]> gcc.gnu.org Git - gcc.git/commit
dwarf2out: Don't move variable sized aggregates to comdat [PR114015]
authorJakub Jelinek <jakub@redhat.com>
Fri, 1 Mar 2024 13:57:15 +0000 (14:57 +0100)
committerJakub Jelinek <jakub@redhat.com>
Fri, 1 Mar 2024 13:57:15 +0000 (14:57 +0100)
commit5b1fb8f8b4fe60745dece9b2f83c155c772ca66d
tree27daaa7b7bc10991110954c0578de6591e64be1d
parent270d0f69cffbfa81f61c5a6b7f7de2c5919703c7
dwarf2out: Don't move variable sized aggregates to comdat [PR114015]

The following testcase ICEs, because we decide to move that
struct { char a[n]; } DW_TAG_structure_type into .debug_types section
/ DW_UT_type DWARF5 unit, but refer from there to a DW_TAG_variable
(created artificially for the array bounds).
Even with non-bitint, I think it is just wrong to use .debug_types
section / DW_UT_type for something that uses DW_OP_fbreg and similar
in it, things clearly dependent on a particular function.
In most cases, is_nested_in_subprogram (die) check results in such
aggregates not being moved, but in the function parameter type case
that is not the case.

The following patch fixes it by returning false from should_move_die_to_comdat
for non-constant sized aggregate types, i.e. when either we gave up on
adding DW_AT_byte_size for it because it wasn't expressable, or when
it is something non-constant (location description, reference, ...).

2024-03-01  Jakub Jelinek  <jakub@redhat.com>

PR debug/114015
* dwarf2out.cc (should_move_die_to_comdat): Return false for
aggregates without DW_AT_byte_size attribute or with non-constant
DW_AT_byte_size.

* gcc.dg/debug/dwarf2/pr114015.c: New test.
gcc/dwarf2out.cc
gcc/testsuite/gcc.dg/debug/dwarf2/pr114015.c [new file with mode: 0644]
This page took 0.066827 seconds and 5 git commands to generate.