Leo Famulari <leo@famulari.name> writes:

> On Mon, Jul 31, 2017 at 10:07:18PM +0200, Marius Bakke wrote:
>> * gnu/packages/patches/python-3.5-fix-tests.patch: Rename to ...
>> * gnu/packages/patches/python-3-fix-tests.patch: ... this. Adjust and disable
>> more tests.
>> * gnu/packages/patches/python-3.5-getentropy-on-old-kernels.patch: Delete file.
>
> I'm curious, do you know if they handled this issue upstream in a
> reliable way, or is getentropy() simply unused in Python 3.6?

The getentropy fix for 3.5 was a port of "random.c" from 3.6 + the fix %
3.5 specific behaviour. Since 3.6.2 was just released it should have
it. Most hunks are there, but didn't verify all.

> Also, do you have any idea about the state of the bytecode timestamp
> issue, which is related to the problem with the bytecode cache? I can
> work on that on python-updates when you push it.

Yes. You might have seen this:

https://github.com/NixOS/nixpkgs/pull/25916

NixOS patches the interpreter to unconditionally set bytecode timestamps
to epoch 1. Now I also found this:

https://github.com/python/cpython/pull/296

So I think we only need that patch and to set PYTHONHASHSEED to some
deterministic value in python-build-system. 

I can give it a go next weekend if you're busy, but more than happy to
share the workload :-)