Hi,
Il 21/04/22 20:58, Zebediah Figura ha scritto:
The intent is that they contain most code corresponding to user actions, and as such do nontrivial interpretation and validation. In the case where we're manually synthesizing instructions I don't want to have to think about whether such functions are doing something we don't want.
I can probably live with it in this case, but I'm concerned that this mental burden will grow.
Maybe there's an argument for helpers which just wrap hlsl_new_* like that, and we can have an add_* family and a parse_* family.
Ok, I can see an argument for having another abstraction layer as you say, but I'd defer this to some later patch, so that this doesn't scope creep too much.
As for all the other comments, they should be addressed in the next series.
Giovanni.