#! /bin/sh
# run the hdf5/bin/snapshot
# Usage:
#	runtest		run the test for the local host
#	runtest <hostname>	run the test for <hostname>
#	runtest -all		run the test for all predefined hosts
#
# Assumptions in knowing where to find the right scripts to execute.
# 1. assume we are at the top level of the hdf5 source.  So, bin/* are
#    where the script files are.
# 2. after the cvs update is completed, we can go to the snapshot area
#    hdf5 source tree and use the bin/* there.
# 3. Cannot use the snapshot area scripts from the beginning because
#    for one, the current directory is renamed as previous right after
#    a snapshot release; and for another, some scripts may be changed
#    by the cvs update while it is being used.

# local setup
DEBUGMODE=""
test -n "$DEBUGMODE" && echo "******** DEBUGMODE is $DEBUGMODE ************"
WHEREAMI='pwd'

# the name of this program
PROGNAME="bin/runtest $DEBUGMODE"

# Setup
HOSTNAME=`hostname | cut -f1 -d.`	# no domain part
H5DIR=$HOME/HDF5/v_1_5/hdf5
TODAY=`date +%y%m%d`
WEEKDAY=`date +%a`
H5VER=			# default to current CVS version
H5VERSTR=		# default to current CVS version

# Default to do checkout (only once) and test, no release.
# If srcdir is not used, don't launched multiple tests
SNAPSHOT="${DEBUGMODE:+echo }bin/snapshot"
SRCDIR="srcdir"
# Default standard Snaptest commands
SNAPCMD="$SRCDIR test clean"
# Default Standard snaptest command options
STANDARD_OPT=""
ENABLE_PARALLEL="--enable-parallel"

#***************
# Various hosts
#***************
# DEC
# Gondolin: DEC
#DECHOST="skydive"	# skydive is dead.
			# gondolin rsh/ssh don't work. run as alone-host.
# HP
# sangamon: HPUX 10
# kelgia: HPUX 11
HPHOST="kelgia sangamon"	# HPUX 10 & 11 

# Linux
# Dangermouse, eirene: Linux
# Dangermouse used to die if gmake -j is used.
LINUXHOST=eirene

# SGI O2K
# modi4:	R10K,  IRIX64 6.5, default to -64,-mips4
#		Testing {parallel,serial}x{-64,-n32}x
O2KHOST=modi4
# regular, unleaded, premium:
#		R10K,  IRIX64 6.5, default to -n32,-mips4
# impact7:	R4400, IRIX 6.5,   default to -n32,-mips3
# o2-N:		R10K,  IRIX 6.5,   default to -n32,-mips4
# paz:		R4400, IRIX 6.5,   default to -n32,-mips3
SGIHOST="regular impact7 o2-3"	

# Sun
SUNHOST="arabica baldric"	# solaris 2.6 and 2.7

# FreeBSD
# AFS does not work well in hawkwind.  Use NFS/local space for its
# test directory.  ssh does not work for it either.
FREEBSDHOST="hawkwind"

# set up default all hosts to test
ALLHOSTS="$O2KHOST $SUNHOST $SGIHOST $HPHOST $LINUXHOST $DECHOST $FREEBSDHOST"

# test hosts
TESTHOST=""

#################################
# Function definitions
#################################

# Print messages to stdout
# Use this to show output heading to stdout
PRINT()
{
    echo "$*"
}

# Show seconds since midnight.
# This is used to calculate seconds elapsed
SecOfDay()
{
    set `date '+%H %M %S'`
    t_sec=`expr $1 \* 3600 + $2 \* 60 + $3`
    echo $t_sec
}

# Calculated the elapsed time (in seconds) between the first
# and second time.  If second time is smaller than the first,
# we assume the clock has passed midnight and calculate appropriately.
ElapsedTime()
{
    if [ $2 -lt $1 ]; then
	t_sec=`expr 3600 \* 24 - $1 + $2`
    else
	t_sec=`expr $2 - $1`
    fi
    echo `expr $t_sec / 60`m `expr $t_sec % 60`s
}

# Report errors
# $1--an error message to be printed
REPORT_ERR()
{
    ERRMSG=$1
    # print it with a banner shifted right a bit
    PRINT "	*************************************"
    PRINT "	`date`"
    PRINT "	$ERRMSG"
    PRINT "	*************************************"
    # report it in the FAILED-LOG file too
    PRINT "$ERRMSG" >> $FAILEDLOG
}

#
# Report results of the last test done
REPORT_RESULT()
{
    if [ $retcode -eq 0 ]; then
	PRINT "PASSED ${HOSTNAME}: $TEST_TYPE" | tee -a $PASSEDLOG
    else
	# test failed.
	REPORT_ERR "****FAILED ${HOSTNAME}: $TEST_TYPE****"
    fi
}

# Print a blank line
PRINT_BLANK()
{
    PRINT
}

# Print test trailer
PRINT_TEST_TRAILER()
{
    PRINT "*** finished $TEST_TYPE tests for $HOSTNAME ***"
    date; EndTime=`SecOfDay`
    PRINT Total time = `ElapsedTime $StartTime $EndTime`
    PRINT_BLANK
}

# Print trailer summary
PRINT_TRAILER()
{
    PRINT "*** finished tests in $HOSTNAME ***"
    date; TotalEndTime=`SecOfDay`
    PRINT Grand total tests time = `ElapsedTime $TotalStartTime $TotalEndTime`
    PRINT_BLANK
}

# Figure out which remote command to use to reach a host.
# Try rsh first, then ssh.
# $1--hostname to reach.
CHECK_RSH()
{
    # Figure out how to use ping command in this host.
    # Some hosts use "ping host count", some use "ping -c count host"
    # Test "ping -c ..." style first because some '-c' machines treat
    # the command 'ping localhost 3' means to ping host '3'.
    if [ -z "$PING" ]; then
	if ping -c 3 localhost >/dev/null 2>&1; then
	    PING='ping -c 3'
	    PINGCOUNT=
	elif ping localhost 3 >/dev/null 2>&1; then
	    PING=ping
	    PINGCOUNT=3
	else	# don't know how to use ping.
	    PING=no_ping
	    PINGCOUNT=
	fi
    fi
    #
    host=$1
    # Try remote command with host if it responds to ping.
    # Still try it if we don't know how to do ping.
    if [ no_ping = "$PING" ] || $PING $host $PINGCOUNT >/dev/null 2>&1; then
	if rsh $host -n hostname >/dev/null 2>&1; then
	    RSH=rsh
	elif ssh $host -n hostname >/dev/null 2>&1; then
	    RSH=ssh
	else
	    PRINT cannot remote command with $host
	    RSH="NoRemoteCommand"
	fi
    else
	RSH="NotReachable"
    fi
}


# Wait for a file for at most number of minutes
# $1--the file
# $2--number of minutes
# WAIT_STATUS set to:
#	-1 if errors encountered
#	0  if file found within time limit
#	1  if file not found within time limit
WAITFOR()
{
    wait_file=$1
    nminutes=$2
    if [ -z "$wait_file" -o ! "$nminutes" -ge 0 ]
    then
	PRINT "errors in argument of WAITFOR(): wait_file($1) or nminutes($2)"
	WAIT_STATUS=-1
	return
    fi
    while [ ! -f $wait_file ]; do
	if [ $nminutes -gt 0 ]; then
	    PRINT "Wait For $wait_file to appear"
	    sleep 60			#sleep 1 minute
	else
	    WAIT_STATUS=1
	    return
	fi
	nminutes=`expr $nminutes - 1`
    done
    WAIT_STATUS=0
    return
}


# Wait till a file disappears for at most number of minutes.
# Useful to wait till a lock is removed by another process.
# $1--the file
# $2--number of minutes
# WAIT_STATUS set to:
#	-1 if errors encountered
#	0  if file disappears within time limit
#	1  if file has not disappeared within time limit
WAITTILL()
{
    wait_file=$1
    nminutes=$2
    if [ -z "$wait_file" -o ! "$nminutes" -ge 0 ]
    then
	PRINT "errors in argument of WAITTILL(): wait_file($1) or nminutes($2)"
	WAIT_STATUS=-1
	return
    fi
    while [ -f $wait_file ]; do
	if [ $nminutes -gt 0 ]; then
	    PRINT "Wait till $wait_file has disappeared"
	    sleep 60			#sleep 1 minute
	else
	    WAIT_STATUS=1
	    return
	fi
	nminutes=`expr $nminutes - 1`
    done
    WAIT_STATUS=0
    return
}


# Run one snapshot test
# $*--Types of test being run
RUNSNAPTEST()
{
    SNAPCMD_OPT="$STANDARD_OPT"		# snapshot test option
    SRCDIRNAME=""
    CC_SAVED="$CC"
    PATH_SAVED=$PATH
    export PATH			# DEC OSF1 needs to export PATH explicitly
    TEST_TYPE=$*
    retcode=0
    date
    PRINT "*** starting $TEST_TYPE tests in $HOSTNAME ***"
    PRINT "Uname -a: `uname -a`"

    # Parse the test type and set options accordingly.
    # See comments of SNAPTEST_CONFIG_PARSE().
    while [ $# -gt 0 ]; do
	case $1 in
	    -n32) # want -n32 option
		SRCDIRNAME=${SRCDIRNAME}-n32
		CC="cc -n32"
		export CC
		;;
	    parallel) # want parallel test
		SNAPCMD_OPT="$SNAPCMD_OPT $ENABLE_PARALLEL"
		SRCDIRNAME=${SRCDIRNAME}-pp
		;;
	    standard) # standard test
		;;
	    --*)
		# option for configure
		SNAPCMD_OPT="$SNAPCMD_OPT $1"
		;;
	    op-configure)
		# option for configure
		SNAPCMD_OPT="$SNAPCMD_OPT $1 $2"
		shift
		;;
	    setenv)
		# set environment variable
		shift
		eval $1="$2"
		export $1
		shift
		;;
	    *) # unknown test
		PRINT "$0: unknown type of test ($1)"
		retcode=1
		;;
	esac
	shift
    done
    [ $retcode -ne 0 ] && errcode=$retcode && return $retcode

    # Track down the HDF4 software
    ans=`$SNAPYARD/current/bin/locate_hdf4`
    H4_SW=`echo $ans | cut -f1 -d' '`
    H4_BIN=`echo $ans | cut -f2 -s -d' '`
    if [ -n "$H4_SW" ]; then
	SNAPCMD_OPT="$SNAPCMD_OPT hdf4 $H4_SW"
    fi
    if [ -n "$H4_BIN" ]; then
	PATH=${PATH}:${H4_BIN}
    fi
	
    if [ -n "${SRCDIRNAME}" ]; then
	SNAPCMD_OPT="$SNAPCMD_OPT srcdirname ${SRCDIRNAME}"
    fi

    # Setup log file name to save test output
    LOGFILE=${LOGBASENAME}${SRCDIRNAME}_${TODAY}
    PRINT "Running snapshot with output saved in"
    PRINT "   $LOGFILE"
    (date; PRINT Hostname=$HOSTNAME) >> $LOGFILE

    (
    cd $SNAPYARD/current
    $SNAPSHOT $SNAPCMD $SNAPCMD_OPT
    ) >> $LOGFILE 2>&1
    retcode=$?
    [ $retcode -ne 0 ] && errcode=$retcode

    date >> $LOGFILE

    # restore CC, PATH
    CC="$CC_SAVED"
    PATH=$PATH_SAVED
}


# configuration parsing.
# Taking configuration from input.
# This should be invoke with configure file as stdin.
# Syntax of the configure file:
#    All lines started with the # are comment lines and are ignored.
#    Blank lines are ignored too.
#    Each config line starts with a "Scope" followed by test types.
#
# Scope can be:
#    standard ...          # what the standard test types are.
#    <host>: <test>        Do <test> for <host> 
#    all: <test>           Do <test> for all hosts. 
#    <weekday>/...         Use this scope if the <weekday> matches.
#                          <weekday> can be {Mon,Tue,Wed,Thu,Fri,Sat,Sun}
#    If no <host>: input for a <host>, the standard test is used.
#
# Test types:
#    standard		tests defined in standard scope.
#    -n32		-n32 mode.  Apply to 64/32 bit OS such as IRIX64.
#    parallel		parallel mode.
#    op-configure <option>	configure option
#    --*		configure option
#    setenv <name> <value>	set environment variable <name> to <value>
SNAPTEST_CONFIG_PARSE()
{
    while read x y ; do
	# Scan for entry for this weekday.
	xd=`echo $x | cut -f1 -d/`
	if [ "$xd" = ${WEEKDAY} ]; then
	    # strip away the weekday/ part.
	    x=`echo $x | cut -f2 -d/`
	fi
	case "$x" in
	    '' | '#'*)
		# blank or comment lines.  Continue.
	    	;;
	    ???/*)
		# Ignore any entry not of this weekday.
		;;
	    standard)
		#standard configuration
		STANDARD_OPT="$y"
		;;
	    all: | ${HOSTNAME}:)
		# types of test for all hosts or this host
		if [ -n "$TEST_TYPES" ]; then
		    TEST_TYPES="$TEST_TYPES ; $y"
		else
		    TEST_TYPES="$y"
		fi
		;;
	    *:)	# ignore types of test for other hosts
		;;
	    *)	# unknown configuration option
		PRINT $x $y
		PRINT "***Unknown configuration option. Ignored.***"
		;;
	esac
    done
}

# Snap Test configuration parsing.
# If TEST_TYPES is not set, set it to do the "standard" test.
SNAPTEST_CONFIG()
{
    TEST_TYPES=
    STANDARD_OPT=
    if [ -f $SNAPTESTCFG ]; then
	SNAPTEST_CONFIG_PARSE < $SNAPTESTCFG
    fi
    TEST_TYPES=${TEST_TYPES:-'standard'}
}


# Flush the AFS files if applicable.
# Hopefully the flushing is done when the tests of this
# host are done rather than when the launching site try
# to pull them in at the same time.  This way, the afs
# server updates are spread out.
FLUSH_FILES()
{
    /usr/afsws/bin/fs flush $SNAPYARD
}


# Show usage page
USAGE()
{
cat <<EOF
Usage: runtest [-h] [-debug] [-r<version>] [-all] [-nocvs] [<host> ...]
    -h
	print this help page
    -debug
	turn on debug mode
    -r<version>
	do runtest for <version>
    -all
	launch tests for all pre-defined testing hosts
    -nocvs
	do not do cvs commands
    <host>
	launch tests for <host>
    
-all and <host> are contradictory and whichever is specified last, is
the one to take effect.  If neither are given, do the test for the
local host.
EOF
}


#################################
# Main
#################################
#################################
# Set up global variables
#################################
retcode=0			# error code of individula task
errcode=0			# error code of the whole test


#################################
# Parse options
#################################
while [ $# -gt 0 ]; do
    case "$1" in
	-h) # help--show usage
	    USAGE
	    exit 0
	    ;;
	-debug*)
	    # set debug mode
	    DEBUGMODE="$1"
	    SNAPSHOT="echo bin/snapshot"
	    PROGNAME="$PROGNAME $DEBUGMODE"
	    PRINT "******** DEBUGMODE is $DEBUGMODE ************"
	    ;;
	-r*)
	    # the version string has a leading _ but not for H5DIR name
	    H5VER="$1"
	    H5VERSTR=_`echo $H5VER | sed -e s/-r// -e s/\\\./_/g`
	    H5DIR=$HOME/HDF5/v$H5VERSTR/hdf5
	    PROGNAME="$PROGNAME $H5VER"
	    ;;
	-all)
	    # cannot assign ALLHOSTS yet.
	    TESTHOST=-all
	    ;;
	-nocvs)
	    # do not do cvs commands
	    NOCVS=nocvs
	    ;;
	-*) # Unknow option
	    PRINT "Unknown option ($1)"
	    USAGE
	    exit 1
	    ;;
	*)
	    TESTHOST=$*
	    break
	    ;;
    esac
    shift
done

#################################
# Setup snapshot test directories
#################################
# Show the real physical path rather than the symbolic path
SNAPYARD=`cd $HOME/snapshots-hdf5${H5VERSTR} && /bin/pwd`
# Log file basename
LOGBASENAME=${SNAPYARD}/log/${HOSTNAME}
PASSEDLOG=${SNAPYARD}/log/PASSED_LOG_${TODAY}
FAILEDLOG=${SNAPYARD}/log/FAILED_LOG_${TODAY}
CVSLOG=${SNAPYARD}/log/CVS_LOG_${TODAY}
CVSLOG_LOCK=${SNAPYARD}/log/CVS_LOG_LOCK_${TODAY}
# Snap Test Configuration file
SNAPTESTCFG=${SNAPYARD}/snaptest.cfg


#################################
# Setup test host(s)
#################################
ALLHOSTSFILE=${SNAPYARD}/allhostfile
if [ -f $ALLHOSTSFILE ]; then
    ALLHOSTS=`grep -v '#' $ALLHOSTSFILE`
fi
if [ "$TESTHOST" = -all ]; then
    TESTHOST=$ALLHOSTS
fi


#################################
# Setup to print a trailer summary when exiting not via
# the normal end of the script.
#################################
trap PRINT_TRAILER 0

#
TotalStartTime=`SecOfDay`

# Process the configuration
SNAPTEST_CONFIG
PRINT STANDARD_OPT=$STANDARD_OPT
PRINT TEST_TYPES=$TEST_TYPES
PRINT_BLANK

# Do a checkout if one has not been done today
# Also check MANIFEST file
if [ -z "$NOCVS" -a ! -f $CVSLOG ]; then
    PRINT "Running CVS checkout with output saved in"
    PRINT "   $CVSLOG"
    # Set CVS lock first
    touch $CVSLOG_LOCK
    ($SNAPSHOT checkout ) >> $CVSLOG 2>&1
    # Save error code and remove the lock
    errcode=$?
    rm -f $CVSLOG_LOCK
    if [ $errcode -ne 0 ]; then
	# test failed.
	REPORT_ERR "****FAILED ${HOSTNAME}: CVS checkout****"
	exit $errcode
    fi
    PRINT Checking MAINFEST file ...
    (cd $SNAPYARD/current; bin/chkmanifest)
    errcode=$?
    if [ $errcode -ne 0 ]; then
	# test failed.
	REPORT_ERR "****FAILED ${HOSTNAME}: MANIFEST check****"
    fi
    PRINT_BLANK
else
    # make sure the cvs update, if done by another host, has completed.
    # First wait for the presence of $CVSLOG which signals some host
    # has started the cvs update.  Then wait for the absense of $CVSLOG_LOCK
    # which signals the host has completed the cvs update.
    WAITFOR $CVSLOG 90
    if [ $WAIT_STATUS -ne 0 ]; then
	errcode=$WAIT_STATUS
	REPORT_ERR "****FAILED ${HOSTNAME}: Time expired waiting CVS update to start****"
	exit $errcode
    fi
    WAITTILL $CVSLOG_LOCK 10
    if [ $WAIT_STATUS -ne 0 ]; then
	errcode=$WAIT_STATUS
	REPORT_ERR "****FAILED ${HOSTNAME}: Time expired waiting CVS update to finish****"
	exit $errcode
    fi
fi

# we can use the version of script in SNAPYARD/current now
PROGNAME="$SNAPYARD/current/$PROGNAME"

# Decide to do test for the local host or for remote hosts
if [ -n "$TESTHOST" -a $HOSTNAME != "$TESTHOST" ]; then
    date
    PRINT "*** launching tests from $HOSTNAME ***"
    PRINT_BLANK
    TEST_TYPE="launching"
    cd ${SNAPYARD}/log
    for h in $TESTHOST; do
	TMP_OUTPUT="#$h.out"
	(PRINT "=============="
	 PRINT "Testing $h"
	 PRINT "==============") > $TMP_OUTPUT
	CHECK_RSH $h
	# run the remote shell command with output to $TMP_OUTPUT
	case "$RSH" in
	    rsh|ssh)
		PRINT $RSH $h -n $PROGNAME
		# launch concurrent tests only if srcdir is used
		if [ -n "$SRCDIR" ]; then
		    $RSH $h -n $PROGNAME &
		else
		    $RSH $h -n $PROGNAME
		fi
		;;
	    NoRemoteCommand)
		PRINT $h does not accept Remote Command
		;;
	    NotReachable)
		PRINT $h is not reachable 
		;;
	    *)
		PRINT "CHECK_RSH for $h returned unknow result ($RSH)"
		;;
	esac >> $TMP_OUTPUT 2>&1
    done
    # wait for all launched tests to finish, then cat them back out.
    wait
    for h in $TESTHOST; do
	TMP_OUTPUT="#$h.out"
	cat $TMP_OUTPUT
	# Verify test script did complete by checking the last lines
	(tail -2 $TMP_OUTPUT | grep -s '^Grand total' > /dev/null 2>&1) ||
	    (REPORT_ERR "****FAILED ${h}: snaptest did not complete****" &&
		PRINT_BLANK)
	rm $TMP_OUTPUT
    done
    exit 0
fi

# run the test(s)
# Note that first field is cut without -s but all subsequent cut
# must use -s.  If -s is not used at all, a $TEST_TYPES that has
# no ';' (only 1 test), will pass through intact in all cut. That
# results in infinite looping.
# If -s is used with the first field, it will suppress completely
# a $TYPE_TYPES that has no ';' (only 1 tst ).  That results in no
# test at all.
# 
n_test=1
TEST="`echo $TEST_TYPES | cut -f1 -d';'`"
while [ -n "$TEST" ]; do
    StartTime=`SecOfDay`
    RUNSNAPTEST $TEST
    REPORT_RESULT
    PRINT_TEST_TRAILER

    n_test=`expr $n_test + 1`
    TEST="`echo $TEST_TYPES | cut -f$n_test -s -d';'`"
done

FLUSH_FILES

PRINT_TRAILER

# disable trailer summary printing since all trailers have been
# printed and we are exiting normally.
trap 0
exit $errcode