remove -Zmiri-panic-on-unsupported flag

[RalfJung]

Fixes #3952, see that issue for discussion.

[saethlin]

nextest is the better option for "continue running other tests if one test failed"

Nextest does not and cannot support doctests, so this would mean we don't have a way to continue running if one doctest fails.

[RalfJung]

Cargo/rustdoc already supports continuing if one doctest fails, I think?

It also runs all doctests concurrently. That's why in miri-test-libstd, doctests are so much faster than unit/integration tests. ;)

[saethlin]

Ah of course it does, this is probably the same reason to get a single-threaded doctest run you need -j1 plus --test-threads=1.

[saethlin]

So if all use of this flag can be replaced by nextest, why should we keep it at all? I don't think I've seen this actually happen, but injecting new unwinding sites into a bunch of unsafe code may cause UB that's only there because of unwinding that can only happen with Miri.

[RalfJung]

Yeah I agree, IMO we should remove the flag. That requires a bit of discussion, so I opened an issue: #3952.

However if possible I'd like to land this PR in the mean time to remove the foot-gun. I doubt unsupported errors due to syscall are super common so I would hope this is an acceptable reduction in functionality -- but OTOH maybe @landaire specifically needed it for syscall and that's why they added it in this function...

[RalfJung]

All right, I now made this just entirely remove the flag. @rust-lang/miri any objections?

[saethlin]

No objections.

[oli-obk]

@bors r=RalfJung,saethlin,oli-obk

[bors]

📌 Commit 2b3deec has been approved by RalfJung,saethlin,oli-obk

It is now in the queue for this repository.

[bors]

⌛ Testing commit 2b3deec with merge bdc100f...

[bors]

☀️ Test successful - checks-actions
Approved by: RalfJung,saethlin,oli-obk
Pushing bdc100f to master...