bug#34525: replace-regexp missing some matches

[Stefan Monnier]

> After this buffer change, adjust_intervals_for_insertion gets called.
> This adds 6 onto the ->position field of each interval "adjusting all of
> its ancestors by adding LENGTH to them", according to the comment at the
> head of adjust_intervals_for_insertion.
>
> Note this only adjusts the ancestors of that interval early in the .h
> file, not all intervals in the tree.

Maybe the problem comes from here.  The lazy adjustment of ->position
may be too lazy here.  This is a fairly delicate/brittle part of intervals.c

> I'm reasonably sure this is what's happening:
> adjust_intervals_for_insertion is failing to adjust the cached intervals
> in gl_state.  It's a nasty cache invalidation problem.

As mentioned earlier, I think gl_state should be sufficiently transient
that this scenario can't happen.

Elsewhere you send a sample patch and Eli asks "Does that slow down the
search in any significant way?".  The answer should be "yes!" because
this patch pretty much defeats the purpose of the gl_state cache,
recomputing the current interval (O(logN)) every time we move from one
character to the next.

Maybe another way to catch the problem is: at the end of
upate_syntax_table, call `interval_of (charpos, object)` and verify that
the returned interval is the same as the one we had selected, plus
verify that the call to `interval_of` has not modified the value
of ->position.


        Stefan