homepage

Consider: should os.fwalk() support dir_fd? I think so. In fact, in retrospect it seems like a bug that os.fwalk *doesn't* already support this.
Georg: is this a feature or a bugfix?
(Wish I'd thought of this Saturday!)
I actually did a little experimenting, and got os.fwalk(fd) to work with very little trouble. os.walk(fd) is harder because the recursive step appends a string to the existing path, and it really needs to be relative to the fd, and there's no way to pass both of those at once with the current signature. It's doable but it would require a separate function for the recursive step that accepted a dir_fd anyway.

My first cut at a patch. Made the logic in posix_listdir easy to follow, fixed up the docstring and docs.

Whoops, wrong issue, ignore that. (Meant for #15176, going there now.)

What I did:
* Added dir_fd=None as a keyword only parameter.
* Gave top a default of ".".
* Passed through dir_fd in two spots--that was all it took!
* Made fwalk contingent on os.stat and os.open both being
 in support_dir_fd, and os.listdir being in support_fd.
* Fixed docstring and docs.
* Added unit test for using dir_fd.

Second revision of patch.
* Removed os.fwalk from os.supports_dir_fd. As Georg points out,
 you can't test it. (Well, you can, but only if the test would
 have succeeded.)
* Changed fstat(fd) calls to stat(fd) calls.

Here's a variant of the patch adding os.walk(dir_fd=) support. I'm not pushing for this. I'm not sure it's a must-have; it seems like a nice-to-have, but we're past the time for nice-to-haves.

Larry, please regenerate patch in Mercurial for review. I have some comments.

Regenerated fwalk(dir_fd=) patch to make Rietveld happy.

Regenerated walk(dir_fd=) patch to make Rietveld happy.

Rietveld is not happy with git diffs.

Fourth time's the charm.

What's better than regenerating a bunch of diffs on the off chance that it'll make Rietveld happy? Nothing, that's what.

Let's keep it to fwalk.

Sure, for 3.3 anyway.
I think walk(dir_fd=) should be okay for 3.4. There's a better implementation anyway, where walk() just calls fwalk() and strips the last element off the yielded stuff. But moving on!

New changeset 7bebd9870c75 by Larry Hastings in branch 'default':
Issue #15177: Added dir_fd parameter to os.fwalk().
http://hg.python.org/cpython/rev/7bebd9870c75 

While I was at it, I changed the subject to accurately reflect the changeset's final disposition.