Non standard extensions used FontForge in True/Open Type

Non standard feature tags

Internal

For TeX
I no longer use these. MS has come up with a MATH table which can contain all the TeX information.

Non standard tables.

Non standard file formats

'PfEd' -- the FontForge extensions table

(the name is 'PfEd' for historical reasons)

The table begins with a table header containing a version number and a count of sub-tables
uint32 version currently 0x00010000
uint32 count

This is followed by a table of contents, there will be count replications of the following structure (ie. a tag and offset for each sub-table
uint32 tag
uint32 offset from start of 'PfEd' table

The format of the subtable depends on the sub-table's tag. There are currently 3 tags supported, these are

'colr' -- the per-glyph color sub-table

The sub-table header begins with a version number, and a count of ranges
uint16 version 0
uint16 count of ranges

After this will be <count> instances of the following structure
uint16 starting glyph index
uint16 ending glyph index
uint32 color expressed as a 24bit rgb value

'cmnt' -- the per-glyph comment sub-table

The sub-table header begins with a version number, and a count of ranges
uint16 version 0/1
uint16 count of ranges

After this will be <count> instances of the following structure
uint16 starting glyph index
uint16 ending glyph index
uint32 offset from the start of this sub-table

The offset points to an array of offsets (<end>-<start>+1+1) elements in the array, so one element for each glyph index mentioned in the range structure above, with one left over which allows readers to compute the length of the last string.
uint32 offset from start of table
...

And each of these offsets points to a unicode (UCS2 for version 0, UTF-8 for version 1) string. The strings are assumed to be consecutive, so the length of each may be calculated by subtracting its offset from the offset to the next string.

'fcmt' -- the font comment (and FONTLOG) sub-table

The sub-table header begins with a version number, and a count of ranges
uint16 version 0/1
uint16 length number of characters in the string

In version 0 this is followed by <length> Unicode (UCS2) characters. In version 1 it is followed by <length> bytes in utf8 encoding.

'cvtc' -- the cvt comments sub-table

The sub-table header begins with a version number, and a count of cvt entries which might have comments
uint16 version 0
uint16 count number of entries in the cvt comments array
which might be smaller than the number of entries in the cvt array itself if we want to save space
uint16 offset[count] offsets to the start of utf8-encoded, NUL terminated strings. Or 0 if this cvt entry has no comment

'GPOS/GSUB' -- lookup names

The sub-table header begins with a version number, and a count of ranges
uint16 version 0
uint16 count of lookups in this table

Then there will be an array of count elements (one for each lookup, ordered as the lookups are in the GPOS or GSUB table)
uint16 offset to lookup name
uint16 offset to lookup subtable structure

These offsets are based on the start of the subtable. The name offset points to a NUL terminated UTF-8 encoded string. The subtable offset points to the following structure:
uint16 count of lookup subtables in this lookup

Then there will be an array of count elements (one for each subtable, ordered as the subtables are in the lookup of the GPOS or GSUB table)
uint16 offset to lookup subtable name
uint16 offset to anchor class structure

These offsets are also based on the start of the subtable. The name offset points to a NUL terminated UTF-8 encoded string. The anchor offset may be 0 if this subtable doesn't have any anchor classes, otherwise it points to the following structure:
uint16 count of anchor classes in this lookupsubtable

Then there will be an array of count elements (one for each anchor class)
uint16 offset to anchor class name

'guid' -- guidelines

The sub-table header begins with a version number, and a count of ranges
uint16 version 1
uint16 vcount number of vertical guidelines
uint16 hcount number of horizontal guidelines
uint16 mbz At some point this may contain info on diagonal guidelines. For now it is undefined
uint16 offset To a full description of the guideline layer

I provide the guidelines in two formats. Either may be omitted. The first format simply describes the horizontal and vertical lines used as guidelines. The second format provides a full description of all contours (curved, straight, horizontal or diagonal) which fontforge uses. I provide both since most apps seem to have a simpler guideline layer than ff does.

This table is followed by two arrays, one for vertical guidelines, one for horizontal guides. Both arrays have the same element type (except that the position is for a different coordinate in the horizontal/vertical cases)
int16 position x location of vertical guides, y location of horizontal ones
uint16 offset to name, a NUL terminated utf8 string. (offset may be 0 if guideline is unnamed).

The offset to the guideline layer points to a variable length structure which is also used in the 'layr' subtable

'glyph-layer' -- Data structure representing all the contours of a layer of a glyph

The sub-table header begins with a version number, and a count of ranges
uint16 count of contours
uint16 count of references (not present in version 0 layers)
uint16 mbz reserved for a count of images

This is followed by an array of structures describing each contour
uint16 offset to start of contour data
uint16 offset to a name for the contour, a utf8, NUL terminated string (or 0 if the contour is unnamed)

All offsets from the start of the glyph-layer structure.

This is followed by an array of structures describing each reference
fixed16.16 transform[6] A PostScript transformaton matrix where each member is a signed 4 byte integers which should be divided by 32768.0 to allow for non-integral values
uint16 gid The Glyph ID of the glyph being referred to

Contour data live in a variable length block. It's basic idea is that it is a list of <command>, <data> pairs. Each command is a byte which consists of two parts, a verb which specifies what happens (and how many items of data are needed) and a modifier which specifies how each data item is represented. The verbs are postscript-like drawing operations: moveto, lineto, curveto, (and quadratic curveto), close path, etc. There are also separate verbs for specifying spiro control points -- these are just the standard spiro type bytes ('v', 'o', 'c', '[' and ']'), no modifier is applied to the spiro commands, their data are always 2 coordinates in fixed notation.

The low order two bits of the command (except for the spiro and close commands) specify the data format:
0 signed byte data for values -128 to 127
1 signed short data for values -32768 to 32767
2 A signed 4 byte integer which should be divided by 256.0 for non-integral coordinates (or for big ones)
3 Undefined and erroneous for now

Each command will start at the current point and draw to the point specified by its data. The data are relative to the last point specified (except for moveto, which is absolute, there being on previous point).

The verb may be one of the following:
0 MoveTo, takes 2 coordinates (x,y). This must begin each contour and may not appear elsewhere within it
4 LineTo, also takes 2 coordinates
8 HLineTo, draws a horizontal line, so only the new x coordinate need be specified.
12 VLineTo, draws a vertical line, so only the new y coordinate need be specified.
16 QCurveTo, takes one off-curve control point and one on-curve point, 4 coordinates total, to draw a quadratic bezier spline
20 QImplicit, only specifies the control point. The on-curve point will be the average of the control point specified here, and the one specified in the next QCurveTo or Q*Implicit command.
24 QHImplicit, Same as above, except only the x coordinate of the new control point is specified
28 QVImplicit, Same as above except only the y coordinate of the new control point is specified.
32 CurveTo, takes two off-curve control point and one on-curve point, 6 coordinates total, to draw a cubic bezier spline
36 VHCurveTo, The first control point is vertical from the current point, so only its y coordinate is specified. The final point is horizontal from the last control point so only its x coordinate is specified. 4 coordinates total y1, x2,y2, x3.
40 HVCurveTo, Reverse of the above x1, x2,y2, y3
44 Close, No data. Closes (and ends) the current contour. Will draw a line from the start point to the end point if needed.
45 End, No data. Ends the current contour, but leaves it open.

These are basically the drawing operators in the type1 charstrings. If my terse descriptions make no sense look there for a more complete description.

examples

suppose we want to draw a box (0,0)->(0,200)->(200,200)->(200,0)->(0,0). Then the glyph-layer would look like:
Header one contour (ushort) 1
Header no images (ushort) 0
First Contour offset to data (ushort) 8 The number of bytes from the start of the glyph-layer to the Contour Data
First Contour no name (ushort) 0
Contour Data Move To 0,0 (byte)0, (byte)0, (byte)0 Both coordinates are <127 and can fit in a byte, so the modifier is 0. The command is also 0, and the coordinates are each 0
Contour Data VLine To [0,]200 (byte)(12+1) (short)200 Vertical motion => VLineTo. Only the new y value need be specified. Is relative to the last position, but that was 0
Contour Data HLine To 200[,200] (byte)(8+1) (short)200 Horizontal motion => HLineTo. Only the new x value need be specified. Is relative to the last position, but that was 0
Contour Data VLine To [200,]0 (byte)(12+1) (short)-200 Vertical motion => HLineTo. Only the new y value need be specified. We move from 200 to 0, so the relative change is -200
Contour Data Close (byte)44 We can draw the final line by closing the path

'layr' -- Background layer data

uint16 version 1
uint16 count number of layers in this sub-table

This is followed by an array of structures describing each layer
uint16 typeflag Low order byte is the type
2=>quadratic splines, 3=>cubic splines, 1=>spiros, other values not defined
High order byte are the flags
0x100 => foreground layer
uint16 offset to the name of this layer, a utf8, NUL terminated string
uint32 offset to the data for this layer

The layer data is a block of ranges specifying which glyphs (by GID) have data for this layer. (the type field is present so that applications can ignore layers which they do not support).
uint16 count of ranges

This is followed by an array of structures one for each range:
uint16 start first GID in the range
uint16 last last GID in the range
uint32 offset to an array of offsets, one for each GID in the range. The offsets in this array may be 0. These offsets in turn point to a glyph-layer structure

All offsets are relative to the start of the 'layr' subtable.

'TeX ' -- the TeX metrics table

The table begins with a table header containing a version number and a count of sub-tables
uint32 version currently 0x00010000
uint32 count

This is followed by a table of contents, there will be count replications of the following structure (ie. a tag and offset for each sub-table
uint32 tag
uint32 offset from start of 'PfEd' table

The format of the subtable depends on the sub-table's tag. There are currently 3 tags supported, these are

'htdp' -- the per-glyph height/depth sub-table

The sub-table header begins with a version number, and a count of glyphs
uint16 version 0
uint16 count of glyphs

After this will be <count> instances of the following structure
uint16 height in em-units
uint16 depth

I store these values in em-units rather than in the fix_word of a tfm file because em-units make more sense in a sfnt and take up less space.

'sbsp' -- the per-glyph sub/super-script sub-table

This sub-table has essentially the same format as the previous one. The sub-table header begins with a version number, and a count of glyphs
uint16 version 0
uint16 count of glyphs

After this will be <count> instances of the following structure
uint16 subscript offset in em-units
uint16 superscript offset

I store these values in em-units rather than in the fix_word of a tfm file because em-units make more sense in a sfnt and take up less space.

'ftpm' -- the font parameter (font dimensions) sub-table

The sub-table header begins with a version number, and a count of parameters
uint16 version 0
uint16 count number of parameters in the font

And this is followed by <count> instances of the following structure:
uint32 tag parameter name
int32 value

I store these values as fix_words and not as em-units because their meaning is not constrained by the spec and the Slant parameter (at the least) can not be converted to em-units.

I have defined the following 4-letter parameter tags
Tag Meaning traditional font parameter number in tfm file (font dimension number)
Slnt Slant 1
Spac Space 2
Stre Stretch 3
Shnk Shrink 4
XHgt XHeight 5
Quad Quad 6
ExSp Extra Space 7 (in text fonts)
MtSp Math Space 7 (in math and math extension fonts)
Num1 Num1 8 (in math fonts)
Num2 Num2 9
Num3 Num2 10
Dnm1 Denom1 11
Dnm2 Denom2 12
Sup1 Sup1 13
Sup2 Sup2 14
Sup3 Sup3 15
Sub1 Sub1 16
Sub2 Sub2 17
SpDp Sup Drop 18
SbDp Sub Drop 19
Dlm1 Delim 1 20
Dlm2 Delim 2 21
AxHt Axis height 22
RlTk Default Rule Thickness 8 (in math extension fonts)
BOS1 Big Op Spacing1 9
BOS2 Big Op Spacing2 10
BOS3 Big Op Spacing3 11
BOS4 Big Op Spacing4 12
BOS5 Big Op Spacing5 13

'BDF ' -- the BDF properties table

The table begins with a table header containing a version number and a count of strikes
uint16 version currently 0x0001
uint16 strike-count
uint32 offset to string table (from start of BDF table)

This is followed by an entry for each strike identifying how many properties that strike has.
uint16 ppem
uint16 property-count

Then there will be the properties, first there with be property-count[1] properties from the first strike, then property-count[2] properties for the second, etc. Each property looks like:
uint32 name offset into the string table of the property's name
uint16 type 0=>string
1=>atom
2=>int
3=>uint
0x10 may be ored into one of the above types to indicate a real property
uint32 value For strings and atoms, this is an offset into the string table
for integers it is the value itself

The string table is a series of ASCII bytes. Each string is NUL terminated.

'FFTM' the FontForge time stamp table

The table begins with a table header containing a version number and is followed by a series of timestamps (same format as the timestamps in the head table -- 64 bit times, seconds since 00:00:00, 1-Jan-1904).

I don't think this is a duplication of the information in the 'head' table. Neither the Apple nor OpenType spec is clear: Does head.creationtime refer to the creation time of the truetype/opentype file, or of the font itself. After examining various fonts of Apple's, it appears that the 'head' entries contain the dates for the font file and not the font. The times in this table are specifically the creation time of the font (the sfd file), while the times in the 'head' table contain the creation time of the truetype or opentype font file.
uint32 version currently 0x00000001
int64 FontForge's own timestamp (the date of the sources for fontforge)
int64 creation date of this font Not the creation date of the tt/ot file,
but the date the sfd file was created.
(not always accurate).
int64 last modification date of this font Not the modification date of the file,
but the time a glyph, etc. was last
changed in the font database.
(not always accurate)