Re: immediate strings

[Eli Zaretskii <eliz@gnu.org>]

> Date: Sat, 26 Nov 2011 00:05:40 -0800
> From: Paul Eggert <eggert@cs.ucla.edu>
> Cc: emacs-devel@gnu.org
> 
> On 11/25/2011 06:15 PM, Dmitry Antipov wrote:
> > I think it's worth playing with such a
> > little complication.
> 
> This looks like a good thing to pursue.

If by "pursue" you mean creating a separate branch and letting people
play with it and report results, then I agree.

But if that means applying this to the trunk, based just on the
preliminary statistics reported by Dmitry, presumably for his usage
pattern and locale, then I think the gains are not substantiated
enough to justify the change.

FWIW, I'd like to see a much more detailed statistics, both for memory
usage and for speed of frequent operations, before I could make up my
mind on whether this kind of micro-optimization is justified.  YMMV.

See also the issue raised by Andreas regarding the effect of packing
`size' and `size_byte' as bit fields.  These members of Lisp strings
are accessed very frequently, including in the innermost code of the
display engine, and also elsewhere where performance counts.  IMO, we
should consider the results very carefully before deciding on such a
change.