homepage

Issue #22936 added support for easily displaying local variables in tracebacks to the unittest and traceback modules.
Would it be worth also making this capability readily available in the logging APIs that display traceback messages?
The main argument against it is that automatically logging local variables in tracebacks is a *really* good way to introduce sensitive information leaks into your log messages. "Safe repr" type objects (which automatically mask sensitive details in __repr__()) can mitigate this, but we don't provide such an object in the standard library (not even for bytes, bytearray or str, which are the typical containers for potentially sensitive data like passwords).

Yes, for debugging etc this can be very useful. I suggest further extending the new traceback interface to allow a filtering/transform hook of some sort, to allow folk more granular control than just repr overloading.

See also 43656.

In this pull request[1] I introduced a `format_locals` which can be provided to customize the serialization of the local variables of each frame. This could be used to filter out variables with certain names, for example. In this context, you could use a certain naming convention for variables with sensitive information and filter them out:
def _format_locals(filename, lineno, name, locals):
 return {k: repr(v) for k,v in locals.items() if not k.endsswith("_sensitive")}
traceback.TracebackException.from_exception(e, capture_locals=True, format_locals=format_locals).format()
(This should be exactly what Robert was suggesting.)
[1] https://github.com/python/cpython/pull/29299