compiletest: be more robust for cross-compile scenarios with dual under-test toolchains

[jieyouxu]

compiletest isn't super robust about distinguishing between configuration and environment for the host versus the target-under-test (for cross-compile scenarios). At places, we sometimes don't bother distinguishing the two for "good enough" cases and when host-only testing is the case where we can make things work.

Furthermore, we do not distinguish between the case where you may actually have two "target" toolchains for the purpose of testing wasm proc macros: there is a under-test target toolchain for wasm proc-macros and there's a under-test target toolchain for $other_things

graph TD
    A["host_target(x86_64-unknown-linux-gnu)"] --> B["target_a(wasm32-wasip2)"]
    A --> C["target_b(i686-unknown-linux-gnu)"]

Loading

[jieyouxu]

@bjorn3 I'm not mistaken with how wasm proc-macros are intended to be tested here right? Under the most "flexible" arrangement of being under cross-compile, but having a different "target-under-test" toolchain for the wasm proc-macros versus the whatever_other_target toolchain? IOW 3 toolchains may be potentially involved?

[bjorn3]

Yeah, there would be 3 targets involved. The host for non-wasm proc-macros and rmake files. wasm32-wasip* for wasm proc-macros and whatever is the target that the final test executable/library should be compiled for.