Re: Emacs port to gcc -fcheck-pointer-bounds

[Paul Eggert <eggert@cs.ucla.edu>]

On 12/13/2017 11:17 AM, Stefan Monnier wrote:
>> The fake pointer catches (at compile-time) common faults like the one 
>> the
>> attached patch fixes, where an int was passed where a Lisp_Object was 
>> expected. These are the most important faults that 
>> --enable-check-lisp-object-type catches. 
> ... it doesn't catch things like `x + n` since adding a constant to a 
> pointer is also a valid operation

Actually it catches even (x + n), because Lisp_Object is 'union Lisp_X 
*', and the union type is deliberately incomplete so the C compiler does 
not know its size and cannot multiply n by sizeof (union Lisp_X). The C 
Standard requires a diagnostic for (x + n) and practice compilers 
invariably issue at least a warning.

There are some things it doesn't catch. Most of these (e.g., 
'Lisp_Object x = 0;', or 'Lisp Object x = FOO, y = BAR; return x == y;') 
are harmless annoyances. The only worrisome thing not caught is 
converting between void * and Lisp_Object, e.g., 'Lisp_Object z = malloc 
(n);'. However, to my mind it's overkill to 
--enable-check-lisp-object-type by default just to catch this, as void * 
is dangerous with every C pointer type and there's little extra harm to 
making it dangerous with Lisp_Object too.

> Maybe we can turn it into a no-op.

Yes, that's my thought too.