[AC21.5] Fix a nasty performange regression in syntax cache

[robert delius royar]

Tue, 23 Aug 2011 (04:38 +0900 UTC) Stephen J. Turnbull wrote:

> APPROVE COMMIT 21.5
>
> Revision 5543:fbe90e6f7a43 "Initialize start and end properly (to new
> markers, not Qnil)" introduced a bad performance regression, resulting
> in map_extents (a fairly expensive function) being called ~15,000,000
> times instead of ~100,000 times while fontifying python-mode.el.

Steven,

Thank you!

I noticed that after the chnage you mentioned, loading the 150K sendmail 
deliver.c file took over 10 minutes. I kept meaning to report the 
problem, but could not get around to it. An 80K elisp file I keep loaded 
also took a long time. For both files, the slowdown appeared to be in 
the "fontifying buffer" process.

robert

-- 
   Wherever they burn books they will also, in the end, burn human beings.
                                                           -Heinrich Heine