[RFC][ODS] Attribute and Type Assembly Formats

[MaheshRavishankar]

This is great! I was just going to write a parse/print method for an AttrDef. Looking forward to this landing

[stellaraccident]

@mehdi_amini

I think what should be possible (and that would help Python bindings as well) is for all these attributes to implement (optionally maybe) a “dictionnary” interface that would allow to lookup fields by name.
I’m not super clear on this though, because we need some type erasure on the value for each field. One way is to wrap this in attribute but that seems costly (and we also now need to have a mechanism to “raise” each field value into its own attribute on demand).

If we can work this out, it would be very much appreciated. General +1 on using interfaces like this to enable reflective access and reduce bespoke bindings.

[stellaraccident]

@mehdi_amini

I was thinking about this last night, and maybe that for inter-op purpose, it would be good enough to have an “DictionnaryCompatibleInterface” of some sort that would have the generated Attribute have to expose a static failureOr<MyAttr> MyAttr::getFromDict(DictionnaryAttr); and DictionnaryAttr MyAttr::toDictionnaryAttr() .

Any thoughts on going the other way (generic construction)?

[mehdi_amini]

@stellaraccident

Any thoughts on going the other way (generic construction)?

I don’t quite get what you mean here?

[stellaraccident]

Ah, sorry - I read the last part DictionnaryAttr MyAttr::toDictionnaryAttr() and missed the first part static failureOr<MyAttr> MyAttr::getFromDict(DictionnaryAttr); which addresses my question (my eyes have trouble reading past color changes). It is still in the realm of what we really need CAPI code generation for to systematize - but it is closer to being usably generic.