[doc] os.walk() consider some symlinks as dirs instead of non-dirs

[socketpair]

BPO	12970
Nosy	@ncoghlan, @vstinner, @benhoyt, @4kir4, @socketpair
Files	z.patch: patch for the problem docs-walk-issue12970.patch: Update os.walk() docs

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = None
closed_at = None
created_at = <Date 2011-09-13.13:02:11.368>
labels = ['3.11', 'type-feature', 'docs']
title = '[doc] os.walk() consider some symlinks as dirs instead of non-dirs'
updated_at = <Date 2021-11-28.13:04:55.425>
user = 'https://github.com/socketpair'

bugs.python.org fields:

activity = <Date 2021-11-28.13:04:55.425>
actor = 'iritkatriel'
assignee = 'docs@python'
closed = False
closed_date = None
closer = None
components = ['Documentation']
creation = <Date 2011-09-13.13:02:11.368>
creator = 'socketpair'
dependencies = []
files = ['23140', '36138']
hgrepos = []
issue_num = 12970
keywords = ['patch']
message_count = 5.0
messages = ['143965', '143967', '152805', '223367', '224167']
nosy_count = 9.0
nosy_names = ['ncoghlan', 'vstinner', 'benhoyt', 'docs@python', 'akira', 'socketpair', 'alexey-smirnov', 'Sung-Yu.Chen', 'ukl']
pr_nums = []
priority = 'normal'
resolution = None
stage = None
status = 'open'
superseder = None
type = 'enhancement'
url = 'https://bugs.python.org/issue12970'
versions = ['Python 3.11']

[socketpair]

Consider code:

for (root, dirs, nondirs) in os.walk(path, followlinks=False):
    print (nondirs)

This code will not print symlinks that refer to some dir. I think it is the bug.

In other words: If followlinks is True, we should consider some symlinks as dirs. If not, any symlink is the non-dir.

Patch included.

Also, please fix documentation about this nuance.

[socketpair]

Also, there is some mis-optimisation for followlinks=False: stat() and then lstat() will be called. Instead of one lstat().

Code may be rewritten as (but I don't know about cross-platform issues):
---------------------------------

if followlinks:
    mode = os.stat(path).st_mode
else:
    mode = os.lstat(path).st_mode

if stat.S_ISDIR(mode):
    dirs.append(path)
else:
    nondir.append(path)

---

It will be much cleaner than current (or patched with my patch) implementation

[ncoghlan]

This behaviour came up recently when implementing os.fwalk() [1]. There are problems with all 3 possible approaches (list as dirs, list as files, don't list at all) when followlinks is False. Since all alternatives are potentially surprising, the current behaviour wins by default (as people will already have written their code to cope with that behaviour and there's no net gain in changing the default, since the desired treatment of such links will vary according to the task at hand).

As a result, I'm converting this to a pure documentation issue - the os.walk() docs should definitely mention this subtlety. The behaviour won't be changing, though.

[1] http://bugs.python.org/issue13734,#msg151077

[ukl]

I like the function as it is documented, i.e. "filenames is a list of the names of the non-directory files in dirpath.". This includes all symlinks (in the followlinks=False cast at least).

I'd say not including symlinks to directories but symlinks to files is a magnitude more surprising than treating a symlink to a directory as a file. And if you consider this as a short comming of the documentation this isn't (IMHO) a subtlety. The (my?) intuition says: all entries of a root (apart from . and .. as documented) are included in either dirnames or filenames.

Yes, changing behaviour here might break some code, but this applies to all changes.

For some usecases it might be right to just skip over symlinks-to-dirs, but if it's not you have to opendir + read all root entries again in the loop to find all symlinks which effectively means reimplementing os.walk.

[4kir4]

I've updated os.walk() documentation to mention that *dirnames* list
includes symlinks to directories.

To imitate the other two cases:

• treat the symlinks as files:

    for dirpath, dirnames, files in os.walk(top):
        dirs = []
        for name in dirnames:
            (files if islink(join(dirpath, name)) else dirs).append(name)
        dirnames = dirs

• don't include in either of the lists:

    for dirpath, dirnames, files in os.walk(top):
        dirnames[:] = [name for name in dirnames
                       if not islink(join(dirpath, name))]

where islink = os.path.islink and join = os.path.join.

I've uploaded the documentation patch. Please, review.