21.5 crash in Gnus
daniel at rimspace.net
Wed Mar 15 18:55:27 EST 2006
Marcus Crestani <crestani at informatik.uni-tuebingen.de> writes:
>>>>>>"DP" == Daniel Pittman <daniel at rimspace.net> writes:
> DP> NEWGC does run very well for me, and the asynchronous finalization
> DP> code means that my previous fix to the image GC code is no longer
> DP> need, which is lovely.
> Do you mean this?
> Subject: [PATCH] [RECOMMEND 21.4] Stopping timeouts during GC causes lossage
> on all GUI platforms (X11, GTK and Win32)
Yes. The timeout is related to animated image glyphs, and the solution
was to introduce an asynchronous finalization for the timeout object.
With your code doing that generically the problem (corruption or an
assert due to a synchronous Lisp object free during a GC run) has gone
away, so my specific fix is no longer necessary.
> DP> I have run into one issue, thus far, which is that a long running
> DP> XEmacs worked slowly up to 608MB of allocated memory before it
> DP> stopped being able to fork and I restarted it.
> DP> On that topic, what is the best way to track down a memory leak
> DP> with the NEWGC code? Is there anything specific that I can look
> DP> at to help work out what the leak is and why it is happening?
> The only thing is to keep an eye on show-object-memory-usage-stats and
> show-mc-alloc-memory-usage and see which objects escape. It might
> also be a fragmentation problem of the allocator.
No problem. I will look to getting a core, memory stats, etc for you.
It looks like reproducing the issue isn't to be a problem; XEmacs has
gained around 96MB VSS over the last day or so, so I should hit the
bloaty problem later.
> Valgrind might be useful, but there are issues with NEWGC's write
> barrier: The virtual dirty bit write barrier provokes SEGFAULTs on
> purpose to keep track of object modifications. Valgrind reports all
> of these SEGFAULTs, which makes it is practically useless.
I might take a look at that, especially if someone else provides the
solution to suppressing the reports; I have never worked with valgrind
I can't promise too much work on debugging this, though, as my time is
currently very over-committed to work. :?
Digital Infrastructure Solutions -- making IT simple, stable and secure
Phone: 0401 155 707 email: contact at digital-infrastructure.com.au
More information about the XEmacs-Beta