bug#34525: replace-regexp missing some matches

[Eli Zaretskii <eliz@gnu.org>]

> Date: Thu, 28 Feb 2019 10:50:25 +0000
> Cc: daniel.lopez999@gmail.com, 34525@debbugs.gnu.org
> From: Alan Mackenzie <acm@muc.de>
> 
> (i) Calculate ->position's in previous_interval and next_interval, as my
>   tentative patch already does.
> (ii) Calculate the ->position's in update_interval, on moving to
>   parents.
> (iii) Do away with update_interval, replacing it in syntax.c with
>   previous/next_interval in while loops.
> 
> At the moment, only (i) has been tried.
> 
> Speed-wise, it seems not to make any difference in an optimised GNU
> build, though it did appear to be significantly (~4%) slower on an
> unoptimised build which scrolls through a C++ file with lots of
> templates.  I don't think it's worth the effort to make a systematic
> speed comparison between the alternatives.
> 
> (iv) Additionally, there is a cleanup wanted, where setting ->position
>   in the chain of parents should be moved from update_syntax_table to
>   find_interval.
> 
> In (i), the convention for ->position would be that it is valid for the
> target interval together with all its parents.  In (ii) and (iii), it
> would only be valid in the final target intervals found by navigation.
> I think this should be explicitly stated in a comment in struct
> interval.
> 
> So, where do we go from here?  If it were up to me, I would probably
> chose (i), simply because it's already been done, but I've no strong
> feelings over it.

I prefer not to do (i) because it has much wider implications than
needed.  Either (ii) or (iii) are okay with me.  The former seems to
be simpler, so I tend to favor it slightly.