Wand should not rely on default encoding on Python 2.x

[jgehrcke]

Example code:

# -*- coding: utf-8 -*-
from wand.drawing import Drawing
from wand.image import Image
import sys


def annotate(path, isovalue):
    with Drawing() as draw:
        with Image(filename=sys.argv[1]) as image:
            draw.font_size = 20
            draw.gravity = "south_west"
            draw.text(0, 0, u'Φ')
            draw(image)
            image.save(filename="out.png")

With Wand 0.3.5, Python 2.7 on Linux (Ubuntu 12.04) this results in

  File "/projects/bioinfp_apps/Python-2.7.3/lib/python2.7/site-packages/wand/drawing.py", line 431, in text
    body = binary(body)
  File "/projects/bioinfp_apps/Python-2.7.3/lib/python2.7/site-packages/wand/compat.py", line 44, in binary
    return string.encode()
UnicodeEncodeError: 'ascii' codec can't encode character u'\u03a6' in position 0: ordinal not in range(128)

The reason is that in compat.py (https://github.com/dahlia/wand/blob/master/wand/compat.py#L44) we are currently doing a simple return string.encode(). If called without arguments, encode uses the current default string encoding which is ASCII in case of Python 2. Obviously, ASCII cannot encode my greek character there and the UnicodeEncodeError is thrown.

We cannot and dont want to change the fact that ASCII is Python 2's default encoding. Relevant discussion about this topic can be found here:

• https://fedoraproject.org/wiki/Features/PythonEncodingUsesSystemLocale
• http://article.gmane.org/gmane.comp.python.devel/109916

So the solution would be to make a well-grounded decision about which encoding to use. I tried UTF-8 and it worked for me (I changed line 44 of compat.py accordingly). I am not entirely sure why this works, for this I do not understand enough of which data format the imagemagick binaries do expect.

[dahlia]

ImageMagick actually seems to provide a property to set string encoding (DrawGetTextEncoding() and DrawSetTextEncoding()) while Wand does not expose them. I will work on this soon. Thanks for your reporting!

[jgehrcke]

Just my thoughts, you may of course do something else :-): then Wand maybe should not directly expose DrawSetTextEncoding, but rather use it internally and set it to UTF-8 (IM's docks say The only character encoding which may be specified at this time is "UTF-8"). Wand could require people to use unicode objects in Python code, which would then (internally, again) become encoded as UTF-8 before passing to ImageMagick.

[dahlia]

I made a small change ☝️ for this.

[jgehrcke]

Great, thanks.