Re: Overlays as an AA-tree

[Eli Zaretskii]

> From: Andreas Politz <address@hidden>
> Cc: address@hidden
> Date: Mon, 13 Feb 2017 10:04:13 +0100
> 
> Eli Zaretskii <address@hidden> writes:
> 
> >> From: Andreas Politz <address@hidden>
> > Thank you for working on this important issue.
> 
> I enjoyed it.

I'm sure we will, too.

> > From my POV, a very important performance issue is that of redisplay.
> 
> Do you expect redisplay performance to improve or is it your aim that it
> does not degrade ?

It should certainly not be worse, otherwise the design and
implementation would need to be improved, IMO.  Display is a very
important client of overlays, so it shouldn't suffer from refactoring.

I would be happy to know that performance gets better, of course.  The
recentering of overlays, which is not a cheap operation, should no
longer be needed, for example.

> BTW how do you figure out which functions need attention,
> performance-wise.

If you mean overlay-specific functions, I can name the most important
ones for you.  I know which ones they are because I know what the
display engine calls in its inner loops.

> Is there any experience using valgrind/callgrind, because I tried it
> and the results seem bogus to me, but I may be mistaken.

Using valgrind/callgrind with Emacs needs some special techniques,
some of them are described in etc/DEBUG.  I think the most important
trick is to measure temacs, not a dumped emacs binary.