rsheeter/gf_glyph_expansion

Exploratory hackery in search of opportunities to add glyphs to Google Fonts to expand their language coverage

gf_glyph_expansion

Exploratory hackery in search of opportunities to add glyphs to Google Fonts to expand their language coverage

Valuing an extension

To assess ROI we need to value both cost and benefit. Both are done in magical ROI units.

Value is taken to be value / cost.

Value

• Popularity gives a base score, loosely tied to magnitude of traffic
  • 50 for top 10
  • 25 for top 50
  • 10 for top 150
  • 5 for top 500
  • otherwise 1
• Add up the populations of the added languages and multiply based on total:
  • 20 for >= 1 billion
  • 10 for >= 0.1 billion
  • 5 for >= 0.01 billion
  • 2 for >= 1 million
  • 1.1 for >= 0.1 million
  • otherwise 1
• Ease of update
• Other value signals
  • For example, we might boost known growth areas

Cost

Based on discussion with @davelab6, assume that:

• Getting setup to modify a font at all is expensive
  • Glyph 1 costs 1 unit
  • Glyph 2..n cost 0.1 units each
  • 1 glyph for 1 unit, 11 glyphs for 2 units, etc
• Modifying multiple static fonts or a variable font costs more than a single static font
  • For a static family, multiply cost by 0.9 + 0.1 * number of fonts
    • x 1.0 for a single static font, x 1.8 for a family with 9 static weights, etc
  • For a variable family, multiply cost by 1 + 2^(num_axes - 1)
    • If there are multiple fonts (e.g. Regular/Italic) multiply by 1.5
    • x 2.0 for weight variable, x 3.0 for weight+width variable, x 17.0 for Recursive with 5 axes
    • This makes adding glyphs to Flex fonts very expensive ... which is perhaps accurate?

TODO: multiply cost by 2 (?) if family is hard to update

• data source https://github.com/googlefonts/gf-dashboard/blob/main/cache.json (courtesy of Simon)
• A family is hard to update unless it has_upstream and last_updated is within ... maybe 2 years ...

Usage

# in a venv
$ pip install -e .
$ gf_glyph_expansion
9295 to add <=  1 and support at least one new language
$ gf_glyph_expansion --max_missing 10
319388 to add <=  10 and support at least one new language
$ gf_glyph_expansion --max_missing 10 --family_filter "^Roboto$"
180 to add <=  10 and support at least one new language

Note that every (family, chars) pair is counted as one so the number of results for one family can be > 1.

Derived from Simon's search for ẞ

# Run at root of google/fonts checkout
import glob
from fontTools.ttLib import TTFont
from gflanguages import LoadLanguages, parse
from gftools import fonts_public_pb2
from gftools.util import google_fonts as fonts
import os

langs = LoadLanguages()

# All the German base exemplars without the capital eszett
cps = parse(langs["de_Latn"].exemplar_chars.base)
cps -= set("ẞ")
counter = 0

for mdfile in glob.glob("ofl/*/METADATA.pb"):
    # Find an exemplar font for this family
    try:
        family_metadata = fonts.ReadProto(fonts_public_pb2.FamilyProto(), mdfile)
        exemplar_font_fp = os.path.join(
            os.path.dirname(mdfile), fonts.GetExemplarFont(family_metadata).filename
        )
        exemplar_font = TTFont(exemplar_font_fp)
    except Exception as e:
        continue
    # All encoded codepoints in the font
    exemplar_cps = set(chr(x) for x in exemplar_font.getBestCmap().keys())
    # Does this font have all the german bases, but does not have capital eszett?
    if exemplar_cps.issuperset(cps) and "ẞ" not in exemplar_cps:
        counter += 1

print(counter)