Re: Memory again

[Eli Zaretskii <eliz@gnu.org>]

> From: Óscar Fuentes <ofv@wanadoo.es>
> Date: Thu, 15 Dec 2011 05:50:17 +0100
> 
> > I don't think there's much fragmented memory in real-life use (barring
> > bugs).  The test cases that exhibited a lot of fragmentation are all
> > toy examples that don't really happen.
> 
> The case I described was real, although seldom happens here.

OK, let's agree these cases are rare.

> I can think of typical scenarios where buffer size grows to large
> sizes: ERC sessions, files monitored with
> auto-revert(-tail)-mode... Even cases where the user makes a mistake
> that creates huge buffers (visiting a GNUs group with tens of
> thousands on unread messages, for instance.)

It has never been proven that just growing a buffer in small chunks
has this effect.  Stefan thinks the reason is intervals.  I'm not sure
he is right (but I'm not an expert on this).  I want to play with the
examples posted here to trace what they do to memory allocation, but I
need to find time to do it.  If someone has time and motivation, it
would help to trace through the various memory-allocation routines in
those cases and report the results.

FWIW, I always have at least one buffer in auto-revert-mode (it
watches my .bzr.log file), and the memory footprint is still quite
moderate, see my previous message.

> Taking a selfish stance, I don't care about this problem because it is
> not harming me, but your response and Stefan's looks a lot like hand
> waving.

I don't understand where you see hand-waving.  Reports about excessive
memory consumption in real-life usage, like in Emacs running normally
doing the usual stuff, are taken very seriously.  They are hard to
track down, especially if they happen on systems whose users cannot
provide hard facts because they lack the ability to dig into the code
with a debugger.  This could explain slow progress, or no progress at
all, in solving these problems.  But no one hand-waves them under the
carpet; saying so is simply unfair.