[COMPILETEST-UNTANGLE 2/N] Make some compiletest errors/warnings/help more visually obvious

[jieyouxu]

This PR makes some compiletest errors/warnings/help more visually obvious. Note that this is only intended to help visually -- the error handling in compiletest is still a mess.

r? ghost

[Kobzol]

r? @Kobzol

The visual changes look great! Maybe we could add another macro, e.g. fatal!, that will do error! + panic!? It seems better than just using the error macro and then an empty panic afterwards, it looks a bit weird.

(you can r=me on the last commit)

[jieyouxu]

Added a fatal!() macro which is basically just error!() + panic!().

[bjorn3]

std::panic::resume_unwind(Box::new(())) would also work to silently panic (no panic message and no backtrace)

[jieyouxu]

Hm right. Though a backtrace might be useful in this case, mostly for panic location?

[jieyouxu]

(Not that randomly panicking in random places is ideal, but)

[bjorn3]

At least some of the cases of fatal!() are user errors, not bugs in compiletest itself. For user errors backtraces aren't that useful.

[jieyouxu]

Changed to std::panic::resume_unwind(Box::new(())) instead.

[Kobzol]

Backtraces are useful in general to make debugging easier though, even if it's a user error, users might sometimes want to take a look at compiletest's source code to see what exactly is triggering the panic, and what should they change (it might not always be obvious from the error message).

[jieyouxu]

That is true. I think I'll retain the explicit panic for the backtrace location at least for now. Eventually, though, I'd like the user error vs compiletest logic bug to be handled differently.

[jieyouxu]

I'm going with the explicit panic for the short-term.

[jieyouxu]

Changed to std::panic::resume_unwind(Box::new(())).

[jieyouxu]

Going with the explicit panic in fatal!() in the short-term.
@bors r=Kobzol rollup

[bors]

📌 Commit c3fb7d1 has been approved by Kobzol

It is now in the queue for this repository.