gistfile1.py

#!/usr/bin/env python

from __future__ import print_function

import collections
import fileinput
import sys

plop = collections.defaultdict(list)

try:
  import fontTools.agl

  def map_character(c):
    try:
      return fontTools.agl.UV2AGL.get(ord(c), c)
    except:
      return c
except ImportError:
  print('WARNING: Package fonttools not found.', file=sys.stderr)
  # Don't have fonttools package.
  # TODO: 5 ---> five etc.
  character_mappings = {
    ' ': 'space',
  }

  def map_character(c):
    return character_mappings.get(c, c)

for word in fileinput.input():
  word = word.strip()
  length = len(word)
  string = ''
  glue = ''
  firstletter = ''

  if length < 6:
    yooniecode = 'uniE602'
  elif length > 11:
    yooniecode = 'uniE600'
  else:
    yooniecode = 'uniE601'
  firstletter = word[0]
  string = ','.join(map(map_character, word[1:]))

  line = '  <Ligature components="' + string + '" glyph="' + yooniecode + '"/>'

  plop[firstletter].append(line)

for ltr in plop:
  print('<LigatureSet glyph="' + ltr + '">')
  print("\n".join(plop[ltr]))
  print('</LigatureSet>' + "\n")

@DonaldTsang I got something messy but working at https://github.com/dperelman/ligature_replacer . All of the work happens inside fontTools in Python.

Basic usage:

./rewrite_ligatures.py SansBullshitSans.ttf --write-word-list words.txt --write-glyphs-json glyphs.json
# edit words.txt
./rewrite_ligatures.py SansBullshitSans.ttf --read-word-list words.txt --read-glyphs-json glyphs.json -o out.ttf

Will read this once I get home. Looks interesting.

Looking more carefully at Project Seen, it actually doesn't work quite the same way. It still uses ligatures, but it defines a separate ligature glyph for each word unlike SansBullshitSans which defines a small number of them and reuses them for words of the approximately the same length. For the "blackout" version of Project Seen it doesn't really matter (although my script will do a poor job of automatically identifying which ligature glyphs to use because it finds them by looking for repeats), but for the versions that leave the text visible, there's actual graphical manipulation involved which is much more complicated. I couldn't find any source code for Project Seen although I assume the author used a script of some sort to generate it from a word list.

the blackout version sounds good enough for me, if the script can detect the length of the combined ligatures that would be useful.