What's (not) cooking

I'll be pushing out the integration branches with some updates, but there is no change in 'next' and below. The following topics I gave a quick look and gave them topic branches, but I had trouble merging them in 'pu' and making them work correctly or pass the tests, so they are not part of 'pu' in today's pushout.

    pk/rebase-in-c
    en/dirty-merge-fixes
    en/t6036-merge-recursive-tests
    en/t6042-insane-merge-rename-testcases
    ds/multi-pack-index

Junio C Hamano <gitster@pobox.com> writes:

Also I ran out of time looking at various interesting things that happened during US holiday, and didn't get around to a few interesting topics. So they are not even in the above "not in pu" list, but they did hit my mailbox. I just haven't got around to them yet.

On Fri, Jul 6, 2018 at 3:57 PM, Junio C Hamano <gitster@pobox.com> wrote:

It looks to me like the main problem is that pu itself has lots of test failures. It seems to bisect down to kg/gc-auto-windows-workaround. If I revert commit ac9d3fdbebbd ("gc --auto: clear repository before auto packing", 2018-07-04), then pu passes tests again for me. With that reverted, I can merge en/t6036-merge-recursive-tests and en/t6042-insane-merge-rename-testcases without conflicts and the tests pass without incident.

The other three topics all have merge conflicts.

en/dirty-merge-fixes has a small conflict with the new topic nd/use-the-index-compat-less, which I mentioned as a possibility in the cover letter to my series. I'm happy to do whatever makes it easiest for you to pick up; I can easily rebase on that topic branch, but I thought you wanted to see that topic redone first (to avoid "useless churn"), so I'm unsure what the right next step is.

Hi Elijah,

On Fri, 6 Jul 2018, Elijah Newren wrote:

Is this the segmentation fault about which I just sent a mail?

Ciao, Dscho

Hi Dscho,

On Sat, Jul 7, 2018 at 5:11 AM, Johannes Schindelin <Johannes.Schindelin@gmx.de> wrote:

Yes, this is is a gc segfault issue. It looks like every test calling git gc will fail because of it, I even saw a "nothing to pack" message or something like that followed by a segfault.

On (07/07/18 08:34), Elijah Newren wrote:

Sorry, I forgot to run the tests for v2 of my patch. I'll be more careful in the future.

-Kim

Hi Junio,

On Fri, 6 Jul 2018, Junio C Hamano wrote:

Quick note that `pu` is broken on Windows:

https://git-for-windows.visualstudio.com/git/git%20Team/_build/results?buildId=11901&view=logs

One quite serious looking symptom is this (line 169728 in the log):

2018-07-07T00:25:33.9171932Z ./test-lib.sh: line 664: 4516 Segmentation fault git gc

As you know, some time ago I tried to implement an automated build that auto-bisects issues like this, and unfortunately I had to disable this because it regularly ran out of time after 4h to bisect through the complex commit history (due to the fact that many branches in `pu` are not based on `master` but on commits that are way back in the past, and an automated build cannot retain information easily such as "this commit was clean, and please do not bother bisecting past it".

And sadly, previous tests of `pu` (see https://git-for-windows.visualstudio.com/git/_build/index?definitionId=1&_a=history) were failing on Windows already in the compile stage, see e.g. https://git-for-windows.visualstudio.com/git/git%20Team/_build/results?buildId=11612&view=logs

So I have no idea what caused this `gc` breakage.

Ciao, Dscho

On 7/6/2018 6:57 PM, Junio C Hamano wrote:

I tested merging ds/multi-pack-index against the latest pu and the only issue I had was with header files being added to 'packfile.c' and 'building/repack.c'. Both were that I added "#include <midx.h>" and cc/remote-odb added "#include <remote-odb.h>".

After resolving those conflicts, I could compile and test as normal. My series did not include any new test failures to these that are already failing on current pu:

t0410-partial-clone.sh                           (Wstat: 256 Tests: 16 
Failed: 2)
   Failed tests:  15-16
   Non-zero exit status: 1
t1302-repo-version.sh                            (Wstat: 256 Tests: 15 
Failed: 1)
   Failed test:  15
   Non-zero exit status: 1
t1304-default-acl.sh                             (Wstat: 256 Tests: 4 
Failed: 1)
   Failed test:  4
   Non-zero exit status: 1
t3306-notes-prune.sh                             (Wstat: 256 Tests: 12 
Failed: 2)
   Failed tests:  3, 10
   Non-zero exit status: 1
t5304-prune.sh                                   (Wstat: 256 Tests: 26 
Failed: 11)
   Failed tests:  1, 4, 6, 12-18, 20
   Non-zero exit status: 1
t5502-quickfetch.sh                              (Wstat: 256 Tests: 7 
Failed: 1)
   Failed test:  7
   Non-zero exit status: 1
t5318-commit-graph.sh                            (Wstat: 256 Tests: 63 
Failed: 1)
   Failed test:  41
   Non-zero exit status: 1
t5500-fetch-pack.sh                              (Wstat: 256 Tests: 356 
Failed: 1)
   Failed test:  43
   Non-zero exit status: 1
t6014-rev-list-all.sh                            (Wstat: 256 Tests: 4 
Failed: 1)
   Failed test:  3
   Non-zero exit status: 1
t6500-gc.sh                                      (Wstat: 256 Tests: 10 
Failed: 5)
   Failed tests:  2-3, 6-7, 9
   Non-zero exit status: 1
t6501-freshen-objects.sh                         (Wstat: 256 Tests: 29 
Failed: 7)
   Failed tests:  7, 12, 20, 25, 27-29
   Non-zero exit status: 1
t7701-repack-unpack-unreachable.sh               (Wstat: 256 Tests: 7 
Failed: 1)
   Failed test:  5
   Non-zero exit status: 1
t9300-fast-import.sh                             (Wstat: 256 Tests: 183 
Failed: 1)
   Failed test:  88
   Non-zero exit status: 1

Should I rebase onto cc/remote-odb to avoid these conflicts in the future? Or, were there more merge conflicts with the other "not cooking" branches?

Thanks, -Stolee

Derrick Stolee <stolee@gmail.com> writes:

I think ds/multi-pack-index didn't have any difficult textual merge conflicts. I ran out of time making 'pu' build with new topics and listed the ones that were left behind. As Elijah already identified, there was a bad apple not listed above that was in 'pu' that made the tests fail, so the above is not even a complete list of "bad" topics. It was merely an "I have them but 'pu' doesn't include them" list.

Thanks.