.OTF File Format – CFF table

Rules in OpenType

• The font has sfntVersion OTTO and no glyf or loca tables; see the table directory.
• The Name INDEX must hold exactly one entry: one font per table. The name need not match name ID 6 in the name table; applications use the name table. In a collection one CFF  table can be shared by several fonts.
• The Top DICT must give CharstringType 2 (the default).
• numGlyphs in maxp (version 0.5 for CFF fonts) must equal the number of charstrings, and OpenType glyph numbers are CFF glyph numbers.
• Advance widths appear twice: in the charstrings and in hmtx.

Data types and layout

All numbers are big-endian and nothing is aligned, so there is no padding anywhere. Offsets are written relative to a reference point: “(0)” means from the start of the CFF data, “(self)” from the start of the structure holding the offset.

The parts, in their usual order. The first five are always at the start, in this order; the rest are reached by offsets and could be anywhere.

Header

A reader that does not know the major version must reject the data; an unknown minor version means extensions it can safely ignore.

INDEX

An INDEX is an array of variable-length objects.

An empty INDEX is just a count of 0: two bytes, with no offSize or offsets. An INDEX can be skipped by jumping to its last offset.

DICT encoding

A DICT is a sequence of key–value pairs written as operands followed by an operator, like PostScript. The first byte tells them apart: 0–21 start an operator; 28, 29, 30 and 32–254 start an operand; 22–27, 31 and 255 are reserved. Operator 12 is an escape: the next byte completes a two-byte operator. An operator may have up to 48 operands before it.

Name INDEX

An INDEX of the PostScript names (FontName or CIDFontName) of the fonts in the set, sorted by byte value, a name sorting before any longer name it is a prefix of. There must be at least one; in OpenType, exactly one. For compatibility, names should be at most 127 characters (63 for Adobe Type Manager) and avoid [ ] ( ) { } < > / %, NUL, space, tab, CR, LF and form feed; the recommended set is codes 33 to 126. A font can be deleted from a set without rebuilding it by setting the first byte of its name to 0.

Top DICT

The Top DICT INDEX holds one DICT per font, in the same order as the Name INDEX. Most keys carry the names of Type 1 font dictionary keys; those marked FontInfo come from Type 1’s FontInfo dictionary.

CIDFonts add these operators:

A synthetic font’s Top DICT starts with SyntheticBase and a CIDFont’s with ROS; any other first operator means an ordinary name-keyed font, so the kind can be told without parsing the whole DICT. All one-byte operators not listed here or under the Private DICT (22–27, 31) and the two-byte operators 12 15, 12 16, 12 24 to 12 29 and 12 39 to 12 255 are reserved.

Strings

Every string except the font names is stored once, in the String INDEX, and referred to by a SID. SIDs 0 to 390 are the 391 standard strings, which a reader keeps in its own table; SID 391 is the first entry of the String INDEX, SID 392 the second, and so on (entry = SID − 391). A font with no strings of its own has an empty String INDEX.

Glyphs, encodings and charsets

Glyphs are numbered from 0 (GID) in CharStrings order. The charset gives each glyph a name (a SID; in CIDFonts a CID), and the encoding gives some glyphs a one-byte character code: think of three parallel arrays indexed by GID. GID 0 is always .notdef, unencoded, so the charset and encoding arrays start at GID 1. In an OpenType font, characters are mapped by cmap and the CFF encoding matters only to PostScript use.

CharStrings INDEX

One charstring per glyph, by GID; the first must be .notdef, and the count is the number of glyphs. CharstringType 2 means Type 2 charstrings; type 1 charstrings are described in Adobe’s Type 1 Font Format book.

Private DICT

The Private DICT holds hinting values and the widths and subroutines the charstrings use. It is required, but may have length 0 if every value is the default.

A glyph whose width equals defaultWidthX leaves it out of its charstring; any other width is stored as the difference from nominalWidthX, chosen so most differences fit in one byte. Earlier versions also had ForceBoldThreshold and lenIV; both were removed in 2000.

Subroutines

Charstrings can call shared pieces of charstring code. Local subroutines belong to one font (or, in a CIDFont, one Private DICT) and are found through the Subrs operator. Global subroutines, in the Global Subr INDEX straight after the String INDEX, can be used by every font in the set; a set without any has an empty INDEX. A charstring calls subroutine n with n − bias, so that small negative numbers are used as well as positive ones. The bias depends on the number of subroutines in that INDEX:

if (CharstringType == 1)   bias = 0;
else if (nSubrs < 1240)    bias = 107;
else if (nSubrs < 33900)   bias = 1131;
else                       bias = 32768;

A font maker that wants most calls to take one byte puts the most used subroutines where the bias makes their numbers small. With bias 107, index 0 to 214 is called as −107 to +107 (one byte) and 215 to 1238 as +108 to +1131 (two bytes). With bias 1131, index 1024 to 1238 takes one byte, 0 to 1023 and 1239 to 2262 two, and 2263 to 33898 three. With bias 32768, index 32661 to 32875 takes one byte, 31637 to 32660 and 32876 to 33899 two, and the rest three.

Synthetic fonts

A synthetic font is another font in the set with a different matrix or encoding, such as an oblique or condensed version. Its Top DICT must begin with SyntheticBase, the index of the base font, and may also contain only FullName, ItalicAngle, FontMatrix and Encoding, which override the base font’s. The base may not be a CIDFont or another synthetic font. Since OpenType allows one font per table, synthetic fonts do not occur in OpenType.

CID-keyed fonts

A CIDFont, usually for Chinese, Japanese or Korean, names glyphs by number (CID) instead of by string, and gets its character mapping from a CMap outside the font (in OpenType, from cmap). Its Top DICT starts with ROS, and:

• FDArray points to a Font DICT INDEX. Each Font DICT applies to a group of glyphs and has its own Private operator, and so its own Private DICT and local subroutines.
• The charset maps GIDs to CIDs instead of SIDs. A complete CIDFont has an identity charset (GID = CID), one range starting at CID 1; subsets usually need more ranges.
• FDSelect says which Font DICT each glyph uses.
• There is no Encoding operator.

Embedded PostScript and the FontSet wrapper

A Top DICT may have at most one PostScript operator, whose string a PostScript interpreter runs after building the font dictionary and before definefont, with that dictionary on the dictionary stack. In a CIDFont it may also be in a Font DICT, where it runs on that sub-dictionary. Software that does not run PostScript ignores it. When an OpenType font is converted to bare CFF for embedding in a document, Adobe recommends recording the embedding permissions this way, for example /FSType 8 def /OrigFontType /OpenType def.

Outside OpenType, a CFF FontSet sent to a printer is wrapped as a PostScript resource: %!PS-Adobe-3.0 Resource-FontSet, DSC comments, /FontSetInit /ProcSet findresource begin, then /name n StartData, one space and n bytes of CFF data, whose (0) offsets count from the byte after that space.

Example

The start of the CFF  table of Noto Sans Javanese Regular (NotoSansJavanese-Regular.otf, shipped with macOS), at file offset 2,176:

00000880  01 00 04 02 00 01 01 01  19 4E 6F 74 6F 53 61 6E   .........NotoSan
00000890  73 4A 61 76 61 6E 65 73  65 2D 52 65 67 75 6C 61   sJavanese-Regula
000008A0  72 00 01 01 01 33 F9 AE  00 F9 AF 01 F9 B0 0C 00   r....3..........
000008B0  F9 B1 02 F9 B2 03 1D FF  FF F9 E1 FE 28 1D 00 00   ............(...
000008C0  08 A1 FA A8 05 1D 00 00  25 75 0F B7 1D 00 00 9E   ........%u......
000008D0  CF 12 1D 00 00 25 82 11  01 98 02 00 01 00 05 00   .....%..........

There is no Encoding (so the default 0) and no ROS, so this is a name-keyed font. The charset at 9,589 is 02 01 87 00 01 00 01 00 00 01 89 01 90: format 2, SIDs 391–392 (NULL and CR, the first two strings), SID 1 (space), then SIDs 393–793, naming all 404 glyphs after .notdef. The CharStrings INDEX starts 01 95 02: 405 glyphs, which matches numGlyphs in maxp. The Private DICT is:

7C 9A F8 AC 9A F7 5B 9A F7 AB 9A 06   BlueValues −15 15 536 15 199 15 279 15
FC BB 9A 07                           OtherBlues −551 15
1E A0 5F 0C 09                        BlueScale 0.05
8B 0C 0B                              BlueFuzz 0
CD 0A                                 StdHW 66
DA 0B                                 StdVW 79
CD 95 0C 0C                           StemSnapH 66 10
DA 92 0C 0D                           StemSnapV 79 7
1D 00 00 05 1D 15                     nominalWidthX 1309
B7 13                                 Subrs 44

The deltas decode to blue zones −15 to 0, 536 to 551, 750 to 765 and 1044 to 1059, another zone −551 to −536, and stem widths 66 and 76, 79 and 86. Subrs 44 means the local Subr INDEX starts right after these 44 bytes. The charstrings page decodes this font’s .notdef.

Standard strings

SIDs 0 to 390, ten to a row.

Predefined encodings and charsets