On Wed, Nov 23, 2016 at 09:11:02AM -0500, Jason Merrill wrote:
> On Tue, Nov 22, 2016 at 3:45 PM, Jakub Jelinek <ja...@redhat.com> wrote:
> > As mentioned in the PR, another option I see is special case
> > REFERENCE_REF_P INDIRECT_REFs and MEM_REFs into which they are gimplified
> > in cp_genericize_r by not changing is_invisiref_parm decls if they are
> > already wrapped in those.
>
> That sounds more robust.
Actually, now that I think about it, it sounds less robust.
E.g. if the invisiref parm is initially used as operand of ADDR_EXPR,
then during the first cp_genericize_tree it will be turned into
ADDR_EXPR <INDIRECT_REF <decl>>, but gimplification will turn that just
into decl, and if we genericize that again, we turn it into INDIRECT_REF <decl>
because we won't see INDIRECT_REF/MEM_REF wrapping it.
Is there any easy way to construct a testcase where VEC_INIT_EXPR
initializer will be an arbitrary expression where such &parm could appear
if parm is passed by invisible reference?
Also, gimplification already performs tons of other folding (through fold_stmt),
making it less likely to recognize those.
Jakub
