Return boolean from ImageMath lambda_eval and unsafe_eval comparison operations

[radarhere]

from PIL import Image, ImageMath
A = Image.new("L", (1, 1))
print(ImageMath.lambda_eval(lambda args: args["A"] == args["A"], A=A))

currently gives <PIL.Image.Image image mode=I size=1x1 at 0x1047906E0>

This PR suggests changing the output to a more meaningful True.

==, !=, <, <=, >, >=, equal() and notequal() would all now return boolean results.

[hugovk]

Is this a breaking change?

[radarhere]

For ImageMath expressions that involve equality or comparison checks, yes.

Here are some alternative ways forward.

1. Keep things as they are. The return type is _Operand, and we can type hint it as such and use a # type: ignore[override] to prevent mypy saying

error: Return type "_Operand" of "__eq__" incompatible with return type "bool" in supertype "object"

2. We could made this breaking change only for lambda_eval() and unsafe_eval(). It would still be a breaking change for those methods (added in 10.3.0), but eval() would stay the same until it is removed in Pillow 12. Not a perfect solution, but an improvement.

[hugovk]

@python-pillow/pillow-team What do you think?

[aclark4life]

Probably one of the few good times to make a breaking change (from 10.x to 11) and worst case you could follow up with options 1 or 2 in 11.0.1 if some unanticipated breakage occurred in the wild. 🤷

[hugovk]

We've usually been cautious about breaking changes without at least a year of deprecations.

I would prefer alternative 1 or 2.

This PR is to improve the API, but motivated by adding type hints, and not user demand, so I would err on bring cautious.

[radarhere]

Ok, I've pushed a commit to implement 2 - only changing the behaviour for lambda_eval() and unsafe_eval(), the methods that were added in 10.3.0.

[radarhere]

I've created #8464 to implement 1 - I've realised that the current form does have conceivably useful functionality. For example, you can invert the upper left quadrant of hopper.

>>> from PIL import Image, ImageMath
>>> im = Image.open("Tests/images/hopper.png").convert("1")
>>> b = Image.new("1", (128, 128))
>>> for x in range(64):
...   for y in range(64):
...     b.putpixel((x, y), 1)
... 
>>> ImageMath.eval("(A != B) * 65535", A=im, B=b).save("out.png")