Should we build noarch Python packages?

[jeanconn]

https://www.anaconda.com/blog/developer-blog/condas-new-noarch-packages/

[jeanconn]

While this increases complexity, I think it might save enough on maintenance to be worth it.

[taldcroft]

+100. Most of our packages should be noarch so this is a huge win!

[jeanconn]

The way it is described here, I don't know if even our more architecture-specific packages could work via this process (I'm thinking Chandra.Time or xija). Will try some cross-arch testing.

[taldcroft]

Can't imagine how that could work for Windows for those...

[jeanconn]

Unless we have a Windows build/test process/machine, I'm not going to have much of a clue about Windows. I think this flag will be relatively easy to turn off (and go back to building individual arch packages) if we need to provide recipes that will work to build from Windows.

[jeanconn]

Ah good, when I tried to build Chandra.Time noarch I got a very reasonable error. So I should be able to see which are noarch-possible on build:

[noarch_python] Error: Binary library or executable found: lib/python3.6/site-packages/Chandra.Time-3.20.1-py3.6-linux-x86_64.egg/Chandra/Time/_axTime3.cpython-36m-x86_64-linux-gnu.so

[taldcroft]

Closed by #7