Different return types for S3Path.stat() and WindowsUPath.stat()

[theogaraj]

Which operating system and Python version are you using?
Windows 11, Python 3.9.6

Which version of this project are you using?
0.1.3

What did you do?
I am attempting to use universal_pathlib in order to have a unified way of handling files whether they are local or in S3.
One of the things I need to do is get all the files in a folder (yes I know S3 doesn't have actual folders, but hopefully you understand what I mean) and get their sizes.

>>> from upath import UPath
>>> lpath = UPath('local_data/files/')
>>>spath = UPath('s3://test_bucket/files/')
>>> lfiles = lpath.iterdir()
>>> lf = next(lfiles)
>>> type(lf.stat())
<class 'os.stat_result'>
>>> sfiles = spath.iterdir()
>>> sf = next(sfiles)
>>> type(sf.stat())
<class 'dict'>
>>> lf.stat().st_size
1657
>>> sf.stat().st_size
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
AttributeError: 'dict' object has no attribute 'st_size'

What did you expect to see?
I expected to be able to use same code to get file sizes whether they are in local directory or in S3

What did you see instead?
The return types of stat() are different for S3Path vs WindowsUPath and I can't get file size in the same way from each.

Would this difference in behavior be something you would consider reconciling? Alternatively, do you have suggestions on another approach to achieving what I'm trying to do?

[ap--]

Hi @theogaraj

This is definitely an issue, and we should create a UPathStatResult class (name up for discussion) that provides an interface compatible with dict (fsspec) and os.stat_result

We need to ensure that the os.stat_result attribute names map to the equivalent type for each of the fsspec filesystems, too.

relevant fsspec issues:

• Add ctime/mtime to list of expected values in info filesystem_spec#526

For now I don't have a good recommendation other than

_stat = pth.stat()
size = _stat["size"] if isinstance(_stat, dict) else _stat.st_size

Cheers,
Andreas

[ap--]

I vaguely remembered that I implemented something related to this already...

getmtime on fsspec

https://github.com/mdshw5/pyfaidx/blob/cac82f24e9c4e334cf87a92e477b92d4615d260f/pyfaidx/__init__.py#L1318-L1345

[theogaraj]

@ap-- thank you for checking and responding so promptly. Because of another problem I had with iterdir (logged as #146) I ended up going with the slightly clunkier spath.fs.ls(str(spath)) and then accessing the ['size] attribute common to both local and S3.

So this is by no means a showstopper. I'll track this issue and can update my code whenever someone is able to resolve these two issues.

[bolkedebruin]

Just ran across this. See https://github.com/apache/airflow/blob/main/airflow/io/store/stat.py for a stat compatible version.

[asford]

I just ran across the same issue here and would be happy to fire a PR with a fix, though it looks like @ap-- may already be investigating?

@bolkedebruin make several updates to the airflow io provider while inheriting from UPath in apache/airflow#35612 which are great prior art here.

I would propose that we just port the changes to support stat into UPath, which is now under:

• stat: https://github.com/apache/airflow/blob/2.8.1/airflow/io/utils/stat.py
• def stat: https://github.com/apache/airflow/blob/2.8.1/airflow/io/path.py#L180-L186

[ap--]

Collected Info

For links to info() dicts of a lot of FileSystem implementations check fsspec/filesystem_spec#526 (comment)

All filesystems have "name", "size" and "type".

• "name" is a string
• "size" is an int or None
• "type" is a string that is almost always "file" or "directory" but it can be "other", or (something else)

For translating to os.stat_result.st_* attributes, these are the keys that could be checked:

Attribute	Possible Info Keys
st_mode	mode, unix.mode, writable, isLink, nlink, permission, isexec
st_ino	ino, name, id, sha, hex, Digest?
st_dev
st_nlink	nlink, isLink
st_uid	uid, owner, uname, unix.owner
st_gid	gid, group, gname, unix.group
st_size	size
st_atime	time, last_accessed_on, accessTime
st_mtime	mtime, last_modified, last_modification_time_ms, timeModified, modify, modificationTime, LastModified, modified_at
st_ctime
st_birthtime	created, creation_time, timeCreated, created_at

Additional info to be considered comes from specific filesystems:

• https://hadoop.apache.org/docs/r1.0.4/webhdfs.html#FileStatus
• https://cloud.google.com/storage/docs/json_api/v1/objects#resource
• https://developers.google.com/drive/api/reference/rest/v3/files#File
• https://dropbox-sdk-python.readthedocs.io/en/latest/api/files.html#dropbox.files.FileMetadata

All data types need to be normalized to int