fix make check-stage1

[arielb1]

r? @alexcrichton

[alexcrichton]

As I've mentioned on #27417 let's just take the route of removing -Z no-landing-pads for now, it's by far the easiest and doesn't require separate code paths. Requiring a stage2 compiletest requires compiling the entire stage1 compiler before check-stage1-rpass can be run which is pretty unfortunate. This PR also uses features like #[thread_local] (not portable) and rt::unwind::register (super unstable) which I would prefer to avoid.

[arielb1]

separate code paths? what are you talking about? I have 1 code-path. It would be better to have a sane panic-control API rather than the crazy interdiction thing I am doing, but the current implementation is also rather ugly.

stage2 compiletest only depends on stage2 libstd etc. (aka make check-stage1), and you need stage2 libstd anyway for doing anything. It does not depend on a stage2 compiler (libsyntax, librustc, etc.) - that's the expensive part.

The moment we get a better panic-control API I will switch to it. Maybe even remove the thread-local - it is not strictly needed.

It is very important to be able to react sanely to panics. Printing an ugly error to stderr and unwinding is not necessarily a sane reaction. Rust will be pretty much unusable if there is no such way.

[arielb1]

r? @nrc

[nrc]

r? @alexcrichton

kicking back to acrichto - I'm not familiar with the background here. Also cc @rust-lang/tools.

[alexcrichton]

Closing due to inactivity, and given the discussion on the discuss thread I've opened #28710 to track this.