Re: [Emacs-diffs] master 05a5a94: Correct calculation of CC Mode's font-lock region.

[Alan Mackenzie <acm@muc.de>]

Hello, Stefan.

On Mon, Mar 30, 2015 at 05:08:38PM -0400, Stefan Monnier wrote:
> >     * cc-mode.el (c-fl-decl-start): Renamed from c-set-fl-decl-start.

> Please use the present tense, even for "rename" ;-)

Do you mean I should change the commit message on that commit?  Can I
actually do that?

> >     (c-extend-after-change-region): Explicitly apply 'fontified properties
> >     to the extended bits of the font lock region.

> Sounds like this is working around a problem elsewhere, so better add
> a comment explaining what's going on.

jit-lock-force-redisplay (that function which is set to run on the
expiry of a 0 second timer, when jit-lock-fontify-now suspects that some
area which has already been rendered requires to be re-rendered) doesn't
seem to work.  At least, I couldn't get it to work.

Its source is this:

1. (defun jit-lock-force-redisplay (start end)
2.   "Force the display engine to re-render START's buffer from START to END.
3. This applies to the buffer associated with marker START."
4.   (when (marker-buffer start)
5.     (with-current-buffer (marker-buffer start)
6.       (with-buffer-prepared-for-jit-lock
7.        (when (> end (point-max))
8.          (setq end (point-max) start (min start end)))
9.        (when (< start (point-min))
10.         (setq start (point-min) end (max start end)))
11.       ;; Don't cause refontification (it's already been done), but just do
12.       ;; some random buffer change, so as to force redisplay.
13.       (put-text-property start end 'fontified t)))))

The exact locations `start' and `end' used in L13 appear to be a red
herring - when I tried (artifically) subtracting 100 from start, to
ensure (start end) covered the location where re-rendering was needed,
the re-rendering still failed to take place.

However, when I tried changing L13 to set 'fontified to nil, the
re-rendering happened even when the location was far before `start'.

I don't think making this non-change change to the buffer is sufficient to
trigger a redisplay, though I was unable to get deeply enough into
xdisp.c to verify this.

*************************************************************************

Assuming that jit-lock-force-redisplay is not going to work, it follows
that any buffer bits to be fontified which come earlier than
after-change-functions's `beg' need to have their 'fontified text
property set to nil, and this needs to happen at after-change time
(during redisplay is too late).  This necessity even applies to
extending change regions to whole lines.

The only functions which can do this extending are those on
jit-lock-after-change-extend-region-functions, i.e. the single function
font-lock-extend-jit-lock-region-after-change.  By the way, I am of the
opinion that the extending of regions to whole lines in
f-l-extend-j-l-r-after-change is not currently an optimisation, but a
crucial part of the functioning of jit-lock.  Without it, after a buffer
change in the middle of a line, the beginning of that line doesn't get
rendered (at least, not until context fontification kicks in).  This was
what Daniel C. saw and complained about.  You recently suggested doing
away with f-l-e-j-l-r-a-c.  I don't think you can do this without
putting something similar, running at after-change time, in its place.

f-l-extend-j-l-r-after-change has a recently fixed bug, where the
extended region is only returned to the caller in certain circumstances.
Those circumstances no longer hold in CC Mode.  To allow CC Mode to
continue to work in already released Emacs versions, it is necessary to
set the 'fontified text-property to nil from
c-extend-after-change-region.

All this would make a rather long comment in lisp/ChangeLog.  ;-(

>         Stefan

-- 
Alan Mackenzie (Nuremberg, Germany).