OcrdExif: Pillow's IFDRational is unsupported

[bertsky]

I get the following error when using OcrdExif on (most of) the TIFFs in the Metastore GT Bags:

14:45:37.787 INFO ocrd.task_sequence - Start processing task 'cis-ocropy-clip -I OCR-D-GT-SEG-LINE -O OCR-D-GT-SEG-LINE-CLIP'
14:45:38.968 INFO processor.OcropyClip - INPUT FILE 0 / phys_0001
Traceback (most recent call last):
  File "env3/bin/ocrd-cis-ocropy-clip", line 11, in <module>
    load_entry_point('cis-ocrd', 'console_scripts', 'ocrd-cis-ocropy-clip')()
  File "env3/lib/python3.6/site-packages/click/core.py", line 764, in __call__
    return self.main(*args, **kwargs)
  File "env3/lib/python3.6/site-packages/click/core.py", line 717, in main
    rv = self.invoke(ctx)
  File "env3/lib/python3.6/site-packages/click/core.py", line 956, in invoke
    return ctx.invoke(self.callback, **ctx.params)
  File "env3/lib/python3.6/site-packages/click/core.py", line 555, in invoke
    return callback(*args, **kwargs)
  File "ocrd_cis/ocropy/cli.py", line 27, in cis_ocrd_ocropy_clip
    return ocrd_cli_wrap_processor(OcropyClip, *args, **kwargs)
  File "env3/lib/python3.6/site-packages/ocrd/decorators.py", line 38, in ocrd_cli_wrap_processor
    run_processor(processorClass, ocrd_tool, mets, workspace=workspace, **kwargs)
  File "env3/lib/python3.6/site-packages/ocrd/processor/base.py", line 65, in run_processor
    processor.process()
  File "ocrd_cis/ocropy/clip.py", line 96, in process
    self.workspace, page, page_id)
  File "ocrd_cis/ocropy/common.py", line 776, in image_from_page
    page_image_info = OcrdExif(page_image)
  File "env3/lib/python3.6/site-packages/ocrd_models/ocrd_exif.py", line 54, in __init__
    self.resolution = round(sqrt(self.xResolution * self.yResolution))
TypeError: unsupported operand type(s) for *: 'IFDRational' and 'IFDRational'
Traceback (most recent call last):
  File "env3/bin/ocrd", line 10, in <module>
    sys.exit(cli())
  File "env3/lib/python3.6/site-packages/click/core.py", line 764, in __call__
    return self.main(*args, **kwargs)
  File "env3/lib/python3.6/site-packages/click/core.py", line 717, in main
    rv = self.invoke(ctx)
  File "env3/lib/python3.6/site-packages/click/core.py", line 1137, in invoke
    return _process_result(sub_ctx.command.invoke(sub_ctx))
  File "env3/lib/python3.6/site-packages/click/core.py", line 956, in invoke
    return ctx.invoke(self.callback, **ctx.params)
  File "env3/lib/python3.6/site-packages/click/core.py", line 555, in invoke
    return callback(*args, **kwargs)
  File "env3/lib/python3.6/site-packages/ocrd/cli/process.py", line 26, in process_cli
    run_tasks(mets, log_level, page_id, tasks)
  File "env3/lib/python3.6/site-packages/ocrd/task_sequence.py", line 95, in run_tasks
    raise Exception("%s exited with non-zero return value %s" % (task.executable, returncode))
Exception: ocrd-cis-ocropy-clip exited with non-zero return value 1

It does not happen on loeber_heuschrecken_1693 and wecker_kochbuch_1598 (but all 34 other bags).

We already chased this exact error in a Travis build for Python 3.4 here, but this time the error occurs on Python 3.6.5 (and Pillow 5.4.1)!

I fail to see any reason why the class IFDRational would not allow the multiplication operator.

[bertsky]

This is critical, because it drags down our current image_from_page with it.

Those 2 bags which evade the error are from the subset we already identified as coming from lossy JPEGs. And, it seems, these 2 are the ones which show no dpi info in Pillow at all (all others do).

The error disappears with Pillow 6.1.0 BTW.

[bertsky]

So can someone please determine whether #238 still happens with 6.1.0?

@kba @finkf I traced your original description in the Gitter log and you reported the problem on a file called eval-grenzboten-shuffle/trainws/179396.tif – can you please publish this somewhere (or try to reproduce with 6.1.0 yourself)?

[kba]

Here's the TIFF. 179470.zip Debugging this now.

[kba]

So can someone please determine whether #238 still happens with 6.1.0?

Unfortunately, yes:

def test_pil_version(self):
    """
    Test segfault issue in PIL TiffImagePlugin

    Run the same code multiple times to make segfaults more probable

    Should fail persistently:
        5.3.1 no
        5.4.1 no
        6.0.0 yes
        6.1.0 yes
    """
    for _ in range(0, 10):
        pil_image = Image.open(assets.path_to('grenzboten-test/data/OCR-D-IMG-BIN/p179470'))
        pil_image.crop(box=[1539, 202, 1626, 271])

throws random segfaults for Pillow 6.0.0 and 6.1.0 :-/

I'll check whether there's a workaround in OcrdExif...