Fix TaggableCLArray copy without queue

[alexfikl]

No description provided.

[inducer]

So I think the underlying issue here is that there's confusion on what copy should do. (Which I should have caught before merging #111, sorry!)

• Taggable assumes copy can be used for (cheap-ish, immutable) return-with-modifications.
• cla.Array says copy copies array data.

IMO:

• It's not possible to cleanly untangle those two use cases without causing un-obvious corner cases.
• TaggableCLArray should not override copy, and Taggable should not assume copy is used to apply tags. The latter makes sense for another reason: If the Taggable is a dataclass, then the correct spelling would be something with replace.

So my vote would be to give Taggable an interface (_with_tags) that a subclass needs to define, with a deprecated fallback to using copy.

What do the both of you think?

[alexfikl]

What do the both of you think?

I think that's mostly what the current version does, except that copy still tries to handle tags. I can remove that once we agree on an approach.

For me this is 👍.

EDIT: Not super worried about deprecating things though, since this was just introduced and nobody uses it outside arraycontext (right?).

[inducer]

inducer/pytools#120 realizes part of this.

EDIT: Not super worried about deprecating things though, since this was just introduced and nobody uses it outside arraycontext (right?).

I'm talking about changing Taggable, and that has uses throughout pytato, here, and loopy.

[alexfikl]

I'm talking about changing Taggable, and that has uses throughout pytato, here, and loopy.

Ah! I read that as a shorthand for TaggableCLArray and got confused. Then yeah, completely agree that needs a bit more care. Thanks for taking care of the pytools side!

[alexfikl]

@inducer This needs a pytools release that includes inducer/pytools#120. Otherwise it should be good for another look!

[inducer]

https://pypi.org/project/pytools/2022.1.1/

[inducer]

Thx! Pushed a pytools version constraint to (hopefully) make mirgecom happy, and underscored unwrap_cl_array. LGTM, in it goes, pending CI. :)

[alexfikl]

Thx! Pushed a pytools version constraint to (hopefully) make mirgecom happy, and underscored unwrap_cl_array. LGTM, in it goes, pending CI. :)

Looks like it picked it up! Thank you.