fix: raise when detecting multiple entry points

[ltalirz]

Due to a slight misunderstanding of the new importlib.metadata API,
AiiDA did no longer raise when detecting multiple entries for a given
entry point (pair of group + name), and instead simply loaded the first
one.

Example illustrating the behavior:

In [8]: eps.select(group='aiida.workflows',name='mcscf.abcChain')
Out[8]:
[EntryPoint(name='mcscf.abcChain', value='aiida_mcscf.workflows:abcChain', group='aiida.workflows'),
 EntryPoint(name='mcscf.abcChain', value='aiida_mcscf.workflows:AbcDefChain', group='aiida.workflows')]

In [9]: result=eps.select(group='aiida.workflows',name='mcscf.abcChain')

In [10]: result.names
Out[10]: {'mcscf.abcChain'}

In [11]: len(result)
Out[11]: 2

In [12]: len(result.names)
Out[12]: 1

In [13]: result['mcscf.abcChain']
Out[13]: EntryPoint(name='mcscf.abcChain', value='aiida_mcscf.workflows:abcChain', group='aiida.workflows')

Fixes #5530

[sphuber]

Thanks @ltalirz . The fix looks fine to me. I have just taken the liberty to add a test. Hope you don't mind. Let me know if you are happy with the test and then I will accept and merge this.

[ltalirz]

Thanks @sphuber, both for the review and for the test!

[sphuber]

Just rereading through your original description in the issue, are you sure that this actually fixes the problem you describe of AiiDAlab where a user installs another version of AiiDA in user space overriding one in global space?

You gave the example exception:

ValueError: aiida.groups:core is not a valid entry point string: Multiple entry points 'core' found in group 'aiida.groups': [EntryPoint(name='core', value='aiida.orm.groups:Group', group='aiida.groups'), EntryPoint(name='core', value='aiida.orm.groups:Group', group='aiida.groups')]

But there the entry points are exactly identical. So even with this fix, this would still raise a MultipleEntryPointError right? You said yourself though that there is nothing we can do here, which I agree with. But would that mean it still poses problems for AiiDAlab?

[ltalirz]

But there the entry points are exactly identical. So even with this fix, this would still raise a MultipleEntryPointError right?

Erm... didn't you just add a test that proves the opposite?

As long as the entry points are exactly identical, it should not raise - which is what we want.

[sphuber]

Of course 🤦 scratch that, not sure what came over me. But, I am a bit surprised that those entry points would end up in the same environment. Is AiiDAlab using virtual envs? Or was aiida-core installed twice, once in system space and one in user space? Not sure how and where the entry points are stored and read from in that case.

[ltalirz]

P.S. And the fix did address the issue for me in the AiiDAlab.

This was only part of the goal of this fix, however - the other was indeed to raise when there are multiple entries for a given entry point that are not identical.

This is how I initially stumbled upon this bug - I was trying to change the value of an entry point but it was not reflected in the one loaded by AiiDA. It took me a while to figure out that apparently some other distribution was adding an entry point with the same name and a different value.

[ltalirz]

But, I am a bit surprised that those entry points would end up in the same environment. Is AiiDAlab using virtual envs? Or was aiida-core installed twice, once in system space and one in user space?

The AiiDAlab uses a conda environment (managed by root), where aiida-core is installed.
This environment cannot be modified by the user - but they can still install their own version (which pip will end up installing with --user automatically), which is often useful for development.

Not sure how and where the entry points are stored and read from in that case.

I didn't look into it further, but I guess it makes sense that both are added to the same entry point map.

[sphuber]

Thanks @ltalirz