1.5.3-rc5 regression on OS X?

Hi,

When I try to build and test 1.5.3-rc5 on my OS X box, I get a failure on "make test"

*   ok 82: 15 (fail) - must match A in O && A && B && O==A && O==B case
* FAIL 83: 16 - A matches in one and B matches in another.
        rm -f .git/index F16 &&
            echo F16 >F16 &&
            git update-index --add F16 &&
            tree0=`git write-tree` &&
            echo E16 >F16 &&
            git update-index F16 &&
            tree1=`git write-tree` &&
            git read-tree -m $tree0 $tree1 $tree1 $tree0 &&
            git ls-files --stage
* failed 1 among 83 test(s)
make[1]: *** [t1000-read-tree-m-3way.sh] Error 1
make: *** [test] Error 2

I was able to build and test git 1.5.2.4 fine, with no errors in the test suite. 

I also just build 1.5.3-rc5 on a redhat box, with no errors in the testsuite. 

My OS X box is a quad G5 running OS X 10.4.10, with gcc version: 

perrin-meyers-power-mac-g5:~/GIT/git perrin$ gcc -v
Using built-in specs.
Target: powerpc-apple-darwin8
Configured with: /private/var/tmp/gcc/gcc-5026.obj~19/src/configure --disable-checking --prefix=/usr --mandir=/share/man --enable-languages=c,objc,c++,obj-c++ --program-transform-name=/^[cg][^+.-]*$/s/$/-4.0/ --with-gxx-include-dir=/include/gcc/darwin/4.0/c++ --build=powerpc-apple-darwin8 --host=powerpc-apple-darwin8 --target=powerpc-apple-darwin8
Thread model: posix
gcc version 4.0.0 (Apple Computer, Inc. build 5026)

I define "NO_EXPAT" when I build on OS X. 

Am I doing something wrong?

Thanks,

Perrin

Perrin Meyer <perrinmeyer@yahoo.com> writes:

The above sounds as if you built 1.5.2.4 on your Mac in not so recent past, and 1.5.3-rc5 on your Mac and RH recently.

Has anything changed on your Mac between the time you built 1.5.2.4 and failed to pass the test for 1.5.3-rc5?

On Aug 15, 2007, at 5:19 PM, Perrin Meyer wrote:

I believe you have an old version of the Developer Tools. I have no issues with that test on my MacBook Pro (2.16 GHz Intel Core Duo).

$ gcc -v
Using built-in specs.
Target: i686-apple-darwin8
Configured with: /private/var/tmp/gcc/gcc-5367.obj~1/src/configure -- 
disable-checking -enable-werror --prefix=/usr --mandir=/share/man -- 
enable-languages=c,objc,c++,obj-c++ --program-transform-name=/^[cg] 
[^.-]*$/s/$/-4.0/ --with-gxx-include-dir=/include/c++/4.0.0 --with- 
slibdir=/usr/lib --build=powerpc-apple-darwin8 --with-arch=nocona -- 
with-tune=generic --program-prefix= --host=i686-apple-darwin8 -- 
target=i686-apple-darwin8
Thread model: posix
gcc version 4.0.1 (Apple Computer, Inc. build 5367)

I don't use NO_EXPAT, but that shouldn't affect read-tree at all.

~~ Brian

Perrin> Hi,

Perrin> When I try to build and test 1.5.3-rc5 on my OS X box, I get a failure on "make test"

My t1000 test succeeded on modern OSX.  But "t1301-shared-repo" failed:

*** t1301-shared-repo.sh ***
* FAIL 1: shared=all

                mkdir sub &&
                cd sub &&
                git init --shared=all &&
                test 2 = $(git config core.sharedrepository)

* FAIL 2: update-server-info honors core.sharedRepository

                : > a1 &&
                git add a1 &&
                test_tick &&
                git commit -m a1 &&
                umask 0277 &&
                git update-server-info &&
                test 444 = $(stat -c %a .git/info/refs)

* failed 2 among 2 test(s)
make[1]: *** [t1301-shared-repo.sh] Error 1

This is on 1467b5fec376b6c43a5fa54bffd356e458a8d57c.

Oh... I notice "stat -c" there.  My stat doesn't know -c:

    localhost.local:~/MIRROR/git-GIT % stat -c %a /
    stat: illegal option -- c
    usage: stat [-FlLnqrsx] [-f format] [-t timefmt] [file ...]

Gotta stop using those non-portable options, I guess. :)

Does this help?

---
 builtin-read-tree.c |    5 ++++-
 1 files changed, 4 insertions(+), 1 deletions(-)

For multi-base merges, we allowed read-tree -m to take more than three trees (the last two are our and their branches, and all the earlier ones, typically one but potentially more, are used as the merge base). Unfortunately, the conversion done by commit 933bf40 broke this.

Signed-off-by: Junio C Hamano <gitster@pobox.com>
---
 builtin-read-tree.c |    5 ++++-
 1 files changed, 4 insertions(+), 1 deletions(-)

On Wed, 15 Aug 2007, Junio C Hamano wrote:

Sorry. I didn't even realize people did this. Oops.

I note that the thign you committed limited the number of trees to 4. May I suggest this patch in addition to your patch? Also, is 4 actually the real maximum? Maybe it should be raised?

What *is* the behaviour for four trees? Reading "unpack_trees()" source implies that all other trees go into stage 1, where the first matching entry is used?

		Linus

---
 builtin-read-tree.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

Linus Torvalds <torvalds@linux-foundation.org> writes:

This is no big deal. It would have triggered if you are using resolve strategy _and_ did a multi merge-base merge, the former of which is quite unusual these days.

Documentation/technical/trivial-merge.txt describes this under three-way case. Essentially this was an attempt to extend the original merge-resolve so that it would catch problematic criss-cross cases (case #16 in the description).

On Thu, 16 Aug 2007, Junio C Hamano wrote:

Yeah, ok. I see what's up. 

Anyway, it sounds like we should just increase MAX_TREES to get the old behaviour (and that requires applying my trivial one-liner patch too, otherwise "git read-tree" will refuse to do more than 4 trees anyway).

Even if nothing apparently actually does it.

The alternative, I think, would be to just say that the three-way merge is really just three ways, and remove the test that tests four different trees. That would allow simplifying the logic in unpack_trees(). But since the logic already exists, there's no real upside.

		Linus

Linus Torvalds <torvalds@linux-foundation.org> writes:

So no worries.  It was triggered by an error in t1000 test.

-- >8 -- [PATCH] Clean-up read-tree error condition.

This is a follow-up to f34f2b0b; list_tree() function is where it first notices that the command line fed too many trees for us to handle, so move the error exit message to there, and raise the MAX_TREES to 8 (not that it matters very much in practice).

Signed-off-by: Junio C Hamano <gitster@pobox.com>
---
 builtin-read-tree.c |    9 +++------
 1 files changed, 3 insertions(+), 6 deletions(-)

On Thu, 16 Aug 2007, Junio C Hamano wrote:

ACK. Obviously.

		Linus