  ;; machine type
			maxload ;; max allowed load
			'()     ;; prevars - environment vars to set for the job
			common:get-disk-space-used  ;; if a proc call it, if a string it is a unix command
			(list testdir)))
	       (apath  (archive:get-archive testname itempath dused)))
	  (jobrunner:run-job
	   flavor
	   maxload
	   '()
	   archive:run-bup
	   (list testdir apath))))))
	  
;; Get archive disks from megatest.config
;;
(define (archive:get-archive-disks)
  (let ((section (configf:get-section *configdat* "archive-disks")))
    (if section
	section
	'())))

;; look for the best candidate archive area, else create new 
;; area
;;
(define (archive:get-archive testname itempath dused)
  ;; look up in archive_allocations if there is a pre-used archive
  ;; with adequate diskspace
  ;;
  (let* ((existing-blocks (rmt:archive-get-allocations testname itempath dused))
	 (candidate-disks (map (lambda (block)
				 (list
				  (vector-ref block 1)   ;; archive-area-name
				  (vector-ref block 2))) ;; disk-path
			       existing-blocks)))
    (or (common:get-disk-with-most-free-space candidate-disks dused)
	(archive:allocate-new-archive-block testname itempath))))

;; allocate a new archive area
;;
(define (archive:allocate-new-archive-block run-area-home testsuite-name dneeded)
  (let* ((adisks    (archive:get-archive-disks))
	 (best-disk (common:get-disk-with-most-free-space adisks dneeded)))
    (if best-disk
	(let* ((bdisk-name    (car best-disk))
	       (bdisk-path    (cdr best-disk))
	       (area-key      (substring (message-digest-string (md5-primitive) run-area-home) 0 5))
	       (bdisk-id      (rmt:archive-register-disk bdisk-name bdisk-path (get-df bdisk-path)))
	       (archive-name  (let ((sec (current-seconds)))
				(conc (time->string (seconds->local-time sec) "%Y")
				      "_q" (seconds->quarter sec) "/"
				      testsuite-name "_" area-key)))
	       (archive-path  (conc bdisk-path "/" archive-name))
	       (block-id      (rmt:archive-register-block-name bdisk-id archive-path)))
	  ;;   (allocation-id (rmt:archive-allocate-testsuite/area-to-block block-id testsuite-name area-key)))
	  (if block-id ;; (and block-id allocation-id)
	      (cons block-id archive-path)
	      #f))
	#f)))

;; archive - run bup
;;
;; 1. create the bup dir if not exists
;; 2. start the du of each directory
;; 3. gen index
;; 4. save
;;
(define (archive:run-bup archive-command run-id run-name tests rp-mutex bup-mutex)
  ;; move the getting of archive space down into the below block so that a single run can 
  ;; allocate as needed should a disk fill up
  ;;
  (let* ((min-space    (string->number (or (configf:lookup *configdat* "archive" "minspace") "1000")))
	 (archive-info (archive:allocate-new-archive-block *toppath* (common:get-testsuite-name) min-space))
	 (archive-dir  (if archive-info (cdr archive-info) #f))
	 (archive-id   (if archive-info (car archive-info) -1))
	 (disk-groups  (make-hash-table))
	 (test-groups  (make-hash-table)) ;; these two (disk and test groups) could be combined nicely
	 (bup-exe      (or (configf:lookup *configdat* "archive" "bup") "bup"))
	 (compress     (or (configf:lookup *configdat* "archive" "compress") "9"))
	 (linktree     (configf:lookup *configdat* "setup" "linktree")))

    (if (not archive-dir) ;; no archive disk found, this is fatal
	(begin
	  (debug:print 0 "FATAL: No archive disks found. Please add disks with at least " min-space " MB space to the [archive-disks] section of megatest.config")
	  (debug:print 0 "       use [archive] minspace to specify minimum available space")
	  (debug:print 0 "   disks: " (string-intersperse (map cadr (archive:get-archive-disks)) "\n         "))
	  (exit 1))
	(debug:print-info 0 "Using path " archive-dir " for archiving"))

    ;; from the test info bin the path to the test by stem
    ;;
    (for-each
     (lambda (test-dat)
       (let* ((item-path         (db:test-get-item-path test-dat))
	      (test-name         (db:test-get-testname  test-dat))
	      (test-id           (db:test-get-id        test-dat))
	      (run-id            (db:test-get-run_id    test-dat))
	      (target            (string-intersperse (map cadr (rmt:get-key-val-pairs run-id)) "/"))
	      
	      (toplevel/children (and (db:test-get-is-toplevel test-dat)
				      (> (rmt:test-toplevel-num-items run-id test-name) 0)))
	      (test-partial-path (conc target "/" run-name "/" (db:test-make-full-name test-name item-path)))
	      ;; note the trailing slash to get the dir inspite of it being a link
	      (test-path         (conc linktree "/" test-partial-path))
	      (mutex-lock! rp-mutex)
	      (test-physical-path (if (file-exists? test-path) 
				      (common:real-path test-path)
				      #f))
	      (mutex-unlock! rp-mutex)
	      (partial-path-index (if test-physical-path (substring-index test-partial-path test-physical-path) #f))
	      (test-base         (if (and partial-path-index 
					  test-physical-path )
				     (substring test-physical-path
						0
						partial-path-index)
				     #f)))
	 
 	 (cond
	  (toplevel/children
	   (debug:print 0 "WARNING: cannot archive " test-name " with id " test-id " as it is a toplevel test with children"))
	  ((not (file-exists? test-path))
	   (debug:print 0 "WARNING: Cannot archive " test-name "/" item-path " as path " test-path " does not exist"))
	  (else
	   (debug:print 0
			"From test-dat=" test-dat " derived the following:\n"
			"test-partial-path  = " test-partial-path "\n"
			"test-path          = " test-path "\n"
			"test-physical-path = " test-physical-path "\n"
			"partial-path-index = " partial-path-index "\n"
			"test-base          = " test-base)
	   (hash-table-set! disk-groups test-base (cons test-physical-path (hash-table-ref/default disk-groups test-base '())))
	   (hash-table-set! test-groups test-base (cons test-dat (hash-table-ref/default test-groups test-base '())))
	   test-path))))
     tests)
    ;; for each disk-group
    (for-each 
     (lambda (disk-group)
       (debug:print 0 "Processing disk-group " disk-group)
       (let* ((test-paths (hash-table-ref disk-groups disk-group))
	      ;; ((string-intersperse (map cadr (rmt:get-key-val-pairs 1)) "-")
	      (bup-init-params  (list "-d" archive-dir "init"))
	      (bup-index-params (append (list "-d" archive-dir "index") test-paths))
	      (bup-save-params  (append (list "-d" archive-dir "save" ;; (conc "--strip-path=" linktree)
					      (conc "-" compress) ;; or (conc "--compress=" compress)
					      "-n" (conc (common:get-testsuite-name) "-" run-id)
					      (conc "--strip-path=" disk-group))
					test-paths))
	      (print-prefix      #f)) ;; "Running: ")) ;; change to #f to turn off printing
	 (if (not (file-exists? archive-dir))
	     (create-directory archive-dir #t))
	 (if (not (file-exists? (conc archive-dir "/HEAD")))
	     (begin
	       ;; replace this with jobrunner stuff enventually
	       (debug:print-info 0 "Init bup in " archive-dir)
	       ;; (mutex-lock! bup-mutex)
	       (run-n-wait bup-exe params: bup-init-params print-cmd: print-prefix)
	       ;; (mutex-unlock! bup-mutex)
	       ))
	 (debug:print-info 0 "Indexing data to be archived")
	 ;; (mutex-lock! bup-mutex)
	 (run-n-wait bup-exe params: bup-index-params print-cmd: print-prefix)
	 (debug:print-info 0 "Archiving data with bup")
	 (run-n-wait bup-exe params: bup-save-params print-cmd: print-prefix)
	 ;; (mutex-unlock! bup-mutex)
	 (for-each
	  (lambda (test-dat)
	    (let ((test-id           (db:test-get-id        test-dat))
		  (run-id            (db:test-get-run_id    test-dat)))
	      (rmt:test-set-archive-block-id run-id test-id archive-id)
	      (if (member archive-command '("save-remove"))
		  (runs:remove-test-directory test-dat 'archive-remove))))
	  (hash-table-ref test-groups disk-group))))
     (hash-table-keys disk-groups))
    #t))

(define (archive:bup-restore archive-command run-id run-name tests rp-mutex bup-mutex)  ;; move the getting of archive space down into the below block so that a single run can 
  ;; allocate as needed should a disk fill up
  ;;
  (let* ((bup-exe      (or (configf:lookup *configdat* "archive" "bup") "bup"))
	 (linktree     (configf:lookup *configdat* "setup" "linktree")))

    ;; from the test info bin the path to the test by stem
    ;;
    (for-each
     (lambda (test-dat)
       ;; When restoring test-dat will initially contain an old and invalid path to the test
       (let* ((best-disk         (get-best-disk *configdat* #f)) ;; BUG: get the testconfig and use it here. Otherwise data pulled out of archive could end up on the wrong kind of disk.
	      (item-path         (db:test-get-item-path test-dat))
	      (test-name         (db:test-get-testname  test-dat))
	      (test-id           (db:test-get-id        test-dat))
	      (run-id            (db:test-get-run_id    test-dat))
	      (keyvals           (rmt:get-key-val-pairs run-id))
	      (target            (string-intersperse (map cadr keyvals) "/"))
	      
	      (toplevel/children (and (db:test-get-is-toplevel test-dat)
				      (> (rmt:test-toplevel-num-items run-id test-name) 0)))
	      (test-partial-path (conc target "/" run-name "/" (db:test-make-full-name test-name item-path)))
	      ;; note the trailing slash to get the dir inspite of it being a link
	      (test-path         (conc linktree "/" test-partial-path))
	      ;; if the old path was not deleted then prev-test-physical-path will end up pointing to a real directory
	      (mutex-lock! rp-mutex)
	      (prev-test-physical-path (if (file-exists? test-path)
					   ;; (read-symbolic-link test-path #t)
					   (common:real-path test-path)
					   #f))
	      (mutex-unlock! rp-mutex)
	      (new-test-physical-path  (conc best-disk "/" test-partial-path))
	      (archive-block-id        (db:test-get-archived test-dat))
	      (archive-block-info      (rmt:test-get-archive-block-info archive-block-id))
	      (archive-path            (if (vector? archive-block-info)
					   (vector-ref archive-block-info 2) ;; look in db.scm for test-get-archive-block-info for the vector record info
					   #f)) ;; no archive found?
	      (archive-internal-path   (conc (common:get-testsuite-name) "-" run-id "/latest/" test-partial-path)))
	 
	 ;; some sanity checks, move an existing path out of the way - iif it is not a toplevel with children
	 ;;
	 (if (and (not toplevel/children)  ;; special handling needed for toplevel with children
		  prev-test-physical-path
		  (file-exists? prev-test-physical-path)) ;; what to do? abort or clean up or link it in?
	     (let* ((base (pathname-directory prev-test-physical-path))
		    (dirn (pathname-file      prev-test-physical-path))
		    (newn (conc base "/." dirn)))
	       (debug:print 0 "ERROR: the old directory " prev-test-physical-path ", still exists! Moving it to " newn)
	       (rename-file prev-test-physical-path newn)))

	 (if (and archive-path ;; no point in proceeding if there is no actual archive
		  (not toplevel/children))
	     (begin
	       ;; CREATE WORK AREA
	       ;; test-src-path == #f     ==> don't copy in data from tests directory
	       ;; itemdat       == string ==> use directly
	       (create-work-area run-id run-name keyvals test-id #f best-disk test-name item-path) ;; #!key (remtries 2))

	       ;; 1. Get the block id from the test info
	       ;; 2. Get the block data given the block id
	       ;; 3. Construct the paths etc. for the following command:
	       ;; 
	       ;; bup -d /tmp/matt/adisk1/2015_q1/fullrun_e1a40/ restore -C /tmp/seeme fullrun-30/latest/ubuntu/nfs/none/w02.1.20.54_b/

	       ;; DO BUP RESTORE
	       (let* ((new-test-dat        (rmt:get-test-info-by-id run-id test-id))
		      (new-test-path       (if (vector? new-test-dat )
					       (db:test-get-rundir new-test-dat)
					       (begin
						 (debug:print 0 "ERROR: unable to get data for run-id=" run-id ", test-id=" test-id)
						 (exit 1))))
		      ;; new-test-path won't work - must use best-disk instead? Nope, new-test-path but tack on /..
		      (bup-restore-params  (list "-d" archive-path "restore" "-C" (conc new-test-path "/..") archive-internal-path)))
		 (debug:print-info 0 "Restoring archived data to " new-test-physical-path " from archive in " archive-path " ... " archive-internal-path)
		 ;; (mutex-lock! bup-mutex)
		 (run-n-wait bup-exe params: bup-restore-params print-cmd: #f)
		 ;; (mutex-unlock! bup-mutex)
		 (mt:test-set-state-status-by-id run-id test-id "COMPLETED" #f #f)))
	     (debug:print 0 "ERROR: No archive path in the record for run-id=" run-id " test-id=" test-id))))
     (filter vector? tests))))
	 

RUN=default.scm

all : batchsim
	./batchsim $(RUN)

batchsim : batchsim.scm
	csc batchsim.scm

(use ezxdisp srfi-18)

(define *ezx* (ezx-init 650 650 "Batch simulator"))
(require-library ezxgui)
(define *green*  (make-ezx-color 0 1 0)) 
(define *black*  (make-ezx-color 0 0 0))
(define *grey*   (make-ezx-color 0.1 0.1 0.1))
(define *blue*   (make-ezx-color 0 0 1)) 
(define *cyan*   (make-ezx-color 0 1 1))
(define *green*  (make-ezx-color 0 1 0))
(define *purple* (make-ezx-color 1 0 1))
(define *red*    (make-ezx-color 1 0 0))
(define *white*  (make-ezx-color 1 1 1))
(define *yellow* (make-ezx-color 1 1 0))

(define *user-colors-palette*
  (list 
   *green*
   *blue*
   *cyan*
   *purple*
   *red*
   *yellow*
   *black*))

(define *dark-green* (get-color "dark-green"))
(define *brown*      (get-color "brown"))

(ezx-select-layer *ezx* 1)
(ezx-wipe-layer *ezx* 1)

;; (ezx-str-2d *ezx* 30 30 "Hello" *white*)
;; (ezx-fillrect-2d *ezx* 100 100 120 120 *brown*)
(ezx-redraw *ezx*)

(define *last-draw* (current-milliseconds))
(define *draw-delta* 40) ;; milliseconds between drawing

(define (wait-for-next-draw-time)
  (let* ((cm    (current-milliseconds))
	 (delta (- *draw-delta* (- cm *last-draw*))))
    (if (> delta 0)
	(thread-sleep! (/ delta 1000)))
    (set! *last-draw* (current-milliseconds))))

(include "events.scm")

;; System spec (to be moved into loaded file)
;;
;;                           x  y  w gap  x-min x-max
(define *cpu-grid* (vector  500 50 15  2  500   600))
(define (make-cpu:grid)(make-vector 6))
(define *queues* (make-hash-table)) ;; name -> (list (list user duration num-cpus num-gigs) ... )
(define *cpus* (make-hash-table)) ;; cpu-name => (vector user job-len num-cpu mem x-loc y-loc)
(define *obj-locations* (make-hash-table)) ;; name -> (x y layer)
(define *queue-spec*
  (vector
   80  ;; start-x
   300 ;; start-y
   300 ;; delta-y how far to next queue
   15  ;; height
   400 ;; length
   ))
(define *use-log* #f)
(define *job-log-scale* 10)

;;======================================================================
;; CPU
;;======================================================================

(define-record cpu name num-cores mem job x y)

;;======================================================================
;; CPU Pool
;;======================================================================

(define-record pool name x y w h gap boxw cpus delta nrows ncols cpunum)

(define (new-pool name x y nrows ncols gap boxw)
  (let* ((delta (+ gap boxw))
	 ;; (nrows (quotient h (+ gap delta)))
	 ;; (ncols (quotient w (+ gap delta)))
	 (w     (+ gap (* nrows delta)))
	 (h     (+ gap (* ncols delta)))
	 (cpus  (make-vector (* nrows ncols) #f))
	 (npool (make-pool name x y w h gap boxw cpus delta nrows ncols 0)))
    npool))

(define (pool:add-cpu pool name num-cores mem)
  (let* ((cpu (make-cpu name num-cores mem #f #f #f)))
    (vector-set! (pool-cpus pool)(pool-cpunum pool) cpu)
    (pool-cpunum-set! pool (+ 1 (pool-cpunum pool)))
    cpu))

(define (pool:draw ezx pool)
  (let ((nrows (pool-nrows pool))
	(ncols (pool-ncols pool))
	(x     (pool-x     pool))
	(y     (pool-y     pool))
	(w     (pool-w     pool))
	(h     (pool-h     pool))
	(gap   (pool-gap   pool))
	(boxw  (pool-boxw  pool))
	(delta (pool-delta pool))
	(cpus  (pool-cpus  pool)))
    (ezx-select-layer ezx 1)
    ;(ezx-wipe-layer   ezx 1)
    ;; draw time at upper right
    (ezx-str-2d ezx x y (pool-name pool) *black*)
    (ezx-rect-2d ezx x y (+ x w)(+ y h) *black* 1)
    (let loop ((row    0)
	       (col    0)
	       (cpunum 0))
      (let* ((cpu  (vector-ref cpus cpunum))
	     (xval (+ x gap (* row delta)))
	     (yval (+ y gap (* col delta))))
	(if cpu
	    (begin
	      (cpu-x-set! cpu xval)
	      (cpu-y-set! cpu yval))
	    (vector-set! cpus cpunum (make-cpu (conc cpunum) 1 1 #f xval yval)))
	;; (print "box at " xval ", " yval)
	(ezx-rect-2d ezx xval yval (+ xval boxw) (+ yval boxw) *grey* 1)
	(if (< col (- ncols 1))
	    (loop row (+ col 1)(+ cpunum 1))
	    (if (< row (- nrows 1))
		(loop (+ row 1) 0 (+ cpunum 1))))))
    (ezx-redraw ezx)))
	       

;;======================================================================
;; Users
;;======================================================================

(define *user-colors* (make-hash-table))

(define (get-user-color user)
  (let ((color (hash-table-ref/default *user-colors* user #f)))
    (if color
	color
	(let* ((color-num (+ (length (hash-table-keys *user-colors*)) 1))
	       (color     (list-ref *user-colors-palette* color-num)))
	  (hash-table-set! *user-colors* user color)
	  color))))

;;======================================================================
;; Job Queues
;;======================================================================

;; jobs

(define (make-queue:job)(make-vector 4))
(define-inline (queue:job-get-user       vec)    (vector-ref  vec 0))
(define-inline (queue:job-get-duration   vec)    (vector-ref  vec 1))
(define-inline (queue:job-get-num-cpu    vec)    (vector-ref  vec 2))
(define-inline (queue:job-get-num-gigs   vec)    (vector-ref  vec 3))
(define-inline (queue:job-set-user!      vec val)(vector-set! vec 0 val))
(define-inline (queue:job-set-duration!  vec val)(vector-set! vec 1 val))
(define-inline (queue:job-set-num-cpu!   vec val)(vector-set! vec 2 val))
(define-inline (queue:job-set-num-gigs!  vec val)(vector-set! vec 3 val))

;; add a job to the queue
;;
(define (add-job queue-name user duration num-cpu num-gigs)
  (let* ((queue-dat (hash-table-ref/default *queues* queue-name '()))
	 (new-queue (append 
		     queue-dat
		     (list (vector user duration num-cpu num-gigs)))))
  (hash-table-set! *queues* queue-name new-queue)
  (draw-queue-jobs queue-name)))

;; peek for jobs to do in queue
;;
(define (peek-job queue-name)
  (let ((queue (hash-table-ref/default *queues* queue-name '())))
    (if (null? queue)
	#f
	(car queue))))

;; take job from queue
;;
(define (take-job queue-name)
  (let ((queue (hash-table-ref/default *queues* queue-name '())))
    (if (null? queue)
	#f
	(begin
	  (hash-table-set! *queues* queue-name (cdr queue))
	  (draw-queue-jobs queue-name)
	  (car queue)))))

;;======================================================================
;; CPUs
;;======================================================================

(define (make-cpu:dat)(make-vector 6 #f))
(define-inline (cpu:dat-get-user      vec)    (vector-ref  vec 0))
(define-inline (cpu:dat-get-job-len   vec)    (vector-ref  vec 1))
(define-inline (cpu:dat-get-num-cpu   vec)    (vector-ref  vec 2))
(define-inline (cpu:dat-get-mem       vec)    (vector-ref  vec 3))
(define-inline (cpu:dat-get-x         vec)    (vector-ref  vec 4))
(define-inline (cpu:dat-get-y         vec)    (vector-ref  vec 5))
(define-inline (cpu:dat-set-user!     vec val)(vector-set! vec 0 val))
(define-inline (cpu:dat-set-job-len!  vec val)(vector-set! vec 1 val))
(define-inline (cpu:dat-set-num-cpu!  vec val)(vector-set! vec 2 val))
(define-inline (cpu:dat-set-mem!      vec val)(vector-set! vec 3 val))
(define-inline (cpu:dat-set-x!        vec val)(vector-set! vec 4 val))
(define-inline (cpu:dat-set-y!        vec val)(vector-set! vec 5 val))

(define-inline (cpu:grid-get-x        vec)    (vector-ref  vec 0))
(define-inline (cpu:grid-get-y        vec)    (vector-ref  vec 1))
(define-inline (cpu:grid-get-w        vec)    (vector-ref  vec 2))
(define-inline (cpu:grid-get-gap      vec)    (vector-ref  vec 3))
(define-inline (cpu:grid-get-x-min    vec)    (vector-ref  vec 4))
(define-inline (cpu:grid-get-x-max    vec)    (vector-ref  vec 5))
(define-inline (cpu:grid-set-x!       vec val)(vector-set! vec 0 val))
(define-inline (cpu:grid-set-y!       vec val)(vector-set! vec 1 val))
(define-inline (cpu:grid-set-w!       vec val)(vector-set! vec 2 val))
(define-inline (cpu:grid-set-gap!     vec val)(vector-set! vec 3 val))
(define-inline (cpu:grid-set-x-min!   vec val)(vector-set! vec 4 val))
(define-inline (cpu:grid-set-x-max!   vec val)(vector-set! vec 5 val))

(define (add-cpu name num-cores mem)
  (let ((x     (cpu:grid-get-x     *cpu-grid*))
	(y     (cpu:grid-get-y     *cpu-grid*))
	(delta (+ (cpu:grid-get-w  *cpu-grid*)(cpu:grid-get-gap *cpu-grid*)))
	(x-max (cpu:grid-get-x-max *cpu-grid*)))
    (hash-table-set! *cpus* name (vector #f #f num-cores mem x y))
    (if (> x x-max)
	(begin
	  (cpu:grid-set-x! *cpu-grid* (cpu:grid-get-x-min *cpu-grid*))
	  (cpu:grid-set-y! *cpu-grid* (+ y delta)))
	(cpu:grid-set-x! *cpu-grid* (+ x delta)))))

;; draw grey box for each cpu on layer 2
;; jobs are drawn on layer 1
;;
(define (draw-cpus) ;; call once after init'ing all cpus
  (ezx-select-layer *ezx* 1)
  (ezx-wipe-layer   *ezx* 1)
  ;; draw time at upper right
  (ezx-str-2d *ezx* 20 20 (seconds->h:m:s *now*) *black*)
  (for-each
   (lambda (cpu)
     (let ((x (cpu:dat-get-x cpu))
	   (y (cpu:dat-get-y cpu))
	   (w (cpu:grid-get-w *cpu-grid*)))
       (ezx-rect-2d *ezx* x y (+ x w) (+ y w) *grey* 1)))
   (hash-table-values *cpus*))
  (ezx-redraw *ezx*))

(define (draw-jobs)
  ;; (draw-cpus)
  (ezx-select-layer *ezx* 2)
  (ezx-wipe-layer   *ezx* 2)
  (for-each
   (lambda (cpu)
     (let* ((x (cpu:dat-get-x cpu))
	    (y (cpu:dat-get-y cpu))
	    (w (cpu:grid-get-w *cpu-grid*))
	    (u (cpu:dat-get-user cpu)))
       (if u ;; job running if not #f
	   (let ((color (get-user-color u)))
	     (ezx-fillrect-2d *ezx* (+ x 2)(+ 2 y)(+ x 9) (+ y 9) color)))))
   (hash-table-values *cpus*))
  (ezx-redraw *ezx*))

(define (end-job cpu-name user)
  (let ((cpu (hash-table-ref/default *cpus* cpu-name #f)))
    (if cpu
	(let ((curr-user (cpu:dat-get-user cpu))) ;; if it is a user name then job is not done - error
	  (if (or (not curr-user)
		  (not (equal? curr-user user)))
	      (print "ERROR: cpu " cpu-name " not running job for " user "!")
	      (begin
		(cpu:dat-set-user!    cpu #f)
		(cpu:dat-set-job-len! cpu #f)
		(draw-jobs)))) ;; hash-table-set! *cpus* cpu-name (make-cpu:dat))))
	(print "ERROR: no cpu " cpu-name " found. Ensure it is registered before addressing it."))))

(define (run-job cpu-name job)
  (let* ((user    (queue:job-get-user     job))
	 (job-len (queue:job-get-duration job))
	 (cpu     (hash-table-ref/default *cpus* cpu-name #f)))
    (if cpu
	(let ((curr-user (cpu:dat-get-user cpu))) ;; if it is a user name then job is not done - error
	  (if curr-user
	      (begin
		(print "ERROR: cpu already busy! Adding more jobs not supported yet. " cpu-name)
		#f)
	      (begin
		(cpu:dat-set-user!    cpu user)
		(cpu:dat-set-job-len! cpu job-len)
		(draw-jobs)
		(hash-table-set! *cpus* cpu-name cpu)
		(event (+ *now* job-len) (lambda ()(end-job cpu-name user)))
		#t)))
	#f)))
    
(define (get-cpu)
  (let ((all-cpus (hash-table-keys *cpus*)))
    (if (null? all-cpus) 
	#f
	(let loop ((hed (car all-cpus))
		   (tal (cdr all-cpus)))
	  (if (cpu:dat-get-user (hash-table-ref/default *cpus* hed '(#f #f))) ;; if user is #f then cpu is available
	      (if (null? tal)
		  #f
		  (loop (car tal)(cdr tal)))
	      hed)))))
  
;;======================================================================
;; Animation
;;======================================================================

;; make-vector-record  queue spec x y delta-y height length
(define (make-queue:spec)(make-vector 5))
(define-inline (queue:spec-get-x         vec)    (vector-ref  vec 0))
(define-inline (queue:spec-get-y         vec)    (vector-ref  vec 1))
(define-inline (queue:spec-get-delta-y   vec)    (vector-ref  vec 2))
(define-inline (queue:spec-get-height    vec)    (vector-ref  vec 3))
(define-inline (queue:spec-get-length    vec)    (vector-ref  vec 4))
(define-inline (queue:spec-set-x!        vec val)(vector-set! vec 0 val))
(define-inline (queue:spec-set-y!        vec val)(vector-set! vec 1 val))
(define-inline (queue:spec-set-delta-y!  vec val)(vector-set! vec 2 val))
(define-inline (queue:spec-set-height!   vec val)(vector-set! vec 3 val))
(define-inline (queue:spec-set-length!   vec val)(vector-set! vec 4 val))

;; queues are drawn on layer 3 but boxes (jobs) are drawn on the numbered layer
;;
(define (draw-queues)
  (let* ((text-offset 3)
	 (queue-names (sort (hash-table-keys *queues*) string>=?))
	 (start-x (vector-ref *queue-spec* 0))
	 (text-x  (+ start-x text-offset))
	 (delta-y (vector-ref *queue-spec* 1))
	 (delta-x (vector-ref *queue-spec* 2))
	 (height  (vector-ref *queue-spec* 3))
	 (length  (vector-ref *queue-spec* 4))
	 (end-x   (+ start-x length)))
    (ezx-select-layer *ezx* 3)
    (ezx-wipe-layer   *ezx* 3)
    (let loop ((y       (vector-ref *queue-spec* 1))
	       (qname   (car queue-names))
	       (tail    (cdr queue-names))
	       (layer   4))
      (print "Drawing queue at x=" start-x ", y=" y)
      (ezx-fillrect-2d  *ezx* start-x y end-x (+ y height) *brown*)
      (ezx-str-2d       *ezx* text-x (- (+ y height) text-offset) qname *white*)
      (hash-table-set! *obj-locations* qname (list start-x y layer))
      (if (not (null? tail))
	  (loop (+ y height delta-y)
		(car tail)
		(cdr tail)
		(+  layer 1))))
    (ezx-redraw *ezx*)))

;; compress queue data to (vector user count) list
;;
(define (draw-queue-compress-queue-data queue-dat)
  (let loop ((hed  (car queue-dat))
	     (tal  (cdr queue-dat))
	     (curr #f) ;; (vector name count)
	     (res  '()))
    (let ((user (queue:job-get-user hed)))
      (cond
       ((not curr) ;; first time through only?
	(if (null? tal)
	    (append res (list (vector user 1)))
	    (loop (car tal)(cdr tal)(vector user 1) res)))
       ((equal? (vector-ref curr 0) user) 
	(vector-set! curr 1 (+ (vector-ref curr 1) 1))
	(if (null? tal)
	    (append res (list curr))
	    (loop (car tal)(cdr tal) curr res)))
       (else ;; names are different, add curr to queue and create new curr
	(let ((newcurr (vector user 1)))
	  (if (null? tal)
	      (append res (list newcurr))
	      (loop (car tal)(cdr tal) newcurr (append res (list curr))))))))))

;; draw jobs for a given queue
;;
(define (draw-queue-jobs queue-name)
  (let* ((queue-dat (hash-table-ref/default *queues*        queue-name #f))  ;; list of jobs in the queue
	 (obj-spec  (hash-table-ref/default *obj-locations* queue-name #f)))  ;; x, y etc. of the drawn queue
    (if obj-spec
	(let ((origin-x  (list-ref obj-spec 0))
	      (origin-y  (list-ref obj-spec 1))
	      (bar-width 10)
	      (queue-len (queue:spec-get-length *queue-spec*))
	      (layer     (list-ref obj-spec 2)))
	  (ezx-select-layer *ezx* layer)
	  (ezx-wipe-layer   *ezx* layer)
	  (if (not (null? queue-dat))
	      (let ((res (draw-queue-compress-queue-data queue-dat)))
		(if (not (null? res))
		    (let loop ((hed (car res))
			       (tal (cdr res))
			       (x2   (+ origin-x queue-len)))
		      (let* ((user (vector-ref hed 0))
			     (h    (let ((numjobs (vector-ref hed 1)))
				     (if *use-log*
					 (inexact->exact (round (log (+ 1 (* *job-log-scale* numjobs)))))
					 numjobs)))
			     (x1   (- x2  bar-width))
			     (y2   (- origin-y h)))
			;; (print "x1 " x1 ", origin-y " origin-y ", x2 " x2 ", y2 " y2)
			(ezx-fillrect-2d *ezx* x1 y2 x2 origin-y (get-user-color user))
			(if (not (null? tal))
			    (loop (car tal)(cdr tal) x1)))))
		(ezx-redraw *ezx*)))))))
	  
(let* ((args  (argv))
       (fname (if (> (length args) 1)
		  (cadr args)
 		  "default.scm")))
  (load (if (file-exists? fname) fname "default.scm")))

;; run sim for four hours
;;
(define *end-time* (* 60 50))

;; create the cpus
;;
(let loop ((count 200))
  (add-cpu (conc "cpu_" count) 1 1)
  (if (>= count 0)(loop (- count 1))))

(draw-cpus)

(define *pool1* (new-pool "generic" 100 100 100 100 2 10))
(let loop ((count 10))
  (pool:add-cpu *pool1* (conc count) 1 1)
  (if (> count 0)
      (loop (- count 1))))

(pool:draw *ezx* *pool1*)

;; init the queues
;;
(hash-table-set! *queues* "normal" '())
(hash-table-set! *queues* "quick"  '())
(draw-queues)

;; user k adds 200 jobs at time zero
;;
(event *start-time*
       (lambda ()
	 (let loop ((count 300)) ;; add 500 jobs
	   (add-job "normal" "k" 600 1 1)
	   (if (>= count 0)(loop (- count 1))))))

;; one minute in user m runs ten jobs
;;
(event (+ 600 *start-time*)
       (lambda ()
	 (let loop ((count 300)) ;; add 100 jobs
	   (add-job "normal" "m" 600 1 1)
	   (if (> count 0)(loop (- count 1))))))

;; every minute user j runs ten jobs
;;
(define *user-j-jobs* 300)
(event (+ 600 *start-time*)
       (lambda ()
	 (let f ()
	   (schedule 60)
	   (if (> *user-j-jobs* 0)
	       (begin
		 (let loop ((count 5)) ;; add 100 jobs
		   (add-job "quick" "j" 600 1 1)
		   (if (> count 0)(loop (- count 1))))
		 (set! *user-j-jobs* (- *user-j-jobs* 5))))
	   (if (and (not *done*)
		    (> *user-j-jobs* 0))
	       (f))))) ;; Megatest user running 200 jobs

;; every minute user j runs ten jobs
;;
(define *user-j-jobs* 300)
(event (+ 630 *start-time*)
       (lambda ()
	 (let f ()
	   (schedule 60)
	   (if (> *user-j-jobs* 0)
	       (begin
		 (let loop ((count 5)) ;; add 100 jobs
		   (add-job "quick" "n" 600 1 1)
		   (if (> count 0)(loop (- count 1))))
		 (set! *user-j-jobs* (- *user-j-jobs* 5))))
	   (if (and (not *done*)
		    (> *user-j-jobs* 0))
	       (f))))) ;; Megatest user running 200 jobs

;; ;;
;; (event *start-time*
;;        (lambda ()
;; 	 (let f ((count 200))
;; 	   (schedule 10)
;; 	   (add-job "normal" "t" 60 1 1)
;; 	   (if (and (not *done*)
;; 		    (>= count 0))
;; 	       (f (- count 1))))))

;; every 3 seconds check for available machines and launch a job
;;
(event *start-time*
       (lambda ()
	 (let f ()
	   (schedule 3)
	   (let ((queue-names (random-sort (hash-table-keys *queues*))))
	     (let loop ((cpu   (get-cpu))
			(count (+ (length queue-names) 4))
			(qname (car queue-names))
			(remq  (cdr queue-names)))
	       (if (and cpu
			(> count 0))
		   (begin
		     (if (peek-job qname) ;; any jobs to do in normal queue
			 (let ((job (take-job qname)))
			   (run-job cpu job)))
		     (loop (get-cpu)
			   (- count 1)
			   (if (null? remq)
			       (car queue-names)
			       (car remq))
			   (if (null? remq)
			       (cdr queue-names)
			       (cdr remq)))))))
	   (if (not *done*)(f)))))

;; screen updates
;;
(event *start-time* (lambda ()
		      (let f ()
			(schedule 60) ;; update the screen every 60 seconds of sim time
			(draw-cpus) ;; (print "Now: " *now* " queue: " (hash-table->alist *queues*))
			(wait-for-next-draw-time)
			(if (not *done*) (f)))))


;; end the simulation
;;
(event *end-time*
       (lambda () 
	 (set! *event-list* '())
	 (set! *done* #t)))

(start)
;; (exit 0)

;;======================================================================
;; Event Processing and Simulator
;;======================================================================

;; The global event list
(define *event-list* '())
(define *start-time* 0)
(define *end-time*   (* 60 60 4)) ;; four hours
(define *now*        *start-time*)
(define *done*       #f)

(define (random-sort l)
  (sort l
        (lambda (x y)
          (equal? 0 (random 2))))) 
				    
;; Each item in the event list is a list of a scheduled time and the thunk
;; (time thunk). Sort the list so that the next event is the earliest.
;;
(define event-sort
  (lambda (@a @b)
    (< (car @a)(car @b))))

(define event
  (lambda ($time $thunk)  ;; add a sort based on scheduled time here -- improve later
                          ;; to use an insert algorythm.
    (set! *event-list* (sort (cons (list $time $thunk) *event-list*) event-sort))))

(define start
  (lambda ()
    (let ((next (car *event-list*)))
      (set! *event-list* (cdr *event-list*))
      (set! *now* (car next))      
      (if (not *done*) ;; note that the second item in the list is the thunk
	  ((car (cdr next)))))))

(define pause
  (lambda ()
    (call/cc 
     (lambda (k)
       (event (lambda () (k #f)))
       (start)))))

(define schedule
  (lambda ($time)
    (call/cc 
     (lambda (k)
       (event (+ *now* $time) (lambda () (k #f)))
       (start)))))

;; (event (lambda () (let f () (pause) (display "h") (f))))

(define years->seconds
  (lambda ($yrs)
    (* $yrs 365 24 3600)))

(define weeks->seconds
  (lambda ($wks)
    (* $wks 7 24 3600)))

(define days->seconds
  (lambda ($days)
    (* $days 24 3600)))

(define months->seconds
  (lambda ($months)
    (* $months (/ 365 12) 24 3600)))

(define seconds->date
  (lambda ($seconds)
    (posix-strftime "%D" (posix-localtime (inexact->exact $seconds)))))

(define (seconds->h:m:s seconds)
  (let* ((hours   (quotient seconds 3600))
	 (rem1    (- seconds (* hours 3600)))
	 (minutes (quotient rem1 60))
	 (rem-sec (- rem1 (* minutes 60))))
    (conc hours "h " minutes "m " rem-sec "s")))

;; run sim for four hours
;;
(define *end-time* (* 60 50))

;; create the cpus
;;
(let loop ((count 200))
  (add-cpu (conc "cpu_" count) 1 1)
  (if (>= count 0)(loop (- count 1))))

;; (draw-cpus)

(define *pool1* (new-pool "generic" 20 20 12 80 2 4))
(let loop ((count 10))
  (pool:add-cpu *pool1* (conc count) 1 1)
  (if (> count 0)
      (loop (- count 1))))

(pool:draw *ezx* *pool1*)

;; ;; init the queues
;; ;;
;; (hash-table-set! *queues* "normal" '())
;; (hash-table-set! *queues* "quick"  '())
;; (draw-queues)
;; 
;; ;; user k adds 200 jobs at time zero
;; ;;
;; (event *start-time*
;;        (lambda ()
;; 	 (let loop ((count 300)) ;; add 500 jobs
;; 	   (add-job "normal" "k" 600 1 1)
;; 	   (if (>= count 0)(loop (- count 1))))))
;; 
;; ;; one minute in user m runs ten jobs
;; ;;
;; (event (+ 600 *start-time*)
;;        (lambda ()
;; 	 (let loop ((count 300)) ;; add 100 jobs
;; 	   (add-job "normal" "m" 600 1 1)
;; 	   (if (> count 0)(loop (- count 1))))))
;; 
;; ;; every minute user j runs ten jobs
;; ;;
;; (define *user-j-jobs* 300)
;; (event (+ 600 *start-time*)
;;        (lambda ()
;; 	 (let f ()
;; 	   (schedule 60)
;; 	   (if (> *user-j-jobs* 0)
;; 	       (begin
;; 		 (let loop ((count 5)) ;; add 100 jobs
;; 		   (add-job "quick" "j" 600 1 1)
;; 		   (if (> count 0)(loop (- count 1))))
;; 		 (set! *user-j-jobs* (- *user-j-jobs* 5))))
;; 	   (if (and (not *done*)
;; 		    (> *user-j-jobs* 0))
;; 	       (f))))) ;; Megatest user running 200 jobs
;; 
;; ;; every minute user j runs ten jobs
;; ;;
;; (define *user-j-jobs* 300)
;; (event (+ 630 *start-time*)
;;        (lambda ()
;; 	 (let f ()
;; 	   (schedule 60)
;; 	   (if (> *user-j-jobs* 0)
;; 	       (begin
;; 		 (let loop ((count 5)) ;; add 100 jobs
;; 		   (add-job "quick" "n" 600 1 1)
;; 		   (if (> count 0)(loop (- count 1))))
;; 		 (set! *user-j-jobs* (- *user-j-jobs* 5))))
;; 	   (if (and (not *done*)
;; 		    (> *user-j-jobs* 0))
;; 	       (f))))) ;; Megatest user running 200 jobs
;; 
;; ;; ;;
;; ;; (event *start-time*
;; ;;        (lambda ()
;; ;; 	 (let f ((count 200))
;; ;; 	   (schedule 10)
;; ;; 	   (add-job "normal" "t" 60 1 1)
;; ;; 	   (if (and (not *done*)
;; ;; 		    (>= count 0))
;; ;; 	       (f (- count 1))))))
;; 
;; ;; every 3 seconds check for available machines and launch a job
;; ;;
;; (event *start-time*
;;        (lambda ()
;; 	 (let f ()
;; 	   (schedule 3)
;; 	   (let ((queue-names (random-sort (hash-table-keys *queues*))))
;; 	     (let loop ((cpu   (get-cpu))
;; 			(count (+ (length queue-names) 4))
;; 			(qname (car queue-names))
;; 			(remq  (cdr queue-names)))
;; 	       (if (and cpu
;; 			(> count 0))
;; 		   (begin
;; 		     (if (peek-job qname) ;; any jobs to do in normal queue
;; 			 (let ((job (take-job qname)))
;; 			   (run-job cpu job)))
;; 		     (loop (get-cpu)
;; 			   (- count 1)
;; 			   (if (null? remq)
;; 			       (car queue-names)
;; 			       (car remq))
;; 			   (if (null? remq)
;; 			       (cdr queue-names)
;; 			       (cdr remq)))))))
;; 	   (if (not *done*)(f)))))
;; 
;; ;; screen updates
;; ;;
(event *start-time* (lambda ()
		      (let f ()
			(schedule 60) ;; update the screen every 60 seconds of sim time
			;; (draw-cpus) ;; (print "Now: " *now* " queue: " (hash-table->alist *queues*))
			(pool:draw *ezx* *pool1*)

			(wait-for-next-draw-time)
			(if (not *done*) (f)))))
;; 
;; 
;; ;; end the simulation
;; ;;
(event *end-time*
       (lambda () 
	 (set! *event-list* '())
	 (set! *done* #t)))
;; 
(start)
;; ;; (exit 0)
;; 

#!/bin/bash 

if [[ $SLEEPRUNNER == "" ]];then
SLEEPRUNNER=0
fi
  
echo "nbfake $@ &> /dev/null" | at now + $SLEEPRUNNER minutes &> /dev/null

;; client:get-signature
(define (client:get-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*)))





;; Not currently used! But, I think it *should* be used!!!
(define (client:logout serverdat)
  (let ((ok (and (socket? serverdat)
		 (cdb:logout serverdat *toppath* (client:get-signature)))))
    ok))

(define (client:connect iface port)
  (case (server:get-transport)
    ((rpc)  (rpc:client-connect  iface port))
    ((http) (http:client-connect iface port))
    ((zmq)  (zmq:client-connect  iface port))
    (else   (rpc:client-connect  iface port))))

(define (client:setup  run-id #!key (remaining-tries 10) (failed-connects 0))
  (case (server:get-transport)
    ((rpc) (rpc-transport:client-setup run-id)) ;;(client:setup-rpc run-id))
    ((http)(client:setup-http run-id))
    (else  (rpc-transport:client-setup run-id)))) ;; (client:setup-rpc run-id))))

;; (define (client:login-no-auto-setup server-info run-id)
;;   (case (server:get-transport)
;;     ((rpc)  (rpc:login-no-auto-client-setup server-info run-id))
;;     ((http) (rmt:login-no-auto-client-setup server-info run-id))
;;     (else   (rpc:login-no-auto-client-setup server-info run-id))))
;; 
;; (define (client:setup-rpc run-id)
;;   (debug:print 0 "INFO: client:setup remaining-tries=" remaining-tries)
;;   (if (<= remaining-tries 0)
;;       (begin
;; 	(debug:print 0 "ERROR: failed to start or connect to server for run-id " run-id)
;; 	(exit 1))
;;       (let ((host-info (hash-table-ref/default *runremote* run-id #f)))
;; 	(debug:print-info 0 "client:setup host-info=" host-info ", remaining-tries=" remaining-tries)
;; 	(if host-info
;; 	    (let* ((iface     (car  host-info))
;; 		   (port      (cadr host-info))
;; 		   (start-res (client:connect iface port))
;; 		   ;; (ping-res  (server:ping-server run-id iface port))
;; 		   (ping-res  (client:login-no-auto-setup start-res run-id)))
;; 	      (if ping-res   ;; sucessful login?
;; 		  (begin
;; 		    (hash-table-set! *runremote* run-id start-res)
;; 		    start-res)  ;; return the server info
;; 		  (if (member remaining-tries '(3 4 6))
;; 		      (begin    ;; login failed
;; 			(debug:print 25 "INFO: client:setup start-res=" start-res ", run-id=" run-id ", server-dat=" host-info)
;; 			(hash-table-delete! *runremote* run-id)
;; 			(open-run-close tasks:server-force-clean-run-record
;; 			 		tasks:open-db
;; 			 		run-id 
;; 			 		(car  host-info)
;; 			 		(cadr host-info)
;; 					" client:setup (host-info=#t)")
;; 			(thread-sleep! 5)
;; 			(client:setup run-id remaining-tries: 10)) ;; (- remaining-tries 1)))
;; 		      (begin
;; 			(debug:print 25 "INFO: client:setup failed to connect, start-res=" start-res ", run-id=" run-id ", host-info=" host-info)
;; 			(thread-sleep! 5)
;; 			(client:setup run-id remaining-tries: (- remaining-tries 1))))))
;; 	    ;; YUK: rename server-dat here
;; 	    (let* ((server-dat (open-run-close tasks:get-server tasks:open-db run-id)))
;; 	      (debug:print-info 0 "client:setup server-dat=" server-dat ", remaining-tries=" remaining-tries)
;; 	      (if server-dat
;; 		  (let* ((iface     (tasks:hostinfo-get-interface server-dat))
;; 			 (port      (tasks:hostinfo-get-port      server-dat))
;; 			 (start-res (http-transport:client-connect iface port))
;; 			 ;; (ping-res  (server:ping-server run-id iface port))
;; 			 (ping-res  (rmt:login-no-auto-client-setup start-res run-id)))
;; 		    (if start-res
;; 			(begin
;; 			  (hash-table-set! *runremote* run-id start-res)
;; 			  start-res)
;; 			(if (member remaining-tries '(2 5))
;; 			    (begin    ;; login failed
;; 			      (debug:print 25 "INFO: client:setup start-res=" start-res ", run-id=" run-id ", server-dat=" server-dat)
;; 			      (hash-table-delete! *runremote* run-id)
;; 			      (open-run-close tasks:server-force-clean-run-record
;; 					      tasks:open-db
;; 					      run-id 
;; 					      (tasks:hostinfo-get-interface server-dat)
;; 					      (tasks:hostinfo-get-port      server-dat)
;; 					      " client:setup (server-dat = #t)")
;; 			      (thread-sleep! 2)
;; 			      (server:try-running run-id)
;; 			      (thread-sleep! 10) ;; give server a little time to start up
;; 			      (client:setup run-id remaining-tries: 10)) ;; (- remaining-tries 1)))
;; 			    (begin
;; 			      (debug:print 25 "INFO: client:setup start-res=" start-res ", run-id=" run-id ", server-dat=" server-dat)
;; 			      (thread-sleep! 5)
;; 			      (client:setup run-id remaining-tries: (- remaining-tries 1))))))
;; 		  (begin    ;; no server registered
;; 		    (if (eq? remaining-tries 2)
;; 			(begin
;; 			  ;; (open-run-close tasks:server-clean-out-old-records-for-run-id tasks:open-db run-id " client:setup (server-dat=#f)")
;; 			  (client:setup run-id remaining-tries: 10))
;; 			(begin
;; 			  (thread-sleep! 2) 
;; 			  (debug:print 25 "INFO: client:setup start-res (not defined here), run-id=" run-id ", server-dat=" server-dat)
;; 			  (if (< (open-run-close tasks:num-in-available-state tasks:open-db run-id) 3)
;; 			      (begin
;; 				;; (open-run-close tasks:server-clean-out-old-records-for-run-id tasks:open-db run-id " client:setup (server-dat=#f)")
;; 				(server:try-running run-id)))
;; 			  (thread-sleep! 10) ;; give server a little time to start up
;; 			  (client:setup run-id remaining-tries: (- remaining-tries 1)))))))))))

;; Do all the connection work, look up the transport type and set up the
;; connection if required.
;;
;; There are two scenarios. 
;;   1. We are a test manager and we received *transport-type* and *runremote* via cmdline
;;   2. We are a run tests, list runs or other interactive process and we must figure out
;;      *transport-type* and *runremote* from the monitor.db
;;
;; client:setup
;;
;; lookup_server, need to remove *runremote* stuff
;;
(define (client:setup-http run-id #!key (remaining-tries 10) (failed-connects 0))
  (debug:print-info 2 "client:setup remaining-tries=" remaining-tries)
  (let* ((tdbdat (tasks:open-db)))
    (if (<= remaining-tries 0)

	(begin
	  (debug:print 0 "ERROR: failed to start or connect to server for run-id " run-id)
	  (exit 1))

	(let* ((server-dat (tasks:get-server (db:delay-if-busy tdbdat) run-id)))
	  (debug:print-info 4 "client:setup server-dat=" server-dat ", remaining-tries=" remaining-tries)
	  (if server-dat
	      (let* ((iface     (tasks:hostinfo-get-interface server-dat))
		     (hostname  (tasks:hostinfo-get-hostname  server-dat))
		     (port      (tasks:hostinfo-get-port      server-dat))
		     (start-res (case *transport-type*
				  ((http)(http-transport:client-connect iface port))

				  ((nmsg)(nmsg-transport:client-connect hostname port))))
		     (ping-res  (case *transport-type* 
				  ((http)(rmt:login-no-auto-client-setup start-res run-id))
				  ((nmsg)(let ((logininfo (rmt:login-no-auto-client-setup start-res run-id)))
 					   (if logininfo
 					       (car (vector-ref logininfo 1))
 					       #f))))))
		(if (and start-res
			 ping-res)
		    (begin
		      (hash-table-set! *runremote* run-id start-res)
		      (debug:print-info 2 "connected to " (http-transport:server-dat-make-url start-res))
		      start-res)
		    (begin    ;; login failed but have a server record, clean out the record and try again
		      (debug:print-info 0 "client:setup, login failed, will attempt to start server ... start-res=" start-res ", run-id=" run-id ", server-dat=" server-dat)
		      (case *transport-type* 

			((http)(http-transport:close-connections run-id)))
		      (hash-table-delete! *runremote* run-id)
		      (tasks:kill-server-run-id run-id)
		      (tasks:server-force-clean-run-record (db:delay-if-busy tdbdat)
							   run-id 
							   (tasks:hostinfo-get-interface server-dat)


							   (tasks:hostinfo-get-port      server-dat)



							   " client:setup (server-dat = #t)")

		      (if (> remaining-tries 8)
			  (thread-sleep! (+ 1 (random 5))) ;; spread out the starts a little
			  (thread-sleep! (+ 15 (random 20)))) ;; it isn't going well. give it plenty of time
		      (server:try-running run-id)
		      (thread-sleep! 5)   ;; give server a little time to start up
		      (client:setup run-id remaining-tries: (- remaining-tries 1))
		      )))
	      (begin    ;; no server registered
		(let ((num-available (tasks:num-in-available-state (db:dbdat-get-db tdbdat) run-id)))
		  (debug:print-info 0 "client:setup, no server registered, remaining-tries=" remaining-tries " num-available=" num-available)
		  (if (< num-available 2)
		      (server:try-running run-id))
		  (thread-sleep! (+ 5 (random (- 20 remaining-tries))))  ;; give server a little time to start up, randomize a little to avoid start storms.
		  (client:setup run-id remaining-tries: (- remaining-tries 1)))))))))

;; keep this as a function to ease future 
(define (client:start run-id server-info)
  (http-transport:client-connect (tasks:hostinfo-get-interface server-info)
				 (tasks:hostinfo-get-port server-info)))

;; ;; client:signal-handler
;; (define (client:signal-handler signum)
;;   (signal-mask! signum)
;;   (set! *time-to-exit* #t)
;;   (handle-exceptions
;;    exn
;;    (debug:print " ... exiting ...")
;;    (let ((th1 (make-thread (lambda ()
;; 			     "") ;; do nothing for now (was flush out last call if applicable)
;; 			   "eat response"))
;; 	 (th2 (make-thread (lambda ()
;; 			     (debug:print 0 "ERROR: Received ^C, attempting clean exit. Please be patient and wait a few seconds before hitting ^C again.")
;; 			     (thread-sleep! 1) ;; give the flush one second to do it's stuff
;; 			     (debug:print 0 "       Done.")
;; 			     (exit 4))
;; 			   "exit on ^C timer")))
;;      (thread-start! th2)
;;      (thread-start! th1)
;;      (thread-join! th2))))
;; 
;; ;; client:launch
;; ;; Need to set the signal handler somewhere other than here as this
;; ;; routine will go away.
;; ;;
;; (define (client:launch run-id)
;;   (set-signal-handler! signal/int  client:signal-handler)
;;   (set-signal-handler! signal/term client:signal-handler)
;;   (if (client:setup run-id)
;;       (debug:print-info 2 "connected as client")
;;       (begin
;; 	(debug:print 0 "ERROR: Failed to connect as client")
;; 	(exit))))
;; 

;;======================================================================
;; Copyright 2006-2012, Matthew Welland.
;; 
;;  This program is made available under the GNU GPL version 2.0 or
;;  greater. See the accompanying file COPYING for details.
;; 
;;  This program is distributed WITHOUT ANY WARRANTY; without even the
;;  implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
;;  PURPOSE.
;;======================================================================

(use srfi-1 posix regex-case base64 format dot-locking csv-xml z3 nanomsg sql-de-lite hostinfo)
(require-extension regex posix)

(require-extension (srfi 18) extras tcp rpc)

(import (prefix sqlite3 sqlite3:))
(import (prefix base64 base64:))

(declare (unit common))

(include "common_records.scm")

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

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

(define getenv get-environment-variable)
(define (safe-setenv key val)
  (if (and (string? val)(string? key))
      (handle-exceptions
       exn
       (debug:print 0 "ERROR: bad value for setenv, key=" key ", value=" val)
       (setenv key val))
      (debug:print 0 "ERROR: bad value for setenv, key=" key ", value=" val)))

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

;; GLOBAL GLETCHES
(define *db-keys* #f)
(define *configinfo* #f)
(define *configdat*  #f)
(define *toppath*    #f)
(define *already-seen-runconfig-info* #f)
(define *waiting-queue*     (make-hash-table))
(define *test-meta-updated* (make-hash-table))
(define *globalexitstatus*  0) ;; attempt to work around possible thread issues
(define *passnum*           0) ;; when running track calls to run-tests or similar
(define *write-frequency*   (make-hash-table)) ;; run-id => (vector (current-seconds) 0))
(define *alt-log-file* #f)  ;; used by -log
(define *common:denoise*    (make-hash-table)) ;; for low noise printing

;; DATABASE
(define *dbstruct-db*  #f)
(define *db-stats*            (make-hash-table)) ;; hash of vectors < count duration-total >
(define *db-stats-mutex*      (make-mutex))
(define *db-sync-mutex*       (make-mutex))
(define *db-multi-sync-mutex* (make-mutex))
(define *db-local-sync*       (make-hash-table)) ;; used to record last touch of db
(define *megatest-db*         #f)
(define *last-db-access*      (current-seconds))  ;; update when db is accessed via server
(define *db-write-access*     #t)
(define *inmemdb*             #f)
(define *task-db*             #f) ;; (vector db path-to-db)
(define *db-access-allowed*   #t) ;; flag to allow access
(define *db-access-mutex*     (make-mutex))

;; SERVER
(define *my-client-signature* #f)
(define *transport-type*    'http)
(define *transport-type*    'http)             ;; override with [server] transport http|rpc|nmsg

(define *runremote*         (make-hash-table)) ;; if set up for server communication this will hold <host port>

(define *max-cache-size*    0)
(define *logged-in-clients* (make-hash-table))
(define *client-non-blocking-mode* #f)
(define *server-id*         #f)
(define *server-info*       #f)
(define *time-to-exit*      #f)
(define *received-response* #f)
(define *default-numtries*  10)
(define *server-run*        #t)
(define *run-id*            #f)
(define *server-kind-run*   (make-hash-table))

(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

  (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)))

;; Generic string database
(define sdb:qry #f) ;; (make-sdb:qry)) ;;  'init #f)
;; Generic path database
(define *fdb* #f)

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

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

(define (common:get-megatest-exe)
  (or (getenv "MT_MEGATEST") "megatest"))

(define (common:read-encoded-string instr)
  (handle-exceptions
   exn
   (handle-exceptions
    exn
    (begin
      (debug:print 0 "ERROR: received bad encoded string \"" instr "\", message: " ((condition-property-accessor 'exn 'message) exn))
      (print-call-chain (current-error-port))
      #f)
    (read (open-input-string (base64:base64-decode instr))))
   (read (open-input-string (z3:decode-buffer (base64:base64-decode instr))))))

;; dot-locking egg seems not to work, using this for now
;; if lock is older than expire-time then remove it and try again
;; to get the lock
;;
(define (common:simple-file-lock fname #!key (expire-time 300))
  (if (file-exists? fname)
      (if (> (- (current-seconds)(file-modification-time fname)) expire-time)
	  (begin
	    (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)
	(with-input-from-file fname
	  (lambda ()
	    (equal? key-string (read-line)))))))
	
(define (common:simple-file-release-lock fname)
  (delete-file* fname))

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

(define *common:std-states*   
  '((0 "COMPLETED")
    (1 "NOT_STARTED")
    (2 "RUNNING")
    (3 "REMOTEHOSTSTART")
    (4 "LAUNCHED")
    (5 "KILLED")
    (6 "KILLREQ")
    (7 "STUCK")
    (8 "ARCHIVED")))

(define *common:std-statuses*
  '((0 "PASS")
    (1 "WARN")
    (2 "FAIL")
    (3 "CHECK")
    (4 "n/a")
    (5 "WAIVED")
    (6 "SKIP")
    (7 "DELETED")
    (8 "STUCK/DEAD")
    (9 "ABORT")))

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

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

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

(define (get-with-default val default)
  (let ((val (args:get-arg val)))
    (if val val default)))

(define (assoc/default key lst . default)
  (let ((res (assoc key lst)))
    (if res (cadr res)(if (null? default) #f (car default)))))

(define (common:get-testsuite-name)
  (or (configf:lookup *configdat* "setup" "testsuite" )
       (pathname-file *toppath*)))

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

(define (common:legacy-sync-recommended)
  (or (args:get-arg "-runtests")
      (args:get-arg "-server")
      (args:get-arg "-set-run-status")
      (args:get-arg "-remove-runs")
      (args:get-arg "-get-run-status")
      ))

(define (common:legacy-sync-required)
  (configf:lookup *configdat* "setup" "megatest-db"))

(define (std-exit-procedure)
  (let ((no-hurry  (if *time-to-exit* ;; hurry up
		       #f
		       (begin
			 (set! *time-to-exit* #t)
			 #t))))
    (debug:print-info 4 "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
			      (let ((run-ids (hash-table-keys *db-local-sync*)))
				(if (and (not (null? run-ids))
					 (configf:lookup *configdat* "setup" "megatest-db"))
				    (if no-hurry (db:multi-db-sync run-ids 'new2old))))
			      (if *dbstruct-db* (db:close-all *dbstruct-db*))
			      (if *inmemdb*     (db:close-all *inmemdb*))
			      (if (and *megatest-db*
				       (sqlite3:database? *megatest-db*))
				  (begin
				    (sqlite3:interrupt! *megatest-db*)
				    (sqlite3:finalize! *megatest-db* #t)
				    (set! *megatest-db* #f)))
			      (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)))))) "Cleanup db exit thread"))
	  (th2 (make-thread (lambda ()
			      (debug:print 4 "Attempting clean exit. Please be patient and wait a few seconds...")
			      (if no-hurry
				  (thread-sleep! 5) ;; give the clean up few seconds to do it's stuff
				  (thread-sleep! 2))
			      (debug:print 4 " ... done")
			      )
			    "clean exit")))
      (thread-start! th1)
      (thread-start! th2)
      (thread-join! th1))))

(define (std-signal-handler signum)
  ;; (signal-mask! signum)
  (set! *time-to-exit* #t)
  (debug:print 0 "ERROR: Received signal " signum " exiting promptly")
  ;; (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

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

;; Convert strings like "5s 2h 3m" => 60x60x2 + 3x60 + 5
(define (common:hms-string->seconds tstr)
  (let ((parts     (string-split tstr))
	(time-secs 0)
	;; s=seconds, m=minutes, h=hours, d=days

	     (if (string-match (regexp modpatt) item)
		 (set! res #t))))
	 (string-split patts ","))
	res)
      #t))

;; (map print (map car (hash-table->alist (read-config "runconfigs.config" #f #t))))
(define (common:get-runconfig-targets #!key (configf #f))
  (sort (map car (hash-table->alist
		  (or configf
		      (read-config "runconfigs.config"
			       #f #t))))
	string<?))

;; '(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
;;======================================================================

(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")))

(define (common:args-get-testpatt rconf)
  (let* ((rtestpatt     (if rconf (runconfigs-get rconf "TESTPATT") #f))
	 (args-testpatt (or (args:get-arg "-testpatt")
			    (args:get-arg "-runtests")
			    "%"))
	 (testpatt    (or (and (equal? args-testpatt "%")
			       rtestpatt)
			  args-testpatt)))
    (if rtestpatt (debug:print-info 0 "TESTPATT from runconfigs: " rtestpatt))
    testpatt))

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

(define (common:args-get-target #!key (split #f))
  (let* ((keys    (keys:config-get-fields *configdat*))
	 (numkeys (length keys))
	 (target  (if (args:get-arg "-reqtarg")
		      (args:get-arg "-reqtarg")
		      (if (args:get-arg "-target")
			  (args:get-arg "-target")
			  (getenv "MT_TARGET"))))
	 (tlist   (if target (string-split target "/" #t) '()))
	 (valid   (if target
		      (and (not (null? tlist))
			   (eq? numkeys (length tlist))
			   (null? (filter string-null? tlist)))
		      #f)))
    (if valid
	(if split
	    tlist
	    target)
	(if target
	    (begin
	      (debug:print 0 "ERROR: Invalid target, spaces or blanks not allowed \"" target "\", target should be: " (string-intersperse keys "/"))
	      #f)
	    #f))))

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

;; return a nice clean pathname made absolute
(define (nice-path dir)
  (normalize-pathname (if (absolute-pathname? dir)
			  dir
			  (conc (current-directory) "/" dir))))















(define (get-cpu-load)
  (car (common:get-cpu-load)))
;;   (let* ((load-res (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)))

(define (get-uname . params)
  (let* ((uname-res (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)
  ;; (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))

(define (get-df path)
  (let* ((df-results (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))
  
;; paths is list of lists ((name path) ... )
;;
(define (common:get-disk-with-most-free-space disks minsize)
  (let ((best     #f)
	(bestsize 0))
    (for-each 
     (lambda (disk-num)
       (let* ((dirpath    (cadr (assoc disk-num disks)))
	      (freespc    (cond
			   ((not (directory? dirpath))
			    (if (common:low-noise-print 50 "disks not a dir " disk-num)
				(debug:print 0 "WARNING: disk " disk-num " at path " dirpath " is not a directory - ignoring it."))
			    -1)
			   ((not (file-write-access? dirpath))
			    (if (common:low-noise-print 50 "disks not writeable " disk-num)
				(debug:print 0 "WARNING: disk " disk-num " at path " dirpath " is not writeable - ignoring it."))
			    -1)
			   ((not (eq? (string-ref dirpath 0) #\/))
			    (if (common:low-noise-print 50 "disks not a proper path " disk-num)
				(debug:print 0 "WARNING: disk " disk-num " at path " dirpath " is not a fully qualified path - ignoring it."))
			    -1)
			   (else
			    (get-df dirpath)))))
	 (if (> freespc bestsize)
	     (begin
	       (set! best     (cons disk-num dirpath))
	       (set! bestsize freespc)))))
     (map car disks))
    (if (and best (> bestsize minsize))
	best
	#f))) ;; #f means no disk candidate found

;;======================================================================
;; E N V I R O N M E N T   V A R S
;;======================================================================
	      
(define (save-environment-as-files fname #!key (ignorevars (list "USER" "HOME" "DISPLAY" "LS_COLORS" "XKEYSYMDB" "EDITOR" "MAKEFLAGS" "MAKEF")))
  (let ((envvars (get-environment-variables))
        (whitesp (regexp "[^a-zA-Z0-9_\\-:,.\\/%$]")))
     (with-output-to-file (conc fname ".csh")
       (lambda ()
          (for-each (lambda (keyval)
		      (let* ((key   (car keyval))
			     (val   (cdr keyval))

(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->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)))

;;======================================================================
;; Colors
;;======================================================================
      
(define (common:name->iup-color name)
  (case (string->symbol (string-downcase name))

  (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 (server:get-best-guess-address hostname)
  (let ((res #f))
    (for-each 
     (lambda (adr)
       (if (not (eq? (u8vector-ref adr 0) 127))
	   (set! res adr)))
     ;; NOTE: This can fail when there is no mention of the host in /etc/hosts. FIXME
     (vector->list (hostinfo-addresses (hostname->hostinfo hostname))))
    (string-intersperse 
     (map number->string
	  (u8vector->list
	   (if res res (hostname->ip hostname)))) ".")))

(define (common:open-nm-req addr)
  (let* ((req (nn-socket 'req))
	 (res (nn-connect req addr)))
    req))

;; (with-output-to-string (lambda ()(serialize obj)))
(define (common:nm-send-receive soc msg)
  (nn-send soc msg)
  (nn-recv soc))

(define (common:close-nm-req soc)
  (nn-close soc))

(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)))
    
(define (common:nm-send-receive-timeout req msg)
  (let* ((key     "ping")
	 (success #f)
	 (keepwaiting #t)
	 (result  #f)
	 (sendrec (make-thread
		   (lambda ()
		     (nn-send req msg)
		     (set! result (nn-recv req))
		     (set! success #t))
		   "send-receive"))
	 (timeout (make-thread (lambda ()
				 (let loop ((count 0))
				   (thread-sleep! 1)
				   (print "still waiting after count seconds...")
				   (if (and keepwaiting (< count 10))
				       (loop (+ count 1))))
				 (if keepwaiting
				     (begin
				       (print "timeout waiting for reply")
				       (thread-terminate! sendrec))))
			       "timeout")))
    (handle-exceptions
     exn
     (begin
       (print-call-chain)
       (print 0 " message: " ((condition-property-accessor 'exn 'message) exn))
       (print "exn=" (condition->list exn)))
     (thread-start! timeout)
     (thread-start! sendrec)
     (thread-join!  sendrec)
     (if success (thread-terminate! timeout)))
    result))
    
(define (common:ping-nm req)
  ;; send a random number and check that we get it back
  (let* ((key     "ping")
	 (success #f)
	 (keepwaiting #t)
	 (ping    (make-thread
		   (lambda ()
		     (print "ping: sending string \"" key "\", expecting " (current-process-id))
		     (nn-send req key)
		     (let ((result  (nn-recv req)))
		       (if (equal? (conc (current-process-id)) result)
			   (begin
			     (print "ping, success: received \"" result "\"")
			     (set! success #t))
			   (begin
			     (print "ping, failed: received key \"" result "\"")
			     (set! keepwaiting #f)
			     (set! success #f)))))
		   "ping"))
	 (timeout (make-thread (lambda ()
				 (let loop ((count 0))
				   (thread-sleep! 1)
				   (print "still waiting after count seconds...")
				   (if (and keepwaiting (< count 10))
				       (loop (+ count 1))))
				 (if keepwaiting
				     (begin
				       (print "timeout waiting for ping")
				       (thread-terminate! ping))))
			       "timeout")))
    (handle-exceptions
     exn
     (begin
       (print-call-chain)
       (print 0 " message: " ((condition-property-accessor 'exn 'message) exn))
       (print "exn=" (condition->list exn))
       (print "ping failed to connect to tcp://" hostport))
     (thread-start! timeout)
     (thread-start! ping)
     (thread-join! ping)
     (if success (thread-terminate! timeout)))
    (if return-socket
	(if success req #f)
	(begin
	  (nn-close req)
	  success))))

;;======================================================================
;; D A S H B O A R D   D B 
;;======================================================================

(define (mddb:open-db)
  (let* ((db (open-database (conc (get-environment-variable "HOME") "/.dashboard.db"))))
    (set-busy-handler! db (busy-timeout 10000))
    (for-each
     (lambda (qry)
       (exec (sql db qry)))
     (list 
      "CREATE TABLE IF NOT EXISTS vars       (id INTEGER PRIMARY KEY,key TEXT, val TEXT, CONSTRAINT varsconstraint UNIQUE (key));"
      "CREATE TABLE IF NOT EXISTS dashboards (
          id         INTEGER PRIMARY KEY,
          pid        INTEGER,
          username   TEXT,
          hostname   TEXT,
          ipaddr     TEXT,
          portnum    INTEGER,
          start_time TIMESTAMP DEFAULT (strftime('%s','now')),
             CONSTRAINT hostport UNIQUE (hostname,portnum)
        );"
      ))
    db))

;; register a dashboard 
;;
(define (mddb:register-dashboard port)
  (let* ((pid      (current-process-id))
	 (hostname (get-host-name))
	 (ipaddr   (server:get-best-guess-address hostname))
	 (username (current-user-name)) ;; (car userinfo)))
	 (db      (mddb:open-db)))
    (print "Register monitor, pid: " pid ", hostname: " hostname ", port: " port ", username: " username)
    (exec (sql db "INSERT OR REPLACE INTO dashboards (pid,username,hostname,ipaddr,portnum) VALUES (?,?,?,?,?);")
	   pid username hostname ipaddr port)
    (close-database db)))

;; unregister a monitor
;;
(define (mddb:unregister-dashboard host port)
  (let* ((db      (mddb:open-db)))
    (print "Register unregister monitor, host:port=" host ":" port)
    (exec (sql db "DELETE FROM dashboards WHERE hostname=? AND portnum=?;") host port)
    (close-database db)))

;; get registered dashboards
;;
(define (mddb:get-dashboards)
  (let ((db (mddb:open-db)))
    (query fetch-column
	   (sql db "SELECT ipaddr || ':' || portnum FROM dashboards;"))))
    
;;======================================================================
;;  T E S T   L A U N C H I N G   P E R   I T E M   W I T H   H O S T   T Y P E S
;;======================================================================
;; 
;; [host-types]
;; general ssh #{getbgesthost general}
;; nbgeneral nbjob run JOBCOMMAND -log $MT_LINKTREE/$MT_TARGET/$MT_RUNNAME.$MT_TESTNAME-$MT_ITEM_PATH.lgo
;; 
;; [hosts]
;; general cubian xena
;; 
;; [launchers]
;; envsetup general
;; xor/%/n 4C16G
;; % nbgeneral
;; 
;; [jobtools]
;; launcher bsub
;; # if defined and not "no" flexi-launcher will bypass launcher unless there is no
;; # match.
;; flexi-launcher yes  

(define (common:get-launcher configdat testname itempath)
  (let ((fallback-launcher (configf:lookup configdat "jobtools" "launcher")))
    (if (and (configf:lookup configdat "jobtools" "flexi-launcher") ;; overrides launcher
	     (not (equal? (configf:lookup configdat "jobtools" "flexi-launcher") "no")))
	(let* ((launchers         (hash-table-ref/default configdat "launchers" '())))
	  (if (null? launchers)
	      fallback-launcher
	      (let loop ((hed (car launchers))
			 (tal (cdr launchers)))
		(let ((patt      (car hed))
		      (host-type (cadr hed)))
		  (if (tests:match patt testname itempath)
		      (begin
			(debug:print-info 0 "Have flexi-launcher match for " testname "/" itempath " = " host-type)
			(let ((launcher (configf:lookup configdat "host-types" host-type)))
			  (if launcher
			      launcher
			      (begin
				(debug:print-info 0 "WARNING: no launcher found for host-type " host-type)
				(if (null? tal)
				    fallback-launcher
				    (loop (car tal)(cdr tal)))))))
		      ;; no match, try again
		      (if (null? tal)
			  fallback-launcher
			  (loop (car tal)(cdr tal))))))))
	fallback-launcher)))
  

;;======================================================================
;; Copyright 2006-2012, Matthew Welland.
;; 
;;  This program is made available under the GNU GPL version 2.0 or
;;  greater. See the accompanying file COPYING for details.
;; 
;;  This program is distributed WITHOUT ANY WARRANTY; without even the
;;  implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
;;  PURPOSE.
;;======================================================================

;; (use trace)

;; Some of these routines use:
;;
;;     http://www.cs.toronto.edu/~gfb/scheme/simple-macros.html
;;
;; Syntax for defining macros in a simple style similar to function definiton,
;;  when there is a single pattern for the argument list and there are no keywords.
;;
;; (define-simple-syntax (name arg ...) body ...)
;;

(define-syntax define-simple-syntax
  (syntax-rules ()
    ((_ (name arg ...) body ...)
     (define-syntax name (syntax-rules () ((name arg ...) (begin body ...)))))))

(define-syntax common:handle-exceptions
  (syntax-rules ()
    ((_ exn-in errstmt ...)(handle-exceptions exn-in errstmt ...))))

(define (debug:calc-verbosity vstr)
  (cond
   ((number? vstr) vstr)
   ((not (string?  vstr))   1)
   ;; ((string-match  "^\\s*$" vstr) 1)
   (vstr           (let ((debugvals  (filter number? (map string->number (string-split vstr ",")))))
		     (cond
		      ((> (length debugvals) 1) debugvals)
		      ((> (length debugvals) 0)(car debugvals))
		      (else 1))))
   ((args:get-arg "-v")   2)
   ((args:get-arg "-q")    0)
   (else                   1)))

;; check verbosity, #t is ok
(define (debug:check-verbosity verbosity vstr)
  (if (not (or (number? verbosity)
	       (list?   verbosity)))
      (begin
	(print "ERROR: Invalid debug value \"" vstr "\"")
	#f)
      #t))

(define (debug:debug-mode n)
  (cond
   ((and (number? *verbosity*)   ;; number number
	 (number? n))
    (<= n *verbosity*))
   ((and (list? *verbosity*)     ;; list   number
	 (number? n))
    (member n *verbosity*))
   ((and (list? *verbosity*)     ;; list   list
	 (list? n))
    (not (null? (lset-intersection! eq? *verbosity* n))))
   ((and (number? *verbosity*)
	 (list? n))
    (member *verbosity* n))))
      
(define (debug:setup)
  (let ((debugstr (or (args:get-arg "-debug")
		      (getenv "MT_DEBUG_MODE"))))
    (set! *verbosity* (debug:calc-verbosity debugstr))
    (debug:check-verbosity *verbosity* debugstr)
    ;; if we were handed a bad verbosity rule then we will override it with 1 and continue
    (if (not *verbosity*)(set! *verbosity* 1))
    (if (or (args:get-arg "-debug")
	    (not (getenv "MT_DEBUG_MODE")))
	(setenv "MT_DEBUG_MODE" (if (list? *verbosity*)
				    (string-intersperse (map conc *verbosity*) ",")
				    (conc *verbosity*))))))
  

(define (debug:print n . params)
  (if (debug:debug-mode n)
      (with-output-to-port (current-error-port)
	(lambda ()
	  (if *logging*
	      (db:log-event (apply conc params))
	      ;; (apply print "pid:" (current-process-id) " " params)
	      (apply print params)
	      )))))

(define (debug:print-info n . params)
  (if (debug:debug-mode n)
      (with-output-to-port (current-error-port)
	(lambda ()
	  (let ((res (format#format #f "INFO: (~a) ~a" n (apply conc params))))
	    (if *logging*
		(db:log-event res)
		;; (apply print "pid:" (current-process-id) " " "INFO: (" n ") " params) ;; res)
		(apply print "INFO: (" n ") " params) ;; res)
		))))))

;; if a value is printable (i.e. string or number) return the value
;; else return an empty string
(define-inline (printable val)
  (if (or (number? val)(string? val)) val ""))

(define configf:section-rx (regexp "^\\[(.*)\\]\\s*$"))
(define configf:blank-l-rx (regexp "^\\s*$"))
(define configf:key-sys-pr (regexp "^(\\S+)\\s+\\[system\\s+(\\S+.*)\\]\\s*$"))
(define configf:key-val-pr (regexp "^(\\S+)(\\s+(.*)|())$"))
(define configf:key-no-val (regexp "^(\\S+)(\\s*)$"))
(define configf:comment-rx (regexp "^\\s*#.*"))
(define configf:cont-ln-rx (regexp "^(\\s+)(\\S+.*)$"))
(define configf:settings   (regexp "^\\[configf:settings\\s+(\\S+)\\s+(\\S+)]\\s*$"))

;; read a line and process any #{ ... } constructs

(define configf:var-expand-regex (regexp "^(.*)#\\{(scheme|system|shell|getenv|get|runconfigs-get|rget)\\s+([^\\}\\{]*)\\}(.*)"))
(define (configf:process-line l ht allow-system)
  (let loop ((res l))
    (if (string? res)
	(let ((matchdat (string-search configf:var-expand-regex res)))
	  (if matchdat
	      (let* ((prestr  (list-ref matchdat 1))
		     (cmdtype (list-ref matchdat 2)) ;; eval, system, shell, getenv
		     (cmd     (list-ref matchdat 3))

					(sect  (car parts))
					(var   (cadr parts)))
				   (conc "(lambda (ht)(config-lookup ht \"" sect "\" \"" var "\"))")))
				((runconfigs-get) (conc "(lambda (ht)(runconfigs-get ht \"" cmd "\"))"))
				((rget)           (conc "(lambda (ht)(runconfigs-get ht \"" cmd "\"))"))
				(else "(lambda (ht)(print \"ERROR\") \"ERROR\")"))))
		;; (print "fullcmd=" fullcmd)
		(handle-exceptions
		 exn
		 (debug:print 0 "ERROR: failed to process config input \"" l "\"")
		 (if (or allow-system
			 (not (member cmdtype '("system" "shell"))))
		     (with-input-from-string fullcmd
		       (lambda ()
			 (set! result ((eval (read)) ht))))
		    (set! result (conc "#{(" cmdtype ") "  cmd "}"))))
		(loop (conc prestr result poststr)))
	      res))
	res)))

;; Run a shell command and return the output as a string
(define (shell cmd)
  (let* ((output (cmd-run->list cmd))

	  (with-output-to-port (current-error-port)
	    (lambda ()
	      (print "ERROR: " cmd " returned bad exit code " status)))
	  ""))))

;; Lookup a value in runconfigs based on -reqtarg or -target
(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))))

;; this was inline but I'm pretty sure that is a hold over from when it was *very* simple ...
;;
(define (configf:read-line p ht allow-processing settings)
  (let loop ((inl (read-line p)))
    (let ((cont-line (and (string? inl)
			  (not (string-null? inl))
			  (equal? "\\" (string-take-right inl 1)))))
      (if cont-line ;; last character is \ 
	  (let ((nextl (read-line p)))
	    (if (not (eof-object? nextl))
		(loop (string-append (if cont-line 
					 (string-take inl (- (string-length inl) 1))
					 inl)
				     nextl))))
	  (let ((res (case allow-processing ;; if (and allow-processing 
		       ;;	   (not (eq? allow-processing 'return-string)))
		       ((#t #f)
			(configf:process-line inl ht allow-processing))
		       ((return-string)
			inl)
		       (else
			(configf:process-line inl ht allow-processing)))))
	    (if (and (string? res)
		     (not (equal? (hash-table-ref/default settings "trim-trailing-spaces" "no") "no")))
		(string-substitute "\\s+$" "" res)
		res))))))
      
;; read a config file, returns hash table of alists

;; read a config file, returns hash table of alists
;; adds to ht if given (must be #f otherwise)
;; envion-patt is a regex spec that identifies sections that will be eval'd
;; in the environment on the fly
;; sections: #f => get all, else list of sections to gather
(define (read-config path ht allow-system #!key (environ-patt #f)(curr-section #f)(sections #f)(settings (make-hash-table)))
  (debug:print-info 5 "read-config " path " allow-system " allow-system " environ-patt " environ-patt " curr-section: " curr-section " sections: " sections " pwd: " (current-directory))
  (debug:print 9 "START: " path)
  (if (not (file-exists? path))
      (begin 
	(debug:print-info 1 "read-config - file not found " path " current path: " (current-directory))
	;; WARNING: This is a risky change but really, we should not return an empty hash table if no file read?
	#f) ;; (if (not ht)(make-hash-table) ht))
      (let ((inp        (open-input-file path))
	    (res        (if (not ht)(make-hash-table) ht)))
	(let loop ((inl               (configf:read-line inp res allow-system settings)) ;; (read-line inp))
		   (curr-section-name (if curr-section curr-section "default"))
		   (var-flag #f);; turn on for key-var-pr and cont-ln-rx, turn off elsewhere
		   (lead     #f))
	  (debug:print-info 8 "curr-section-name: " curr-section-name " var-flag: " var-flag "\n   inl: \"" inl "\"")
	  (if (eof-object? inl) 
	      (begin
		(close-input-port inp)
		(hash-table-delete! res "") ;; we are using "" as a dumping ground and must remove it before returning the ht
		(debug:print 9 "END: " path)
		res)
	      (regex-case 
	       inl 
	       (configf:comment-rx _                  (loop (configf:read-line inp res allow-system settings) curr-section-name #f #f))
	       (configf:blank-l-rx _                  (loop (configf:read-line inp res allow-system settings) curr-section-name #f #f))
	       (configf:settings   ( x setting val  ) (begin
							(hash-table-set! settings setting val)
							(loop (configf:read-line inp res allow-system settings) curr-section-name #f #f)))
	       (configf:include-rx ( x include-file ) (let* ((curr-conf-dir (pathname-directory path))
							     (full-conf     (if (absolute-pathname? include-file)
										include-file
										(nice-path 
										 (conc (if curr-conf-dir
											   curr-conf-dir
											   ".")
										       "/" include-file)))))
							(if (file-exists? full-conf)
							    (begin
							      ;; (push-directory conf-dir)
							      (debug:print 9 "Including: " full-conf)
							      (read-config full-conf res allow-system environ-patt: environ-patt curr-section: curr-section-name sections: sections settings: settings)
							      ;; (pop-directory)
							      (loop (configf:read-line inp res allow-system settings) curr-section-name #f #f))
							    (begin
							      (debug:print '(2 9) "INFO: include file " include-file " not found (called from " path ")")
							      (debug:print 2 "        " full-conf)
							      (loop (configf:read-line inp res allow-system settings) curr-section-name #f #f)))))
	       (configf:section-rx ( x section-name ) (loop (configf:read-line inp res allow-system settings)
							    ;; if we have the sections list then force all settings into "" and delete it later?
							    (if (or (not sections) 
								    (member section-name sections))
								section-name "") ;; stick everything into ""
							    #f #f))
	       (configf:key-sys-pr ( x key cmd      ) (if allow-system
							  (let ((alist (hash-table-ref/default res curr-section-name '()))

												 " output: " cmdres)
										    (exit 1)))
									      (if (null? res)
										  ""
										  (string-intersperse res " "))))))
							    (hash-table-set! res curr-section-name 
									     (config:assoc-safe-add alist
									   			    key 
												    (case allow-system
												      ((return-procs) val-proc)
												      ((return-string) cmd)
												      (else (val-proc)))))
							    (loop (configf:read-line inp res allow-system settings) curr-section-name #f #f))
							  (loop (configf:read-line inp res allow-system settings) curr-section-name #f #f)))
	       (configf:key-val-pr ( x key unk1 val unk2 ) (let* ((alist   (hash-table-ref/default res curr-section-name '()))
								  (envar   (and environ-patt (string-search (regexp environ-patt) curr-section-name)))
								  (realval (if envar
									       (config:eval-string-in-environment val)
									       val)))
							     (debug:print-info 6 "read-config env setting, envar: " envar " realval: " realval " val: " val " key: " key " curr-section-name: " curr-section-name)
							     (if envar (safe-setenv key realval))

							     (debug:print 10 "   setting: [" curr-section-name "] " key " = " val)

							     (hash-table-set! res curr-section-name 
									      (config:assoc-safe-add alist key realval))
							     (loop (configf:read-line inp res allow-system settings) curr-section-name key #f)))
	       (configf:key-no-val ( x key val)             (let* ((alist   (hash-table-ref/default res curr-section-name '())))
							      (debug:print 10 "   setting: [" curr-section-name "] " key " = #t")
							      (hash-table-set! res curr-section-name 
									       (config:assoc-safe-add alist key #t))
							      (loop (configf:read-line inp res allow-system settings) curr-section-name key #f)))
	       ;; if a continued line
	       (configf:cont-ln-rx ( x whsp val     ) (let ((alist (hash-table-ref/default res curr-section-name '())))
						(if var-flag             ;; if set to a string then we have a continued var
						    (let ((newval (conc 
								   (config-lookup res curr-section-name var-flag) "\n"
								   ;; trim lead from the incoming whsp to support some indenting.
								   (if lead
								       (string-substitute (regexp lead) "" whsp)
								       "")
								   val)))
						      ;; (print "val: " val "\nnewval: \"" newval "\"\nvarflag: " var-flag)
						      (hash-table-set! res curr-section-name 
								       (config:assoc-safe-add alist var-flag newval))
						      (loop (configf:read-line inp res allow-system settings) curr-section-name var-flag (if lead lead whsp)))
						    (loop (configf:read-line inp res allow-system settings) curr-section-name #f #f))))
	       (else (debug:print 0 "ERROR: problem parsing " path ",\n   \"" inl "\"")
		     (set! var-flag #f)
		     (loop (configf:read-line inp res allow-system settings) curr-section-name #f #f))))))))
  
;; pathenvvar will set the named var to the path of the config
(define (find-and-read-config fname #!key (environ-patt #f)(given-toppath #f)(pathenvvar #f))
  (let* ((curr-dir   (current-directory))
         (configinfo (find-config fname toppath: given-toppath))
	 (toppath    (car configinfo))
	 (configfile (cadr configinfo)))

	       (lambda (sheet-name)
		 (let* ((dat-path  (conc refdb-path "/" sheet-name ".dat"))
			(ref-dat   (configf:read-file dat-path #f #t))
			(ref-assoc (map (lambda (key)
					  (list key (hash-table-ref ref-dat key)))
					(hash-table-keys ref-dat))))
				   ;; (hash-table->alist ref-dat)))
		   ;; (set! data (append data (list (list sheet-name ref-assoc))))))
		   (set! data (cons (list sheet-name ref-assoc) data))))
	       sheets)
	      (list data "NO ERRORS"))))))

;; map over all pairs in a three level hierarchial alist and apply a function to the keys/val
;;
(define (configf:map-all-hier-alist data proc #!key (initproc1 #f)(initproc2 #f)(initproc3 #f))
  (for-each 

	    (for-each
	     (lambda (varname)
	       (let* ((valtmp (assoc varname sectiondat))
		      (val    (if valtmp (cadr valtmp) "")))
		 (proc sheetname sectionname varname val)))
	     (map car sectiondat))))
	(map car sheetdat))))
   (map car data))
  data)

;;======================================================================
;;  C O N F I G   T O / F R O M   A L I S T
;;======================================================================

(define (configf:config->alist cfgdat)
  (hash-table->alist cfgdat))

(define (configf:alist->config adat)
  (let ((ht (make-hash-table)))
    (for-each
     (lambda (section)
       (hash-table-set! ht (car section)(cdr section)))
     adat)
    ht))

(define (configf:read-alist fname)
  (configf:alist->config
   (with-input-from-file fname read)))

(define (configf:write-alist cdat fname)
  (with-output-to-file fname
    (lambda ()
      (pp (configf:config->alist cdat)))))
     

(use sqlite3 srfi-1 posix regex regex-case srfi-69)
(import (prefix sqlite3 sqlite3:))

(declare (unit dashboard-tests))
(declare (uses common))
(declare (uses db))
(declare (uses gutils))
(declare (uses rmt))
(declare (uses ezsteps))
;; (declare (uses sdb))
;; (declare (uses filedb))

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

;;======================================================================
;; C O M M O N
;;======================================================================

(define *dashboard-comment-share-slot* #f)

(define (dtests:get-pre-command #!key (default-override #f))
  (let ((cfg-ovrd (configf:lookup *configdat* "dashboard" "pre-command")))
    (or cfg-ovrd default-override "xterm -geometry 180x20 -e \"")))

(define (dtests:get-post-command #!key (default-override #f))
  (let ((cfg-ovrd (configf:lookup *configdat* "dashboard" "post-command")))

			 (iup:label "TestComment                             "
				    #:expand "HORIZONTAL")
			 (lambda (testdat)
			   (let ((newcomment (db:test-get-comment testdat)))
			     (if *dashboard-comment-share-slot*
				 (if (not (equal? (iup:attribute *dashboard-comment-share-slot* "VALUE")
						  newcomment))
				     (iup:attribute-set! *dashboard-comment-share-slot*
							 "VALUE"
							 newcomment)))
			     newcomment)))
	    (store-label "testid"
			 (iup:label "TestId                             "
				    #:expand "HORIZONTAL")
			 (lambda (testdat)

	   (append (map (lambda (val)
			  (iup:label val ; #:expand "HORIZONTAL"
				     ))
			(list "Author: "
			      "Owner: "
			      "Reviewed: "
			      "Tags: "
			      "Description: "))

		   (list (iup:label "" #:expand "VERTICAL"))))
    (apply iup:vbox  ; #:expand "YES"
	   (list 
	    (store-meta "author"
			 (iup:label (db:testmeta-get-author testmeta) #:expand "HORIZONTAL")
			 (lambda (testmeta)(db:testmeta-get-author testmeta)))
	    (store-meta "owner"

			   (test-meta-panel-get-description testmeta)))
	    )))))


;;======================================================================
;; Run info panel
;;======================================================================
(define (run-info-panel db keydat testdat runname)
  (let* ((run-id     (db:test-get-run_id testdat))
	 (rundat     (db:get-run-info db run-id))
	 (header     (db:get-header rundat))
	 (event_time (db:get-value-by-header (db:get-rows rundat)
					     (db:get-header rundat)
					     "event_time")))
    (iup:frame 
     #:title "Megatest Run Info" ; #:expand "YES"
     (iup:hbox ; #:expand "YES"
      (apply iup:vbox ; #:expand "YES"
	     (append (map (lambda (keyval)

			  (iup:label val ; #:expand "HORIZONTAL"
				     ))
			(list "Hostname: "
			      "Uname -a: "
			      "Disk free: "
			      "CPU Load: "
			      "Run duration: "
			      "Logfile: "
			      "Top process id: "))
		   (iup:label "" #:expand "VERTICAL")))
    (apply iup:vbox ; #:expand "YES"
	   (list
	    ;; NOTE: Yes, the host can change!
	    (store-label "HostName"
			 (iup:label ;; (sdb:qry 'getstr 
			  (db:test-get-host testdat) ;; )
			  #:expand "HORIZONTAL")
			 (lambda (testdat)(db:test-get-host testdat)))
	    (store-label "Uname"
			 (iup:label "                                                   " #:expand "HORIZONTAL")
			 (lambda (testdat) ;; (sdb:qry 'getstr 
			   (db:test-get-uname testdat))) ;; )
	    (store-label "DiskFree"
			 (iup:label (conc (db:test-get-diskfree testdat)) #:expand "HORIZONTAL")
			 (lambda (testdat)(conc (db:test-get-diskfree testdat))))
	    (store-label "CPULoad"
			 (iup:label (conc (db:test-get-cpuload testdat)) #:expand "HORIZONTAL")
			 (lambda (testdat)(conc (db:test-get-cpuload testdat))))
	    (store-label "RunDuration"
			 (iup:label (conc (seconds->hr-min-sec (db:test-get-run_duration testdat))) #:expand "HORIZONTAL")
			 (lambda (testdat)(conc (seconds->hr-min-sec (db:test-get-run_duration testdat)))))
	    (store-label "LogFile"
			 (iup:label (conc (db:test-get-final_logf testdat)) #:expand "HORIZONTAL")
			 (lambda (testdat)(conc (db:test-get-final_logf testdat))))
	    (store-label "ProcessId"
			 (iup:label (conc (db:test-get-process_id testdat)) #:expand "HORIZONTAL")
			 (lambda (testdat)(conc (db:test-get-process_id testdat))))
	    )))))

;; use a global for setting the buttons colors
;;                           state status teststeps
(define *state-status* (vector #f #f #f))
(define (update-state-status-buttons testdat)
  (let* ((state  (db:test-get-state  testdat))
	 (status (db:test-get-status testdat))
	 (color  (car (gutils:get-color-for-state-status state status))))
    ((vector-ref *state-status* 0) state color)
    ((vector-ref *state-status* 1) status color)))

(define *dashboard-test-db* #t)
(define *dashboard-comment-share-slot* #f)

;;======================================================================
;; Set fields 
;;======================================================================
(define (set-fields-panel dbstruct run-id test-id testdat #!key (db #f))
  (let ((newcomment #f)
	(newstatus  #f)
	(newstate   #f)
	(wtxtbox    #f))
    (iup:frame
     #:title "Set fields"
     (iup:vbox
      (iup:hbox (iup:label "Comment:")
		(let ((txtbox (iup:textbox #:action (lambda (val a b)
						      (rmt:test-set-state-status-by-id run-id test-id #f #f b)
						      ;; IDEA: Just set a variable with the proc to call?
						      (rmt:test-set-state-status-by-id run-id test-id #f #f b)
						      (set! newcomment b))
					   #:value (db:test-get-comment testdat)
					   #:expand "HORIZONTAL")))
		  (set! wtxtbox txtbox)
		  txtbox))
		  
      (apply iup:hbox
	     (iup:label "STATE:" #:size "30x")
	     (let* ((btns  (map (lambda (state)
				  (let ((btn (iup:button state
							 #:expand "HORIZONTAL" #:size "50x" #:font "Courier New, -10"
							 #:action (lambda (x)
								    (rmt:test-set-state-status-by-id run-id test-id state #f #f)
								    (db:test-set-state! testdat state)))))
				    btn))
				(map cadr *common:std-states*)))) ;; (list "COMPLETED" "NOT_STARTED" "RUNNING" "REMOTEHOSTSTART" "LAUNCHED" "KILLED" "KILLREQ"))))
	       (vector-set! *state-status* 0
			    (lambda (state color)
			      (for-each 
			       (lambda (btn)
				 (let* ((name     (iup:attribute btn "TITLE"))
					(newcolor (if (equal? name state) color "192 192 192")))
				   (if (not (colors-similar? newcolor (iup:attribute btn "BGCOLOR")))
				       (iup:attribute-set! btn "BGCOLOR" newcolor))))
			       btns)))
	       btns))
      (apply iup:hbox
	     (iup:label "STATUS:" #:size "30x")
	     (let* ((btns  (map (lambda (status)
				  (let ((btn (iup:button status
							 #:expand "HORIZONTAL" #:size "50x" #:font "Courier New, -10"
							 #:action (lambda (x)
								    (let ((t (iup:attribute x "TITLE")))
								      (if (equal? t "WAIVED")
									  (iup:show (dashboard-tests:waiver run-id testdat 
													    (if wtxtbox (iup:attribute wtxtbox "VALUE") #f)
													    (lambda (c)
													      (set! newcomment c)
													      (if wtxtbox 
														  (begin
														    (iup:attribute-set! wtxtbox "VALUE" c)
														    (if (not *dashboard-comment-share-slot*)
															(set! *dashboard-comment-share-slot* wtxtbox)))
														  ))))
									  (begin
									    (rmt:test-set-state-status-by-id run-id test-id #f status #f)
									    (db:test-set-status! testdat status))))))))
				    btn))
				(map cadr *common:std-statuses*)))) ;; (list  "PASS" "WARN" "FAIL" "CHECK" "n/a" "WAIVED" "SKIP"))))
	       (vector-set! *state-status* 1
			    (lambda (status color)
			      (for-each 
			       (lambda (btn)
				 (let* ((name     (iup:attribute btn "TITLE"))
					(newcolor (if (equal? name status) color "192 192 192")))
				   (if (not (colors-similar? newcolor (iup:attribute btn "BGCOLOR")))

					 (conc "ezstep run from step " stepname)))))
    ;; (iup:button "Refresh test data"
    ;;     	#:expand "HORIZONTAL"
    ;;     	#:action (lambda (obj)
    ;;     		   (print "Refresh test data " stepname))
    )))

(define (dashboard-tests:waiver run-id testdat ovrdval cmtcmd)
  (let* ((wpatt (configf:lookup *configdat* "setup" "waivercommentpatt"))
	 (wregx (if (string? wpatt)(regexp wpatt) #f))
	 (wmesg (iup:label (if wpatt (conc "Comment must match pattern " wpatt) "")))
	 (comnt (iup:textbox #:action (lambda (val a b)
					(if wpatt
					    (if (string-match wregx b)
						(iup:attribute-set! wmesg "TITLE" (conc "Comment matches " wpatt))

			      #:expand "HORIZONTAL"
			      #:action (lambda (obj)
					 (let ((comment (iup:attribute comnt "VALUE"))
					       (test-id (db:test-get-id testdat)))
					   (if (or (not wpatt)
						   (string-match wregx comment))
					       (begin
						 (rmt:test-set-state-status-by-id run-id test-id #f "WAIVED" comment)
						 (db:test-set-status! testdat "WAIVED")
						 (cmtcmd comment)
						 (iup:destroy! dlog))))))
		  (iup:button "Cancel"
			      #:expand "HORIZONTAL" 
			      #:action (lambda (obj)
					 (iup:destroy! dlog)))))))
    dlog))


;;======================================================================
;;
;;======================================================================
(define (examine-test run-id test-id) ;; run-id run-key origtest)
  (let* ((db-path       (db:dbfile-path run-id)) ;; (conc (configf:lookup *configdat* "setup" "linktree") "/db/" run-id ".db"))
	 (dbstruct      (make-dbr:dbstruct path:  (db:dbfile-path #f) ;; (configf:lookup *configdat* "setup" "linktree") 
					   local: #t))
	 (testdat       (db:get-test-info-by-id dbstruct run-id test-id))
	 (db-mod-time   0) ;; (file-modification-time db-path))
	 (last-update   0) ;; (current-seconds))
	 (request-update #t))
    (if (not testdat)
	(begin
	  (debug:print 2 "ERROR: No test data found for test " test-id ", exiting")
	  (exit 1))
	(let* (;; (run-id        (if testdat (db:test-get-run_id testdat) #f))
	       (keydat        (if testdat (db:get-key-val-pairs dbstruct run-id) #f))
	       (rundat        (if testdat (db:get-run-info dbstruct run-id) #f))
	       (runname       (if testdat (db:get-value-by-header (db:get-rows rundat)
								  (db:get-header rundat)
								  "runname") #f))
	       ;; (tdb           (tdb:open-test-db-by-test-id-local dbstruct run-id test-id))
	       ;; These next two are intentional bad values to ensure errors if they should not
	       ;; get filled in properly.
	       (logfile       "/this/dir/better/not/exist")
	       (rundir        (if testdat 
				  (db:test-get-rundir testdat)
				  logfile))
	       ;; (testdat-path  (conc rundir "/testdat.db")) ;; this gets recalculated until found 
	       (teststeps     (if testdat (tests:get-compressed-steps dbstruct run-id test-id) '()))
	       (testfullname  (if testdat (db:test-get-fullname testdat) "Gathering data ..."))
	       (testname      (if testdat (db:test-get-testname testdat) "n/a"))
	       (testmeta      (if testdat 
				  (let ((tm (db:testmeta-get-record dbstruct testname)))
				    (if tm tm (make-db:testmeta)))
				  (make-db:testmeta)))

	       (keystring  (string-intersperse 
			    (map (lambda (keyval)
				   ;; (conc ":" (car keyval) " " (cadr keyval)))
				   (cadr keyval))

						  (conc "-e " (get-environment-variable "SHELL"))
						  "")))
				   (system (conc "cd " rundir 
						 ";mt_xterm -T \"" (string-translate testfullname "()" "  ") "\" " shell "&")))
				 (message-window  (conc "Directory " rundir " not found")))))
	       (widgets    (make-hash-table))
	       (refreshdat (lambda ()
			     (let* ((curr-mod-time (file-modification-time db-path))
				                   ;;     (max ..... (if (file-exists? testdat-path)
						   ;;      	      (file-modification-time testdat-path)
						   ;;      	      (begin
						   ;;      		(set! testdat-path (conc rundir "/testdat.db"))
						   ;;      		0))))
				    (need-update   (or (and (>= curr-mod-time db-mod-time)
							    (> (current-milliseconds)(+ last-update 250))) ;; every half seconds if db touched
						       (> (current-milliseconds)(+ last-update 10000))     ;; force update even 10 seconds
						       request-update))
				    (newtestdat (if need-update 
						    ;; NOTE: BUG HIDER, try to eliminate this exception handler
						    (handle-exceptions
						     exn 
						     (debug:print-info 0 "test db access issue in examine test for run-id " run-id ", test-id " test-id ": " ((condition-property-accessor 'exn 'message) exn))

						     (db:get-test-info-by-id dbstruct run-id test-id )))))












			       ;; (debug:print-info 0 "need-update= " need-update " curr-mod-time = " curr-mod-time)




			       (cond
				((and need-update newtestdat)
				 (set! testdat newtestdat)
				 (set! teststeps    (tests:get-compressed-steps dbstruct run-id test-id))
				 (set! logfile      (conc (db:test-get-rundir testdat) "/" (db:test-get-final_logf testdat)))
				 (set! rundir       ;; (filedb:get-path *fdb* 
				       (db:test-get-rundir testdat)) ;; )
				 (set! testfullname (db:test-get-fullname testdat))
				 ;; (debug:print 0 "INFO: teststeps=" (intersperse teststeps "\n    "))
				 
				 ;; I don't see why this was implemented this way. Please comment it ...
				 ;; (if (eq? curr-mod-time db-mod-time) ;; do only once if same
				 ;;     (set! db-mod-time (+ curr-mod-time 1))
				 ;;     (set! db-mod-time curr-mod-time))

			     (conc "megatest -target " keystring " -runname "  runname 
				   " -set-state-status KILLREQ,n/a -testpatt %/% "
				   " -state RUNNING"))))
	       (run-test  (lambda (x)
			    (iup:attribute-set! 
			     command-text-box "VALUE"
			     (conc "megatest -target " keystring " -runname " runname 
				   " -run -testpatt " (conc testname "/" (if (equal? item-path "")
									"%" 
									item-path))
				   ))))
	       (remove-test (lambda (x)
			      (iup:attribute-set!
			       command-text-box "VALUE"
			       (conc "megatest -remove-runs -target " keystring " -runname " runname

	       (remove-test (lambda (x)
			      (iup:attribute-set!
			       command-text-box "VALUE"
			       (conc "megatest -remove-runs -target " keystring " -runname " runname
				     " -testpatt " (conc testname "/" (if (equal? item-path "")
									  "%"
									  item-path))
				     " -v"))))
	       (archive-test  (lambda (x)
				(iup:attribute-set! 
				 command-text-box "VALUE"
				 (conc "megatest -target " keystring " -runname " runname 
				       " -archive save-remove -testpatt " (conc testname "/" (if (equal? item-path "")
												 "%" 
												 item-path))
				       )))))
	  (cond
	   ((not testdat)(begin (print "ERROR: bad test info for " test-id)(exit 1)))
	   ((not rundat)(begin (print "ERROR: found test info but there is a problem with the run info for " run-id)(exit 1)))
	   (else
	    ;;  (test-set-status! db run-id test-name state status itemdat)
	    (set! self ; 
		  (iup:dialog #:close_cb (lambda (a)(exit)) ; #:expand "YES"
			      #:title testfullname
			      (iup:vbox ; #:expand "YES"
			       ;; The run and test info
			       (iup:hbox  ; #:expand "YES"
				(run-info-panel dbstruct keydat testdat runname)
				(test-info-panel testdat store-label widgets)
				(test-meta-panel testmeta store-meta))
			       (host-info-panel testdat store-label)
			       ;; The controls
			       (iup:frame #:title "Actions" 
					  (iup:vbox
					   (iup:hbox 
					    (iup:button "View Log"      #:action viewlog      #:size "80x")
					    (iup:button "Start Xterm"   #:action xterm        #:size "80x")
					    (iup:button "Run Test"      #:action run-test     #:size "80x")
					    (iup:button "Clean Test"    #:action remove-test  #:size "80x")
					    (iup:button "CleanRunExecute!"    #:action clean-run-execute #:size "80x")
					    (iup:button "Kill All Jobs" #:action kill-jobs    #:size "80x")
					    (iup:button "Archive Test"  #:action archive-test #:size "80x")
					    (iup:button "Close"         #:action (lambda (x)(exit)) #:size "80x"))
					   (apply 
					    iup:hbox
					    (list command-text-box command-launch-button))))
			       (set-fields-panel dbstruct run-id test-id testdat)
			       (let ((tabs 
				      (iup:tabs
				       ;; Replace here with matrix
				       (let ((steps-matrix (iup:matrix
							    #:font   "Courier New, -8"
							    #:expand "YES"
							    #:scrollbar "YES"

					 (iup:attribute-set! steps-matrix "0:5" "Duration")
					 (iup:attribute-set! steps-matrix "0:6" "Log File")
					 (iup:attribute-set! steps-matrix "ALIGNMENT1" "ALEFT")
					 ;; (iup:attribute-set! steps-matrix "FIXTOTEXT" "C1")
					 (iup:attribute-set! steps-matrix "RESIZEMATRIX" "YES")
					 (let ((proc
						(lambda (testdat)



						  (dcommon:populate-steps teststeps steps-matrix))))































					   (hash-table-set! widgets "StepsMatrix" proc)
					   (proc testdat))
					 steps-matrix)
				       ;; populate the Test Data panel
				       (iup:frame
					#:title "Test Data"
					(let ((test-data

											      (db:test-data-get-value    x)
											      (db:test-data-get-expected x)
											      (db:test-data-get-tol      x)
											      (db:test-data-get-status   x)
											      (db:test-data-get-units    x)
											      (db:test-data-get-type     x)
											      (db:test-data-get-comment  x)))
										    (db:read-test-data dbstruct run-id test-id "%")))
									      "\n")))
							       (if (not (equal? currval newval))
								   (iup:attribute-set! test-data "VALUE" newval ))))) ;; "TITLE" newval)))))
					  test-data))
				       ;;(dashboard:run-controls)
				       )))
				 (iup:attribute-set! tabs "TABTITLE0" "Steps")

;; (declare (uses dashboard-main))
(declare (uses megatest-version))
(declare (uses mt))

(include "common_records.scm")
(include "db_records.scm")
(include "run_records.scm")
(include "megatest-fossil-hash.scm")

(define help (conc 
"Megatest Dashboard, documentation at http://www.kiatoa.com/fossils/megatest
  version " megatest-version "
  license GPL, Copyright (C) Matt Welland 2012-2014

Usage: dashboard [options]
  -h                   : this help
  -server host:port    : connect to host:port instead of db access
  -test run-id,test-id : control test identified by testid
  -guimonitor          : control panel for runs

Misc
  -rows N         : set number of rows
"))

;; process args
(define remargs (args:get-args 
		 (argv)
		 (list  "-rows"
			"-run"
			"-test"
			"-debug"
			"-host" 
			"-transport"
			) 
		 (list  "-h"
			"-use-server"
			"-guimonitor"
			"-main"
			"-v"
			"-q"

      (exit)))

(if (not (launch:setup-for-run))
    (begin
      (print "Failed to find megatest.config, exiting") 
      (exit 1)))

(define *useserver* (or (args:get-arg "-use-server")
			(configf:lookup *configdat* "dashboard" "use-server")))


(define *dbdir* (db:dbfile-path #f)) ;; (conc (configf:lookup *configdat* "setup" "linktree") "/.db"))
(define *dbstruct-local*  (make-dbr:dbstruct path:  *dbdir*
					     local: #t))
(define *db-file-path* (db:dbfile-path 0))



;; HACK ALERT: this is a hack, please fix.
(define *read-only* (not (file-read-access? *db-file-path*)))


(define toplevel #f)
(define dlg      #f)
(define max-test-num 0)
(define *keys*   (if *useserver*
		     (rmt:get-keys)
		     (db:get-keys *dbstruct-local*)))

(define *dbkeys*  (append *keys* (list "runname")))

(define *header*       #f)
(define *allruns*     '())
(define *allruns-by-id* (make-hash-table)) ;; 
(define *runchangerate* (make-hash-table))

(define *buttondat*    (make-hash-table)) ;; <run-id color text test run-key>
(define *alltestnamelst* '())
(define *searchpatts*  (make-hash-table))
(define *num-runs*      8)
(define *tot-run-count* (if *useserver*
			    (rmt:get-num-runs "%")
			    (db:get-num-runs *dbstruct-local* "%")))

;; (define *tot-run-count* (db:get-num-runs *dbstruct-local* "%"))

;; Update management
;;
(define *last-update*   (current-seconds))
(define *last-db-update-time* 0)
(define *please-update-buttons* #t)
(define *delayed-update* 0)
(define *update-is-running* #f)
(define *update-mutex* (make-mutex))

(define *all-item-test-names* '())
(define *num-tests*     15)
(define *start-run-offset*  0)
(define *start-test-offset* 0)
(define *examine-test-dat* (make-hash-table))
(define *exit-started* #f)
(define *status-ignore-hash* (make-hash-table))
(define *state-ignore-hash*  (make-hash-table))



(define *tests-sort-options* (vector (vector "Sort +a" 'testname   "ASC")
				     (vector "Sort -a" 'testname   "DESC")
				     (vector "Sort +t" 'event_time "ASC")
				     (vector "Sort -t" 'event_time "DESC")
				     (vector "Sort +s" 'statestatus "ASC")
				     (vector "Sort -s" 'statestatus "DESC")
				     (vector "Sort +a" 'testname   "ASC")))

	 (c2    (map string->number (string-split color2)))
	 (delta (map (lambda (a b)(abs (- a b))) c1 c2)))
    (null? (filter (lambda (x)(> x 3)) delta))))

;; keypatts: ( (KEY1 "abc%def")(KEY2 "%") )
(define (update-rundat runnamepatt numruns testnamepatt keypatts)
  (let* ((referenced-run-ids '())
	 (allruns     (if *useserver*
			  (rmt:get-runs runnamepatt numruns *start-run-offset* keypatts)
			  (db:get-runs *dbstruct-local* runnamepatt numruns ;; (+ numruns 1) ;; (/ numruns 2))
				      *start-run-offset* keypatts)))
	 (header      (db:get-header allruns))
	 (runs        (db:get-rows   allruns))
	 (result      '())
	 (maxtests    0)
	 (states      (hash-table-keys *state-ignore-hash*))
	 (statuses    (hash-table-keys *status-ignore-hash*))
	 (sort-info   (get-curr-sort))
	 (sort-by     (vector-ref sort-info 1))
	 (sort-order  (vector-ref sort-info 2))
	 (bubble-type (if (member sort-order '(testname))
			  'testname
			  'itempath)))
    ;; 
    ;; trim runs to only those that are changing often here
    ;; 
    (for-each (lambda (run)
		(let* ((run-id      (db:get-value-by-header run header "id"))
		       (tests       (if *useserver*
					(rmt:get-tests-for-run run-id testnamepatt states statuses
							       #f #f
							       *hide-not-hide*
							       sort-by
							       sort-order
							       'shortlist)
					(db:get-tests-for-run *dbstruct-local* run-id testnamepatt states statuses
							      #f #f
							      *hide-not-hide*
							      sort-by
							      sort-order
							      'shortlist)))
		       ;; NOTE: bubble-up also sets the global *all-item-test-names*
		       ;; (tests       (bubble-up tmptests priority: bubble-type))
		       (key-vals    (if *useserver* 
					(rmt:get-key-vals run-id)
					(db:get-key-vals *dbstruct-local* run-id))))
		  ;; NOTE: 11/01/2013 This routine is *NOT* getting called excessively.
		  ;; (debug:print 0 "Getting data for run " run-id " with key-vals=" key-vals)
		  ;; Not sure this is needed?
		  (set! referenced-run-ids (cons run-id referenced-run-ids))
		  (if (> (length tests) maxtests)
		      (set! maxtests (length tests)))
		  (if (or (not *hide-empty-runs*) ;; this reduces the data burden when set

					     (car matching))))
			   (testname   (db:test-get-testname  test))
			   (itempath   (db:test-get-item-path test))
			   (testfullname (test:test-get-fullname test))
			   (teststatus (db:test-get-status   test))
			   (teststate  (db:test-get-state    test))
			   ;;(teststart  (db:test-get-event_time test))
			   ;;(runtime    (db:test-get-run_duration test))
			   (buttontxt  (cond
					((member teststate '("COMPLETED" "ARCHIVED")) teststatus)
					((and (equal? teststate "NOT_STARTED")
					      (member teststatus '("ZERO_ITEMS" "BLOCKED" "PREQ_FAIL" "PREQ_DISCARDED" "TIMED_OUT" "KEEP_TRYING" "TEN_STRIKES")))
					 teststatus)
					(else
					 teststate)))
			   (button     (vector-ref columndat rown))
			   (color      (car (gutils:get-color-for-state-status teststate teststatus)))

	  (if (not (null? values))
	      (let ((newval (car values)))
		(iup:attribute-set! lb "VALUE" newval)
		newval))))))

(define (dashboard:update-target-selector key-lbs #!key (action-proc #f))
  (let* ((runconf-targs (common:get-runconfig-targets))
	 (db-target-dat (if *useserver* 
			    (rmt:get-targets)
			    (db:get-targets *dbstruct-local*)))
	 (header        (vector-ref db-target-dat 0))
	 (db-targets    (vector-ref db-target-dat 1))
	 (all-targets   (append db-targets
				(map (lambda (x)
				       (list->vector
					(take (append (string-split x "/")
						      (make-list (length header) "na"))

				 #:dropdown "YES"
				 #:action (lambda (obj val index lbstate)
					    (iup:attribute-set! tb "VALUE" val)
					    (dboard:data-set-run-name! *data* val)
					    (dashboard:update-run-command))))
		(refresh-runs-list (lambda ()
				     (let* ((target        (dboard:data-get-target-string *data*))
					    (runs-for-targ (if *useserver*
							       (rmt:get-runs-by-patt *keys* "%" target #f #f #f)
							       (db:get-runs-by-patt *dbstruct-local* *keys* "%" target #f #f #f)))
					    (runs-header   (vector-ref runs-for-targ 0))
					    (runs-dat      (vector-ref runs-for-targ 1))
					    (run-names     (cons default-run-name 
								 (map (lambda (x)
									(db:get-value-by-header x runs-header "runname"))
								      runs-dat))))
				       (iup:attribute-set! lb "REMOVEITEM" "ALL")

		    combos)))
	  (iup:hbox
	   ;; Text box for STATES
	   (iup:frame
	    #:title "States"
	    (dashboard:text-list-toggle-box 
	     ;; Move these definitions to common and find the other useages and replace!
	     (map cadr *common:std-states*) ;; '("COMPLETED" "RUNNING" "STUCK" "INCOMPLETE" "LAUNCHED" "REMOTEHOSTSTART" "KILLED")
	     (lambda (all)
	       (dboard:data-set-states! *data* all)
	       (dashboard:update-run-command))))
	   ;; Text box for STATES
	   (iup:frame
	    #:title "Statuses"
	    (dashboard:text-list-toggle-box 
	     (map cadr *common:std-statuses*) ;; '("PASS" "FAIL" "n/a" "CHECK" "WAIVED" "SKIP" "DELETED" "STUCK/DEAD")
	     (lambda (all)
	       (dboard:data-set-statuses! *data* all)
	       (dashboard:update-run-command))))))))
      
       (iup:frame
	#:title "Tests and Tasks"
	(let* ((updater #f)

;;      )))

;;======================================================================
;; S U M M A R Y 
;;======================================================================
;;
;; General info about the run(s) and megatest area
(define (dashboard:summary db)
  (let ((rawconfig        (read-config (conc *toppath* "/megatest.config") #f #f))) ;; changed to #f since I want #{} to be expanded by [system ...] to NOT be expanded. WAS: 'return-string)))
    (iup:vbox
     (iup:split
      #:value 500
      (iup:frame 
       #:title "General Info"
       (iup:vbox
	(iup:hbox
	 (iup:label "Area Path")
	 (iup:textbox #:value *toppath* #:expand "HORIZONTAL"))
	(iup:hbox 
	 (dcommon:keys-matrix rawconfig)
	 (dcommon:general-info)
	 )))
      (iup:frame
       #:title "Server"
       (dcommon:servers-table)))
     (iup:frame 
      #:title "Megatest config settings"
      (iup:hbox
       (dcommon:section-matrix rawconfig "setup" "Varname" "Value")
       (iup:vbox
	(dcommon:section-matrix rawconfig "server" "Varname" "Value")
	;; (iup:frame
	;; #:title "Disks Areas"
	(dcommon:section-matrix rawconfig "disks" "Disk area" "Path"))))
     (iup:frame
      #:title "Run statistics"
      (dcommon:run-stats db)))))

;;======================================================================
;; R U N
;;======================================================================
;;
;; display and manage a single run at a time

(define (tree-path->run-id path)
  (if (not (null? path))
      (hash-table-ref/default (dboard:data-get-path-run-ids *data*) path #f)
      #f))

(define dashboard:update-run-summary-tab #f)

;; (define (tests window-id)
(define (dashboard:one-run db)
  (let* ((tb      (iup:treebox
		   #:value 0
		   #:name "Runs"
		   #:expand "YES"
		   #:addexpanded "NO"
		   #:selection-cb
		   (lambda (obj id state)
		     ;; (print "obj: " obj ", id: " id ", state: " state)
		     (let* ((run-path (tree:node->path obj id))
			    (run-id   (tree-path->run-id (cdr run-path))))
		       (if (number? run-id)
			   (begin
			     (dboard:data-set-curr-run-id! *data* run-id)
			     (dashboard:update-run-summary-tab))
			   (debug:print 0 "ERROR: tree-path->run-id returned non-number " run-id)))
		       ;; (print "path: " (tree:node->path obj id) " run-id: " run-id)
		       )))
	 (cell-lookup (make-hash-table))
	 (run-matrix (iup:matrix
		      #:expand "YES"
		      #:click-cb
		      (lambda (obj lin col status)
			(let* ((toolpath (car (argv)))
			       (key      (conc lin ":" col))
			       (test-id  (hash-table-ref/default cell-lookup key -1))
			       (cmd      (conc toolpath " -test " (dboard:data-get-curr-run-id *data*) "," test-id "&")))
			  (system cmd)))))
	 (updater  (lambda ()
		     (let* ((runs-dat     (db:get-runs-by-patt db *keys* "%" #f #f #f #f))
			    (runs-header  (vector-ref runs-dat 0)) ;; 0 is header, 1 is list of records
			    (run-id       (dboard:data-get-curr-run-id *data*))
			    (tests-dat    (let ((tdat (db:get-tests-for-run db run-id 
									    (hash-table-ref/default *searchpatts* "test-name" "%/%")
									    (hash-table-keys *state-ignore-hash*) ;; '()
									    (hash-table-keys *status-ignore-hash*) ;; '()
									    #f #f
									    *hide-not-hide*
									    #f #f
									    "id,testname,item_path,state,status"))) ;; get 'em all
					    (sort tdat (lambda (a b)
							 (let* ((aval (vector-ref a 2))
								(bval (vector-ref b 2))
								(anum (string->number aval))
								(bnum (string->number bval)))
							   (if (and anum bnum)
							       (< anum bnum)

     tb
     run-matrix)))

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

(define (make-dashboard-buttons db nruns ntests keynames)
  (let* ((nkeys   (length keynames))
	 (runsvec (make-vector nruns))
	 (header  (make-vector nruns))
	 (lftcol  (make-vector ntests))
	 (keycol  (make-vector ntests))
	 (controls '())
	 (lftlst  '())

	      (let ((hideit (iup:button "HideTests" #:action (lambda (obj)
							       (set! *hide-not-hide* (not *hide-not-hide*))
							       (iup:attribute-set! obj "TITLE" (if *hide-not-hide* "HideTests" "NotHide"))
							       (mark-for-update)))))
		(set! *hide-not-hide-button* hideit)
		hideit))
	     (iup:hbox
	      (iup:button "Quit"      #:action (lambda (obj)
						 ;; (if *dbstruct-local* (db:close-all *dbstruct-local*))
						 (exit)))
	      (iup:button "Refresh"   #:action (lambda (obj)
						 (mark-for-update)))
	      (iup:button "Collapse"  #:action (lambda (obj)
						 (let ((myname (iup:attribute obj "TITLE")))
						   (if (equal? myname "Collapse")
						       (begin
							 (for-each (lambda (tname)

	      (map (lambda (status)
		     (iup:toggle status  #:action   (lambda (obj val)
						      (mark-for-update)
						      (if (eq? val 1)
							  (hash-table-set! *status-ignore-hash* status #t)
							  (hash-table-delete! *status-ignore-hash* status))
						      (set-bg-on-filter))))
		   (map cadr *common:std-statuses*))) ;; '("PASS" "FAIL" "WARN" "CHECK" "WAIVED" "STUCK/DEAD" "n/a" "SKIP")))
	     (apply 
	      iup:hbox
	      (map (lambda (state)
		     (iup:toggle state   #:action   (lambda (obj val)
						      (mark-for-update)
						      (if (eq? val 1)
							  (hash-table-set! *state-ignore-hash* state #t)
							  (hash-table-delete! *state-ignore-hash* state))
						      (set-bg-on-filter))))
		   (map cadr *common:std-states*))) ;; '("RUNNING" "COMPLETED" "INCOMPLETE" "LAUNCHED" "NOT_STARTED" "KILLED" "DELETED")))
	     (iup:valuator #:valuechanged_cb (lambda (obj)
					       (let ((val (inexact->exact (round (/ (string->number (iup:attribute obj "VALUE")) 10))))
						     (oldmax   (string->number (iup:attribute obj "MAX")))
						     (maxruns  *tot-run-count*))
						 (set! *start-run-offset* val)
						 (mark-for-update)
						 (debug:print 6 "*start-run-offset* " *start-run-offset* " maxruns: " maxruns ", val: " val " oldmax: " oldmax)

				       #:size "60x15" 
				       #:expand "HORIZONTAL"
				       #:fontsize "10" 
				       #:action (lambda (x)
						  (let* ((toolpath (car (argv)))
							 (buttndat (hash-table-ref *buttondat* button-key))
							 (test-id  (db:test-get-id (vector-ref buttndat 3)))
							 (run-id   (db:test-get-run_id (vector-ref buttndat 3)))
							 (cmd  (conc toolpath " -test " run-id "," test-id "&")))
					;(print "Launching " cmd)
						    (system cmd))))))
	  (hash-table-set! *buttondat* button-key (vector 0 "100 100 100" button-key #f #f)) 
	  (vector-set! testvec testnum butn)
	  (loop runnum (+ testnum 1) testvec (cons butn res))))))
    ;; now assemble the hdrlst and bdylst and kick off the dialog
    (iup:show

					(apply iup:hbox (reverse hdrlst))
					(apply iup:hbox (reverse bdylst))))))
			 controls))
	     (tabs (iup:tabs
		    #:tabchangepos-cb (lambda (obj curr prev)
					(set! *please-update-buttons* #t)
					(set! *current-tab-number* curr))
		    (dashboard:summary db)
		    runs-view
		    (dashboard:one-run db)
		    (dashboard:run-controls)
		    )))
	;; (set! (iup:callback tabs tabchange-cb:) (lambda (a b c)(print "SWITCHED TO TAB: " a " " b " " c)))
	(iup:attribute-set! tabs "TABTITLE0" "Summary")
	(iup:attribute-set! tabs "TABTITLE1" "Runs")
	(iup:attribute-set! tabs "TABTITLE2" "Run Summary")
	(iup:attribute-set! tabs "TABTITLE3" "Run Control")

(define *tim* (iup:timer))
(define *ord* #f)
(iup:attribute-set! *tim* "TIME" 300)
(iup:attribute-set! *tim* "RUN" "YES")

;; Move this stuff to db.scm? I'm not sure that is the right thing to do...
;;
(define *last-db-update-time* (file-modification-time *db-file-path*)) ;; (conc *toppath* "/db/main.db")))
(define *last-recalc-ended-time* 0)

(define (dashboard:been-changed)
  (> (file-modification-time *db-file-path*) *last-db-update-time*))

(define (dashboard:set-db-update-time)
  (set! *last-db-update-time* (file-modification-time *db-file-path*)))

(define (dashboard:recalc modtime please-update-buttons last-db-update-time)
  (or please-update-buttons
      (and (> (current-milliseconds)(+ *last-recalc-ended-time* 150))
	   (> modtime last-db-update-time)
	   (> (current-seconds)(+ last-db-update-time 1)))))

(define *monitor-db-path* (conc *dbdir* "/monitor.db"))
(define *last-monitor-update-time* 0)

;; Force creation of the db in case it isn't already there.
(tasks:open-db)

(define (dashboard:get-youngest-run-db-mod-time)
  (handle-exceptions
   exn
   (begin
     (debug:print 0 "WARNING: error in accessing databases in get-youngest-run-db-mod-time: " ((condition-property-accessor 'exn 'message) exn))
     (current-seconds)) ;; something went wrong - just print an error and return current-seconds
   (apply max (map (lambda (filen)
		     (file-modification-time filen))
		   (glob (conc *dbdir* "/*.db"))))))

(define (dashboard:run-update x)
  (let* ((modtime         (dashboard:get-youngest-run-db-mod-time)) ;; (file-modification-time *db-file-path*))
	 (monitor-modtime (if (file-exists? *monitor-db-path*)
			      (file-modification-time *monitor-db-path*)
			      -1))
	 (run-update-time (current-seconds))
	 (recalc          (dashboard:recalc modtime *please-update-buttons* *last-db-update-time*)))
    (if (and (eq? *current-tab-number* 0)
	     (or (> monitor-modtime *last-monitor-update-time*)
		 (> (- run-update-time *last-monitor-update-time*) 5))) ;; update every 1/2 minute just in case
	(begin
	  (set! *last-monitor-update-time* run-update-time) ;; monitor-modtime)
	  (if dashboard:update-servers-table (dashboard:update-servers-table))))
    (if recalc
	(begin	
	  (case *current-tab-number* 
	    ((0) 
	     (if dashboard:update-summary-tab (dashboard:update-summary-tab)))
	    ((1) ;; The runs table is active

(cond 
 ((args:get-arg "-run")
  (let ((runid (string->number (args:get-arg "-run"))))
    (if runid
	(begin
	  (lambda (x)
	    (on-exit std-exit-procedure)

	    (examine-run *dbstruct-local* runid)))
	(begin
	  (print "ERROR: runid is not a number " (args:get-arg "-run"))
	  (exit 1)))))
 ((args:get-arg "-test") ;; run-id,test-id
  (let* ((dat     (let ((d (map string->number (string-split (args:get-arg "-test") ","))))
		    (if (> (length d) 1)
			d
			(list #f #f))))
	 (run-id  (car dat))
	 (test-id (cadr dat)))
    (if (and (number? run-id)
	     (number? test-id)
	     (>= test-id 0))
	(examine-test run-id test-id)
	(begin
	  (debug:print 3 "INFO: tried to open test with invalid run-id,test-id. " (args:get-arg "-test"))
	  (exit 1)))))
 ((args:get-arg "-guimonitor")
  (gui-monitor *dbstruct-local*))
 (else
  (set! uidat (make-dashboard-buttons *dbstruct-local* *num-runs* *num-tests* *dbkeys*))
  (iup:callback-set! *tim*
		     "ACTION_CB"
		     (lambda (x)
		       (let ((update-is-running #f))
			 (mutex-lock! *update-mutex*)
			 (set! update-is-running *update-is-running*)
			 (if (not update-is-running)

			  ;;   (if (not *please-update-buttons*)
			  ;;       (loop))))))
      (th2 (make-thread iup:main-loop "Main loop")))
  (thread-start! th1)
  (thread-start! th2)
  (thread-join! th2))

;; (iup:main-loop)(db:close-all *dbstruct-local*)

# Read in the users vars first (so the offical data cannot be overridden
[include ~/.datashare.config]

# Read in local overrides
[include datashare.config]

# Replace [storage] with settings entry - more secure
[settings]

storage /tmp/#{getenv USER}/datashare/disk1 \
        /tmp/#{getenv USER}/datashare/disk2

basepath #{scheme (or (getenv "BASEPATH") "/tmp/#{getenv USER}")}

[areas]
synthesis  asic/synthesis
verilog    asic/verilog
customlibs custom/oalibs
megatest   tools/megatest

[quality]
0 untested
1 lightly tested
2 tested
3 full QA

[database]
location /tmp/#{getenv USER}/datashare

[pathmaps]
SHELF /tmp/#{getenv USER}/theshelf

[buildmethods]
customlibs make setup;make install

(use canvas-draw)
(import canvas-draw-iup)

(use sqlite3 srfi-1 posix regex regex-case srfi-69)
(import (prefix sqlite3 sqlite3:))

(declare (uses configf))
(declare (uses tree))
(declare (uses margs))
;; (declare (uses dcommon))
;; (declare (uses launch))
;; (declare (uses gutils))
;; (declare (uses db))
;; (declare (uses synchash))
;; (declare (uses server))
;; (declare (uses megatest-version))
;; (declare (uses tbd))

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

;;
;; GLOBALS
;;
(define *datashare:current-tab-number* 0)
(define *args-hash* (make-hash-table))
(define datashare:help (conc "Usage: datashare [action [params ...]]

Note: run datashare without parameters to start the gui.

  list-areas                          : List the allowed areas

  list-versions <area>                : List versions available in <area>
         options : -full, -vpatt patt

  publish <path> <area> <version>     : Publish data for area and with version

  get <area> <version>                : Get a link to data, put the link in destpath
         options : -i iteration

  update <area>                       : Update the link to data to the latest iteration.

Part of the Megatest tool suite.
Learn more at http://www.kiatoa.com/fossils/megatest

Version: " megatest-fossil-hash)) ;; "

;;======================================================================
;; RECORDS
;;======================================================================

;; make-vector-record "testing" datastore pkg id area version_name store_type copied source_path iteration submitter datetime storegrp datavol quality disk_id comment
;; testing
(define (make-datashare:pkg)(make-vector 15))
(define-inline (datashare:pkg-get-id             vec)    (vector-ref  vec 0))
(define-inline (datashare:pkg-get-area           vec)    (vector-ref  vec 1))
(define-inline (datashare:pkg-get-version_name   vec)    (vector-ref  vec 2))
(define-inline (datashare:pkg-get-store_type     vec)    (vector-ref  vec 3))
(define-inline (datashare:pkg-get-copied         vec)    (vector-ref  vec 4))
(define-inline (datashare:pkg-get-source_path    vec)    (vector-ref  vec 5))
(define-inline (datashare:pkg-get-iteration      vec)    (vector-ref  vec 6))
(define-inline (datashare:pkg-get-submitter      vec)    (vector-ref  vec 7))
(define-inline (datashare:pkg-get-datetime       vec)    (vector-ref  vec 8))
(define-inline (datashare:pkg-get-storegrp       vec)    (vector-ref  vec 9))
(define-inline (datashare:pkg-get-datavol        vec)    (vector-ref  vec 10))
(define-inline (datashare:pkg-get-quality        vec)    (vector-ref  vec 11))
(define-inline (datashare:pkg-get-disk_id        vec)    (vector-ref  vec 12))
(define-inline (datashare:pkg-get-comment        vec)    (vector-ref  vec 13))
(define-inline (datashare:pkg-get-stored_path    vec)    (vector-ref  vec 14))
(define-inline (datashare:pkg-set-id!            vec val)(vector-set! vec 0 val))
(define-inline (datashare:pkg-set-area!          vec val)(vector-set! vec 1 val))
(define-inline (datashare:pkg-set-version_name!  vec val)(vector-set! vec 2 val))
(define-inline (datashare:pkg-set-store_type!    vec val)(vector-set! vec 3 val))
(define-inline (datashare:pkg-set-copied!        vec val)(vector-set! vec 4 val))
(define-inline (datashare:pkg-set-source_path!   vec val)(vector-set! vec 5 val))
(define-inline (datashare:pkg-set-iteration!     vec val)(vector-set! vec 6 val))
(define-inline (datashare:pkg-set-submitter!     vec val)(vector-set! vec 7 val))
(define-inline (datashare:pkg-set-datetime!      vec val)(vector-set! vec 8 val))
(define-inline (datashare:pkg-set-storegrp!      vec val)(vector-set! vec 9 val))
(define-inline (datashare:pkg-set-datavol!       vec val)(vector-set! vec 10 val))
(define-inline (datashare:pkg-set-quality!       vec val)(vector-set! vec 11 val))
(define-inline (datashare:pkg-set-disk_id!       vec val)(vector-set! vec 12 val))
(define-inline (datashare:pkg-set-comment!       vec val)(vector-set! vec 13 val))
(define-inline (datashare:pkg-set-stored_path!   vec val)(vector-set! vec 14 val))

;;======================================================================
;; DB
;;======================================================================

(define (datashare:initialize-db db)
  (for-each
   (lambda (qry)
     (sqlite3:execute db qry))
   (list 
    "CREATE TABLE pkgs 
         (id           INTEGER PRIMARY KEY,
          area         TEXT,
          version_name TEXT,
          store_type   TEXT DEFAULT 'copy',
          copied       INTEGER DEFAULT 0,
          source_path  TEXT,
          stored_path  TEXT,
          iteration    INTEGER DEFAULT 0,
          submitter    TEXT,
          datetime     TIMESTAMP DEFAULT (strftime('%s','now')),
          storegrp     TEXT,
          datavol      INTEGER,
          quality      TEXT,
          disk_id      INTEGER,
          comment      TEXT);"
    "CREATE TABLE refs
         (id        INTEGER PRIMARY KEY,
          pkg_id    INTEGER,
          destlink  TEXT);"
    "CREATE TABLE disks
         (id         INTEGER PRIMARY KEY,
          storegrp   TEXT,
          path       TEXT);")))

(define (datashare:register-data db area version-name store-type submitter quality source-path comment)
  (let ((iter-qry       (sqlite3:prepare db "SELECT max(iteration) FROM pkgs WHERE area=? AND version_name=?;"))
	(next-iteration 0))
    (sqlite3:with-transaction
     db
     (lambda ()
       (sqlite3:for-each-row
	(lambda (iteration)
	  (if (and (number? iteration)
		   (>= iteration next-iteration))
	      (set! next-iteration (+ iteration 1))))
	iter-qry area version-name)
       ;; now store the data
       (sqlite3:execute db "INSERT INTO pkgs (area,version_name,iteration,store_type,submitter,source_path,quality,comment) 
                                 VALUES (?,?,?,?,?,?,?,?);"
			area version-name next-iteration (conc store-type) submitter source-path quality comment)))
    (sqlite3:finalize! iter-qry)
    next-iteration))

(define (datashare:get-id db area version-name iteration)
  (let ((res #f))
    (sqlite3:for-each-row
     (lambda (id)
       (set! res id))
     db
     "SELECT id FROM pkgs WHERE area=? AND version_name=? AND iteration=?;"
     area version-name iteration)
    res))

(define (datashare:set-stored-path db id path)
  (sqlite3:execute db "UPDATE pkgs SET stored_path=? WHERE id=?;" path id))

(define (datashare:set-copied db id value)
  (sqlite3:execute db "UPDATE pkgs SET copied=? WHERE id=?;" value id))
  
(define (datashare:get-pkg-record db area version-name iteration)
  (let ((res #f))
    (sqlite3:for-each-row
     (lambda (a . b)
       (set! res (apply vector a b)))
     db 
     "SELECT * FROM pkgs WHERE area=? AND version_name=? AND iteration=?;"
     area 
     version-name
     iteration)
    res))

;; take version-name iteration and register or update "lastest/0"
;;
(define (datashare:set-latest db id area version-name iteration)
  (let* ((rec         (datashare:get-pkg-record db area version-name iteration))
	 (latest-id   (datashare:get-id db area "latest" 0))
	 (stored-path (datashare:pkg-get-stored_path rec)))
    (if latest-id ;; have a record - bump the link pointer
	(datashare:set-stored-path db latest-id stored-path)
	(datashare:register-data db area "latest" 'link "auto" "na" stored-path "latest data"))))

;; set a package ref, this is the location where the link back to the stored data 
;; is put. 
;;
;; if there is nothing at that location then the record can be removed
;; if there are no refs for a particular pkg-id then that pkg-id is a 
;; candidate for removal
;;
(define (datashare:record-pkg-ref db pkg-id dest-link)
  (sqlite3:execute db "INSERT INTO refs (pkg_id,destlink) VALUES (?,?);" pkg-id dest-link))
  
(define (datashare:count-refs db pkg-id)
  (let ((res 0))
    (sqlite3:for-each-row
     (lambda (count)
       (set! res count))
     db
     "SELECT count(id) FROM refs WHERE pkg_id=?;"
     pkg-id)
    res))

;; Create the sqlite db
(define (datashare:open-db configdat) 
  (let ((path (configf:lookup configdat "database" "location")))
    (if (and path
	     (directory? path)
	     (file-read-access? path))
	(let* ((dbpath    (conc path "/datashare.db"))
	       (writeable (file-write-access? dbpath))
	       (dbexists  (file-exists? dbpath))
	       (handler   (make-busy-timeout 136000)))
	  (handle-exceptions
	   exn
	   (begin
	     (debug:print 2 "ERROR: problem accessing db " dbpath
			  ((condition-property-accessor 'exn 'message) exn))
	     (exit))
	   (set! db (sqlite3:open-database dbpath)))
	  (if *db-write-access* (sqlite3:set-busy-handler! db handler))
	  (if (not dbexists)
	      (begin
		(datashare:initialize-db db)))
	  db)
	(print "ERROR: invalid path for storing database: " path))))

(define (open-run-close-exception-handling proc idb . params)
  (handle-exceptions
   exn
   (let ((sleep-time (random 30))
         (err-status ((condition-property-accessor 'sqlite3 'status #f) exn)))
     (case err-status
       ((busy)
        (thread-sleep! sleep-time))
       (else
        (print "EXCEPTION: database overloaded or unreadable.")
        (print " message: " ((condition-property-accessor 'exn 'message) exn))
        (print "exn=" (condition->list exn))
        (print " status:  " ((condition-property-accessor 'sqlite3 'status) exn))
        (print-call-chain (current-error-port))
        (thread-sleep! sleep-time)
        (print "trying db call one more time....this may never recover, if necessary kill process " (current-process-id) " on host " (get-host-name) " to clean up")))
     (apply open-run-close-exception-handling proc idb params))
   (apply open-run-close-no-exception-handling proc idb params)))

(define (open-run-close-no-exception-handling  proc idb . params)
  ;; (print "open-run-close-no-exception-handling START given a db=" (if idb "yes " "no ") ", params=" params)
  (let* ((db (cond
	      ((sqlite3:database? idb)     idb)
	      ((not idb)                   (print "ERROR: cannot open-run-close with #f anymore"))
	      ((procedure? idb)            (idb))
	      (else                        (print "ERROR: cannot open-run-close with #f anymore"))))
	 (res #f))
    (set! res (apply proc db params))
    (if (not idb)(sqlite3:finalize! dbstruct))
    ;; (print "open-run-close-no-exception-handling END" )
    res))

(define open-run-close open-run-close-no-exception-handling)

(define (datashare:get-pkgs db area-filter version-filter iter-filter)
  (let ((res '()))
    (sqlite3:for-each-row ;; replace with fold ...
     (lambda (a . b)
       (set! res (cons (list->vector (cons a b)) res)))
     db 
     (conc "SELECT id,area,version_name,store_type,copied,source_path,iteration,submitter,datetime,storegrp,datavol,quality,disk_id,comment,stored_path "
	   " FROM pkgs WHERE area like ? AND version_name LIKE ? AND iteration " iter-filter ";")
     area-filter version-filter)
    (reverse res)))

(define (datashare:get-pkg db area-name version-name #!key (iteration #f))
  (let ((dat '())
	(res #f))
    (sqlite3:for-each-row ;; replace with fold ...
     (lambda (a . b)
       (set! dat (cons (list->vector (cons a b)) dat)))
     db 
     (conc "SELECT id,area,version_name,store_type,copied,source_path,iteration,submitter,datetime,storegrp,datavol,quality,disk_id,comment,stored_path "
	   " FROM pkgs WHERE area=? AND version_name=? ORDER BY iteration ASC;")
     area-name version-name)
    ;; now filter for iteration, either max if #f or specific one
    (if (null? dat)
	#f
	(let loop ((hed (car dat))
		   (tal (cdr dat))
		   (cur 0))
	  (let ((itr (datashare:pkg-get-iteration hed)))
	    (if (equal? itr iteration) ;; this is the one if iteration is specified
		hed
		(if (null? tal)
		    hed
		    (loop (car tal)(cdr tal)))))))))

(define (datashare:get-versions-for-area db area-name #!key (version-patt #f))
  (let ((res '())
	(data (make-hash-table)))
    (sqlite3:for-each-row
     (lambda (version-name submitter iteration submitted-time comment)
       ;;                                              0           1         2           3           4
       (hash-table-set! data version-name (vector version-name submitter iteration submitted-time comment)))
     db 
     "SELECT version_name,submitter,iteration,datetime,comment FROM pkgs WHERE area='megatest' AND version_name != 'latest' AND version_name LIKE ? ORDER BY datetime asc;"
     (or version-patt "%"))
    (map (lambda (x)(hash-table-ref data x))(sort (hash-table-keys data) string-ci>=))))

;;======================================================================
;; DATA IMPORT/EXPORT
;;======================================================================

(define (datashare:import-data configdat source-path dest-path area version iteration)
  (let* ((space-avail (car dest-path))
	 (disk-path   (cdr dest-path))
	 (targ-path   (conc disk-path "/" area "/" version "/" iteration))
	 (id          (datashare:get-id db area version iteration))
	 (db          (datashare:open-db configdat)))
    (if (> space-avail 10000) ;; dumb heuristic
	(begin
	  (create-directory targ-path #t)
	  (datashare:set-stored-path db id targ-path)
	  (print "Running command: rsync -av " source-path "/ " targ-path "/")
	  (let ((th1 (make-thread (lambda ()
				    (let ((pid (process-run "rsync" (list "-av" (conc source-path "/") (conc targ-path "/")))))
				      (process-wait pid)
				      (datashare:set-copied db id "yes")
				      (sqlite3:finalize! db)))
				   "Data copy")))
	    (thread-start! th1))
	  #t)
	(begin
	  (print "ERROR: Not enough space in storage area " dest-path)
	  (datashare:set-copied db id "no")
	  (sqlite3:finalize! db)
	  #f))))

(define (datashare:get-areas configdat)
  (let* ((areadat (configf:get-section configdat "areas"))
	 (areas   (if areadat (map car areadat) '())))
    areas))

(define (datashare:publish configdat publish-type area-name version comment spath submitter quality)
  ;; input checks
  (cond 
   ((not (member area-name (datashare:get-areas configdat)))
    (cons #f (conc "Illegal area name \"" area-name "\"")))
   (else
    (let ((db          (datashare:open-db configdat))
	  (iteration   (datashare:register-data db area-name version publish-type submitter quality spath comment))
	  (dest-store  (datashare:get-best-storage configdat)))
      (if iteration
	  (if (eq? 'copy publish-type)
	      (begin
		(datashare:import-data configdat spath dest-store area-name version iteration)
		(let ((id (datashare:get-id db area-name version iteration)))
		  (datashare:set-latest db id area-name version iteration)))
	      (let ((id (datashare:get-id db area-name version iteration)))
		(datashare:set-stored-path db id spath)
		(datashare:set-copied db id "yes")
		(datashare:set-copied db id "n/a")
		(datashare:set-latest db id area-name version iteration)))
	  (print "ERROR: Failed to get an iteration number"))
      (sqlite3:finalize! db)
      (cons #t "Successfully saved data")))))

(define (datashare:get-best-storage configdat)
  (let* ((storage     (configf:lookup configdat "settings" "storage"))
	 (store-areas (if storage (string-split storage) '())))
    (print "Looking for available space in " store-areas)
    (datashare:find-most-space store-areas)))

;; (string->number (list-ref (with-input-from-pipe "df -B1000000 /tmp" (lambda ()(read-line)(string-split (read-line)))) 3))

(define (datashare:find-most-space paths)
  (fold (lambda (area res)
	  ;; (print "area=" area " res=" res)
	  (let ((maxspace (car res))
		(currpath (cdr res)))
	    ;; (print currpath " " maxspace)
	    (if (file-write-access? area)
		(let ((currspace (string->number
				  (list-ref
				   (with-input-from-pipe 
				    ;; (conc "df --output=avail " area)
				    (conc "df -B1000000 " area)
				    ;; (lambda ()(read)(read))
				    (lambda ()(read-line)(string-split (read-line))))
				   3))))
		  (if (> currspace maxspace) 
		      (cons currspace area)
		      res))
		res)))
	(cons 0 #f)
	paths))

;; remove existing link and if possible ...
;; create path to next of tip of target, create link back to source
(define (datashare:build-dir-make-link source target)
  (if (file-exists? target)(datashare:backup-move target))
  (create-directory (pathname-directory target) #t)
  (create-symbolic-link source target))

(define (datashare:backup-move path)
  (let* ((trashdir  (conc (pathname-directory path) "/.trash"))
	 (trashfile (conc trashdir "/" (current-seconds) "-" (pathname-file path))))
    (create-directory trashdir #t)
    (if (directory? path)
	(system (conc "mv " path " " trashfile))
	(file-move path trash-file))))

;;======================================================================
;; GUI
;;======================================================================

;; The main menu 
(define (datashare:main-menu)

		       ;;  					     ;; set positon using coordinates or center, start, top, left, end, bottom, right, parent-center, current
		       ;;  					     ;; #:x 'mouse
		       ;;  					     ;; #:y 'mouse
		       ;;  )					     
		       ))))

(define (datashare:publish-view configdat)
  ;; (pp (hash-table->alist configdat))
  (let* ((areas       (configf:get-section configdat "areas"))
	 (label-size  "70x")
	 (areas-sel   (iup:listbox #:expand "HORIZONTAL" #:dropdown "YES"))
	 (version-tb  (iup:textbox #:expand "HORIZONTAL")) ;;  #:size "50x"))
	 (areas-sel   (iup:listbox #:expand "HORIZONTAL" #:dropdown "YES"))
	 (component   (iup:listbox #:expand "HORIZONTAL" #:dropdown "YES" ))
	 (version-val (iup:textbox #:expand "HORIZONTAL" #:size "50x"))
	 ;; (copy-link   (iup:toggle  #:expand "HORIZONTAL"))
	 ;; (iteration   (iup:textbox #:expand "YES" #:size "20x"))
	 ;; (iteration   (iup:textbox #:expand "HORIZONTAL" #:size "20x"))
	 (area-filter (iup:textbox #:expand "HORIZONTAL" #:value "%"))
	 (comment-tb  (iup:textbox #:expand "YES" #:multiline "YES"))
	 (source-tb   (iup:textbox #:expand "HORIZONTAL"
				   #:value (or (configf:lookup configdat "settings" "basepath")
					       "")))
	 (publish     (lambda (publish-type)
			(let* ((area-num    (or (string->number (iup:attribute areas-sel "VALUE")) 0))
			       (area-dat    (if (> area-num 0)(list-ref areas (- area-num 1))'("NOT SELECTED" "NOT SELECTED")))
			       (area-path   (cadr area-dat))
			       (area-name   (car  area-dat))
			       (version     (iup:attribute version-tb "VALUE"))
			       (comment     (iup:attribute comment-tb "VALUE"))
			       (spath       (iup:attribute source-tb  "VALUE"))
			       (submitter   (current-user-name))
			       (quality     2))
			  (datashare:publish configdat publish-type area-name version comment spath submitter quality))))
	 (copy        (iup:button "Copy and Publish"
				  #:expand "HORIZONTAL"
				  #:action (lambda (obj)
					     (publish 'copy))))
	 (link        (iup:button "Link and Publish"
				  #:expand "HORIZONTAL"
				  #:action (lambda (obj)
					     (publish 'link))))
	 (browse-btn  (iup:button "Browse"
				  #:size "40x"
				  #:action (lambda (obj)
					     (let* ((fd  (iup:file-dialog #:dialogtype "DIR"))
						    (top (iup:show fd #:modal? "YES")))
					       (iup:attribute-set! source-tb "VALUE"
								   (iup:attribute fd "VALUE"))
					       (iup:destroy! fd))))))
    (print "areas")
    ;; (pp areas)
    (fold (lambda (areadat num)
	    ;; (print "Adding num=" num ", areadat=" areadat)
	    (iup:attribute-set! areas-sel (conc num) (car areadat))
	    (+ 1 num))
	  1 areas)
    (iup:vbox
     (iup:hbox (iup:label "Area:"        #:size label-size) ;; area-filter 
	       areas-sel)
     (iup:hbox (iup:label "Version:"     #:size label-size)   version-tb)
     ;; (iup:hbox (iup:label "Link only"    #:size label-size)   copy-link)
     ;; 	       (iup:label "Iteration:")   iteration)
     (iup:hbox (iup:label "Comment:"     #:size label-size)   comment-tb)
     (iup:hbox (iup:label "Source base path:" #:size label-size)   source-tb browse-btn)
     (iup:hbox copy link))))

(define (datashare:lst->path pathlst)
  (conc "/" (string-intersperse (map conc pathlst) "/")))

(define (datashare:path->lst path)
  (string-split path "/"))

(define (datashare:pathdat-apply-heuristics configdat path)
  (cond
   ((file-exists? path) "found")
   (else (conc path " not installed"))))

(define (datashare:get-view configdat)
  (iup:vbox
   (iup:hbox
    (let* ((label-size     "60x")
	   ;; filter elements
	   (area-filter    "%")
	   (version-filter "%")
	   (iter-filter    ">= 0")
	   ;; reverse lookup from path to data for src and installed
	   (srcdat         (make-hash-table)) ;; reverse lookup
	   (installed-dat  (make-hash-table))
	   ;; config values
	   (basepath       (configf:lookup configdat "settings" "basepath"))
	   ;; gui elements
	   (submitter      (iup:label "" #:expand "HORIZONTAL"))
	   (date-submitted (iup:label "" #:expand "HORIZONTAL"))
	   (comment        (iup:label "" #:expand "HORIZONTAL"))
	   (copy-link      (iup:label "" #:expand "HORIZONTAL"))
	   (quality        (iup:label "" #:expand "HORIZONTAL"))
	   (installed-status (iup:label "" #:expand "HORIZONTAL"))
	   ;; misc 
	   (curr-record    #f)
	   ;; (source-data    (iup:label "" #:expand "HORIZONTAL"))
	   (tb             (iup:treebox
			    #:value 0
			    #:name "Packages"
			    #:expand "YES"
			    #:addexpanded "NO"
			    #:selection-cb
			    (lambda (obj id state)
			      ;; (print "obj: " obj ", id: " id ", state: " state)
			      (let* ((path   (datashare:lst->path (cdr (tree:node->path obj id))))
				     (record (hash-table-ref/default srcdat path #f)))
				(if record
				    (begin
				      (set! curr-record record)
				      (iup:attribute-set! submitter      "TITLE" (datashare:pkg-get-submitter record))
				      (iup:attribute-set! date-submitted "TITLE" (time->string (seconds->local-time (datashare:pkg-get-datetime record))))
				      (iup:attribute-set! comment        "TITLE" (datashare:pkg-get-comment record))
				      (iup:attribute-set! quality        "TITLE" (datashare:pkg-get-quality record))
				      (iup:attribute-set! copy-link      "TITLE" (datashare:pkg-get-store_type record))
				      ))
				;; (print  "id=" id " path=" path " record=" record);; (tree:node->path obj id) " run-id: " run-id)
				))))
	   (tb2             (iup:treebox
			    #:value 0
			    #:name "Installed"
			    #:expand "YES"
			    #:addexpanded "NO"
			    #:selection-cb
			    (lambda (obj id state)
			      ;; (print "obj: " obj ", id: " id ", state: " state)
			      (let* ((path   (datashare:lst->path (cdr (tree:node->path obj id))))
				     (status (hash-table-ref/default installed-dat path #f)))
				(iup:attribute-set! installed-status "TITLE" (if status status ""))
				))))
	   (refresh        (lambda (obj)
			     (let* ((db    (datashare:open-db configdat))
				    (areas (or (configf:get-section configdat "areas") '())))
			       ;;
			       ;; first update the Sources
			       ;;
			       (for-each
				(lambda (pkgitem)
				  (let* ((pkg-path   (list (datashare:pkg-get-area  pkgitem)
							   (datashare:pkg-get-version_name pkgitem)
							   (datashare:pkg-get-iteration pkgitem)))
					 (pkg-id     (datashare:pkg-get-id          pkgitem))
					 (path       (datashare:lst->path pkg-path)))
				    ;; (print "tree:add-node tb=" tb ", pkg-path=" pkg-path ", pkg-id=" pkg-id)
				    (if (not (hash-table-ref/default srcdat path #f))
					(tree:add-node tb "Packages" pkg-path userdata: (conc "pkg-id: " pkg-id)))
				    ;; (print "path=" path " pkgitem=" pkgitem)
				    (hash-table-set! srcdat path pkgitem)))
				(datashare:get-pkgs db area-filter version-filter iter-filter))
			       ;;
			       ;; then update the installed
			       ;;
			       (for-each
				(lambda (area)
				  (let* ((path     (conc "/" (cadr area)))
					 (fullpath (conc basepath path)))
				    (if (not (hash-table-ref/default installed-dat path #f))
					(tree:add-node tb2 "Installed" (datashare:path->lst path)))
				    (hash-table-set! installed-dat path (datashare:pathdat-apply-heuristics configdat fullpath))))
				areas)
			       (sqlite3:finalize! db))))
	   (apply          (iup:button "Apply"
				       #:action
				       (lambda (obj)
					 (if curr-record
					     (let* ((area        (datashare:pkg-get-area        curr-record))
						    (stored-path (datashare:pkg-get-stored_path curr-record))
						    (source-type (datashare:pkg-get-store_type  curr-record))
						    (source-path (case source-type ;;  (equal? source-type "link"))
								   ((link)(datashare:pkg-get-source-path curr-record))
								   ((copy)stored-path)
								   (else #f)))
						    (dest-stub   (configf:lookup configdat "areas" area))
						    (target-path (conc basepath "/" dest-stub)))
					       (datashare:build-dir-make-link stored-path target-path)
					       (print "Creating link from " stored-path " to " target-path)))))))
      (iup:vbox 
       (iup:hbox tb tb2)
       (iup:frame 
	#:title "Source Info"
	(iup:vbox
	 (iup:hbox (iup:button "Refresh" #:action refresh) apply)
	 (iup:hbox (iup:label "Submitter: ") ;;  #:size label-size)
		   submitter 
		   (iup:label "Submitted on: ") ;;  #:size label-size)
		   date-submitted)
	 (iup:hbox (iup:label "Data stored: ")
		   copy-link
		   (iup:label "Quality: ")
		   quality)
	 (iup:hbox (iup:label "Comment: ")
		   comment)))
       (iup:frame
	#:title "Installed Info"
	(iup:vbox
	 (iup:hbox (iup:label "Installed status/path: ") installed-status)))
       )))))

(define (datashare:manage-view configdat)
  (iup:vbox
   (iup:hbox 
    (iup:button "Pushme"
		#:expand "YES"
		))))

(define (datashare:gui configdat)
  (iup:show
   (iup:dialog 
    #:title (conc "DataShare dashboard " (current-user-name) ":" (current-directory))   
    #:menu (datashare:main-menu)
    (let* ((tabs (iup:tabs
		  #:tabchangepos-cb (lambda (obj curr prev)
				      (set! *datashare:current-tab-number* curr))
		  (datashare:publish-view configdat)
		  (datashare:get-view configdat)
		  (datashare:manage-view configdat)
		  )))
	;; (set! (iup:callback tabs tabchange-cb:) (lambda (a b c)(print "SWITCHED TO TAB: " a " " b " " c)))
	(iup:attribute-set! tabs "TABTITLE0" "Publish")
	(iup:attribute-set! tabs "TABTITLE1" "Get")
	(iup:attribute-set! tabs "TABTITLE2" "Manage")
	;; (iup:attribute-set! tabs "BGCOLOR" "190 190 190")
	tabs)))
  (iup:main-loop))

;;======================================================================
;; MISC
;;======================================================================


(define (datashare:do-as-calling-user proc)
  (let ((eid (current-effective-user-id))
        (cid (current-user-id)))
    (if (not (eq? eid cid)) ;; running suid
            (set! (current-effective-user-id) cid))
    ;; (print "running as " (current-effective-user-id))
    (proc)
    (if (not (eq? eid cid))
        (set! (current-effective-user-id) eid))))

(define (datashare:find name paths)
  (if (null? paths)
      #f
      (let loop ((hed (car paths))
		 (tal (cdr paths)))
	(if (file-exists? (conc hed "/" name))
	    hed
	    (if (null? tal)
		#f
		(loop (car tal)(cdr tal)))))))

;;======================================================================
;; MAIN
;;======================================================================

(define (datashare:load-config exe-dir exe-name)
  (let* ((fname   (conc exe-dir "/." exe-name ".config")))
    (ini:property-separator-patt " *  *")
    (ini:property-separator #\space)
    (if (file-exists? fname)
	;; (ini:read-ini fname)
	(read-config fname #f #t)
	(make-hash-table))))

(define (datashare:process-action configdat action . args)
  (case (string->symbol action)
    ((get)
     (if (< (length args) 2)
	 (begin 
	   (print "ERROR: Missing arguments; " (string-intersperse args ", "))
	   (exit 1))
	 (let* ((basepath    (configf:lookup configdat "settings" "basepath"))
		(db          (datashare:open-db configdat))
		(area        (car args))
		(version     (cadr args)) ;;    iteration
		(remargs     (args:get-args args '("-i") '() args:arg-hash 0))
		(iteration   (if (args:get-arg "-i")(string->number (args:get-arg "-i")) #f))
		(curr-record (datashare:get-pkg db area version iteration: iteration)))
	   (if (not curr-record)
	       (begin
		 (print "ERROR: No matching record found; area=" area ", version=" version ", iteration=" (if iteration iteration "(max)"))
		 (exit 1))
	       (let* ((stored-path (datashare:pkg-get-stored_path curr-record))
		      (source-type (datashare:pkg-get-store_type  curr-record))
		      (source-path (case source-type ;;  (equal? source-type "link"))
				     ((link) (datashare:pkg-get-source-path curr-record))
				     ((copy) stored-path)
				     (else #f)))
		      (dest-stub   (configf:lookup configdat "areas" area))
		      (target-path (conc basepath "/" dest-stub)))
		 (datashare:build-dir-make-link stored-path target-path)
		 (datashare:record-pkg-ref db (datashare:pkg-get-id curr-record) target-path)
		 (sqlite3:finalize! db)
		 (print "Creating link from " stored-path " to " target-path))))))
    ((publish)
     (if (< (length args) 3)
	 (begin 
	   (print "ERROR: Missing arguments; " (string-intersperse args ", "))
	   (exit 1))
	 (let* ((srcpath  (list-ref args 0))
		(areaname (list-ref args 1))
		(version  (list-ref args 2))
		(remargs  (args:get-args (drop args 2)
					 '("-type" ;; link or copy (default is copy)
					   "-m")
 					 '()
 					 args:arg-hash
 					 0))
		(publish-type (if (equal? (args:get-arg "-type") "link") 'link 'copy))
		(comment      (or (args:get-arg "-m") ""))
		(submitter    (current-user-name))
		(quality      (args:get-arg "-quality"))
		(publish-res  (datashare:publish configdat publish-type areaname version comment srcpath submitter quality)))
	   (if (not (car publish-res))
	       (begin
		 (print "ERROR: " (cdr publish-res))
		 (exit 1))))))
    ((list-versions)
     (let ((area-name (car args)) ;;      version patt   full print
	   (remargs   (args:get-args args '("-vpatt") '("-full") args:arg-hash 0))
	   (db        (datashare:open-db configdat))
	   (versions  (datashare:get-versions-for-area db (car args) version-patt: (args:get-arg "-vpatt"))))
       ;; (print "area-name=" area-name " args=" args " *args-hash*=" (hash-table->alist *args-hash*))
       (map (lambda (x)
	      (if (args:get-arg "-full")
		  (format #t 
			  "~10a~10a~4a~27a~30a\n"
			  (vector-ref x 0)
			  (vector-ref x 1) 
			  (vector-ref x 2) 
			  (conc "\"" (time->string (seconds->local-time (vector-ref x 3))) "\"")
			  (conc "\"" (vector-ref x 4) "\""))
		  (print (vector-ref x 0))))
	    versions)
       (sqlite3:finalize! db)))))

;; ease debugging by loading ~/.dashboardrc - REMOVE FROM PRODUCTION!
(let ((debugcontrolf (conc (get-environment-variable "HOME") "/.datasharerc")))
  (if (file-exists? debugcontrolf)
      (load debugcontrolf)))

(define (main)
  (let* ((args      (argv))
	 (prog      (car args))
	 (rema      (cdr args))
	 (exe-name  (pathname-file (car (argv))))
	 (exe-dir   (or (pathname-directory prog)
			(datashare:find exe-name (string-split (get-environment-variable "PATH") ":"))))
	 (configdat (datashare:load-config exe-dir exe-name)))
    (cond
     ;; one-word commands
     ((eq? (length rema) 1)
      (case (string->symbol (car rema))
	((help -h -help --h --help)
	 (print datashare:help))
	((list-areas)
	 (map print (datashare:get-areas configdat)))
	(else
	 (print "ERROR: Unrecognised command. Try \"datashare help\""))))
     ;; multi-word commands
     ((null? rema)(datashare:gui configdat))
     ((>= (length rema) 2)
      (apply datashare:process-action configdat (car rema)(cdr rema)))
     (else (print "ERROR: Unrecognised command. Try \"datashare help\"")))))

(main)

;;  PURPOSE.
;;======================================================================

;;======================================================================
;; Database access
;;======================================================================

(require-extension (srfi 18) extras tcp)


(use sqlite3 srfi-1 posix regex regex-case srfi-69 csv-xml s11n md5 message-digest base64 format dot-locking z3)
(import (prefix sqlite3 sqlite3:))
(import (prefix base64 base64:))




(declare (unit db))
(declare (uses common))
(declare (uses keys))
(declare (uses ods))

(declare (uses client))
(declare (uses mt))

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

(define *rundb-mutex* (make-mutex)) ;; prevent problems opening/closing rundb's
(define *number-of-writes* 0)
(define *number-non-write-queries* 0)

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

;; convert to -inline
(define (db:first-result-default db stmt default . params)
  (handle-exceptions
   exn
   (let ((err-status ((condition-property-accessor 'sqlite3 'status #f) exn)))
     ;; check for (exn sqlite3) ((condition-property-accessor 'exn 'message) exn)
     (if (eq? err-status 'done)
	 default
	 (begin
	   (debug:print 0 "ERROR:  query " stmt " failed, params: " params ", error: " ((condition-property-accessor 'exn 'message) exn))
	   (print-call-chain (current-error-port))
	   default)))
   (apply sqlite3:first-result db stmt params)))

;; Get/open a database
;;    if run-id => get run specific db
;;    if #f     => get main db
;;    if db already open - return inmem
;;    if db not open, open inmem, rundb and sync then return inmem
;;    inuse gets set automatically for rundb's
;;
(define (db:get-db dbstruct run-id)
  (if (sqlite3:database? dbstruct) ;; pass sqlite3 databases on through
      dbstruct
      (begin
	(let ((dbdat (if (or (not run-id)
			     (eq? run-id 0))
			 (db:open-main dbstruct)
			 (db:open-rundb dbstruct run-id)
			 )))
	  dbdat))))

(define (db:dbdat-get-db dbdat)
  (if (pair? dbdat)
      (car dbdat)
      dbdat))

(define (db:dbdat-get-path dbdat)
  (if (pair? dbdat)
      (cdr dbdat)
      #f))

;; mod-read:
;;     'mod   modified data
;;     'read  read data
;;
(define (db:done-with dbstruct run-id mod-read)
  (if (not (sqlite3:database? dbstruct))
      (begin
	(mutex-lock! *rundb-mutex*)
	(if (eq? mod-read 'mod)
	    (dbr:dbstruct-set-mtime! dbstruct (current-milliseconds))
	    (dbr:dbstruct-set-rtime! dbstruct (current-milliseconds)))
	(dbr:dbstruct-set-inuse! dbstruct #f)
	(mutex-unlock! *rundb-mutex*))))

;; (db:with-db dbstruct run-id sqlite3:exec "select blah from blaz;")
;; r/w is a flag to indicate if the db is modified by this query #t = yes, #f = no
;;
(define (db:with-db dbstruct run-id r/w proc . params)
  (let* ((dbdat (if (vector? dbstruct)
		    (db:get-db dbstruct run-id)
		    dbstruct)) ;; cheat, allow for passing in a dbdat
	 (db    (db:dbdat-get-db dbdat)))
    (db:delay-if-busy dbdat)
    (handle-exceptions
     exn
     (begin
       (debug:print 0 "ERROR: sqlite3 issue in db:with-db, dbstruct=" dbstruct ", run-id=" run-id ", proc=" proc ", params=" params " error: " ((condition-property-accessor 'exn 'message) exn))
       (print-call-chain (current-error-port)))
     (let ((res (apply proc db params)))
       (if (vector? dbstruct)(db:done-with dbstruct run-id r/w))
       res))))

;;======================================================================
;; K E E P   F I L E D B   I N   dbstruct
;;======================================================================

;; (define (db:get-filedb dbstruct run-id)
;;   (let ((db (vector-ref dbstruct 2)))
;;     (if db
;; 	db
;; 	(let ((fdb (filedb:open-db (conc *toplevel* "/db/files.db"))))
;; 	  (vector-set! dbstruct 2 fdb)
;; 	  fdb))))
;; 
;; ;; Can also be used to save arbitrary strings
;; ;;
;; (define (db:save-path dbstruct path)
;;   (let ((fdb (db:get-filedb dbstruct)))b
;;     (filedb:register-path fdb path)))
;; 
;; ;; Use to get a path. To get an arbitrary string see next define
;; ;;
;; (define (db:get-path dbstruct id)
;;   (let ((fdb (db:get-filedb dbstruct)))
;;     (filedb:get-path db id)))

;; NB// #f => return dbdir only
;;      (was planned to be;  zeroth db with name=main.db)
;;
(define (db:dbfile-path run-id)
  (let* ((dbdir           (or (configf:lookup *configdat* "setup" "dbdir")
			      (conc (configf:lookup *configdat* "setup" "linktree") "/.db")))
	 (fname           (if run-id
			      (if (eq? run-id 0) "main.db" (conc run-id ".db"))
			      #f)))
    (handle-exceptions
     exn
     (begin
       (debug:print 0 "ERROR: Couldn't create path to " dbdir)
       (exit 1))
     (if (not (directory? dbdir))(create-directory dbdir #t)))
    (if fname
	(conc dbdir "/" fname)
	dbdir)))
	       
(define (db:set-sync db)
  (let ((syncprag (configf:lookup *configdat* "setup" "sychronous")))
    (sqlite3:execute db (conc "PRAGMA synchronous = " (or syncprag 1) ";"))))

;; open an sql database inside a file lock
;;
;; returns: db existed-prior-to-opening
;;
(define (db:lock-create-open fname initproc)
  ;; (if (file-exists? fname)
  ;;     (let ((db (sqlite3:open-database fname)))
  ;;       (sqlite3:set-busy-handler! db (make-busy-timeout 136000))
  ;;       (db:set-sync db) ;; (sqlite3:execute db "PRAGMA synchronous = 0;")
  ;;       db)
  (let* ((parent-dir   (pathname-directory fname))
	 (dir-writable (file-write-access? parent-dir))
	 (file-exists  (file-exists? fname))
	 (file-write   (if file-exists
			   (file-write-access? fname)
			   dir-writable )))
    (if file-write ;; dir-writable
	(let (;; (lock    (obtain-dot-lock fname 1 5 10))
	      (db      (sqlite3:open-database fname)))
	  (sqlite3:set-busy-handler! db (make-busy-timeout 136000))
	  (db:set-sync db) ;; (sqlite3:execute db "PRAGMA synchronous = 0;")
	  (if (not file-exists)(initproc db))
	  ;; (release-dot-lock fname)


	  db)
	(begin
	  (debug:print 2 "WARNING: opening db in non-writable dir " fname)
	  (sqlite3:open-database fname))))) ;; )

;; This routine creates the db. It is only called if the db is not already opened
;; 
(define (db:open-rundb dbstruct run-id #!key (attemptnum 0)(do-not-open #f)) ;;  (conc *toppath* "/megatest.db") (car *configinfo*)))
  (let* ((local  (dbr:dbstruct-get-local dbstruct))
	 (rdb    (if local
		     (dbr:dbstruct-get-localdb dbstruct run-id)
		     (dbr:dbstruct-get-inmem dbstruct)))) ;; (dbr:dbstruct-get-runrec dbstruct run-id 'inmem)))
    (if (or rdb
	    do-not-open)
	rdb
	(begin
	  (mutex-lock! *rundb-mutex*)
	  (let* ((dbpath       (db:dbfile-path run-id)) ;; (conc toppath "/db/" run-id ".db"))
		 (dbexists     (file-exists? dbpath))
		 (inmem        (if local #f (db:open-inmem-db)))
		 (refdb        (if local #f (db:open-inmem-db)))
		 (db           (db:lock-create-open dbpath ;; this is the database physically on disk
						    (lambda (db)
						      (handle-exceptions
						       exn
						       (begin
							 ;; (release-dot-lock dbpath)
							 (if (> attemptnum 2)
							     (debug:print 0 "ERROR: tried twice, cannot create/initialize db for run-id " run-id ", at path " dbpath)
							     (db:open-rundb dbstruct run-id attemptnum (+ attemptnum 1))))
						       (db:initialize-run-id-db db)
						       (sqlite3:execute 
							db
							"INSERT OR IGNORE INTO tests (id,run_id,testname,event_time,item_path,state,status) VALUES (?,?,'bogustest',strftime('%s','now'),'nowherepath','DELETED','n/a');"
							(* run-id 30000) ;; allow for up to 30k tests per run
							run-id)
						       ;; do a dummy query to test that the table exists and the db is truly readable
						       (sqlite3:execute db "SELECT * FROM tests WHERE id=?;" (* run-id 30000))
						       )))) ;; add strings db to rundb, not in use yet
		 ;;   )) ;; (sqlite3:open-database dbpath))
		 (olddb        (if *megatest-db*
				   *megatest-db* 
				   (let ((db (db:open-megatest-db)))
				     (set! *megatest-db* db)
				     db)))
		 (write-access (file-write-access? dbpath))
		 ;; (handler      (make-busy-timeout 136000))
		 )
	    (if (and dbexists (not write-access))
		(set! *db-write-access* #f)) ;; only unset so other db's also can use this control
	    (dbr:dbstruct-set-rundb!  dbstruct (cons db dbpath))
	    (dbr:dbstruct-set-inuse!  dbstruct #t)
	    (dbr:dbstruct-set-olddb!  dbstruct olddb)
	    ;; (dbr:dbstruct-set-run-id! dbstruct run-id)
	    (mutex-unlock! *rundb-mutex*)
	    (if local
		(begin
		  (dbr:dbstruct-set-localdb! dbstruct run-id db) ;; (dbr:dbstruct-set-inmem! dbstruct db) ;; direct access ...
		  db)
		(begin
		  (dbr:dbstruct-set-inmem!  dbstruct inmem)
		  ;; dec 14, 2014 - keep deleted records available. hunch is that they are needed for id placeholders
		  ;; (sqlite3:execute db "DELETE FROM tests WHERE state='DELETED';") ;; they just slow us down in this context
		  (db:sync-tables db:sync-tests-only db inmem)
		  (db:delay-if-busy refdb) ;; dbpath: (db:dbdat-get-path refdb)) ;; What does delaying here achieve? 
		  (dbr:dbstruct-set-refdb!  dbstruct refdb)
		  (db:sync-tables db:sync-tests-only inmem refdb) ;; use inmem as the reference, don't read again from db
		  ;; sync once more to deal with delays?
		  ;; (db:sync-tables db:sync-tests-only db inmem)
		  ;; (db:sync-tables db:sync-tests-only inmem refdb)
		  inmem)))))))

;; This routine creates the db. It is only called if the db is not already ls opened
;;
(define (db:open-main dbstruct) ;;  (conc *toppath* "/megatest.db") (car *configinfo*)))
  (let ((mdb (dbr:dbstruct-get-main dbstruct)))
    (if mdb

	mdb
	(begin
	  (mutex-lock! *rundb-mutex*)
	  (let* ((dbpath       (db:dbfile-path 0))
		 (dbexists     (file-exists? dbpath))
		 (db           (db:lock-create-open dbpath db:initialize-main-db))
		 (olddb        (db:open-megatest-db))
		 (write-access (file-write-access? dbpath))
		 (dbdat        (cons db dbpath)))
	    (if (and dbexists (not write-access))
		(set! *db-write-access* #f))
	    (dbr:dbstruct-set-main!   dbstruct dbdat)
	    (dbr:dbstruct-set-olddb!  dbstruct olddb) ;; olddb is already a (cons db path)
	    (mutex-unlock! *rundb-mutex*)
	    (if (and (not dbexists)
		     *db-write-access*) ;; did not have a prior db and do have write access
		(db:multi-db-sync #f 'old2new))  ;; migrate data from megatest.db automatically
	    dbdat)))))

;; Make the dbstruct, setup up auxillary db's and call for main db at least once
;;
(define (db:setup run-id #!key (local #f))
  (let* ((dbdir    (db:dbfile-path #f)) ;; (conc (configf:lookup *configdat* "setup" "linktree") "/.db"))
	 (dbstruct (make-dbr:dbstruct path: dbdir local: local)))
    dbstruct))

;; Open the classic megatest.db file in toppath
;;
(define (db:open-megatest-db)
  (let* ((dbpath       (conc *toppath* "/megatest.db"))
	 (dbexists     (file-exists? dbpath))
	 (db           (db:lock-create-open dbpath
					    (lambda (db)
					      (db:initialize-main-db db)
					      (db:initialize-run-id-db db))))
	 (write-access (file-write-access? dbpath)))
    (if (and dbexists (not write-access))
	(set! *db-write-access* #f))
    (cons db dbpath)))

;; sync run to disk if touched
;;
(define (db:sync-touched dbstruct run-id #!key (force-sync #f))
  (let ((mtime  (dbr:dbstruct-get-mtime dbstruct))
	(stime  (dbr:dbstruct-get-stime dbstruct))
	(rundb  (dbr:dbstruct-get-rundb dbstruct))
	(inmem  (dbr:dbstruct-get-inmem dbstruct))
	(maindb (dbr:dbstruct-get-main  dbstruct))
	(refdb  (dbr:dbstruct-get-refdb dbstruct))
	(olddb  (dbr:dbstruct-get-olddb dbstruct))
	;; (runid  (dbr:dbstruct-get-run-id dbstruct))
	)
    (debug:print-info 4 "Syncing for run-id: " run-id)
    ;; (mutex-lock! *http-mutex*)
    (if (eq? run-id 0)
	;; runid equal to 0 is main.db
	(if maindb
	    (if (or (not (number? mtime))
		    (not (number? stime))
		    (> mtime stime)
		    force-sync)
		(begin
		  (db:delay-if-busy maindb)
		  (db:delay-if-busy olddb)
		  (let ((num-synced (db:sync-tables (db:sync-main-list maindb) maindb olddb)))
		    (dbr:dbstruct-set-stime! dbstruct (current-milliseconds))
		    num-synced)
		  0))
	    (begin
	      ;; this can occur when using local access (i.e. not in a server)
	      ;; need a flag to turn it off.
	      ;;
	      (debug:print 3 "WARNING: call to sync main.db to megatest.db but main not initialized")
	      0))
	;; any other runid is a run
	(if (or (not (number? mtime))
		(not (number? stime))
		(> mtime stime)
		force-sync)
	    (begin
	      (db:delay-if-busy rundb)
	      (db:delay-if-busy olddb)
	      (dbr:dbstruct-set-stime! dbstruct (current-milliseconds))
	      (let ((num-synced (db:sync-tables db:sync-tests-only inmem refdb rundb olddb)))
		;; (mutex-unlock! *http-mutex*)
		num-synced)
	      (begin
		;; (mutex-unlock! *http-mutex*)
		0))))))

(define (db:close-main dbstruct)
  (let ((maindb (dbr:dbstruct-get-main dbstruct)))
    (if maindb
	(begin
	  (sqlite3:finalize! (db:dbdat-get-db maindb))
	  (dbr:dbstruct-set-main! dbstruct #f)))))

(define (db:close-run-db dbstruct run-id)
  (let ((rdb (db:open-rundb dbstruct run-id do-not-open: #t)))
    (if (and rdb
	     (sqlite3:database? rdb))
	(begin
	  (sqlite3:finalize! rdb)
	  (dbr:dbstruct-set-localdb! dbstruct run-id #f)
	  (dbr:dbstruct-set-inmem! dbstruct #f)))))

;; close all opened run-id dbs
(define (db:close-all dbstruct)
  ;; finalize main.db
  (db:sync-touched dbstruct 0 force-sync: #t)
  ;;(common:db-block-further-queries)
  ;; (mutex-lock! *db-sync-mutex*) ;; with this perhaps it isn't necessary to use the block-further-queries mechanism?

  (db:close-main dbstruct)
  
  (let ((locdbs (dbr:dbstruct-get-locdbs dbstruct)))
    (if (hash-table? locdbs)
	(for-each (lambda (run-id)
		    (db:close-run-db dbstruct run-id))
		  (hash-table-keys locdbs))))

  ;; (let* ((local (dbr:dbstruct-get-local dbstruct))
  ;;        (rundb (db:dbdat-get-db (dbr:dbstruct-get-rundb dbstruct))))
  ;;   (if local
  ;;       (for-each
  ;;        (lambda (dbdat)
  ;;          (let ((db (db:dbdat-get-db dbdat)))
  ;;            (if (sqlite3:database? db)
  ;;       	 (begin
  ;;       	   (sqlite3:interrupt! db)
  ;;       	   (sqlite3:finalize! db #t)))))
  ;;        ;; TODO: Come back to this and rework to delete from hashtable when finalized
  ;;        (hash-table-values (dbr:dbstruct-get-locdbs dbstruct))))
  ;;   (thread-sleep! 3)
  ;;   (if (and rundb
  ;;            (sqlite3:database? rundb))
  ;;       (handle-exceptions
  ;;        exn
  ;;        (begin 
  ;;          (debug:print 0 "WARNING: database files may not have been closed correctly. Consider running -cleanup-db")
  ;;          (debug:print 0 " message: " ((condition-property-accessor 'exn 'message) exn))
  ;;          (debug:print 0 " db: " rundb)
  ;;          (print-call-chain (current-error-port))
  ;;          #f)
  ;;        (sqlite3:interrupt! rundb)
  ;;        (sqlite3:finalize! rundb #t))))
  ;; ;; (mutex-unlock! *db-sync-mutex*)
  )

(define (db:open-inmem-db)
  (let* ((db      (sqlite3:open-database ":memory:"))
	 (handler (make-busy-timeout 3600)))
    (sqlite3:set-busy-handler! db handler)
    (db:initialize-run-id-db db)
    (cons db #f)))

;; just tests, test_steps and test_data tables
(define db:sync-tests-only
  (list
   ;; (list "strs"
   ;;       '("id"             #f)
   ;;       '("str"            #f))
   (list "tests" 
	 '("id"             #f)
	 '("run_id"         #f)
	 '("testname"       #f)
	 '("host"           #f)
	 '("cpuload"        #f)
	 '("diskfree"       #f)
	 '("uname"          #f)
	 '("rundir"         #f)
	 '("shortdir"       #f)
	 '("item_path"      #f)
	 '("state"          #f)
	 '("status"         #f)
	 '("attemptnum"     #f)
	 '("final_logf"     #f)
	 '("logdat"         #f)
	 '("run_duration"   #f)
	 '("comment"        #f)
	 '("event_time"     #f)
	 '("fail_count"     #f)
	 '("pass_count"     #f)
	 '("archived"       #f))
  (list "test_steps"
	 '("id"             #f)
	 '("test_id"        #f)
	 '("stepname"       #f)
	 '("state"          #f)
	 '("status"         #f)
	 '("event_time"     #f)
	 '("comment"        #f)
	 '("logfile"        #f))
   (list "test_data"
	 '("id"             #f)
	 '("test_id"        #f)
	 '("category"       #f)
	 '("variable"       #f)
	 '("value"          #f)
	 '("expected"       #f)
	 '("tol"            #f)
	 '("units"          #f)
	 '("comment"        #f)
	 '("status"         #f)
	 '("type"           #f))))

;; needs db to get keys, this is for syncing all tables
;;
(define (db:sync-main-list db)
  (let ((keys  (db:get-keys db)))
    (list
     (list "keys"
	   '("id"        #f)
	   '("fieldname" #f)
	   '("fieldtype" #f))
     (list "metadat" '("var" #f) '("val" #f))
     (append (list "runs" 
		   '("id"  #f))
	     (map (lambda (k)(list k #f))
		  (append keys
			  (list "runname" "state" "status" "owner" "event_time" "comment" "fail_count" "pass_count"))))
     (list "test_meta"
	   '("id"             #f)
	   '("testname"       #f)
	   '("owner"          #f)
	   '("description"    #f)
	   '("reviewed"       #f)
	   '("iterated"       #f)
	   '("avg_runtime"    #f)
	   '("avg_disk"       #f)
	   '("tags"           #f)
	   '("jobgroup"       #f)))))

;; use bunch of Unix commands to try to break the lock and recreate the db
;;
(define (db:move-and-recreate-db dbdat)
  (let* ((dbpath   (db:dbdat-get-path        dbdat))
	 (dbdir    (pathname-directory       dbpath))
	 (fname    (pathname-strip-directory dbpath))
	 (fnamejnl (conc fname "-journal"))
	 (tmpname  (conc fname "." (current-process-id)))
	 (tmpjnl   (conc fnamejnl "." (current-process-id))))
    (debug:print 0 "ERROR: " fname " appears corrupted. Making backup \"old/" fname "\"")
    (system (conc "cd " dbdir ";mkdir -p old;cat " fname " > old/" tmpname))
    (system (conc "rm -f " dbpath))
    (if (file-exists? fnamejnl)
	(begin
	  (debug:print 0 "ERROR: " fnamejnl " found, moving it to old dir as " tmpjnl)
	  (system (conc "cd " dbdir ";mkdir -p old;cat " fnamejnl " > old/" tmpjnl))
	  (system (conc "rm -f " dbdir "/" fnamejnl))))
    ;; attempt to recreate database
    (system (conc "cd " dbdir ";sqlite3 old/" tmpname " .dump | sqlite3 " fname))))
    
;; return #f to indicate the dbdat should be closed/reopened
;; else return dbdat
;;
(define (db:repair-db dbdat #!key (numtries 1))
  (let* ((dbpath   (db:dbdat-get-path        dbdat))
	 (dbdir    (pathname-directory       dbpath))
	 (fname    (pathname-strip-directory dbpath)))
    (debug:print-info 0 "Checking db " dbpath " for errors.")
    (cond
     ((not (file-write-access? dbdir))
      (debug:print 0 "WARNING: can't write to " dbdir ", can't fix " fname)
      #f)

     ;; handle special cases, megatest.db and monitor.db
     ;; 
     ;;  NOPE: apply this same approach to all db files
     ;;
     (else ;; ((equal? fname "megatest.db") ;; this file can be regenerated if needed
      (handle-exceptions
       exn
       (begin
	 ;; (db:move-and-recreate-db dbdat)
	 (if (> numtries 0)
	     (db:repair-db dbdat numtries: (- numtries 1))
	     #f)
	 (debug:print 0 "FATAL: file " dbpath " was found corrupted, an attempt to fix has been made but you must start over.")
	 (debug:print 0
		      "   check the following:\n"
		      "      1. full directories, look in ~/ /tmp and " dbdir "\n"
		      "      2. write access to " dbdir "\n\n"
		      "   if the automatic recovery failed you may be able to recover data by doing \"" 
		      (if (member fname '("megatest.db" "monitor.db"))
			  "megatest -cleanup-db"
			  "megatest -import-megatest.db;megatest -cleanup-db")
		      "\"\n")
	 (exit) ;; we can not safely continue when a db was corrupted - even if fixed.
	 )
       ;; test read/write access to the database
       (let ((db (sqlite3:open-database dbpath)))
	 (cond
	  ((equal? fname "megatest.db")
	   (sqlite3:execute db "DELETE FROM tests WHERE state='DELETED';"))
	  ((equal? fname "main.db")
	   (sqlite3:execute db "DELETE FROM runs WHERE state='deleted';"))
	  ((string-match "\\d.db" fname)
	   (sqlite3:execute db "UPDATE tests SET state='DELETED' WHERE state='DELETED';"))
	  ((equal? fname "monitor.db")
	   (sqlite3:execute "DELETE FROM servers WHERE state LIKE 'defunct%';"))
	  (else
	   (sqlite3:execute db "vacuum;")))
	 
	 (finalize! db)
	 #t))))))
    
;; tbls is ( ("tablename" ( "field1" [#f|proc1] ) ( "field2" [#f|proc2] ) .... ) )
;; db's are dbdat's
;;
(define (db:sync-tables tbls fromdb todb . slave-dbs)
  (mutex-lock! *db-sync-mutex*)
  (handle-exceptions
   exn
   (begin
     (mutex-unlock! *db-sync-mutex*)
     (debug:print 0 "EXCEPTION: database probably overloaded or unreadable in db:sync-tables.")
     (print-call-chain (current-error-port))
     (debug:print 0 " message: " ((condition-property-accessor 'exn 'message) exn))
     (print "exn=" (condition->list exn))
     (debug:print 0 " status:  " ((condition-property-accessor 'sqlite3 'status) exn))
     (debug:print 0 " src db:  " (db:dbdat-get-path fromdb))
     (for-each (lambda (dbdat)
		 (let ((dbpath (db:dbdat-get-path dbdat)))
		   (debug:print 0 " dbpath:  " dbpath)
		   (if (not (db:repair-db dbdat))
		       (begin
			 (debug:print 0 "ERROR: Failed to rebuild " dbpath ", exiting now.")
			 (exit)))))
	       (cons todb slave-dbs))
     
     0)
;;      (if *server-run* ;; we are inside a server, throw a sync-failed error
;; 	 (signal (make-composite-condition
;; 		 (make-property-condition 'sync-failed 'message "db:sync-tables failed in a server context.")))
;; 	 0)) ;; return zero for num synced

	 ;; (set! *time-to-exit* #t) ;; let watch dog know that it is time to die.
	 ;; (tasks:server-set-state! (db:delay-if-busy tdbdat) server-id "shutting-down")
	 ;; (portlogger:open-run-close portlogger:set-port port "released")
	 ;; (exit 1)))
   (cond
    ((not fromdb) (debug:print 3 "WARNING: db:sync-tables called with fromdb missing") -1)
    ((not todb)   (debug:print 3 "WARNING: db:sync-tables called with todb missing") -2)
    ((not (sqlite3:database? (db:dbdat-get-db fromdb)))
     (debug:print 0 "ERROR: db:sync-tables called with fromdb not a database " fromdb) -3)
    ((not (sqlite3:database? (db:dbdat-get-db todb)))
     (debug:print 0 "ERROR: db:sync-tables called with todb not a database " todb) -4)
    (else
     (let ((stmts       (make-hash-table)) ;; table-field => stmt
	   (all-stmts   '())              ;; ( ( stmt1 value1 ) ( stml2 value2 ))
	   (numrecs     (make-hash-table))
	   (start-time  (current-milliseconds))
	   (tot-count   0))
       (for-each ;; table
	(lambda (tabledat)
	  (let* ((tablename  (car tabledat))
		 (fields     (cdr tabledat))
		 (num-fields (length fields))
		 (field->num (make-hash-table))
		 (num->field (apply vector (map car fields)))
		 (full-sel   (conc "SELECT " (string-intersperse (map car fields) ",") 
				   " FROM " tablename ";"))
		 (full-ins   (conc "INSERT OR REPLACE INTO " tablename " ( " (string-intersperse (map car fields) ",") " ) "
				   " VALUES ( " (string-intersperse (make-list num-fields "?") ",") " );"))
		 (fromdat    '())
		 (fromdats   '())
		 (totrecords 0)
		 (batch-len  (string->number (or (configf:lookup *configdat* "sync" "batchsize") "10")))
		 (todat      (make-hash-table))
		 (count      0))

	    ;; set up the field->num table
	    (for-each
	     (lambda (field)
	       (hash-table-set! field->num field count)
	       (set! count (+ count 1)))
	     fields)

	    ;; read the source table
	    (sqlite3:for-each-row
	     (lambda (a . b)
	       (set! fromdat (cons (apply vector a b) fromdat))
	       (if (> (length fromdat) batch-len)
		   (begin
		     (set! fromdats (cons fromdat fromdats))
		     (set! fromdat  '())
		     (set! totrecords (+ totrecords 1)))))
	     (db:dbdat-get-db fromdb)
	     full-sel)
	    
	    ;; tack on remaining records in fromdat
	    (if (not (null? fromdat))
		(set! fromdats (cons fromdat fromdats)))

	    (debug:print-info 4 "found " totrecords " records to sync")

	    ;; read the target table
	    (sqlite3:for-each-row
	     (lambda (a . b)
	       (hash-table-set! todat a (apply vector a b)))
	     (db:dbdat-get-db todb)
	     full-sel)

	    ;; first pass implementation, just insert all changed rows
	    (for-each 
	     (lambda (targdb)
	       (let* ((db     (db:dbdat-get-db targdb))
		      (stmth  (sqlite3:prepare db full-ins)))
		 ;; (db:delay-if-busy targdb) ;; NO WAITING
		 (for-each
		  (lambda (fromdat-lst)
		    (sqlite3:with-transaction
		     db
		     (lambda ()
		       (for-each ;; 
			(lambda (fromrow)
			  (let* ((a    (vector-ref fromrow 0))
				 (curr (hash-table-ref/default todat a #f))
				 (same #t))
			    (let loop ((i 0))
			      (if (or (not curr)
				      (not (equal? (vector-ref fromrow i)(vector-ref curr i))))
				  (set! same #f))
			      (if (and same
				       (< i (- num-fields 1)))
				  (loop (+ i 1))))
			    (if (not same)
				(begin
				  (apply sqlite3:execute stmth (vector->list fromrow))
				  (hash-table-set! numrecs tablename (+ 1 (hash-table-ref/default numrecs tablename 0)))))))
			fromdat-lst))
		  ))
		  fromdats)
		 (sqlite3:finalize! stmth)))
	     (append (list todb) slave-dbs))))
	tbls)
       (let* ((runtime      (- (current-milliseconds) start-time))
	      (should-print (common:low-noise-print 30 "db sync" (> runtime 500)))) ;; low and high sync times treated as separate.
	 (if should-print (debug:print 0 "INFO: db sync, total run time " runtime " ms"))
	 (for-each 
	  (lambda (dat)
	    (let ((tblname (car dat))
		  (count   (cdr dat)))
	      (set! tot-count (+ tot-count count))
	      (if (> count 0)
		  (if should-print (debug:print 0 (format #f "    ~10a ~5a" tblname count))))))
	  (sort (hash-table->alist numrecs)(lambda (a b)(> (cdr a)(cdr b))))))
       tot-count)))
   (mutex-unlock! *db-sync-mutex*)))

;; options:
;;
;;  'killservers  - kills all servers
;;  'dejunk       - removes junk records
;;  'adj-testids  - move test-ids into correct ranges
;;  'old2new      - sync megatest.db records to .db/{main,1,2 ...}.db
;;  'new2old      - sync .db/{main,1,2,3 ...}.db to megatest.db
;;  'closeall     - close all opened dbs
;;
;;  run-ids: '(1 2 3 ...) or #f (for all)
;;
(define (db:multi-db-sync run-ids . options)
  (let* ((toppath  (launch:setup-for-run))
	 (dbstruct (if toppath (make-dbr:dbstruct path: toppath) #f))
	 (mtdb     (if toppath (db:open-megatest-db)))
	 (allow-cleanup (if run-ids #f #t))
	 (run-ids  (if run-ids 
		       run-ids
		       (if toppath (begin
				     (db:delay-if-busy mtdb)
				     (db:get-all-run-ids mtdb)))))
	 (tdbdat  (tasks:open-db))
	 (servers (tasks:get-all-servers (db:delay-if-busy tdbdat))))
    
    ;; kill servers
    (if (member 'killservers options)
	(for-each
	 (lambda (server)
	   (tasks:server-delete-record (db:delay-if-busy tdbdat) (vector-ref server 0) "dbmigration")
	   (tasks:kill-server (vector-ref server 2)(vector-ref server 1)))
	 servers))

    ;; clear out junk records
    ;;
    (if (member 'dejunk options)
	(begin
	  (db:delay-if-busy mtdb)
	  (db:clean-up mtdb)))

    ;; adjust test-ids to fit into proper range
    ;;
    (if (member 'adj-testids options)
	(begin
	  (db:delay-if-busy mtdb)
	  (db:prep-megatest.db-for-migration mtdb)))

    ;; sync runs, test_meta etc.
    ;;
    (if (member 'old2new options)
	(begin
	  (db:sync-tables (db:sync-main-list mtdb) mtdb (db:get-db dbstruct #f))
	  (for-each 
	   (lambda (run-id)
	     (db:delay-if-busy mtdb)
	     (let ((testrecs (db:get-all-tests-info-by-run-id mtdb run-id))
		   (dbstruct (if toppath (make-dbr:dbstruct path: toppath local: #t) #f)))
	       (debug:print 0 "INFO: Propagating " (length testrecs) " records for run-id=" run-id " to run specific db")
	       (db:replace-test-records dbstruct run-id testrecs)
	       (sqlite3:finalize! (db:dbdat-get-db (dbr:dbstruct-get-rundb dbstruct)))))
	   run-ids)))

    ;; now ensure all newdb data are synced to megatest.db
    ;; do not use the run-ids list passed in to the function
    ;;
    (if (member 'new2old options)
	(let* ((maindb      (make-dbr:dbstruct path: toppath local: #t))
	       (src-run-ids (db:get-all-run-ids (db:dbdat-get-db (db:get-db maindb 0))))
	       (all-run-ids (sort (delete-duplicates (cons 0 src-run-ids)) <))
	       (count       1)
	       (total       (length all-run-ids))
	       (dead-runs  '()))
	  (for-each
	   (lambda (run-id)
	     (debug:print 0 "Processing run " (if (eq? run-id 0) " main.db " run-id) ", " count " of " total)
	     (set! count (+ count 1))
	     (let* ((fromdb (if toppath (make-dbr:dbstruct path: toppath local: #t) #f))
		    (frundb (db:dbdat-get-db (db:get-db fromdb run-id))))
	       ;; (db:delay-if-busy frundb)
	       ;; (db:delay-if-busy mtdb)
	       ;; (db:clean-up frundb)
	       (if (eq? run-id 0)
		   (begin
		     (db:sync-tables (db:sync-main-list dbstruct) (db:get-db fromdb #f) mtdb)
		     (set! dead-runs (db:clean-up-maindb (db:get-db fromdb #f))))
		   (begin
		     ;; NB// must sync first to ensure deleted tests get marked as such in megatest.db
		     (db:sync-tables db:sync-tests-only (db:get-db fromdb run-id) mtdb)
		     (db:clean-up-rundb (db:get-db fromdb run-id))
		     ))))
	   all-run-ids)
	  ;; removed deleted runs
	  (let ((dbdir (tasks:get-task-db-path)))
	    (for-each (lambda (run-id)
			(let ((fullname (conc dbdir "/" run-id ".db")))
			  (if (file-exists? fullname)
			      (begin
				(debug:print 0 "Removing database file for deleted run " fullname)
				(delete-file fullname)))))
		      dead-runs))))
    ;; (db:close-all dbstruct)
    ;; (sqlite3:finalize! mdb)
    ))

;; keeping it around for debugging purposes only
(define (open-run-close-no-exception-handling  proc idb . params)
  (debug:print-info 11 "open-run-close-no-exception-handling START given a db=" (if idb "yes " "no ") ", params=" params)
  (if (or *db-write-access*
	  (not (member proc *db:all-write-procs*)))
      (let* ((db (cond
		  ((pair? idb)                 (db:dbdat-get-db idb))
		  ((sqlite3:database? idb)     idb)
		  ((not idb)                   (debug:print 0 "ERROR: cannot open-run-close with #f anymore"))
		  ((procedure? idb)            (idb))
		  (else   	               (debug:print 0 "ERROR: cannot open-run-close with #f anymore"))))
	     (res #f))
	(set! res (apply proc db params))
	(if (not idb)(sqlite3:finalize! dbstruct))
	(debug:print-info 11 "open-run-close-no-exception-handling END" )
	res)
      #f))

(define (open-run-close-exception-handling proc idb . params)
  (handle-exceptions
   exn
   (let ((sleep-time (random 30))
	 (err-status ((condition-property-accessor 'sqlite3 'status #f) exn)))
     (case err-status
       ((busy)
	(thread-sleep! sleep-time))
       (else
	(debug:print 0 "EXCEPTION: database probably overloaded or unreadable.")
	(debug:print 0 " message: " ((condition-property-accessor 'exn 'message) exn))
	(print "exn=" (condition->list exn))
	(debug:print 0 " status:  " ((condition-property-accessor 'sqlite3 'status) exn))
	(print-call-chain (current-error-port))
	(thread-sleep! sleep-time)
	(debug:print-info 0 "trying db call one more time....this may never recover, if necessary kill process " (current-process-id) " on host " (get-host-name) " to clean up")))
     (apply open-run-close-exception-handling proc idb params))
   (apply open-run-close-no-exception-handling proc idb params)))

;; (define open-run-close 
(define open-run-close open-run-close-exception-handling)
		;;	   open-run-close-no-exception-handling


;;			   open-run-close-exception-handling)





;;)













(define (db:initialize-main-db dbdat)

  (let* ((configdat (car *configinfo*))  ;; tut tut, global warning...
	 (keys     (keys:config-get-fields configdat))
	 (havekeys (> (length keys) 0))
	 (keystr   (keys->keystr keys))
	 (fieldstr (keys->key/field keys))
	 (db       (db:dbdat-get-db dbdat)))
    (for-each (lambda (key)
		(let ((keyn key))
		  (if (member (string-downcase keyn)
			      (list "runname" "state" "status" "owner" "event_time" "comment" "fail_count"
				    "pass_count"))
		      (begin
			(print "ERROR: your key cannot be named " keyn " as this conflicts with the same named field in the runs table, you must remove your megatest.db and <linktree>/.db before trying again.")

			(exit 1)))))
	      keys)
    (sqlite3:with-transaction
     db
     (lambda ()
       (sqlite3:execute db "CREATE TABLE IF NOT EXISTS keys (id INTEGER PRIMARY KEY, fieldname TEXT, fieldtype TEXT, CONSTRAINT keyconstraint UNIQUE (fieldname));")
       (for-each (lambda (key)
		   (sqlite3:execute db "INSERT OR REPLACE INTO keys (fieldname,fieldtype) VALUES (?,?);" key "TEXT"))
		 keys)
       (sqlite3:execute db (conc 
			    "CREATE TABLE IF NOT EXISTS runs (id INTEGER PRIMARY KEY, \n			 " 
			    fieldstr (if havekeys "," "") "















			 runname    TEXT DEFAULT 'norun',






			 state      TEXT DEFAULT '',
			 status     TEXT DEFAULT '',

			 owner      TEXT DEFAULT '',

			 event_time TIMESTAMP DEFAULT (strftime('%s','now')),
			 comment    TEXT DEFAULT '',

			 fail_count INTEGER DEFAULT 0,
			 pass_count INTEGER DEFAULT 0,

			 CONSTRAINT runsconstraint UNIQUE (runname" (if havekeys "," "") keystr "));"))



       (sqlite3:execute db "CREATE TABLE IF NOT EXISTS test_meta (
                                     id          INTEGER PRIMARY KEY,













                                     testname    TEXT DEFAULT '',
                                     author      TEXT DEFAULT '',
                                     owner       TEXT DEFAULT '',
                                     description TEXT DEFAULT '',
                                     reviewed    TIMESTAMP,
                                     iterated    TEXT DEFAULT '',
                                     avg_runtime REAL,
                                     avg_disk    REAL,
                                     tags        TEXT DEFAULT '',
                                     jobgroup    TEXT DEFAULT 'default',
                                CONSTRAINT test_meta_constraint UNIQUE (testname));")
       (sqlite3:execute db "CREATE TABLE IF NOT EXISTS tasks_queue (id INTEGER PRIMARY KEY,
                                action TEXT DEFAULT '',
                                owner TEXT,
                                state TEXT DEFAULT 'new',
                                target TEXT DEFAULT '',
                                name TEXT DEFAULT '',
                                testpatt TEXT DEFAULT '',
                                keylock TEXT,
                                params TEXT,
                                creation_time TIMESTAMP DEFAULT (strftime('%s','now')),
                                execution_time TIMESTAMP);")
       ;; archive disk areas, cached info from [archive-disks]
       (sqlite3:execute db "CREATE TABLE IF NOT EXISTS archive_disks (
                                id INTEGER PRIMARY KEY,
                                archive_area_name TEXT,
                                disk_path TEXT,
                                last_df INTEGER DEFAULT -1,
                                last_df_time TIMESTAMP DEFAULT (strftime('%s','now')),
                                creation_time TIMESTAMP DEFAULT (strftime('%','now')));")
       ;; individual bup (or tar) data chunks
       (sqlite3:execute db "CREATE TABLE IF NOT EXISTS archive_blocks (
                                id INTEGER PRIMARY KEY,
                                archive_disk_id INTEGER,
                                disk_path TEXT,
                                last_du INTEGER DEFAULT -1,
                                last_du_time TIMESTAMP DEFAULT (strftime('%s','now')),
                                creation_time TIMESTAMP DEFAULT (strftime('%','now')));")
       ;; tests allocated to what chunks. reusing a chunk for a test/item_path is very efficient
       ;; NB// the per run/test recording of where the archive is stored is done in the test
       ;;      record. 
       (sqlite3:execute db "CREATE TABLE IF NOT EXISTS archive_allocations (
                                id INTEGER PRIMARY KEY,
                                archive_block_id INTEGER,
                                testname TEXT,
                                item_path TEXT,
                                creation_time TIMESTAMP DEFAULT (strftime('%','now')));")
       ;; move this clean up call somewhere else
       (sqlite3:execute db "DELETE FROM tasks_queue WHERE state='done' AND creation_time < ?;" (- (current-seconds)(* 24 60 60))) ;; remove older than 24 hrs
       (sqlite3:execute db (conc "CREATE INDEX IF NOT EXISTS runs_index ON runs (runname" (if havekeys "," "") keystr ");"))
       ;; (sqlite3:execute db "CREATE VIEW runs_tests AS SELECT * FROM runs INNER JOIN tests ON runs.id=tests.run_id;")
       (sqlite3:execute db "CREATE TABLE IF NOT EXISTS extradat (id INTEGER PRIMARY KEY, run_id INTEGER, key TEXT, val TEXT);")
       (sqlite3:execute db "CREATE TABLE IF NOT EXISTS metadat (id INTEGER PRIMARY KEY, var TEXT, val TEXT,
                                  CONSTRAINT metadat_constraint UNIQUE (var));")
       (sqlite3:execute db "CREATE TABLE IF NOT EXISTS access_log (id INTEGER PRIMARY KEY, user TEXT, accessed TIMESTAMP, args TEXT);")
       ;; Must do this *after* running patch db !! No more. 
       ;; cannot use db:set-var since it will deadlock, hardwire the code here
       (sqlite3:execute db "INSERT OR REPLACE INTO metadat (var,val) VALUES (?,?);" "MEGATEST_VERSION" megatest-version)
       (debug:print-info 11 "db:initialize END")))))

;;======================================================================
;; R U N   S P E C I F I C   D B 
;;======================================================================

(define (db:initialize-run-id-db db)
  (sqlite3:with-transaction 
   db
   (lambda ()
     (sqlite3:execute db "CREATE TABLE IF NOT EXISTS tests 
                    (id INTEGER PRIMARY KEY,
                     run_id       INTEGER   DEFAULT -1,
                     testname     TEXT      DEFAULT 'noname',
                     host         TEXT      DEFAULT 'n/a',
                     cpuload      REAL      DEFAULT -1,
                     diskfree     INTEGER   DEFAULT -1,
                     uname        TEXT      DEFAULT 'n/a', 
                     rundir       TEXT      DEFAULT '/tmp/badname',
                     shortdir     TEXT      DEFAULT '/tmp/badname',
                     item_path    TEXT      DEFAULT '',
                     state        TEXT      DEFAULT 'NOT_STARTED',
                     status       TEXT      DEFAULT 'FAIL',
                     attemptnum   INTEGER   DEFAULT 0,
                     final_logf   TEXT      DEFAULT 'logs/final.log',
                     logdat       TEXT      DEFAULT '', 
                     run_duration INTEGER   DEFAULT 0,
                     comment      TEXT      DEFAULT '',
                     event_time   TIMESTAMP DEFAULT (strftime('%s','now')),
                     fail_count   INTEGER   DEFAULT 0,
                     pass_count   INTEGER   DEFAULT 0,
                     archived     INTEGER   DEFAULT 0, -- 0=no, > 1=archive block id where test data can be found
                        CONSTRAINT testsconstraint UNIQUE (run_id, testname, item_path));")
     (sqlite3:execute db "CREATE INDEX IF NOT EXISTS tests_index ON tests (run_id, testname, item_path);")
     (sqlite3:execute db "CREATE TABLE IF NOT EXISTS test_steps 
                              (id INTEGER PRIMARY KEY,
                               test_id INTEGER, 
                               stepname TEXT, 
                               state TEXT DEFAULT 'NOT_STARTED', 
                               status TEXT DEFAULT 'n/a',
                               event_time TIMESTAMP,
                               comment TEXT DEFAULT '',
                               logfile TEXT DEFAULT '',
                               CONSTRAINT test_steps_constraint UNIQUE (test_id,stepname,state));")
     ;;   (sqlite3:execute db "CREATE TABLE IF NOT EXISTS test_data 
     ;;                               (id          INTEGER PRIMARY KEY,
     ;;                                      reviewed    TIMESTAMP DEFAULT (strftime('%s','now')),
     ;;                                      iterated    TEXT DEFAULT '',
     ;;                                      avg_runtime REAL DEFAULT -1,
     ;;                                      avg_disk    REAL DEFAULT -1,
     ;;                                      tags        TEXT DEFAULT '',
     ;;                                      jobgroup    TEXT DEFAULT 'default',
     ;;                                 CONSTRAINT test_meta_constraint UNIQUE (testname));")
     (sqlite3:execute db "CREATE TABLE IF NOT EXISTS test_data (id INTEGER PRIMARY KEY,
                                test_id INTEGER,
                                category TEXT DEFAULT '',
                                variable TEXT,
	                        value REAL,
	                        expected REAL,
	                        tol REAL,
                                units TEXT,
                                comment TEXT DEFAULT '',
                                status TEXT DEFAULT 'n/a',
                                type TEXT DEFAULT '',
                              CONSTRAINT test_data_constraint UNIQUE (test_id,category,variable));")
     ;; Why use FULL here? This data is not that critical
     ;; (sqlite3:execute db "PRAGMA synchronous = FULL;")
     (sqlite3:execute db "CREATE TABLE IF NOT EXISTS test_rundat (

                              id           INTEGER PRIMARY KEY,

                              test_id      INTEGER,

                              update_time  TIMESTAMP,
                              cpuload      INTEGER DEFAULT -1,
                              diskfree     INTEGER DEFAULT -1,
                              diskusage    INTGER DEFAULT -1,
                              run_duration INTEGER DEFAULT 0);")
     (sqlite3:execute db "CREATE TABLE IF NOT EXISTS archives (
                              id           INTEGER PRIMARY KEY,
                              test_id      INTEGER,


                              state        TEXT DEFAULT 'new',
                              status       TEXT DEFAULT 'n/a',
                              archive_type TEXT DEFAULT 'bup',
                              du           INTEGER,




                              archive_path TEXT);")))
  db)



;;======================================================================
;; A R C H I V E S
;;======================================================================


;; dneeded is minimum space needed, scan for existing archives that 
;; are on disks with adequate space and already have this test/itempath
;; archived
;;
(define (db:archive-get-allocations dbstruct testname itempath dneeded)
  (let* ((dbdat        (db:get-db dbstruct #f)) ;; archive tables are in main.db
	 (db           (db:dbdat-get-db dbdat))
	 (res          '())
	 (blocks       '())) ;; a block is an archive chunck that can be added too if there is space
    (sqlite3:for-each-row


     (lambda (id archive-disk-id disk-path last-du last-du-time)
       (set! res (cons (vector id archive-disk-id disk-path last-du last-du-time) res)))







     db
     "SELECT b.id,b.archive_disk_id,b.disk_path,b.last_du,b.last_du_time FROM archive_blocks AS b
        INNER JOIN archive_allocations AS a ON a.archive_block_id=b.id
        WHERE a.testname=? AND a.item_path=?;" 
     testname itempath)
    ;; Now res has list of candidate paths, look in archive_disks for candidate with potential free space
    (if (null? res)
	'()
	(sqlite3:for-each-row
	 (lambda (id archive-area-name disk-path last-df last-df-time)
	   (set! blocks (cons (vector id archive-area-name disk-path last-df last-df-time) blocks)))
	 db 
	 (conc
	  "SELECT d.id,d.archive_area_name,disk_path,last_df,last_df_time FROM archive_disks AS d
             INNER JOIN archive_blocks AS b ON d.id=b.archive_disk_id
             WHERE b.id IN (" (string-intersperse (map conc res) ",") ") AND
         last_df > ?;")
	 dneeded))
    blocks))
    
;; returns id of the record, register a disk allocated to archiving and record it's last known
;; available space
;;
(define (db:archive-register-disk dbstruct bdisk-name bdisk-path df)
  (let* ((dbdat        (db:get-db dbstruct #f)) ;; archive tables are in main.db
	 (db           (db:dbdat-get-db dbdat))
	 (res          #f))
    (sqlite3:for-each-row
     (lambda (id)
       (set! res id))
     db
     "SELECT id FROM archive_disks WHERE archive_area_name=? AND disk_path=?;"
     bdisk-name bdisk-path)
    (if res ;; record exists, update df and return id
	(begin


	  (sqlite3:execute db "UPDATE archive_disks SET last_df=?,last_df_time=(strftime('%s','now'))
                                  WHERE archive_area_name=? AND disk_path=?;"
			   df bdisk-name bdisk-path)
	  res)
	(begin
	  (sqlite3:execute
	   db
	   "INSERT OR REPLACE INTO archive_disks (archive_area_name,disk_path,last_df)
                VALUES (?,?,?);"
	   bdisk-name bdisk-path df)
	  (db:archive-register-disk dbstruct bdisk-name bdisk-path df)))))


;; record an archive path created on a given archive disk (identified by it's bdisk-id)
;; if path starts with / then it is full, otherwise it is relative to the archive disk
;; preference is to store the relative path.
;;
(define (db:archive-register-block-name dbstruct bdisk-id archive-path #!key (du #f))
  (let* ((dbdat        (db:get-db dbstruct #f)) ;; archive tables are in main.db
	 (db           (db:dbdat-get-db dbdat))
	 (res          #f))
    ;; first look to see if this path is already registered
    (sqlite3:for-each-row
     (lambda (id)
       (set! res id))
     db
     "SELECT id FROM archive_blocks WHERE archive_disk_id=? AND disk_path=?;"
     bdisk-id archive-path)
    (if res ;; record exists, update du if applicable and return res
	(begin
	  (if du (sqlite3:exectute db "UPDATE archive_blocks SET last_du=?,last_du_time=(strftime('%s','now'))
                                          WHERE archive_disk_id=? AND disk_path=?;"
				   bdisk-id archive-path du))
	  res)
	(begin
	  (sqlite3:execute db "INSERT OR REPLACE INTO archive_blocks (archive_disk_id,disk_path,last_du)
                                                        VALUES (?,?,?);"
			   bdisk-id archive-path (or du 0))
	  (db:archive-register-block-name dbstruct bdisk-id archive-path du: du)))))


;; The "archived" field in tests is overloaded; 0 = not archived, > 0 archived in block with given id
;;
(define (db:test-set-archive-block-id dbstruct run-id test-id archive-block-id)
  (db:with-db
   dbstruct
   run-id
   #f
   (lambda (db)
     (sqlite3:execute db "UPDATE tests SET archived=? WHERE id=?;"









		      archive-block-id test-id))))







 
;; Look up the archive block info given a block-id
;;
(define (db:test-get-archive-block-info dbstruct archive-block-id)
  (db:with-db
   dbstruct
   #f
   #f
   (lambda (db)

     (let ((res #f))

       (sqlite3:for-each-row 
	;;        0         1           2        3          4           5
	(lambda (id archive-disk-id disk-path last-du last-du-time creation-time)
	  (set! res (vector id archive-disk-id disk-path last-du last-du-time creation-time)))

	db
	"SELECT id,archive_disk_id,disk_path,last_du,last_du_time,creation_time FROM archive_blocks WHERE id=?;"

	archive-block-id)
       res))))

;; (define (db:archive-allocate-testsuite/area-to-block block-id testsuite-name areakey)
;;   (let* ((dbdat        (db:get-db dbstruct #f)) ;; archive tables are in main.db
;; 	 (db           (db:dbdat-get-db dbdat))
;; 	 (res          '())
;; 	 (blocks       '())) ;; a block is an archive chunck that can be added too if there is space
;;     (sqlite3:for-each-row  #f)

;;======================================================================
;; L O G G I N G    D B 
;;======================================================================

(define (open-logging-db) ;;  (conc *toppath* "/megatest.db") (car *configinfo*)))
  (let* ((dbpath    (conc (if *toppath* (conc *toppath* "/") "") "logging.db")) ;; fname)
	 (dbexists  (file-exists? dbpath))
	 (db        (sqlite3:open-database dbpath))
	 (handler   (make-busy-timeout (if (args:get-arg "-override-timeout")
					   (string->number (args:get-arg "-override-timeout"))
					   136000)))) ;; 136000)))
    (sqlite3:set-busy-handler! db handler)
    (if (not dbexists)
	(begin
	  (sqlite3:execute db "CREATE TABLE IF NOT EXISTS log (id INTEGER PRIMARY KEY,event_time TIMESTAMP DEFAULT (strftime('%s','now')),logline TEXT,pwd TEXT,cmdline TEXT,pid INTEGER);")
	  (db:set-sync db) ;; (sqlite3:execute db (conc "PRAGMA synchronous = 0;"))
	  ))
    db))

(define (db:log-local-event . loglst)
  (let ((logline (apply conc loglst)))



    (db:log-event logline)))

(define (db:log-event logline)
  (let ((db (open-logging-db)))
    (sqlite3:execute db "INSERT INTO log (logline,pwd,cmdline,pid) VALUES (?,?,?,?);"
		     logline
		     (current-directory)
		     (string-intersperse (argv) " ")
		     (current-process-id))
    (sqlite3:finalize! db)
    logline))

;;======================================================================
;; D B   U T I L S


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
















































































;;======================================================================
;; M A I N T E N A N C E
;;======================================================================

(define (db:have-incompletes? dbstruct run-id ovr-deadtime)
  (let* ((dbdat        (db:get-db dbstruct run-id))
	 (db           (db:dbdat-get-db dbdat))
	 (incompleted '())
	 (oldlaunched '())
	 (toplevels   '())
	 (deadtime-str (configf:lookup *configdat* "setup" "deadtime"))
	 (deadtime     (if (and deadtime-str
				(string->number deadtime-str))
			   (string->number deadtime-str)
			   7200))) ;; two hours
    (if (number? ovr-deadtime)(set! deadtime ovr-deadtime))
    
    ;; in RUNNING or REMOTEHOSTSTART for more than 10 minutes
    ;;
    ;; HOWEVER: this code in run:test seems to work fine
    ;;              (> (- (current-seconds)(+ (db:test-get-event_time testdat)
    ;;                     (db:test-get-run_duration testdat)))
    ;;                    600) 
    (db:delay-if-busy dbdat)
    (sqlite3:for-each-row 
     (lambda (test-id run-dir uname testname item-path)
       (if (and (equal? uname "n/a")
		(equal? item-path "")) ;; this is a toplevel test
	   ;; what to do with toplevel? call rollup?
	   (begin
	     (set! toplevels   (cons (list test-id run-dir uname testname item-path run-id) toplevels))
	     (debug:print-info 0 "Found old toplevel test in RUNNING state, test-id=" test-id))
	   (set! incompleted (cons (list test-id run-dir uname testname item-path run-id) incompleted))))
     db
     "SELECT id,rundir,uname,testname,item_path FROM tests WHERE run_id=? AND (strftime('%s','now') - event_time) > (run_duration + ?) AND state IN ('RUNNING','REMOTEHOSTSTART');"
     run-id deadtime)

    ;; in LAUNCHED for more than one day. Could be long due to job queues TODO/BUG: Need override for this in config
    ;;
    (db:delay-if-busy dbdat)
    (sqlite3:for-each-row
     (lambda (test-id run-dir uname testname item-path)
       (if (and (equal? uname "n/a")
		(equal? item-path "")) ;; this is a toplevel test
	   ;; what to do with toplevel? call rollup?
	   (set! toplevels   (cons (list test-id run-dir uname testname item-path run-id) toplevels))
	   (set! oldlaunched (cons (list test-id run-dir uname testname item-path run-id) oldlaunched))))
     db
     "SELECT id,rundir,uname,testname,item_path FROM tests WHERE run_id=? AND (strftime('%s','now') - event_time) > 86400 AND state IN ('LAUNCHED');"
     run-id)
    
    (debug:print-info 18 "Found " (length oldlaunched) " old LAUNCHED items, " (length toplevels) " old LAUNCHED toplevel tests and " (length incompleted) " tests marked RUNNING but apparently dead.")
    (if (and (null? incompleted)
	     (null? oldlaunched)
	     (null? toplevels))
	#f
	#t)))

;;  select end_time-now from
;;      (select testname,item_path,event_time+run_duration as
;;                          end_time,strftime('%s','now') as now from tests where state in
;;      ('RUNNING','REMOTEHOSTSTART','LAUNCED'));

(define (db:find-and-mark-incomplete dbstruct run-id ovr-deadtime)
  (let* ((dbdat        (db:get-db dbstruct run-id))
	 (db           (db:dbdat-get-db dbdat))
	 (incompleted '())
	 (oldlaunched '())
	 (toplevels   '())
	 (deadtime-str (configf:lookup *configdat* "setup" "deadtime"))
	 (deadtime     (if (and deadtime-str
				(string->number deadtime-str))
			   (string->number deadtime-str)
			   7200))) ;; two hours

    (if (number? ovr-deadtime)(set! deadtime ovr-deadtime))


    
    ;; in RUNNING or REMOTEHOSTSTART for more than 10 minutes
    ;;



    ;; HOWEVER: this code in run:test seems to work fine
    ;;              (> (- (current-seconds)(+ (db:test-get-event_time testdat)
    ;;                     (db:test-get-run_duration testdat)))
    ;;                    600) 
    (db:delay-if-busy dbdat)
    (sqlite3:for-each-row 
     (lambda (test-id run-dir uname testname item-path)
       (if (and (equal? uname "n/a")
		(equal? item-path "")) ;; this is a toplevel test
	   ;; what to do with toplevel? call rollup?
	   (begin
	     (set! toplevels   (cons (list test-id run-dir uname testname item-path run-id) toplevels))
	     (debug:print-info 0 "Found old toplevel test in RUNNING state, test-id=" test-id))
	   (set! incompleted (cons (list test-id run-dir uname testname item-path run-id) incompleted))))
     db
     "SELECT id,rundir,uname,testname,item_path FROM tests WHERE run_id=? AND (strftime('%s','now') - event_time) > (run_duration + ?) AND state IN ('RUNNING','REMOTEHOSTSTART');"
     run-id deadtime)

    ;; in LAUNCHED for more than one day. Could be long due to job queues TODO/BUG: Need override for this in config
    ;;
    (db:delay-if-busy dbdat)
    (sqlite3:for-each-row
     (lambda (test-id run-dir uname testname item-path)
       (if (and (equal? uname "n/a")
		(equal? item-path "")) ;; this is a toplevel test
	   ;; what to do with toplevel? call rollup?
	   (set! toplevels   (cons (list test-id run-dir uname testname item-path run-id) toplevels))
	   (set! oldlaunched (cons (list test-id run-dir uname testname item-path run-id) oldlaunched))))
     db
     "SELECT id,rundir,uname,testname,item_path FROM tests WHERE run_id=? AND (strftime('%s','now') - event_time) > 86400 AND state IN ('LAUNCHED');"
     run-id)
    
    (debug:print-info 18 "Found " (length oldlaunched) " old LAUNCHED items, " (length toplevels) " old LAUNCHED toplevel tests and " (length incompleted) " tests marked RUNNING but apparently dead.")

    ;; These are defunct tests, do not do all the overhead of set-state-status. Force them to INCOMPLETE.
    ;;
    (db:delay-if-busy dbdat)
    (let* (;; (min-incompleted (filter (lambda (x)
	   ;;      		      (let* ((testpath (cadr x))
	   ;;      			     (tdatpath (conc testpath "/testdat.db"))
	   ;;      			     (dbexists (file-exists? tdatpath)))
	   ;;      			(or (not dbexists) ;; if no file then something wrong - mark as incomplete
	   ;;      			    (> (- (current-seconds)(file-modification-time tdatpath)) 600)))) ;; no change in 10 minutes to testdat.db - she's dead Jim
	   ;;      		    incompleted))
	   (min-incompleted-ids (map car incompleted)) ;; do 'em all
	   (all-ids             (append min-incompleted-ids (map car oldlaunched))))
      (if (> (length all-ids) 0)
	  (begin
	    (debug:print 0 "WARNING: Marking test(s); " (string-intersperse (map conc all-ids) ", ") " as INCOMPLETE")
	    (sqlite3:execute 
	     db
	     (conc "UPDATE tests SET state='INCOMPLETE' WHERE id IN (" 
		   (string-intersperse (map conc all-ids) ",")
		   ");")))))

    ;; Now do rollups for the toplevel tests
    ;;
    (db:delay-if-busy dbdat)
    (for-each
     (lambda (toptest)
       (let ((test-name (list-ref toptest 3)))
;;	     (run-id    (list-ref toptest 5)))
	 (db:top-test-set-per-pf-counts dbstruct run-id test-name)))
     toplevels)))

(define (db:top-test-set-per-pf-counts dbstruct run-id test-name)
  (db:general-call (db:get-db dbstruct run-id) 'top-test-set-per-pf-counts (list test-name test-name test-name test-name test-name test-name test-name test-name test-name test-name test-name test-name test-name test-name test-name test-name test-name))) 
 
		     
;; Clean out old junk and vacuum the database
;;
;; Ultimately do something like this:
;;
;; 1. Look at test records either deleted or part of deleted run:
;;    a. If test dir exists, set the the test to state='UNKNOWN', Set the run to 'unknown'
;;    b. If test dir gone, delete the test record
;; 2. Look at run records
;;    a. If have tests that are not deleted, set state='unknown'
;;    b. ....
;;
(define (db:clean-up dbdat)
  ;; (debug:print 0 "WARNING: db clean up not fully ported to v1.60, cleanup action will be on megatest.db")
  (let* ((db         (db:dbdat-get-db dbdat))
	 (count-stmt (sqlite3:prepare db "SELECT (SELECT count(id) FROM tests)+(SELECT count(id) FROM runs);"))
	(statements
	 (map (lambda (stmt)
		(sqlite3:prepare db stmt))
	      (list
	       ;; delete all tests that belong to runs that are 'deleted'
	       "DELETE FROM tests WHERE run_id in (SELECT id FROM runs WHERE state='deleted');"
	       ;; delete all tests that are 'DELETED'
	       "DELETE FROM tests WHERE state='DELETED';"
	       ;; delete all tests that have no run
	       "DELETE FROM tests WHERE run_id NOT IN (SELECT DISTINCT id FROM runs);"
	       ;; delete all runs that are state='deleted'
	       "DELETE FROM runs WHERE state='deleted';"
	       ;; delete empty runs
	       "DELETE FROM runs WHERE id NOT IN (SELECT DISTINCT r.id FROM runs AS r INNER JOIN tests AS t ON t.run_id=r.id);"
	       ))))
    (db:delay-if-busy dbdat)
    (sqlite3:with-transaction 
     db
     (lambda ()
       (sqlite3:for-each-row (lambda (tot)
			       (debug:print-info 0 "Records count before clean: " tot))
			     count-stmt)
       (map sqlite3:execute statements)
       (sqlite3:for-each-row (lambda (tot)
			       (debug:print-info 0 "Records count after  clean: " tot))
			     count-stmt)))
    (map sqlite3:finalize! statements)
    (sqlite3:finalize! count-stmt)
    ;; (db:find-and-mark-incomplete db)
    (db:delay-if-busy dbdat)
    (sqlite3:execute db "VACUUM;")))

;; Clean out old junk and vacuum the database
;;
;; Ultimately do something like this:
;;
;; 1. Look at test records either deleted or part of deleted run:
;;    a. If test dir exists, set the the test to state='UNKNOWN', Set the run to 'unknown'
;;    b. If test dir gone, delete the test record
;; 2. Look at run records
;;    a. If have tests that are not deleted, set state='unknown'
;;    b. ....
;;
(define (db:clean-up-rundb dbdat)
  ;; (debug:print 0 "WARNING: db clean up not fully ported to v1.60, cleanup action will be on megatest.db")
  (let* ((db         (db:dbdat-get-db dbdat))
	 (count-stmt (sqlite3:prepare db "SELECT (SELECT count(id) FROM tests);"))
	(statements
	 (map (lambda (stmt)
		(sqlite3:prepare db stmt))
	      (list
	       ;; delete all tests that belong to runs that are 'deleted'
	       ;; (conc "DELETE FROM tests WHERE run_id NOT IN (" (string-intersperse (map conc valid-runs) ",") ");")
	       ;; delete all tests that are 'DELETED'
	       "DELETE FROM tests WHERE state='DELETED';"
	       ))))
    (db:delay-if-busy dbdat)
    (sqlite3:with-transaction 
     db
     (lambda ()
       (sqlite3:for-each-row (lambda (tot)
			       (debug:print-info 0 "Records count before clean: " tot))
			     count-stmt)
       (map sqlite3:execute statements)
       (sqlite3:for-each-row (lambda (tot)
			       (debug:print-info 0 "Records count after  clean: " tot))
			     count-stmt)))
    (map sqlite3:finalize! statements)
    (sqlite3:finalize! count-stmt)
    ;; (db:find-and-mark-incomplete db)
    (db:delay-if-busy dbdat)
    (sqlite3:execute db "VACUUM;")))

;; Clean out old junk and vacuum the database
;;
;; Ultimately do something like this:
;;
;; 1. Look at test records either deleted or part of deleted run:
;;    a. If test dir exists, set the the test to state='UNKNOWN', Set the run to 'unknown'
;;    b. If test dir gone, delete the test record
;; 2. Look at run records
;;    a. If have tests that are not deleted, set state='unknown'
;;    b. ....
;;
(define (db:clean-up-maindb dbdat)
  ;; (debug:print 0 "WARNING: db clean up not fully ported to v1.60, cleanup action will be on megatest.db")
  (let* ((db         (db:dbdat-get-db dbdat))
	 (count-stmt (sqlite3:prepare db "SELECT (SELECT count(id) FROM runs);"))
	 (statements
	  (map (lambda (stmt)
		 (sqlite3:prepare db stmt))
	       (list
		;; delete all tests that belong to runs that are 'deleted'
		;; (conc "DELETE FROM tests WHERE run_id NOT IN (" (string-intersperse (map conc valid-runs) ",") ");")
		;; delete all tests that are 'DELETED'
		"DELETE FROM runs WHERE state='deleted';"
		)))
	 (dead-runs '()))
    (sqlite3:for-each-row
     (lambda (run-id)
       (set! dead-runs (cons run-id dead-runs)))
       db
       "SELECT id FROM runs WHERE state='deleted';")
    (db:delay-if-busy dbdat)
    (sqlite3:with-transaction 
     db
     (lambda ()
       (sqlite3:for-each-row (lambda (tot)
			       (debug:print-info 0 "Records count before clean: " tot))
			     count-stmt)
       (map sqlite3:execute statements)
       (sqlite3:for-each-row (lambda (tot)
			       (debug:print-info 0 "Records count after  clean: " tot))
			     count-stmt)))
    (map sqlite3:finalize! statements)
    (sqlite3:finalize! count-stmt)
    ;; (db:find-and-mark-incomplete db)
    (db:delay-if-busy dbdat)
    (sqlite3:execute db "VACUUM;")
    dead-runs))

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

;; M E T A   G E T   A N D   S E T   V A R S
;;======================================================================

;; returns number if string->number is successful, string otherwise
;; also updates *global-delta*
;;
;; Operates on megatestdb
;;
(define (db:get-var dbstruct var)

  (let* ((start-ms (current-milliseconds))
         (throttle (let ((t  (config-lookup *configdat* "setup" "throttle")))
		     (if t (string->number t) t)))
	 (res      #f)
	 (dbdat    (db:get-db dbstruct #f))
	 (db       (db:dbdat-get-db dbdat)))
    (db:delay-if-busy dbdat)
    (sqlite3:for-each-row
     (lambda (val)
       (set! res val))
     db
     "SELECT val FROM metadat WHERE var=?;" var)
    ;; convert to number if can
    (if (string? res)
	(let ((valnum (string->number res)))
	  (if valnum (set! res valnum))))
    ;; scale by 10, average with current value.
    (set! *global-delta* (/ (+ *global-delta* (* (- (current-milliseconds) start-ms)
						 (if throttle throttle 0.01)))
			    2))
    (if (> (abs (- *last-global-delta-printed* *global-delta*)) 0.08) ;; don't print all the time, only if it changes a bit
	(begin
	  (debug:print-info 4 "launch throttle factor=" *global-delta*)
	  (set! *last-global-delta-printed* *global-delta*)))

    res))

(define (db:set-var dbstruct var val)
  (let ((dbdat (db:get-db dbstruct #f))
	(db    (db:dbdat-get-db dbdat)))
    (db:delay-if-busy dbdat)
    (sqlite3:execute db "INSERT OR REPLACE INTO metadat (var,val) VALUES (?,?);" var val)))


(define (db:del-var dbstruct var)

  ;; (db:delay-if-busy)
  (db:with-db dbstruct #f #t 
	      (lambda (db)
		(sqlite3:execute db "DELETE FROM metadat WHERE var=?;" var))))


;; use a global for some primitive caching, it is just silly to
;; re-read the db over and over again for the keys since they never
;; change

;; why get the keys from the db? why not get from the *configdat*
;; using keys:config-get-fields?

(define (db:get-keys dbstruct)
  (if *db-keys* *db-keys* 
      (let ((res '()))
	(db:with-db dbstruct #f #f
		    (lambda (db)
		      (sqlite3:for-each-row 
		       (lambda (key)
			 (set! res (cons key res)))
		       db
		       "SELECT fieldname FROM keys ORDER BY id DESC;")))
	(set! *db-keys* res)
	res)))

;; look up values in a header/data structure
(define (db:get-value-by-header row header field)

  (if (null? header) #f
      (let loop ((hed (car header))
		 (tal (cdr header))
		 (n   0))
	(if (equal? hed field)
	    (vector-ref row n)
	    (if (null? tal) #f (loop (car tal)(cdr tal)(+ n 1)))))))

;; Accessors for the header/data structure
;; get rows and header from 
(define (db:get-header vec)(vector-ref vec 0))
(define (db:get-rows   vec)(vector-ref vec 1))

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

(define (db:get-run-name-from-id dbstruct run-id)
  (db:with-db 
   dbstruct
   #f ;; this is for the main runs db
   #f ;; does not modify db
   (lambda (db)
     (let ((res #f))
       (sqlite3:for-each-row
	(lambda (runname)
	  (set! res runname))
	db
	"SELECT runname FROM runs WHERE id=?;"
	run-id)
       res))))

(define (db:get-run-key-val dbstruct run-id key)
  (db:with-db
   dbstruct
   #f
   #f
   (lambda (db)
     (let ((res #f))
       (sqlite3:for-each-row
	(lambda (val)
	  (set! res val))
	db
	(conc "SELECT " key " FROM runs WHERE id=?;")
	run-id)
       res))))

;; keys list to key1,key2,key3 ...
(define (runs:get-std-run-fields keys remfields)
  (let* ((header    (append keys remfields))
	 (keystr    (conc (keys->keystr keys) ","
			  (string-intersperse remfields ","))))
    (list keystr header)))