This is a proposed traceback module enhancement deriving from the "python -m" traceback verbosity discussion in bpo-16217.

Specifically, I'm thinking a good way of cleaning up those tracebacks would be to use TracebackException.format() to get the formatted exception output, but designate particular modules to be skipped during the iteration (specifically runpy, _frozen_importlib, importlib).

I'm filing this as a separate issue so the discussion of the public API for the capability in the traceback module can be separated from the discussion of applying that API in runpy.

I think a more general facility to strip unwanted frames from traceback objects would be useful for more people (including importlib, I think). i.e. actually mutate the traceback object, not remove the unwanted frames when printing (since it only works if you're the one doing the printing).

(note mutating a traceback object is very easy in C, it's just not possible in pure Python, except perhaps with nasty ctypes hacks)

Also, the ignore_modules would be more powerful as a generic filter callback, I think.

Yes, a public built-in facility to modify a traceback would certainly be useful, although there should be a standard way to "hide" unwanted frames without mutating the traceback, too, and the traceback module seems a perfect place to add it.

I have no objections to the filter argument, though filtering out frames based on modules to me feels by far more common a task than anything else filter would be capable of.

I can update the PR if there's a consensus that filter is favored over ignore_modules.

Ping. (this issue needs a decision on ignore_modules vs filter callback, and/or patch review)

I only recently realized that trace.Trace accepts two similar arguments, namely ignoremods and ignoredirs. Should we try to make the API and implementation of the functionality proposed here conform to that of trace.Trace's ignoremods? Would Python-Ideas be a more appropriate place to discuss this?

I have prepared a patch (attached) with a possible implementation of the C function that can filter traceback objects modifying the traceback chain. I can make a PR if the patch looks good enough to start iterating over it. In case an alternative implementation is needed I am more than happy to prepare another patch/PR.

I asked for some input on this issue in January (https://mail.python.org/pipermail/python-dev/2018-January/151590.html). Since then an important PR was merged (#4793), however it mutates the traceback object. Should there be a way to hide frames while keeping the original traceback object intact? If not, this issue should probably be closed. Otherwise I would appreciate a review of the PR.

It would seem no one is actually interested in this proposed enhancement. I'm closing my PR, since I'm not interested in resolving the file conflict. I'll probably submit a traceback-mutating patch to the bpo-16217. This issue can be closed.

In bpo-42247 the unit test framework uses limit as a hack to remove its own frame, and then the traceback module applies the limit to the chained exception as well.

If the filter suggested here existed, unit test could use that instead of limit.

New changeset 863154c by Irit Katriel in branch 'main':
bpo-31299: make it possible to filter out frames from tracebacks (GH-28067)
863154c