Re: Memory again

[Ted Zlatanov <tzz@lifelogs.com>]

On Mon, 23 Jan 2012 16:49:37 +0000 Nix <nix@esperi.org.uk> wrote: 

N> On 19 Dec 2011, Stefan Monnier outgrape:
>>>> I agree that we're probably going to see better overall results by
>>>> improving general memory use than by trying to attack
>>>> fragmentation problems.
>>> Among this list's subscribers, Nix <nix@[hidden]> is constantly
>>> reporting an enormous memory usage caused by Gnus.
>> 
>> I think we still have a leak there somewhere.

N> Confirmed. With Dmitry's immediate-string patch (which works perfectly
N> otherwise):

N>    RSS    VSZ   USER     STIME
N> 720696 1008752  gnus     Jan07
N> 242740 525592   hacking  2011

N> Despite running for less than half the time, the Gnus Emacs is twice the
N> size. It bloats more slowly than without Dmitry's patch, but it still
N> bloats.

N> I suspect that the only way I'll be able to provide a reproducible
N> recipe for this is to instrument Emacs to record all allocations and GCs
N> and when they happen, so that someone else can play this back to observe
N> WTF is going on. Doing this compactly enough to run over weeks without
N> serving as a DoS-attack on my own system may be... interesting. I
N> haven't done any work toward this yet though.

Some valgrind recipes were posted that may help.  Can you try them?

I could not put together a test case to cause memory bloat with GnuTLS
usage alone.  Do you want to try that?  If such a test case can be shown
to cause bloat without Gnus, we'll know where to look.

Ted