C++: Implement proper coroutine support in IR

[MathiasVP]

This PR fixes the IR generation for co_await and co_yield. We added IR support for these in #16187. However, we didn't quite capture the right semantics of the constructs. This PR fixes this by generating the (hopefully) correct semantics by making use of the children of co_await and co_yield as emitted by our frontend.

This gets rid of all the dead ends in the IR. I haven't quite figured out why we get new nonUniqueIRVariable inconsistencies, though (and whether this is a QL or extractor bug).

[MathiasVP]

This was a drive-by fix I did while debugging. This tag didn't have a result previously.

[jketema]

This gets rid of all the dead ends in the IR. I haven't quite figured out why we get new nonUniqueIRVariable inconsistencies, though (and whether this is a QL or extractor bug).

I think this is related to the issue I reported to our frontend provider (but let me double check that).

[MathiasVP]

I think this is related to the issue I reported to our frontend provider (but let me double check that).

Awesome. That will save me half a morning of debugging then 🎉

[jketema]

I think this is related to the issue I reported to our frontend provider (but let me double check that).

Awesome. That will save me half a morning of debugging then 🎉

Yeah, those are exactly the local variables that we filed an issue for with our frontend provider.

If these are not causing any serious problems, I propose with just add a comment to the open issue on this that once we receive a patch we should also fix this (if that's not automatic).

[jketema]

One question: how do the translations of the reuse expression look (they are a bit hard to find)? In the intermediate state we had before this PR, the use and reuse we in separate blocks. Is that solved now?

[jketema]

One question: how do the translations of the reuse expression look (they are a bit hard to find)? In the intermediate state we had before this PR, the use and reuse we in separate blocks. Is that solved now?

Ah, they still seem to be in different blocks, but the order of the blocks seems such that we get something that is initialised.

[MathiasVP]

Ah, they still seem to be in different blocks, but the order of the blocks seems such that we get something that is initialised.

Indeed. It seems like the frontend is doing sensible things here 🎉

[MathiasVP]

If these are not causing any serious problems, I propose with just add a comment to the open issue on this that once we receive a patch we should also fix this (if that's not automatic).

I would expect this to be fixed automagically, but let's note it in the internal issue anyway 👍

[jketema]

If these are not causing any serious problems, I propose with just add a comment to the open issue on this that once we receive a patch we should also fix this (if that's not automatic).

I would expect this to be fixed automagically, but let's note it in the internal issue anyway 👍

Added.