Add pathlib.chown method

[vajrasky]

BPO	20779
Nosy	@pitrou, @vajrasky, @sedrubal, @eamanu, @y0urself
PRs	gh-64978: Add chown() to pathlib.Path #31212
Files	add_chown_to_pathlib.patch

^{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 2014-02-26.09:51:50.192>
labels = ['type-feature', 'library', '3.11']
title = 'Add pathlib.chown method'
updated_at = <Date 2022-02-08.11:17:25.947>
user = 'https://github.com/vajrasky'

bugs.python.org fields:

activity = <Date 2022-02-08.11:17:25.947>
actor = 'y0urself'
assignee = 'none'
closed = False
closed_date = None
closer = None
components = ['Library (Lib)']
creation = <Date 2014-02-26.09:51:50.192>
creator = 'vajrasky'
dependencies = []
files = ['34227']
hgrepos = []
issue_num = 20779
keywords = ['patch']
message_count = 7.0
messages = ['212245', '212378', '212411', '212425', '396629', '398695', '412787']
nosy_count = 8.0
nosy_names = ['pitrou', 'cvrebert', 'neologix', 'vajrasky', 'sedrubal', 'eamanu', 'Tom Cook', 'y0urself']
pr_nums = ['31212']
priority = 'normal'
resolution = None
stage = 'patch review'
status = 'open'
superseder = None
type = 'enhancement'
url = 'https://bugs.python.org/issue20779'
versions = ['Python 3.11']

Linked PRs

• GH-64978: Add pathlib.Path.chown() #104183

[vajrasky]

For pragmatic and philosophical reasons, I would argue that we should add chown to pathlib library.

[pitrou]

I don't know about philosophical reasons ;-), but I don't think chown is very often used. Do you know of a context where it is?

[vajrasky]

Pragmatic reasons:

I use chown moderately often. Usually as root, I want to change the uid of the files after generating a bunch of files. But I almost never changed gid.

I am in two mind whether I should add lchown as well or not (just like lchmod) and make the signature of pathlib.chown same as os.chown (with third parameter, follow_symbolic_links).

If you close this as won't fix, I will not hold grudge against you, though. ;)

[pitrou]

Well, I don't know yet, but if we ever want chown() in pathlib, I think it should be higher-level and accept symbolic uids as well.

[TomCook]

+1 this.

I have a program that opens a UNIX socket as root for other processes to communicate with it. I need to set the permissions to 0o775 and set the owner gid to a specific group so that members of that group can communicate with the process. It's annoying to have to drop back to os.chown(str(path), ...) for this.

[eamanu]

Hello!

I also use it frequently, but looking this issue I can agree with pitrou that there aren't lot of people that use chown.

If I'm not wrong, there's a trend to move os to pathlib, perhaps chown should be in pathlib too :)

I can work on this.

[y0urself]

I would love to use chown for a pathlib Path, too.

It may not be used often, but if anyone want's to change all the os references from a file, it would be cool to have it on pathlib!

[barneygale]

I'm -1 on this: I don't think pathlib should start accumulating further functionality from os unless its used to implement high-level path behaviour (like walking directory trees).

[CAM-Gerlach]

To note, I can't confirm myself since I'm on a Windows machine at the moment, but os.chown is documented as natively supporting pathlib.Paths since Python 3.6. There's also the higher level shutil.chown which has a nicer API with more Pythonic argument handling and accepts user/group names as well as IDs, unlike the os version. As it simply passes path through to the underlying os.chown, it in turn supports pathlib.Paths as well.

This seems to address the major concrete issue expressed by users here (e.g. "It's annoying to have to drop back to os.chown(str(path), ...)", "I would love to use chown for a pathlib Path, too."), as well as exposing a nicer higher-level interface without the need to add yet another chown in pathlib.

(Sidenote: Perhaps, if there's interest, it's worth allowing passing through dir_fd or particularly follow_symlinks in shutil.chown to os.chown so the latter is a proper superset of the former, but that's an entirely separate issue of course and unrelated to pathlib.)

[AlexWaygood]

I agree with @barneygale, @CAM-Gerlach, and @brettcannon (Brett voiced his opinions on the PR here: #31212 (comment)). There seems to be little point adding this method, if it's just a thin wrapper around os.chown().

[anki-code]

It will be super cool to have this method because there is awesome python-powered xonsh shell and Path objects are using in xonsh broadly. You can do this in the shell using path-strings:

pip install -U 'xonsh[full]'
xonsh

cd /tmp && echo world > hello.txt 

p'hello.txt'
# Path('hello')
p'hello.txt'.read_text()
# 'world\n'

for f in p'/tmp'.glob('hello*'):
    print(f, f.exists())
# /tmp/hello.txt True

It will be cool to run p'hello.txt'.chown('root') instead of running subprocess chown root hello.txt (it's slower) or do import os etc.

I'm for reopening this issue and favor for enriching pathlib.

[barneygale]

Re-opening the issue for the sake of discussion.

I'm ±0 on this idea.

On the one hand, I don't want pathlib to accumulate simple wrappers of os functions, because there's a large number of them and it's hard to draw the line. The Zen of Python says "There should be one-- and preferably only one --obvious way to do it."

On the other hand, the implementation of chown() in shutil (not os) feels high-level enough to belong in pathlib, as it deals with user and group names like pathlib's owner() and group(). I have a vague plan to move shutil things into pathlib, so it fits reasonably well. In the very distant future we might consider deprecating the shutil version, which makes me feel better about the intervening duplication.

Would love to hear thoughts from others!

[anki-code]

I prefer shutil.chown() vs os.chown() and when wrote about xonsh's use cases I mean shutil's version. So +1

[CAM-Gerlach]

Personally, FWIW, I'm +0.5 so long as the idea fits with the long-term plan and the implementation does not add more than minimal additional maintenance burden. It's a little unfortunate there are two different chown functions with different APIs, but if this could eventually lead to deprecating one of them, leading to the same net number, while adding the convinence of calling methods directly on Paths rather than importing from os, then it seems reasonable to consider.

[barneygale]

Thanks both.

IMO we should move the implementation from shutil to pathlib. The version in shutil becomes a thin wrapper:

# shutil.py

def chown(path, user=None, group=None):
    pathlib.Path(path).chown(user, group)

However, we'd need to add support for bytes paths to pathlib, as they're accepted by shutil.chown(). There's no feature request logged for that yet. It's not easy, but it's worth doing I think.

[brettcannon]

we'd need to add support for bytes paths to pathlib, as they're accepted by shutil.chown(). There's no feature request logged for that yet. It's not easy, but it's worth doing I think.

FYI I disagree with that as we want people to move towards string-based paths as they are much easier to work with (just like we want people to use UTF-8 more).

[barneygale]

Hm. I wouldn't want to encourage users to use bytes if they don't need to, or to put the thought in their minds. If we were to add it, it would only get a brief mention in the docs.

But POSIX doesn't specify a particular character encoding for paths, and portable POSIX applications can't assume one. Pathnames are byte strings that may be in no particular encoding. If we can find a way to support this in pathlib without compromising the usability for people who don't care, I'd like to pursue it. Maybe.

[AlexWaygood]

Hm. I wouldn't want to encourage users to use bytes if they don't need to, or to put the thought in their minds. If we were to add it, it would only get a brief mention in the docs.

I view pathlib refusing to handle bytes as a feature. If I have a function that accepts an arbitrary Path object x, I know for sure that x.parts[0] will be a str. That's one of the things that makes the pathlib API easy to fit in my brain, and easy to reason about.

Even if we can add support for bytes paths without complicating the implementation or slowing things down, I think I'd still be against adding support for bytes paths in pathlib 😕 it would raise backwards-compatibility concerns and increase the cognitive load associated with the module, and I'm unsure that there's a strong use case for it.

[barneygale]

I find that quite convincing! Thanks.

It sounds like we can't implement shutil.chown() using pathlib.Path.chown() then, as we'd be breaking support for bytes. It probably makes sense to do the opposite: call shutil.chown() from pathlib. This will make it slightly more complicated to move other shutil functionality to pathlib in a future where AbstractPath is a thing, but that's a problem for another day.

[pitrou]

If users were allowed to pass bytes to pathlib, they would have to be converted to str internally anyway. I'd rather people do their own conversion if they need a special one.

[barneygale]

If users were allowed to pass bytes to pathlib, they would have to be converted to str internally anyway. I'd rather people do their own conversion if they need a special one.

I don't follow, sorry - why would they have to be converted to str? AFAIK the only reason to use bytes paths is specifically to avoid decoding to str, because the encoding can't be known with certainty on POSIX.

[barneygale]

Relevant SO: https://stackoverflow.com/questions/45724240/processing-non-utf-8-posix-filenames-using-python-pathlib

[AlexWaygood]

Relevant SO: https://stackoverflow.com/questions/45724240/processing-non-utf-8-posix-filenames-using-python-pathlib

(I mainly use Windows so apologies if this question exposes my ignorance about bytes paths!)

Are there any drawbacks to the solution given in the answer to that Stack Overflow question? It looks pretty "clean" to me — if there are no drawbacks, maybe we should just document in the pathlib docs that this is how you can use pathlib if you have a bytes path?

[barneygale]

I think what I'm missing is a proper understanding of surrogateescape encoding. I'll get back to you :-).

[pitrou]

PEP 383 is the fundamental reference here.

[pitrou]

I don't follow, sorry - why would they have to be converted to str?

Because pathlib uses str internally.

[barneygale]

PR available:

• GH-64978: Add pathlib.Path.chown() #104183

[nineteendo]

Barney, what's the status here?

[barneygale]

I'm still undecided as to whether this is a good idea.

IIUC we can't make shutil use pathlib because pathlib strips trailing slashes, and so shutil.chown('not_a_dir/', ...) would succeed where previously it failed with NotADirectoryError. This is fixable if we introduce a private superclass of Path that skips normalization, but there's little appetite for that right now.

Consequently we'd either need to make pathlib use shutil, which feels wrong to some core devs and I'm trying to avoid if I can, or we'd just copy-paste the shutil implementation into pathlib, which I feel pretty "meh" about too.

[barneygale]

#73991 is the more important shutil/pathlib issue. I hope to solve that first, and by doing so, shed some light on whether this request is worth doing or not.

[barneygale]

I'm going to close this as "won't do". It turns out that Path.chown() won't be helpful for copying file metadata in #73991, and I can see no other compelling reason to add it. Plus it's POSIX-only and usually needs root. Use os.chown() or shutil.chown() instead please! And feel free to leave a comment if you disagree and would like to see this re-opened.