bug#34525: replace-regexp missing some matches

[Alan Mackenzie <acm@muc.de>]

Hello, Stefan.

On Wed, Feb 27, 2019 at 10:40:19 -0500, Stefan Monnier wrote:
> >> Here, in seeking the next interval, we go down a chain of `left's.  We
> >> do not set the ->position field of these intervals, except for the last
> >> one, which we return.
> >> So the returned interval doesn't satisfy the condition that all its
> >> parents have their ->position's set correctly.
> [...]
> > I've done this, and it appears to have fixed the bug.  :-)

> AFAICT the only place where we need the parents to have
> a valid ->position is in update_interval.  So maybe another fix is to
> change update_interval so it computes the parent's ->position rather
> than rely on it having the right value.

I'll think about this.

> I personally don't have a preference and I'm not sure which option would
> be better performancewise.

I've done some speed testing with my function M-: (time-scroll), which
scrolls through a buffer a screenful at a time, redisplaying each place
it stops.

On xdisp.c, there was no detectable difference between versions with the
bug fix and without.

On a largish C++ file with lots of template delimiters, the corrected
version was about 4% slower on unoptimised builds.  Between comparable
optimised builds, the differences were not detectable.

> If we opt (like your patch does) to have the invariant that
> the ->position of parents is kept up-to-date, then maybe we should
> change find_interval to guarantee this (which would basically be
> a matter of moving the corresponding code from update_syntax_table where
> we update the parents's ->position after calling find_interval) ?

This would be an excellent idea, something I was going to suggest
myself.  :-)

>         Stefan

-- 
Alan Mackenzie (Nuremberg, Germany).