[usmanmateen]

Fixes #10372.

Changes proposed in this pull request:

• Clip F to I conversion to the INT32 range in f2i, and map NaN to 0. The cast had no range check, so on x86 every value above the range, inf and NaN became -2147483648, and an I image at 2147483647 converted to F and back did too. These casts are undefined behavior in C, so the result also depended on the CPU.
• Add a test with values at and past both ends of the range, inf, -inf and NaN.

>>> Image.new("F", (1, 1), float("inf")).convert("I").getpixel((0, 0))
-2147483648

With this change, it is 2147483647, the same clipping clip32() in Filter.c does. NaN becomes 0, which is what F to L and F to I;16 give.

[radarhere]

At the moment in main, NaN is rejected by L mode images, but not F.

>>> from PIL import Image
>>> Image.new("L", (1, 1), float("nan"))
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "PIL/Image.py", line 3277, in new
    return im._new(core.fill(mode, size, color))
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
TypeError: color must be int or single-element tuple
>>> Image.new("F", (1, 1), float("nan"))
<PIL.Image.Image image mode=F size=1x1 at 0x105DCE3D0>

I've pushed a commit to reject it for F mode as well, and also rearranged the code to be more in line with

Pillow/src/libImaging/Convert.c

Line 585 in 22e8391

*out = (UINT8)(v <= 0 ? 0 : (v >= 255 ? 255 : v));

See what you think.

[radarhere]

While my last commit added helpful validation, somehow, I missed that it did not actually raise an error for NaN. I've created #10441 to follow up.

[radarhere]

#10372 (comment)

NaN mostly comes in with the data rather than from Image.new(). Image.fromarray(), Image.frombytes() and opening a float TIFF with NaN pixels all keep it, and all three convert to I as -2147483648. NaN is valid float data in those cases (it's often the nodata value in float TIFFs), so I don't think rejecting it would be right.

Ah, sorry, forgot about this. Just as well that I missed it then.

[radarhere]

Here are the current results with the merged form of this PR. Let me know what you think.

>>> from PIL import Image
>>> Image.new("F", (1, 1), float("inf")).convert("I").getpixel((0, 0))
2147483647
>>> Image.new("F", (1, 1), float("nan")).convert("I").getpixel((0, 0))
-2147483648
>>> Image.new("F", (1, 1), float("-inf")).convert("I").getpixel((0, 0))
-2147483648