MAINT: pin anaconda-client and add pythran build dependency

[tylerjereddy]

Fixes #115

Forward-port the fix that worked for the v1.6.x wheels branch anaconda-client upload issue. Does not affect Azure Pipelines because they use conda to obtain the dependency instead of pip.

• pin the anaconda-client commit hash used for
  Travis and Appveyor pip installs to avoid a
  problematic dependency addition to requirements.txt

[tylerjereddy]

oh right, master branch of the wheels repo is currently a mess because pythran, our new dependency in main repo master, is not installed. I'll probably try to resolve that in this same PR.

[tylerjereddy]

@serge-sans-paille @rgommers revising to use latest stable release of pythran for SciPy wheel builds; my reasoning is that this is the same specification used in main repo pyproject.toml at the moment

[rgommers]

yep that's the right thing to do, thanks @tylerjereddy

[tylerjereddy]

Still needs work; for old manylinux1 is there a plan for gcc: error: unrecognized command line option -std=c++14? Just going to bump up manylinux version moving forward with 1.7.0?

Windows CI fail is also showing Pythran using C++ features not available in the current build environment there.

Travis aarch64 build has some test failures--those may have simply slipped through as main repo no long probes that arch.

[mattip]

I think it is time to abandon manylinux1. The distribution it is based on is EOL. The problem is that pip<19 does not recognize manylinux2010, so people using old pip versions will do pip install scipy, pip will fail to find a binary wheel, it will download the sdist sources and try to build scipy from source. Any python>=3.7.3 ships with pip>=19. The question is how many people are using python<3.7.3 and not upgrading their pip?

[rgommers]

I think it is time to abandon manylinux1. The distribution it is based on is EOL.

I agree, it's time.

The question is how many people are using python<3.7.3 and not upgrading their pip?

Probably not that many. Python 3.7.3 is from March 2019, so if people have managed to install with an old pip into such an environment for more than 2 years, it's time for them to clean up or run pip install -U pip.

[rgommers]

Travis aarch64 build has some test failures--those may have simply slipped through as main repo no long probes that arch.

Yes that looks pretty harmless, I'll look at fixing that now.

This does fix a couple of jobs and looks good, so let's merge this and continue in a next PR. Thanks @tylerjereddy and @mattip.