bug#60234: Build failure on mac os 12.6 / gcc 12.2

bug#60234: Build failure on mac os 12.6 / gcc 12.2

[lloda]

The error is

.../libguile/threads.h:194:43: error: 'scm_i_current_thread' is defined with tls model global-dynamic
  194 | SCM_INTERNAL SCM_THREAD_LOCAL scm_thread *scm_i_current_thread;
      |                                           ^
.../libguile/threads.c:357:30: note: previously defined here as local-dynamic
  357 | SCM_THREAD_LOCAL scm_thread *scm_i_current_thread = NULL;

Simply repeating SCM_INTERNAL in the .c fixes it...

regards

  Daniel

bug#60234: Build failure on mac os 12.6 / gcc 12.2

[lloda]

Patched in f859e0f58b211eedcb0dce4f2382cfebf37010d7.

bug#60234: Build failure on mac os 12.6 / gcc 12.2

[Ludovic Courtès]

Hi Daniel,

lloda <lloda@sarc.name> skribis:

> .../libguile/threads.h:194:43: error: 'scm_i_current_thread' is defined with tls model global-dynamic
>   194 | SCM_INTERNAL SCM_THREAD_LOCAL scm_thread *scm_i_current_thread;
>       |                                           ^
> .../libguile/threads.c:357:30: note: previously defined here as local-dynamic
>   357 | SCM_THREAD_LOCAL scm_thread *scm_i_current_thread = NULL;
>
> Simply repeating SCM_INTERNAL in the .c fixes it...

The problem is that ‘SCM_INTERNAL’ is synonymous with ‘extern’, which
makes no sense for a definition (threads.c:357), so rightfully GCC
GNU/Linux rightfully complains:

--8<---------------cut here---------------start------------->8---
  CC       libguile_3.0_la-threads.lo
threads.c:358:43: warning: 'scm_i_current_thread' initialized and declared 'extern'
  358 | SCM_INTERNAL SCM_THREAD_LOCAL scm_thread *scm_i_current_thread = NULL;
      |                                           ^~~~~~~~~~~~~~~~~~~~
--8<---------------cut here---------------end--------------->8---

It’s just a warning, but still not looking good.

Is there something else at play, such as a ‘-ftls-model’ flag being
passed to GCC somehow (info “(gcc) Code Gen Options")?

If not, should we have:

  #define SCM_THREAD_LOCAL \
    __thread __attribute__ ((__tls_model__ ("global-dynamic")))

instead (info "(gcc) Common Variable Attributes")?

Would that work with Clang?

Ludo’.

bug#60234: Build failure on mac os 12.6 / gcc 12.2

[lloda]

[-- Attachment #1: Type: text/plain, Size: 2012 bytes --]



> On 18 Jan 2023, at 23:16, Ludovic Courtès <ludo@gnu.org <mailto:ludo@gnu.org>> wrote:
> 
> Hi Daniel,
> 
> lloda <lloda@sarc.name <mailto:lloda@sarc.name>> skribis:
> 
>> .../libguile/threads.h:194:43: error: 'scm_i_current_thread' is defined with tls model global-dynamic
>> 194 | SCM_INTERNAL SCM_THREAD_LOCAL scm_thread *scm_i_current_thread;
>>     |                                           ^
>> .../libguile/threads.c:357:30: note: previously defined here as local-dynamic
>> 357 | SCM_THREAD_LOCAL scm_thread *scm_i_current_thread = NULL;
>> 
>> Simply repeating SCM_INTERNAL in the .c fixes it...
> 
> The problem is that ‘SCM_INTERNAL’ is synonymous with ‘extern’, which
> makes no sense for a definition (threads.c:357), so rightfully GCC
> GNU/Linux rightfully complains:
> 
> --8<---------------cut here---------------start------------->8---
> CC       libguile_3.0_la-threads.lo
> threads.c:358:43: warning: 'scm_i_current_thread' initialized and declared 'extern'
> 358 | SCM_INTERNAL SCM_THREAD_LOCAL scm_thread *scm_i_current_thread = NULL;
>     |                                           ^~~~~~~~~~~~~~~~~~~~
> --8<---------------cut here---------------end--------------->8---
> 
> It’s just a warning, but still not looking good.

Hi,

Agreed, I had the same warning on mac os. Looked like the least bad choice...

> Is there something else at play, such as a ‘-ftls-model’ flag being
> passed to GCC somehow (info “(gcc) Code Gen Options")?
> 
> If not, should we have:
> 
> #define SCM_THREAD_LOCAL \
>   __thread __attribute__ ((__tls_model__ ("global-dynamic")))
> 
> instead (info "(gcc) Common Variable Attributes")?
> 
> Would that work with Clang?
> 
> Ludo’.

I don't seem to have any such flags.

The attribute does fix the issue on mac os with gcc 12.

I hadn't tried clang before, but I did now, and clang 14 works fine with or without the attribute. So the fix would only be for gcc.

Thanks

 Daniel

[-- Attachment #2: Type: text/html, Size: 5727 bytes --]