compiletest does not support taking the disjunction/union of multiple "only" directives

[RalfJung]

Say I want to run a particular test (or a particular revision) on all x86 targets -- 32bit and 64bit. Currently, there's no good way to do that: only-x86 and only-x86_64 select one of these cases, respectively, but I can't say "either one". If I use multiple only annotations, that takes the intersection, i.e. they all have to apply -- which makes sense as semantics, but is not helpful here. So instead what I usually do is have a separate 32bit and 64bit revision, and make them identical, but that can have significant overhead -- in particular for FileCheck tests where one cannot say "this annotation is active in the following two revisions".

I think the biggest problem here is coming up with a good syntax.^^ Ideally, we'd have a concept of an "only annotation", so e.g. instead of only-x86 we'd use something like only: x86 . Then it would be easy to say that only: a b c means that at least one of the conditions a, b, c must be satisfied.

Cc @jieyouxu
(not sure whom else to ping for compiletest things)

[jieyouxu]

I wouldn't want it to be only only-* (heh), but something like only-target-arch: x86 would be very reasonable to me.

I think this is kinda like a general "revision matrix", where you have two axis: the "normal" revisions axis against target-ptr-width axis or sth

| revision / target-ptr-width | 32 bit | 64 bit |
| --------------------------- | ------ | ------ |
| foo                         | ?      | ?      |
| bar                         | ?      | ?      |

lol. I'll keep this in mind when experimenting w/ compiletest directive handling rework (it's not the first time some kind of "revision matrix" request has come up).

[RalfJung]

It's not specific to the target arch though. compiletest supports a long list of conditions that can be checked (x86 and rustc_abi-x86-sse2 and all the other things that can come after only/ignore), but it is very limited in terms of what one can do with these conditions: we can "ignore if one of N conditions is met" and "run only if all of N conditions are met".

[jieyouxu]

Sure. Something that I've thought about is maybe some kind of testing DSL (like cfg)...

//@ cfg: any(target_arch = "x86", target_arch = "x86_64")
//@ cfg: all(not(target_os = "wasm"), supports_threads, supports_subprocess)

but will need some designing.

[RalfJung]

That's the super complicated version.^^ Might be a bit overkill. ;)