[PATCH v2] gas/ELF: don't accumulate .type settings

Jan Beulich JBeulich@suse.com
Mon Jul 8 06:54:00 GMT 2019


On 08.07.2019 07:42, Alan Modra wrote:
> On Wed, Jul 03, 2019 at 08:05:48AM +0000, Jan Beulich wrote:
>> 	* config/obj-elf.c (obj_elf_type): Check for conflicts between
>> 	old and new types.
>> 	* config/tc-hppa.h (md_elf_symbol_type_change): New.
>> 	* doc/as.texi: Mention warning behavior for the ELF flavor of
>> 	.type.
>> 	* testsuite/gas/elf/type-2.e, testsuite/gas/elf/type-2.l,
>> 	testsuite/gas/elf/type-2.s: New.
>> 	* testsuite/gas/elf/elf.exp: Run new test.
> 
> aarch64_be-linux-gnu_ilp32  +FAIL: elf type-2 list
> aarch64-linux  +FAIL: elf type-2 list
> nds32be-elf  +FAIL: elf type-2 list
> nds32le-linux  +FAIL: elf type-2 list
> rl78-elf  +FAIL: elf type-2 list
> 
> Plus a large number of mips-sgi-irix6 tests, all due to "symbol ..
> already has its type set" warnings.

Well, I can see about looking into these, but not before some time
next week. Until then all I can suggest is for someone to revert
the commit if the failures need to be taken care of quickly. I have
to admit though that it is quite unobvious to me why the test would
behave differently for different targets (target specific maintainer
input [or action] might be helpful here), _except_ ones like hppa
which truly customize the type setting. And afaics it's just hppa
which has been making use of md_elf_symbol_type(), and which hence
had to be extended to also make use of the new hook.

Jan



More information about the Binutils mailing list