Debug trait somehow getting lost (?) making debugging difficult

[burntsushi]

Debug trait somehow getting lost (?) making debugging difficult

[BurntSushi]

That error happens when the value that is panicked doesn't implement the Debug trait. I'm not sure what exactly is causing the panic, but you have a few options:

1. QuickCheck is actually giving you more information. In particular, it's telling your the arguments used that caused the failure (this is the "witness"): Arguments: ([0]).
2. You may have more luck by setting the environment variable RUST_BACKTRACE=1 when running cargo test. The key will be to look for the backtraces produced as part of the panic while tests are running. Right now, they appear as this in your build output: thread 'safe' panicked at 'calledOption::unwrap()on aNonevalue', ../src/libcore/option.rs:362.

[kazagistar]

I ended up finding the bug with the backtrace option (thanks!), but I am still unsure why Quickcheck was not able to print the value. The value that panicked was an Option<i32>... however, in the context, it was generic (A value of Option<T> where T: Ord + Clone + Debug + Arbitrary). Either way, thanks for the help.

[BurntSushi]

Strange. This is the relevant code: https://github.com/BurntSushi/quickcheck/blob/master/src/tester.rs#L409-L419

I'll leave this open. Ideally, we could get a smaller test case that is easier to debug.

[bluss]

match any_err.downcast_ref::<&Debug>() is very “optimistic”. Downcasting an Any is only possible to the exact type that was packaged up into the Any.

[BurntSushi]

Right, it basically never works. Are we SOL here? Or is there some other way to show the panic? If most panics are strings, then maybe we should try &str and String too.

[huonw]

I think it's basically impossible to literally satisfy &Debug (that would require panicking with a &'static Debug trait object), so it just make sense to try &str and String, which should cover typical panics, including those from the assert* macros. I guess it doesn't cost much to also check fancier things like Box<Debug> but they seem very unlikely to be triggered.

[yggie]

Any updates on this issue? I am running into lots of "UNABLE TO SHOW RESULT OF PANIC." messages and it is hiding the cause of the panics in my code.

[bluss]

Sure, I have posted a PR to fix. Depends on the resolution of shrinking prs first though.

[bluss]

@yggie Do you have interesting nonstandard panics? My fix only handles &str and String cases, which should be all assertions and all panics that use strings as the message.

[yggie]

@bluss I don’t do anything special with panics, I send &str most of the time, so that would solve my problem!