[RFC PATCH] -Wsizeof-pointer-memaccess warning

[Jakub Jelinek]

On Wed, Aug 15, 2012 at 03:39:29PM +0000, Joseph S. Myers wrote:
> On Wed, 18 Jul 2012, Jakub Jelinek wrote:
> 
> > +	  if (warn_sizeof_pointer_memaccess
> > +	      && sizeof_arg != NULL_TREE)
> > +	    sizeof_pointer_memaccess_warning (c_last_sizeof_arg_loc,
> > +					      expr.value, exprlist,
> > +					      sizeof_arg,
> > +					      sizeof_ptr_memacc_comptypes);
> 
> Why do you pass a local variable sizeof_arg but a global 
> c_last_sizeof_arg_loc here?  I'd have expected the same approach to apply 
> for both arguments (e.g. passing down a pointer to a location alongside 
> &sizeof_arg).

I guess I could pass it the same way as sizeof_arg or stick the location_t
into sizeof_arg (surround the value into NOP_EXPR with the location_t).

I was mainly interested in whether such an approach is acceptable, or
whether I need to stop evaluating sizeof right away, create SIZEOF_EXPR
and only fold it during fully_fold*.  I've briefly looked at that today,
and see several problems with that though: 1) fully_fold* is in c-family/,
so can't use e.g. c_vla_type_p (at least not easily, it could use a
langhook) 2) c_expr_sizeof_{expr,type} currently calls pop_maybe_used,
I don't see how that could be easily deferred till later when the expression
is fully folded.  So, we might need to do what c_expr_sizeof_{expr,type}
does right now (including fully folding), to call pop_maybe_used with the
right argument, and then either create SIZEOF_EXPR with the original
argument and fully fold it again during fully folding of the SIZEOF_EXPR,
or add an argument to SIZEOF_EXPR which would contain the folded expression
(i.e. have two arguments, the original expression and the value of the
sizeof).

	Jakub