Re: [PATCH 0/11] Misc. pull/merge/am improvements

From: Junio C Hamano <junkio@cox.net>
Date: Dec 28, 2006, 11:04 UTC
Message-ID: <7vtzzguvrp.fsf@assigned-by-dhcp.cox.net>
In-Reply-To: <20061228084245.GA18150@spearce.org>

Shawn Pearce <spearce@spearce.org> writes:

Show 10 quoted lines

> Junio C Hamano <junkio@cox.net> wrote:
> ...
>> I am very tempted to have sliding window mmap() if it helps
>> people on cygwin, for example.
>
> Especially now that NO_MMAP is the default on that platform.
> At this point it may be ready to graduate to next to try and get a
> wider audience.  Since fixing that segfault in pack-objects I can't
> break it.  Of course I couldn't break it before you found that error,
> so take my words with a grain of salt... ;-)

Well, I have a bad news for you.

"ng refs/heads/master n/a (unpacker error)" is back. I cannot push things out.

But a bad news is that the problem does not decompose so easily; git-push has too many players involved. I _think_ I have the list of positive and negative objects fed to rev-list --objects --thin --stdin, whose output is in turn fed to pack-objects, but manually running these steps in isolation seems to produce an Ok result that index-pack --stdin --fix-thin accepts happily.

While I was looking at the problem, I noticed something a bit easier to reproduce and should be lot easier to diagnose. At http://userweb.kernel.org/~junio/broken.tar, I have a tarball of git.git repository.

When you extract it, you will have a directory "broken/" with a single directory ".git/" in it. The repository passes "fsck-objects --full" from master just fine.

However, if you try to "git repack -a -d" it using "next" plus "sp/mmap" with your latest fix ("pu" also has it), you will see:

        Generating pack...
        Done counting 32054 objects.
        fatal: internal error: pack revindex corrupt

In that sample repository, I have pack.window, and repack.deltabaseoffset. These settings do not seem to affect the breakage.