feanor 0.3.0

Install with cargo install feanor-cli, or add the library with cargo add feanor.

Added

• Rules about what lookups do, checked on the feature code of every Google
  Fonts family and against HarfBuzz where shaping decides:
  • R001 unreachable-rule: a contextual rule an earlier rule in the same
    lookup always matches first, while doing something else
  • R002 duplicate-substitution: a glyph or ligature substituted twice with
    different results (fea-rs keeps the last, fontTools rejects the file)
  • R003 duplicate-pair: a glyph pair positioned twice with different values;
    the later one is ignored (safe fix: remove it)
  • R004 wrong-lookup-type: a positioning lookup called from a substitution,
    or the reverse
  • R005 lookup-has-no-effect: a lookup called on glyphs it has no rule for
  • R006 redundant-rule: a rule that repeats what an earlier one does (hint;
    safe fix: remove it)
  • C005 mark-class-conflict: mark classes sharing a glyph in one lookup
  • L003 empty-block: a feature or lookup with no rules
  • E002 stray-semicolon: a ; with no statement (safe fix: remove it)
  • E003 languagesystem-after-feature (unsafe fix: move it up)
• feanor dev compile: fea-rs's full errors and warnings for a file.

Changed

• Linting one file reports each include as a P001 hint to lint the whole
  project, instead of an E001 error.
• Glyph name suggestions skip names that differ in a code point (uFF10,
  u1F10), a one-letter part (f_f, f_j) or only digits (ss01, ss02),
  which name different glyphs rather than typos.
• format lays out long rules differently (below), so code formatted by 0.2
  can change once when formatted again: about 1 file in 6 across the Google
  Fonts sources. Compiled output is unchanged.
• format: a glyph class that starts on a new line after [ keeps the
  author's line breaks, so rows that group glyphs, or that line up with a
  parallel class, survive formatting. Only lines that are too long are
  wrapped. Classes written on one line are packed to the line width as before.
• format: when a long rule needs a class split over lines, by stays on the
  rule's line if it fits there.
• format: a contextual rule whose glyph sequence is too long for one line
  puts each class on its own line, instead of splitting short classes
  across lines; a run of single glyphs fills each line. A class is still
  split when it can't fit on a line of its own.

Fixed

• Indent detection no longer picks an unusual unit from uneven nesting (a
  file indented 3 spaces, then 8, was formatted with 5); a tab or up to four
  spaces wins whenever the file uses one.
• A bare include with no file in parentheses is reported as an error
  instead of crashing (also in 0.1.0).
• format: rows that line up without needing any padding are formatted as
  ordinary statements, not as a table.
• format reports # fmt: off regions in untouched, as documented.
• Incremental (and the WebAssembly Document) report what lint reports
  for a feature body too, instead of syntax errors. A text is tried as a
  feature body only when it has syntax errors and no feature block that
  parses, since a body can't contain one.