OSError: [Errno 22] Invalid argument when loading an SLP file

[aaprasad]

Currently when I try to load a .slp file with a bad video file name eg:

print(sio.load_slp(slp_file))

I get the following error:

OSError                                   Traceback (most recent call last)
Cell In[8], line 2
      1 for slp_file, vid_file in zip(slp_files, vid_files):
----> 2     print(sio.load_slp(slp_file))

File /opt/conda/envs/dreem/lib/python3.11/site-packages/sleap_io/io/main.py:19, in load_slp(filename)
     10 def load_slp(filename: str) -> Labels:
     11     """Load a SLEAP dataset.
     12 
     13     Args:
   (...)
     17         The dataset as a `Labels` object.
     18     """
---> 19     return slp.read_labels(filename)

File /opt/conda/envs/dreem/lib/python3.11/site-packages/sleap_io/io/slp.py:1012, in read_labels(labels_path)
   1003 """Read a SLEAP labels file.
   1004 
   1005 Args:
   (...)
   1009     The processed `Labels` object.
   1010 """
   1011 tracks = read_tracks(labels_path)
-> 1012 videos = read_videos(labels_path)
   1013 skeletons = read_skeletons(labels_path)
   1014 points = read_points(labels_path)

File /opt/conda/envs/dreem/lib/python3.11/site-packages/sleap_io/io/slp.py:130, in read_videos(labels_path)
    126 for video_ind, video_data in enumerate(
    127     read_hdf5_dataset(labels_path, "videos_json")
    128 ):
    129     video_json = json.loads(video_data)
--> 130     video = make_video(labels_path, video_json, video_ind=video_ind)
    131     videos.append(video)
    132 return videos

File /opt/conda/envs/dreem/lib/python3.11/site-packages/sleap_io/io/slp.py:72, in make_video(labels_path, video_json, video_ind)
     69 if not video_path.exists():
     70     # Check for the same filename in the same directory as the labels file.
     71     video_path_ = Path(labels_path).parent / video_path.name
---> 72     if video_path_.exists():
     73         video_path = video_path_
     74     else:
     75         # TODO (TP): Expand capabilities of path resolution to support more
     76         # complex path finding strategies.

File /opt/conda/envs/dreem/lib/python3.11/pathlib.py:1235, in Path.exists(self)
   1231 """
   1232 Whether this path exists.
   1233 """
   1234 try:
-> 1235     self.stat()
   1236 except OSError as e:
   1237     if not _ignore_error(e):

File /opt/conda/envs/dreem/lib/python3.11/pathlib.py:1013, in Path.stat(self, follow_symlinks)
   1008 def stat(self, *, follow_symlinks=True):
   1009     """
   1010     Return the result of the stat() system call on this path, like
   1011     os.stat() does.
   1012     """
-> 1013     return os.stat(self, follow_symlinks=follow_symlinks)

OSError: [Errno 22] Invalid argument: 'train/top\\top-10072022134804-0000_h265_CRF12_denoised.mp4'

I know theres a Labels.replace_filename but you need to be able to load the file first. Therefore, it would be nice if either A. you can pass in a videos argument to override the current video OR if it can't open it just ignore and warn the user so you can at least access the actual labels

PS:
sleap.load_file(slp_file)
works fine so I'm not sure whats different at an implementation level

[talmo]

Here are the relevant offending lines:

sleap-io/sleap_io/io/slp.py

Lines 57 to 80 in 5b79a22

	backend_metadata = video_json["backend"]
	video_path = backend_metadata["filename"]
	# Marker for embedded videos.
	source_video = None
	is_embedded = False
	if video_path == ".":
	video_path = labels_path
	is_embedded = True
	# Basic path resolution.
	video_path = Path(video_path)
	if not video_path.exists():
	# Check for the same filename in the same directory as the labels file.
	video_path_ = Path(labels_path).parent / video_path.name
	if video_path_.exists():
	video_path = video_path_
	else:
	# TODO (TP): Expand capabilities of path resolution to support more
	# complex path finding strategies.
	pass
	# Convert video path to string.
	video_path = video_path.as_posix()

From the original issue, the path stored in the SLP file is:

G:/.shortcut-targets-by-id/1ABZ_lZYLzT3qOjXEaUZSd9-JFyvk5lSl/2021-10-27_Talmo/Scratch/Proofreading/2022-10-07/10072022134804/top\\top-10072022134804-0000_h265_CRF12_denoised.mp4

There's clearly both forward slashes and (escaped) backslashes, but pathlib should handle it.

Follow up investigation: cannot reproduce on Colab.

Not only does the SLP file load, but I can't even get the same error when running the actual erroring line:

import os
os.stat("train/top\\top-10072022134804-0000_h265_CRF12_denoised.mp4")

Output:

FileNotFoundError                         Traceback (most recent call last)
[<ipython-input-18-45758ebc4f0a>](https://localhost:8080/#) in <cell line: 3>()
      1 import os
      2 
----> 3 os.stat("train/top\\top-10072022134804-0000_h265_CRF12_denoised.mp4")

FileNotFoundError: [Errno 2] No such file or directory: 'train/top\\top-10072022134804-0000_h265_CRF12_denoised.mp4'

Note that it's FileNotFoundError: [Errno 2], and not OSError: [Errno 22] Invalid argument.

This tells me that it's maybe a file system issue?

Either way, keeping this open as a reminder that we should probably expose the logic in Labels.replace_filenames to load_slp so we can do filename replacement at load time using the same signature as Labels.replace_filenames.

[talmo]

Relevant pathlib sections:

• Path.exists(): https://github.com/python/cpython/blob/d542a9be51776e8d589363ee15164dec8dbd3a76/Lib/pathlib.py#L1230-L1243
• Ignored errors (FileNotFound is ENOENT, which is ignored): https://github.com/python/cpython/blob/d542a9be51776e8d589363ee15164dec8dbd3a76/Lib/pathlib.py#L31

But it does seem like errno 22 means invalid argument.

Maybe on some platforms we need to sanitize the filename manually?

[talmo]

Fixed in sleap-io v0.1.7.