On Tue, Feb 27, 2024 at 6:48 PM Conor Dooley <conor@xxxxxxxxxx> wrote: > > No, I didn't actually try anything. Between trying to test the kcfi > stuff on arm64 and the other work I was doing today I did not have a > chance to actually play with that yet. Apologies, I didn't mean it that way -- I was just wondering if it didn't work. No rush on my side (in fact, I thought this was for 6.10 anyway). > It comes down to you though I > suppose - would you rather have generate_rust_target enable the > compressed instructions depending on the config option or have the > Makefile disable it if compressed instructions are not enabled and use > a builtin target? So the custom target support is there for flexibility purposes: we were told is that since the target spec is too tied to LLVM, it is unlikely to get stabilized (at least as it is), and thus they preferred that we ask for whatever flags in `rustc` would be needed to tweak things in an existing builtin target (or add new built-in targets if needed). Thus, when a new architecture is added, the question is whether one can already use the flags approach or not. For instance, to disable the compressed instructions, from what I can tell, the flag I mentioned seems to work, so that is fine. However, for something like tweaking the data layout for `n64` instead of `n32:64`, I am not aware of a way to do so via a flag (but I see newer LLVM uses `n32:64`, so that may be the better one anyway: https://godbolt.org/z/Eh4cfdeMr). So it all depends on whether you are happy with what the flags approach already give you. I hope that clarifies a bit! Cheers, Miguel
