Refactor the Orbital code and fix bugs

[giovannipizzi]

The code was not tested (and now we add tests)
Moreover, the code required quite some cleanup and
refactoring.
Now the code should be cleaner (even if probably still
not perfect). The intended aim of this commit is to avoid
breaking backward-compatibility.

This fixes #2419

[giovannipizzi]

Marked as [WIP] because I'd like @DanielMarchand to test if now this works (you can check out the branch in my fork) and if I didn't change the API.

Also, @sphuber: I apparently get some weird errors from prospector in my tests (at least In my previous runs, like cyclic imports etc.) - is this related to the caching of pip or to something else?

[coveralls]

Coverage decreased (-5.4%) to 65.057% when pulling de8ccf0f3ea08f82f540441aefd67102c90ff886 on giovannipizzi:fix_2419_realhydrogen_orbitals into 899d803 on aiidateam:develop.

[coveralls]

Coverage decreased (-5.4%) to 65.057% when pulling de8ccf0f3ea08f82f540441aefd67102c90ff886 on giovannipizzi:fix_2419_realhydrogen_orbitals into 899d803 on aiidateam:develop.

[coveralls]

Coverage increased (+0.4%) to 70.919% when pulling 85442e8 on giovannipizzi:fix_2419_realhydrogen_orbitals into d0ab2b8 on aiidateam:develop.

[DanielMarchand]

Marked as [WIP] because I'd like @DanielMarchand to test if now this works (you can check out the branch in my fork) and if I didn't change the API.

I have just tested and the import works now! Thanks!!

[giovannipizzi]

@sphuber I fixed your comments and I also added a couple more tests - I think this is ready to be merged (if tests pass)

[sphuber]

@giovannipizzi the linter is known to be a bit buggy when it comes to cyclic imports. It is a bit weird that touching other files seems to trigger it. You could just try to add the disable statement in one of the files that it highlights as problematic. The only thing I can see quickly is that aiida.engine.processes.ports import validate_link_label from aiida.orm.utils.links and in that same module, LinkManager.nested calls from aiida.engine.processes.ports import PORT_NAMESPACE_SEPARATOR. I thought that if these things are within the scope of methods they should be lazily called and so fine.

[giovannipizzi]

The thing that is actually needed is PORT_NAMESPACE_SEPARATOR. Can I move it one level up, in aiida/engine/processes/__init__.py? Anyway it's used also in aiida/engine/processes/process.py so it seems to me that it is a more appropriate place, and this would avoid the cyclic import?
Should I move also PORT_NAME_MAX_CONSECUTIVE_UNDERSCORES?

[sphuber]

The thing that is actually needed is PORT_NAMESPACE_SEPARATOR. Can I move it one level up, in aiida/engine/processes/__init__.py? Anyway it's used also in aiida/engine/processes/process.py so it seems to me that it is a more appropriate place, and this would avoid the cyclic import?
Should I move also PORT_NAME_MAX_CONSECUTIVE_UNDERSCORES?

👍 Seems fair enough

[giovannipizzi]

I added a commit that should solve the cyclic import problem - to avoid other analogous problems, the right approach seems to be to define the constants in a submodule constants.py and then expose the relevant values one level up.

[giovannipizzi]

py3.6 linter is pickier... and the problem got worse!
https://travis-ci.org/giovannipizzi/aiida_core/jobs/518832224#L1541
Any further suggestions?

[sphuber]

@giovannipizzi I ended up fixing the problem in a different way, no more cyclic imports

[sphuber]

thanks a lot @giovannipizzi

[sphuber]

Thanks a lot @DanielMarchand