Deprecate or remove high-level protocols

[ntessore]

I propose we deprecate or remove the high-level protocols that bind together multiple method and attribute protocols.

While the idea is nice in principle, in practice, these groupings of functionality are hard to map to real use-cases of the API.

A much more flexible and powerful approach would be to suggest that libraries using the API define their own high-level protocols in terms of the low-level CosmoAPI protocols. This removes the need to define arbitrary groups in the standard.

It also makes it much simpler to expand the API to perturbations. We do not have to figure out what a Perturbation object would be, we only have to start adding things like HasGrowthFactor.

[nstarman]

I can see the point about StandardCosmology and the high-levels. But the intermediates, like BaryonComponent still seems useful. When you need Omega_b it's common to also need Omega_b0.

[ntessore]

I can see the point about StandardCosmology and the high-levels. But the intermediates, like BaryonComponent still seems useful. When you need Omega_b it's common to also need Omega_b0.

Right, this is something I also thought about a lot. Perhaps xyz0 should just be part of HasXyz? Since any library that does not already provide the constant can simply return their z=0 value?

[nstarman]

Protocols are checked in O(n) time where n is the number of members. So that's a 2x slowdown for the person who doesn't need both. Is it so bad to have the atomic protocols and a set of these useful goupings? We can deprecate things like DistanceMeasures

[ntessore]

I think they revoke your Python license if you put a @runtime_checkable instance check in an inner loop

[ntessore]

Discussing the possibility of adopting this for the new CLOE code with @gcanasherrera today brought up that this also isn't a good idea because of the reverse direction: codes are much more likely to have the constant but not the function.

[nstarman]

So let's keep both the atomic and intermediate Protocols.
High level protocols can be moved to groups and maybe deprecated.

[gcanasherrera]

For the record, another example where most codes depend on constants rather than the function itself is neutrinos. In this case, you typically specify the mass and the number of relativistic species.