Segment strings by grapheme cluster instead of by character when shaping and rendering text. Store glyphs, glyph requests, and glyph positions by grapheme cluster instead of by codepoint string. Added a simple polyfill for older versions of Firefox.

Increased the buffer around locally rendered glyphs.

Removed hard-coded fudge factors based on the baseline in Arial Unicode MS.

If a grapheme cluster begins with a combining diacritical mark or ends with an invisible stacker, combine it with an adjacent grapheme cluster to avoid drawing diacritics over dotted circles or placeholder diacritics where adjacent characters should be ligated instead.

Added a script that fetches the latest Unicode character database’s property file for Indic syllable categories and generates a function for combining graphemes based on it.

Replace zero-width joiners with temporary strip markers to prevent ICU from stripping them.

Preemptively swap combining marks with the characters they modify to visual order, so that the RTL plugin will swap them back to logical order.

Replaced custom word break heuristics when determining line breaks with a word segmenter. Added a simple polyfill for older versions of Firefox.

Fixed an issue where vertical CJK text was shifted upwards by one em.

Iterate over graphemes instead of words, looking for word boundaries to use as line breaking opportunities. This eliminates the possibility of word-wrapping in the middle of a grapheme cluster, which is valid in some writing systems such as Thai, but mitigates the risk of an invalid section index in Burmese, because the word segmenter considers some modifiers to be “words”.