How to render stacked Tibetan letters

Hello Pillow folks, I've come across this issue with stacked texts in Tibetan (po-ed stackoverflow question here).

from PIL import Image, ImageFont, ImageDraw
import matplotlib.pyplot as plt

test_letters = [u'ཀ', u'ར',u"སྤ",u'རྐ',u'༄',u'ཀྱ',u'ཛྷ']

font = ImageFont.truetype("/Libarary/Fonts/Kokonor.ttf", 20, encoding='unic')

for letter in test_letters:
    img = Image.new("RGBA", (28, 28))
    draw = ImageDraw.Draw(img)
    draw.text((6, 0),letter,(0,0,0),font)
    plt.imshow(img)
    plt.grid()
    plt.show()

One of the output is like this one, displayed apart.
.

There are some links on Tibetan font rendering, which might of some value.

Thanks for any illumination.

I suspect it's the same underlying issue that's fixed in #2284.

@wiredfool

@wiredfool with libraqm compiled and installed, and pip installed Fahad-Alsaidi/Pillow@CTL branch, do I need to alter anything to solve this issue? It seems that the letters still don't stack.

If that's the case, then there are 3 possibilities:

The CTL branch didn't actually compile correctly and the feature isn't available. One issue with that PR is that the feature reporting is incorrect. I've got a fix, and another PR that it relied on is merged, so it's time to get back to that and push it off my machine. It's possible that you have raqm installed, but Pillow couldn't find harfbuzz or fribdi.
The CTL branch is supposed to handle this case, and doesn't do it correctly.
It's a separate issue. From a quick google, it does appear that harfbuzz is supposed to to Tibetan script, so I'm going to count this as unlikely.

So, I've got a branch where I'm testing out the complex text layout PR (#2576) and using the following script, I've got something that appears to match what I'm seeing in the browser and libreoffice on Ubuntu 16.04. This is using the font package 'fonts-tibetan-machine', and displaying the whole thing as one string rather than individual glyphs in matplotlib.

#!/usr/bin/env python
# -*- coding: utf-8 -*-

from PIL import Image, ImageFont, ImageDraw

test_str = u'ཀ ར སྤ རྐ ༄ ཀྱ ཛྷ'

font = ImageFont.truetype("TibetanMachineUni.ttf", 40, encoding='unic')

img = Image.new("RGBA", (320, 60), 'white')
draw = ImageDraw.Draw(img)
draw.text((6, 0),test_str,(0,0,0),font)
img.save('issue_2572.png')

To my eye, it looks correct, but I'm not an expert by any means.

[edit] Now this is strange, it appears to just work without the complex text support as well, on my old 12.04 instance which doesn't have the pre-requisites for raqm. So maybe it's a font thing?

I get this on master (Python 2.7.13, macOS, with no RAQM pre-requisites):

And this on wiredfool/pr_2284:

Is the top one a cropping issue, or is it a rendering thing?

Looks like rendering. With 640px width:

Highlighting for emphasis:

That's odd. Wonder if it's a freetype version issue, or if it's a font thing. It looks like you have the same font, so that's not likely to be it.

I used this font: https://github.com/cscott/mw-ocg-latexer/blob/master/fonts/tibetan-machine/TibetanMachineUni.ttf

Freetype version?

Freetype 2.7.1: /usr/local/Cellar/freetype/2.7.1

12.04:

(vpy27)erics@builder-1204-x64:~/Pillow$ dpkg -l | grep -i freetype
ii  libfreetype6                         2.4.8-1ubuntu2.4                    FreeType 2 font engine, shared library files

16.04:

(vpy27) erics@builder-16:~/Pillow$ dpkg -l | grep freetype
ii  libfreetype6:amd64                2.6.1-0.1ubuntu2.3                         amd64        FreeType 2 font engine, shared library files

Hmm.

@hugovk

Thanks @hugovk and @wiredfool It seems that fonts are handled with some preference. To notice although some of them are stacked, they are not rendered properly. Kokonor and Kailasa are default on macOS, and rendered all right when typed as texts.

I've checked this with the mac fonts on an Ubuntu 16.04 machine, and it's the font, not the freetype version. So there's something different in the fonts.

@Fahad-Alsaidi

@Fahad-Alsaidi Can you take a look at this and see if raqm + harfbuzz is properly parsing this as Tibetan and enabling the appropriate script options? If not, is there a way to override any detected script and pass in Tibetan, either as a feature or otherwise? (i.e., I'm seeing 'TIB ' and 'Tibt' in the harfbuzz source, but neither seem to affect the rendering)

well, have you tried this raqm_set_language (). I think this isn't implemented yet in the original PR.

Mac system fonts are likely to be AAT not OpenType fonts, so they will not work on Windows or Linux correctly. On Mac AAT fonts should work if HarfBuzz is compiled with Core Text support.