Update of Pr #2284, Complex Text Support

[wiredfool]

Includes #2284.

So far, fixes:

• rebased
• feature detection
• Testing both branches when raqm is installed

Todo/test/check

• install docs/alternate platforms
  • 5 of the dockers now have raqm available.
• install detection/enable/disable
• font license
• review docs
• ppa check

Additional note, we should add the Deja license to the font licence directory as well: https://dejavu-fonts.github.io/License.html

[Fahad-Alsaidi]

to build libraqm natively on windows or OSX, you can libraqm-cmake repo.

[khaledhosny]

Any idea what corresponds to these failing lines in Makefile.am?

[wiredfool]

No, I'm afraid I haven't dived into automake.

But I just tried it on Freebsd with the cmake version and it builds and passes tests with that version. So that part is good anyway. I've noticed that the cmake and other repo are slightly out of sync -- would it be possible to sync them up and declare a 0.3.0 release, or add the cmake portions to the main repo?

[wiredfool]

With the cmake version of libraqm, I'm getting one test failure on osx:

(vpy27)Erics-MacBook-Air:Pillow erics$ ./test-installed.py Tests/test_imagefontctl.py
..F.......
======================================================================
FAIL: TestImagecomplextext.test_complex_unicode_text
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/Users/erics/Pillow/Tests/test_imagefontctl.py", line 55, in test_complex_unicode_text
    self.assert_image_similar(im, target_img, .5)
  File "/Users/erics/Pillow/Tests/helper.py", line 110, in assert_image_similar
    ave_diff, epsilon))
AssertionError:  average pixel value difference 12.5302 > epsilon 0.5000

With an image like this:

Which is clearly completely failing.

Install procedure here was roughly:

brew install fribidi harfbuzz cmake

• clone libraqm-cmake
• build with prefix /usr/local
• build pillow.

[Fahad-Alsaidi]

It is weird. I see this draw.text((0, 0), u'السلام عليكم', font=ttf, fill=500) but there is no u notation in the other tests. Is it OK? try to remove it to be synced with the other tests.

[wiredfool]

Ok. Current state is:

• Generally, all un*x platforms work that I've tried. linux, freebsd, mac/brew
• Windows is going to be a treat due to getting harfbuzz and firbidi compiled with msvc.
• Just added direction/features to getsize. Need to test that with some of the features, kerning, possibly ttb.

Almost ready at this point.

[Fahad-Alsaidi]

For windows, you can benefit from Raqm dependencies project .

[wiredfool]

The trouble with the dependencies on windows is that as far as I know, Python needs the compiler to be consistent between the Python version and its extensions. So mingw isn't going to work.

I'm going to merge for now and we'll deal with windows later. The compilation issues are only going to get better over time.