Re: PATCH: PR target/39082: union with long double doesn't follow x86-64 psABI

[Uros Bizjak <ubizjak at gmail dot com>]

On Wed, Feb 4, 2009 at 7:26 PM, H.J. Lu <hjl.tools@gmail.com> wrote:

>>> I am not convinced such ABI changes should be done without adding
>>> a diagnostic for the cases that changed.  And I agree that release branches
>>> should not be touched.
>>>
>>
>> This problem happens deep in the backend. Adding a proper diagnostic
>> may not be easy without FE help. I will see what I can do.
>>
>
> Here is the updated patch with warning for ABI change. OK for trunk?

Warning w.r.t. to ABI changes was discussed in [1] when the meaning of
sync nand builtins was changed. Mark suggested that instead of a
warning, a note should be emitted once per compilation file and there
should be a way to disable this note/warning.

Also, I think we should say at which point the ABI has changed: "The
ABI of passing union with long double has changed in GCC 4.4".

All these suggestions were implemented in [2]. I think that patch at
[2] is a good and accepted example of this approach.

[1] http://gcc.gnu.org/ml/gcc-patches/2008-11/msg00760.html
[2] http://gcc.gnu.org/ml/gcc-patches/2008-11/msg00806.html

Thanks,
Uros.