gh-89770: Implement PEP-678 - Exception notes

[iritkatriel]

No description provided.

[bedevere-bot]

🤖 New build scheduled with the buildbot fleet by @iritkatriel for commit 1555087 🤖

If you want to schedule another build, you need to add the ":hammer: test-with-buildbots" label again.

[gvanrossum]

Where's the descriptor for "__notes__"?

[gvanrossum]

"Errors should never pass silently."

[gvanrossum]

Yeah, I like this!

[iritkatriel]

How about error handling when printing the traceback? Should we print something like "cannot print notes: not a sequence", and similarly if a note is not a str?

[gvanrossum]

How about error handling when printing the traceback? Should we print something like "cannot print notes: not a sequence", and similarly if a note is not a str?

How about printing repr(self.__notes__) if it's not a list, and str(note) if an individual note isn't a string? The former makes it clear what it is, while the latter provides a feature that Barry has been asking for...

If either of those fails, either raise an exception from print_exception() or swallow the error and print a simple message. It probably doesn't really matter, repr() or str() failing should be extremely obscure.

[iritkatriel]

How about error handling when printing the traceback? Should we print something like "cannot print notes: not a sequence", and similarly if a note is not a str?

How about printing repr(self.__notes__) if it's not a list, and str(note) if an individual note isn't a string? The former makes it clear what it is, while the latter provides a feature that Barry has been asking for...

If either of those fails, either raise an exception from print_exception() or swallow the error and print a simple message. It probably doesn't really matter, repr() or str() failing should be extremely obscure.

I'll do something like what we do for str(exc).

[Zac-HD]

Heya - I think we should have an explicit test that if you .split() an ExceptionGroup which already has __notes__, it's shallow-copied rather than simply assigned to the new instances. It'd be pretty confusing to call .add_note() on one and have it added to other instances too!

[iritkatriel]

Heya - I think we should have an explicit test that if you .split() an ExceptionGroup which already has __notes__, it's shallow-copied rather than simply assigned to the new instances. It'd be pretty confusing to call .add_note() on one and have it added to other instances too!

That's a good point. I added the test and fixed the code to copy the notes in split(), so long as they are a sequence (producing a list). If people assign arbitrary objects to notes I don't know how to copy them.
Is there a "copy protocol" we can demand?

[bedevere-bot]

🤖 New build scheduled with the buildbot fleet by @iritkatriel for commit 95be670 🤖

If you want to schedule another build, you need to add the ":hammer: test-with-buildbots" label again.

[iritkatriel]

Sorry about the noise (I messed it up multitasking). Getting there now.

[bedevere-bot]

🤖 New build scheduled with the buildbot fleet by @iritkatriel for commit 602b4c4 🤖

If you want to schedule another build, you need to add the ":hammer: test-with-buildbots" label again.

[iritkatriel]

The buildbots are happy with this. Let me know if you would like to re-review, otherwise I'll merge it tomorrow or so.

[serhiy-storchaka]

Opened #106033.

[serhiy-storchaka]

Do not use PyObject_HasAttr. It is broken by design.