svn commit: r557295 - in head/sysutils/slurm-wlm: . files

[Jason Bacon]

On 12/8/20 10:44 AM, Jason Bacon wrote:
> On 12/8/20 10:28 AM, Kyle Evans wrote:
>> On Tue, Dec 8, 2020 at 10:21 AM Jason W. Bacon <jwb at freebsd.org> wrote:
>>> Author: jwb
>>> Date: Tue Dec  8 16:21:16 2020
>>> New Revision: 557295
>>> URL: https://svnweb.freebsd.org/changeset/ports/557295
>>>
>>> Log:
>>>    sysutils/slurm-wlm: Back out -fno-common patches, add -fcommon flag
>>>
>>>    Fixes runtime regression caused by r546150
>>>
>>>    Reported by:    jrm
>>>    MFH:            2020Q4
>>>
>> I guess the regression was from the patch to
>> src/plugins/select/cons_tres/job_test.c and an issue somewhere without
>> slurmctld? That patch almost certainly should have made them weak
>> references instead of removing them entirely, based on the comment in
>> the vicinity.
>>
>> Thanks,
>>
>> Kyle Evans
> It's more complicated than that.  I tried converting to weak refs, but 
> the files are linked into multiple executables + the use of dlopen 
> might obscure things to the compiler/linker.  I'd have to review the 
> rules to verify the latter.
>
> I'll see if upstream has made -fcommon unnecessary in the next release 
> when I have time to work on an upgrade a few weeks out.
>
> Best,
>
>     JB
>

I do wonder why a team doing mass updates would take a risky approach 
like back-porting changes from a later upstream commit, rather than the 
safer approach of just enabling -fcommon.  I think source patches like 
this should generally be left to upstream or maintainers, who are more 
likely to take the time to fully test them.

I appreciate all the work that people put into these mass fixes but 
would suggest being as conservative as possible with them.

Best,

     JB

-- 
Earth is a beta site.