Re: Another bug in CC Mode's state cache.

Re: Another bug in CC Mode's state cache.

[Leo]

On 2012-04-21 04:04 +0800, Alan Mackenzie wrote:
> The solution is to use the cache's memory of c rather than the paren in
> the buffer to bound the search.  This works.  I have tested it on 100,000
> random points in xdisp.c and 20,000 points in lisp.h.  The results with
> the new bugfix are identical to those without.

I'd love to see this fixed and this seems to indicate it is reasonably
safe.

Leo

Another bug in CC Mode's state cache.

[Alan Mackenzie]

Hi, Yidong.

I've found and corrected another CC Mode state cache bug.  It occurred in
situations like this:

         {      }        (         {  }
	 a      b        c         e  f

Insert a closing paren after c:


         {      }        (    )     {  }
	 a      b        c    d     e  f

Previously, the backward search from e for a brace pair was bounded by
the paren at c.  After the insertion of d, the the paren pair c-d is
invisible to backward searching, hence the brace pair a-b is spuriously
found, leading to a cache corruption.

The solution is to use the cache's memory of c rather than the paren in
the buffer to bound the search.  This works.  I have tested it on 100,000
random points in xdisp.c and 20,000 points in lisp.h.  The results with
the new bugfix are identical to those without.

This clearly isn't a "regression".  Would it nevertheless be OK to
install this fix in the Emacs 24 branch?

-- 
Alan Mackenzie (Nuremberg, Germany).

Re: Another bug in CC Mode's state cache.

[Stefan Monnier]

> This clearly isn't a "regression".  Would it nevertheless be OK to
> install this fix in the Emacs 24 branch?

Since it's not a regression, I'd rather not put it on the
`emacs-24' branch.


        Stefan "who found your explanation scary"

Re: Another bug in CC Mode's state cache.

[Alan Mackenzie]

Hi, Stefan,

On Fri, Apr 20, 2012 at 10:30:14PM -0400, Stefan Monnier wrote:
> > This clearly isn't a "regression".  Would it nevertheless be OK to
> > install this fix in the Emacs 24 branch?

> Since it's not a regression, I'd rather not put it on the
> `emacs-24' branch.

OK.  I'll put it into the trunk.

>         Stefan "who found your explanation scary"

Hmm.  The code for that cache _is_ somewhat scary.  I thought my
explanation was quite clear.

Funnily enough, the bug only shows itself very rarely.  I could only
reproduce it in the CC Mode branch I was working on.  A corrupt state
cache makes its presence felt very quickly.

-- 
Alan Mackenzie (Nuremberg, Germany).

Re: Another bug in CC Mode's state cache.

[Alan Mackenzie]

Hi, Leo.

On Fri, Apr 20, 2012 at 02:19:37PM +0800, Leo wrote:
> On 2012-04-21 04:04 +0800, Alan Mackenzie wrote:
> > The solution is to use the cache's memory of c rather than the paren in
> > the buffer to bound the search.  This works.  I have tested it on 100,000
> > random points in xdisp.c and 20,000 points in lisp.h.  The results with
> > the new bugfix are identical to those without.

> I'd love to see this fixed and this seems to indicate it is reasonably
> safe.

I doubt you'd see much difference.  Like I said to Stefan, its symptoms
only appear very rarely.  But it'll go into the trunk.

> Leo

-- 
Alan Mackenzie (Nuremberg, Germany).