- Codepoints get mapped, one-to-one, to glyphs. This turns your string into a sequence of glyph IDs. Each of these glyph IDs is associated with an “advance” which is a 2D vector from the origin of each glyph to the next.
- Font features run arbitrary modifications on the sequence of glyphs and advances
- Each glyph is rendered at its respective location.
OpenType’s model is simpler, so let’s start there. OpenType encapsulates font features into two tables, GPOS and GSUB. GPOS is for glyph positioning, and GSUB is for glyph substitution. Both of these tables actually share the same overall structure. The shared structure is as follows:
- The table is represented by a collection of “scripts.”
- A “script” is represented by a collection of “language systems.”
- A “language system” is represented by a collection of “features.”
- A “feature” is represented by a collection of “lookups.”
- A “lookup” is represented by a collection of typed tables.
Each of the typed tables has an associated “coverage” table. This simply holds a set of glyphs that the typed table acts upon. When the software is laying out some glyphs, if none of the glyphs appear in the coverage table, the software can safely ignore that typed table.
Everything else is shared between GPOS and GSUB, and is just an aggregation of the level below it. Each “feature” has a four-letter code associated with it, such as “liga” or “smcp.” These codes are reported in the program UI, and the user is allowed to switch on or off each of these codes independently. If the user switched off a particular code, the stuff inside that feature’s “lookup” tables is not performed. The lookup tables are performed one by one, in the order the file describes.
Scripts and language systems each have their own four-character tags associated with them. The reason for the distinction is that many different languages may share the same script. For example, both English and Spanish share the Latin script. Because the languages and scripts are expected to have very different glyph patterns, separating out the features into these buckets seems like a good idea. Also, there is a default script tag and a default language system. Microsoft keeps a registry of all of the scripts and language systems, as well as their associated 4-character tags.
TrueType font features are completely different. In particular, there are many tables, each with a specialized purpose. For example, instead of the general “GPOS” table, there is a much more specific “kern” (or “kerx”) table, which is only designed for kerning glyphs. Similarly, there is a separate, more-specific, “opbd” table for the optical bounds of the font. Many of the OpenType features have corresponding specific-purpose tables inside TrueType. These tables do not provide the same flexibility as the GPOS table.
However, unlike most of the TrueType tables, the “morx” table is quite general. Conceptually, it describes the same sort of things that the OpenType GSUB table describes. However, its model of font features is very different.
The “morx” table doesn’t mess around with any of that “script” and “language system” stuff. Instead, the table is represented by a collection of “chains.” Each chain is comprised of a collection of typed subtables. These typed subtables describe things like “Rearrange this sequence of glyphs according to these predefined rules” or “insert a specific glyph into the middle of this particular glyph sequence.”
The rest of the chain is devoted to describing which typed subtables should be performed.
Similar to before, each chain gets a coverage table. However, in addition, there is a pretty interesting mechanism to determine which features cause which typed subtable to be performed.
Inside the chain, a sequence of bitwise operations is described, which results in a particular bitfield. Each typed subtable in the chain also has an associated bitfield. If the bitwise and of those two values is nonzero, then the typed subtable is applied.
That sequence of bitwise operations is where individual features come into play. Instead of representing a particular feature as a four-character tag, like in OpenType, TrueType represents a feature as a tuple of (feature type, feature selector). The font describes a sequence of features and corresponding bit masks to be ANDed / ORed. When a feature is enabled in the application’s UI, its associated bitwise operations are performed. When the feature is disabled, its associated bitwise operations are not performed.
The software starts with a known bit pattern. It then runs through the sequence. For each item in the sequence, if its associated font feature is enabled, it performs the appropriate bitwise operations. When it’s done, the result is a particular bit pattern value.
Each of the typed subtables also has an associated pattern value. If the bitwise AND of the calculated bit pattern and a subtable’s bit pattern is nonzero, then the table is performed.
Exclusive and Default Features
In OpenType, the font file has no way of expressing that certain features conflict or that certain features should be on by default. Each feature is simply represented by a four-character tag, and it is either enabled or disabled. The software may (or may not) decide to enable certain features by default based entirely on the semantics of the feature itself (not based on the font file). For example, the OpenType feature “rlig” is mean for “required ligatures” and is generally enabled by default.
The representation of a font feature in TrueType conveys more information than an OpenType font feature. In TrueType, a feature is represented by a tuple of a “type” and a “selector.” For a particular “type,” there will be a variety of different selectors which are relevant. Many selectors are only relevant to a particular type. An example would be “type”: “kLigaturesType” and “selector”: “kHistoricalLigaturesOnSelector”. Another example would be “type”: “kCharacterShapeType” and “selector”: “kTraditionalCharactersSelector”.
There is a fairly interesting difference between these two examples, though. The first type, kLigaturesType, and “on” and “off” selectors. The second type, kCharacterShapeType, doesn’t. This means that kCharacterShape is what is known as an “exclusive” type, where setting a value for it overrides all other values for it. On the other hand, kLigaturesType is a nonexclusive type, so you can set both kRareLigaturesOnSelector and kCommonLigaturesOnSelector for it. For all the feature types your font knows about, you can declare whether they are exclusive or nonexclusive in the font’s ‘feat’ table, and the software will behave accordingly.
In addition, you can mark a feature as on by default if you set the chain’s default flags to intersect with the feature’s enable flags. Logically, this makes sense - if the feature turns on something which is already on, then that feature should be on by default. However, because the feature is on by default, the software might decide not to show it in the list of available font features. This makes sense for something like required ligatures; turning that feature off would make the font completely illegible, so the software shouldn’t even give the user that option. This behavior allows the font file to determine which features get turned on by default, rather than the software making a guess.
This also means that, for all your exclusive features, you need to designate one selector as the default, even if that selector doesn't actually trigger any typed subtables to be performed. Again, this logically makes sense, and allows you to perform some shaping by default.