[Distutils] manylinux vs 3.6.0

[Robin Becker]

On 12/10/2016 22:09, Nathaniel Smith wrote:
> On Wed, Oct 12, 2016 at 7:05 AM, Robin Becker <robin at reportlab.com> wrote:
>> Any manylinux experts here?
>......
>> +OPENSSL_ROOT=openssl-1.0.2j
>> +OPENSSL_HASH=e7aff292be21c259c6af26469c7a9b3ba26e9abaaffd325e3dccc9785256c431
>
> If you could submit a quick PR for this to the manylinux repo then
> that'd be helpful.
>
I will attempt this, but I'm not too familiar with github as yet. I guess I have 
to fork etc etc.


> [...]
............
>
> It's not the pydtrace stuff that's new, it's that Benjamin switched
> them to use some exotic C99 features to get feedback on whether those
> features are usable:
>     https://bugs.python.org/issue28092#msg276164
>
> I suspect that this will get reverted again before 3.6-final given
> that it does in fact cause problems...

ok, probably that should have been conditional on gcc version or something.

>
> That said, I'm surprised that it's causing problems in the manylinux
> docker image. The default compiler on CentOS 5 is gcc 4.1, which
> definitely can't build 3.6.0b2. But the build scripts should be
> downloading and using gcc 4.8 instead, and 4.8 should be able to build
> 3.6.0b2. So apparently one of those shoulds is wrong. First step is
> probably to figure out exactly what version of gcc the Python build is
> using.
>
I can volunteer some effort for trying to figure out what's going wrong. I did 
not use the deploy.sh script so perhaps my dockering was wrong somehow

I did this

 > docker build -f Dockerfile-x86_64 -t rl/manylinux-x86_64 .

> -n
>


-- 
Robin Becker