Skip to content

FONTS_AND_TEXT.md

Latest commit

 

History

History
114 lines (93 loc) · 5.58 KB

FONTS_AND_TEXT.md

File metadata and controls

114 lines (93 loc) · 5.58 KB
Raw

Fonts and text

Version 0.4.0-alpha.1 ports the read-only font and text responsibilities needed before the graphics interpreter. All decoding executes in managed C#. The implementation does not use FreeType, Fontconfig, HarfBuzz or a native font library.

Character pipeline

Simple fonts consume one PDF character code at a time. Differences and a clear-text embedded Type 1 encoding map glyph names through the Adobe naming conventions, including uniXXXX, uXXXXX, underscore sequences and common ligature names. WinAnsi, MacRoman, Standard, Symbol and a useful Zapf subset have managed fallbacks. Explicit ToUnicode mappings always take precedence.

Type 0 fonts keep two mappings separate:

  1. the Encoding CMap consumes one to four bytes and maps the source code to a CID;
  2. the ToUnicode CMap maps the original source code to Unicode.

This is required for custom CMaps where the source value and CID differ. Codespace, cidchar, cidrange, bfchar and bfrange entries are supported. Identity-H and Identity-V are built in. CID ranges remain compressed to avoid allocating one entry per CID.

Release 0.8.0-beta.1 can also resolve named encoding and ToUnicode CMaps from explicit PdfReadOptions.CMapDirectories and conventional system poppler-data locations. Both dictionary /UseCMap and PostScript usecmap inheritance are bounded by byte and depth limits. Explicit directories take priority; UseSystemCMaps = false makes resolution fully controlled by the application.

When a PDF omits ToUnicode, an embedded TrueType or OpenType font can supply a fallback through sfnt cmap format 4 or 12. Format 0 byte-encoding tables are retained as direct source-character-code-to-glyph maps. CIDToGIDMap is applied before the reverse Unicode lookup. Adobe-Identity/Adobe-UCS fonts finally fall back to identity Unicode where a valid scalar exists.

Text runs retain their decoded source character codes and CIDs internally. Rasterization therefore selects subset glyphs without attempting the lossy round trip PDF code → Unicode → glyph ID; this also keeps a single ligature glyph intact when ToUnicode expands it to multiple Unicode scalars.

Metrics and placement

  • simple-font FirstChar, Widths and descriptor MissingWidth;
  • canonical Courier, Helvetica, Times, Symbol and ZapfDingbats metrics, including all Base-14 style variants, when a standard font omits Widths;
  • Type 3 FontMatrix width scaling;
  • CID default and exceptional widths through DW and both W forms;
  • vertical defaults and exceptions through DW2 and both W2 forms;
  • font descriptor ascent/descent for run bounds;
  • character spacing, word spacing, horizontal scale, rise and TJ adjustments;
  • horizontal and vertical text-matrix advancement.

TextBox.WritingMode and TextBox.IsRightToLeft expose directional information. TextLayout.Physical clusters baselines and respects dominant right-to-left direction. TextLayout.NonRawNonPhysical adds a conservative two-column reading-order heuristic. TextLayout.RawOrder preserves content stream order.

Font inspection API

Page.Fonts returns one FontInfo per page resource. It reports the resource and base names, PDF font type, encoding, writing mode, subset marker, ToUnicode presence and embedded Type 1, CFF, TrueType or OpenType container. The same data is available through:

poppler-net fonts input.pdf

Embedded byte counts refer to the decoded font stream when its filters are supported, otherwise to the retained encoded payload. Font programs remain owned by the document and are not exposed as mutable buffers.

Resource limits

PdfReadOptions.MaximumCMapMappings defaults to 250,000. It covers expanded Unicode entries and explicit CID entries. CID ranges themselves use a bounded range record. MaximumExternalCMapBytes defaults to 16 MiB per file and MaximumCMapUseDepth defaults to 16. Existing input, decoded stream, object and collection limits also apply to fonts and CMaps.

Deliberate limits

  • release 0.8.0-beta.1 rasterizes common embedded TrueType, CFF1/CFF2 Type 2 and Type 1 outlines, plus Type 3 CharProcs. CFF2 uses its default variation instance; complete region interpolation, rare charstring operators, hinting and Type 1 seac remain unsupported;
  • GSUB processing covers non-contextual vert/vrt2 single substitutions and exact liga/rlig ligatures. Contextual GSUB, GPOS, complex-script shaping and the full Unicode Bidirectional Algorithm remain unsupported;
  • managed file substitution is available, but it is simpler than Fontconfig/FreeType matching and depends on local files unless explicit FontDirectories are supplied. Narrow/Condensed and Expanded/Extended traits participate in scoring, several ranked candidates are tried for a glyph, and horizontal replacement outlines are fitted to the authoritative PDF/Base-14 advance to avoid collisions;
  • raw CFF charset/encoding fallback is partial when ToUnicode is absent;
  • encrypted/eexec Type 1 programs are decoded, but uncommon OtherSubrs and synthetic/flex behavior remain partial;
  • external CMaps support the common declarative mapping and inheritance syntax, but do not execute arbitrary PostScript procedures;
  • advanced Type 3 color/glyph behavior and text clipping through Type 3 outlines remain partial.

Raster text is now part of the graphics display list, including Form-nested text, exact operator interleaving and all eight fill/stroke/clip modes. Pattern and special-color text inherit the limits of the corresponding vector brush; contextual shaping remains future work; simple vertical alternates are applied through vert/vrt2.