Format python traceback in impl Debug for PyErr.

[nicolasavru]

Fixes #4863.

[Icxolu]

Thanks! This looks reasonable to me. Just a quick thought before landing this: Does it make sense to keep the inner Result here, or should we flatten it into the outer Option for simplicity?

[davidhewitt]

I guess if it errors we would try to render the error and maybe get into an infinite descent? I suggest we flatten and use logic similar to the PyAny::display in the case of errors.

[davidhewitt]

Sorry, I meant impl Display for Bound

pyo3/src/instance.rs

Line 484 in 2c732a7

fn python_format(

[nicolasavru]

Done. I thought about using tb.format().map_err() instead of the match but the match is more readable IMO.

This gives output like "traceback: Some(\"<unformattable <traceback object at 0x7f645f7c18c0>>\")".

Is there a good way to test this?

[codspeed-hq]

CodSpeed Performance Report

Merging #4900 will not alter performance

_{Comparing nicolasavru:perr-debug-traceback (9546f5f) with main (5be233c)}

Summary

✅ 88 untouched benchmarks

[LilyFoote]

Can we construct an unraisable traceback in a test to have coverage here?

[nicolasavru]

Don't know, that's what I meant in my last reply with "Is there a good way to test this?". If anyone here knows how I can construct an unformattable traceback, I'm happy to add a test for that.

[nicolasavru]

I could do something like mock out io.StringIO in the python, that should cause the format to fail, not sure if we want to go down that route.

[nicolasavru]

Added a test with that approach.

[Icxolu]

Looks good to me now, thank you very much!