Draft: Add very basic implementation of PEP657 style expression markers in tracebacks

[ammaraskar]

Hey folks this is just a proof of concept for #10224 to serve as a reference for if any contributor wants to give this a full shot.

This adds a very basic implementation of the PEP657 style expression highlighting in the short and long traceback modes. It's not the full fledged python traceback.py implementation where we try to differentiate which parts contain operators.

I don't think I'll have time to finish this out fully, there's some interesting problems here like mypy complaining because typeshed doesn't have some of the newer traceback FrameSummary attrs like colno. Additionally, since the python offsets are from the start of the line, we have to know the line before it gets de-dented in Source.

---

Example of what it looks like:

and with tb short:

[Zac-HD]

I don't think I'll have time to finish this out fully

Thanks for taking it this far! Whether or not you keep going, this is a neat contribution 😄

There's some interesting problems here like mypy complaining because typeshed doesn't have some of the newer traceback.FrameSummary attrs like colno.

It looks like typeshed does have them, so I think we're probably seeing mypy pointing out that we need to version-guard those accesses; see comment below.

Additionally, since the python offsets are from the start of the line, we have to know the line before it gets de-dented in Source.

I think we could fetch it with linecache (as used by the traceback module), and then correct for the different length of leading whitespace? Alternatively we could ast.parse() the Source line, and see if there's an expression which matches the expected start and end colnos for various candidate dedents; it's a heuristic but likely to be unambiguous almost all the time.

[Zac-HD]

Here, and in end_colno below, we'll need to handle the case where on earlier Python versions these attributes don't exist. I'd consider defining the properties differently based on version, so that we can easily remove the backwards-compatibility code once we're ready to drop support for Python 3.10

[Zac-HD]

It's not the full fledged python traceback.py implementation where we try to differentiate which parts contain operators.

That would be very nice to have, but I'd be happy to merge something without that and open a follow-up issue: this is already a nice improvement on the status quo!

[ammaraskar]

Oh yup, you're totally right about them being behind a version check in typeshed, I missed that. Added a version guard and now mypy seems happy with it.

Also added version guards for the one unit test I added and changed.

[Zac-HD]

Thanks again @ammaraskar!

I think the only thing left is to add yourself to the AUTHORS file, and create changelog/10224.improvement.rst containing something like "Partial support for PEP-657 precise traceback locations in Pytest's alternative traceback formatting."