Fix #189 — OpenTypeFont reads TrueType outlines and font collections

[sbeitzel]

Fix #189

What changed

• TrueTypeGlyphs (new): reads glyf through loca, in both the short and long formats.
  • Simple glyphs: run-length flags and short/long coordinate deltas. A run of off-curve points implies on-curve midpoints between them, and contours may start off-curve or have no on-curve point at all.
  • Contours start where fontTools and FreeType start them: at the first on-curve point, or else at the midpoint of the last and first points. A contour that returns to its start along a straight line ends with Z alone.
  • Composites:
    • ARG_1_AND_2_ARE_WORDS and byte offsets;
    • WE_HAVE_A_SCALE, X_AND_Y_SCALE and TWO_BY_TWO;
    • SCALED_COMPONENT_OFFSET (unless UNSCALED_COMPONENT_OFFSET is also set);
    • nesting up to 8 levels, after which compositeTooDeep is thrown, so a self-including composite fails instead of recursing forever.
  • Point-matched components (ARGS_ARE_XY_VALUES clear) are skipped. The issue allows this for now; there is no diagnostic channel this deep in font parsing.
  • Hinting instructions are ignored.
• GlyphPath:
  • Adds .quadCurve(control:to:), written as SVG Q. Quadratics are kept as they are rather than converted to cubics, which keeps the path data smaller.
  • Adds an affine transformed(…) for placing composite components.
• Collections:
  • OpenTypeFont.parse(_:faceIndex:), parse(_:postScriptName:) and postScriptNames(in:), the last reading name ID 6.
  • The new property postScriptName.
  • A new error, faceNotFound.
• Embedding: OpenTypeFont.embedding: EmbeddingPermissions, read from OS/2.fsType. allowsOutlineEmbedding refuses Restricted License, bitmap-only and no-subsetting faces; an outline document embeds only the glyphs it draws, which is a subset. Font provider: use host-registered and system fonts instead of bundling more faces #190 will use this to fall back to a bundled face.
• Bundled faces: they are CFF and go through the same path as before; the snapshot tests are unchanged.

Fixtures

Tests/CeolKitSVGRendererTests/Resources/truetype-fixtures.py generates three fonts with CeolKit's own outlines, so there is no third-party licence. truetype-fixtures-PROVENANCE.txt describes them.

• CeolKitTest-Short.ttf and CeolKitTest-Long.ttf: the same glyphs with each loca format. The long format is forced by giving one glyph an odd-length hinting program.
• CeolKitTest.ttc: two faces with different PostScript names and fsType values (0 and Restricted).

The fixtures cover on-curve-only contours, implied midpoints, an off-curve start, an all-off-curve contour, composites with byte and word offsets, all three transforms, a nested composite and a point-matched component. The oracle is fontTools' own reading of the same bytes, recorded in truetype-reference.json.

Tests (TrueTypeOutlineTests, 10)

• Every glyph matches the reference in both loca formats, along with its advance width.
• The Q path data for an all-off-curve contour.
• A point-matched component is skipped.
• A composite cycle, made by patching the fixture's bytes, throws compositeTooDeep.
• Collection faces selected by index and by name, and both faces outline correctly.
• fsType decoding.
• The bundled faces still parse and allow outline embedding.
• macOS only: Courier New and Times New Roman from /System/Library/Fonts/Supplemental are compared with CoreText for glyph IDs and control-point bounding boxes across AgQé&ß. This test is skipped where the files are missing. It catches misplaced or missing contours and composite parts, but not errors in implied midpoints; the reference tests cover those.

With the midpoint arithmetic deliberately broken, the reference tests fail (17 issues). swift test: all tests pass.

Out of scope (per the issue)

• Shaping, kerning, GPOS/GSUB.
• Variable fonts (read at their default instance).
• WOFF/WOFF2.
• Bitmap and colour glyphs.
• The glyf cubic-flag extension.

🤖 Generated with Claude Code