llvm dev - Jun 2015 - [LLVMdev] .thumb_set

On 13 June 2015 at 04:12, Saleem Abdulrasool <compnerd at compnerd.org>
wrote:
> The behavior was taken from GAS.  I don't know if there are
applications
> using this and dependent on the behavior.  Given that this is a GNU
> extension, I think its better to keep the behavior similar to GAS rather
> than implement and diverge.
I kind of agree with Saleem, here. We don't want to deviate, even from
silly behaviour, when implementing a GNU extension.

If this brings problems, we may add a warning, but not much more, I'd say.

cheers,
--renato

Hi Renato
> On Jun 13, 2015, at 6:19 AM, Renato Golin <renato.golin at linaro.org
<mailto:renato.golin at linaro.org>> wrote:
> 
> On 13 June 2015 at 04:12, Saleem Abdulrasool <compnerd at compnerd.org
<mailto:compnerd at compnerd.org>> wrote:
>> The behavior was taken from GAS.  I don't know if there are
applications
>> using this and dependent on the behavior.  Given that this is a GNU
>> extension, I think its better to keep the behavior similar to GAS
rather
>> than implement and diverge.
> 
> I kind of agree with Saleem, here. We don't want to deviate, even from
> silly behaviour, when implementing a GNU extension.
I totally agree with matching whatever GAS says, which is
(https://sourceware.org/binutils/docs/as/ARM-Directives.html
<https://sourceware.org/binutils/docs/as/ARM-Directives.html>)

"This performs the equivalent of a .set directive in that it creates a
symbol which is an alias for another symbol (possibly not yet defined). This
directive also has the added property in that it marks the aliased symbol as
being a thumb function entry point, in the same way that the .thumb_func
directive does.”

The first part is most important here. Its the equivalent of a .set. 

I would argue that in this case it should throw exactly the same errors in
exactly the same situations as .set. Any more/less and it wouldn’t be
equivalent.

Sorry if that comes across as heavy handed, but i just want to be clear how i
interpret the behavior here.  Ideally i’d like to take whatever tests we have
for .set errors/warnings and port all of them to ARM to make sure we have the
same behavior.

Cheers,
Pete
> 
> If this brings problems, we may add a warning, but not much more, I'd
say.
> 
> cheers,
> --renato
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20150614/921b5b81/attachment.html>

On 15 June 2015 at 03:21, Pete Cooper <peter_cooper at apple.com>
wrote:
> I would argue that in this case it should throw exactly the same errors in
> exactly the same situations as .set. Any more/less and it wouldn’t be
> equivalent.
Sigh... You are, obviously, correct.

I hate when GCC/GAS documents something and does another. When you
bring this to them all you get is "we can't change now because
legacy".

In the past, we have chosen being compatible with the docs over
bug-for-bug, so in this case, I think we better implement .thumb_set
as .set and add a big comment about this in the code, so that future
bug reports can be quickly dismissed after a short investigation.

Saleem, does that sound reasonable?

cheers,
--renato