Re: [PATCH] fix PR40026, failure from COMPOUND_LITERAL_EXPR move to middle-end

[Richard Guenther <richard dot guenther at gmail dot com>]

On Mon, May 11, 2009 at 4:55 PM, Paolo Bonzini <bonzini@gnu.org> wrote:
> This happened because optimize_compound_literals_in_ctor was previously
> handled separately in the C front-end's gimplification hook, while now
> it's part of the gimplifier proper.
>
> What the C front-end was doing was to replace the CONSTRUCTOR in the
> MODIFY_EXPR that was hosting it; I thought this was unnecessary when
> doing it in gimplify_init_constructor, but I was wrong and the analysis
> in the PR comments says why.
>
> The old C front-end code was doing
>
> - ? ? ? ? if (ctor != TREE_OPERAND (*expr_p, 1))
> - ? ? ? ? ? {
> - ? ? ? ? ? ? *expr_p = copy_node (*expr_p);
> - ? ? ? ? ? ? TREE_OPERAND (*expr_p, 1) = ctor;
> - ? ? ? ? ? ? return GS_OK;
> - ? ? ? ? ? }
>
> while I just patch TREE_OPERAND (*expr_p, 1). ?I decided to do this
> because gimplify.c never copies trees, it just patches in place all the
> time.
>
> I took the opportunity to tighten an if into an assertion. ?That it
> holds should be obvious from inspection of the callers of
> gimplify_init_constructor.
>
> Bootstrapped/regtested i686-pc-linux-gnu, ok for trunk?

Ok.

Thanks,
Richard.

> Paolo
>
> 2009-05-11 ?Paolo Bonzini ?<bonzini@gnu.org>
>
> ? ? ? ?PR tree-optimization/40026
> ? ? ? ?* gimplify.c (gimplify_init_constructor): Change initial conditional
> ? ? ? ?to assertion. ?Rewrite TREE_OPERAND (*expr_p, 1) after
> ? ? ? ?optimize_compound_literals_in_ctor.
>
>
> 2009-05-11 ?Paolo Bonzini ?<bonzini@gnu.org>
>
> ? ? ? ?PR tree-optimization/40026
> ? ? ? ?* gcc.c-torture/compile/pr40026.c: New testcase.
>
> Index: gimplify.c
> ===================================================================
> --- gimplify.c ?(revision 147360)
> +++ gimplify.c ?(working copy)
> @@ -3643,14 +3643,11 @@ static enum gimplify_status
> ?gimplify_init_constructor (tree *expr_p, gimple_seq *pre_p, gimple_seq *post_p,
> ? ? ? ? ? ? ? ? ? ? ? ? ? bool want_value, bool notify_temp_creation)
> ?{
> - ?tree object, new_ctor;
> - ?tree ctor = TREE_OPERAND (*expr_p, 1);
> - ?tree type = TREE_TYPE (ctor);
> + ?tree object, ctor, type;
> ? enum gimplify_status ret;
> ? VEC(constructor_elt,gc) *elts;
>
> - ?if (TREE_CODE (ctor) != CONSTRUCTOR)
> - ? ?return GS_UNHANDLED;
> + ?gcc_assert (TREE_CODE (TREE_OPERAND (*expr_p, 1)) == CONSTRUCTOR);
>
> ? if (!notify_temp_creation)
> ? ? {
> @@ -3661,8 +3658,10 @@ gimplify_init_constructor (tree *expr_p,
> ? ? }
>
> ? object = TREE_OPERAND (*expr_p, 0);
> - ?new_ctor = optimize_compound_literals_in_ctor (ctor);
> - ?elts = CONSTRUCTOR_ELTS (new_ctor);
> + ?ctor = TREE_OPERAND (*expr_p, 1) =
> + ? ?optimize_compound_literals_in_ctor (TREE_OPERAND (*expr_p, 1));
> + ?type = TREE_TYPE (ctor);
> + ?elts = CONSTRUCTOR_ELTS (ctor);
> ? ret = GS_ALL_DONE;
>
> ? switch (TREE_CODE (type))
>
>
>
>
>
> typedef struct {
> ? ?unsigned long bits;
> } S;
> struct T {
> ? ?S span;
> ? ?int flags;
> };
>
> struct T f(int x)
> {
> ?return (struct T) {
> ? ? ?.span = (S) { 0UL },
> ? ? ?.flags = (x ? 256 : 0),
> ?};
> }
>
>