Figure out why egg-info needs to be included

[eteq]

While working on astropy/astropy#4682, @astrofrog and I were surprised to find that the astropy_helpers.egg-info had to be included in the base package's MANIFEST.in for the helpers offline machinery to work. Without that, a The requested path 'astropy_helpers' for importing astropy_helpers does not exist, or does not contain a copy of the astropy_helpers package. came up and the build fails if you don't have internet (or use --offline). This seems surprising and counter-intuitive.

@embray, do you have any insight about this?

[eatdust]

Related to this, if one removes the egg files, the main astropy package does not compile any more due to that very same error.

Within astropy source file, in: ah_bootstrap.py

def _directory_import(self):

This function is checking if "setup.py" is provided within astropy_helpers/. If not found, you get the error. After adding the setup.py (from astropy_helpers branches) the eggs are correctly generated by the main astropy setup.py and no longer required. So either this function should be corrected accordingly, of the setup.py file provided.

Cheers.

[embray]

Bizarre--I just saw this ticket. I believe @eatdust is on the mark. I think the main problem is that you're excluding the setup.py for astropy_helpers from the source tarball. I would fix that, and remove the needless .egg-info stuff from the MANIFEST.in.

---

There's a lot about astropy_helpers and how it is used that needs to be reworked. Almost all the boilerplate that uses it in Astropy's setup.py should go away, for example. It's a complete mess (though it does look a little better now than it used to). Before I departed (more or less) from the project it was my intention to clean that up but it was "working" so it sort of kept getting pushed down the list. I wonder if I'll ever have time to revisit that...

[pllim]

There were talks to reduce helpers into a single file and get rid of it as a submodule entirely...

[bsipocz]

There were talks to reduce helpers into a single file and get rid of it as a submodule entirely...

@pllim - do you think there would be enough interest to actually get this done as a sprint during the coordination meeting? Would you think we can add it to the provisory agenda and look for reactions?

[astrofrog]

I think it should definitely be a discussion item. Last year we decided to at least break out the sphinx and pytest helpers, so the code base is smaller now (which would make the single-file thing easier)

[epassaro]

Hi @astrofrog @eteq @keflavich. After working a month on our conda-forge package TARDIS, I still I can't make it work when using a GitHub release tar.gz file (no PyPI package).

I'm using the same script command in meta.yaml, and added the same astropy_helpers lines in MANIFEST.in that you are using in Astroquery package.

Do you have any idea why it's not working for us?