Fabian Ruffy

Yeah we should add some compile-only tests. Please take a look at the BMv2 PSA implementation and the way it sets up compilation. https://github.com/p4lang/p4c/blob/main/backends/bmv2/CMakeLists.txt#L222 We do not need the check...

What is the purpose of the program structure classes if the generators are derived from them? Is it just separation of concerns. For moving things forward, it seems reasonable to...

A clarification question, what do you mean with consistent? In general, the pointers should be consistent for a single IR tree. But it often happens that the entire program is...

> The sanitizer failure is very plausible since the constructor for `MoveInitializers` leaves the `loopsBackToStart` boolean uninitialized, but it is not particularly likely that this PR could itself contain a...

> The only place I can reproduce this error is inside a docker with a locally-compiled GCC 9.5 running on CentOS 7. So, the link problem is almost certainly a...

> Hi @fruffy, yes I do think `CMAKE_INTERPROCEDURAL_OPTIMIZATION` is the canonical way to do this. If so, could you replace ENABLE_LTO with it? And for backward compatibility use `ENABLE_LTO` to...

> @vlstill can we merge this? It still needs a rebase and @vlstill needs to approve the PR.

@vlstill Could you give this a renewed review?

Thanks for investigating these issues. Unfortunately, it looks like the eBPF-PSA back end is currently not maintained. It might take a while until this issue is fixed.

> > Are the test failures unrelated to this PR? Other than that, the changes LGTM. > > I can't see any connection between the failure and this change --...