> On 22 Dec 2015, at 14:01, Ward Poelmans wrote:
>
> Hi Franky,
>
>> On 22-12-15 12:02, Backeljauw Franky wrote:
>>
>> I’ve just installed EasyBuild 2.5.0 and noticed there are several
>> intel-2016b.xx toolchains now:
>
> Yes, that's because we haven't decided about which one is going to be
> intel/2016a. Kenneth was going to send a mail about it but he forgot.
I didn't forget, I gave priority to releasing EasyBuild v2.5.0, and then ended
up not starting a thread to discuss the 2016a toolchains, since I wasn't going
to be able to follow it up closely...
>
> But for the intel case, most likely the 2016.01 will be promoted to
> 2016a.
I disagree, I feel we should stay away from the 2016 versions of the Intel
compilers for a bit longer, especially given the amount of effort that was
needed to add support for installing them and since there are known
compatibility issues with GCC v5.x.
We've had bad experiences before with switching to a new major release of the
Intel compilers, so jumping in 'early' is probably not a good idea.
> On the foss toolchain we still have to decide whether to stick
> with GCC 4.9.x or go to 5.x.
Personally, I would stick with the latest GCC 4.x for a little longer, but I
have to admit this more of a personal preference than having solid technical
arguments.
For intel/2016a, we're pretty much stuck with GCC v4.9.x because of known
issues with using GCC v5.x as a 'base' for the Intel compilers.
And I would rather stick to the same GCC(core) version for both the intel and
foss toolchains, just for simplicity sake.
If there are compelling (technical) arguments for using GCC v5.x, I'm keen on
hearing them.
Ideally, whatever composition we pick for the 2016a versions of the 'common'
toolchains should be thoroughly tested before we set it in stone, i.e. build a
significant amount of libraries and applications with it.
I'll try and flesh out a proposal for both toolchains later this week (probably
via a WIP pull request), and hope to get to testing them between Christmas and
New Years.
If anyone is up for doing it that earlier, please don't hold back.
>
>> Another final question: any idea how to get rid of the UserWarning?
>
> Yes, that shouldn't be there. Open a bug report so we don't forget about it.
This may be related to the known Python packaging issues that I intended to
solve for EasyBuild v2.5.0, but didn't have time for.
So yes, please open an issue for it (in the framework repo on GitHub), so we
don't forget about it.
regards,
Kenneth
