Re: Indentation and gc

[Eli Zaretskii <eliz@gnu.org>]

> From: Ihor Radchenko <yantar92@posteo.net>
> Cc: arne_bab@web.de, spacibba@aol.com, emacs-devel@gnu.org
> Date: Sun, 12 Mar 2023 14:20:33 +0000
> 
> > I'm talking about basis for the 0.7% figure.
> 
> I used 0.7%*RAM because total RAM is the only reasonable metrics. What
> else can we use to avoid memory over-consumption on low-end machines?

It could be the amount of VM on low-memory machines, but something
else on high-memory machines.  No one said it has to be derived from
the total VM on both systems with 2 GiB and systems with 128 GiB.

> Of course, I used implicit assumption that memory usage will scale with
> gc-cons-threshold linearly. IMHO, it is a safe assumption - the real
> memory usage increase is slower than linear. For example, see my Emacs
> loading data for different threshold values:

We are talking about changing the threshold for the session itself,
not just for the initial load.  So the statistics of the initial load
is not what is needed.

> The 0.7% is to ensure safe 800kb lower bound on low-end computers.

I don't see why it would be the value we need to adhere to.  That it's
the current default doesn't make it sacred, and using it as basis for
relative figures such as 0.7% has no real basis.

> > Anyway, how about if you try running with the threshold you think we
> > should adopt, and report back after a month or so, say?
> 
> I am using 250Mb threshold for the last 3 years or so.
> GCs are sometimes noticeable, but not annoying:
> 
> - gc-elapsed 297 sec / gcs-done 290 -> ~1 sec per GC

IMO, 1 sec per GC is pretty annoying.  It's around 0.165 sec in my
production session, and it still quite noticeable.  I'd be interested
to hear what others think.

> - Emacs uptime 2 days 5 hours 21 minutes -> 1 GC per 10 minutes

I'm running with gc-cons-threshold of 1.8 MiB, and get about 1 GC per
minute.  (Actually, it's about twice that, because Emacs stays up at
night, but does nothing.)

> - memory-limit 6,518,516, stable

??? That's 6 GiB.  Didn't you say your memory footprint stabilizes at
1 GiB?

Anyway, we need such statistics from many people and many different
values of the threshold, and then we will be in a position to decide
on better default values, and perhaps also on some more dynamic
adjustments to it.  We are not ready for that yet.