[PDFBOX-3759] TTFSubsetter scrambles OpenType ligature and kerning tables (and probably others as well)

• Type: Bug
• Status: Open
• Priority: Critical
• Resolution: Unresolved
• Affects Version/s: 2.0.5
• Component/s: FontBox
• Assignee: Unassigned
• Reporter: Tobias Fischer
• Created: 18/Apr/17 12:04
• Updated: 19/Apr/17 19:10

Description

When using the TTFSubsetter as standalone software and subsetting an OpenType/TrueType font, the liga OpenType ligature table gets scrambled "randomly".

E.g. when viewing the source font "DejaVuSans.ttf" in FontLab Studio, the OpenType liga table is shown as follows:

feature liga {
 script arab;
lookup liga33 {
lookupflag IgnoreMarks RightToLeft;
    sub uniFEDF uniFE88 by uniFEF9;
    sub uniFEDF uniFE84 by uniFEF7;
    sub uniFEDF uniFE82 by uniFEF5;
    sub uniFEE0 uniFE88 by uniFEFA;
    sub uniFEE0 uniFE84 by uniFEF8;
    sub uniFEE0 uniFE82 by uniFEF6;
} liga33;
lookup liga35 {
lookupflag RightToLeft;
    sub space uni0652 by uniFE7E;
    sub space uni0651 by uniFE7C;
    sub space uni0650 by uniFE7A;
    sub space uni064F by uniFE78;
    sub space uni064E by uniFE76;
    sub space uni064D by uniFE74;
    sub space uni064C by uniFE72;
    sub space uni064B by uniFE70;
    sub uni0627 uni0655 by uni0625;
    sub uni0627 uni0654 by uni0623;
    sub uni0627 uni0653 by uni0622;
    sub uni0640 uni0652 by uniFE7F;
    sub uni0640 uni0651 by uniFE7D;
    sub uni0640 uni0650 by uniFE7B;
    sub uni0640 uni064F by uniFE79;
    sub uni0640 uni064E by uniFE77;
    sub uni0640 uni064B by uniFE71;
    sub uni0648 uni0654 by uni0624;
    sub uni064A uni0654 by uni0626;
} liga35;
 language SND ;
 language URD ;
 language KUR ;
 script latn;
    sub f f l by uniFB04;
    sub f f i by uniFB03;
    sub f l by fl;
    sub f i by fi;
    sub f f by uniFB00;
 language MOL ;
 language ROM ;
 language ISM ;
 language KSM ;
 language LSM ;
 language NSM ;
 language SSM ;
 language SKS ;
} liga;

This ligature table shows for example, in the latn section, the ffl substitution with the ffl ligature \uFB04.

However, when viewing a subsetted version of the font (like the one attached), the liga table is scrambled:

feature liga {
 script arab;
lookup liga33 {
lookupflag IgnoreMarks RightToLeft;
    sub _5337 _5250 by _5363;
    sub _5337 _5246 by _5361;
    sub _5337 _5244 by _5359;
    sub _5338 _5250 by _5364;
    sub _5338 _5246 by _5362;
    sub _5338 _5244 by _5360;
} liga33;
lookup liga35 {
lookupflag RightToLeft;
    sub numbersign _1403 by _5240;
    sub numbersign _1402 by _5238;
    sub numbersign _1401 by _5236;
    sub numbersign _1400 by _5234;
    sub numbersign _1399 by _5232;
    sub numbersign _1398 by _5231;
    sub numbersign _1397 by _5229;
    sub numbersign _1396 by _5227;
    sub _1365 _1406 by _1363;
    sub _1365 _1405 by _1361;
    sub _1365 _1404 by _1360;
    sub _1385 _1403 by _5241;
    sub _1385 _1402 by _5239;
    sub _1385 _1401 by _5237;
    sub _1385 _1400 by _5235;
    sub _1385 _1399 by _5233;
    sub _1385 _1396 by _5228;
    sub _1393 _1405 by _1362;
    sub _1395 _1405 by _1364;
} liga35;
 language SND ;
 language URD ;
 language KUR ;
 script latn;
    sub o o u by _5045;
    sub o o r by _5044;
    sub o u by _5043;
    sub o r by _5042;
    sub o o by _5041;
 language MOL ;
 language ROM ;
 language ISM ;
 language KSM ;
 language LSM ;
 language NSM ;
 language SSM ;
 language SKS ;
} liga;

The liga table of the subsetted font now want's to replace "oou" instead of "ffl".

The result, for example in InDesign, is devastating:

The subsetted font indeed tries to replace "oou" with a non-existing glyph and scrambles the whole text in the rendering. The font subset is unusable.

I know that it's hard to maintain OpenType features in a font subset, but I would at least expect that the OpenType tables are kept "as-is"...