Firstly, .getArray() should be spelled .array these days.
Secondly, if this has to be done, please make it short (a single line or two) — unlike the table printing, which can be thousands of lines long.

At the very least I'd love it if str() gave the dimensions and origin information. I agree that you don't want a 4Kx4K dump but on the other hand, doesn't numpy have a parameter for controlling how much content is reported for str() so shouldn't we respect that setting?

That was my point of delegating to .getArray(): just print whatever that does, plus dimensions and origin.

Tim Jenness, Paul Price: Since you both expressed opinions about this, I figured you'd want to review it. It delegates the array collapsing to numpy, and includes the bbox. I think that means it also tells you the xy0 (as the minimum of the bbox): we weren't able to come up with a test case that behaved otherwise, but maybe you know of one?

Here's what an example str(maskedImage) looks like.

>>> print(lsst.afw.image.MaskedImage(100,100))

image=[[0. 0. 0. ... 0. 0. 0.]

 [0. 0. 0. ... 0. 0. 0.]

 [0. 0. 0. ... 0. 0. 0.]

 ...

 [0. 0. 0. ... 0. 0. 0.]

 [0. 0. 0. ... 0. 0. 0.]

 [0. 0. 0. ... 0. 0. 0.]], bbox=(minimum=(0, 0), maximum=(99, 99))

mask=[[0 0 0 ... 0 0 0]

 [0 0 0 ... 0 0 0]

 [0 0 0 ... 0 0 0]

 ...

 [0 0 0 ... 0 0 0]

 [0 0 0 ... 0 0 0]

 [0 0 0 ... 0 0 0]], bbox=(minimum=(0, 0), maximum=(99, 99)), maskPlaneDict={'BAD': 0, 'CR': 3, 'DETECTED': 5, 'DETECTED_NEGATIVE': 6, 'EDGE': 4, 'INTRP': 2, 'NO_DATA': 8, 'SAT': 1, 'SUSPECT': 7}

variance=[[0. 0. 0. ... 0. 0. 0.]

 [0. 0. 0. ... 0. 0. 0.]

 [0. 0. 0. ... 0. 0. 0.]

 ...

 [0. 0. 0. ... 0. 0. 0.]

 [0. 0. 0. ... 0. 0. 0.]

 [0. 0. 0. ... 0. 0. 0.]], bbox=(minimum=(0, 0), maximum=(99, 99))

Pair coded with Gabor Kovacs [X] and Krzysztof Findeisen.

Jenkins run: https://ci.lsst.codes/blue/organizations/jenkins/stack-os-matrix/detail/stack-os-matrix/28903/pipeline

Looking only at the example, right now:

• Can you get rid of the duplicated bbox entries (probably requires special-casing MaskedImage).
• I think the class name should be in there somewhere, (MaskedImage(...))?
• It would be good to include information about the pixel type, either by making this explicit (e.g., dtype=float32) or through the class name (e.g., MaskedImageF).
• Should have commas between the image, mask and variance elements.
• Does __repr__ work the same way?

Usually repr should show the class but str tends not to. Not that we've been overly consistent with this.

Pulling out of review since both reviewers have commented.

  Can you get rid of the duplicated bbox entries (probably requires special-casing MaskedImage).

Our thought with this one was that we didn't see explicit checks that the three components of a MaskedImage had to have the same bbox. But I just tested it, and it looks like you can't make a MaskedImage with different bboxes with the (Image, Mask, Variance) constructor. So, I'll special case and get rid of the extra bbox printing. Unfortunate, since it would be nice to keep them in sync for future changes.

• I think the class name should be in there somewhere, (MaskedImage(...))?
• It would be good to include information about the pixel type, either by making this explicit (e.g., dtype=float32) or through the class name (e.g., MaskedImageF).
• Does _repr_ work the same way?

As Tim Jenness says, that's what repr is for, and what it does, in this case (prepends the namespace and class name). For comparison, look at repr vs. str for a numpy array.

Should have commas between the image, mask and variance elements.

Done.

Thanks for the review. Merged and done.