Re: leaky cherry-pick

From: Nguyen Thai Ngoc Duy <pclouds@gmail.com>
Date: Jan 10, 2012, 01:38 UTC
Message-ID: <CACsJy8DpOY6UUVBjPK4k5BbvbfAckc+N5hoDTwRq5c7koZ=OgQ@mail.gmail.com>
In-Reply-To: <20120109223737.GA1589@padd.com>

On Tue, Jan 10, 2012 at 5:37 AM, Pete Wyckoff <pw@padd.com> wrote:

Show 14 quoted lines

> I've got a big tree, and rebased 200 commits today using
> cherry-pick.  It ran out of memory.  Both 1.7.4 and recent
> master (eac2d83 (Git 1.7.9-rc0, 2012-01-06)) behave similarly.
>
> Here's a valgrind dump for rebasing just 9 commits, if someone is
> interested to track this down.  This was invoked as:
>
>    valgrind --log-file=/tmp/vg.out --leak-check=full \
>    /home/pw/src/git/git cherry-pick 8d535b6^..2cf53a0
>
> I can re-test if you like, or provide more detail if this seems
> unusual.
>
> Scroll to the end to see the biggest leaks.

Index entries are leaked since probably forever because we allocated entries in bulk and could not release them individually, until debed2a (read-cache.c: allocate index entries individually). Putting discard_index() to read_and_refresh_cache() might help, I think.

-- 
Duy