Exif headers with zero-denominator rational (fraction) types raise DivideByZero exception unnecessarily

[Hobson Lane]

Applies to pyexiv2 version 0.3
Python 2.7.1+ (r271:86832, Apr 11 2011, 18:05:24)
distributed with Ubuntu Natty Narwhal

#!/bin/python
# show_bug2.py
import pyexiv2
im = pyexiv2.ImageMetadata('bora_bora--stone_tiki.JPG')
im.read()
k = 'Exif.Nikon3.FocusDistance'
s = im[k].value;
print "{0}: {1}".format(k,s)

hobs@hobs-laptop:~/Photos/malformed_exif_examples$ ipython show_bug2.py
ERROR: An unexpected error occurred while tokenizing input
The following traceback may be corrupted or invalid
The error message is: ('EOF in multi-line statement', (33, 0))

---------------------------------------------------------------------------
ExifValueError Traceback (most recent call last)

/media/Win7/Users/Hobs/Documents/Photos/malformed_exif_examples/show_bug2.py in <module>()
      4 im.read()
      5 k = 'Exif.Nikon3.FocusDistance'
----> 6 s = im[k].value;
      7 print "{0}: {1}".format(k,s)
      8

/usr/lib/python2.7/dist-packages/pyexiv2/exif.pyc in _get_value(self)
    187 def _get_value(self):
    188 if self._value_cookie:
--> 189 self._compute_value()
    190 return self._value
    191

/usr/lib/python2.7/dist-packages/pyexiv2/exif.pyc in _compute_value(self)
    182 return
    183
--> 184 self._value = self._convert_to_python(self._raw_value)
    185 self._value_cookie = False
    186

/usr/lib/python2.7/dist-packages/pyexiv2/exif.pyc in _convert_to_python(self, value)
    320 r = make_fraction(value)
    321 except (ValueError, ZeroDivisionError):
--> 322 raise ExifValueError(value, self.type)
    323 else:
    324 if self.type == 'Rational' and r.numerator < 0:

ExifValueError: Invalid value for EXIF type [Rational]: [0/0]
WARNING: Failure executing file: <show_bug2.py>

[Hobson Lane (hobs)]

Here's a work-around patch for this bug and a related bug, #781464.

[Olivier Tilloy (osomon)]

pyexiv2 is voluntarily strict with value parsing, so as a rule of thumb accessing the value should always be enclosed in a try…catch statement if the metadata is not known to be always well-formed. You can still access the raw value (unparsed) without errors ("0/0" in this case).

However in this specific case it seems that it’s not uncommon to get invalid fractions of "0/0", so pyexiv2 could indeed be made more fault-tolerant.

Your patch is correct but too invasive though, it contains unrelated changes. I’ll keep the relevant changes and update the unit tests accordingly, thanks a lot!