bug#37137: Setting font-lock-case-fold-search to t causes hangs on certain types in c-mode

[Alan Mackenzie <acm@muc.de>]

Hello, Zachary.

In article <mailman.18.1566429065.1922.bug-gnu-emacs@gnu.org> you wrote:
> [-- text/plain, encoding 7bit, charset: UTF-8, 31 lines --]

> I have in my .emacs file only these two lines:

>     (custom-set-variables
>      '(font-lock-keywords-case-fold-search t))

May I ask why you're making this setting?

I think font-lock-keywords-case-fold-search is really intended to be set
by major modes, rather than as a user customisation.

> Then if I visit an empty or nonexistent C file and just type "LONG"
> (in all caps or with at least one letter being capital), Emacs will
> hang after typing the final G but before the G appears on
> screen. Typing C-g a couple of times gets Emacs unstuck until I
> continue to type. The same behavior also happens if I type "SHORT".

Yes.  The keywords in all the CC Mode languages are case sensitive.

Almost exactly the same bug came up in 2013, when the solution was to
bind case-fold-search to nil at every "entry point" to CC Mode.  I missed
font-lock-keywords-case-fold-search at the time.

> This happens in both Emacs 26.1 in text mode on a Linux machine and in
> Emacs 25.3 in GUI mode on Windows.

> After messing with GDB a bit, I am pretty sure the hang is at the
> while loop in c-forward-type at cc-engine.el:7657 and its getting
> stuck because looking-at matches LONG because its case insensitive,
> but the c-forward-keyword-clause doesn't move forward because the
> first thing it does is try to find "LONG" in c-keywords-obarray via
> `intern-soft` and immediately returns because it doesn't find it
> there. Here is the while loop:

>     (while (progn
>              (setq safe-pos (point))
>              (looking-at c-opt-type-component-key))
>       (when (and c-record-type-identifiers
>                  (looking-at c-primitive-type-key))
>         (c-record-type-id (cons (match-beginning 1)
>                                 (match-end 1))))
>       (c-forward-keyword-clause 1))

That's an excellent piece of debugging.  :-)

I think the best fix for the bug would be for CC Mode to set
font-lock-keywords-case-fold-search explicitly to nil (buffer locally) at
C Mode (etc.) initialisation.  This would undo the effect of your
custom-set-variables form in C Mode buffers.  Would this be a problem?

-- 
Alan Mackenzie (Nuremberg, Germany).