Make yt.units a shim that wraps unyt

[ngoldbaum]

This issue tracks the task of replacing yt.units with unyt: https://github.com/yt-project/unyt

There are some differences in unyt with respect to yt.units so we will need to be careful to make sure that everything works the same after integrating. We also need to check the codebase to see if there are places where we can clean things up. These differences are summarized below:

• unyt allows you to do e.g. data_with_units + data_without_units*data_with_units.units, which obviates the need to use unit_quantity or uq.
• SI <-> MKS Electromagnetic conversions do not need to use an equivalence
• Results of operations that have complex units that have dimensions equal to a known unit name in the unit system being used are reduced to an equivalent unit in that unit system. For example, if something comes out of an operation with units like erg * cm/km, by default the result will be presented in erg (e.g. the cm/km will be converted to a constant that gets multiplied by the value of the object).
• MKS is the default unit system
• If you do unyt_array([1, 2, 3], 'code_length') you don't get a yt-specific error message suggesting to use ds.arr like if you do YTArray([1, 2, 3], 'code_length').

[chummels]

Consistent with our discussion two weeks ago on a PR in unyt: yt-project/unyt#2 , Will you make it so the astrophysics context stuff defaults to cgs? I think you're going to find a lot of the current users upset if all of a sudden their projections are in kg/m^2 and so forth.

[ngoldbaum]

Yes, that's why I included "MKS is the default unit system" in the bullet points.

[chummels]

I saw that, but I guess we're talking past each other. So your bullet points list the changes in unyt that we need to effectively cover up so that the yt experience is the same for users?

[jzuhone]

@chummels my view is that the experience from the point of view of yt is identical--the current default unit system in yt is cgs, so it should be that way when we use unyt as a backend.

[jzuhone]

Though I should say that I consider the first three bullet points enhancements and yt 4.0 should have those changes.

[ngoldbaum]

Yup, exactly, although with the qualification that data from fields that we don't support yet might load data with MKS or other unit systems that are commonly used in that field. The astrophysics domain context will definitely have CGS units by default.

[chummels]

OK cool, then we're all on the same page. Sorry for my mis-reading of your description.

[munkm]

I think this is is satisfied now, so I'm going to close it. Feel free to reopen if you all think that yt.units still hasn't outsourced enough to unyt.