The test for empty strings in calls to pydoc.apropos() uses a chmod call that seems non-sensical:

(Line 1313 in particular)

This test was added as part of resolving #65747. Part of the triage of that issue included the observation that the problem could be reproduced by removing execute permissions from a directory. However, this appears to be unrelated to the reported problem.

By removing the x permission from the directory, the module is converted into a namespace package - at which point, it can't have a docstring. However, you can also validate this by... having a module without a docstring. Which the current test implements.

Adding to the confusion - the test only removes user execute permissions. Group and Global execute permissions are retained. The original report removed all execute permissions, not just user execute permissions; so the test isn't reproducing the triage example. The only explanation I can find for this is that the buildbots run with a umask of 0o77, which means removing user execute permissions is effectively equivalent to chmod a-x... but it won't be for anyone who isn't running under buildbot.

This was discovered in the process of developing #136740. This did reveal a bug in Emscripten's implementation of readdir() - but the actual test itself doesn't make any sense to me (or @hoodmane).

Linked PRs

• gh-136744: Remove unnecessary chmod from pydoc.apropos() test. #136746