Stacked Borrows beautififcation, update for EscapeToRaw

[RalfJung]

Matches rust-lang/rust#55716 which has landed and should be in the next nightly.

[RalfJung]

I have a blog post already written that I'd like to publish once this patch hits the miri on playground. So @oli-obk it'd be great if we could get this merged tomorrow, and then I will try to get @shepmaster to trigger a redeploy of the miri-on-playground so I don't have to wait another night. ;)

[shepmaster]

I don't have to wait another night

Looks like I'll be the one waiting another night 😝

[RalfJung]

This should work against the currently nightly now, and hence is ready for landing.

[oli-obk]

Travis is unhappy

constant evaluation error: Borrow being accessed (Shr(None)) does not exist on the stack, or is guarded by a barrier

[RalfJung]

Interesting, probably one of those test failures that only happen without full MIR? (The libstd then doesn't have retag/escape-to-raw.)

[RalfJung]

Yeah, that's it. Quite a few tests though. I see two options:

• Move them to fullmir.
• Do not do validation when we have a libstd without full MIR. Though then we should be careful that people do not think validation was checked when truly, it was not. Maybe require a -Zmiri-libstd-without-mir flag to even work without full MIR?

[RalfJung]

A third option would be to make people pass -Zmiri-disable-validation, and do that automatically in the tests.

[oli-obk]

Does the "no mir" case make sense for miri anymore at all? Can we just move to "get a libstd with full mir, sorry"?

[RalfJung]

It still makes some sense without stacked borrows (but there is no switch to disable stacked borrows but keep validation enabled, and too many switches might also be confusing).

Before dropping support for missing MIR, I'd like to have cargo miri setup or so which automatically does all the steps to build the libstd with full MIR. I hope to implement that next week.

[RalfJung]

With these changes, the test suite passes without full MIR locally. Let's see what Travis says.