Re: Should native compilation be enabled by default?

[Eli Zaretskii]

> From: Richard Stallman <rms@gnu.org>
> Cc: emacs-devel@gnu.org
> Date: Mon, 04 Mar 2024 22:42:32 -0500
> 
>   > > Enabling native compilation has disadvantages as well as advantages.
> 
>   > What are the disadvantages?
> 
> If Emacs compilation by default causes native compilation, it will be
> slower.

This is true, but the slowdown is IME insignificant on a reasonably
modern system.  And if Emacs is installed from a distro (which AFAIU
is what most users do nowadays), the slowdown doesn't exist at all,
because the *.eln files come prebuilt with the binaries.

> It will also be more fragile.  Byte-compilation is a self-contained
> Emacs feature, and aside from occasional bugs that affect specific
> code, it never breaks.  On general principles we can see that native
> compilation is likely to go wrong because something has disappeared,
> or because of bugs in other programs that you wouldnt otherwise ever
> use.

That was a valid consideration when native compilation was first added
to Emacs 28.  Which is why we kept it off by default for 2 major
releases.  Now we have enough experience to tell that the problems,
such as they were, are all but solved, and native compilation is
nowadays no more fragile than byte compilation, what with all the
improvements and optimizations introduced lately into the byte
compiler.

> Native compilation is useful mainly for power users who want
> to run Lisp programs that normally are too slow.

I disagree, based on my experience.  I see a significant speedup when
using Rmail for just reading email, something that I cannot
characterize as "power use" of Emacs.  Even restarting a large Emacs
session by restoring from a desktop file is noticeably faster.  I
don't think having a lot of buffers in a long-running session
qualifies me as a "power user" just because of this factor.

And there are other significant speed advantages.

> The people who participate in emacs-devel tend to be power users.
> What they think is not a good guide to what is useful for most users.
> To find that out, we should poll the users.

Which is why we didn't decide on the default before collecting a lot
of experience and processing the bug reports we had about the related
issues.  Having solved them, I think we can be quite sure users aren't
unhappy with native compilation, quite the contrary.