Feature/detect format function

[cleong110]

DRAFT PR:

Adding a function to fix #142. Still TODO:

• go through the file and replace manual checks with this function.
• do some more testing to make sure there are no side effects

[cleong110]

OK, I've updated throughout generic.py to use the new detect_known_format function. I also added some pytests in, though not comprehensively. But it seems to successfully work with the fake_pose function, detecting the types I expect it to type.

Various functions had various levels of support for the different types. I've gone through and tried to make it more consistent, such that

• if a type is known not supported it will give an ImportError
• if a type is unknown it will give a ValueError

[cleong110]

OK, ready for the next round, I believe.

• detect function takes a pose or poseheader now
• various formatting changes.
• added some unknown-format poses to the tests, detect_known_format correctly raises an exception.
• removed the unreachable ValueErrors
• various type annotations

[AmitMY]

Very cool, thank you :)

need to only fix the tests.
instead of from pose_format import Pose i think it needs from pose_format.pose import Pose

[cleong110]

Taking a crack at fixing the tests. They run on my machine with python 3.10, but crash when run on github on 3.8. Why?

Error:

The error in question:

______________________ ERROR collecting pose_format/utils ______________________
/opt/hostedtoolcache/Python/3.8.18/x64/lib/python3.8/importlib/__init__.py:127: in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
<frozen importlib._bootstrap>:1014: in _gcd_import
    ???
<frozen importlib._bootstrap>:991: in _find_and_load
    ???
<frozen importlib._bootstrap>:975: in _find_and_load_unlocked
    ???
<frozen importlib._bootstrap>:671: in _load_unlocked
    ???
/opt/hostedtoolcache/Python/3.8.18/x64/lib/python3.8/site-packages/_pytest/assertion/rewrite.py:184: in exec_module
    exec(co, module.__dict__)
pose_format/utils/conftest.py:4: in <module>
    from pose_format import Pose
pose_format/__init__.py:1: in <module>
    from pose_format.pose import Pose
pose_format/pose.py:13: in <module>
    from pose_format.utils.generic import pose_normalization_info
pose_format/utils/generic.py:6: in <module>
    from pose_format import Pose
E   ImportError: cannot import name 'Pose' from partially initialized module 'pose_format' (most likely due to a circular import) (/home/runner/work/pose/pose/src/python/pose_format/__init__.py)

Files

src\python\pose_format\__init__.py has

from pose_format.pose import Pose
from pose_format.pose_body import PoseBody
from pose_format.pose_header import PoseHeader

src\python\pose_format\utils\__init__.py is empty

src\python\pose_format\utils\generic_test.py has

from pose_format.pose import Pose

Ah, perhaps conftest.py? It had:

from pose_format import Pose

what if I change everywhere?

• changing every single instance of pose_format import Pose to pose_format.pose import Pose everywhere in the whole library. Still got errors.

What if I trace the error more closely?

Here's the error output:

pose_format/__init__.py:1: in <module>
    from pose_format.pose import Pose
pose_format/pose.py:13: in <module>
    from pose_format.utils.generic import pose_normalization_info
pose_format/utils/generic.py:6: in <module>
    from pose_format.pose import Pose

That genuinely is a circular import!

OK, so we can't use pose_format.utils.generic in pose_format.pose.pose, because pose_format.utils.generic imports pose_format.pose... how to fix this?

[cleong110]

I'm kinda stumped by this pose import circle. Which I think I introduced actually, with the "quick and simple" PR where pose.normalize() would call it by default.

Pose imports utils.generic to get pose_normalization_info
utils imports Pose
^ thus a circle.

Where I'm stuck:

• Simplest solution might be to remove pose_normalization_info import from Pose
• but we do want to have that feature, so do we move the implementation into Pose?
• But then pose_shoulders() needs to move too...

[AmitMY]

Fixed the circular issue at 56e6717

[cleong110]

Fixed the circular issue at 56e6717

That'll do it!

[cleong110]

Pipe annotations causing failures

OK, latest problem is with the pipe type annotations, which prior to 3.10 will throw errors.

https://peps.python.org/pep-0604/

I can either remove all those, or we drop support for 3.8 and 3.9

NotImplementedErrors

Tests for versions 3.10-3.12 are currently expected to fail with NotImplementedErrors. We can either leave them failing as a reminder to implement, or just add a with pytest.raises to catch those.

[AmitMY]

I susggest we use Union instead of the pipe, to support older versions

[cleong110]

While trying to get tests to pass, I discovered two problems with correct_wrist:

1. It was modifying the input Pose's data despite returning a Pose object. Added a deepcopy operation to prevent this.
2. It was creating stacked_conf with shape 3 points (XYZ), even if the format called for 2 (XY). Changed out the hardcoded "3" to check the shape of the data, thus:

point_coordinate_count = wrist.shape[-1]
    stacked_conf = np.stack([wrist_conf] * point_coordinate_count, axis=-1) 

[cleong110]

All tests now green on my machine!

[AmitMY]

great! thanks for the hard work :)