Added IGNORE_DECODING_ERRORS flag to ImageFile allowing corrupted images to load

[cezarsa]

Since commit 33bf5d9 PIL is no longer able to load this image even after setting ImageFile.LOAD_TRUNCATED_IMAGES = True because of the t == 0 check that was introduced then.

Because I'm not sure about what use cases @d-schmidt had in mind when the size check was added, I've introduced a new flag called IGNORE_DECODING_ERRORS that if True will never raise IOError. This should make everyone happy.

[aclark4life]

Thanks

[d-schmidt]

A broken image? Great :) Will check this out tomorrow.

[d-schmidt]

I didn't know t can be zero if image data is loaded. I just assumed, without proof, that t is only 0 if there is no image data at all.

[cezarsa]

You have to set it manually after importing ImageFile.

from PIL import ImageFile
ImageFile.IGNORE_DECODING_ERRORS = True

Maybe this extra flag is unnecessary and we could just change the behavior of LOAD_TRUNCATED_IMAGES. I'll look for more partially corrupted jpeg images and write a test case using them. It should make it easier to figure out if we need both.

[aclark4life]

Got it, thanks

[d-schmidt]

Imho revert this.
The file opens just fine with only my old flag set true. Using current pillow and jpeglib8. Don't know whats your problem @cezarsa. t is 65406 the moment it breaks truncated.
Please test it again with just

ImageFile.LOAD_TRUNCATED_IMAGES = True

[wiredfool]

Works for me (ubuntu 12.04/py2,3) with @d-schmidt 's old flag. +1 revert

[aclark4life]

Yep, already reverted