From: Jeff King <peff@peff.net>
Date: Tue, 27 Jan 2009 04:48:38 GMT
Subject: Re: Valgrind updates
Message-ID: <20090127044838.GA735@coredump.intra.peff.net>
In-Reply-To: <alpine.LFD.2.00.0901261934450.3123@localhost.localdomain>

On Mon, Jan 26, 2009 at 07:38:56PM -0800, Linus Torvalds wrote:

> > ==valgrind== Syscall param write(buf) points to uninitialised byte(s)
> > ==valgrind==    at 0x5609E40: __write_nocancel (in /lib/libpthread-2.6.1.so)
> > ==valgrind==    by 0x4D0380: xwrite (wrapper.c:129)
> > ==valgrind==    by 0x4D046E: write_in_full (wrapper.c:159)
> > ==valgrind==    by 0x4C0697: write_buffer (sha1_file.c:2275)
> > ==valgrind==    by 0x4C0B1C: write_loose_object (sha1_file.c:2387)
> 
> Looks entirely bogus.
> 
> I suspect that valgrind for some reason doesn't see the writes made by 
> zlib as being initialization, possibly due to some incorrect valgrind 
> annotations on deflate().  We've just totally initialized that whole 
> buffer with deflate().
> 
> It definitely does not look like a git bug, but a valgrind run issue.

Yes, this is exactly the issue I ran into when doing the valgrind stuff
a few months ago. I spent several hours looking carefully at the code
and came to the same conclusion. Anything zlib touches needs to be
manually suppressed for uninitialized writes (which I _thought_ was
covered in the suppressions I sent out originally, but maybe they need
to be tweaked for Dscho's system).

-Peff