@mhils don't feel obligated to review my code, but my work here reminded me of your comments about project structure on #68.

This PR adds another module (arxiv.query), which is reexported in __init__.py alongside arxiv.arxiv. If you're interested in weighing in, I'd appreciate your feedback; does this feel like an antipattern?

It looks like you want to publicly expose your entire API on the top-level arxiv namespace, but you want to split up the code internally into different submodules ? In that case I personally would do what httpx does: Explicit imports + __all__ + internal modules prefixed with an underscore.

https://github.com/encode/httpx/blob/master/httpx/__init__.py

There really is no super well-established convention here though, it boils down to personal preference a lot. I guess the only thing I would avoid is * imports, it makes it hard for trace where things are actually coming from. For example when I'm interested in the definition of QuantitativeBiology, __init__.py currently has no hint on which submodule I need to look into. Does that make sense?

@mhils yes, that makes sense––thanks for the input!

It looks like you want to publicly expose your entire API on the top-level arxiv namespace, but you want to split up the code internally into different submodules?

Yeah, that's about right! With one complication: if possible, I want to avoid breaking the original import arxiv usage.

If I were writing this package from scratch, I'd use named submodules:

• arxiv.api (the original module, currently named arxiv.arxiv)
• arxiv.query (unchanged)
• arxiv.fixtures module for consts/enums defined by arXiv themselves. This is porbably a better name than arxiv.categories... I'll think about renaming in this PR.

The problem is that the current organization encourages import arxiv, but also permits from arxiv import arxiv.

I'm inclined to only allow explicit imports, but there are two issues:

• import arxiv usage is broken.
• I'd like to rename arxiv.arxiv to arxiv.api to avoid from arxiv import arxiv, but that's a breaking change.

I suppose I could move the API logic into an arxiv.api module; then––to avoid breaking changes––a deprected arxiv.arxiv module can * export arxiv.api, and __init__.py can auto-export arxiv.arxiv.

Oh, looks like I jumped to conclusions! I can turn arxiv.query and arxiv.arxiv internal (arxiv._query and arxiv._arxiv respectively) and have init.py explicitly import certain values from them, like httpx does.

I'll think more about this.

Closing: stale.