Automagically wrap AbstractOperation and Reduction in Field inside OutputWriters?

[glwagner]

This came up on #2235.

The idea is to support syntax like

outputs = (ζ = ∂x(v) - ∂y(u), U = Average(u, dims=1))
output_writer = JLD2OutputWriter(model, outputs, ... )

rather than requiring things to be wrapped in Field to work.

This wasn't easy to support previously, before we consolidate all the Field implementations (we had ComputedField, Field, AveragedField...). But now we can support it and it would reduce boilerplate (eg everything is Field(this), Field(that) by quite a bit for the majority of use cases.

I think the downside is that users aren't immediately exposed to Field infrastructure. A good "Fields tutorial" in the docs might help with that though.

[tomchor]

This came up on #2235.

The idea is to support syntax like

outputs = (ζ = ∂x(v) - ∂y(u), U = Average(u, dims=1))
output_writer = JLD2OutputWriter(model, outputs, ... )

rather than requiring things to be wrapped in Field to work.

I think this is certainly more intuitive for users!

This wasn't easy to support previously, before we consolidate all the Field implementations (we had ComputedField, Field, AveragedField...). But now we can support it and it would reduce boilerplate (eg everything is Field(this), Field(that) by quite a bit for the majority of use cases.

I think the downside is that users aren't immediately exposed to Field infrastructure. A good "Fields tutorial" in the docs might help with that though.

Pardon if this question is naive but why is it important that users to exposed to the Field infrastructure?

[glwagner]

Pardon if this question is naive but why is it important that users to exposed to the Field infrastructure?

Because that way they are prepared to do sophisticated post-processing! But I think you're right, maybe it's not a real "downside".

[glwagner]

Done