[cargo-miri] Interpret unit tests of proc-macro crates

[ghost]

This PR is the successor of #1674.

Will fix #1660.

[RalfJung]

This PR attempts to do that by using cargo metadata to detect the unit tests and calling cargo test again to interpret them in Miri.

Hm, that's not what I had in mind... I thought we could detect this by looking at the command-line arguments cargo passes to us.

I will take a closer look at this, but I don't have a lot of spare time currently, so I cannot dig into this myself or provide an in-depth review currently -- sorry for that.

[ghost]

Hm, that's not what I had in mind... I thought we could detect this by looking at the command-line arguments cargo passes to us.

I will take a closer look at this, but I don't have a lot of spare time currently, so I cannot dig into this myself or provide an in-depth review currently -- sorry for that.

It's okay. Meanwhile I'm going to try if I can continue #1674 (that's "looking at the command-line arguments cargo passes to us"). Thanks for taking a look!

[ghost]

I thought we could detect this by looking at the command-line arguments cargo passes to us.

I tried to continue #1674 here and stopped using cargo metadata.
#1674 failed with found possibly newer version of crate `std` which `num_cpus` depends on, so I tried to set --sysroot, but I'm getting an ICE (compiler/rustc_mir/src/monomorphize/collector.rs:826:9: no MIR available for DefId(2:6849 ~ core[25bd]::fmt::num::DEC_DIGITS_LUT)) now.

[RalfJung]

no MIR available

This sounds like it is using the host rustc sysroot when it needs the Miri sysroot.
So are you trying to run the test in Miri? Looks like that won't work.

[ghost]

This sounds like it is using the host rustc sysroot when it needs the Miri sysroot.

It's strange that I explicitly set --sysroot to the Miri sysroot and that was even shown up in the verbose output. I must miss something. 😕

[RalfJung]

I don't understand your goal here. Didn't we say that these crates (proc-macro tests) would not be interpreted with Miri, and that they would not be run at all?

[ghost]

@RalfJung That's not what I was trying to do here. I updated #1675 and did that there.

[ghost]

I don't understand your goal here. Didn't we say that these crates (proc-macro tests) would not be interpreted with Miri, and that they would not be run at all?

@RalfJung I meant I had not given up this PR yet and feel free to close this PR (next week or more later when you have time, good luck to your current work) if you think this approach is undesirable or unlikely to succeed.

[RalfJung]

I think #1675 is more promising, so given that this here also doesn't seem to work (if I understood you correctly, and judging from the CI status), let's close this and focus on the other PR.