Avoid a non-writable-by-the-user file when copying xattrs.

[rsync/rsync.git] / runtests.sh

#! /bin/sh

# Copyright (C) 2001, 2002 by Martin Pool <mbp@samba.org>
# Copyright (C) 2003, 2004, 2005, 2006 Wayne Davison

# This program is free software; you can redistribute it and/or modify
# it under the terms of the GNU General Public License version
# 2 as published by the Free Software Foundation.
#
# This program 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
# Lesser General Public License for more details.
# 
# You should have received a copy of the GNU Lesser General Public
# License along with this program; if not, write to the Free Software
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA

# rsync top-level test script -- this invokes all the other more
# detailed tests in order.  This script can either be called by `make
# check' or `make installcheck'.  `check' runs against the copies of
# the program and other files in the build directory, and
# `installcheck' against the installed copy of the program.  

# In either case we need to also be able to find the source directory,
# since we read test scripts and possibly other information from
# there.

# Whenever possible, informational messages are written to stdout and
# error messages to stderr.  They're separated out by the build farm
# display scripts.

# According to the GNU autoconf manual, the only valid place to set up
# directory locations is through Make, since users are allowed to (try
# to) change their mind on the Make command line.  So, Make has to
# pass in all the values we need.

# For other configured settings we read ./config.sh, which tells us
# about shell commands on this machine and similar things.

# rsync_bin gives the location of the rsync binary.  This is either
# builddir/rsync if we're testing an uninstalled copy, or
# install_prefix/bin/rsync if we're testing an installed copy.  On the
# build farm rsync will be installed, but into a scratch /usr.

# srcdir gives the location of the source tree, which lets us find the
# build scripts.  At the moment we assume we are invoked from the
# source directory.

# This script must be invoked from the build directory.  

# A scratch directory, 'testtmp', is used in the build directory to
# hold per-test subdirectories.

# This script also uses the $loglevel environment variable.  1 is the
# default value, and 10 the most verbose.  You can set this from the
# Make command line.  It's also set by the build farm to give more
# detail for failing builds.


# NOTES FOR TEST CASES:

# Each test case runs in its own shell. 

# Exit codes from tests:

#    1  tests failed
#    2  error in starting tests
#   77  this test skipped (random value unlikely to happen by chance, same as
#       automake)

# HOWEVER, the overall exit code to the farm is different: we return
# the *number of tests that failed*, so that it will show up nicely in
# the overall summary.

# rsync.fns contains some general setup functions and definitions.


# NOTES ON PORTABILITY:

# Both this script and the Makefile have to be pretty conservative
# about which Unix features they use.

# We cannot count on Make exporting variables to commands, unless
# they're explicitly given on the command line.

# Also, we can't count on 'cp -a' or 'mkdir -p', although they're
# pretty handy (see function makepath for the latter).

# I think some of the GNU documentation suggests that we shouldn't
# rely on shell functions.  However, the Bash manual seems to say that
# they're in POSIX 1003.2, and since the build farm relies on them
# they're probably working on most machines we really care about.

# You cannot use "function foo {" syntax, but must instead say "foo()
# {", or it breaks on FreeBSD.

# BSD machines tend not to have "head" or "seq".

# You cannot do "export VAR=VALUE" all on one line; the export must be
# separate from the assignment.  (SCO SysV)

# Don't rely on grep -q, as that doesn't work everywhere -- just redirect
# stdout to /dev/null to keep it quiet.


# STILL TO DO:

# We need a good protection against tests that hang indefinitely.
# Perhaps some combination of starting them in the background, wait,
# and kill?

# Perhaps we need a common way to cleanup tests.  At the moment just
# clobbering the directory when we're done should be enough.

# If any of the targets fail, then (GNU?) Make returns 2, instead of
# the return code from the failing command.  This is fine, but it
# means that the build farm just shows "2" for failed tests, not the
# number of tests that actually failed.  For more details we might
# need to grovel through the log files to find a line saying how many
# failed.


set -e

. "./shconfig"

RUNSHFLAGS='-e'
export RUNSHFLAGS

# for Solaris
[ -d /usr/xpg4/bin ] && PATH="/usr/xpg4/bin/:$PATH"

if [ "x$loglevel" != x ] && [ "$loglevel" -gt 8 ]; then
    if set -x; then
	# If it doesn't work the first time, don't keep trying.
	RUNSHFLAGS="$RUNSHFLAGS -x"
    fi
fi

POSIXLY_CORRECT=1 
if test x"$TOOLDIR" = x; then
    TOOLDIR=`pwd`
fi
srcdir=`dirname $0`
if test x"$srcdir" = x -o x"$srcdir" = x.; then
    srcdir="$TOOLDIR"
fi
if test x"$rsync_bin" = x; then
    rsync_bin="$TOOLDIR/rsync"
fi

# This allows the user to specify extra rsync options -- use carefully!
RSYNC="$rsync_bin $*"
#RSYNC="valgrind $rsync_bin $*"

TLS_ARGS=''
if egrep '^#define HAVE_LUTIMES 1' config.h >/dev/null; then
    TLS_ARGS="$TLS_ARGS -l"
fi
if egrep '#undef CHOWN_MODIFIES_SYMLINK' config.h >/dev/null; then
    TLS_ARGS="$TLS_ARGS -L"
fi

export POSIXLY_CORRECT TOOLDIR srcdir RSYNC TLS_ARGS

echo "============================================================"
echo "$0 running in $TOOLDIR"
echo "    rsync_bin=$RSYNC"
echo "    srcdir=$srcdir"
echo "    TLS_ARGS=$TLS_ARGS"

if [ -f /usr/bin/whoami ]; then
    testuser=`/usr/bin/whoami`
elif [ -f /usr/ucb/whoami ]; then
    testuser=`/usr/ucb/whoami`
elif [ -f /bin/whoami ]; then
    testuser=`/bin/whoami`
else
    testuser=`id -un 2>/dev/null || echo ${LOGNAME:-${USERNAME:-${USER:-'UNKNOWN'}}}`
fi

echo "    testuser=$testuser"
echo "    os=`uname -a`"

# It must be "yes", not just nonnull
if [ "x$preserve_scratch" = xyes ]; then
    echo "    preserve_scratch=yes"
else
    echo "    preserve_scratch=no"
fi    

# Check if setacl/setfacl is around and if it supports the -k or -s option.
if setacl -k u::7,g::5,o:5 testsuite 2>/dev/null; then
    setfacl_nodef='setacl -k'
elif setfacl --help 2>&1 | grep ' -k,\|\[-[a-z]*k' >/dev/null; then
    setfacl_nodef='setfacl -k'
elif setfacl -s u::7,g::5,o:5 testsuite 2>/dev/null; then
    setfacl_nodef='setfacl -s u::7,g::5,o:5'
else
    # The "true" command runs successfully, but does nothing.
    setfacl_nodef=true
fi

export setfacl_nodef

if [ ! -f "$rsync_bin" ]; then
    echo "rsync_bin $rsync_bin is not a file" >&2
    exit 2
fi

if [ ! -d "$srcdir" ]; then
    echo "srcdir $srcdir is not a directory" >&2
    exit 2
fi

skipped=0
missing=0
passed=0
failed=0

# Directory that holds the other test subdirs.  We create separate dirs
# inside for each test case, so that they can be left behind in case of
# failure to aid investigation.  We don't remove the testtmp subdir at
# the end so that it can be configured as a symlink to a filesystem that
# has ACLs and xattr support enabled (if desired).
scratchbase="$TOOLDIR"/testtmp
echo "    scratchbase=$scratchbase"
[ -d "$scratchbase" ] || mkdir "$scratchbase"

suitedir="$srcdir/testsuite"
TESTRUN_TIMEOUT=300

export scratchdir suitedir TESTRUN_TIMEOUT

prep_scratch() {
    [ -d "$scratchdir" ] && chmod -R u+rwX "$scratchdir" && rm -rf "$scratchdir"
    mkdir "$scratchdir"
    # Get rid of default ACLs and dir-setgid to avoid confusing some tests.
    $setfacl_nodef "$scratchdir" || true
    chmod g-s "$scratchdir"
    case "$srcdir" in
    /*) ln -s "$srcdir" "$scratchdir/src" ;;
    *)  ln -s "$TOOLDIR/$srcdir" "$scratchdir/src" ;;
    esac
    return 0
}

maybe_discard_scratch() {
    [ x"$preserve_scratch" != xyes ] && [ -d "$scratchdir" ] && rm -rf "$scratchdir"
    return 0
}

if [ "x$whichtests" = x ]; then
    whichtests="*.test"
fi

for testscript in $suitedir/$whichtests
do
    testbase=`echo $testscript | sed -e 's!.*/!!' -e 's/.test\$//'`
    scratchdir="$scratchbase/$testbase"

    prep_scratch

    case "$testscript" in
    *hardlinks*) TESTRUN_TIMEOUT=600 ;;
    *) TESTRUN_TIMEOUT=300 ;;
    esac

    set +e
    "$TOOLDIR/"testrun $RUNSHFLAGS "$testscript" >"$scratchdir/test.log" 2>&1
    result=$?
    set -e

    if [ "x$always_log" = xyes -o \( $result != 0 -a $result != 77 -a $result != 78 \) ]
    then
	echo "----- $testbase log follows"
	cat "$scratchdir/test.log"
	echo "----- $testbase log ends"
	if [ -f "$scratchdir/rsyncd.log" ]; then
	    echo "----- $testbase rsyncd.log follows"
	    cat "$scratchdir/rsyncd.log"
	    echo "----- $testbase rsyncd.log ends"
	fi
    fi

    case $result in
    0)
	echo "PASS    $testbase"
	passed=`expr $passed + 1`
	maybe_discard_scratch
	;;
    77)
	# backticks will fill the whole file onto one line, which is a feature
	whyskipped=`cat "$scratchdir/whyskipped"`
	echo "SKIP    $testbase ($whyskipped)"
	skipped=`expr $skipped + 1`
	maybe_discard_scratch
	;;
    78)
        # It failed, but we expected that.  don't dump out error logs, 
	# because most users won't want to see them.  But do leave
	# the working directory around.
	echo "XFAIL   $testbase"
	failed=`expr $failed + 1`
	;;
    *)
	echo "FAIL    $testbase"
	failed=`expr $failed + 1`
	if [ "x$nopersist" = xyes ]; then
	    exit 1
	fi
    esac
done

echo '------------------------------------------------------------'
echo "----- overall results:"
echo "      $passed passed"
[ "$failed" -gt 0 ]  && echo "      $failed failed"
[ "$skipped" -gt 0 ] && echo "      $skipped skipped"
[ "$missing" -gt 0 ] && echo "      $missing missing"
echo '------------------------------------------------------------'

# OK, so expr exits with 0 if the result is neither null nor zero; and
# 1 if the expression is null or zero.  This is the opposite of what
# we want, and if we just call expr then this script will always fail,
# because -e is set.

result=`expr $failed + $missing || true`
echo "overall result is $result"
exit $result