Abort in "assert_*" intrinsics

[RalfJung]

These three intrinsics should, IMO, abort instead of panic. The main reason we make them panic in rustc is that we want a good backtrace, and doing that from codegen is hard. But causing a panic is easy thanks to libcore having panic support, and std hooking into it.

But in Miri, we can show great backtraces on abort! So, there is no reason to use panics IMO. So I propose that we extend this with an abort message:

miri/src/eval.rs

Line 56 in 704228d

Abort,

and then cause an abort from the intrinsic.

@elichai that they would look into this.

[eddyb]

Might make more sense to name the intrinsics e.g assert_inhabited instead of panic_if_uninhabited.
That would also make aborting instead of panicking in miri less odd-looking.

[elichai]

Had a medical procedure, so sadly I'll only be able to work on it in a few days

[RalfJung]

No worries, this is not urgent at all. :) Get well soon!

[elichai]

How should I modify these tests to capture the aborts?
Should I make a separate .rs file for each? :/
Should I write the code into a new .rs file, and compile+run it in a new process? something else?

(maybe I can "patch" the abort function so I can catch it? but it might defeat the purpose)

We could also fork, but that will be unix specific

Another option would be to hook on SIGABRT and loop there, and launch a separate thread per function so we can catch all aborts like that without the process being terminated something like: https://play.rust-lang.org/?version=stable&mode=debug&edition=2018&gist=4363a89d68c089f122fd06cd37598a75

just remembered miri doesn't support threading yet

[RalfJung]

Just remove these things from the catch test. And then add one or two new compile-fail tests instead. See here for an example of another test that aborts.