bug#14786: 24.3.50; `field-end' is now very slow

[Drew Adams <drew.adams@oracle.com>]

> > The result appears after a few seconds (I'm guessing about 4 sec
> > in the directory I used).
> >
> > Do the same thing in previous Emacs versions (including as recent
> > as 24.3, at least), and the result appears immediately.
> 
> Not exactly immediately, but approximately 4 to 5 times faster.

OK.  I was describing the user experience.  Less than a second is
quick enough for a large Dired buffer.  Four seconds it not.

> The problem is that dired-details floods the Dired buffer with
> overlays.  When there are so many overlays, searching for an end of
> a non-existent property takes a lot of time, because Emacs needs to
> examine every overlay in the buffer.

OK.

> What happened between Emacs 24.3 and the current trunk is that the
> call GET_OVERLAYS_AT (XINT (position), overlay_vec, noverlays, NULL, 0);
> in get_char_property_and_overlay got roughly 4 to 5 times slower.  I
> don't know why this slowdown happened, but it just moved the time
> needed by field-end from below the annoyance threshold to well above
> it.

I agree with your characterization in the last phrase.

> If someone can find out why GET_OVERLAYS_AT is now slower, we could
> see if that could be fixed.  But it's quite possible that the change
> which caused that fixed some bug that we don't want to re-introduce.
> In any case, I think having so many overlays in a buffer is asking for
> trouble.

Agreed, in general.  Emacs does make heavy use of overlays here and
there, and I don't see that possibility disappearing.  I hope this bug
(performance regression) can be fixed.  It is not particular to
dired-details.el, though that is one place that it can easily be
noticed.

For my info: When such a code change is made, is there no log of the
reason behind the change?  If it was to fix a bug, for instance,
wouldn't that bug # be noted?