Re: [PATCH] Correctly determine libc.so 'OUTPUT_FORMAT' when cross-compiling.

[Carlos O'Donell <carlos@redhat.com>]

On 10/13/22 16:56, Fangrui Song wrote:
> On 2022-10-07, Ludovic Courtès via Libc-alpha wrote:
>> Hi Carlos,
>>
>> Carlos O'Donell <carlos@redhat.com> skribis:
>>
>>> On Thu, Jul 01, 2021 at 11:00:19PM +0200, Ludovic Courtès via Libc-alpha wrote:
>>>> Commit 87d583c6e8cd0e49f64da76636ebeec033298b4d replaces the sed script
>>>> with an "objdump -f" invocation to determine the 'OUTPUT_FORMAT' bit of
>>>> the libc.so linker script.
>>>>
>>>> However, when cross-compiling, for example from x86_64-linux-gnu to
>>>> aarch64-linux-gnu, "objdump -f" would report the wrong
>>>> format ("elf64-little").  Conversely, "aarch64-linux-gnu-objdump -f"
>>>> reports "elf64-littleaarch64" as expected.
>>>>
>>>> This patch changes 'configure.ac' to use AC_CHECK_TOOL rather than
>>>> '$CC -print-prog-name=objdump' to determine the value of the OBJDUMP
>>>> variable.  That way, OBJDUMP is set to TRIPLET-objdump when
>>>> cross-compiling for TRIPLET.
>>>
>>> I've been tackling a backlog of old glibc patches, and this one is up
>>> next. Yes it's been over a year, but this patch still applies and the
>>> idea is sound. I've tested this with build-many-glibcs (bmg) on x86_64
>>> and it has no impact because bmg always sets OBJDUMP. My opinion is that
>>> bmg is the "base standard" for how we build native and cross tooling and
>>> so it your changes work here, they should work in other instances.
>>>
>>> The change looks good to me.
>>>
>>> No regresions on x86_64.
>>>
>>> Would you like me to commit this? :-)
>>
>> I don’t have commit access AFAIK so I’ll happily defer to you.  :-)
>>
>> Thanks for taking the time to look at the backlog!
>>
>> Ludo’.
> 
> Thanks. This looks good.
 
Tested again with bmg to cross x86_64 to aarch64. No regressions.

Pushed. Thank you!

-- 
Cheers,
Carlos.