Fix up vla, vm and [*] semantics for C

[Joseph S. Myers]

On Fri, 12 May 2006, Mike Stump wrote:

> Review points:
> 
> Ok to change typeof(vm) evaluation semantics for gnu89?
> 
> Ok to change sizeof (vla) evaluation semantics for gnu89?

Yes.

> Ok to narrow vla usage in structs and unions in gnu89?

VLAs and VM types in structs and unions are a well-established GNU C 
extension.  They have also caused many problems (especially with 
variable-size function arguments).  I don't know how widely used it is, 
but widely enough to get bug reports (unlike parameter predeclarations).

The appropriate handling is simply to pedwarn-if-pedantic for VLAs and VM 
types in structures/unions, for both gnu89 and gnu99.

> Ok to narrow vm with extern in gnu89?

I think so.

> Ok to narrow gnu99 to be more like c99 wrt vla/vm/[*]?
> 
> I think based upon the testing results, gnu89 and gnu99 need to leave vlas in
> structures and unions alone, agree?  Joseph, want to propose this for the next
> C standard?

New feature proposals aren't for now being considered except through the 
TR procedure.  I don't think this is a particularly good feature, but if 
it's to be deprecated then it needs the full deprecation procedure of 
deprecating for release series and removing in a subsequent release 
series.

> +bool
> +c_vla_type_p (tree t)

Needs comment.

> +  /* True when these areguments had [*].  */

Typo, should be "arguments".

+       if (c_vla_type_p (t1) || c_vla_type_p (t2))
+           break;

Why is this needed when there's

        if (d1_variable || d2_variable)
          break;

shortly above which should do the same thing?

You use

+      if (was_vm)
+	c_finish_expr_stmt (expr.value);

to evaluate the argument of typeof.  Will this ensure the argument only 
gets evaluated when the typeof gets evaluated?  For example, the typeof 
may not get evaluated inside sizeof, or in a cast in conditionally 
executed code.  sizeof(typeof(VLA)) evaluates the VLA, sizeof(typeof(VM)) 
doesn't evaluate the non-VLA VM.  (If this is hard to fix - involving the 
notion of when a type may or may not get evaluated - then fixing the 
ordinary C99 cases not involving side-effects in array sizes etc. is still 
useful.)

-- 
Joseph S. Myers               http://www.srcf.ucam.org/~jsm28/gcc/
    jsm@polyomino.org.uk (personal mail)
    joseph@codesourcery.com (CodeSourcery mail)
    jsm28@gcc.gnu.org (Bugzilla assignments and CCs)