On 04/23/2015 02:46 PM, Marek Polacek wrote:
This PR points out a problem with enum forward declarations (so C++ is out as
these are forbidden in C++). If we forward declare an enum, and later on
declare the enum with __attribute__ ((packed)), the attribute is ignored. The
reason is that when we first see the forward declaration, parser_xref_tag sets
up a forward-reference node, which has a default layout like unsigned int, so
among other things it does
TYPE_PRECISION (ref) = TYPE_PRECISION (unsigned_type_node);
But then in finish_enum we were confused: the code thinks that the precision
was specified with the mode attribute so it uses wrong TYPE_PRECISION. After
playing with this a bit, I think easiest fix is to check for a mode attribute.
If we can't find it, we know that TYPE_PRECISION was set when processing the
forward declaration.
We also need to copy TYPE_ALIGN, otherwise subsequent layout_type won't DTRT.
While at it, I noticed that we don't test the "specified mode too small" error
at all, so I've added enum-mode-1.c test.
It's unclear what to do with a combination of "packed" and "mode" attributes; I
suppose that should be invalid, but I'm not changing it now.
Bootstrapped/regtested on x86_64-linux, ok for trunk?
2015-04-23 Marek Polacek <pola...@redhat.com>
PR c/52085
* c-decl.c (finish_enum): Copy over TYPE_ALIGN. Also check for "mode"
attribute.
* gcc.dg/enum-incomplete-2.c: New test.
* gcc.dg/enum-mode-1.c: New test.
What happens if we have used the enum inside an aggregate? Can we just
blindly change the alignment/precision like that?
My inclination would to to issue an error if we had a forward enum decl
followed by a later packed enum.
jeff
