Suppress generator error on panic

[jamesbornholt]

When a thread panics in a Shuttle test, generator produces an error that gets printed to stdout via tracing. This error is a red herring: it mentions generator::gen_impl and so suggests that maybe something went wrong in that code, when really it's just that a Shuttle thread panicked and that panic is being propagated. It would be nice to suppress this error so we don't lead users down the wrong debugging path.

[bkragl]

+1

I have a Shuttle test running inside a proptest, where a failing execution aborts before showing the input used by proptest (and probably even before proptest tries to minimize).

After the stack backtrace the output finishes as follows:

Aug 12 11:16:32.499 ERROR generator::gen_impl: set panicked inside generator    
thread panicked while panicking. aborting.
error: test failed, to rerun pass '-p foo --lib'

Caused by:
  process didn't exit successfully: `/path/to/foo-7e20baac4e3ab4e7 'foo:bar' --exact --nocapture` (signal: 4, SIGILL: illegal instruction)

Will suppressing the generator error resolve the nested panicking and SIGILL?

[jamesbornholt]

No, the generator error is only an error message printed when the Shuttle thread is panicking -- it doesn't actually create its own panic. Nested panics like this usually happen because a Drop handler panics during stack unwinding. We could consider creating an option to disable stack unwinding after a panic (by leaking the generator threads) to make minimization work.