bug#15156: 24.3; !MEM FULL!

[Sebastien Vauban]

Eli Zaretskii wrote:
>> From: "Sebastien Vauban" <sva-news@mygooglest.com>
>> Cc: Sebastien Vauban <sva-news@mygooglest.com>,  15156@debbugs.gnu.org
>> Date: Fri, 06 Sep 2013 10:55:55 +0200
>> 
>> > All I see in the screencast is that the memory footprint grows, then
>> > shrinks back.  Assuming you have something going on in Emacs that can
>> > explain several hundreds of MBs of memory consumption, that's actually
>> > normal: Emacs uses up memory when it needs it, then releases it when
>> > it no longer does.  So maybe there's no problem here at all.
>> 
>> You take the precaution of "assuming I have something explaining several
>> hundreds of MBs of memory consumption", and that's where the problem lies: I
>> don't have anything that could explain that.
>
> You are using helm, aren't you?

Yes, I do -- can't live without it anymore (to find a file, just type part of
its name, instead of typing every directory from the root). [1]

> That can explain anything at all, as it runs subprocesses on every
> keystroke, AFAIR.

Yes, that's what I remember as well. Though, I wasn't using Helm at the moment
the situation become visible to me; but the full memory could have been
triggered a little while before, and only apparent when I was typing my commit
message.

> Anyway, putting a break at xmalloc conditioned by some large
> allocation size might show who is requesting this much memory.  I
> don't see how this can be investigated otherwise without some
> debugging.

If you have a recipe which does not constrain too much my daily usage of
Emacs, I'm willing to apply it.

>> What's weird as well is that the memory shrinks back to a lower level -- even
>> if still quite important. No apparent reason for that.
>
> GC is normally the reason: Emacs relinquishes memory it doesn't need
> anymore.

OK, clear.

>> And, anyway, I still can't use Emacs as you saw: I'm forced to kill
>> Emacs from the Task Manager.
>
> Why can't you exit Emacs "normally"?  Memory full condition does not
> prevent that.  What happens if you try exiting?

The problem is that Emacs did not respond at all my attempts to get back
control of it. In the video, I typed C-g numerous times in the last minute
(certainly > 10 times). Never got a working state back. My keys simply where
ignored (or not processed yet). I sometimes could type a couple of letters,
but then Emacs went back in his state where it does not listen to my keys
anymore. It really stays completely deaf.

It'd be interesting to get a "key logger" with on-screen display, but I still
did not find one convenient for Windows. That way, you'd see that it simply
stayed unusable.

Another proof of that is that the CPU usage never drops anymore under ~30%
(which seems to be equal to infloop on my i7-3517U CPU, 2 cores, 4 logical
processors).

[1] I'd be interested by knowing what something similar you use...