This is another regression present on the mainline. The compiler aborts on an
aggregate whose type is a record type with a misaligned scalar component and
with a representation clause for this component that gives it a size smaller
than the default. We're incorrectly laying out the type.
Tested on x86_64-suse-linux, applied on the mainline.
2013-03-06 Eric Botcazou <ebotca...@adacore.com>
* gcc-interface/decl.c (gnat_to_gnu_field): Remove the wrapper around
a misaligned integral type if a size is specified for the field.
2013-03-06 Eric Botcazou <ebotca...@adacore.com>
* gnat.dg/specs/aggr6.ads: New test.
--
Eric Botcazou-- { dg-do compile }
package Aggr6 is
type B15_T is mod 2 ** 15;
for B15_T'Size use 15;
for B15_T'Alignment use 1;
type B17_T is mod 2 ** 17;
for B17_T'Size use 17;
for B17_T'Alignment use 1;
type Rec_T is record
A : B17_T;
B : B15_T;
end record;
for Rec_T use record
A at 0 range 0 .. 16;
B at 0 range 17 .. 31;
end record;
for Rec_T'Size use 32;
C : constant Rec_T := (A => 1, B => 0);
end Aggr6;
Index: gcc-interface/decl.c
===================================================================
--- gcc-interface/decl.c (revision 196487)
+++ gcc-interface/decl.c (working copy)
@@ -6619,6 +6619,13 @@ gnat_to_gnu_field (Entity_Id gnat_field,
<= 0)
gnu_field_type = TREE_TYPE (TYPE_FIELDS (gnu_field_type));
+ /* Similarly if the field's type is a misaligned integral type, but
+ there is no restriction on the size as there is no justification. */
+ if (!needs_strict_alignment
+ && TYPE_IS_PADDING_P (gnu_field_type)
+ && INTEGRAL_TYPE_P (TREE_TYPE (TYPE_FIELDS (gnu_field_type))))
+ gnu_field_type = TREE_TYPE (TYPE_FIELDS (gnu_field_type));
+
gnu_field_type
= make_type_from_size (gnu_field_type, gnu_size,
Has_Biased_Representation (gnat_field));
