Fabian Ruffy

> Looks good to me, but why did the tests fail? I believe it's an odd side effect of https://github.com/p4lang/p4c/blob/main/midend/actionSynthesis.cpp#L169 Not sure yet whether that is doing the right thing.

~~It looks like the root cause is the predication pass, which does not preserve source information well. Thankfully, we do not use it in P4Testgen, because it has [problems](https://github.com/p4lang/p4c/issues?q=is%3Aissue+predication+label%3A%22Predication+pass%22).~~ Nevermind,...

@antoninbas Do you perhaps recall why this was done this way?

I think we should stick with whatever definition of uniqueness source info gives us for now. Just to keep things simple. Based on a program coverage view this behavior technically...

>BTW. I've now realized this touches on something more complex still. For example, if there is a sub-control invoked from multiple controls (say in a multi-pipeline architecture, or from both...

> Cache keys are immutable. So with this strategy, our main branch cache entry will be created once and then go more and more stale over time. Are you sure?...

We could adopt the strategy gRPC uses: https://github.com/grpc/grpc/blob/master/.github/workflows/pr-auto-fix.yaml#L24

Could you add a PR with this example program and predication enabled in P4Test? I believe the pass is not used in many back ends because of how bug prone...

Actually, removing https://github.com/p4lang/p4c/blob/main/backends/p4tools/common/CMakeLists.txt#L44 appears to fix this issue. There seem to be some side-effects in the frontend lib... this whole linking business is giving me a headache.

>@fruffy Are there any CONVERT_PRIMITIVE(modify_field) macro usage in downstream? If yes, then you'd need to bump its priority. No, no further usage. Could be a unity build problem?