GIT+CYGWIN annoying test failure

Hello,

First of all, this problem always existed to me. I was lazy to report. Please check the failure logs and my GIT build script below. My build script runs 'make test' twice because of this annoying failure. It does not mean that the first test always failed, but very often. If it occurs, the first test failed at 't1000-read-tree-m-3way.sh' always. And in this case, 2nd test always succeeded.

Thanks,

GIT test failure logs
====================================================================

2007-01-15
==============================================================================
GIT_VERSION = 1.5.0.rc1.gf4b6c-dirty

...
* FAIL 57: 5 - must match in !O && A && B && A==B case.

        rm -f .git/index LL &&
             cp .orig-A/LL LL &&
             git-update-index --add LL &&
             echo extra >>LL &&
             git-read-tree -m 3fe085e0589de4327971d50e416fc292dd00fbfe

997bbc4a0a51e0574168a4f637739380edebe4d7 76d47d681d7f1d4fa975334a9a0ba8a6eeea2226 &&
check_result
...
* FAIL 62: 9 - must match and be up-to-date in O && A && !B && O!=A case

        rm -f .git/index MD &&
             cp .orig-A/MD MD &&
             git-update-index --add MD &&
             git-read-tree -m 3fe085e0589de4327971d50e416fc292dd00fbfe

997bbc4a0a51e0574168a4f637739380edebe4d7 76d47d681d7f1d4fa975334a9a0ba8a6eeea2226 &&
check_result
...
* FAIL 65: 10 - must match and be up-to-date in O && A && !B && O==A case

        rm -f .git/index ND &&
             cp .orig-A/ND ND &&
             git-update-index --add ND &&
             git-read-tree -m 3fe085e0589de4327971d50e416fc292dd00fbfe

997bbc4a0a51e0574168a4f637739380edebe4d7 76d47d681d7f1d4fa975334a9a0ba8a6eeea2226 &&
check_result
...
* failed 3 among 83 test(s)
make[1]: *** [t1000-read-tree-m-3way.sh] Error 1 make[1]: Leaving directory `/c/user/namsh/unix/git/t'
make: *** [test] Error 2

2007-01-18
==============================================================================
GIT_VERSION = 1.5.0.rc1.g556b-dirty

...
* FAIL 50: 3 - must match A in !O && A && !B case.

        rm -f .git/index AN &&
             cp .orig-A/AN AN &&
             git-update-index --add AN &&
             git-read-tree -m 3fe085e0589de4327971d50e416fc292dd00fbfe

997bbc4a0a51e0574168a4f637739380edebe4d7 76d47d681d7f1d4fa975334a9a0ba8a6eeea2226 &&
check_result
...
* FAIL 53: 4 - must match and be up-to-date in !O && A && B && A!=B case.

        rm -f .git/index AA &&
             cp .orig-A/AA AA &&
             git-update-index --add AA &&
             git-read-tree -m 3fe085e0589de4327971d50e416fc292dd00fbfe

997bbc4a0a51e0574168a4f637739380edebe4d7 76d47d681d7f1d4fa975334a9a0ba8a6eeea2226 &&
check_result
...
* failed 2 among 83 test(s)
make[1]: *** [t1000-read-tree-m-3way.sh] Error 1 make[1]: Leaving directory `/c/user/namsh/unix/git/t'
make: *** [test] Error 2

2007-01-24
==============================================================================
GIT_VERSION = 1.5.0.rc2.gc9a89-dirty

...
* FAIL 50: 3 - must match A in !O && A && !B case.

        rm -f .git/index AN &&
             cp .orig-A/AN AN &&
             git-update-index --add AN &&
             git-read-tree -m 3fe085e0589de4327971d50e416fc292dd00fbfe

997bbc4a0a51e0574168a4f637739380edebe4d7 76d47d681d7f1d4fa975334a9a0ba8a6eeea2226 &&
check_result
...
* failed 1 among 83 test(s)
make[1]: *** [t1000-read-tree-m-3way.sh] Error 1 make[1]: Leaving directory `/c/user/namsh/unix/git/t'
make: *** [test] Error 2

GIT build script
====================================================================
...
if test ! -r ./configure || test ./configure.ac -nt ./configure then
printf "Run autoconf..."
autoconf
echo " DONE!"
fi
if test ! -r ./config.mak.autogen \
|| test ./configure -nt ./config.mak.autogen
then
./configure --prefix=$prefix
fi

# generate personal configuration makefile
[ -r config.mak ] && rm -f config.mak
touch config.mak
[ -r config.mak ] || die "Cannot generate config.mak" $E "CFLAGS=$CFLAGS" >> config.mak
$E "DESTDIR=$DESTDIR" >> config.mak
# git's configure script could not find the libiconv for CYGWIN. $E "NEEDS_LIBICONV=1\nNO_ICONV=" >> config.mak
$E "NO_PERL_MAKEMAKER=1" >> config.mak # have no PERL MakeMaker $E "NO_TRUSTABLE_FILEMODE=" >> config.mak # default filemode = true # FAT32 has issues with 'mmap', but I only have NTFS. $E "NO_MMAP=" >> config.mak

if make all doc
then
# I don't use CVS/Subversion. So skip those tests. export GIT_SKIP_TESTS='t91?? t92??'
if ! make test
then
${E} "\n*** ${R}Test failed${N} ***\n"

# I don't know why... Sometimes first 'make test' failed. printf "Try again... "
if ! make test >/dev/null 2>&1
then
die "2nd test failed also."
fi
printf "OK!"
fi

rm -f "$gitLog"
if ! make install install-doc > "$gitLog" 2>&1 then
die "Install failed. Check $gitLog"
fi
rm -f "$gitLog"
...
fi

Hi,

It is not directly apparent what is causing this error. Could you please change your script to use "-i -v" on the tests? I.e. instead of:

this:

	    if ! (cd t; GIT_TEST_OPTS="-i -v" make)

It should give us a better idea what happened, since we also see the error messages of the git commands.

If it is still not apparent what went wrong, there is unfortunately no way around doing that particular test case manually, as to find out what went wrong. Since you seem to be the only one who can reproduce this bug, it boils down to you having to do the leg work.

Please keep us posted!

Ciao,
Dscho

Hello,

I'm not sure I can believe my system, Windows or CYGWIN. It seems GIT has no problem.

Thanks,

2007-01-25
=========================================================================
* expecting success: rm -fr [NDMALTS][NDMALTSF] Z &&

     rm .git/index &&
     git-read-tree 997bbc4a0a51e0574168a4f637739380edebe4d7 &&
     git-checkout-index -f -u -a &&
     git-read-tree -m 3fe085e0589de4327971d50e416fc292dd00fbfe

997bbc4a0a51e0574168a4f637739380edebe4d7 76d47d681d7f1d4fa975334a9a0ba8a6eeea2226 &&
check_result
t1000-read-tree-m-3way.sh: line 133: current: Permission denied
* FAIL 46: 3-way merge with git-read-tree -m, match H

        rm -fr [NDMALTS][NDMALTSF] Z &&
             rm .git/index &&
             git-read-tree 997bbc4a0a51e0574168a4f637739380edebe4d7 &&
             git-checkout-index -f -u -a &&
             git-read-tree -m 3fe085e0589de4327971d50e416fc292dd00fbfe

997bbc4a0a51e0574168a4f637739380edebe4d7
76d47d681d7f1d4fa975334a9a0ba8a6eeea2226 &&
check_result
make: *** [t1000-read-tree-m-3way.sh] Error 1

2007-01-29
=========================================================================
* expecting success: rm -f .git/index LL &&

     cp .orig-A/LL LL &&
     git-update-index --add LL &&
     echo extra >>LL &&
     git-read-tree -m 3fe085e0589de4327971d50e416fc292dd00fbfe

997bbc4a0a51e0574168a4f637739380edebe4d7 76d47d681d7f1d4fa975334a9a0ba8a6eeea2226 &&
check_result
t1000-read-tree-m-3way.sh: line 133: current: Permission denied
* FAIL 57: 5 - must match in !O && A && B && A==B case.

        rm -f .git/index LL &&
             cp .orig-A/LL LL &&
             git-update-index --add LL &&
             echo extra >>LL &&
             git-read-tree -m 3fe085e0589de4327971d50e416fc292dd00fbfe

997bbc4a0a51e0574168a4f637739380edebe4d7
76d47d681d7f1d4fa975334a9a0ba8a6eeea2226 &&
check_result
make: *** [t1000-read-tree-m-3way.sh] Error 1

Hi,

Are you perchance on a FAT32 system?

Ciao,
Dscho

No, I have 3 NTFS partitions (C=program, D=work/data, E=temp). There's no FAT32.

Though I have no FAT32, I even tried the NO_MMAP thing. And it didn't help.

Thanks,
namsh

Hello,

It seems the patch below work-around the problem. I tested this patch several times after knowing the problem.

Thanks,
namsh

index d0af8c3..422922e 100755
@@ -130,6 +130,7 @@ _x40='[0-9a-f][0-9a-f][0-9a-f][0-9a-f][0-9a-f]'
 _x40="$_x40$_x40$_x40$_x40$_x40$_x40$_x40$_x40"
 
 check_result () {
+    [ -r current ] && rm -f current
     git-ls-files --stage | sed -e 's/ '"$_x40"' / X /' >current &&
     diff -u expected current
 }

Hi,

Are you having a restrictive umask, or what are the permissions on that file?

Ciao,
Dscho

Hello,

There's no problem with the umask.

$ umask
0022

I guess there is some race condition in CYGWIN or windows. And my system triggers that condition.

Check the logic above. I cannot think there is any chance that 'rm -f current && echo x >current' is success, but 'echo x >current' is fail?

Thanks,
namsh

I would have written:

	check_result () {
        	rm -f current &&
                git-ls-files --stage | ... &&
                diff -u expected current
	}

If the above still makes the tests pass for your environment, that would be good. I am not very happy about the lack of "&&" after "rm -f current" in your patch.

It seems as if 'current' is sometimes writable and sometimes not, and when the shell couldn't open it to redirect sed output the whole test fails for you. That does not sound like a sane, reliable filesystem at all, and even if you worked around the problem in that single shell function, we tend to use shell redirection to create or overwrite files everywhere (both in tests and in the real Porcelain scripts) so that means git is a disaster waiting to happen on your system, which is not good.

What I am trying to get at is to find out what is causing the problem first, before working it around. Then, if we are lucky, we may be able to add a notice in the INSTALL file that says "It was reported that Filfre virus scanner makes shell redirect in Cygwin occasionally fail, and both git itself and its testsuite may not work as expected --- please do not use it", or something like that.

"Filfre virus scanner" is totally made up, but I think you get the point.