Re: Troubles when directory is replaced by symlink

From: Jeff King <peff@peff.net>
Date: Jun 11, 2009, 11:48 UTC
Message-ID: <20090611114846.GC4409@coredump.intra.peff.net>
In-Reply-To: <c6c947f60906090118n78d3c40fq11d1390f8776c2c0@mail.gmail.com>

[Kjetil: I bisected this bug to your 92604b4. Details below.]

On Tue, Jun 09, 2009 at 12:18:16PM +0400, Alexander Gladysh wrote:

> Any help here? This bug is pretty annoying (especially that it
> requires double hard resets).
> 
> Should I provide any extra information?

Actually, I think the problem happens earlier than you realize; I see problems when switching back to master:

-- >8 -- # make a repo mkdir repo && cd repo && git init

# content in alpha mkdir alpha && echo content >alpha/file && git add . && git commit -m one # and duplicate content inside beta mkdir beta && cp -R alpha beta && git add . && git commit -m two

# now replace the duplicate with a symlink git checkout -b branch rm -rf beta/alpha && git add -u && git commit -m deleted ln -s ../alpha beta/alpha && git add . && git commit -m symlink

# now checkout master again; alpha/file will be missing, even # though it wasn't touched at all git checkout master git status -- 8< --

We handled this correctly in v1.6.1, so I was able to bisect it down to:

    commit 92604b466344b2157efc42ef3521dac22d7906a2
    Author: Kjetil Barvik <barvik@broadpark.no>
    Date:   Sun Jan 18 16:14:50 2009 +0100

        lstat_cache(): more cache effective symlink/directory detection

        Make the cache functionality more effective.  Previously when
        A/B/C/D was in the cache and A/B/C/E/file.c was called for,
        there was no match at all from the cache.  Now we use the fact
        that the paths "A", "A/B" and "A/B/C" are already tested, and we
        only need to do an lstat() call on "A/B/C/E".

I don't have time to look more closely now, but I'm guessing we are treating the symlink to the directory like the directory itself in a place that we shouldn't.

-Peff