Font Development Best Practices

Technical guidance regarding font development and production

View project onGitHub

6.3 OpenType

Links to the OpenType specification, as well as other font specifications, can be found on ScriptSource.

Lookup Orders

The OpenType specification suggests that software should:

assemble all lookups from the set of chosen features, and apply the lookups in the order given in the LookupList table.

However in actual implementations this isn’t how it is usually done. Instead, features are divided into sets, and the sets are processed in a specific order. Within a set, all the lookups associated with the included features are executed (in lookup order) before moving on to the next set of features.

It is important, therefore, that font developers know how features are divided into sets, and the order in which these sets are executed. To complicate matters, this is:

  • not always consistent among shaping engines
  • script-specific, and
  • not well documented.

The most useful information is gleaned by reading the Harfbuzz code1 since (a) the code is opensource and thus accessible, and (b) Harfbuzz tries to be compatible with Microsoft’s Uniscribe.

The following tables attempt to document what the Harfbuzz code indicates for GSUB execution. Each row represents a set of features, each feature identified by its 4-character tag.

  Default shaper
GSUB rvrn
  ltra ltrm frac numr dnom rand ccmp locl rlig calt clig liga rclt userfeatures
GPOS mark mkmk curs kern userfeatures
GSUB rvrn
syllable analysis ltra ltrm rtla rtlm frac numr dnom rand trak HARF
  locl ccmp nukt akhn
  rkrf abvr blwf half pstf vatu cjct
insert dotted circles, reorder isol init med fina
  abvs blws haln pres psts BUZZ ccmp locl rlig calt clig liga rclt vert userfeatures
GPOS dist abvm blwm mark mkmk curs kern userfeatures
  Arabic Syriac
GSUB rvrn rvrn
  rtla rtlm frac numr dnom rand rtla rtlm frac numr dnom rand
  ccmp locl ccmp locl
  isol isol
  fina fina
  medi medi
  init init
  rlig rlig
  rclt calt rclt calt
  mset rlig clig liga userfeatures mset rlig clig liga userfeatures
GPOS mark mkmk curs kern mark mkmk curs kern
  Indic Khmer
GSUB rvrn rvrn
  ltra ltrm frac numr dnom rand ltra ltrm frac numr dnom rand
  ccmp locl ccmp locl
  pref pref blwf abvf
  init pres abvs blws psts haln rlig calt clig rclt pres abvs blws psts rlig calt clig rclt userfeatures
GPOS curs kern dist abvm blwm curs dist abvm blwm

where userfeatures includes everything else that the user (or CSS) might have requested, such as dlig, ssxx, cvxx, smcp, onum, etc.

Language Tags

(This section has not yet been written.)

Feature Tags

The OpenType specification provides for Layout Feature tables, each of which is identified by a 4-character tag.

(This rest of this section has not yet been written.)

Stylistic Sets and Character Variants

Tags for Layout Feature tables which are in the range “ss01” to “ss20” are registered as Stylistic Set 1 through 20. Tags in the range “cv01” to “cv99” are registered as Character Variant 1 through 99.

When should one use Stylistic Sets and when should one use Character Variants?

According to the OpenType spec, Character Variants (cvxx tags) should be preferred when only one character or very closely related characters are affected. Examples include:

  • matching lower and upper case characters
  • a lower case character and its small capital form
  • all characters that use a particular diacritic, where the diacritic has two forms

Stylistic Sets (ssxx tags) are preferred when systematically related changes affect more distinct characters. Examples include:

  • ‘barred b’, ‘barred d’, and ‘barred g’ characters with different placements for the bar
  • ‘a’ and ‘g’ characters with literacy forms

Having said that, there are some technical and practical distinctions that may require violating these guidelines:

  • The cvxx tags can utilize GSUB “alternate” (type 3) lookups, thus allowing a feature to have more than on/off values.
  • There are only 20 registered ssxx features, but 99 cvxx features.
  • Some apps (MS Word for example) assume that users would need to turn on only one ssxx feature at a time.
  • Not as many apps support cvxx (yet).

GSUB only

Note that only substitution type lookups are allowed in Stylistic Set and Character Variant features and the features must be in the GSUB table. There may be cases where variant positioning behavior is needed, such as a feature that controls whether a particular diacritic is drawn touching its base or separated from it. While it may seem reasonable to use positioning type lookups and place the features in the GPOS table, it is unlikely that rendering engines will actually process such features.

Mark classes and filter sets

(This section has not yet been written.)

  1. The Harfbuzz source is maintained on Github, and the pertinent information gleaned from the collect_features methods of the various shaper modules (,,, etc.)