hb-subset: Ligatures and GDEF/GPOS/GSUB not included in the subset font

[papandreou]

Probably a known issue, but I was adviced to open an issue here so it can be tracked 🤗

This Roboto-400.ttf that I downloaded from Google Web Fonts produces an incomplete file when I subset it with hb-subset, resulting in ligatures not rendering. Here's a visual diff with magenta showing where the rendering is different from the original:

I tried using ttx to compare the output of hb-subset to that of pyftsubset invoked with similar options:

pyftsubset --text="Waffle stuffings" --output-file=roboto-pyftsubset.ttf Roboto-400.ttf
../harfbuzz/build/util/hb-subset --output-file=roboto-hbsubset.ttf Roboto-400.ttf "Waffle stuffings"
diff -U1000 <(ttx -o - roboto-pyftsubset.ttf) <(ttx -o - roboto-hbsubset.ttf) > pyftsubset_vs_hbsubset_ttx.diff

The ttx diff shows that the 4 ligature glyphs are missing from <GlyphOrder> and <glyf>, and that the <GDEF>, <GPOS>, and <GSUB> blocks are missing altogether.

Tested on Ubuntu 19.10 with harfbuzz compiled from source (8e5bc53).

[ebraminio]

@garretrieger @qxliu76 can you have a look if anything unexpected here, weird that all of the tables are dropped, I thought some GPOS subsetting was implemented.

[garretrieger]

Since GSUB/GPOS subsetting is not yet fully implemented its disable by default and the two tables are just dropped. You can re-enable subsetting those tables by adding "--drop-tables-=GSUB,GPOS,GDEF" to your subset command. We currently support nearly all lookup types so that should work fine for something like Roboto.

[papandreou]

@garretrieger, thanks! Adding --drop-tables-=GSUB,GPOS,GDEF to the command line makes all my primitive test cases pass 😌

Does anyone know how to achieve the equivalent of that when using hb-subset.wasm via harfbuzzjs? (context: harfbuzz/harfbuzzjs#9 (comment))

[garretrieger]

Similar process, use "hb_subset_input_drop_tables_set" to get the set of tables to be dropped and then remove GSUB, GPOS, and GDEF from the set. For example see the subset fuzzer code: https://github.com/harfbuzz/harfbuzz/blob/master/test/fuzzing/hb-subset-fuzzer.cc#L26

[ebraminio]

Oooh, you'll need something like this then,

    exports.hb_set_del(export.hb_subset_input_drop_tables_set(input), HB_TAG('G', 'S', 'U', 'B'));
    exports.hb_set_del(export.hb_subset_input_drop_tables_set(input), HB_TAG('G', 'P', 'O', 'S'));
    exports.hb_set_del(export.hb_subset_input_drop_tables_set(input), HB_TAG('G', 'D', 'E', 'F'));

hb_set_del weren't in hb-subset wasm build harfbuzz/harfbuzzjs@8c27ab1 and this should make it work, the only left part will be a HB_TAG like thing, which you'll need something like this or you can hard code the answer anyway.

[papandreou]

@ebraminio @garretrieger, thank you so much! It works!

I'm looking forward to feature-parity with pyftsubset and beyond, but this is probably good enough to get the harfbuzz support into subfont and put it behind an "experimental" switch 🚀

(Would be nice with a new harfbuzzjs release soon so I don't have to ship the hb-subset.wasm I compiled myself 🙏)

[ebraminio]

(Would be nice with a new harfbuzzjs release soon so I don't have to ship the hb-subset.wasm I compiled myself pray)

Oh, sure! https://github.com/harfbuzz/harfbuzzjs/releases/tag/0.1.3 and of course the updated hb-subset.wasm is included in npm also.

btw, chunkName.split('').reduce(function (a, x) { return (a << 8) + x.charCodeAt(0); }, 0); also can be used but that may make it too cryptic so use that only if you also feel better with it :)

Guess you can close this and report other issues separately. Thanks! :)

[papandreou]

Yes, thanks a lot! 🙇