Also roundtrip all characters in the Cf category.
Characters with the DI (Default Ignorable) property are now
preserved but not rendered and treated as zero-width
as per the unicode standard.
See https://www.unicode.org/faq/unsup_char.html
The new infinite ligature cache needed to be keyed by total number of
cells as well as total number of glyphs, otherwise a space based
ligature which is the same glyph rendered in different numbers of cells
wold render incorrectly when present in multiple forms. Fixes#3729
An infinite length ligature (longer than 9 glyphs) that has long
sequences of a repeated glyph now does not store a separate sprite
for every repeat.
For example in the ligature ---------- even though there are 10 hyphen
glyphs only four will be stored on the GPU.
Apparently on CoreText, harfbuzz gives incorrect values for glyph
positions. So we use it only for selection and grouping of glyphs.
Actual positioning is done using CoreText. This means sophisticated
positioning using GPOS tables is probably broken, but that isn't really
useable in a character grid anyway.
Also remove the hack where glyph_centering was done for calt ligatures
as it seems to not be needed with modern FiraCode and CoreText
rendering. Fixes#3372
Instead use the larger of the two alphas. This gives better results
for overlapping text, such as the infinite length ligature glyphs used
by FiraCode. I dont think adding ever gave better results anyway.
Now such ligatures are detected based on glyph naming convention.
This removes the gap in the ligatures at cell boundaries. However, at
least in Fira Code the infinite length ligature glyphs define a negative
left side bearing. This means they overlap when drawn consecutively,
leading to unsightly bumps at the joins.
Fixes#2695
For some reason it puts empty glyphs after the ligature glyph rather than before it.
There is a possibility this fix might break something else, we will see.
Fixes#3313
