Integer::new(std::i64::MIN) panics

[turboladen]

I feel like I must be overlooking something, but I ran across this while writing some tests. I tried whittling it down to find the breaking point; some of these panic (signal: 11, SIGSEGV: invalid memory reference); some don't:

// Panic
let i = std::i64::MIN;  // -9223372036854775808
let _ = Integer::new(i);

// Panic
let i = std::i64::MIN + std::u32::MAX as i64; // -9223372032559808513
let _ = Integer::new(i);

// Panic
let i = std::i64::MAX;  // 9223372036854775807
let _ = Integer::new(i);

// Panic
let i = std::i64::MAX - 1;
let _ = Integer::new(i);

// Ok
let i = (std::i32::MAX as i64) + 1; // 2147483648
let _ = Integer::new(i);

// Panic
let i = (std::i32::MIN as i64).pow(2); // 4611686018427387904
let _ = Integer::new(i);

// Ok! Looks like this is a threshold on the positive end.
let i = (std::i32::MIN as i64).pow(2) -  1;
let _ = Integer::new(i);

// Ok!
let i = (std::i32::MIN as i64).pow(2) * -1; // -4611686018427387904
let _ = Integer::new(i);

// Panic; looks like the -1 here is where it fails, at least on the negative end
let i = (std::i32::MIN as i64).pow(2) * -1 - 1;  // -4611686018427387905
let _ = Integer::new(i);

Ruby version: ruby 2.6.2p47 (2019-03-13 revision 67232) [x86_64-darwin18] Rust version: rustc 1.39.0 (4560ea788 2019-11-04)

If this is a Ruby limitation (only handling values in between 32- and 64-bit int types), I'm not sure what a reasonable solution would be, but it at least seems worth documenting.

[danielpclark]

I've tested each of these on Linux and they all work without issue. I've added tests to the main code base to see how they'll do across operating systems.

[danielpclark]

The tests pass CI for Mac. I is likely a system architecture and customized build of Ruby issue.

[turboladen]

FWIW, it was a standard install of Ruby, using chruby, nothing fancy.

[danielpclark]

@turboladen I've noticed with rbenv on Mac it tends to have a statically linked Ruby built rather than a dynamically linked one. That causes its own issues. I've not heard any other reports with chruby, and I haven't tried it yet.

That may be something I should add to the test suite… using different Ruby version managers.

[turboladen]

Oh ya, good point—I usually forget that (chruby behaves the same) and have to reinstall Ruby with the —enable-shared (I think that’s the one at least) flag. ...which I suppose negates my “nothing fancy” comment, since it could be picking up anything weird from my env.