Re: Emacs 24.5 second release candidate

[Alan Mackenzie <acm@muc.de>]

Hello, Eli.

On Thu, Apr 02, 2015 at 08:22:50PM +0300, Eli Zaretskii wrote:
> > Date: Thu, 2 Apr 2015 16:12:56 +0000
> > From: Alan Mackenzie <acm@muc.de>
> > Cc: rgm@gnu.org, nicolas@petton.fr, monnier@iro.umontreal.ca,
> > 	emacs-devel@gnu.org

> > > Do you have an estimation of how much time you will need?

> > Perhaps this evening, perhaps tomorrow morning.  It's one of these nasty
> > little bugs which looks totally wierd and random until one stumbles upon
> > the cause, after which it takes 2 minutes to fix.

> > > Also, was this bug caused by a recent commit, and if so, which one?

> > No, I don't think it was.  I think it was "unmasked" by a recent bug
> > "fix", d69e9f1c24f37f36af33b0468c5a4e100dbb09b6 from 2015-02-01.  Oleh
> > Krehel, the OP of the pertinent bug #20245, says that with that change
> > reversed, the bug symptoms don't manifest themselves.  I haven't checked
> > this yet, but it sounds plausible.  That fix is actually incomplete,
> > requiring Monday/yesterday's fix to make it complete.

> > I'm beginning to come round to the view that perhaps it might be best
> > just to take d69e9f1c24f37f36af33b0468c5a4e100dbb09b6 out of Emacs 24.5,
> > as Oleh Krehel has explicitly suggested.

> Sounds plausible.  Please share your opinions after you study the
> issue.

I now understand precisely what the problem is.  Sorry it's taken me a
little longer than "perhaps tomorrow morning".

The problem is in `c-fl-decl-start', a routine called (indirectly) from
every `jit-lock-fontify-now'.  It attempts to check whether `start' is
within a declaration.

It does this by calling `c-beginning-of-decl-1' with a 500 byte search
limit.  c-b-o-d-1 wrongly returns the same value for "hit search limit"
as "found beginning of declaration".

So CC Mode's fontification is directed to start 500 bytes back, in the
middle of a #include directive.  This is not immediately apparent, only
becoming visible at the next redisplay.

This is indeed an old bug, unmasked by the d69e.... change rather than
caused by it.

I don't want to try to fix this in a few hours; a few days would be more
realistic.

-- 
Alan Mackenzie (Nuremberg, Germany).