Re: prevent creation of invalid static constructors from gimplifier

[Olivier Hainque <hainque at adacore dot com>]

Ian Lance Taylor wrote:
> I was thinking more of checking it in the CONSTRUCTOR case in
> categorize_ctor_elements_1. We want categorize_ctor_elements and
> initializer_constant_valid_p to consistently return the same value.

> Your change assumes that initializer_constant_valid_p will return true
> for any TREE_CONSTANT initializer, but that is not the case.

 Indeed. I misinterpreted your former comment

 << And, of course, if we do add that new return value, we might as
    well remove the p_nc_elts return value, since it will no longer be
    used.
 >>

 as to be applied to both categorize and categorize_1. I didn't know
 how to match this with your other comment about only checking in the
 CONSTRUCTOR case in categorize_1 and made incorrect assumptions from
 there.

 [...]

 Thanks for the example.

 I think I see what you meant now. Something like the change you suggested
 to categorize_1 for the CONSTRUCTOR case only, plus categorize changed into

     HOST_WIDE_INT nc_elts = 0;
     ...
     categorize_ctor_elements_1 (ctor, p_nz_elts, &nc_elts, p_elt_count,
				 p_must_clear);

     ...
     return nc_elts == 0 && TREE_CONSTANT (ctor);

 Is this better/correct understanding ?