Replace `shape` assignment with `reshape` calls by jakirkham · Pull Request #1106 · rapidsai/cucim

jakirkham · 2026-06-24T18:05:31Z

NumPy is deprecating direct shape assignment. The underlying code path simply goes through reshape. So just switch to reshape to eliminate the deprecation warning and future proof the code.

ref: https://github.com/numpy/numpy/blob/2a674fc909ef47631e4f6550e25817731534987f/numpy/_core/src/multiarray/getset.c#L64

NumPy is deprecating direct `shape` assignment. The underlying code path simply goes through `reshape`. So just switch to `reshape` to eliminate the deprecation warning and future proof the code. ref: https://github.com/numpy/numpy/blob/2a674fc909ef47631e4f6550e25817731534987f/numpy/_core/src/multiarray/getset.c#L64

jakirkham · 2026-06-24T18:16:47Z

    if not hasattr(data, "__cuda_array_interface__"):
        data = np.frombuffer(data, np.uint8)
-    data.shape = tile_shape
+    data = data.reshape(tile_shape, copy=False)


The current behavior of .shape assignment is to call reshape and error if a copy occurs. Equivalent behavior can be implemented by calling .reshape(...) with the copy keyword argument as False

Based on the discussion below, all of these were dropped because CuPy's reshape doesn't support the copy keyword argument and they were either unneeded in that context or the intended behavior could be captured other ways.

grlee77

Thanks, @jakirkham this changes to avoid shape assignment look good to me

seberg

Thanks, looks all better and safe. Not sure if the last can see NumPy arrays but it hardly matters.
(The first feels like it might be a latent fix, but probably not in practice.)

jakirkham · 2026-06-24T18:59:35Z

+    x_view = x.reshape((ndim, 2), copy=False)
    return x_view.tolist()


Given we call .tolist(...) here, avoiding a copy probably doesn't matter much. After all this is an array that is already created as C-contiguous above, so the copy shouldn't happen anyways.

This would better match the same upstream change: cupy/cupy#10037

Suggested change

x_view = x.reshape((ndim, 2), copy=False)

return x_view.tolist()

x_view = x.reshape((ndim, 2))

return x_view.tolist()

Included in commit: 9c75cbb

Yeah. Reshapes are always no-copy by default if possible, so unless any of these could error it doesn't matter.
(But yes, it adds the guarantee, which may makes sense for the other cases)

This array is constructed so that copying shouldn't happen and it is a small array (extents). Plus `the `.tolist()` call will copy it in the end anyways. So drop the `copy=False` keyword argument. Co-authored-by: jakirkham <jakirkham@gmail.com>

seberg · 2026-06-24T19:34:16Z

    if not hasattr(data, "__cuda_array_interface__"):
        data = np.frombuffer(data, np.uint8)
-    data.shape = tile_shape
+    data = data.reshape(tile_shape, copy=False)


Hmmm, this one could be a problem, though? Because cupy doesn't have copy=False (should probably add that).
And the hasattr() above looks like the input may or may not be a CuPy array?

Good point. Looking at CuPy's reshape logic it appears to behave similar to NumPy's copy=None. Besides the NumPy case is already starting from a 1-D contiguous array so shouldn't copy anyways. So perhaps we can just drop this keyword argument

Suggested change

data = data.reshape(tile_shape, copy=False)

data = data.reshape(tile_shape)

Included in commit: bfb7cd0

jakirkham · 2026-06-24T20:08:20Z

+    # https://github.com/cupy/cupy/blob/v14.1.1/cupy/_padding/pad.py#L408
+    "ignore:Setting the shape on a NumPy array:DeprecationWarning:cupy",
+    # https://github.com/scikit-image/scikit-image/blob/v0.26.0/skimage/measure/_marching_cubes_lewiner.py#L237
+    "ignore:Setting the shape on a NumPy array:DeprecationWarning:skimage",


Have picked up @bdice fixes from PR: #1105

Was still seeing some test failures without them. Hoping by combining them we can get a clean CI run

Also the warnings in CuPy are being fixed by @seberg in PR: cupy/cupy#10037

The warnings in scikit-image will be fixed in 0.27 with PR: scikit-image/scikit-image#8020

Co-authored-by: jakirkham <jakirkham@gmail.com>

jakirkham · 2026-06-24T23:56:14Z

+        out_view = out.view()
+        out_view.shape = (-1,)  # no-copy reshape/ravel


Have reverted to the original syntax here as CuPy's reshape doesn't support the copy keyword argument and we really must avoid a copy here. So the best way to do this with CuPy is to assign to .shape. This is unfortunate, but it should work as intended.

Edit - Filed an upstream request to add the copy keyword argument to reshape: cupy/cupy#10039

bdice

Great, thanks!

jakirkham · 2026-06-25T16:03:04Z

Thanks all! 🙏

jakirkham · 2026-06-25T16:03:56Z

/merge

jakirkham requested a review from a team as a code owner June 24, 2026 18:05

jakirkham added bug Something isn't working non-breaking Introduces a non-breaking change labels Jun 24, 2026

jakirkham mentioned this pull request Jun 24, 2026

Ignore CuPy NumPy shape deprecation warning #1105

Closed

jakirkham added 2 commits June 24, 2026 11:10

Update copyright to 2026

6792ab4

Disallow copying in reshape when view required

25150c4

jakirkham commented Jun 24, 2026

View reviewed changes

grlee77 approved these changes Jun 24, 2026

View reviewed changes

seberg approved these changes Jun 24, 2026

View reviewed changes

jakirkham mentioned this pull request Jun 24, 2026

Update CuPy and NumPy dependency specs #1104

Open

jakirkham commented Jun 24, 2026

View reviewed changes

Drop copy=False from reshape in _as_pairs

9c75cbb

This array is constructed so that copying shouldn't happen and it is a small array (extents). Plus `the `.tolist()` call will copy it in the end anyways. So drop the `copy=False` keyword argument. Co-authored-by: jakirkham <jakirkham@gmail.com>

seberg reviewed Jun 24, 2026

View reviewed changes

bdice and others added 3 commits June 24, 2026 14:38

Ignore CuPy NumPy shape deprecation warning

8cb4254

Drop copy=False from reshape in GDS examples

bfb7cd0

Merge in PR rapidsai#1105

42e0515

jakirkham requested a review from a team as a code owner June 24, 2026 20:06

jakirkham requested a review from jameslamb June 24, 2026 20:06

jakirkham commented Jun 24, 2026

View reviewed changes

Comment thread python/cucim/src/cucim/skimage/util/_map_array.py Outdated

Revert to .shape assignment for CuPy array flattening

149f355

Co-authored-by: jakirkham <jakirkham@gmail.com>

jakirkham commented Jun 24, 2026

View reviewed changes

bdice approved these changes Jun 25, 2026

View reviewed changes

rapids-bot Bot merged commit b4ad404 into rapidsai:main Jun 25, 2026
68 checks passed

jakirkham deleted the use_reshape branch June 25, 2026 16:11

		x_view = x.reshape((ndim, 2), copy=False)
		return x_view.tolist()

	data = data.reshape(tile_shape, copy=False)
	data = data.reshape(tile_shape)

		out_view = out.view()
		out_view.shape = (-1,) # no-copy reshape/ravel

Uh oh!

Conversation

jakirkham commented Jun 24, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

grlee77 left a comment

Choose a reason for hiding this comment

Uh oh!

seberg left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jakirkham Jun 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

jakirkham Jun 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

bdice left a comment

Choose a reason for hiding this comment

Uh oh!

jakirkham commented Jun 25, 2026

Uh oh!

jakirkham commented Jun 25, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

jakirkham Jun 24, 2026 •

edited

Loading

jakirkham Jun 24, 2026 •

edited

Loading