Add numpy array interface support for 16 and 32 bit integer modes

[cgohlke]

Fix issue #273.
Please review.
Needs test.

[wiredfool]

I'll try to exercise this tomorrow morning.

Having added this, do you have any ideas about why this is labeled broken? https://github.com/cgohlke/Pillow/blob/9bff0b14e0b4e3b6f53d2251c186f208125bd7d9/PIL/Image.py#L217

[cgohlke]

It looks like the tobytes function does not work correctly for mode 1 images. For a 73x76 1 bit TIFF file the function returns 760 bytes (not 5548 bytes), the size of the undecoded image data in the file.

[wiredfool]

Ok, I've had a chance to review this. The Unsigned 16bit Integer modes work as expected. The others appear correct, but I don't think you can actually make an image with those modes. You can't make a new image or convert to them. It looks like you can read a 16bit signed tiff, and there are several plugins that read 32 bit int images. It appears that they're always stored in local endianess.

Summary: I think they're right, and even if they're not, it's not going to be that damaging since it's really hard to make that type of images, and if there is a problem, we can fix the format code easily.

[gregcotten]

So what is the exact status of this?

[cgohlke]

This PR has been merged and closed.