TIFF_LZW decompression is incorrect for 16bpc image

[giorgiosironi]

Pillow was refusing to open this TIFF in version 4.2.1. In version 4.3.0, it loads it, but then saving it in any format including TIFF produces a silent corruption. Python 2.7.14 in both cases.

This is the reproducing script:

from PIL import Image                                                                                                                      

# https://s3-external-1.amazonaws.com/prod-elife-published/articles/32330/elife-32330-fig1-v1.tif 
src_fp = 'elife-32330-fig1-v1.tif'
im = Image.open(src_fp)
target_fp = open('result.tif', 'w')
im.save(target_fp)

In 4.2.1:

  File "reproducing.py", line 5, in <module>
    im = Image.open(src_fp)
  File "/opt/loris/venv/local/lib/python2.7/site-packages/PIL/Image.py", line 2519, in open
    % (filename if filename else fp))
IOError: cannot identify image file 'elife-32330-fig1-v1.tif'

In 4.3.0, no output, file gets stored and corrupted.
elife-32330-fig1-v1.tif.gz
result.tif.gz

I tried to track how the Image class changed from 4.2.1 to 4.3.0, but there were too many changes to identify the source. I don't expect it to start to work, but to continue throwing an exception if it can't be read safely.

[wiredfool]

If you use libtiff instead of the internal tiff decoder, it works:

from PIL import Image, TiffImagePlugin
TiffImagePlugin.READ_LIBTIFF=True

im = Image.open('issue_2839.tif')
im.show()

This image is a 16bit RGBA image, with tiff_lzw compression. Pillow 4.3.0 added support for the 16 bit RGBA type, though the output image is a lower precision 8 bit per channel RGBA image. Prior to that, there was no support for this image.

Pillow (and PIL before it) have had internal support for tiff_lzw compression, but apparently it's either buggy or cannot handle this particular image mode.

Forcing the TiffImagePlugin to use the libtiff decoder allows Pillow to properly decode this image.