Re: [Adonthell-devel] Re: Waste's Edge runnable with worldtest, but ...

[Kai Sterker]

On Sat, Mar 20, 2010 at 2:28 AM, Rian Shelley <address@hidden> wrote:

> Valgrind actually works quite well with C++. It uses the debugging
> information, so it will even point out the source file line number
> information.

My problem with valgrind is that even though v3.5.0 finally support
OSX without external patches, it only runs on OSX 10.5. I have both
10.4 and 10.6 ... so again I had to resort to patching. So not sure if
valgrind runs as well as it should.

So yesterday, I turned to a few other tools. The OSX built-in malloc debuggers
(http://developer.apple.com/mac/library/releasenotes/DeveloperTools/RN-MallocOptions/index.html)
and electric fence, which can be coerced to work on OSX as well.
(http://lists.apple.com/archives/xcode-users/2005/Oct/msg00792.html).

No real luck, however. Running with libgmalloc, with word alignment
did not trigger the issue. So it must be a buffer overrun of only a
byte or two. However, when I turned on byte alignment, I did not get
further than paths.cc:101 on the 32bit/OSX 10.4 system. Either a
problem with libgmalloc or the underlying libstdc++..

OTOH, the same test on 64bit/OSX 10.6 did not yield anything. Neither
did efence turn up with anything. Makes me wonder if the problem is
caused by a 3rd party lib, since I don't think versions of those are
identical across both systems.

I came across a blog post with some helpful hints
(http://intquestion.wordpress.com/2008/09/14/how-to-debug-a-corrupted-stack/),
maybe following that through leads to something.

In the meantime, maybe one of you guys can try out what happens when
running worldtest with the wastesedge data. Would be interesting to
see if only my version is affected, or if other systems have the same
problem. Of course, if there is truly a memory corruption, different
systems might be affected in different ways ...

Kai

P.S.: Glad you're still around, Rian :-).