When writing the unit tests for running doctests from inside a zipfile,
I initially had "<anon>" as the name being passed to the
doctest.DocTestFinder.find method.

With that name, the line numbers coming back for all of the examples
were being incremented by 1 beyond what test_doctest expected (and hence
my tests were failing).

Changing the name to "anon" eliminated the errors. (Both "<anon" and
"anon>" also failed in the same fashion as "<anon>")

I suspect there may be a problem with __LINECACHE_FILENAME_RE when "<"
or ">" are embedded in the name used for the doctest execution.

@Nick: could you provide a patch that addresses this issue?

On Sat, Jul 31, 2010 at 9:30 PM, Mark Lawrence <report@bugs.python.org> wrote:
>
> Mark Lawrence <breamoreboy@yahoo.co.uk> added the comment:
>
> @Nick: could you provide a patch that addresses this issue?

Probably, but the "don't do that" work around is trivial enough that
my own motivation to fix it is really low (otherwise I would have
fixed it by now).

It's still a genuine bug though. Since my original post isn't all that
helpful in reproducing it, I'll provide some more details for anyone
else that wants to tackle it.

To reproduce it, edit Lib/test/test_zipimport_support.py to pass
"<anon>" instead of name as the second argument to the finder.find
call inside _run_object_doctest(). Running "python -m test.regrtest
test_zipimport_support" will fail (with 6 of 33 doctests failing as of
2.7)

Then pass "anon" (no angle brackets) to see that the reported doctest
errors are different (1 of 19 reported as failing in 2.7) and, unlike
the errors with "<anon>" the error in this latter case is directly
attributable to having the wrong module name in the pdb output.

(Note: regrtest's "-v" option is actually more confusing than helpful
in debugging this test case due to the sheer amount of output it
generates from the doctest tests themselves)

Note that this appears to have been fixed already in the 3.x series (I haven't investigated the differences to see what makes this a 2.x only problem)

I have a vague memory of changing some code, in linecache I think, that involved anonymous names.  I might have fixed it by accident.

I tested on 3.10 (macos) and can confirm that it is fixed. 

Details: I changed _run_object_doctest as Nick explained, and made it display the output of the finder.find() call - the line number was the same with and without the <>s.