Adrian Sampson

Sounds reasonable! FWIW, some of the decisions that might look a little odd are a result of adherence to the [JSON API](https://jsonapi.org) spec. We used that as a kind of...

Cool! I'd be happy to take a look at that separately too if it would be helpful.

After staring at this for a little while, it seems to me that there is a second big difference between the existing backends and the new LLVM backend: statefulness. In...

OK, after having done my fiddling, I think that's my conclusion for now: let's stick with building a totally separate backend that doesn't reuse much of the existing infrastructure. If...

This is great; thanks! Perhaps the next step would be to group these into rough units that we might work on adding. For example, here are some patterns I see:...

Also, it strikes me after reading this list that it would be nice to support much of these functions on the CPU as well as on the GPU. We already...

Ah, nice work boiling this down to a small example! This is incredibly helpful to look at! It looks like the optimizations run ahead of Diospyros are already smart enough...

Hmm; a crash in `printInstruction` looks suspicious! Probably indicates malformed IR… running the validation pass could help, as could attempting to narrow down the crashing instruction.

Super good idea.

I don't think you're doing anything wrong! We just don't have a very good system for displaying large trees. The CSS/etc. work OK for very small trees where everything happens...