Megatest

# make install CSCOPTS='-accumulate-profile -profile-name $(PWD)/profile-ww$(shell date +%V.%u)'
# rm <files>.o ; make install CSCOPTS='-profile' ; ... ;  chicken-profile | less
SHELL=/bin/bash
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		\
           process.scm runs.scm tasks.scm tests.scm genexample.scm	\
           http-transport.scm filedb.scm tdb.scm client.scm mt.scm	\
           ezsteps.scm lock-queue.scm sdb.scm rmt.scm api.scm		\
           subrun.scm portlogger.scm archive.scm env.scm		\
           diff-report.scm cgisetup/models/pgdb.scm

# module source files
MSRCFILES = 

# ftail.scm rmtmod.scm commonmod.scm removed
# MSRCFILES = ducttape-lib.scm pkts.scm stml2.scm cookie.scm mutils.scm	\
#             mtargs.scm commonmod.scm dbmod.scm adjutant.scm ulex.scm	\
#             rmtmod.scm apimod.scm




GUISRCF = dashboard-context-menu.scm dashboard-tests.scm		\
          dashboard-guimonitor.scm gutils.scm dcommon.scm tree.scm	\
          vg.scm

OFILES   = $(SRCFILES:%.scm=%.o)
GOFILES  = $(GUISRCF:%.scm=%.o)

MOFILES = $(addprefix mofiles/,$(MSRCFILES:%.scm=%.o))
# compiled import files
MOIMPFILES = $(MSRCFILES:%.scm=%.import.o)

%.import.o : %.import.scm
	csc $(CSCOPTS) -unit $*.import -c $*.import.scm -o $*.import.o

# I'm not sure the cp is a good idea, changed a lot of things and it may not have been necessary...
# mofiles/%.o  %.import.scm : %.scm megatest-fossil-hash.scm
#	@[ -e mofiles ] || mkdir -p mofiles
#	csc $(CSCOPTS) -I mofiles -I $* -J -c $< -o $*.o
#	cp $*.o mofiles/$*.o
#	@touch $*.import.scm # ensure it is touched after the .o is made









mofiles/%.o : %.scm














	mkdir -p mofiles




















	csc $(CSCOPTS) -J -c $< -o mofiles/$*.o


ADTLSCR=mt_laststep mt_runstep mt_ezstep
HELPERS=$(addprefix $(PREFIX)/bin/,$(ADTLSCR))
DEPLOYHELPERS=$(addprefix deploytarg/,$(ADTLSCR))
MTESTHASH=$(shell fossil info|grep checkout:| awk '{print $$2}')

ifeq ($(MTESTHASH),)
$(error MTESTHASH is broken!)
endif

# CSIPATH=$(shell which csi)
# CKPATH=$(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=$(shell if [[ -e /usr/bin/sw_vers ]]; then /usr/bin/sw_vers -productVersion; else lsb_release -sr; fi)
# ARCHSTR=$(shell bash -c "echo \$$MACHTYPE")

PNGFILES = $(shell cd docs/manual;ls *png)

# all : $(PREFIX)/bin/.$(ARCHSTR) mtest dboard mtut ndboard
all : $(PREFIX)/bin/.$(ARCHSTR) mtest dboard mtut tcmt

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

showmtesthash:
	@echo $(MTESTHASH)

dboard : $(OFILES) $(GOFILES) dashboard.scm $(MOFILES) $(MOIMPFILES) megatest-version.scm megatest-fossil-hash.scm
	csc $(CSCOPTS) $(OFILES) dashboard.scm $(GOFILES) $(MOFILES) $(MOIMPFILES) -o dboard

mtut: $(OFILES) $(MOFILES) megatest-fossil-hash.scm mtut.scm megatest-version.scm
	csc $(CSCOPTS) $(OFILES) $(MOFILES) mtut.scm -o mtut

# include makefile.inc

TCMTOBJS = \
	api.o \
	archive.o \

	tests.o \
	subrun.o \
        ezsteps.o

#        mofiles/rmtmod.o \
#        mofiles/commonmod.o \

tcmt : $(TCMTOBJS) tcmt.scm megatest-version.scm
	csc $(CSCOPTS) $(TCMTOBJS) $(MOFILES) $(MOIMPFILES) tcmt.scm -o tcmt

# install documentation to $(PREFIX)/docs
# DOES NOT REBUILD DOCS
#
$(PREFIX)/share/docs/megatest_manual.html : docs/manual/megatest_manual.html
	mkdir -p $(PREFIX)/share/docs
	$(INSTALL) docs/manual/megatest_manual.html $(PREFIX)/share/docs/megatest_manual.html
	for png in $(PNGFILES);do $(INSTALL) docs/manual/$$png $(PREFIX)/share/docs/$$png;done

# add a fake dependency so this doens't copy everytime
$(PREFIX)/share/js/jquery-3.1.0.slim.min.js : # .fslckout
	mkdir -p $(PREFIX)/share/js
	fossil wiki export java-script-lib > $(PREFIX)/share/js/jquery-3.1.0.slim.min.js

$(PREFIX)/share/db/mt-pg.sql : mt-pg.sql
	mkdir -p $(PREFIX)/share/db
	$(INSTALL) mt-pg.sql $(PREFIX)/share/db/mt-pg.sql

# Special dependencies for the includes
$(MOFILE) $(MOIMPFILES) : megatest-fossil-hash.scm

# common.o : mofiles/commonmod.o megatest-fossil-hash.scm
mofiles/dbmod.o : mofiles/configfmod.o

# commonmod.o dashboard.o megatest.o tcmt.o apimod.o : megatest-fossil-hash.scm

tests.o db.o launch.o runs.o dashboard-tests.o				\
dashboard-context-menu.o dashboard-guimonitor.o dashboard-main.o	\
monitor.o dashboard.o archive.o megatest.o : db_records.scm megatest-fossil-hash.scm

tests.o runs.o dashboard.o dashboard-tests.o dashboard-context-menu.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 megatest-version.scm

rmt.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 : common_records.scm megatest-version.scm

common_records.scm : altdb.scm

# mofiles/stml2.o : mofiles/cookie.o
# configf.o : mofiles/commonmod.o

vg.o dashboard.o : vg_records.scm megatest-version.scm

dcommon.o : run_records.scm

mofiles/stml2.o : mofiles/cookie.o

# # special include based modules
# mofiles/pkts.o      : pkts/pkts.scm
# mofiles/stml2.o     : cookie.o
# # mofiles/mtargs.o    : mtargs/mtargs.scm
# # mofiles/mtconfigf.o : mtconfigf/mtconfigf.scm
# mofiles/ulex.o      : ulex/ulex.scm
# mofiles/mutils.o    : mutils/mutils.scm
# mofiles/cookie.o    : stml2/cookie.scm
# mofiles/stml2.o     : stml2/stml2.scm

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

# for the modularized stuff
mofiles/rmtmod.o : mofiles/commonmod.o

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 $(MOFILES)
	csc $(CSCOPTS) -c $< $(MOFILES)

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

$(PREFIX)/bin/.$(ARCHSTR)/lib/libxcb-xlib.so.0 : lib/libxcb-xlib.so.0
	if [[ $(ARCHSTR) == 12.5 ]]; then \
	mkdir -p $(PREFIX)/bin/.$(ARCHSTR)/lib; \
	$(INSTALL) lib/libxcb-xlib.so.0 $(PREFIX)/bin/.$(ARCHSTR)/lib/libxcb-xlib.so.0; \
        fi

install : $(PREFIX)/bin/.$(ARCHSTR) $(PREFIX)/bin/.$(ARCHSTR)/mtest $(PREFIX)/bin/megatest \
          $(PREFIX)/bin/.$(ARCHSTR)/dboard $(PREFIX)/bin/dashboard $(HELPERS) $(PREFIX)/bin/nbfake \
          $(PREFIX)/bin/.$(ARCHSTR)/mtexec $(PREFIX)/bin/mtexec $(PREFIX)/bin/serialize-env \
	  $(PREFIX)/bin/nbfind $(PREFIX)/bin/mtrunner $(PREFIX)/bin/viewscreen $(PREFIX)/bin/mt_xterm \
	  $(PREFIX)/share/docs/megatest_manual.html $(PREFIX)/bin/remrun \
	  $(PREFIX)/share/docs/megatest_manual.html $(PREFIX)/bin/remrun $(PREFIX)/bin/mtutil \
	  $(PREFIX)/bin/tcmt $(PREFIX)/share/db/mt-pg.sql \
          $(PREFIX)/share/js/jquery-3.1.0.slim.min.js \
          $(PREFIX)/bin/.$(ARCHSTR)/lib/libpangox-1.0.so \
          $(PREFIX)/bin/.$(ARCHSTR)/lib/libpangox-1.0.so.0 \
          $(PREFIX)/bin/.$(ARCHSTR)/lib/libxcb-xlib.so.0
#         $(PREFIX)/bin/.$(ARCHSTR)/ndboard


# $(PREFIX)/bin/newdashboard






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

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

#======================================================================
# Make the records files
#======================================================================

# vg_records.scm : records.sh
#	./records.sh

#======================================================================
# Deploy section (not complete yet)
#======================================================================

$(DEPLOYHELPERS) : utils/mt_*
	$(INSTALL) $< $@
	chmod a+X $@

deploytarg/apropos.so : Makefile
	chicken-install -p deploytarg -deploy -keep-installed $(EGGS)

deploy : deploytarg/mtest deploytarg/dboard $(DEPLOYHELPERS) deploytarg/nbfake deploytarg/remrun deploytarg/viewsceen deploytarg/nbfind deploytarg/apropos.so

# puts deployed megatest in directory "megatest"
deploytarg/mtest : $(OFILES) megatest.o deploytarg/apropos.so
	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

datashare-testing/sd : datashare.scm $(OFILES)
	csc $(CSCOPTS) datashare.scm $(OFILES) -o datashare-testing/sd

datashare-testing/sdat: sharedat.scm $(OFILES)
	csc $(CSCOPTS) sharedat.scm $(OFILES) -o datashare-testing/sdat

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 &)

datashare-testing/spublish : spublish.scm $(OFILES) megatest-version.scm
	csc $(CSCOPTS) spublish.scm margs.o process.o common.o  -o datashare-testing/spublish

datashare-testing/sretrieve : sretrieve.scm $(OFILES) megatest-version.scm
	csc $(CSCOPTS) sretrieve.scm margs.o process.o common.o  -o datashare-testing/sretrieve


datashare-testing/sauthorize : sauthorize.scm $(OFILES) megatest-version.scm
	 csc $(CSCOPTS) sauthorize.scm margs.o process.o common.o  -o datashare-testing/sauthorize

sauth-init:
	mkdir -p  datashare-testing
	rm  datashare-testing/sauthorize
	rm  datashare-testing/sretrieve 
	rm  datashare-testing/spublish

sauth : sauth-init datashare-testing/sauthorize datashare-testing/sretrieve datashare-testing/spublish 

readline-fix.scm :
	if [[ $(shell chicken-status | grep readline | awk '{print $4}' | cut -d. -f1) -gt 3 ]];then \
	   echo "(define *use-new-readline* #f)" > readline-fix.scm; \
	else \
	   echo "(define *use-new-readline* #t)" > readline-fix.scm;\
	fi

altdb.scm :
	echo ";; optional alternate db setup" > altdb.scm
	echo "(define *available-db* (make-hash-table))" >> altdb.scm
	if  csi -ne '(use mysql-client)';then \
           echo "(use mysql-client)(hash-table-set! *available-db* 'mysql #t)" >> altdb.scm; \
	fi
	if csi -ne '(use postgresql)';then \
	   echo "(use postgresql)(hash-table-set! *available-db* 'postgresql #t)" >> altdb.scm;\
	fi

portlogger-example : portlogger-example.scm api.o archive.o client.o common.o configf.o dashboard-tests.o dashboard-context-menu.o db.o dcommon.o ezsteps.o filedb.o genexample.o gutils.o http-transport.o items.o keys.o launch.o lock-queue.o margs.o mt.o ods.o portlogger.o process.o rmt.o runconfig.o runs.o sdb.o server.o synchash.o tasks.o tdb.o tests.o tree.o
	csc $(CSCOPTS) portlogger-example.scm api.o archive.o client.o common.o configf.o dashboard-tests.o dashboard-context-menu.o db.o dcommon.o ezsteps.o filedb.o genexample.o gutils.o http-transport.o items.o keys.o launch.o lock-queue.o margs.o mt.o ods.o portlogger.o process.o rmt.o runconfig.o runs.o sdb.o server.o synchash.o tasks.o tdb.o tests.o tree.o

# create a pdf dot graphviz diagram from notations in rmt.scm
rmt.pdf : rmt.scm
	grep ';;DOT' rmt.scm | sed -e 's/.*;;DOT //' > rmt.dot;dot -Tpdf rmt.dot -o rmt.pdf

buildmanual:
	cd docs/manual && make

wikipage=plan
editwiki:
	cd docs/manual && ../../utils/editwiki $(wikipage)

viewmanual:
	arora docs/manual/megatest_manual.html

targets:
	@grep : Makefile | perl -ne '/^([A-Za-z0-9_-]+):/ && print "$$1\n"'


unit :
	cd tests;make unit

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

(declare (unit adjutant))

(module adjutant *

(import scheme chicken data-structures extras files)
(import (prefix sqlite3 sqlite3:) posix typed-records srfi-18 srfi-69
	md5 message-digest
	regex srfi-1)

(define (adjutant-run)
  (print "Running the adjutant!"))

)

;;======================================================================
;; Copyright 2006-2013, Matthew Welland.
;; 
;; This file is part of Megatest.
;; 
;;     Megatest is free software: you can redistribute it and/or modify
;;     it under the terms of the GNU General Public License as published by
;;     the Free Software Foundation, either version 3 of the License, or
;;     (at your option) any later version.
;; 
;;     Megatest is distributed in the hope that it will be useful,
;;     but WITHOUT ANY WARRANTY; without even the implied warranty of
;;     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
;;     GNU General Public License for more details.
;; 
;;     You should have received a copy of the GNU General Public License
;;     along with Megatest.  If not, see <http://www.gnu.org/licenses/>.
;;
;;======================================================================

(use srfi-69 posix)

(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-var
    get-keys
    get-key-vals
    test-toplevel-num-items
    get-test-info-by-id
    get-steps-info-by-id
    get-data-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
    get-run-state
    get-run-stats
    get-run-times
    get-targets
    get-target
    ;; register-run
    get-tests-tags
    get-test-times
    get-tests-for-run
    get-tests-for-run-state-status
    get-test-id
    get-tests-for-runs-mindata
    get-tests-for-run-mindata
    get-run-name-from-id
    get-runs
    simple-get-runs
    get-num-runs
    get-runs-cnt-by-patt
    get-all-run-ids
    get-prev-run-ids
    get-run-ids-matching-target
    get-runs-by-patt
    get-steps-data
    get-steps-for-test
    read-test-data
    read-test-data-varpatt
    login
    tasks-get-last
    testmeta-get-record
    have-incompletes?
    ;; synchash-get
    get-changed-record-ids
    get-run-record-ids 
    get-not-completed-cnt))

(define api:write-queries
  '(
    get-keys-write ;; dummy "write" query to force server start

    ;; SERVERS
    start-server
    kill-server

    ;; TESTS
    test-set-state-status-by-id
    delete-test-records
    delete-old-deleted-test-records
    test-set-state-status
    test-set-top-process-pid
    set-state-status-and-roll-up-items
    
    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
    set-state-status-and-roll-up-run
    ;; STEPS
    teststep-set-status!
    delete-steps-for-test
    ;; TEST DATA
    test-data-rollup
    csv->test-data

    ;; MISC
    sync-inmem->db
    drop-all-triggers
    create-all-triggers
    update-tesdata-on-repilcate-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)))
     (debug:print 0 *default-log-port* "WARNING: api:execute-requests received an exception from peer, dat=" dat ", exn=" exn)
     (print-call-chain (current-error-port))
     (debug:print 0 *default-log-port* " message: "  ((condition-property-accessor 'exn 'message) exn))       
     (vector #f (vector exn call-chain dat))) ;; return some stuff for debug if an exception happens
   (cond
    ((not (vector? dat))                    ;; it is an error to not receive a vector
     (vector #f (vector #f "remote must be called with a vector")))
    ((> *api-process-request-count* 20) ;; 20)
     (debug:print 0 *default-log-port* "WARNING: api:execute-requests received an overloaded message.")
     (set! *server-overloaded* #t)
     (vector #f (vector #f 'overloaded))) ;; the inner vector is what gets returned. nope, don't know why. please refactor!
    (else  
     (let* ((cmd-in            (vector-ref dat 0))
            (cmd               (if (symbol? cmd-in)
				   cmd-in
				   (string->symbol cmd-in)))
            (params            (vector-ref dat 1))
            (start-t           (current-milliseconds))
            (readonly-mode     (dbr:dbstruct-read-only dbstruct))
            (readonly-command  (member cmd api:read-only-queries))
            (writecmd-in-readonly-mode (and readonly-mode (not readonly-command)))
            #;(foo               (begin
                                 (common:telemetry-log (conc "api-in:"(->string cmd))
                                                       payload: `((params . ,params)))
                                 
                                 #t))
            (res    
             (if writecmd-in-readonly-mode
                 (conc "attempt to run write command "cmd" on a read-only database")
                 (case cmd
                   ;;===============================================
                   ;; READ/WRITE QUERIES
                   ;;===============================================

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

                   ;; TESTS

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

                    ;; (define (db:set-state-status-and-roll-up-items dbstruct run-id test-name item-path state status comment)
                    (db:set-state-status-and-roll-up-items
                     dbstruct
                     (list-ref params 0) ; run-id
                     (list-ref params 1) ; test-name
                     #f                  ; item-path
                     (list-ref params 2) ; state
                     (list-ref params 3) ; status
                     (list-ref params 4) ; comment
                     ))
                   
                   ((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-state-status)           (apply db:test-set-state-status dbstruct params))
                   ((test-set-top-process-pid)        (apply db:test-set-top-process-pid dbstruct params))
                   ((set-state-status-and-roll-up-items) (apply db:set-state-status-and-roll-up-items dbstruct params))
                   ((set-state-status-and-roll-up-run) (apply db:set-state-status-and-roll-up-run dbstruct 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))
                   ((update-run-stats)             (apply db:update-run-stats dbstruct params))
                   ((set-var)                      (apply db:set-var dbstruct params))
                   ((inc-var)                      (apply db:inc-var dbstruct params))
		   ((dec-var)                      (apply db:dec-var dbstruct params))
                   ((del-var)                      (apply db:del-var dbstruct params))
		   ((add-var)                      (apply db:add-var dbstruct params))

                   ;; STEPS
                   ((teststep-set-status!)         (apply db:teststep-set-status! dbstruct params))
                   ((delete-steps-for-test!)        (apply db:delete-steps-for-test! 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))
                   ((create-all-triggers)          (db:create-all-triggers dbstruct))
                   ((drop-all-triggers)            (db:drop-all-triggers dbstruct)) 

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

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

		   ;; NO SYNC DB
		   ((no-sync-set)               (apply db:no-sync-set         *no-sync-db* params))
		   ((no-sync-get/default)       (apply db:no-sync-get/default *no-sync-db* params))
		   ((no-sync-del!)              (apply db:no-sync-del!        *no-sync-db* params))
		   ((no-sync-get-lock)          (apply db:no-sync-get-lock    *no-sync-db* 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-target)                      (apply db:get-target 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))
                   ((get-not-completed-cnt)           (apply db:get-not-completed-cnt  dbstruct params)) 
                   ;; ((synchash-get)                    (apply synchash:server-get dbstruct params))
                   ((get-raw-run-stats)               (apply db:get-raw-run-stats dbstruct params))
		   ((get-test-times)                  (apply db:get-test-times dbstruct params))

                   ;; RUNS
                   ((get-run-info)                 (apply db:get-run-info dbstruct params))
                   ((get-run-status)               (apply db:get-run-status dbstruct params))
                   ((get-run-state)                (apply db:get-run-state dbstruct params))
                   ((set-run-status)               (apply db:set-run-status dbstruct params))
                   ((set-run-state-status)  			 (apply db:set-run-state-status dbstruct params))
                   ((update-tesdata-on-repilcate-db) (apply db:update-tesdata-on-repilcate-db  dbstruct params)) 
                   ((get-tests-for-run)            (apply db:get-tests-for-run dbstruct params))
                   ((get-tests-for-run-state-status) (apply db:get-tests-for-run-state-status 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-tests-for-runs-mindata)   (apply db:get-tests-for-runs-mindata dbstruct params))
                   ((get-runs)                     (apply db:get-runs dbstruct params))
                   ((simple-get-runs)              (apply db:simple-get-runs dbstruct params))
                   ((get-num-runs)                 (apply db:get-num-runs dbstruct params))
                   ((get-runs-cnt-by-patt)         (apply db:get-runs-cnt-by-patt 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))
                   ((get-main-run-stats)           (apply db:get-main-run-stats dbstruct params))
                   ((get-var)                      (apply db:get-var dbstruct params))
                   ((get-run-stats)                (apply db:get-run-stats dbstruct params))
                   ((get-run-times)                (apply db:get-run-times 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))
		   ((get-steps-info-by-id)         (apply db:get-steps-info-by-id dbstruct params))

                   ;; TEST DATA
                   ((read-test-data)               (apply db:read-test-data dbstruct params))
                   ((read-test-data-varpatt)       (apply db:read-test-data-varpatt dbstruct params))
                   ((get-data-info-by-id)          (apply db:get-data-info-by-id dbstruct params)) 

                   ;; MISC
                   ((get-latest-host-load)         (apply db:get-latest-host-load dbstruct params))
                   ((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:general-call dbstruct stmtname realparams)))
                   ((sdb-qry)                      (apply sdb:qry params))
                   ((ping)                         (current-process-id))
		   ((get-changed-record-ids)       (apply db:get-changed-record-ids dbstruct params))
		   ((get-run-record-ids) 	   (apply db:get-run-record-ids dbstruct params))	
                   ;; TESTMETA
                   ((testmeta-get-record)       (apply db:testmeta-get-record dbstruct params))

                   ;; TASKS 
                   ((find-task-queue-records)   (apply tasks:find-task-queue-records dbstruct params))
		   (else
		    (debug:print 0 *default-log-port* "ERROR: bad api call " cmd)
		    (conc "ERROR: BAD api call " cmd))))))

       
       ;; save all stats
       (let ((delta-t (- (current-milliseconds)
			 start-t)))
	 (hash-table-set! *db-api-call-time* cmd
			  (cons delta-t (hash-table-ref/default *db-api-call-time* cmd '()))))
       (if writecmd-in-readonly-mode
           (begin
             #;(common:telemetry-log (conc "api-out:"(->string cmd))
                                   payload: `((params . ,params)
                                              (ok-res . #t)))
	     (vector #f res))
           (begin
             #;(common:telemetry-log (conc "api-out:"(->string cmd))
                                   payload: `((params . ,params)
                                              (ok-res . #f)))
             (vector #t res))))))))

;; 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
  (debug:print 4 *default-log-port* "server-id:"  *server-id*)
  (let* ((cmd     ($ 'cmd))
	 (paramsj ($ 'params))
         (key     ($ 'key))   
	 (params  (db:string->obj paramsj transport: 'http))) ;; incoming data from the POST (or is it a GET?)
    (debug:print 4 *default-log-port* "cmd:" cmd " with params " params "key " key)
    (if (equal? key *server-id*)
      (begin
        (set! *api-process-request-count* (+ *api-process-request-count* 1))
 	(let* ((resdat  (api:execute-requests dbstruct (vector cmd params))) ;; process the request, resdat = #( flag result )
	       (success (vector-ref resdat 0))
	       (res     (vector-ref resdat 1))) ;; (vector flag payload), get the payload, ignore the flag (why?)
	  (debug:print 4 *default-log-port* "res:" res)
	  (if (not success)
	      (debug:print 0 *default-log-port* "ERROR: success flag is #f for " cmd " with params " params))
	  (if (> *api-process-request-count* *max-api-process-requests*)
	      (set! *max-api-process-requests* *api-process-request-count*))
	  (set! *api-process-request-count* (- *api-process-request-count* 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)))
	(begin
	  (debug:print 0 *default-log-port*   "Server refused to process request. Sever id mismatch. recived " key " expected:  " *server-id* ".\nOther arguments recived: cmd=" cmd " params = " params) 
	  (db:obj->string (conc "Server refused to process request server-id mismatch: " key ", " *server-id*) transport: 'http)))))

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

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

(declare (unit apimod))
(declare (uses commonmod))

(declare (uses ulex))



(module apimod
	*
	
(import scheme chicken data-structures extras)








(import (prefix sqlite3 sqlite3:) posix typed-records srfi-18)



































































































































































































(import commonmod)












(import (prefix ulex ulex:))



























































































































































































)

;; Copyright 2006-2014, Matthew Welland.
;; 
;; This file is part of Megatest.
;; 
;;     Megatest is free software: you can redistribute it and/or modify
;;     it under the terms of the GNU General Public License as published by
;;     the Free Software Foundation, either version 3 of the License, or
;;     (at your option) any later version.
;; 
;;     Megatest is distributed in the hope that it will be useful,
;;     but WITHOUT ANY WARRANTY; without even the implied warranty of
;;     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
;;     GNU General Public License for more details.
;; 
;;     You should have received a copy of the GNU General Public License
;;     along with Megatest.  If not, see <http://www.gnu.org/licenses/>.
;;

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

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

(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 (common: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 #f #f #f)))) ;; BROKEN. testname itempath))))

;; allocate a new archive area
;;
(define (archive:allocate-new-archive-block blockid-cache run-area-home testsuite-name dneeded target run-name test-name)
  (let ((key (conc testsuite-name "/" target "/" run-name "/" test-name)))
    (if (hash-table-exists? blockid-cache key)
	(hash-table-ref blockid-cache key)
	(let* ((pscript     (configf:lookup *configdat* "archive" "pathscript"))
	       (pscript-cmd (conc pscript " " testsuite-name " " target " " run-name " " test-name))
	       (apath       (if pscript
				(handle-exceptions
				 exn
				 (begin
				   (debug:print 0 *default-log-port* "ERROR: script \"" pscript-cmd "\" failed to run properly. exn=" exn)
				   (exit 1))
				 (with-input-from-pipe
				  pscript-cmd
				  read-line))
				#f)) ;; this is the user-calculated archive path
	       (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  (if apath
					apath
					(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)
		    (let ((res (cons block-id archive-path)))
		      (hash-table-set! blockid-cache key res)
		      res)
		    (begin
		      (debug:print 0 *default-log-port* "WARNING: no disk found for " target ", " run-name ", " test-name ",  archive-path=" archive-path)
		      #f)))
	      (begin
		(debug:print 0 *default-log-port* "WARNING: no disk found for " target ", " run-name ", " test-name )
		#f)))))) ;; no best disk found

;; 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* ((blockid-cache  (make-hash-table))
	 (tsname         (common:get-testsuite-name))
         (target         (string-intersperse (map cadr (rmt:get-key-val-pairs run-id)) "/"))
	 (min-space      (string->number (or (configf:lookup *configdat* "archive" "minspace") "1000")))
	 (arch-groups    (make-hash-table)) ;; archive groups, each corrosponds to a bup area
	 (disk-groups    (make-hash-table)) ;; 
	 (test-groups    (make-hash-table)) ;; these two (disk and test groups) could be combined nicely
	 (test-dirs      (make-hash-table))
	 (bup-exe        (or (configf:lookup *configdat* "archive" "bup") "bup"))
	 (compress       (or (configf:lookup *configdat* "archive" "compress") "9"))
	 (linktree       (common:get-linktree)) ;; (configf:lookup *configdat* "setup" "linktree")))
	 (archiver       (let ((s (configf:lookup *configdat* "archive" "archiver")))
			   (if s (string->symbol s) 'bup)))
	 (archiver-cmd   (case archiver
			   ((tar) "tar cfj ARCHIVE_NAME.tar.bz2 ")
			   ((7z)  " 7z u -t7z -m0=lzma -mx=9 -mfb=64 -md=32m -ms=on ARCHIVE_NAME.7z ")
			   (else #f)))
         (src-archive-linktree (rmt:get-var "src-archive-linktree"))  
	 (print-prefix      "Running: ") ;; change to #f to turn off printing
	 (preclean-spec  (configf:get-section *configdat* "archive-preclean")))

     (if (or (not src-archive-linktree) (not (equal? src-archive-linktree linktree)))      
         (rmt:set-var "src-archive-linktree"  linktree))
    ;;     (tests:match patt testname itempath)
    
    ;; 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))
	      	      
	      (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 (common: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))
	      ;; we need our archive dir checked for every test to enable folks who want to store other ways.
	      (archive-info (archive:allocate-new-archive-block blockid-cache *toppath* tsname min-space target run-name test-name))
	      (archive-dir  (if archive-info (cdr archive-info) #f))
	      (archive-id   (if archive-info (car archive-info) -1)))
	 
	 (if (not archive-dir) ;; no archive disk found, this is fatal
	     (begin
	       (debug:print 0 *default-log-port* "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 *default-log-port* "       use [archive] minspace to specify minimum available space")
	       (debug:print 0 *default-log-port* "   disks: "
			    (string-intersperse (map cadr (archive:get-archive-disks)) "\n         "))
	       (exit 1))
	     (debug:print-info 0 *default-log-port* "Using path " archive-dir " for archiving test " test-path))

	 ;; preclean the test directory per the spec if provided
	 (if (not (null? preclean-spec)) ;; we've been asked to preclean before archiving
	     (let loop ((spec (car preclean-spec))
			(tail (cdr preclean-spec)))
	       (if (> (length spec) 1)
		   (let ((testspec (car spec))
			 (rules    (cadr spec)))
		     (if (tests:match testspec test-name item-path)
			 (begin
			   (debug:print 0 *default-log-port* "INFO: cleanup requested for " test-physical-path)
			   (common:dir-clean-up test-physical-path rules remove-empty: #t))
			 (if (not (null? tail))
			     (loop (car tail)(cdr tail)))))
		   (begin
		     (debug:print 0 *default-log-port* "ERROR: bad spec line in [archive-preclean] section. \"" spec "\"")
		     (if (not (null? tail))(loop (car tail)(cdr tail)))))))
	 (cond
	  (toplevel/children
	   (debug:print 0 *default-log-port* "WARNING: cannot archive " test-name " with id " test-id
			" as it is a toplevel test with children"))
	  ((not (common:file-exists? test-path))
	   (debug:print 0 *default-log-port* "WARNING: Cannot archive " test-name "/" item-path
			" as path " test-path " does not exist"))
	  (else
	   (debug:print 2 *default-log-port*
			"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 '())))
	   (hash-table-set! arch-groups test-base
			    (cons archive-info (hash-table-ref/default arch-groups test-base '())))
	   (hash-table-set! test-dirs test-id test-path)))))
	   ;; test-path))))
     tests)
    (debug:print 0 *default-log-port* "INFO: DISK GROUPS=" (hash-table->alist disk-groups))
    ;; for each disk-group, initialize the bup area if needed
    (for-each 
     (lambda (test-base)
       (let* ((disk-group (hash-table-ref disk-groups test-base))
	      (arch-group (hash-table-ref arch-groups test-base))
	      (arch-info  (car arch-group)) ;; don't know yet how this will work, can I get more than one possibility?
	      (archive-id    (car arch-info))
	      (archive-dir   (cdr arch-info)))
	 (debug:print 0 *default-log-port* "Processing disk-group " test-base)
	 (let* ((test-paths-in (hash-table-ref disk-groups test-base))
		(test-paths    (if (args:get-arg "-include")
				   (let ((subpaths (string-split (args:get-arg "-include") ",")))
				     (apply append
					    (map (lambda (p)
						   (map (lambda (subp)
							  (conc p "/" subp))
							subpaths))
						 test-paths-in)))
				   test-paths-in)))
	   (if (not (common:file-exists? archive-dir))
	       (create-directory archive-dir #t))
	   (case archiver
	     ((bup) ;; Archive using bup
	      (let* ((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) "-"(string-substitute "/" "-" target " "))
						     (conc "--strip-path=" (conc test-base target "/" )) ;; if we push to the directory do we need this?
						     )
					       test-paths)))
		(if (not (common:file-exists? (conc archive-dir "/HEAD")))
		    (begin
		      ;; replace this with jobrunner stuff enventually
		      (debug:print-info 2 *default-log-port* "Init bup in " archive-dir)
		      ;; (mutex-lock! bup-mutex)
		      (let-values (((pid-val exit-status exit-code) (run-n-wait bup-exe params: bup-init-params print-cmd: print-prefix)))
                        (if (not (eq? exit-code 0))
                              (begin    
                              (debug:print-error 0 *default-log-port* "There was an error initializing bup. Archive failed.")
                              (exit 1))))
		      ;; (mutex-unlock! bup-mutex)
		      ))
		(debug:print-info 2 *default-log-port* "Indexing data to be archived")
		;; (mutex-lock! bup-mutex)
		(let-values (((pid-val exit-status exit-code) (run-n-wait bup-exe params: bup-index-params print-cmd: print-prefix)))
                   (if (not (eq? exit-code 0))
                              (begin    
                              (debug:print-error 0 *default-log-port* "There was an error Indexing bup. Archive failed.")
                              (exit 1))))
		(debug:print-info 2 *default-log-port* "Archiving data with bup")
		(let-values (((pid-val exit-status exit-code) (run-n-wait bup-exe params: bup-save-params print-cmd: print-prefix)))
                     (if (not (eq? exit-code 0))
                              (begin    
                              (debug:print-error 0 *default-log-port* "There was an archiving data with bup. Archive failed.")
                              (exit 1))))))
	     ((7z tar)
	      (for-each
	       (lambda (test-dat)
		 (let* ((test-id           (db:test-get-id        test-dat))
			(test-name         (db:test-get-testname  test-dat))
			(item-path         (db:test-get-item-path test-dat))
			(test-full-name    (db:test-make-full-name test-name item-path))
			(run-id            (db:test-get-run_id    test-dat))
			(target            (string-intersperse (map cadr (rmt:get-key-val-pairs run-id)) "/"))
			(run-name          (rmt:get-run-name-from-id run-id))
			(source-dir        (hash-table-ref test-dirs test-id)) ;; (conc test-base "/" test-name "/" item-path))
			(target-dir        (string-substitute "/$" "" (conc archive-dir "/" target "/" run-name "/" test-full-name))))
		   ;; create the test and item-path levels under archive-dir
		   (create-directory (pathname-directory target-dir) #t)
		   (run-n-wait
		    (conc
		     (string-substitute "ARCHIVE_NAME" target-dir archiver-cmd) " "
		     "."
		     )
		    print-cmd: print-prefix
		    run-dir: source-dir)))
	       (hash-table-ref test-groups test-base))))
	   ;; (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 (symbol->string archive-command) '("save-remove"))
                    (begin 
                     (debug:print-info 0 *default-log-port* "remove testdat")
		    (runs:remove-test-directory test-dat 'archive-remove)))))
	    (hash-table-ref test-groups test-base)))))
       (hash-table-keys disk-groups))
    #t))

(define (archive:megatest-db target-patt run-patt)
 (let* ((blockid-cache  (make-hash-table))
        (tsname         (common:get-testsuite-name))
        (min-space      (string->number (or (configf:lookup *configdat* "archive" "minspace") "1000")))
        (bup-exe        (or (configf:lookup *configdat* "archive" "bup") "bup"))
	(compress       (or (configf:lookup *configdat* "archive" "compress") "9"))
        (archiver       (let ((s (configf:lookup *configdat* "archive" "archiver")))
			   (if s (string->symbol s) 'bup)))
        (rsync-exe (or (configf:lookup *configdat* "archive" "rsync") "rsync"))   
        (print-prefix      "Running: ") 
        (archive-info (archive:allocate-new-archive-block blockid-cache *toppath* tsname min-space target-patt run-patt "megatest-db"))
	(archive-dir  (if archive-info (cdr archive-info) #f))
	(archive-id   (if archive-info (car archive-info) -1))
        (home-host (common:get-homehost))
        (archive-time (seconds->std-time-str (current-seconds)))
        (archive-staging-db (conc *toppath* "/.db-snapshot/archive_" archive-time))
        (tmp-db-path (conc (common:get-db-tmp-area) "/megatest.db"))
        (dbfile             (conc  archive-staging-db "/megatest.db"))) 
        (create-directory archive-staging-db #t)
        (let-values (((pid-val exit-status exit-code) (run-n-wait rsync-exe params: (list "-v" (conc (car home-host) ":"tmp-db-path) archive-staging-db) print-cmd: print-prefix)))
            (if (eq? exit-code 0)   
              (case archiver
	        ((bup) ;; Archive using bup
	          (let* ((bup-init-params  (list "-d" archive-dir "init"))
		         (bup-index-params (list "-d" archive-dir "index" archive-staging-db))
		         (bup-save-params  (list "-d" archive-dir "save" ;; (conc "--strip-path=" linktree)
						     (conc "-" compress) ;; or (conc "--compress=" compress)
						     "-n" (conc tsname "-megatest-db" )
						     (conc "--strip-path=" archive-staging-db ) ;; if we push to the directory do we need this?
						     dbfile)))
                    (if (not (common:file-exists? (conc archive-dir "/HEAD")))
		      (begin
		        ;; replace this with jobrunner stuff enventually
		        (debug:print-info 2 *default-log-port* "Init bup in " archive-dir)
		         (let-values (((pid-val exit-status exit-code)(run-n-wait bup-exe params: bup-init-params print-cmd: print-prefix)))
                          (if (not (eq? exit-code 0))
                              (begin    
                              (debug:print-error 0 *default-log-port* "There was an error initializing bup. Archive failed.")
                              (exit 1))))))
		     (debug:print-info 2 *default-log-port* "Indexing data to be archived")
		     (let-values (((pid-val exit-status exit-code) (run-n-wait bup-exe params: bup-index-params print-cmd: print-prefix)))
                        (if (not (eq? exit-code 0))
                              (begin    
                              (debug:print-error 0 *default-log-port* "There was an error Indexing bup. Archive failed.")
                              (exit 1))))
		     (debug:print-info 2 *default-log-port* "Archiving data with bup")
		     (let-values (((pid-val exit-status exit-code) (run-n-wait bup-exe params: bup-save-params print-cmd: print-prefix)))
                         (if (not (eq? exit-code 0))
                              (begin    
                              (debug:print-error 0 *default-log-port* "There was an error archiving data with bup. Archive failed.")
                              (exit 1))
                             (debug:print-info 0 *default-log-port* "To restore megatest.db run megatest -archive replicacte-db -source archive-dir -time-stamp <ts>. Current timestamp: " (seconds->std-time-str (current-seconds))))))) 
               (else
                   (debug:print-info 0 *default-log-port* "No support for databse archiving with " archiver)))
               (debug:print-error 0 *default-log-port* "There was an error rsyncing tmp database")))))

(define (archive:restore-db archive-path ts)
   (let* ((bup-exe               (or (configf:lookup *configdat* "archive" "bup") "bup"))
         (archive-internal-path (conc (common:get-testsuite-name) "-megatest-db/" ts "/megatest.db" ))
         (bup-restore-params  (list "-d" archive-path "restore" "-C" *toppath* archive-internal-path)))
		 (debug:print-info 2 *default-log-port* "Restoring archived data to " *toppath* " from archive in " archive-path " ... " archive-internal-path)
		 (run-n-wait bup-exe params: bup-restore-params print-cmd: "Running:"))
      (sleep 2)
      (db:multi-db-sync 
       (db:setup #f)
       'killservers
       ;'dejunk
       ;'adj-testids
       'old2new
       )
      (debug:print-info 1 *default-log-port* "dropping triggers to update linktree") 
      (rmt:drop-all-triggers)
    (let* ((linktree       (common:get-linktree)) ;; (configf:lookup *configdat* "setup" "linktree")))
	  (src-archive-linktree (rmt:get-var "src-archive-linktree")))
        (if (not (equal? src-archive-linktree linktree))
           (rmt:update-tesdata-on-repilcate-db src-archive-linktree linktree))
              (debug:print-info 1 *default-log-port* "creating triggers after updating linktree")   
       (rmt:create-all-triggers)
))  

(define (archive:ls->list  bup-exe archive-dir internal-path)
  (let ((cmd (conc bup-exe " -d " archive-dir  " ls -l " internal-path "| awk '{print $6}' | sort"))
        (res '()))
    (debug:print-info 0 *default-log-port* cmd)
    (handle-exceptions
     exn
     #f ;; anything goes wrong - assume the process in NOT running.
     (with-input-from-pipe 
      cmd
      (lambda ()
	(let* ((inl (read-lines)))
	  (reverse inl)))))))

(define (time-string->seconds tstr ds-flag)
 (let* ((atime (string->time tstr "%Y-%m-%d-%H%M%S")))
       (vector-set! atime 8 ds-flag)
     (local-time->seconds atime)))

(define (seconds->std-time-str sec)
  (time->string 
   (seconds->local-time sec)
   "%Y-%m-%d-%H%M%S"))
 

(define (archive:get-timestamp-dir bup-exe archive-dir testsuite-name target test-partial-path test-last-update)
    (debug:print-info 0 *default-log-port* "Test last update time:" (seconds->std-time-str test-last-update)) 
    (let* ((internal-path (conc testsuite-name "-" target))
           (archive-update-delay (string->number (or (configf:lookup *configdat* "archive" "test-update-delay") "900" )))  
           (ts-list (archive:ls->list  bup-exe archive-dir internal-path))
           (ds-flag (vector-ref (seconds->local-time) 8)))
           (let loop ((hed (car ts-list))
                       (tail (cdr ts-list)))
                   (if (and (null? tail) (equal? hed "latest"))
                        #f
                    (if (and (not (null? tail)) (equal? hed "latest"))
                        (loop (car tail) (cdr tail))
                  (let* ((archive-seconds (time-string->seconds hed ds-flag)))
                    (if (< (abs (- archive-seconds test-last-update)) archive-update-delay)
                    (let* ((test-list (archive:ls->list  bup-exe archive-dir (conc internal-path "/" hed "/" test-partial-path))))
                           (if (> (length test-list) 0)
                               hed
                               (if (not (null? tail)) 
                                 (loop (car tail) (cdr tail))
                                 #f)))
                       (if (null? tail)
                          #f
                          (loop (car tail) (cdr tail))))))))))

(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     (common:get-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  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 (common: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))
              (test-last-update        (db:test-get-last_update 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-timestamp-dir   (if archive-path (archive:get-timestamp-dir bup-exe archive-path (common:get-testsuite-name) (string-substitute "/" "-" target " ") test-partial-path test-last-update) #f))  
	      (archive-internal-path   (conc (common:get-testsuite-name) "-" (string-substitute "/" "-" target " ") "/" archive-timestamp-dir "/" test-partial-path))
	      (include-paths           (args:get-arg "-include"))
	      (exclude-pattern         (args:get-arg "-exclude-rx"))
	      (exclude-file            (args:get-arg "-exclude-rx-from")))
	 (if (not archive-timestamp-dir)
               (debug:print-error 0 *default-log-port* "Archive not found for testsuite" (common:get-testsuite-name) " run/test/itempath" test-partial-path)
         (begin    
	 ;; some sanity checks, move an existing path out of the way - iif it is not a toplevel with children
         (debug:print-info 0 *default-log-port* "Archive time: " archive-timestamp-dir)
	 (if (and (not toplevel/children)  ;; special handling needed for toplevel with children
		  prev-test-physical-path
		  (common: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-error 0 *default-log-port* "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-error 0 *default-log-port* "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 *default-log-port* "Restoring archived data to " new-test-physical-path " from archive in " archive-path " ... " archive-internal-path)
                 (debug:print-info 0 *default-log-port* bup-exe " " (string-join  bup-restore-params " "))
		 ;; (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-error 0 *default-log-port* "No archive path in the record for run-id=" run-id " test-id=" test-id))))))
     (filter vector? tests))))

(define (common:get-youngest-test tests)
  (if (null? tests)
      #f
      (let ((res #f))
	(for-each
	 (lambda (test-dat)
	   (let ((event-time (db:test-get-event_time test-dat)))
	     (if (or (not res)
		     (> event-time (db:test-get-event_time res)))
		 (set! res test-dat))))
	 tests)
	res)))
	   
;; from an archive get a specific path - works ONLY with bup for now
;;
(define (archive:bup-get-data archive-command run-id-in run-name-in tests rp-mutex bup-mutex)
  (if (null? tests)
      (debug:print-info 0 *default-log-port* "get-data called with no matching tests to operate on.")
      
      (let* ((bup-exe      (or (configf:lookup *configdat* "archive" "bup") "bup"))
	     (linktree     (common:get-linktree)) ;; (configf:lookup *configdat* "setup" "linktree")))
	     ;; (test-dat     (common:get-youngest-test tests))
	     (destpath     (args:get-arg "-dest")))
	(cond
	 ((null? tests)
	  (debug:print-error 0 *default-log-port*
			     "No test matching provided target, runname pattern and test pattern found."))
	 ((file-exists? destpath)
	  (debug:print-error 0 *default-log-port*
			     "Destination path alread exists! Please remove it before running get."))
	 (else
	  (let loop ((rem-tests tests))
	    (let* ((test-dat          (common:get-youngest-test rem-tests))
		   (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))
		   (run-name          (rmt:get-run-name-from-id run-id))
		   (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))
		   (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)
						#f))
		   (archive-internal-path   (conc (common:get-testsuite-name) "-" run-id
						  "/latest/" test-partial-path))
		   (include-paths           (args:get-arg "-include"))
		   (exclude-pattern         (args:get-arg "-exclude-rx"))
		   (exclude-file            (args:get-arg "-exclude-rx-from")))
	      
	      (if (and archive-path ;; no point in proceeding if there is no actual archive
		       (not toplevel/children))
		  (begin
		    (let* ((bup-restore-params (append (list "-d" archive-path "restore" "-C" (or destpath "data"))
						       ;; " " ;; What is the empty string for?
						       (if include-paths
							   (map (lambda (p)
								  (conc archive-internal-path "/" p))
								(string-split include-paths ","))
							   (list archive-internal-path)))))
		      (debug:print-info 0 *default-log-port* "Restoring archived data to " (or destpath "data")
					" from archive in " archive-path " ... " archive-internal-path)
		      (run-n-wait bup-exe params: bup-restore-params print-cmd: #t)))
		  (let ((new-rem-tests (filter (lambda (tdat)
						 (or (not (eq? (db:test-get-id tdat) test-id))
						     (not (eq? (db:test-get-run_id tdat) run-id))))
					       rem-tests) ))
		    (debug:print-info 0 *default-log-port*
				      "No archive path in the record for run-id=" run-id
				      " test-id=" test-id ", skipping.")
		    (if (null? new-rem-tests)
			(begin
			  (debug:print-info 0 *default-log-port* "No archives found for " target "/" run-name "...")
			  #f)
			(loop new-rem-tests)))))))))))
  

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

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

(declare (unit pgdb))
(declare (uses configf))

;; I don't know how to mix compilation units and modules, so no module here.
;;
;; (module pgdb
;;     (
;;      open-pgdb
;;      )
;; 
;; (import scheme)
;; (import data-structures)
;; (import chicken)

(use typed-records (prefix dbi dbi:))

;; given a configdat lookup the connection info and open the db
;;
(define (pgdb:open configdat #!key (dbname #f)(dbispec #f))  
  (let ((pgconf (or dbispec
		    (args:get-arg "-pgsync")
		    (if configdat

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

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

(use srfi-18 extras tcp s11n srfi-1 posix regex srfi-69 hostinfo md5
     message-digest matchable spiffy uri-common intarweb http-client
     spiffy-request-vars uri-common intarweb directory-utils)

(declare (unit client))

(declare (uses common))
(declare (uses db))
(declare (uses tasks)) ;; tasks are where stuff is maintained about what is running.

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

;; 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)
  (http-transport: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 areapath #!key (remaining-tries 100) (failed-connects 0))
  (client:setup-http areapath remaining-tries: remaining-tries failed-connects: failed-connects)
  #;(case (server:get-transport)
    ((rpc) (rpc-transport:client-setup remaining-tries: remaining-tries failed-connects: failed-connects)) ;;(client:setup-rpc run-id))
    ((http)(client:setup-http areapath remaining-tries: remaining-tries failed-connects: failed-connects))
    (else  (rpc-transport:client-setup remaining-tries: remaining-tries failed-connects: failed-connects)))) ;; (client:setup-rpc run-id))))

;; 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 areapath #!key (remaining-tries 100) (failed-connects 0)(area-dat #f))
  (debug:print-info 2 *default-log-port* "client:setup remaining-tries=" remaining-tries)
  (server:start-and-wait areapath)
  (if (<= remaining-tries 0)
      (begin
	(debug:print-error 0 *default-log-port* "failed to start or connect to server")
	(exit 1))
      ;;
      ;; Alternatively here, we can get the list of candidate servers and work our way
      ;; through them searching for a good one.
      ;;
      (let* ((server-dat (server:get-rand-best areapath)) ;; (server:get-first-best areapath))
	     (runremote  (or area-dat *runremote*)))
	(if (not server-dat) ;; no server found
	    (client:setup-http areapath remaining-tries: (- remaining-tries 1))
	    (let ((host  (cadr  server-dat))
		  (port  (caddr server-dat))
                  (server-id (caddr (cddr server-dat))))
	      (debug:print-info 4 *default-log-port* "client:setup server-dat=" server-dat ", remaining-tries=" remaining-tries)
	      (if (and (not area-dat)
		       (not *runremote*))
                  (begin       
		    (set! *runremote* (make-remote))
                    (let* ((server-info (remote-server-info *runremote*))) 
                      (if server-info
                        (begin
                          (remote-server-url-set! *runremote* (server:record->url server-info))
                          (remote-server-id-set! *runremote* (server:record->id server-info)))))))
	      (if (and host port server-id)
		  (let* ((start-res (case *transport-type*
				      ((http)(http-transport:client-connect host port server-id))))
			 (ping-res  (case *transport-type* 
				      ((http)(rmt:login-no-auto-client-setup start-res)))))
		    (if (and start-res
			     ping-res)
			(let ((runremote (or area-dat *runremote*))) ;; it might have been generated only a few statements ago
			  (remote-conndat-set! runremote start-res) ;; (hash-table-set! runremote run-id start-res)
			  (debug:print-info 2 *default-log-port* "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 *default-log-port* "client:setup, login unsuccessful, will attempt to start server ... start-res=" start-res ", server-dat=" server-dat) ;; had runid.  Fixes part of Randy;s ticket 1405717332
			  (case *transport-type* 
			    ((http)(http-transport:close-connections)))
			  (remote-conndat-set! runremote #f)  ;; (hash-table-delete! runremote run-id)
			  (thread-sleep! 1)
			  (client:setup-http areapath remaining-tries: (- remaining-tries 1))
			  )))
		  (begin    ;; no server registered
		    ;; (server:kind-run areapath)
		    (server:start-and-wait areapath)
		    (debug:print-info 0 *default-log-port* "client:setup, no server registered, remaining-tries=" remaining-tries)
		    (thread-sleep! 1) ;; (+ 5 (random (- 20 remaining-tries))))  ;; give server a little time to start up, randomize a little to avoid start storms.
		    (client:setup-http areapath remaining-tries: (- remaining-tries 1)))))))))

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

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

(use srfi-1 data-structures posix regex-case (prefix base64 base64:)
     format dot-locking csv-xml z3 udp ;; sql-de-lite
     hostinfo md5 message-digest typed-records directory-utils stack
     matchable regex posix (srfi 18) extras ;; tcp 
     (prefix nanomsg nmsg:)
     (prefix sqlite3 sqlite3:)
     pkts (prefix dbi dbi:)
     )


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

(include "common_records.scm")


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

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


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

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

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


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

  

;; GLOBALS

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

;; ;; safe method for accessing a context given a toppath
;; ;;
;; (define (common:with-cxt toppath proc)
;;   (mutex-lock! *context-mutex*)
;;   (let ((cxt (hash-table-ref/default *contexts* toppath #f)))
;;     (if (not cxt)
;;         (set! cxt (let ((x (make-cxt)))(hash-table-set! *contexts* toppath x) x)))
;;     (let ((cxt-mutex (cxt-mutex cxt)))
;;       (mutex-unlock! *context-mutex*)
;;       (mutex-lock! cxt-mutex)
;;       (let ((res (proc cxt)))
;;         (mutex-unlock! cxt-mutex)
;;         res))))
        
;; A hash table that can be accessed by #{scheme ...} calls in
;; config files. Allows communicating between confgs
;;
(define *user-hash-data* (make-hash-table))

(define *db-keys* #f)

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

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

;; DATABASE
(define *dbstruct-db*         #f) ;; used to cache the dbstruct in db:setup. Goal is to remove this.
;; db stats
(define *db-stats*            (make-hash-table)) ;; hash of vectors < count duration-total >
(define *db-stats-mutex*      (make-mutex))
;; db access
(define *db-last-access*      (current-seconds)) ;; last db access, used in server
(define *db-write-access*     #t)
;; db sync
(define *db-last-sync*        0)                 ;; last time the sync to megatest.db happened
(define *db-sync-in-progress* #f)                ;; if there is a sync in progress do not try to start another
(define *db-multi-sync-mutex* (make-mutex))      ;; protect access to *db-sync-in-progress*, *db-last-sync*
;; task db
(define *task-db*             #f) ;; (vector db path-to-db)
(define *db-access-allowed*   #t) ;; flag to allow access
(define *db-access-mutex*     (make-mutex))
(define *db-transaction-mutex* (make-mutex))
(define *db-cache-path*       #f)
(define *db-with-db-mutex*    (make-mutex))
(define *db-api-call-time*    (make-hash-table)) ;; hash of command => (list of times)
;; no sync db
(define *no-sync-db*          #f)

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

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

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

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

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

;; Miscellaneous
(define *triggers-mutex*     (make-mutex))     ;; block overlapping processing of triggers
(define *numcpus-cache* (make-hash-table))

(use posix-extras pathname-expand files)

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

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

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

(define *common:this-exe-fullpath* (common:get-this-exe-fullpath))
(define *common:this-exe-dir*      (pathname-directory *common:this-exe-fullpath*))
(define *common:this-exe-name*     (pathname-strip-directory *common:this-exe-fullpath*))

(define (common:get-sync-lock-filepath)
  (let* ((tmp-area     (common:get-db-tmp-area))
         (lockfile     (conc tmp-area "/megatest.db.sync-lock")))
    lockfile))

;;======================================================================
;; when called from a wrapper I need sometimes to find the calling
;; wrapper, this is for dashboard to find the correct megatest.
;;
(define (common:find-local-megatest #!optional (progname "megatest"))
  (let ((res (filter file-exists?
		     (map (lambda (updir)
			    (let* ((lm  (car (argv)))
				   (dir (pathname-directory lm))
				   (exe (pathname-strip-directory lm)))
			      (conc (if dir (conc dir "/") "")
				    (case (string->symbol exe)
				      ((dboard)    (conc updir progname))
				      ((mtest)     (conc updir progname))
				      ((dashboard) progname)
				      (else exe)))))
			  '("../../" "../")))))
    (if (null? res)
	(begin
	  (debug:print 0 *default-log-port* "Failed to find this executable! Using what can be found on the path")
	  progname)
	(car res))))

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

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

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

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

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

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

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

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

(define *host-loads*         (make-hash-table))

;; cache environment vars for each run here
(define *env-vars-by-run-id* (make-hash-table))

;; Testconfig and runconfig caches. 
(define *testconfigs*        (make-hash-table)) ;; test-name => testconfig
(define *runconfigs*         (make-hash-table)) ;; target    => runconfig

;; This is a cache of pre-reqs met, don't re-calc in cases where called with same params less than
;; five seconds ago
(define *pre-reqs-met-cache* (make-hash-table))

;; cache of verbosity given string
;;
(define *verbosity-cache*    (make-hash-table))

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

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

(define *last-launch* (current-seconds)) ;; use for throttling the launch rate. Would be better to use the db and last time of a test in LAUNCHED state.

;;======================================================================
;; V E R S I O N
;;======================================================================

(define (common:get-full-version)
  (conc megatest-version "-" megatest-fossil-hash))

(define (common:version-signature)
  (conc megatest-version "-" (substring megatest-fossil-hash 0 4)))

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

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

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

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

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

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

;;======================================================================
;; Move me elsewhere ...
;; RADT => Why do we meed the version check here, this is called only if version misma
;;
(define (common:cleanup-db dbstruct #!key (full #f))
  (apply db:multi-db-sync 
   dbstruct
   'schema
   ;; 'new2old
   'killservers
   'adj-target
   ;; 'old2new
   'new2old
   ;; (if full
       '(dejunk)
       ;; '())
       )
  (if (common:api-changed?)
      (common:set-last-run-version)))

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

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


        ;; if age(<file>.mins.gz) >= 1h:
        ;;     copy <file>.mins.gz <file>.hrs.gz
        ;;     copy <prev file> <file>.mins.gz
        (when (>= (age-mins minsfile) 1)
          (copy minsfile hrsfile)
          (copy+zip filepath  minsfile))
        
        ;; if age(<file>.hrs.gz) >= 1d:
        ;;     copy <file>.hrs.gz <file>.days.gz
        ;;     copy <file>.mins.gz <file>.hrs.gz
        (when (>= (age-days hrsfile) 1)
          (copy hrsfile daysfile)
          (copy minsfile hrsfile))
        
        ;; if age(<file>.days.gz) >= 1w:
        ;;     copy <file>.days.gz <file>.weeks.gz
        ;;     copy <file>.hrs.gz <file>.days.gz
        (when (>= (age-wks daysfile) 1)
          (copy daysfile wksfile)
          (copy hrsfile daysfile))
        #t)
      #f))

;;======================================================================
;; Rotate logs, logic: 
;;                 if > 500k and older than 1 week:
;;                     remove previous compressed log and compress this log
;; WARNING: This proc operates assuming that it is in the directory above the
;;          logs directory you wish to log-rotate.
;;
(define (common:rotate-logs)
  (let* ((all-files (make-hash-table))
	 (stats     (make-hash-table))
	 (inc-stat  (lambda (key)
		      (hash-table-set! stats key (+ (hash-table-ref/default stats key 0) 1))))
	(max-allowed (string->number (or (configf:lookup *configdat* "setup" "max-logfiles") "300")))) ;; name -> age
    (if (not (directory-exists? "logs"))(create-directory "logs"))
    (directory-fold 
     (lambda (file rem)
       (handle-exceptions
	exn
	(begin
	  (debug:print-info 2 *default-log-port* "unable to rotate log " file ", probably handled by another process, this is safe to ignore. exn=" exn)
	  (debug:print 2 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
	  ;; (print-call-chain (current-error-port)) ;; 
	  )
	(let* ((fullname  (conc "logs/" file))
	       (mod-time  (file-modification-time fullname))
	       (file-age  (- (current-seconds) mod-time))
	       (file-old  (> file-age (* 48 60 60)))
	       (file-big  (> (file-size fullname) 200000)))
	  (hash-table-set! all-files file mod-time)
	  (if (or (and (string-match "^.*.log" file)
		       file-old
		       file-big)
		  (and (string-match "^server-.*.log" file)
		       file-old))
	      (let ((gzfile (conc fullname ".gz")))
		(if (common:file-exists? gzfile)
		    (begin
		      (debug:print-info 0 *default-log-port* "removing " gzfile)
		      (delete-file* gzfile)
		      (hash-table-delete!  all-files gzfile) ;; needed?
		      ))
		(debug:print-info 0 *default-log-port* "compressing " file)
		(system (conc "gzip " fullname))
		(inc-stat "gzipped")
		(hash-table-set! all-files (conc file ".gz") file-age)  ;; add the .gz file and remove the base file
		(hash-table-delete! all-files file)
		)
	      (if (and (> file-age (* (string->number (or (configf:lookup *configdat* "setup" "log-expire-days") "30")) 24 3600))
		       (file-exists? fullname)) ;; just in case it was gzipped - will get it next time
		  (handle-exceptions
		   exn
		   #f
		   (if (directory? fullname)
		       (begin
			 (debug:print-info 0 *default-log-port* fullname " in logs directory is a directory! Cannot rotate it, it is best to not put subdirectories in the logs dir.")
			 (inc-stat "directories"))
		       (begin
			 (delete-file* fullname)
			 (inc-stat "deleted")))
		   (hash-table-delete! all-files file)))))))
     '()
     "logs")
    (for-each
     (lambda (category)
       (let ((quant (hash-table-ref/default stats category 0)))
	 (if (> quant 0)
	     (debug:print-info 0 *default-log-port* category " log files: " quant))))
     `("deleted" "gzipped" "directories"))
    (let ((num-logs (hash-table-size all-files)))
      (if (> num-logs max-allowed) ;; because NFS => don't let number of logs exceed 300
	  (let ((files (take (sort (hash-table-keys all-files)
				   (lambda (a b)
				     (< (hash-table-ref all-files a)(hash-table-ref all-files b))))
			     (- num-logs max-allowed))))
	    (for-each
	     (lambda (file)
	       (let* ((fullname (conc "logs/" file)))
		 (if (directory? fullname)
		     (debug:print-info 0 *default-log-port* fullname " in logs directory is a directory! Cannot rotate it, it is best to not put subdirectories in the logs dir.")
		     (handle-exceptions
		      exn
		      (debug:print-error 0 *default-log-port* "failed to remove " fullname ", exn=" exn)
		      (delete-file* fullname)))))
	     files)
	    (debug:print-info 0 *default-log-port* "Deleted " (length files) " files from logs, keeping " max-allowed " files."))))))

;;======================================================================
;; Force a megatest cleanup-db if version is changed and skip-version-check not specified
;; Do NOT check if not on homehost!
;;
(define (common:exit-on-version-changed)
  (if (common:on-homehost?)
      (if (common:api-changed?)
	  (let* ((mtconf (conc (get-environment-variable "MT_RUN_AREA_HOME") "/megatest.config"))
                (dbfile  (conc (get-environment-variable "MT_RUN_AREA_HOME") "/megatest.db"))
                (read-only (not (file-write-access? dbfile)))
                (dbstruct (db:setup #t)))
	    (debug:print 0 *default-log-port*
			 "WARNING: Version mismatch!\n"
			 "   expected: " (common:version-signature) "\n"
			 "   got:      " (common:get-last-run-version))
            (cond
             ((get-environment-variable "MT_SKIP_DB_MIGRATE") #t)
             ((and (common:file-exists? mtconf) (common:file-exists? dbfile) (not read-only)
                   (eq? (current-user-id)(file-owner mtconf))) ;; safe to run -cleanup-db
              (debug:print 0 *default-log-port* "   I see you are the owner of megatest.config, attempting to cleanup and reset to new version")
              (handle-exceptions
               exn
               (begin
                 (debug:print 0 *default-log-port* "Failed to switch versions. exn=" exn)
                 (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
                 (print-call-chain (current-error-port))
                 (exit 1))
               (common:cleanup-db dbstruct)))
             ((not (common:file-exists? mtconf))
              (debug:print 0 *default-log-port* "   megatest.config does not exist in this area.  Cannot proceed with megatest version migration.")
              (exit 1))
             ((not (common:file-exists? dbfile))
              (debug:print 0 *default-log-port* "   megatest.db does not exist in this area.  Cannot proceed with megatest version migration.")
              (exit 1))
             ((not (eq? (current-user-id)(file-owner mtconf)))
              (debug:print 0 *default-log-port* "   You do not own megatest.db in this area.  Cannot proceed with megatest version migration.")
              (exit 1))
             (read-only
              (debug:print 0 *default-log-port* "   You have read-only access to this area.  Cannot proceed with megatest version migration.")
              (exit 1))
             (else
              (debug:print 0 *default-log-port* " to switch versions you can run: \"megatest -cleanup-db\"")
              (exit 1)))))))
;;======================================================================
;;      (begin
;;	(debug:print 0 *default-log-port* "ERROR: cannot migrate version unless on homehost. Exiting.")
;;	(exit 1))))

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

(define (make-sparse-array)
  (let ((a (make-sparse-vector)))
    (sparse-vector-set! a 0 (make-sparse-vector))
    a))

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

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

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

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

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

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

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

;; convert things to an alist or assoc list, #f gets converted to ""
;;
(define (common:to-alist dat)
  (cond
   ((list? dat)   (map common:to-alist dat))
   ((vector? dat)
    (map common:to-alist (vector->list dat)))
   ((pair? dat)
    (cons (common:to-alist (car dat))
	  (common:to-alist (cdr dat))))
   ((hash-table? dat)
    (map common:to-alist (hash-table->alist dat)))
   (else
    (if dat
	dat
	""))))

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

(define (common:low-noise-print waitval . keys)
  (let* ((key      (string-intersperse (map conc keys) "-" ))
	 (lasttime (hash-table-ref/default *common:denoise* key 0))
	 (currtime (current-seconds)))
    (if (> (- currtime lasttime) waitval)
	(begin
	  (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-error 0 *default-log-port* "received bad encoded string \"" instr "\", message: " ((condition-property-accessor 'exn 'message) exn) ", exn=" 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))
  (let ((fmod-time (handle-exceptions
		       ext
		     (current-seconds)
		     (file-modification-time fname))))
    (if (common:file-exists? fname)
	(if (> (- (current-seconds) fmod-time) expire-time)
	    (begin
	      (handle-exceptions exn #f (delete-file* fname))	
	      (common:simple-file-lock fname expire-time: expire-time))
	    #f)
	(let ((key-string (conc (get-host-name) "-" (current-process-id))))
	  (with-output-to-file fname
	    (lambda ()
	      (print key-string)))
	  (thread-sleep! 0.25)
	  (if (common:file-exists? fname)
	      (handle-exceptions exn
                #f 
                (with-input-from-file fname
	  	  (lambda ()
		    (equal? key-string (read-line)))))
	      #f)))))

(define (common:simple-file-lock-and-wait fname #!key (expire-time 300))
  (let ((end-time (+ expire-time (current-seconds))))
    (let loop ((got-lock (common:simple-file-lock fname expire-time: expire-time)))
      (if got-lock
	  #t
	  (if (> end-time (current-seconds))
	      (begin
		(thread-sleep! 3)
		(loop (common:simple-file-lock fname expire-time: expire-time)))
	      #f)))))

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

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

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

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

;;======================================================================
;; BBnote: *common:std-statuses* dashboard filter control and test control status buttons defined here; used in set-fields-panel and dboard:make-controls
;; note these statuses are sorted from better to worse.
;; This sort order is important to dcommon:status-compare3 and db:set-state-status-and-roll-up-items
(define *common:std-statuses*
  '(;; (0 "DELETED")  
    (1 "n/a")
    (2 "PASS")
    (3 "SKIP")
    (4 "WARN")
    (5 "WAIVED")
    (6 "CHECK")
    (7 "STUCK/DEAD")
    (8 "DEAD")
    (9 "FAIL")
    (10 "PREQ_FAIL")
    (11 "PREQ_DISCARDED")
    (12 "ABORT")))

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

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

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

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

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

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

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

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

;;======================================================================
;; ;; given a toplevel with currstate, currstatus apply state and status
;; ;;  => (newstate . newstatus)
;; (define (common:apply-state-status currstate currstatus state status)
;;   (let* ((cstate  (string->symbol (string-downcase currstate)))
;;          (cstatus (string->symbol (string-downcase currstatus)))
;;          (sstate  (string->symbol (string-downcase state)))
;;          (sstatus (string->symbol (string-downcase status)))
;;          (nstate  #f)
;;          (nstatus #f))
;;     (set! nstate
;;           (case cstate
;;             ((completed not_started killed killreq stuck archived) 
;;              (case sstate ;; completed -> sstate
;;                ((completed killed killreq stuck archived) completed)
;;                ((running remotehoststart launched)        running)
;;                (else                                      unknown-error-1)))
;;             ((running remotehoststart launched)
;;              (case sstate
;;                ((completed killed killreq stuck archived) #f) ;; need to look at all items
;;                ((running remotehoststart launched)        running)
;;                (else                                      unknown-error-2)))
;;             (else unknown-error-3)))
;;     (set! nstatus
;;           (case sstatus
;;             ((pass)
;;              (case nstate
;;                ((pass n/a deleted)     pass)
;;                ((warn)                 warn)
;;                ((fail)                 fail)
;;                ((check)               check)
;;                ((waived)             waived)
;;                ((skip)                 skip)
;;                ((stuck/dead)          stuck)
;;                ((abort)               abort)
;;                (else        unknown-error-4)))
;;             ((warn)
;;              (case nstate
;;                ((pass warn n/a skip deleted)   warn)
;;                ((fail)                         fail)
;;                ((check)                       check)
;;                ((waived)                     waived)
;;                ((stuck/dead)                  stuck)
;;                (else                unknown-error-5)))
;;             ((fail)
;;              (case nstate
;;                ((pass warn fail check n/a waived skip deleted stuck/dead stuck)  fail)
;;                ((abort)                                                         abort)
;;                (else                                                  unknown-error-6)))
;;             (else    unknown-error-7)))
;;     (cons 
;;      (if nstate  (symbol->string nstate)  nstate)
;;      (if nstatus (symbol->string nstatus) nstatus))))
               
;;======================================================================
;; 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" "area-name") ;; megatest is a flexible tool, testsuite is too limiting a description.
      (configf:lookup *configdat* "setup" "testsuite" )
      (getenv "MT_TESTSUITE_NAME")
      (pathname-file (or (if (string? *toppath* )
			     (pathname-file *toppath*)
			     #f)
			 (common:get-toppath #f)))
      "please-set-setup-area-name")) ;; (pathname-file (current-directory)))))

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

;;======================================================================
;; redefine for future cleanup (converge on area-name, the more generic
;;
(define common:get-area-name common:get-testsuite-name)

(define (common:get-db-tmp-area . junk)
  (if *db-cache-path*
      *db-cache-path*
      (if *toppath* ;; common:get-create-writeable-dir
	  (handle-exceptions
	      exn
	      (begin
		(debug:print-error 0 *default-log-port* "Couldn't create path to " *db-cache-path* ", exn=" exn)
		(exit 1))
	      (let* ((tsname (common:get-testsuite-name))
		     (dbpath (common:get-create-writeable-dir
			      (list (conc "/tmp/" (current-user-name)
					  "/megatest_localdb/"
					  tsname "/"
					  (string-translate *toppath* "/" "."))
				    (conc "/tmp/" (current-process-id) ;; just in case we have an issue with the dir by own user name
					  "/megatest_localdb/"
					  tsname
					  (string-translate *toppath* "/" "."))
				    ))))
		(set! *db-cache-path* dbpath)
		dbpath))
	  #f)))

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

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

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

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

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

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

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


(define (std-exit-procedure)
  ;;(common:telemetry-log-close)
  (on-exit (lambda () 0))
  ;;(debug:print-info 13 *default-log-port* "std-exit-procedure called; *time-to-exit*="*time-to-exit*)
  (let ((no-hurry  (if *time-to-exit* ;; hurry up
		       #f
		       (begin
			 (set! *time-to-exit* #t)
			 #t))))
    (debug:print-info 4 *default-log-port* "starting exit process, finalizing databases.")
    (if (and no-hurry (debug:debug-mode 18))
	(rmt:print-db-stats))
    (let ((th1 (make-thread (lambda () ;; thread for cleaning up, give it five seconds
                              (if *dbstruct-db* (db:close-all *dbstruct-db*)) ;; one second allocated
			      (if *task-db*    
				  (let ((db (cdr *task-db*)))
				    (if (sqlite3:database? db)
					(begin
					  (sqlite3:interrupt! db)
					  (sqlite3:finalize! db #t)
					  ;; (vector-set! *task-db* 0 #f)
					  (set! *task-db* #f)))))
                              (http-client#close-all-connections!)
                              ;; (if (and *runremote*
                              ;;          (remote-conndat *runremote*))
                              ;;     (begin
                              ;;       (http-client#close-all-connections!))) ;; for http-client
                              (if (not (eq? *default-log-port* (current-error-port)))
                                  (close-output-port *default-log-port*))
			      (set! *default-log-port* (current-error-port))) "Cleanup db exit thread"))
	  (th2 (make-thread (lambda ()
			      (debug:print 4 *default-log-port* "Attempting clean exit. Please be patient and wait a few seconds...")
			      (if no-hurry
                                  (begin
                                    (thread-sleep! 5)) ;; give the clean up few seconds to do it's stuff
                                  (begin
      				  (thread-sleep! 2)))
      			      (debug:print 4 *default-log-port* " ... done")
      			      )
			    "clean exit")))
      (thread-start! th1)
      (thread-start! th2)
      (thread-join! th1)
      )
    )

  0)

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

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


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

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

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

;;======================================================================
;; convert stuff to a number if possible
(define (any->number val)
  (cond 
   ((number? val) val)
   ((string? val) (string->number val))
   ((symbol? val) (any->number (symbol->string val)))
   (else #f)))

(define (any->number-if-possible val)
  (let ((num (any->number val)))
    (if num num val)))

(define (patt-list-match item patts)
  (debug:print-info 8 *default-log-port* "patt-list-match item=" item " patts=" patts)
  (if (and item patts)  ;; here we are filtering for matches with item patterns
      (let ((res #f))   ;; look through all the item-patts if defined, format is patt1,patt2,patt3 ... wildcard is %
	(for-each 
	 (lambda (patt)
	   (let ((modpatt (string-substitute "%" ".*" patt #t)))
	     (debug:print-info 10 *default-log-port* "patt " patt " modpatt " modpatt)
	     (if (string-match (regexp modpatt) item)
		 (set! res #t))))
	 (string-split patts ","))
	res)
      #t))

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

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

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

;;======================================================================
;; return the youngest timestamp . filename
;;
(define (common:get-youngest glob-list)
  (let ((all-files (apply append
			  (map (lambda (patt)
				 (handle-exceptions
				     exn
				     '()
				   (glob patt)))
			       glob-list))))
    (fold (lambda (fname res)
	    (let ((last-mod (car res))
		  (curmod   (handle-exceptions
				exn
				0
			      (file-modification-time fname))))
	      (if (> curmod last-mod)
		  (list curmod fname)
		  res)))
	  '(0 "n/a")
	  all-files)))

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

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

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

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

;;======================================================================
;; 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))))

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

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

(define (common:file-exists? path-string #!key (silent #f))
  ;;;; TODO: catch permission denied exceptions and emit appropriate warnings. Might be able to get rid of with chicken 5?
  (common:false-on-exception (lambda () (file-exists? path-string))
                             message: (if (not silent)
                                          (conc "Unable to access path: " path-string)
                                          #f)
                             ))

(define (common:directory-exists? path-string)
  ;;;; TODO: catch permission denied exceptions and emit appropriate warnings
  (common:false-on-exception (lambda () (directory-exists? path-string))
                             message: (conc "Unable to access path: " path-string)
                             ))

;;======================================================================
;; does the directory exist and do we have write access?
;;
;;    returns the directory or #f
;;
(define (common:directory-writable? path-string)
  (handle-exceptions
      exn
    (begin
      (debug:print 0 *default-log-port* "Failed to identify access to " path-string ", exn=" exn)
      #f)
   (if (and (directory-exists? path-string)
            (file-write-access? path-string))
       path-string
       #f)))

(define (common:get-linktree)
  (or (getenv "MT_LINKTREE")
      (if *configdat*
	  (configf:lookup *configdat* "setup" "linktree")
	  #f)
      (if (or *toppath* (getenv "MT_RUN_AREA_HOME"))
	  (conc (or *toppath* (getenv "MT_RUN_AREA_HOME")) "/lt")
	  #f)
      (let* ((tp (common:get-toppath #f))
	     (lt (conc tp "/lt")))
	(if (not tp)(debug:print 0 *default-log-port* "WARNING: cannot calculate best path for linktree, using " lt))
	lt)))

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

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

(define (common:args-get-target #!key (split #f)(exit-if-bad #f))
  (let* ((keys    (if (hash-table? *configdat*) (common:get-fields *configdat*) '()))
	 (numkeys (length keys))
	 (target  (or (args:get-arg "-reqtarg")
		      (args:get-arg "-target")
		      (getenv "MT_TARGET")))
	 (tlist   (if target (string-split target "/" #t) '()))
	 (valid   (if target
		      (or (null? keys) ;; probably don't know our keys yet
			  (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-error 0 *default-log-port* "Invalid target, spaces or blanks not allowed \"" target "\", target should be: " (string-intersperse keys "/") ", have " tlist " for elements")
	      (if exit-if-bad (exit 1))
	      #f)
	    #f))))

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

;;======================================================================
;; logic for getting homehost. Returns (host . at-home)
;; IF *toppath* is not set, wait up to five seconds trying every two seconds
;; (this is to accomodate the watchdog)
;;
(define (common:get-homehost #!key (trynum 5))
  ;; called often especially at start up. use mutex to eliminate collisions
  (mutex-lock! *homehost-mutex*)
  (cond
   (*home-host*
    (mutex-unlock! *homehost-mutex*)
    *home-host*)
   ((not *toppath*)
    (mutex-unlock! *homehost-mutex*)
    (launch:setup) ;; safely mutexed now
    (if (> trynum 0)
	(begin
	  (thread-sleep! 2)
	  (common:get-homehost trynum: (- trynum 1)))
	#f))
   (else
    (let* ((currhost (get-host-name))
	   (bestadrs (server:get-best-guess-address currhost))
	   ;; first look in config, then look in file .homehost, create it if not found
	   (homehost (or (configf:lookup *configdat* "server" "homehost" )
			 (handle-exceptions
			     exn
			     (if (> trynum 0)
				 (let ((delay-time (* (- 5 trynum) 5)))
				   (mutex-unlock! *homehost-mutex*)
				   (debug:print 0 *default-log-port* "ERROR: ["(common:human-time)"] Failed to read .homehost file, delaying "
						delay-time " seconds and trying again, message: "  ((condition-property-accessor 'exn 'message) exn)
						", exn=" exn)
				   (thread-sleep! delay-time)
				   (common:get-homehost trynum: (- trynum 1)))
				 (begin
				   (mutex-unlock! *homehost-mutex*)
				   (debug:print 0 *default-log-port* "ERROR: ["(common:human-time)
						"] Failed to read .homehost file after trying five times. Giving up and exiting, message: "
						((condition-property-accessor 'exn 'message) exn))
				   (exit 1)))
			   (let ((hhf (conc *toppath* "/.homehost")))
			     (if (common:file-exists? hhf)
				 (with-input-from-file hhf read-line)
				 (if (file-write-access? *toppath*)
				     (begin
				       (with-output-to-file hhf
					 (lambda ()
					   (print bestadrs)))
				       (begin
					 (mutex-unlock! *homehost-mutex*)
					 (car (common:get-homehost))))
				     #f))))))
	   (at-home  (or (equal? homehost currhost)
			 (equal? homehost bestadrs))))
      (set! *home-host* (cons homehost at-home))
      (mutex-unlock! *homehost-mutex*)
      *home-host*))))

;;======================================================================
;; am I on the homehost?
;;
(define (common:on-homehost?)
  (let ((hh (common:get-homehost)))
    (if hh
	(cdr hh)
	#f)))

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

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

;;======================================================================
;; items in lista are matched value and position in listb
;; return the remaining items in listb or #f
;;
(define (common:list-is-sublist lista listb)
  (if (null? lista)
      listb ;; all items in listb are "remaining"
      (if (> (length lista)(length listb)) 
	  #f
	  (let loop ((heda (car lista))
		     (tala (cdr lista))
		     (hedb (car listb))
		     (talb (cdr listb)))
	    (if (equal? heda hedb)
		(if (null? tala) ;; we are done
		    talb
		    (loop (car tala)
			  (cdr tala)
			  (car talb)
			  
			  (cdr talb)))
		#f)))))

;;======================================================================
;; Needed for long lists to be sorted where (apply max ... ) dies
;;
(define (common:max inlst)
  (let loop ((max-val (car inlst))
	     (hed     (car inlst))
	     (tal     (cdr inlst)))
    (if (not (null? tal))
	(loop (max hed max-val)
	      (car tal)
	      (cdr tal))
	(max hed max-val))))

;;======================================================================
;; get min or max, use > for max and < for min, this works around the limits on apply
;;
(define (common:min-max comp lst)
  (if (null? lst)
      #f ;; better than an exception for my needs
      (fold (lambda (a b)
	      (if (comp a b) a b))
	    (car lst)
	    lst)))

;;======================================================================
;; get min or max, use > for max and < for min, this works around the limits on apply
;;
(define (common:sum lst)
  (if (null? lst)
      0
      (fold (lambda (a b)
	      (+ a b))
	    (car lst)
	    lst)))

;;======================================================================
;; path list to hash-table tree
;;   ((a b c)(a b d)(e b c)) => ((a (b (d) (c))) (e (b (c))))
;;
(define (common:list->htree lst)
  (let ((resh (make-hash-table)))
    (for-each
     (lambda (inlst)
       (let loop ((ht  resh)
		  (hed (car inlst))
		  (tal (cdr inlst)))
	 (if (hash-table-ref/default ht hed #f)
	     (if (not (null? tal))
		 (loop (hash-table-ref ht hed)
		       (car tal)
		       (cdr tal)))
	     (begin
	       (hash-table-set! ht hed (make-hash-table))
	       (loop ht hed tal)))))
     lst)
    resh))

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

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

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

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

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

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

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

;;======================================================================
;; lazy-safe get file mod time. on any error (file not existing etc.) return 0
;;
(define (common:lazy-modification-time fpath)
  (handle-exceptions
      exn
    (begin
      (debug:print 0 *default-log-port* "Failed to get modifcation time for " fpath ", treating it as zero. exn=" exn)
      0)
    (if (file-exists? fpath)
	(file-modification-time fpath)
	0)))

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

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

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

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

;;======================================================================
;; returns *effective load* (not normalized)
;;
(define (common:get-intercept onemin fivemin)
  (if (< onemin fivemin) ;; load is decreasing, just use the onemin load
      onemin
      (let* ((load-change (- onemin fivemin))
	     (tchange     (- 300 60)))
	(max (+ onemin (* 60 (/ load-change tchange))) 0))))

;;======================================================================
;; calculate a delay number based on a droop curve
;;    inputs are:
;;      - load-in, load as from uptime, NOT normalized
;;      - numcpus, number of cpus, ideally use the real cpus, not threads
;;
(define (common:get-delay load-in numcpus)
  (let* ((ratio (/ load-in numcpus))
         (new-option (configf:lookup *configdat* "load" "new-load-method"))
	 (paramstr   (or (configf:lookup *configdat* "load" "exp-params")
			 "15 12 1281453987.9543 0.75")) ;; 5 4 10 1"))
	 (paramlst   (map string->number (string-split paramstr))))
    (if new-option
	(begin
          (cond ((and (>= ratio 0) (< ratio .5))
		 0)
		((and (>= ratio 0.5) (<= ratio .9))
		 (* ratio (/ 5 .9)))
		((and (> ratio .9) (<= ratio 1.1))
		 (+ 5 (* (- ratio .9) (/ 55 .2)))) 
		((> ratio 1.1)
		 60)))
	(match paramlst
	  ((r1 r2 s1 s2)
	   (debug:print 3 *default-log-port* "Using params r1=" r1 " r2=" r2 " s1=" s1 " s2=" s2)
	   (min (max (/ (expt r1 (* r2 s2 ratio)) s1) 0) 30))
	  (else
	   (debug:print 0 *default-log-port* "BAD exp-params, should be \"r1 r2 s1 s2\" but got " paramstr)
	   30)))))

(define (common:print-delay-table)
  (let loop ((x 0))
    (print x "," (common:get-delay x 1))
    (if (< x 2)
	(loop (+ x 0.1)))))

(define (get-cpu-load #!key (remote-host #f))
  (car (common:get-cpu-load remote-host)))

;;======================================================================
;;   (let* ((load-res (process:cmd-run->list "uptime"))
;; 	 (load-rx  (regexp "load average:\\s+(\\d+)"))
;; 	 (cpu-load #f))
;;     (for-each (lambda (l)
;; 		(let ((match (string-search load-rx l)))
;; 		  (if match
;; 		      (let ((newval (string->number (cadr match))))
;; 			(if (number? newval)
;; 			    (set! cpu-load newval))))))
;; 	      (car load-res))
;;     cpu-load))

;;======================================================================
;; get values from cached info from dropping file in logs dir
;;  e.g. key is host and dtype is normalized-load
;;
(define (common:get-cached-info key dtype #!key (age 10))
  (if *toppath*
      (let* ((fullpath (conc *toppath* "/.sysdata/" key "-" dtype ".log"))
	     (delfile  (lambda (exn)
			 (debug:print-info 1 *default-log-port* " removing bad file " fullpath ", exn=" exn)
			 (delete-file* fullpath)
			 #f)))
	(if (and (file-exists? fullpath)
		 (file-read-access? fullpath))
	    (handle-exceptions
		exn
	      (begin
		(debug:print 0 *default-log-port* "failed to get cached info from " fullpath ", exn=" exn)
		#f)
	     (debug:print 2 *default-log-port* "reading file " fullpath)
	     (let ((real-age (- (current-seconds)
				(handle-exceptions
				    exn
				  (begin
				    (debug:print 1 *default-log-port* "Failed to read mod time on file "
						 fullpath ", using 0, exn=" exn)
				    0)
				  (file-change-time fullpath)))))
	       (if (< real-age age)
		   (handle-exceptions
		       exn
		     (delfile exn)
		     (let* ((res (with-input-from-file fullpath read)))
		       (if (eof-object? res)
			   (begin
			     (delfile "n/a")
			     #f)
			   res)))
		   (begin
		     (debug:print-info 2 *default-log-port* "file " fullpath
				       " is too old (" real-age" seconds) to trust, skipping reading it")
		     #f))))
	    (begin
	      (debug:print 2 *default-log-port* "not reading file " fullpath)
	      #f)))
      #f))
 
(define (common:write-cached-info key dtype dat)
  (if *toppath*
      (let* ((fulldir  (conc *toppath* "/.sysdata"))
	     (fullpath (conc fulldir "/" key "-" dtype ".log")))
	(if (not (file-exists? fulldir))(create-directory fulldir #t))
	(handle-exceptions
	    exn
	  (begin
	    (debug:print 0 *default-log-port* "failed to write file " fullpath ", exn=" exn)
	    #f)
	  (with-output-to-file fullpath (lambda ()(pp dat)))))
      #f))

(define (common:raw-get-remote-host-load remote-host)
  (handle-exceptions
      exn
    (begin
      (debug:print 0 *default-log-port* "failed to ssh to " remote-host " and get loadavg. exn=" exn)
      #f) ;; more specific handling of errors needed
    (with-input-from-pipe 
     (conc "ssh " remote-host " cat /proc/loadavg")
     (lambda ()(list (read)(read)(read))))))

;;======================================================================
;; get cpu load by reading from /proc/loadavg, return all three values
;;
(define (common:get-cpu-load remote-host)
  (handle-exceptions
      exn
    (begin
      (debug:print 0 *default-log-port* "failed to ssh or read loadavg from host " remote-host ", exn=" exn)
      '(-99 -99 -99))
    (let* ((actual-hostname (or remote-host (get-host-name) "localhost")))
      (or (common:get-cached-info actual-hostname "cpu-load")
	  (let ((result (if remote-host
			    (map (lambda (res)
				   (if (eof-object? res) 9e99 res))
			         (with-input-from-pipe 
				  (conc "ssh " remote-host " cat /proc/loadavg")
				  (lambda ()(list (read)(read)(read)))))
			    (with-input-from-file "/proc/loadavg" 
			      (lambda ()(list (read)(read)(read)))))))
	    (match
		result
	      ((l1 l2 l3)
	       (if (and (number? l1)
		      (number? l2)
		      (number? l3))
		   (begin
		     (common:write-cached-info actual-hostname "cpu-load" result)
		     result)
		   '(-1 -1 -1))) ;; -1 is bad result
	      (else '(-2 -2 -2))))))))

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

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

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

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

;;======================================================================
;; see defstruct host at top of file.
;;    host: reachable last-update last-used last-cpuload
;;
(define (common:update-host-loads-table hosts-raw)
  (let* ((hosts (filter (lambda (x)
                          (string-match (regexp "^\\S+$") x))
                        hosts-raw)))
    (for-each
     (lambda (hostname)
       (let* ((rec       (let ((h (hash-table-ref/default *host-loads* hostname #f)))
                          (if h
                              h
                              (let ((h (make-host)))
                                (hash-table-set! *host-loads* hostname h)
                                h))))
              (host-info         (common:get-host-info hostname))
              (is-reachable      (car host-info))
              (last-reached-time (cadr host-info))
              (load              (caddr host-info)))
         (host-reachable-set!    rec is-reachable)
         (host-last-update-set!  rec last-reached-time)
         (host-last-cpuload-set! rec load)))
     hosts)))

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

(define (common:wait-for-homehost-load maxnormload msg)
  (let* ((hh-dat (if (common:on-homehost?) ;; if we are on the homehost then pass in #f so the calls are local.
                     #f
                     (common:get-homehost)))
         (hh     (if hh-dat (car hh-dat) #f)))
    (common:wait-for-normalized-load maxnormload msg hh)))

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

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

;;======================================================================
;; DO NOT CALL THIS DIRECTLY. It is called from common:wait-for-normalized-load
;;   count     - count down to zero, at some point we'd give up if the load never drops
;;   num-tries - count down to zero number tries to get numcpus
;;
(define (common:wait-for-cpuload maxnormload numcpus-in
				 #!key (count 1000)
				 (msg #f)(remote-host #f)(num-tries 5))
  (let* ((loadavg     (common:get-cpu-load remote-host))
	  ;; not possible to have zero.  If we get 1, it's possible that we got the previous default, and we should check again
	 (numcpus     (if (<= 1 numcpus-in)
			  (common:get-num-cpus remote-host)
			  numcpus-in))
	 (first       (car loadavg))
	 (next        (cadr loadavg))
	 (adjmaxload  (* maxnormload (max 1 numcpus))) ;; possible bug
						   ;; where numcpus
						   ;; (or could be
						   ;; maxload) is
						   ;; zero, crude
						   ;; fallback is to
						   ;; at least use 1
	 ;; effective load accounts for load jumps, this should elminate all the first-next-avg, adjwait, load-jump-limit
	 ;; etc.
	 (effective-load    (common:get-intercept first next))
	 (recommended-delay (common:get-delay effective-load numcpus))
	 (effective-host    (or remote-host "localhost"))
	 (normalized-effective-load (/ effective-load numcpus))
	 (will-wait                 (> normalized-effective-load maxnormload)))
    (if (> recommended-delay 1) 
	(let* ((actual-delay (min recommended-delay 30)))
	  (if (common:low-noise-print 30 (conc (round actual-delay) "-safe-load"))
	      (debug:print-info 0 *default-log-port* "Load control, delaying "
				actual-delay " seconds to maintain safe load. current normalized effective load is "
				normalized-effective-load". maxnormload = " maxnormload " numcpus = " numcpus " loadavg = " loadavg " effective-load = " effective-load))
	  (thread-sleep! actual-delay)))
    
    (cond
     ;; bad data, try again to get the data
     ((not will-wait)
      (if (common:low-noise-print 3600 (conc (round normalized-effective-load) "-load-acceptable-" effective-host))
      	  (debug:print 0 *default-log-port* "Effective load on " effective-host " is acceptable at " effective-load " continuing.")))

     ((and (< first 0) ;; this indicates the loadavg data is bad - machine may not be reachable
	   (> num-tries 0))
      (debug:print 0 *default-log-port* "WARNING: received bad data from get-cpu-load "
		   first ", we'll sleep 10s and try " num-tries " more times.")
      (thread-sleep! 10)
      (common:wait-for-cpuload maxnormload numcpus-in
			       count: count remote-host: remote-host num-tries: (- num-tries 1)))

     ;; need to wait for load to drop
     ((and will-wait ;; (> first adjmaxload)
	   (> count 0))
      (debug:print-info 0 *default-log-port*
			"Delaying 15" ;; adjwait
			" seconds due to normalized effective load " normalized-effective-load ;; first
			" exceeding max of " adjmaxload
			" on server " (or remote-host (get-host-name))
			" (normalized load-limit: " maxnormload ") " (if msg msg ""))
      (thread-sleep! 15) ;; adjwait)
      (common:wait-for-cpuload maxnormload numcpus count: (- count 1) msg: msg remote-host: remote-host)
      ;; put the message here to indicate came out of waiting
      (debug:print-info 1 *default-log-port*
			"On host: " effective-host
			", effective load: " effective-load
			", numcpus: " numcpus
			", normalized effective load: " normalized-effective-load
			))
     ;; overloaded and count expired (i.e. went to zero)
     (else
      (if (> num-tries 0) ;; should be "num-tries-left". 
	  (if (common:low-noise-print 30 (conc (round effective-load) "-load-acceptable-" effective-host))
	      (debug:print 0 *default-log-port* "Load on " effective-host " is acceptable at effective normalized load of "
			   effective-normalized-load "  continuing."))
	  (debug:print 0 *default-log-port* "Load on " effective-host ", "
		       first" could not be retrieved. Giving up and continuing."))))))

;;======================================================================
;; DO NOT CALL THIS DIRECTLY. It is called from common:wait-for-normalized-load
;;
;; (define (common:wait-for-cpuload maxload-in numcpus-in waitdelay #!key (count 1000) (msg #f)(remote-host #f)(force-maxload #f)(num-tries 5))
;;   (let* ((loadavg (common:get-cpu-load remote-host))
;; 	 (numcpus (if (<= 1 numcpus-in) ;; not possible to have zero.  If we get 1, it's possible that we got the previous default, and we should check again
;; 		      (common:get-num-cpus remote-host)
;; 		      numcpus-in))
;; 	 (maxload (if force-maxload
;; 		      maxload-in
;; 		      (if (number? maxload-in)
;; 			  (max maxload-in 0.5)
;; 			  0.5))) ;; so maxload must be greater than 0.5 for now BUG - FIXME?
;; 	 (first   (car loadavg))
;; 	 (next    (cadr loadavg))
;; 	 (adjmaxload (* maxload (max 1 numcpus))) ;; possible bug where
;; 					       ;; numcpus (or could be
;; 					       ;; maxload) is zero,
;; 					       ;; crude fallback is to
;; 					       ;; at least use 1
;; 	 (loadjmp (- first (if (> next (* numcpus 0.7)) ;; could do something with average of first and next?
;; 			       0
;; 			       next))) ;; we will force a conservative calculation any time next is large.
;; 	 (first-next-avg    (/ (+ first next) 2))
;; 	 ;; add some randomness to the time to break any alignment
;; 	 ;; where netbatch dumps many jobs to machines simultaneously
;;          (adjwait           (min (+ 300 (random 10)) (abs (* (+ (random 10)
;; 								(/ (- 1000 count) 10)
;; 								waitdelay)
;; 							     (- first adjmaxload) ))))
;; 	 (load-jump-limit   (configf:lookup-number *configdat* "setup" "load-jump-limit"))
;; 	 ;; effective load accounts for load jumps, this should elminate all the first-next-avg, adjwait, load-jump-limit
;; 	 ;; etc.
;; 	 (effective-load    (common:get-intercept first next))
;; 	 (effective-host    (or remote-host "localhost"))
;; 	 (normalized-effective-load (/ effective-load numcpus))
;; 	 (will-wait                 (> normalized-effective-load maxload)))
;; 	 
;;     ;; let's let the user know once in a long while that load checking
;;     ;; is happening but not constantly report it
;;     #;(if (common:low-noise-print 30 (conc "cpuload" (or remote-host "localhost"))) ;; (> (random 100) 75) ;; about 25% of the time
;; 	(debug:print-info 1 *default-log-port* "Checking cpuload on " (or remote-host "localhost") ", maxload: " maxload
;;     ", load: " first ", adjmaxload: " adjmaxload ", loadjmp: " loadjmp))
;; 
;;     (debug:print-info 1 *default-log-port*
;; 		      "On host: " effective-host
;; 		      ", effective load: " effective-load
;; 		      ", numcpus: " numcpus
;; 		      ", normalized effective load: " normalized-effective-load
;; 		      )
;;     
;;     (cond
;;      ;; bad data, try again to get the data
;;      ((and (< first 0) ;; this indicates the loadavg data is bad - machine may not be reachable
;; 	   (> num-tries 0))
;;       (debug:print 0 *default-log-port* "WARNING: received bad data from get-cpu-load " first ", we'll sleep 10s and try " num-tries " more times.")
;;       (thread-sleep! 10)
;;       (common:wait-for-cpuload maxload-in numcpus-in waitdelay
;; 			       count: count remote-host: remote-host force-maxload: force-maxload num-tries: (- num-tries 1)))
;;      ;; need to wait for load to drop
;;      ((and will-wait ;; (> first adjmaxload)
;; 	   (> count 0))
;;       (debug:print-info 0 *default-log-port*
;; 			"Delaying " 15 ;; adjwait
;; 			" seconds due to normalized effective load " normalized-effective-load ;; first
;; 			" exceeding max of " adjmaxload
;; 			" on server " (or remote-host (get-host-name))
;; 			" (normalized load-limit: " maxload ") " (if msg msg ""))
;;       (thread-sleep! 15) ;; adjwait)
;;       (common:wait-for-cpuload maxload numcpus waitdelay count: (- count 1) msg: msg remote-host: remote-host))
;;      ((and (> loadjmp (cond
;; 		       (load-jump-limit load-jump-limit)
;; 		       ((> numcpus 8)(/ numcpus 2))
;; 		       ((> numcpus 4)(/ numcpus 1.2))
;; 		       (else 0.5)))
;; 	   (> count 0))
;;       (debug:print-info 0 *default-log-port* "waiting " adjwait " seconds due to possible load jump " loadjmp ". "
;; 			(if msg msg ""))
;;       (thread-sleep! adjwait)
;;       (common:wait-for-cpuload maxload numcpus waitdelay count: (- count 1) msg: msg remote-host: remote-host))
;;      (else
;;       (if (> num-tries 0)
;; 	  (if (common:low-noise-print 30 (conc (round first) "-load-acceptable-" (or remote-host "localhost")))
;; 	      (debug:print 0 *default-log-port* "Load on " (or remote-host "localhost") " is acceptable at " first " continuing."))
;; 	  (debug:print 0 *default-log-port* "Load on " (or remote-host "localhost") ", "first" could not be retrieved. Giving up and continuing."))))))
;; 
(define (get-uname . params)
  (let* ((uname-res (process:cmd-run->list (conc "uname " (if (null? params) "-a" (car params)))))
	 (uname #f))
    (if (null? (car uname-res))
	"unknown"
	(caar uname-res))))

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

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

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

;;======================================================================
;; given path get free space, allows override in [setup]
;; with free-space-script /path/to/some/script.sh
;;
(define (get-df path)
  (if (configf:lookup *configdat* "setup" "free-space-script")
      (with-input-from-pipe 
       (conc (configf:lookup *configdat* "setup" "free-space-script") " " path)
       (lambda ()
	 (let ((res (read-line)))
	   (if (string? res)
	       (string->number res)))))
      (get-unix-df path)))

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

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

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

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

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

;;======================================================================
;; check available space in dbdir, exit if insufficient
;;
(define (common:check-db-dir-and-exit-if-insufficient)
  (let* ((spacedat (car (common:check-db-dir-space))) ;; look only at worst for now
	 (is-ok    (car spacedat))
	 (dbspace  (cadr spacedat))
	 (required (caddr spacedat))
	 (dbdir    (cadddr spacedat)))
    (if (not is-ok)
	(begin
	  (debug:print-error 0 *default-log-port* "Insufficient space in " dbdir ", require " required ", have " dbspace  ", exiting now.")
	  (exit 1)))))

;;======================================================================
;; paths is list of lists ((name path) ... )
;;
(define (common:get-disk-with-most-free-space disks minsize)
  (let* ((best     #f)
	(bestsize 0)
        (default-min-inodes-string "1000000")
        (default-min-inodes (string->number default-min-inodes-string))
        (min-inodes (or (string->number (if (configf:lookup *configdat* "setup" "min_inodes") (configf:lookup *configdat* "setup" "min_inodes") default-min-inodes-string)) default-min-inodes)))

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

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

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

;;======================================================================
;; E N V I R O N M E N T   V A R S
;;======================================================================

(define (bb-check-path #!key (msg "check-path: "))
  (let ((path (or (get-environment-variable "PATH") "none")))
    (debug:print-info 0 *default-log-port* (conc msg" : $PATH="path))
    (if (string-match "^.*/isoenv-core/.*" path)
        (debug:print-error 0 *default-log-port* (conc msg" : !!ISOENV PRESENT!!")) ;; remove for prod
        (debug:print-info 1 *default-log-port* (conc msg" : **no isoenv present**")))))
	      
(define (save-environment-as-files fname #!key (ignorevars (list "USER" "HOME" "DISPLAY" "LS_COLORS" "XKEYSYMDB" "EDITOR" "MAKEFLAGS" "MAKEF" "MAKEOVERRIDES")))
  ;;(bb-check-path msg: "save-environment-as-files entry")
  (let ((envvars (get-environment-variables))
        (whitesp (regexp "[^a-zA-Z0-9_\\-:,\\.\\/%$]"))
	(mungeval (lambda (val)
		    (cond
		     ((eq? val #t) "") ;; convert #t to empty string
		     ((eq? val #f) #f) ;; convert #f to itself (still thinking about this one
		     (else val)))))
    (with-output-to-file (conc fname ".csh")
       (lambda ()
          (for-each (lambda (keyval)
		      (let* ((key   (car keyval))
			     (val   (cdr keyval))
			     (delim (if (string-search whitesp val) 
					"\""
					"")))
			(print (if (or (member key ignorevars)
				       (string-search whitesp key))
				   "# setenv "
				   "setenv ")
			       key " " delim (mungeval val) delim)))
		    envvars)))
     (with-output-to-file (conc fname ".sh")
       (lambda ()
          (for-each (lambda (keyval)
		      (let* ((key (car keyval))
			     (val (cdr keyval))
			     (delim (if (string-search whitesp val) 
					"\""
					"")))
			(print (if (or (member key ignorevars)
				       (string-search whitesp key)
				       (string-search ":" key)) ;; internal only values to be skipped.
				   "# export "
				   "export ")
			       key "=" delim (mungeval val) delim)))
                    envvars)))))

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

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

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

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

(define (common:without-vars proc . var-patts)
  (let ((vars (make-hash-table)))
    (for-each
     (lambda (vardat) ;; each env var
       (for-each
	(lambda (var-patt)
	  (if (string-match var-patt (car vardat))
	      (let ((var (car vardat))
		    (val (cdr vardat)))
		(hash-table-set! vars var val)
		(unsetenv var))))
	var-patts))
     (get-environment-variables))
    (cond
     ((string? proc)(system proc))
     (proc          (proc)))
    (hash-table-for-each
     vars
     (lambda (var val)
       (setenv var val)))
    vars))

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

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

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

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

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

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

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

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

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

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

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

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

;;======================================================================
;; expand a complex cron string to a list of cron strings
;;
;;  x/y   => x, x+y, x+2y, x+3y while x+Ny<max_for_field
;;  a,b,c => a, b ,c
;;
;;   NOTE: with flatten a lot of the crud below can be factored down.
;;
(define (common:cron-expand cron-str)
  (if (list? cron-str)
      (flatten
       (fold (lambda (x res)
	       (if (list? x)
		   (let ((newres (map common:cron-expand x)))
		     (append x newres))
		   (cons x res)))
	     '()
	     cron-str)) ;; (map common:cron-expand cron-str))
      (let ((cron-items (string-split cron-str))
	    (slash-rx   (regexp "(\\d+)/(\\d+)"))
	    (comma-rx   (regexp ".*,.*"))
	    (max-vals   '((min        . 60)
			  (hour       . 24)
			  (dayofmonth . 28) ;;; BUG!!!! This will be a bug for some combinations
			  (month      . 12)
			  (dayofweek  . 7))))
	(if (< (length cron-items) 5) ;; bad spec
	    cron-str ;; `(,cron-str)              ;; just return the string, something downstream will fix it
	    (let loop ((hed  (car cron-items))
		       (tal  (cdr cron-items))
		       (type 'min)
		       (type-tal '(hour dayofmonth month dayofweek))
		       (res  '()))
	      (regex-case
		  hed
		(slash-rx ( _ base incr ) (let* ((basen          (string->number base))
						 (incrn          (string->number incr))
						 (expanded-vals  (common:expand-cron-slash basen incrn (alist-ref type max-vals)))
						 (new-list-crons (fold (lambda (x myres)
									 (cons (conc (if (null? res)
											 ""
											 (conc (string-intersperse res " ") " "))
										     x " " (string-intersperse tal " "))
									       myres))
								       '() expanded-vals)))
					    ;; (print "new-list-crons: " new-list-crons)
					    ;; (fold (lambda (x res)
					    ;; 	    (if (list? x)
					    ;; 		(let ((newres (map common:cron-expand x)))
					    ;; 		  (append x newres))
					    ;; 		(cons x res)))
					    ;; 	  '()
					    (flatten (map common:cron-expand new-list-crons))))
		;;					    (map common:cron-expand (map common:cron-expand new-list-crons))))
		(else (if (null? tal)
			  cron-str
			  (loop (car tal)(cdr tal)(car type-tal)(cdr type-tal)(append res (list hed)))))))))))

;;======================================================================
;; given a cron string and the last time event was processed return #t to run or #f to not run
;;
;;  min    hour   dayofmonth month  dayofweek
;; 0-59    0-23   1-31       1-12   0-6          ### NOTE: dayofweek does not include 7
;;
;;  #t => yes, run the job
;;  #f => no, do not run the job
;;
(define (common:cron-event cron-str now-seconds-in last-done) ;; ref-seconds = #f is NOW.
  (let* ((cron-items     (map string->number (string-split cron-str)))
	 (now-seconds    (or now-seconds-in (current-seconds)))
	 (now-time       (seconds->local-time now-seconds))
	 (last-done-time (seconds->local-time last-done))
	 (all-times      (make-hash-table)))
    ;; (print "cron-items: " cron-items "(length cron-items): " (length cron-items))
    (if (not (eq? (length cron-items) 5)) ;; don't even try to figure out junk strings
	#f
	(match-let (((     cmin chour cdayofmonth cmonth    cdayofweek)
		     cron-items)
		    ;; 0     1    2        3         4    5      6
		    ((nsec nmin nhour ndayofmonth nmonth nyr ndayofweek n7 n8 n9)
		     (vector->list now-time))
		    ((lsec lmin lhour ldayofmonth lmonth lyr ldayofweek l7 l8 l9)
		     (vector->list last-done-time)))
	  ;; create all possible time slots
	  ;; remove invalid slots due to (for example) day of week
	  ;; get the start and end entries for the ref-seconds (current) time
	  ;; if last-done > ref-seconds => this is an ERROR!
	  ;; does the last-done time fall in the legit region?
	  ;;    yes => #f  do not run again this command
	  ;;    no  => #t  ok to run the command
	  (for-each ;; month
	   (lambda (month)
	     (for-each ;; dayofmonth
	      (lambda (dom)
		(for-each
		 (lambda (hr) ;; hour
		   (for-each
		    (lambda (minute) ;; minute
		      (let ((copy-now (apply vector (vector->list now-time))))
			(vector-set! copy-now 0 0) ;; force seconds to zero
			(vector-set! copy-now 1 minute)
			(vector-set! copy-now 2 hr)
			(vector-set! copy-now 3 dom)  ;; dom is already corrected for zero referenced
			(vector-set! copy-now 4 month)
			(let* ((copy-now-secs (local-time->seconds copy-now))
			       (new-copy      (seconds->local-time copy-now-secs))) ;; remake the time vector
			  (if (or (not cdayofweek)
				  (equal? (vector-ref new-copy 6)
					  cdayofweek)) ;; if the day is specified and a match OR if the day is NOT specified
			      (if (or (not cdayofmonth)
				      (equal? (vector-ref new-copy 3)
					      (+ 1 cdayofmonth))) ;; if the month is specified and a match OR if the month is NOT specified
				  (hash-table-set! all-times copy-now-secs new-copy))))))
		    (if cmin
			`(,cmin)  ;; if given cmin, have to use it
			(list (- nmin 1) nmin (+ nmin 1))))) ;; minute
		 (if chour
		     `(,chour)
		     (list (- nhour 1) nhour (+ nhour 1))))) ;; hour
	      (if cdayofmonth
		  `(,cdayofmonth)
		  (list (- ndayofmonth 1) ndayofmonth (+ ndayofmonth 1)))))
	   (if cmonth
	       `(,cmonth)
	       (list (- nmonth 1) nmonth (+ nmonth 1))))
	  (let ((before #f)
		(is-in  #f))
	    (for-each
	     (lambda (moment)
	       (if (and before
			(<= before now-seconds)
			(>= moment now-seconds))
		   (begin
		     ;; (print)
		     ;; (print "Before: " (time->string (seconds->local-time before)))
		     ;; (print "Now:    " (time->string (seconds->local-time now-seconds)))
		     ;; (print "After:  " (time->string (seconds->local-time moment)))
		     ;; (print "Last:   " (time->string (seconds->local-time last-done)))
		     (if (<  last-done before)
			 (set! is-in before))
		     ))
	       (set! before moment))
	     (sort (hash-table-keys all-times) <))
	    is-in)))))

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

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

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

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

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

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

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

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

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

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

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

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

;;======================================================================
;; N A N O M S G   C L I E N T
;;======================================================================
;; 
;; 
;; 
;; (define (common:send-dboard-main-changed)
;;   (let* ((dashboard-ips (mddb:get-dashboards)))
;;     (for-each
;;      (lambda (ipadr)
;;        (let* ((soc (common:open-nm-req (conc "tcp://" ipadr)))
;; 	      (msg (conc "main " *toppath*))
;; 	      (res (common:nm-send-receive-timeout soc msg)))
;; 	 (if (not res) ;; couldn't reach that dashboard - remove it from db
;; 	     (print "ERROR: couldn't reach dashboard " ipadr))
;; 	 res))
;;      dashboard-ips)))
;;     
;;     
;; ;;======================================================================
;; ;; D A S H B O A R D   D B 
;; ;;======================================================================
;; 
;; (define (mddb:open-db)
;;   (let* ((db (open-database (conc (get-environment-variable "HOME") "/.dashboard.db"))))
;;     (set-busy-handler! db (busy-timeout 10000))
;;     (for-each
;;      (lambda (qry)
;;        (exec (sql db qry)))
;;      (list 
;;       "CREATE TABLE IF NOT EXISTS vars       (id INTEGER PRIMARY KEY,key TEXT, val TEXT, CONSTRAINT varsconstraint UNIQUE (key));"
;;       "CREATE TABLE IF NOT EXISTS dashboards (
;;           id         INTEGER PRIMARY KEY,
;;           pid        INTEGER,
;;           username   TEXT,
;;           hostname   TEXT,
;;           ipaddr     TEXT,
;;           portnum    INTEGER,
;;           start_time TIMESTAMP DEFAULT (strftime('%s','now')),
;;              CONSTRAINT hostport UNIQUE (hostname,portnum)
;;         );"
;;       ))
;;     db))
;; 
;; ;; register a dashboard 
;; ;;
;; (define (mddb:register-dashboard port)
;;   (let* ((pid      (current-process-id))
;; 	 (hostname (get-host-name))
;; 	 (ipaddr   (server:get-best-guess-address hostname))
;; 	 (username (current-user-name)) ;; (car userinfo)))
;; 	 (db      (mddb:open-db)))
;;     (print "Register monitor, pid: " pid ", hostname: " hostname ", port: " port ", username: " username)
;;     (exec (sql db "INSERT OR REPLACE INTO dashboards (pid,username,hostname,ipaddr,portnum) VALUES (?,?,?,?,?);")
;; 	   pid username hostname ipaddr port)
;;     (close-database db)))
;; 
;; ;; unregister a monitor
;; ;;
;; (define (mddb:unregister-dashboard host port)
;;   (let* ((db      (mddb:open-db)))
;;     (print "Register unregister monitor, host:port=" host ":" port)
;;     (exec (sql db "DELETE FROM dashboards WHERE hostname=? AND portnum=?;") host port)
;;     (close-database db)))
;; 
;; ;; get registered dashboards
;; ;;
;; (define (mddb:get-dashboards)
;;   (let ((db (mddb:open-db)))
;;     (query fetch-column
;; 	   (sql db "SELECT ipaddr || ':' || portnum FROM dashboards;"))))
    
;;======================================================================
;;  T E S T   L A U N C H I N G   P E R   I T E M   W I T H   H O S T   T Y P E S
;;======================================================================
;; 
;; [hosts]
;; arm cubie01 cubie02
;; x86_64 zeus xena myth01
;; allhosts #{g hosts arm} #{g hosts x86_64}
;; 
;; [host-types]
;; general #MTLOWESTLOAD #{g hosts allhosts}
;; arm     #MTLOWESTLOAD #{g hosts arm}
;; nbgeneral nbjob run JOBCOMMAND -log $MT_LINKTREE/$MT_TARGET/$MT_RUNNAME.$MT_TESTNAME-$MT_ITEM_PATH.lgo
;;
;; [host-rules]
;; # maxnload   => max normalized load
;; # maxnjobs   => max jobs per cpu
;; # maxjobrate => max jobs per second
;; general maxnload=1.1; maxnjobs=1.2; maxjobrate=0.1 
;; 
;; [launchers]
;; envsetup general
;; xor/%/n 4C16G
;; % nbgeneral
;; 
;; [jobtools]
;; # if defined and not "no" flexi-launcher will bypass "launcher" unless no match.
;; flexi-launcher yes  
;; launcher nbfake
;;
(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 2 *default-log-port* "Have flexi-launcher match for " testname "/" itempath " = " host-type)
			(let ((launcher (configf:lookup configdat "host-types" host-type)))
			  (if launcher
			      (let* ((launcher-parts (string-split launcher))
				     (launcher-exe   (car launcher-parts)))
				(if (equal? launcher-exe "#MTLOWESTLOAD") ;; this is our special case, we will find the lowest load and craft a nbfake commandline
				    (let host-loop ((targ-host (common:get-least-loaded-host (cdr launcher-parts) host-type configdat))
						    (count     100))
				      (if targ-host
					  (conc "remrun " targ-host)
					  (if (> count 0)
					      (begin
						(debug:print 0 *default-log-port* "INFO: Waiting for a host for host-type " host-type)
						(thread-sleep! (- 101 count))
						(host-loop (common:get-least-loaded-host (cdr launcher-parts) host-type configdat)
							   (- count 1)))
					      (begin
						(debug:print 0 *default-log-port* "FATAL: Failed to find a host from #MTLOWESTLOAD for host-type " host-type)
						(exit)))))
				    launcher))
			      (begin
				(debug:print-info 0 *default-log-port* "WARNING: no launcher found for host-type " host-type)
				(if (null? tal)
				    fallback-launcher
				    (loop (car tal)(cdr tal)))))))
		      ;; no match, try again
		      (if (null? tal)
			  fallback-launcher
			  (loop (car tal)(cdr tal))))))))
	fallback-launcher)))

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

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

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

;;======================================================================
;; H I E R A R C H I C A L   H A S H   T A B L E S
;;======================================================================
;;
;; Every element including top element is a vector:
;;   <vector subhash value>

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

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

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

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

;;======================================================================
;; Manage pkts, used in servers, tests and likely other contexts so put
;; in common
;;======================================================================

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

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

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

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

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

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

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

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

(define (common:join-backgrounded-threads)
  ;; may need to trap and ignore exceptions -- dunno how atomic threads are...
  (for-each
   (lambda (thread-name)
     (let* ((thread (hash-table-ref/default *common:thread-punchlist* thread-name #f)))
       (if thread
           (handle-exceptions
               exn
	     (begin
	       (debug:print 0 *default-log-port* "joining threads failed. exn=" exn)
               #t) ;; just ignore it, it might have died in the meantime so joining it will throw an exception
             (thread-join! thread))
           )))
   (hash-table-keys *common:thread-punchlist*)))

;;======================================================================
;; (define *common:telemetry-log-state* 'startup)
;; (define *common:telemetry-log-socket* #f)
;; 
;; (define (common:telemetry-log-open)
;;   (if (eq? *common:telemetry-log-state* 'startup)
;;       (let* ((serverhost (configf:lookup *configdat* "telemetry" "host"))
;;              (serverport (configf:lookup-number *configdat* "telemetry" "port"))
;;              (user (or (get-environment-variable "USER") "unknown"))
;;              (host (or (get-environment-variable "HOST") "unknown")))
;;         (set! *common:telemetry-log-state*
;;               (handle-exceptions
;;                exn
;;                (begin
;;                  (debug:print-info 0 *default-log-port* "common-telemetry-log open udp port failure")
;;                  'broken)
;;                (if (and serverhost serverport user host)
;;                    (let* ((s (udp-open-socket)))
;;                      ;;(udp-bind! s #f 0)
;;                      (udp-connect! s serverhost serverport)
;;                      (set! *common:telemetry-log-socket* s)
;;                      'open)
;;                    'not-needed))))))
;;   
;; (define (common:telemetry-log event #!key (payload '()))
;;   (if (eq? *common:telemetry-log-state* 'startup)
;;       (common:telemetry-log-open))
;; 
;;   (if (eq? 'open *common:telemetry-log-state*)
;;       (handle-exceptions
;;        exn
;;        (begin
;;          (debug:print-info 0 *default-log-port* "common-telemetry-log comms failure ; disabled (no server?)")
;;          ;;(define *common:telemetry-log-state* 'broken-or-no-server-preclose)
;;          ;;(common:telemetry-log-close)
;;          (define *common:telemetry-log-state* 'broken-or-no-server)
;;          (set! *common:telemetry-log-socket* #f)
;;          )
;;        (if (and *common:telemetry-log-socket* event) ;; TODO - filter on event against telemetry.want-events
;;            (let* ((user (or (get-environment-variable "USER") "unknown"))
;;                   (host (or (get-environment-variable "HOST") "unknown"))
;;                   (start (conc "[megatest "event"]"))
;;                   (toppath (or *toppath* "/dev/null"))
;;                   (payload-serialized
;;                    (base64:base64-encode
;;                     (z3:encode-buffer
;;                      (with-output-to-string (lambda () (pp payload))))))
;;                   (msg     (conc user":"host":"start":"(current-process-id)":"(car (argv))":"
;;                                  toppath":"payload-serialized)))
;;              (udp-send *common:telemetry-log-socket* msg))))))
;;   
;; (define (common:telemetry-log-close)
;;   (when (or (member *common:telemetry-log-state* '(broken-or-no-server-preclose open)) *common:telemetry-log-socket*)
;;     (handle-exceptions
;;      exn
;;      (begin
;;        (define *common:telemetry-log-state* 'closed-fail)
;;        (debug:print-info 0 *default-log-port* "common-telemetry-log closure failure")
;;        )
;;      (begin
;;        (define *common:telemetry-log-state* 'closed)
;;        (udp-close-socket *common:telemetry-log-socket*)
;;        (set! *common:telemetry-log-socket* #f)))))