RTEMS 5 gcc-fb371a33fa6 vs gcc-9.2.0

[Chris Johns]

On 14/8/19 3:02 pm, Sebastian Huber wrote:
> On 14/08/2019 01:52, Chris Johns wrote:
>> On 13/8/19 3:02 pm, Sebastian Huber wrote:
>>
>>> the patch just changed GCC 9.1 to 9.2 on all targets which use GCC 9, these are
>>> or1k, riscv, and x86_64.
>>
>> The change as broken MacOS due to this bug ...
>>
>> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=81797
>>
>> I posted build results showing the error is still present. I had a patch in the
>> previous version and it should apply.
> 
> Now I am a bit confused. This patch
> 
> https://git.rtems.org/rtems-source-builder/commit/?id=5a0dba77b13eec49f86ce3812fb74f65dfa1e98d
> 
> 
> changes the GCC version from 9.1.0 to 9.2.0 on or1k, riscv, and x86_64. I didn't
> change the patches:
> 
> https://git.rtems.org/rtems-source-builder/tree/rtems/config/tools/rtems-gcc-9.1.0-newlib-6661a67.cfg?id=5a0dba77b13eec49f86ce3812fb74f65dfa1e98d

OK and thanks. I will have a look next week to see what is happening.

> The GCC bug doesn't seem to be a GCC 9.1.0 to 9.2.0 regression. In your build
> log only riscv failed, or1k and x86_64 passed. Is this a sporadic problem?

Yes. No one has chased the issue down and the reason it happens so solutions and
fixes currently are and nothing more than speculation. It may not fail and it
does not fail for some. On a faster Mac then my mini it is easier to have happen.

>>> Since there will be probably no RTEMS 5 in the near future, maybe we should move
>>> to GCC 9.2 in general.
>>
>> This is a catch-22, I hope to start on the release process soon but things like
>> this add to the complexity and add to the time it takes.
>>
>>> If we do this on PowerPC, then the SPE is no longer supported.
>>
>> Joel has stated many times in talks I have seen that we follow the architectures
>> GCC supports and if one is removed we remove support.
>>
>> Does this mean 5.1 is the last version of RTEMS to support SPE?
> 
> This was my plan.

OK.

>> Should the specific BSPs be moved to tier 4 and marked for removal?
> 
> The SPE support should bit rot for a while. The chips are still in production
> and used with RTEMS, e.g. by EPICS users.

When we release we need to make sure these users know this is the case. I am not
sure how we should do that.

>>> On ARM a breaking change in compiler options is necessary.
>>
>> What options are these?
> 
> ARM changed the FPU options in GCC 8 and later.
> 
>>
>> Also this seems back to front to me. Should all hosts be on 9.2.0 and those that
>> cannot have specific versions?
> 
> Only PowerPC should stay at GCC 7 until the next RTEMS release from my point of
> view. Everything else can move on.

Awesome. To make sure I understand the steps ...

1. Move gcc-fb371a33fa6 to 5/rtems-powerpc
2. Set the default to gcc-9.2.0
3. Move all other archs to the defaults

?