Dmitry Vyukov

Nice! I think in general this is the right approach and it will provide a good base for implementation of other features on top. We also need some user docs...

> What would be some examples that come to your mind? Simple example: ``` foo { a int8 b int32 [if a == 0] c int8 } ``` If b...

> > In particular, I am interested in the alignment details for struct/union fields. This may also need some additional tests for alignment, I see lots of potential for bugs...

> I've run some side-to-side comparisons with the current upstream version and the version from this PR seemed ~20% slower (apparently due to extra computations we now perform). > >...

> Maybe we really don't need that optimization. Though it's really weird to see that these seemingly reasonable changes made the benchmark even slower. Maybe we need a different, better...

Add this to pkg/compiler.TestFuzz ;) ``` type r[N][ R r[t[r[t[N,NN,NN,N,T]],N,N,N,T]] ] n(a r[t]) ```

Do simple Go binaries run on the machine?

I am not sure we have riscv experts here who can answer this without debugging.

It would be good to have a test that ensures that export respects access rules within namespace and don't export bugs from other namespaces.

Yes, this is intentional to get better reproducibility and get more interesting interactions in this range between different mm syscalls and various uses of that memory. Potentially program minimization/simplification during...