Fix MIR check for MIN % -1 in exact_div

[elichai]

Hi,
I hope I understood the bug correctly, because I'm not too familiar with the code here.

I've found this while playing with Miri, it seems like this if condition is no longer executed: https://github.com/rust-lang/rust/blob/master/src/librustc_mir/interpret/intrinsics.rs#L414
Because this commit 28f85c6 made it return 0 in the MIN % -1 case, while the if explicitly tests if it's not 0

So I reverted to the old behavior which returned the left side (MIN) instead of 0.

r? @RalfJung

FWIW, Before this the exact_div4 test in Miri failed and after this it passes

[RalfJung]

Thanks a lot for helping to fix Miri! However, I'm afraid I already took care of this one... #69126

So I reverted to the old behavior which returned the left side (MIN) instead of 0.

The old behavior was wrong, though. int_min % -1 returns 0 when you actually run it.

[elichai]

@RalfJung Ha. That PR wasn't there when I started diving into rustc_mir 😉

int_min % -1 returns 0 when you actually run it.

I think I don't quite understand what the mir::interpret is (ie when exactly does it run and on what) but maybe we can take this convo offline :)

[RalfJung]

I suggest https://rust-lang.github.io/rustc-guide/const-eval.html (and its sub-section) to start with :)