bug#16585: c-invalidate-state-cache fails if called when narrowed

bug#16585: c-invalidate-state-cache fails if called when narrowed

[Daniel Colascione]

c-invalidate-state-cache can fail if called in a narrowed region: we end 
up calling c-clear-char-property for bob even if that's before 
point-min, causing remote-text-properties to raise (args-out-of-range 1 2).

We get into this situation when expanding yasnippet snippets; we're 
calling c-invalidate-state-cache from c-before-change.

To repro manually, first write this to some buffer.

#include <stdio.h>

void foo() {

}


Then (this is important) run M-x c-mode. After starting the mode, move 
point to inside foo's body and eval (save-restriction (narrow-to-region 
(point-at-bol) (point-at-eol)) (c-invalidate-state-cache (point))). You 
should break into the debugger.

The patch below seems correct and fixes the problem for me, but it'd be 
nice if Alan could take a look at it before I push it to trunk. The 
version of c-invalidate-state-cache in cc-mode trunk is very different.

=== modified file 'lisp/progmodes/cc-engine.el'
--- lisp/progmodes/cc-engine.el	2014-01-19 12:32:47 +0000
+++ lisp/progmodes/cc-engine.el	2014-01-29 02:07:44 +0000
@@ -3318,15 +3318,17 @@
    ;; of all parens in preprocessor constructs, except for any such 
construct
    ;; containing point.  We can then call `c-invalidate-state-cache-1' 
without
    ;; worrying further about macros and template delimiters.
-  (c-with-<->-as-parens-suppressed
-   (if (and c-state-old-cpp-beg
-	    (< c-state-old-cpp-beg here))
-       (c-with-all-but-one-cpps-commented-out
-	c-state-old-cpp-beg
-	(min c-state-old-cpp-end here)
-	(c-invalidate-state-cache-1 here))
-     (c-with-cpps-commented-out
-      (c-invalidate-state-cache-1 here)))))
+  (save-restriction
+    (widen)
+    (c-with-<->-as-parens-suppressed
+     (if (and c-state-old-cpp-beg
+              (< c-state-old-cpp-beg here))
+         (c-with-all-but-one-cpps-commented-out
+          c-state-old-cpp-beg
+          (min c-state-old-cpp-end here)
+          (c-invalidate-state-cache-1 here))
+       (c-with-cpps-commented-out
+        (c-invalidate-state-cache-1 here))))))

  (defmacro c-state-maybe-marker (place marker)
    ;; If PLACE is non-nil, return a marker marking it, otherwise nil.

bug#16585: c-invalidate-state-cache fails if called when narrowed

[Glenn Morris]

See also http://debbugs.gnu.org/13713 .

BTW, if you use

Package: emacs,cc-mode

at the start of the initial report, all mails will also go to bug-cc-mode.

bug#16585: c-invalidate-state-cache fails if called when narrowed

[Daniel Colascione]

On 01/28/2014 06:49 PM, Glenn Morris wrote:
>
> See also http://debbugs.gnu.org/13713 .
>
> BTW, if you use
>
> Package: emacs,cc-mode
>
> at the start of the initial report, all mails will also go to bug-cc-mode.
>

Thanks. Adding bug-cc-mode manually now. I'd like to commit the change 
to Emacs before release.

bug#16585: c-invalidate-state-cache fails if called when narrowed

[Alan Mackenzie]

Hi, Daniel.

On Tue, Jan 28, 2014 at 06:09:04PM -0800, Daniel Colascione wrote:
> c-invalidate-state-cache can fail if called in a narrowed region: we end 
> up calling c-clear-char-property for bob even if that's before 
> point-min, causing remote-text-properties to raise (args-out-of-range 1 2).

> We get into this situation when expanding yasnippet snippets; we're 
> calling c-invalidate-state-cache from c-before-change.

> To repro manually, first write this to some buffer.

> #include <stdio.h>

> void foo() {

> }


> Then (this is important) run M-x c-mode. After starting the mode, move 
> point to inside foo's body and eval (save-restriction (narrow-to-region 
> (point-at-bol) (point-at-eol)) (c-invalidate-state-cache (point))). You 
> should break into the debugger.

> The patch below seems correct and fixes the problem for me, but it'd be 
> nice if Alan could take a look at it before I push it to trunk. The 
> version of c-invalidate-state-cache in cc-mode trunk is very different.

I think I'd like to check that nothing can go awry in
c-invalidate-state-cache-1 with the widened buffer.  Give me a day or
two.

-- 
Alan Mackenzie (Nuremberg, Germany).

bug#16585: c-invalidate-state-cache fails if called when narrowed

[Alan Mackenzie]

Hi, Daniel.

On Tue, Jan 28, 2014 at 06:09:04PM -0800, Daniel Colascione wrote:
> c-invalidate-state-cache can fail if called in a narrowed region: we end 
> up calling c-clear-char-property for bob even if that's before 
> point-min, causing remote-text-properties to raise (args-out-of-range 1 2).

> We get into this situation when expanding yasnippet snippets; we're 
> calling c-invalidate-state-cache from c-before-change.

> To repro manually, first write this to some buffer.

> #include <stdio.h>

> void foo() {

> }


> Then (this is important) run M-x c-mode. After starting the mode, move 
> point to inside foo's body and eval (save-restriction (narrow-to-region 
> (point-at-bol) (point-at-eol)) (c-invalidate-state-cache (point))). You 
> should break into the debugger.

Thanks for this crystal clear bug report, and for taking the trouble to
debug it.

> The patch below seems correct and fixes the problem for me, but it'd be 
> nice if Alan could take a look at it before I push it to trunk. The 
> version of c-invalidate-state-cache in cc-mode trunk is very different.

I don't think (but I'm not quite sure) that this patch is correct, since
it leaves c-invalidate-state-cache-1 running widened.  That might well
lead to incorrect "state cache" manipulations.  Instead I've widened the
buffer around the two calls which clear and set the text properties.
Would you please try out the following patch, which should fix the
problem.


=== modified file 'lisp/progmodes/cc-defs.el'
*** lisp/progmodes/cc-defs.el	2014-02-02 10:25:29 +0000
--- lisp/progmodes/cc-defs.el	2014-02-08 22:21:01 +0000
***************
*** 1293,1302 ****
    ;; suppressed.
    `(unwind-protect
         (c-save-buffer-state ()
! 	 (c-clear-cpp-delimiters ,beg ,end)
  	 ,`(c-with-cpps-commented-out ,@forms))
       (c-save-buffer-state ()
!        (c-set-cpp-delimiters ,beg ,end))))
  \f
  (defsubst c-intersect-lists (list alist)
    ;; return the element of ALIST that matches the first element found
--- 1293,1306 ----
    ;; suppressed.
    `(unwind-protect
         (c-save-buffer-state ()
! 	 (save-restriction
! 	   (widen)
! 	   (c-clear-cpp-delimiters ,beg ,end))
  	 ,`(c-with-cpps-commented-out ,@forms))
       (c-save-buffer-state ()
!        (save-restriction
! 	 (widen)
! 	 (c-set-cpp-delimiters ,beg ,end)))))
  \f
  (defsubst c-intersect-lists (list alist)
    ;; return the element of ALIST that matches the first element found

-- 
Alan Mackenzie (Nuremberg, Germany).

bug#16585: c-invalidate-state-cache fails if called when narrowed

[Alan Mackenzie]

Hi, Daniel.

On Sat, Feb 08, 2014 at 10:37:17PM +0000, Alan Mackenzie wrote:

[ .... ]

> I don't think (but I'm not quite sure) that this patch is correct, since
> it leaves c-invalidate-state-cache-1 running widened.  That might well
> lead to incorrect "state cache" manipulations.  Instead I've widened the
> buffer around the two calls which clear and set the text properties.
> Would you please try out the following patch, which should fix the
> problem.

I just forgot to say, after applying the patch, please compile first
cc-defs.el, then cc-engine.el.

-- 
Alan Mackenzie (Nuremberg, Germany).

bug#16585: c-invalidate-state-cache fails if called when narrowed

[Alan Mackenzie]

Bug fixed.

-- 
Alan Mackenzie (Nuremberg, Germany).