RIP, Vernon...

[ttfautohint.git] / doc / ttfautohint-1.pandoc

% ttfautohint
% Werner Lemberg
%

<!--
  Copyright (C) 2011-2016 by Werner Lemberg.

  This file is part of the ttfautohint library, and may only be used,
  modified, and distributed under the terms given in `COPYING'.  By
  continuing to use, modify, or distribute this file you indicate that you
  have read `COPYING' and understand and accept it fully.

  The file `COPYING' mentioned in the previous paragraph is distributed
  with the ttfautohint library.
-->



Introduction
============

**ttfautohint** is a library written in\ C that takes a TrueType font as
the input, removes its bytecode instructions (if any), and returns a new
font where all glyphs are bytecode hinted using the information given by
FreeType's auto-hinting module.  The idea is to provide the excellent
quality of the auto-hinter on platforms that don't use FreeType.

The library has a single API function, `TTF_autohint`, which is described
[below](#the-ttfautohint-api).

Bundled with the library there are two front-end programs, [`ttfautohint`
and `ttfautohintGUI`](#ttfautohint-and-ttfautohintgui), being a command line
program and an application with a Graphics User Interface (GUI),
respectively.


What exactly are hints?
-----------------------

To cite [Wikipedia](http://en.wikipedia.org/wiki/Font_hinting):

> **Font hinting** (also known as **instructing**) is the use of
> mathematical instructions to adjust the display of an outline font so that
> it lines up with a rasterized grid.  At low screen resolutions, hinting is
> critical for producing a clear, legible text.  It can be accompanied by
> antialiasing and (on liquid crystal displays) subpixel rendering for
> further clarity.

and Apple's [TrueType Reference
Manual](https://developer.apple.com/fonts/TrueType-Reference-Manual/RM03/Chap3.html#features):

> For optimal results, a font instructor should follow these guidelines:
>
>  - At small sizes, chance effects should not be allowed to magnify small
>    differences in the original outline design of a glyph.
>
>  - At large sizes, the subtlety of the original design should emerge.

In general, there are three possible ways to hint a glyph.

 1. The font contains hints (in the original sense of this word) to guide
    the rasterizer, telling it which shapes of the glyphs need special
    consideration.  The hinting logic is partly in the font and partly in
    the rasterizer.  More sophisticated rasterizers are able to produce
    better rendering results.

    This is how Type\ 1 and CFF hints work.

 2. The font contains exact instructions (also called *bytecode*) on how to
    move the points of its outlines, depending on the resolution of the
    output device, and which intentionally distort the (outline) shape to
    produce a well-rasterized result.  The hinting logic is in the font;
    ideally, all rasterizers simply process these instructions to get the
    same result on all platforms.

    This is how TrueType hints work.

 3. The font gets auto-hinted (at run-time).  The hinting logic is
    completely in the rasterizer.  No hints in the font are used or needed;
    instead, the rasterizer scans and analyzes the glyphs to apply
    corrections by itself.

    This is how FreeType's auto-hinter works; see
    [below](#background-and-technical-details) for more.


What problems can arise with TrueType hinting?
----------------------------------------------

While it is relatively easy to specify PostScript hints (either manually or
by an auto-hinter that works at font creation time), creating TrueType
hints is far more difficult.  There are at least two reasons:

  - TrueType instructions form a programming language, operating at a very
    low level.  They are comparable to assembler code, thus lacking all
    high-level concepts to make programming more comfortable.

    Here an example how such code looks like:

    ```
        SVTCA[0]
        PUSHB[ ]  /* 3 values pushed */
        18 1 0
        CALL[ ]
        PUSHB[ ]  /* 2 values pushed */
        15 4
        MIRP[01001]
        PUSHB[ ]  /* 3 values pushed */
        7 3 0
        CALL[ ]
    ```

    Another major obstacle is the fact that font designers usually aren't
    programmers.

  - It is very time consuming to manually hint glyphs.  Given that the
    number of specialists for TrueType hinting is very limited, hinting a
    large set of glyphs for a font or font family can become very expensive.


Why ttfautohint?
----------------

The ttfautohint library brings the excellent quality of FreeType rendering
to platforms that don't use FreeType, yet require hinting for text to look
good -- like Microsoft Windows.  Roughly speaking, it converts the glyph
analysis done by FreeType's auto-hinting module to TrueType bytecode.
Internally, the auto-hinter's algorithm resembles PostScript hinting
methods; it thus combines all three hinting methods discussed
[previously](#what-exactly-are-hints).

The simple interface of the front-ends (both on the command line and with
the GUI) allows quick hinting of a whole font with a few mouse clicks or a
single command on the prompt.  As a result, you get better rendering results
with web browsers, for example.

Across Windows rendering environments today, fonts processed with
ttfautohint look best with ClearType enabled.  This is the default for
Windows\ 7.  Good visual results are also seen in recent MacOS\ X versions
and GNU/Linux systems (including Android, ChromeOS, and other mobile
operating systems) that use FreeType for rendering glyphs.

The goal of the project is to generate a 'first pass' of hinting that font
developers can refine further for ultimate quality.


'Smooth' hinting
----------------

Fundamentally, there are two approaches to hinting. The older approach,
let's call it 'sharp', popular when text was rendered in pure
black-and-white, was to make all stems round to full pixels so that in a
text line, all stems would be either one pixel or (at a larger point size)
two pixels.  When grayscale antialiasing came about, this approach actually
started harming the rendering rather than helping it, because the horizontal
and vertical stems would render very dark but round or diagonal stems would
render very light.

So a new approach was developed, let's call it 'fuzzy', where all stems and
other elements are equalized so that in grayscale (or ClearType) rendering,
they all are of roughly equal color.  This means that stems are not rounded
to full pixels but in fact to fractions of a pixel.  However, with
black-and-white renderers, this approach yields poor results because in
black-and-white you cannot render a fraction of a pixel, so some stems
become one pixel and some become two.

The TrueType autohinters in [FontForge] and [FontLab Studio], to name two
well-known font editors, take the 'sharp' approach, while the TrueType
autohinter in ttfautohint takes the 'fuzzy' approach.

In theory, a hybrid approach is possible, using TrueType conditional hints:
If the rasterizer is black-and-white, 'sharp' rendering could happen, while
if the rasterizer is ClearType, the 'fuzzy' rendering could be used.  It is
not intended to add black-and-white auto-hinting to ttfautohint.  However,
it is planned to develop an interface so that ttfautohint can cooperate with
font editors, providing this hybrid hinting.



`ttfautohint` and `ttfautohintGUI`
==================================

On all supported platforms (GNU/Linux, Windows, and Mac OS\ X), the GUI
looks quite similar; the used toolkit is [Qt], which in turn uses the
platform's native widgets.

![`ttfautohintGUI` on GNU/Linux running KDE](img/ttfautohintGUI.png)

Both the GUI and console version share the same features, to be discussed in
the next subsection.

**Warning: ttfautohint cannot always process a font a second time.**
If the font contains composite glyphs, and option [`-c`](#hint-composites)
is used, reprocessing with ttfautohint will fail.  For this reason it is
strongly recommended to *not* delete the original, unhinted font so that you
can always rerun ttfautohint.


Calling `ttfautohint`
---------------------

```
    ttfautohint [OPTION]... [IN-FILE [OUT-FILE]]
```

The command-line binary, `ttfautohint`, works like a Unix filter, this is,
it reads data from standard input if no input file name is given, and it
sends its output to standard output if no output file name is specified.

A typical call looks like the following.

```
    ttfautohint -v -f latn foo.ttf foo-autohinted.ttf
```

For demonstration purposes, here the same using a pipe and redirection.
Note that Windows's default command line interpreter, `cmd.exe`, doesn't
support piping with binary files, unfortunately.

```
    cat foo.ttf | ttfautohint -v -f latn > foo-autohinted.ttf
```


Calling `ttfautohintGUI`
------------------------

```
    ttfautohintGUI [OPTION]...
```

`ttfautohintGUI` doesn't send any output to a console; however, it accepts
the same command line options as `ttfautohint`, setting default values for
the GUI.


Options
-------

Long options can be given with one or two dashes, and with and without an
equal sign between option and argument.  This means that the following forms
are acceptable: `-foo=`*bar*, `--foo=`*bar*, `-foo`\ *bar*, and
`--foo`\ *bar*.

Below, the section title refers to the command's label in the GUI (if
applicable), then comes the name of the corresponding long command line
option and its short equivalent, followed by a description.

Background and technical details on the meaning of the various options are
given [afterwards](#background-and-technical-details).

### Hint Set Range Minimum, Hint Set Range Maximum

See '[Hint Sets](#hint-sets)' for a definition and explanation.

`--hinting-range-min=`*n*, `-l`\ *n*
:   The minimum PPEM value (in pixels) at which hint sets are created.  The
    default value for *n* is\ 8.

`--hinting-range-max=`*n*, `-r`\ *n*
:   The maximum PPEM value (in pixels) at which hint sets are created.  The
    default value for *n* is 50.

Increasing the range given by `-l` and `-r` normally makes the font's
bytecode larger.

### Default Script

`--default-script=`*s*, `-D`\ *s*
:   Set default script to tag *s*, which is a string consisting of four
    lowercase characters like `latn` or `dflt`.  It is needed to specify the
    OpenType default script: After applying all features that are handled
    specially (like small caps or superscript), ttfautohint uses this value
    for the remaining features.  The default value is `latn`.  See
    [below](#opentype-features) for more details.

### Fallback Script

`--fallback-script=`*s*, `-f`\ *s*
:   Set fallback script to tag *s*, which is a string consisting of four
    characters like `latn` or `dflt`.  It gets used for for all glyphs that
    can't be assigned to a script automatically.  The default value is
    `none`.  See [below](#scripts) for more details.

`--fallback-scaling`, `-S`
:   Use scaling for glyphs covered by the fallback script, not hinting.  See
    [below](#scripts) for more details.

### Hinting Limit

`--hinting-limit=`*n*, `-G`\ *n*
:   The *hinting limit* is the PPEM value (in pixels) where hinting gets
    switched off (using the `INSTCTRL` bytecode instruction, not the `gasp`
    table data); it does not influence the file size.  The default value for
    *n* is 200, which means that the font is not hinted for PPEM values
    larger than 200.

    Note that hinting in the range 'hinting-range-max' up to 'hinting-limit'
    uses the hinting configuration for 'hinting-range-max'.

    To omit a hinting limit, use `--hinting-limit=0` (or check the 'No
    Hinting Limit' box in the GUI).  Since this causes internal math
    overflow in the rasterizer for large pixel values (>\ 1500px approx.) it
    is strongly recommended to not use this except for testing purposes.

### x Height Increase Limit

`--increase-x-height=`*n*, `-x`\ *n*
:   Normally, ttfautohint rounds the x\ height to the pixel grid, with a
    slight preference for rounding up (to use the terminology of TrueType's
    'Super Round' bytecode instruction, the threshold is 5/8px).  If this
    flag is set, values in the range 6\ PPEM to *n*\ PPEM are much more
    often rounded up (setting the threshold to 13/16px).  The default value
    for *n* is 14.  Use this flag to increase the legibility of small sizes
    if necessary; you might get weird rendering results otherwise for glyphs
    like 'a' or 'e', depending on the font design.

    To switch off this feature, use `--increase-x-height=0` (or check the
    'No x\ Height Increase' box in the GUI).  To switch off rounding the
    x\ height to the pixel grid in general, either partially or completely,
    see '[x Height Snapping Exceptions](#x-height-snapping-exceptions)'.

    The following FontForge snapshot images use the font '[Mertz
    Bold](https://github.com/vernnobile/mertzFont/tree/master/FINAL/Mertz-Bold)'
    from Vernon Adams.

    ![At 17px, without option `-x` and '`-w ""`', the hole in glyph 'e'
      looks very grey in the FontForge snapshot, and the GDI ClearType
      rendering (which is the default on older Windows versions) fills it
      completely with black because it uses B/W rendering along the y\ axis.
      FreeType's 'light' autohint mode (which corresponds to ttfautohint's
      'smooth' stem width algorithm) intentionally aligns horizontal lines
      to non-integer (but still discrete) values to avoid large glyph shape
      distortions.](img/e-17px-x14.png)

    ![The same, this time with option `-x 17` (and
      '`-w ""`').](img/e-17px-x17.png)

### x Height Snapping Exceptions

`--x-height-snapping-exceptions=`*string*, `-X`\ *string*
:   A list of comma separated PPEM values or value ranges at which no
    x\ height snapping shall be applied.  A value range has the form
    *value*~1~`-`*value*~2~, meaning *value*~1~\ <= PPEM <=\ *value*~2~.
    *value*~1~ or *value*~2~ (or both) can be missing; a missing value is
    replaced by the beginning or end of the whole interval of valid PPEM
    values, respectively (6\ to 32767).  Whitespace is not significant;
    superfluous commas are ignored, and ranges must be specified in
    increasing order.  For example, the string `"7-9, 11, 13-"` means the
    values 7, 8, 9, 11, 13, 14, 15, etc.  Consequently, if the supplied
    argument is `"-"`, no x\ height snapping takes place at all.  The
    default is the empty string (`""`), meaning no snapping exceptions.

    Normally, x\ height snapping means a slight increase in the overall
    vertical glyph size so that the height of lowercase glyphs gets aligned
    to the pixel grid (this is a global feature, affecting *all* glyphs of a
    font).  However, having larger vertical glyph sizes is not always
    desired, especially if it is not possible to adjust the `usWinAscent`
    and `usWinDescent` values from the font's `OS/2` table so that they are
    not too tight.  See '[Windows Compatibility](#windows-compatibility)'
    for more details.

### Fallback Stem Width

`--fallback-stem-width=`*n*, `-H`\ *n*
:   Set the horizontal stem width (hinting) value for all scripts that lack
    proper standard characters in the font.  The value is given in font
    units and must be a positive integer.  If not set, ttfautohint uses a
    hard-coded default (50\ units at 2048 units per EM, and linearly scaled
    for other UPEM values, for example 24\ units at 1000 UPEM).

    For symbol fonts, you need option `--fallback-script` too (to set up a
    script at all).

    In the GUI, uncheck the 'Default Fallback Stem Width' box to activate
    this feature.

### Windows Compatibility

`--windows-compatibility`, `-W`
:   This option makes ttfautohint add two artificial blue zones, positioned
    at the `usWinAscent` and `usWinDescent` values (from the font's `OS/2`
    table).  The idea is to help ttfautohint so that the hinted glyphs stay
    within this horizontal stripe since Windows clips everything falling
    outside.

    There is a general problem with tight values for `usWinAscent` and
    `usWinDescent`; a good description is given in the [Vertical Metrics
    How-To](http://typophile.com/node/13081).  Additionally, there is a
    special problem with tight values if used in combination with
    ttfautohint because the auto-hinter tends to slightly increase the
    vertical glyph dimensions at smaller sizes to improve legibility.  This
    enlargement can make the heights and depths of glyphs exceed the range
    given by `usWinAscent` and `usWinDescent`.

    If ttfautohint is part of the font creation tool chain, and the font
    designer can adjust those two values, a better solution instead of using
    option `-W` is to reserve some vertical space for 'padding': For the
    auto-hinter, the difference between a top or bottom outline point before
    and after hinting is less than 1px, thus a vertical padding of 2px is
    sufficient.  Assuming a minimum hinting size of 6ppem, adding two pixels
    gives an increase factor of 8÷6 = 1.33.  This is near to the default
    baseline-to-baseline distance used by TeX and other sophisticated text
    processing applications, namely 1.2×designsize, which gives satisfying
    results in most cases.  It is also near to the factor 1.25 recommended
    in the abovementioned how-to.  For example, if the vertical extension of
    the largest glyph is 2000 units (assuming that it approximately
    represents the designsize), the sum of `usWinAscent` and `usWinDescent`
    could be 1.25×2000 = 2500.

    In case ttfautohint is used as an auto-hinting tool for fonts that can
    be no longer modified to change the metrics, option `-W` in combination
    with '`-X "-"`' to suppress any vertical enlargement should prevent
    almost all clipping.

### Adjust Subglyphs

`--adjust-subglyphs`, `-p`
:   *Adjusting subglyphs* makes a font's original bytecode be applied to all
    glyphs before it is replaced with bytecode created by ttfautohint.  This
    makes only sense if your font already has some hints in it that modify
    the shape even at EM size (normally 2048px); in particular, some CJK
    fonts need this because the bytecode is used to scale and shift
    subglyphs (hence the option's long name).  For most fonts, however, this
    is not the case.

### Hint Composites

`--composites`, `-c`
:   By default, the components of a composite glyph get hinted separately.
    If this flag is set, the composite glyph itself gets hinted (and the
    hints of the components are ignored).  Using this flag increases the
    bytecode size a lot, however, it might yield better hinting results.

    If this option is used (and a font actually contains composite glyphs),
    ttfautohint currently cannot reprocess its own output for technical
    reasons, see [below](#the-.ttfautohint-glyph).

### Symbol Font

`--symbol`, `-s`
:   Process a font that ttfautohint would refuse otherwise because it can't
    find a single standard character for any of the supported scripts.

    For all scripts that lack proper standard characters, ttfautohint uses a
    default (hinting) value for the standard stem width instead of deriving
    it from a script's set of standard characters (for the latin script, one
    of them is character 'o').

    Use this option (usually in combination with the
    [`--fallback-script`](#fallback-script) and/or
    [`--fallback-stem-width`](#fallback-stem-width) option) to hint symbol
    or dingbat fonts or math glyphs, for example.

### Dehint

`--dehint`, `-d`
:   Strip off all hints without generating new hints.  Consequently, all
    other hinting options are ignored.  This option is intended for testing
    purposes.

### ttfautohint Info

`--no-info`, `-n`
:   Don't add ttfautohint version and command line information to the
    version string or strings (with name ID\ 5) in the font's `name` table.
    In the GUI, it corresponds to value 'None' in the 'ttfautohint
    info' combo box.

    This option is mutually exclusive with option `-I`.

`--detailed-info`, `-I`
:   Add ttfautohint version and command line information to the version
    string or strings (with name ID\ 5) in the font's `name` table.  In the
    GUI, it corresponds to value 'Version and Parameters' in the
    'ttfautohint info' combo box.

    This option is mutually exclusive with option `-n`.

If neither `-n` nor `-I` is set, the string '`ttfautohint (vNNN)`' gets
added to the `name` table (with *NNN* the current version); this correponds
to value 'Version' in the 'ttfautohint info' combo box.

### Add TTFA Info Table

`--ttfa-table`, `-t`
:   Add an SFNT table called `TTFA` to the output font that holds a dump of
    all parameters; the data resembles the format of the `--debug` option's
    parameter listing.  In particular, it lists all ttfautohint control
    instructions (which are *not* shown in the `name` table info).  This
    option is mainly for archival purposes so that all information used to
    create a font is stored in the font itself.  Note that such a `TTFA`
    table gets ignored by all TrueType rendering engines.

    Forthcoming versions of the ttfautohint front-ends will be able to use
    this data so that a font can be processed another time with exactly the
    same parameters, thus providing a means for round-tripping fonts.

### Family Suffix

`--family-suffix=`*string*, `-F`\ *string*
:   A string that gets appended to the family name in entries with IDs 1, 4,
    6, 16, and\ 21 in the font's `name` table.  Allowed input is ASCII in
    the range 0x20-0x7E except characters `%()/<>[]{}`.

    Assuming an input family name 'Foo', a full name 'Foo Bold', and a
    family suffix '\ 1', the output family name will be 'Foo 1' and the
    full name 'Foo 1 Bold'.  For the PostScript name in ID\ 6, ttfautohint
    uses the suffix with space characters removed (for example 'Foo1Bold').

    This option is mainly for testing purposes, enabling the operating
    system to simultaneously display several instances of a font that are
    processed with different ttfautohint parameters.

### Strong Stem Width and Positioning

`--strong-stem-width=`*string*, `-w`\ *string*
:   ttfautohint offers two different routines to handle (horizontal) stem
    widths and stem positions: 'smooth' and 'strong'.  The former uses
    discrete values that slightly increase the stem contrast with almost no
    distortion of the outlines, while the latter snaps both stem widths and
    stem positions to integer pixel values as much as possible, yielding a
    crisper appearance at the cost of much more distortion.

    These two routines are mapped onto three possible rendering targets:

    - grayscale rendering, with or without optimization for subpixel
      positioning (e.g. Android)

    - 'GDI ClearType' rendering: the rasterizer version, as returned by the
      GETINFO bytecode instruction, is in the range 36\ <= version <\ 38 and
      ClearType is enabled (e.g. Windows XP)

    - 'DirectWrite ClearType' rendering: the rasterizer version, as returned
      by the GETINFO bytecode instruction, is >=\ 38, ClearType is enabled,
      and subpixel positioning is enabled also (e.g. Internet Explorer\ 9
      running on Windows\ 7)

    GDI ClearType uses a mode similar to B/W rendering along the vertical
    axis, while DW ClearType applies grayscale rendering.  Additionally,
    only DW ClearType provides subpixel positioning along the x\ axis.  For
    what it's worth, the rasterizers version\ 36 and version\ 38 in
    Microsoft Windows are two completely different rendering engines.

    The command line option expects *string* to contain up to three letters
    with possible values '`g`' for grayscale, '`G`' for GDI ClearType, and
    '`D`' for DW ClearType.  If a letter is found in *string*, the strong
    stem width routine is used for the corresponding rendering target (and
    smooth stem width handling otherwise).  The default value is '`G`', which
    means that strong stem width handling is activated for GDI ClearType
    only.  To use smooth stem width handling for all three rendering
    targets, use the empty string as an argument, usually connoted with
    '`""`'.

    In the GUI, simply set the corresponding check box to select the strong
    width routine for a given rendering target.  If you unset the check box,
    the smooth width routine gets used.

    The following images again use the font 'Mertz Bold'.

    ![The left part shows the glyph 'g' unhinted at 26px, the right part
     with hints, using the 'smooth' stem algorithm.](img/ff-g-26px.png)

    ![The same, but this time using the 'strong'
     algorithm.  Note how the stems are aligned to the pixel
     grid.](img/ff-g-26px-wD.png)

### Control Instructions File
`--control-file=`*file*, `-m`\ *file* (not in `ttfautohintGUI`)
:   Specify the name of a control instructions file to manually tweak the
    hinting process.  This feature can be used to correct glitches in
    ttfautohint's hinting algorithm.  The syntax used in a control
    instructions file is given [below](#control-instructions).

### Miscellaneous

Watch input files (`ttfautohintGUI` only)
:   If this checkbox is set, automatically regenerate the output file as
    soon as an input file (either the font or the control instructions file)
    gets modified.

    Pressing the 'Run' button starts watching.  If an error occurs, watching
    stops and must be restarted with the 'Run' button.

`--ignore-restrictions`, `-i`
:   By default, fonts that have bit\ 1 set in the 'fsType' field of the
    `OS/2` table are rejected.  If you have a permission of the font's legal
    owner to modify the font, specify this command line option.

    If this option is not set, `ttfautohintGUI` shows a dialogue to handle
    such fonts if necessary.

`--help`, `-h`
:   On the console, print a brief documentation on standard output and exit.
    This doesn't work with `ttfautohintGUI` on MS Windows.

`--version`, `-v`
:   On the console, print version information on standard output and exit.
    This doesn't work with `ttfautohintGUI` on MS Windows.

`--ttfa-info`, `-T` (not in `ttfautohintGUI`)
:   Print [`TTFA` table](#add-ttfa-info-table) of the input font on standard
    output if present, then exit.

`--debug`
:   Print *a lot* of debugging information on standard error while
    processing a font (you should redirect stderr to a file).  This
    doesn't work with `ttfautohintGUI` on MS Windows.

    To reduce the amount of debug data it is recommended to restrict the
    hinting process to a single PPEM value, e.g.,

    ```
       ttfautohint --debug -l 15 -r 15 ... > debug.txt 2>&1
    ```



Background and Technical Details
================================

[Real-Time Grid Fitting of Typographic
Outlines](http://www.tug.org/TUGboat/tb24-3/lemberg.pdf) is a scholarly
paper that describes FreeType's auto-hinter in some detail.  Regarding the
described data structures it is slightly out of date, but the algorithm
itself hasn't changed in general.

The next few subsections are mainly based on this article, introducing some
important concepts.  Note that ttfautohint only does hinting along the
vertical direction (modifying y\ coordinates only).


Segments and Edges
------------------

A glyph consists of one or more *contours* (this is, closed curves).  For
example, glyph 'O' consists of two contours, while glyph 'I' has only one.

![The letter 'O' has two contours, an inner and an outer one, while letter
  'I' has only an outer contour.](img/o-and-i)

A *segment* is a series of consecutive points of a contour (including its
Bézier control points) that are approximately aligned along a coordinate
axis.  A segment has one of three possible directions: left, right, or none
(which means neither left nor right), derived from the TrueType outline
directions.  ttfautohint itself creates segments that contain at least two
points.  Using control instructions, however, it is possible to create
one-point segments, which are useful for fine-tuning the hinting process.

![A serif.  Contour and control points are represented by squares and
  circles, respectively.  The bottom 'line' DE is approximately aligned
  along the horizontal axis, thus it forms a segment of 7\ points.  Together
  with the two other horizontal segments, BC and FG, they form two edges
  (BC+FG, DE).](img/segment-edge)

An *edge* corresponds to a single coordinate value (allowing for a small
threshold) on the main dimension that collects one or more segments, all
pointing into the same direction (either left or right, all others are
ignored).  While finding segments is done on the unscaled outline, finding
edges is bound to the device resolution.  See [below](#hint-sets) for an
example.

In general, segments and edges pointing into different directions 'repel'
each other, thus preventing alignment on the same vertical coordinate if
they are near.  Note that this is a simplification, but it should help
understand how to manipulate and/or create segments in control instructions
files.

The analysis to find segments and edges is specific to a writing
system, see [below](#writing-systems).


Feature Analysis
----------------

The auto-hinter analyzes a font in two steps.  Right now, everything
described here happens for the horizontal axis only, providing vertical
hinting.

  * Global Analysis

    This affects the hinting of all glyphs, trying to give them a uniform
    appearance.

      + Compute standard horizontal stem width of the font.  The value
        is normally taken from glyphs that resemble letter 'o'.

      + Compute blue zones, see [below](#blue-zones).

    If the stem widths of single glyphs differ by a large value, or if
    ttfautohint fails to find proper blue zones, hinting becomes quite poor,
    possibly leading even to severe shape distortions.


Table: script-specific standard characters of the 'latin' writing system

    Script    Standard characters
  ----------  ---------------------
  `arab`      'ـ', U+0640, ARABIC TATWEEL
              'ل', U+0644, ARABIC LETTER LAM
              'ح', U+062D, ARABIC LETTER HAH
  `armn`      'օ', U+0585, ARMENIAN SMALL LETTER OH
              'Օ', U+0555, ARMENIAN CAPITAL LETTER OH
  `beng`      '০', U+09E6, BENGALI DIGIT ZERO
              '৪', U+09EA, BENGALI DIGIT FOUR
  `cyrl`      'о', U+043E, CYRILLIC SMALL LETTER O
              'О', U+041E, CYRILLIC CAPITAL LETTER O
  `cher`      'Ꭴ', U+13A4, CHEROKEE LETTER U
              'Ꮕ', U+13C5, CHEROKEE LETTER NV
              'ꮕ', U+AB95, CHEROKEE SMALL LETTER NV
  `deva`      'ठ', U+0920, DEVANAGARI LETTER TTHA
              'व', U+0935, DEVANAGARI LETTER VA
              'ट', U+091F, DEVANAGARI LETTER TTA
  `ethi`      'ዐ', U+12D0, ETHIOPIC SYLLABLE PHARYNGEAL A
  `geor`      'ი', U+10D8, GEORGIAN LETTER IN
              'ე', U+10D4, GEORGIAN LETTER EN
              'ა', U+10D0, GEORGIAN LETTER AN
  `geok`      'Ⴖ', U+10B6, GEORGIAN CAPITAL LETTER GHAN
              'Ⴑ', U+10B1, GEORGIAN CAPITAL LETTER SAN
              'ⴙ', U+2D19, GEORGIAN SMALL LETTER CHIN
  `grek`      'ο', U+03BF, GREEK SMALL LETTER OMICRON
              'Ο', U+039F, GREEK CAPITAL LETTER OMICRON
  `gujr`      'ટ', U+0A9F, GUJARATI LETTER TTA
              '૦', U+0AE6, GUJARATI DIGIT ZERO
  `guru`      'ਠ', U+0A20, GURMUKHI LETTER TTHA
              'ਰ', U+0A30, GURMUKHI LETTER RA
              '੦', U+0A66, GURMUKHI DIGIT ZERO
  `hebr`      'ם', U+05DD, HEBREW LETTER FINAL MEM
  `knda`      '೦', U+0CE6, KANNADA DIGIT ZERO
              'ಬ', U+0CAC, KANNADA LETTER BA
  `khmr`      '០', U+17E0, KHMER DIGIT ZERO
  `lao`       '໐', U+0ED0, LAO DIGIT ZERO
  `latn`      'o', U+006F, LATIN SMALL LETTER O
              'O', U+004F, LATIN CAPITAL LETTER O
              '0', U+0030, DIGIT ZERO
  `mlym`      'ഠ', U+0D20, MALAYALAM LETTER TTHA
              'റ', U+0D31, MALAYALAM LETTER RRA
  `mymr`      'ဝ', U+101D, MYANMAR LETTER WA
              'င', U+1004, MYANMAR LETTER NGA
              'ဂ', U+1002, MYANMAR LETTER GA
  `sinh`      'ට', U+0DA7, SINHALA LETTER ALPAPRAANA TTAYANNA
  `taml`      '௦', U+0BE6, TAMIL DIGIT ZERO
  `telu`      '౦', U+0C66, TELUGU DIGIT ZERO
              '౧', U+0C67, TELUGU DIGIT ONE
  `thai`      'า', U+0E32, THAI CHARACTER SARA AA
              'ๅ', U+0E45, THAI CHARACTER LAKKHANGYAO
              '๐', U+0E50, THAI DIGIT ZERO


Table: standard characters of the 'latin' writing system, special scripts

    Script    Standard characters
  ----------  ---------------------
  `khms`      '᧡', U+19E1, KHMER SYMBOL MUOY KOET
              '᧪', U+19EA, KHMER SYMBOL DAP KOET
  `latb`      'ₒ', U+2092, LATIN SUBSCRIPT SMALL LETTER O
              '₀', U+2080, SUBSCRIPT ZERO
  `latp`      'ᵒ', U+1D52, MODIFIER LETTER SMALL O
              'ᴼ', U+1D3C, MODIFIER LETTER CAPITAL O
              '⁰', U+2070, SUPERSCRIPT ZERO


  * Glyph Analysis

    This is a per-glyph operation.

      + Find segments and edges.

      + Link edges together to find stems and serifs.  The abovementioned
        paper gives more details on what exactly constitutes a stem or a
        serif and how the algorithm works.


Blue Zones
----------

![Two blue zones relevant to the glyph 'a'.  Vertical point coordinates of
  *all* glyphs within these zones are aligned, provided the blue zone is
  active (this is, its vertical size is smaller than
  3/4\ pixels).](img/blue-zones)

Outlines of certain characters are used to determine *blue zones*.  This
concept is the same as with Type\ 1 fonts: All glyph points that lie in
certain small horizontal zones get aligned vertically.

Here a series of tables that show the blue zone characters of the latin
writing system's available scripts; the values are hard-coded in the source
code.  Since the auto-hinter takes mean values it is not necessary that all
characters of a zone are present.

'Round' characters in blue zones (e.g., the top and bottom of 'O' or the
bottom of 'g') are used to control overshoot handling.

Blue zones marked with an asterisk are x\ height blue zones, which are
adjusted to be on the pixel grid (to improve rendering at small sizes) by
scaling the remaining blue zones before they are adjusted to the grid.  See
also option [`--increase-x-height`](#x-height-increase-limit).


Table: `arab` blue zones

  ID    Blue zone                              Characters
  ----  -----------                            ------------
  1     top of letters with vertical stroke    ا إ ل ك ط ظ
  2     bottom of letters                      ت ث ط ظ ك
  3     glyph joining                          ـ


Table: `armn` blue zones

  ID    Blue zone                              Characters
  ----  -----------                            ------------
  1     top of capital letters                 Ա Մ Ւ Փ Բ Գ Դ Օ
  2     bottom of capital letters              Ւ Ո Փ Ճ Շ Ս Տ Օ
  3     top of ascenders of small letters      ե է ի մ վ փ ֆ փ
  4*    top of small letters                   ա յ ւ ս գ ջ ր օ
  5     bottom of small letters                հ ո ճ ա ե ծ ս օ
  6     bottom of descenders of small letters  բ ը ի լ ղ պ փ ց


Table: `beng` blue zones

  ID    Blue zone                              Characters
  ----  -----------                            ------------
  1     baseline (flat glyphs only)            অ ড ত ন ব ভ ল ক
  2     top of ascenders                       ই ট ঠ ি ী ৈ ৗ
  3*    top of baseline                        ও এ ড ত ন ব ল ক
  4     bottom of base characters              অ ড ত ন ব ভ ল ক

Contrary to scripts like latin, the baseline in Bengali is on the top, and
we hint from top to bottom.


Table: `cher` blue zones

  ID    Blue zone                              Characters
  ----  -----------                            ------------
  1     top of capital letters                 Ꮖ Ꮋ Ꭼ Ꮓ Ꭴ Ꮳ Ꭶ Ꮥ
  2     bottom of capital letters              Ꮖ Ꮋ Ꭼ Ꮓ Ꭴ Ꮳ Ꭶ Ꮥ
  3     top of ascenders of small letters      ꮒ ꮤ ꮶ ꭴ ꭾ ꮗ ꮝ ꮿ
  4*    top of small letters                   ꮖ ꭼ ꮓ ꮠ ꮳ ꭶ ꮥ ꮻ
  5     bottom of small letters                ꮖ ꭼ ꮓ ꮠ ꮳ ꭶ ꮥ ꮻ
  6     bottom of descenders of small letters  ᏸ ꮐ ꭹ ꭻ


Table: `cyrl` blue zones

  ID    Blue zone                              Characters
  ----  -----------                            ------------
  1     top of capital letters                 Б В Е П З О С Э
  2     bottom of capital letters              Б В Е Ш З О С Э
  3*    top of small letters                   х п н ш е з о с
  4     bottom of small letters                х п н ш е з о с
  5     bottom of descenders of small letters  р у ф


Table: `deva` blue zones

  ID    Blue zone                              Characters
  ----  -----------                            ------------
  1     top of ascenders                       ई ऐ ओ औ ि ी ो ौ
  2     top of baseline                        क म अ आ थ ध भ श
  3*    top of baseline (flat glyphs only)     क न म उ छ ट ठ ड
  4     bottom of base characters              क न म उ छ ट ठ ड
  5     bottom of descenders                   ु ृ

Contrary to scripts like latin, the baseline in Devanagari is on the top,
and we hint from top to bottom.  Note that some fonts have extreme variation
in the height of the round elements in Zone\ 3; for this reason we also
define Zone\ 1, which must be always present.


Table: `ethi` blue zones

  ID    Blue zone                              Characters
  ----  -----------                            ------------
  1     top of letters                         ሀ ሃ ዘ ፐ ማ በ ዋ ዐ
  2     bottom of letters                      ለ ሐ በ ዘ ሀ ሪ ዐ ጨ


Table: `geok` blue zones

  ID    Blue zone                              Characters
  ----  -----------                            ------------
  1     top of Asomtavruli letters             Ⴑ Ⴇ Ⴙ Ⴜ Ⴄ Ⴅ Ⴓ Ⴚ
  2     bottom of Asomtavruli letters          Ⴄ Ⴅ Ⴇ Ⴈ Ⴆ Ⴑ Ⴊ Ⴋ
  3*    top of Nuskhuri letters                ⴁ ⴗ ⴂ ⴄ ⴅ ⴇ ⴔ ⴖ
  4     bottom of Nuskhuri letters             ⴈ ⴌ ⴖ ⴎ ⴃ ⴆ ⴋ ⴢ
  5     top of ascender Nuskhuri letters       ⴐ ⴑ ⴓ ⴕ ⴙ ⴛ ⴡ ⴣ
  6     bottom of Nuskhuri descender letters   ⴄ ⴅ ⴔ ⴕ ⴁ ⴂ ⴘ ⴝ

Georgian Asomtavruli and Nuskhuri form the old ecclesiastical script,
Khutsuri.  Note that fonts show a great variation in height and depth of
ascender and descender letter forms.


Table: `geor` blue zones

  ID    Blue zone                              Characters
  ----  -----------                            ------------
  1*    top of Mkhedruli letters               გ დ ე ვ თ ი ო ღ
  2     bottom of Mkhedruli letters            ა ზ მ ს შ ძ ხ ჰ
  3     top of ascender Mkhedruli letters      ს ხ ქ ზ მ შ ჩ წ
  4     bottom of descender Mkhedruli letters  ე ვ ჟ ტ უ ფ ქ ყ

Georgian Mkhedruli support is incomplete; it doesn't yet contain characters
for Mtavruli (which are not yet encoded in Unicode), the uppercase glyph
variants of Mkhedruli.


Table: `grek` blue zones

  ID    Blue zone                              Characters
  ----  -----------                            ------------
  1     top of capital letters                 Γ Β Ε Ζ Θ Ο Ω
  2     bottom of capital letters              Β Δ Ζ Ξ Θ Ο
  3     top of 'small beta' like letters       β θ δ ζ λ ξ
  4*    top of small letters                   α ε ι ο π σ τ ω
  5     bottom of small letters                α ε ι ο π σ τ ω
  6     bottom of descenders of small letters  β γ η μ ρ φ χ ψ


Table: `gujr` blue zones

  ID    Blue zone                              Characters
  ----  -----------                            ------------
  1*    top of letters                         ત ન ઋ ઌ છ ટ ર ૦
  2     bottom of letters                      ખ ગ ઘ ઞ ઇ ઈ ઠ જ
  3     top of ascenders                       ઈ ઊ િ ી લી શ્ચિ જિ સી
  4     bottom of descenders                   ુ ૃ ૄ ખુ છૃ છૄ
  5     top of Gujarati digits                 ૦ ૧ ૨ ૩ ૭


Table: `guru` blue zones

  ID    Blue zone                              Characters
  ----  -----------                            ------------
  1     top of ascenders                       ਇ ਈ ਉ ਏ ਓ ੳ ਿ ੀ
  2     top of baseline                        ਕ ਗ ਙ ਚ ਜ ਤ ਧ ਸ
  3*    top of baseline (flat glyphs only)     ਕ ਗ ਙ ਚ ਜ ਤ ਧ ਸ
  4     bottom of characters                   ਅ ਏ ਓ ਗ ਜ ਠ ਰ ਸ
  5     top of Gurmukhi digits                 ੦ ੧ ੨ ੩ ੭


Table: `hebr` blue zones

  ID    Blue zone                              Characters
  ----  -----------                            ------------
  1     top of letters                         ב ד ה ח ך כ ם ס
  2     bottom of letters                      ב ט כ ם ס צ
  3     bottom of descenders of letters        ק ך ן ף ץ


Table: `knda` blue zones

  ID    Blue zone                              Characters
  ----  -----------                            ------------
  1     top of letters                         ಇ ಊ ಐ ಣ ಸಾ ನಾ ದಾ ರಾ
  2     bottom of letters                      ಅ ಉ ಎ ಲ ೦ ೨ ೬ ೭


Table: `khmr` blue zones

  ID    Blue zone                              Characters
  ----  -----------                            ------------
  1*    top of letters                         ខ ទ ន ឧ ឩ ា
  2     top of subscript cluster components    ក្ក ក្ខ ក្គ ក្ថ
  3     bottom of letters                      ខ ឃ ច ឋ ប ម យ ឲ
  4     bottom of descenders                   ត្រ រៀ ឲ្យ អឿ
  5     bottom of large descenders             ន្ត្រៃ ង្ខ្យ ក្បៀ ច្រៀ ន្តឿ ល្បឿ


Table: `khms` blue zones

  ID    Blue zone                              Characters
  ----  -----------                            ------------
  1*    top of symbols for waxing              ᧠ ᧡
  2     bottom of symbols for waning           ᧶ ᧹

Khmer symbols are used for lunar dates.


Table: `lao` blue zones

  ID    Blue zone                              Characters
  ----  -----------                            ------------
  1*    top of letters                         າ ດ ອ ມ ລ ວ ຣ ງ
  2     bottom of letters                      າ ອ ບ ຍ ຣ ຮ ວ ຢ
  3     top of ascenders                       ປ ຢ ຟ ຝ
  4     top of large ascenders                 ໂ ໄ ໃ
  5     bottom of descenders                   ງ ຊ ຖ ຽ ໆ ຯ


Table: `latb` blue zones

  ID    Blue zone                                 Characters
  ----  -----------                               ------------
  1     top of capital characters                 ₀ ₃ ₅ ₇ ₈
  2     bottom of capital characters              ₀ ₁ ₂ ₃ ₈
  3     top of 'small f' like characters          ᵢ ⱼ ₕ ₖ ₗ
  4*    top of small characters                   ₐ ₑ ₒ ₓ ₙ ₛ ᵥ ᵤ ᵣ
  5     bottom of small characters                ₐ ₑ ₒ ₓ ₙ ₛ ᵥ ᵤ ᵣ
  6     bottom of descenders of small characters  ᵦ ᵧ ᵨ ᵩ ₚ

Subscript latin characters are similar to normal latin characters.


Table: `latn` blue zones

  ID    Blue zone                              Characters
  ----  -----------                            ------------
  1     top of capital letters                 T H E Z O C Q S
  2     bottom of capital letters              H E Z L O C U S
  3     top of 'small f' like letters          f i j k d b h
  4*    top of small letters                   x z r o e s c
  5     bottom of small letters                x z r o e s c
  6     bottom of descenders of small letters  p q g j y


Table: `latp` blue zones

  ID    Blue zone                                 Characters
  ----  -----------                               ------------
  1     top of capital characters                 ⁰ ³ ⁵ ⁷ ᵀ ᴴ ᴱ ᴼ
  2     bottom of capital characters              ⁰ ¹ ² ³ ᴱ ᴸ ᴼ ᵁ
  3     top of 'small f' like characters          ᵇ ᵈ ᵏ ʰ ʲ ᶠ ⁱ
  4*    top of small characters                   ᵉ ᵒ ʳ ˢ ˣ ᶜ ᶻ
  5     bottom of small characters                ᵉ ᵒ ʳ ˢ ˣ ᶜ ᶻ
  6     bottom of descenders of small characters  ᵖ ʸ ᵍ

Superscript latin characters are similar to normal latin characters.


Table: `mlym` blue zones

  ID    Blue zone                              Characters
  ----  -----------                            ------------
  1     top of letters                         ഒ ട ഠ റ ച പ ച്ച പ്പ
  2     bottom of letters                      ട ഠ ധ ശ ഘ ച ഥ ല


Table: `mymr` blue zones

  ID    Blue zone                              Characters
  ----  -----------                            ------------
  1*    top of letters                         ခ ဂ င ဒ ဝ ၥ ၊ ။
  2     bottom of letters                      င ဎ ဒ ပ ဗ ဝ ၊ ။
  3     top of ascenders of characters         ဩ ြ ၍ ၏ ၆ ါ ိ
  3     bottom of descenders of letters        ဉ ည ဥ ဩ ဨ ၂ ၅ ၉


Table: `sinh` blue zones

  ID    Blue zone                              Characters
  ----  -----------                            ------------
  1     top of letters                         ඉ ක ඝ ඳ ප ය ල ෆ
  2     bottom of letters                      එ ඔ ඝ ජ ට ථ ධ ර
  3     bottom of descenders of letters        ද ඳ උ ල තූ තු බු දු


Table: `taml` blue zones

  ID    Blue zone                              Characters
  ----  -----------                            ------------
  1     top of letters                         உ ஒ ஓ ற ஈ க ங ச
  2     bottom of letters                      க ச ல ஶ உ ங ட ப


Table: `telu` blue zones

  ID    Blue zone                              Characters
  ----  -----------                            ------------
  1     top                                    ఇ ఌ ఙ ఞ ణ ఱ ౯
  2     bottom                                 అ క చ ర ఽ ౨ ౬


Table: `thai` blue zones

  ID    Blue zone                              Characters
  ----  -----------                            ------------
  1*    top                                    บ เ แ อ ก า
  2     bottom                                 บ ป ษ ฯ อ ย ฮ
  3     ascender                               ป ฝ ฟ
  4     large ascender                         โ ใ ไ
  5     descender                              ฎ ฏ ฤ ฦ
  6     large descender                        ญ ฐ
  7     top of Thai digits                     ๐ ๑ ๓


![This image shows the relevant glyph terms for vertical blue zone
  positions.](img/glyph-terms)


Grid Fitting
------------

Aligning outlines along the grid lines is called *grid fitting*.  It doesn't
necessarily mean that the outlines are positioned *exactly* on the grid,
however, especially if you want a smooth appearance at different sizes.
This is the central routine of the auto-hinter; its actions are highly
dependent on the used writing system.  Currently, only one writing system is
available (latin), providing support for scripts like Latin or Greek.

  * Align edges linked to blue zones.

  * Fit edges to the pixel grid.

  * Align serif edges.

  * Handle remaining 'strong' points.  Such points are not part of an edge
    but are still important for defining the shape.  This roughly
    corresponds to the `IP` TrueType instruction.

  * Everything else (the 'weak' points) is handled with an 'IUP'
    instruction.

The following images illustrate the hinting process, using glyph 'a' from
the freely available font '[Ubuntu Book](http://font.ubuntu.com)'.  The
manual hints were added by [Dalton Maag Ltd], the used application to create
the hinting debug snapshots was [FontForge].

![Before hinting.](img/a-before-hinting.png)

![After hinting, using manual hints.](img/a-after-hinting.png)

![After hinting, using ttfautohint.  Note that the hinting process
  doesn't change horizontal positions.](img/a-after-autohinting.png)


Hint Sets
---------

In ttfautohint terminology, a *hint set* is the *optimal* configuration for
a given PPEM (pixel per EM) value.

In the range given by the `--hinting-range-min` and `--hinting-range-max`
options, ttfautohint creates hint sets for every PPEM value.  For each
glyph, ttfautohint automatically determines whether a new set should be
emitted for a PPEM value if it finds that it differs from a previous one.
For some glyphs it is possible that one set covers, say, the range
8px-1000px, while other glyphs need 10 or more such sets.

In the PPEM range below `--hinting-range-min`, ttfautohint always uses just
one set, in the PPEM range between `--hinting-range-max` and
`--hinting-limit`, it also uses just one set.

One of the hinting configuration parameters is the decision which segments
form an edge.  For example, let us assume that two segments get aligned on a
single horizontal edge at 11px, while two edges are used at 12px.  This
change makes ttfautohint emit a new hint set to accomodate this situation.
The next images illustrate this, using a Cyrillic letter (glyph 'afii10108')
from the 'Ubuntu book' font, processed with ttfautohint.

![Before hinting, size 11px.](img/afii10108-11px-before-hinting.png)

![After hinting, size 11px.  Segments 43-27-28 and 14-15 are aligned on a
  single edge, as are segments 26-0-1 and
  20-21.](img/afii10108-11px-after-hinting.png)

![Before hinting, size 12px.](img/afii10108-12px-before-hinting.png)

![After hinting, size 12px.  The segments are not aligned.  While
  segments 43-27-28 and 20-21 now have almost the same horizontal position,
  they don't form an edge because the outlines passing through the segments
  point into different directions.](img/afii10108-12px-after-hinting.png)

Obviously, the more hint sets get emitted, the larger the bytecode
ttfautohint adds to the output font.  To find a good value\ *n* for
`--hinting-range-max`, some experimentation is necessary since *n* depends
on the glyph shapes in the input font.  If the value is too low, the hint
set created for the PPEM value\ *n* (this hint set gets used for all larger
PPEM values) might distort the outlines too much in the PPEM range given
by\ *n* and the value set by `--hinting-limit` (at which hinting gets
switched off).  If the value is too high, the font size increases due to
more hint sets without any noticeable hinting effects.

Similar arguments hold for `--hinting-range-min` except that there is no
lower limit at which hinting is switched off.

An example.  Let's assume that we have a hinting range 10\ <= ppem <=\ 100,
and the hinting limit is set to 250.  For a given glyph, ttfautohint finds
out that four hint sets must be computed to exactly cover this hinting
range: 10-15, 16-40, 41-80, and 81-100.  For ppem values below 10ppem, the
hint set covering 10-15ppem is used, for ppem values larger than 100 the
hint set covering 81-100ppem is used.  For ppem values larger than 250, no
hinting gets applied.


Composite Glyphs
----------------

The ttfautohint library (and programs) supports two solutions for handling
composite glyphs, to be controlled with option
[`--composites`](#hint-composites).  This section contains some general
information, then covers the case where the option is off, while the next
section describes how ttfautohint behaves if this option is activated.

Regardless of the `--composites` option, ttfautohint performs a scan over
all composite glyphs to assure that components of a composite glyph inherit
its style, as described [later](#opentype-features).  However, components
that are shifted vertically will be skipped.  For example, if the glyph
'Agrave' uses a shifted 'grave' accent glyph, the accent is ignored.  On the
other hand, if there is a glyph 'agrave' that uses the same 'grave' glyph
vertically unshifted, 'grave' does inherit the style.

If `--composites` is off, components are hinted separately, then put
together.  Separate hinting implies that the current style's blue zones are
applied to all subglyphs in its original, unshifted positions.  In case you
want to shift components vertically, it is *mandatory* to set bit\ 2
(value\ 4), `ROUND_XY_TO_GRID`, in the flag variable of the composite glyph
description to get visually pleasing results, as the images below
demonstrate.

![Here, the subscript glyphs are composites each having a single element
  that is shifted down.  If option `--composites` is not used, subglyphs are
  hinted before they are glued together (possibly applying scaling and
  shifting).  Because the `ROUND_XY_TO_GRID` flag isn't set, the vertical
  translation doesn't align the subglyph to the pixel grid, causing severe
  distortions.](img/composite-no-round-xy-to-grid.png)

![The same as before, but with `ROUND_XY_TO_GRID` set.  Now the subscript
  glyphs look identical to the
  superscripts.](img/composite-round-xy-to-grid.png)

![For comparison purposes, here the result *with* option `--composites` (and
  no `ROUND_XY_TO_GRID`).  The composite glyphs as a whole get hinted;
  consequently, the subscript glyphs get separate blue zones.  At the
  displayed size of 16ppem the vertical positions of the subscript blue
  zones are rounded differently if compared to the superscript zones, thus
  the smaller glyph height.](img/composite-no-round-xy-to-grid-option-c.png)


The '\.ttfautohint' Glyph
-------------------------

If option [`--composites`](#hint-composites) is used, ttfautohint doesn't
hint subglyphs of composite glyphs separately.  Instead, it hints the whole
glyph, this is, composites get recursively expanded internally so that they
form simple glyphs, then hints are applied -- this is the normal working
mode of FreeType's auto-hinter.

One problem, however, must be solved: Hinting for subglyphs (which usually
are used as normal glyphs also) must be deactivated so that nothing but the
final bytecode of the composite gets executed.

The trick used by ttfautohint is to prepend a composite element called
'\.ttfautohint', a dummy glyph with a single point, and which has a single
job: Its bytecode increases a variable (to be more precise, it is a CVT
register called `cvtl_is_subglyph` in the source code), indicating that we
are within a composite glyph.  The final bytecode of the composite glyph
eventually decrements this variable again.

As an example, let's consider composite glyph 'Agrave' ('À'), which has the
subglyph 'A' as the base and 'grave' as its accent.  After processing with
ttfautohint it consists of three components: '\.ttfautohint', 'A', and
'grave' (in this order).

  Bytecode of    Action
  -------------  --------
  .ttfautohint   increase `cvtl_is_subglyph` (now: 1)
  A              do nothing because `cvtl_is_subglyph` > 0
  grave          do nothing because `cvtl_is_subglyph` > 0
  Agrave         decrease `cvtl_is_subglyph` (now: 0)\
                 apply hints because `cvtl_is_subglyph` == 0

Some technical details (which you might skip): All glyph point indices get
adjusted since each '\.ttfautohint' subglyph shifts all following indices by
one.  This must be done for both the bytecode and one subformat of
OpenType's `GPOS` anchor tables.

While this approach works fine on all tested platforms, there is one single
drawback: Direct rendering of the '\.ttfautohint' subglyph (this is,
rendering as a stand-alone glyph) disables proper hinting of all glyphs in
the font!  Under normal circumstances this never happens because
'\.ttfautohint' doesn't have an entry in the font's `cmap` table.  (However,
some test and demo programs like FreeType's `ftview` application or other
glyph viewers that are able to bypass the `cmap` table might be affected.)


Writing Systems
---------------

In FreeType terminology, a writing system is a set of functions that
provides auto-hinting for certain scripts.  Right now, only two writing
systems from FreeType's auto-hinter are available in ttfautohint: 'dummy'
and 'latin'.  The former handles the 'no-script' case; details to 'latin'
follow in the next section.


Scripts
-------

ttfautohint needs to know which script should be used to hint a specific
glyph.  To do so, it checks a glyph's Unicode character code whether it
belongs to a given script.

See '[Character Ranges](#character-ranges)' for a complete list of all
handled scripts and its ranges.  This list is auto-generated from a source
code file, covering the 'latin' writing system.  It also covers some
non-latin scripts (in the Unicode sense) that have similar typographical
properties.

In ttfautohint, scripts are identified by four-character tags (if there are
less characters, spaces are appended).  The value `none` indicates 'no
script'.

Each script is represented by two tables to handle 'base' and 'non-base'
characters.  For ttfautohint, a non-base character is something that should
not be affected by blue zones, regardless of whether this is a spacing or
no-spacing glyph.  In other words, non-base characters are hinted using a
script's default stem width without applying blue zones.

Right now, there are two pseudo-scripts that are used as fallbacks: `latb`
and `latp`, used for latin subscript and superscript characters,
respectively.  Its main usage is support of phonetic alphabets like the IPA,
which intermix those characters with normal characters sitting on the
baseline, and which are not specially handled in corresponding OpenType
features like `sups`.

If a glyph's character code is not covered by a script range, it is handled
by a *fallback script*.  By default, the fallback script is `none`, which
indicates handling by the 'latin' writing system without applying
script-specific blue zones (but aligning stems to the grid if possible).
The fallback script can be changed; see option
[`--fallback-script`](#fallback-script).

The user can also select whether uncovered glyphs are either hinted (which
is the default) or scaled only with the fallback script's scaling
parameters.  This can be controlled with option
[`--fallback-scaling`](#fallback-script).  Note that fallback scaling only
makes sense if the fallback script has x\ height blue zones, e.g., `cyrl` or
`latn`.

As a special case, specifying `none` as a fallback script and switching on
fallback scaling ('`-f none -S`'), no hinting is applied at all to uncovered
glyphs – using `none` always implies a scaling factor of\ 1.


OpenType Features
-----------------

(Please read the [OpenType specification] for details on *features*, `GSUB`,
and `GPOS` tables, and how they relate to scripts.)

For modern OpenType fonts, character ranges are not sufficient to handle
scripts.

  * Due to glyph substitution in the font (as specified in a font's `GSUB`
    table), which handles ligatures and similar typographic features, there
    is no longer a one-to-one mapping from an input Unicode character to a
    glyph index.  Some ligatures, like 'fi', actually do have Unicode values
    for historical reasons, but most of them don't.  While it is possible to
    map ligature glyphs into Unicode's Private Use Area (PUA), code values
    from this area are arbitrary by definition and thus unusable for
    ttfautohint.

  * Some features like `sups` (for handling superscript) completely change
    the appearance and even vertical position of the affected glyphs.
    Obviously, the blue zones for 'normal' glyphs no longer fit, thus the
    auto-hinter puts them into a separate group (called *style* in FreeType
    speak), having its own set of blue zones.


Table: OpenType features handled specially by ttfautohint

    Feature tag    Description
  ---------------  -------------
  `c2cp`           petite capitals from capitals
  `c2sc`           small capitals from capitals
  `ordn`           ordinals
  `pcap`           petite capitals
  `sinf`           scientific inferiors
  `smcp`           small capitals
  `subs`           subscript
  `sups`           superscript
  `titl`           titling


There are two conditions to get a valid style for a feature in a given
script.

 1. One of the script's standard characters must be available in the
    feature.

 2. The feature must provide characters to form at least one blue zone; see
    [above](#blue-zones).

An additional complication is that features from the above table might use
data not only from the `GSUB` but also from the `GPOS` table, containing
information for glyph positioning.  For example, the `sups` feature for
superscripts might use the same glyphs as the `subs` feature for subscripts,
simply moved up.  ttfautohint skips such vertically shifted glyphs (except
for accessing standard characters) because glyph positioning happens after
hinting.  Continuing our example, the `sups` feature wouldn't form a style,
contrary to `subs`, which holds the unshifted glyphs.

The remaining OpenType features of a script are not handled specially; the
affected glyphs are simply hinted together with the 'normal' glyphs of the
script.

Note that a font might still contain some features not covered yet: OpenType
has the concept of a *default script*; its data gets used for all scripts
that aren't explicitly handled in a font.  By default, ttfautohint unifies
all affected glyphs from default script features with the `latn` script.
This can be changed with option [`--default-script`](#default-script), if
necessary.


ttfautohint uses the [HarfBuzz] library for handling OpenType features.


SFNT Tables
-----------

ttfautohint touches almost all SFNT tables within a TrueType or OpenType
font.  Note that only OpenType fonts with TrueType outlines are supported.
OpenType fonts with a `CFF` table (this is, with PostScript outlines) won't
work.

  * `glyf`: All hints in the table are replaced with new ones.  If option
    [`--composites`](#hint-composites) is used, one glyph gets added (namely
    the '\.ttfautohint' glyph) and all composites get an additional
    component.

  * `cvt`, `prep`, and `fpgm`: These tables get replaced with data
    necessary for the new hinting bytecode.

  * `gasp`: Set up to always use grayscale rendering, for all sizes, with
    grid-fitting for standard hinting, and symmetric grid-fitting and
    symmetric smoothing for horizontal subpixel hinting (ClearType).

  * `DSIG`: If it exists, it gets replaced with a dummy version.
    ttfautohint can't digitally sign a font; you have to do that afterwards.

  * `name`: The 'version' entries are modified to add information about the
    parameters that have been used for calling ttfautohint.  This can be
    controlled with the [`--no-info`](#ttfautohint-info) option.

  * `GPOS`, `hmtx`, `loca`, `head`, `maxp`, `post`: Updated to fit the
    additional '\.ttfautohint' glyph, the additional subglyphs in
    composites, and the new hinting bytecode.

  * `LTSH`, `hdmx`: Since ttfautohint doesn't do any horizontal hinting,
    those tables are superfluous and thus removed.

  * `VDMX`: Removed, since it depends on the original bytecode, which
    ttfautohint removes.  A font editor might recompute the necessary data
    later on.


Problems
--------

### Interaction With FreeType

Recent versions of FreeType have an experimental extension for handling
subpixel hinting; it is off by default and can be activated by defining the
macro `TT_CONFIG_OPTION_SUBPIXEL_HINTING` at compile time.  This code has
been contributed mainly by [Infinality], being a subset of his original
patch.  Many GNU/Linux distributions activate this code, or provide packages
to activate it.

This extension changes the behaviour of many bytecode instructions to get
better rendering results.  However, not all changes are global; some of them
are specific to certain fonts.  For example, it contains font-specific
improvements for the '[DejaVu] Sans' font family.  The list of affected
fonts is hard-coded; it can be found in FreeType's source code file
`ttsubpix.c`.

If you are going to process such specially-handled fonts with ttfautohint,
serious rendering problems might show up.  Since ttfautohint (intentionally)
doesn't change the font name in the `name` table, the Infinality extension
has no chance to recognize that the hints are different.  All such problems
vanish if the font gets renamed in its `name` table (the name of the font
file itself doesn't matter).

### Incorrect Unicode Character Map

Fonts with an incorrect Unicode `cmap` table will not be properly hinted by
ttfautohint.  Especially older fonts do cheat; for example, there exist
Hebrew fonts that map its glyphs to character codes 'A', 'B', etc., to make
them work with non-localized versions of Windows\ 98, say.

Since ttfautohint needs to find both standard and blue zone characters, it
relies on correct Unicode values.  If you want to handle such fonts, please
fix their `cmap` tables accordingly.

### Irregular Glyph Heights

The central concept of ttfautohint's hinting algorithm, as discussed
[above](#segments-and-edges), is to identify horizontal segments at extremum
positions, especially for blue zones.  If such a segment is missing, it
cannot be associated with a blue zone, possibly leading to irregular heights
for the particular glyph.

Normally, a segment has a horizontal length of at least 20\ font units
(assuming 2048 units per EM)^[To be more precise, the sum of the height and
length of a segment must be at least 20 font units, and the height multiplied
by\ 14 must not exceed the length.  Thus (19,1) is also a valid minimum
(length,height) pair, while (18,2) isn't.  The value\ 20 is heuristic and
hard-coded, as is the value\ 14 (corresponding to a slope of approx.
4.1°).].  Using a [Control Instructions File](#control-instructions-file),
however, it is possible to define additional segments at arbitrary points
that help overcome this restriction, making it possible to fix (most of)
such problems.

### Diagonals

ttfautohint doesn't handle diagonal lines specially.  For thin outlines,
this might lead to strokes that look too thick at smaller sizes.  A font
designer might compensate this to a certain amount by slightly reducing the
stroke width of diagonal lines.  However, in many cases the sub-optimal
appearance of a stroke with borders that don't exactly fit the pixel grid is
not the outline itself but an incorrect gamma value of the monitor: People
tend to not properly adjust it, and the default values of most operating
systems are too low, causing too much darkening of such strokes.  It is thus
of vital importance to compare ttfautohint's results with similar fonts to
exclude any systematic effect not related to the outlines themselves.



Control Instructions
====================

An entry in a control instructions file has various syntax forms, which are
discussed here.  Brackets indicate optional elements.


Common Syntax Elements
----------------------

*font‑idx* gives the index of the font in a TrueType Collection, starting
with value\ 0.  If missing, it is set to zero.  For normal TrueType fonts,
only value zero is valid.  A font index can be specified in decimal, octal,
or hexadecimal format, the latter two indicated by the prefixes `0` and
`0x`, respectively.

*glyph‑id* is either a glyph's name as listed in the `post` SFNT table or a
glyph index.  A glyph name consists of characters from the set
'`A-Za-z0-9._`' only and does not start with a digit or period, with the
exceptions of the names '`.notdef`' and '`.null`'.  A glyph index starts
with value\ 0 can be specified in decimal, octal, or hexadecimal format, the
latter two indicated by the prefixes `0` and `0x`, respectively.  Glyph
names are internally converted to glyph indices.

*points* are number ranges, see '[x Height Snapping
Exceptions](#x-height-snapping-exceptions)' for the syntax.

Similar to the Bourne shell (`sh` or `bash`), a comment starts with
character '`#`'; the rest of the line is ignored.  An empty line is ignored
also.  Both the newline character and '`;`' can be used as a separator
between exception entries.  A trailing '`\`' at the end of a line continues
the current line on the next one.

A control instructions file is parsed line by line; later entries override
earlier entries (in case there is something to override).


Style Adjustments
-----------------

This syntax form makes it possible to override the style assignment
algorithm of ttfautohint; see '[Scripts](#scripts)' and '[OpenType
Features](#opentype-features)' for more details.

> *\[*\ font-idx\ *\]*\ \ script\ \ feature\ \ *`@`*\ \ glyph-ids

*script* is a four-letter name of one of the scripts supported by
ttfautohint.  *feature* is one of the four-letter names of features
supported by ttfautohint.

The elements of *glyph-ids* are a list of comma separated *glyph-id* values
or value ranges.  Note that is not necessary that elements are specified in
increasing order.

Assuming that a font contains superscript digits 'zero.sups' to 'nine.sups'
together with the glyphs 'a.sups' and 'o.sups', use a line

```
    cyrl sups @ zero.sups-nine.sups, a.sups, o.sups
```

to add those glyphs to the style handling Cyrillic superscript glyphs.
However, it is still necessary that the selected script contains proper
[Blue Zone characters](#blue-zones), otherwise those glyphs aren't handled
at all.

Use the `--debug` command line option to see how ttfautohint assigns glyph
indices of a font to styles.


Glyph Adjustments
-----------------

The following syntax forms allows adjustments of a glyph's hinting process.

### Change Direction of Points, Artificial Segments

> *\[*\ font‑idx\ *\]*\ \ glyph‑id\ \ *`l`\[`eft`\]|`r`\[`ight`\]*\ \ points\ \ *\[*\ *`(`*\ left‑offset\ *`,`*\ right‑offset\ *`)`*\ *\]*\

The mutually exclusive parameters `left` and `right` (which can be
abbreviated as '`l`' and '`r`', respectively) indicate that the following
points have left or right 'out' direction, respectively, overriding
ttfautohint's algorithm for setting point directions.  The 'out direction'
of a point is the direction of the outline *leaving* the point (or passing
the control point).  If the specified direction is identical to what
ttfautohint computes, nothing special happens.  Otherwise, a one-point
segment with the specified direction gets created, see
[above](#segments-and-edges).  By default, its length is zero.  Setting
*left‑offset* and *right‑offset*, you can change the segment's horizontal
start and end position relative to the point position.  *left‑offset* and
*right‑offset* are integers measured in font units.

The following five images, displaying glyphs 'O' and 'Q' from the font
[Halant-Regular](http://www.google.com/fonts/specimen/Halant), demonstrate
how to use direction changes.

![The outlines of glyphs 'O' and 'Q', as displayed in FontForge.  They are
  sufficiently similar to expect that ttfautohint hints them equally.
  However, this is not the case.](img/Halant-Regular-O-Q.png)

![The same glyphs, shown at 12px before hinting.  [Please ignore the outline
  distortion in the upper right of glyph 'O'; this is a bug in FontForge
  while running the TrueType
  debugger.]](img/Halant-Regular-O-Q-unhinted-12px.png)

![Using only ttfautohint's '`-w gGD`' parameter to force strong stem width
  and positioning, the hinting of glyph 'Q' is really bad, making the glyph
  vertically two pixels larger!  Reason is that this glyph doesn't contain a
  horizontal segment at the baseline blue zone (*y*\ =\ 1; this corresponds
  to the segment 13-14 in the 'O' glyph).  Normally, segment 1-2 would form
  a 'stem' with the baseline segment (as segment 7-8 does in glyph 'O').
  Instead, it forms a stem with segment 19-20, which gets moved down
  (*y*\ =\ −1) because the whole glyph appears to be
  stretched.](img/Halant-Regular-O-good-Q-badly-hinted-12px.png)

![To fix the problem, we change the direction of point\ 38 to 'left' by
  writing a line '`Q left 38`' (without the quotes) to a control description
  file `Halant-Regular.txt`.  Adding option '`-m Halant-Regular.txt`' to
  ttfautohint, we get the shown image as a result, which is much better:
  Segment 1-2 now properly forms a stem with our artificial one-point
  segment\ 38, and the 'O'-like shape is properly positioned.  However,
  there is still room for improvement: Segment 19-20 is also positioned at
  the baseline, making the connection between the 'O' shape and the tail too
  thin.](img/Halant-Regular-O-good-Q-better-hinted-12px.png)

![By giving the one-point segment\ 38 a horizontal width, we can prevent
  that segment 19-20 gets positioned at the baseline: Replace the line in
  the previous image description with '`Q left 38 (−70,20)`', making the
  segment extend 70 font units to the left and 20 to the right of point\ 38.
  The exact offset values don't matter; it's only important to start left of
  point\ 19.  Another solution to the problem is to artificially change the
  direction of segment 19-20 by adding a second line '`Q right 19-20`' to
  the control instructions file; for our 'Q' glyph, this produces almost
  exactly the same hinting results.  Note that such direction changes only
  influence the hinting process; an outline's direction won't be changed at
  all.](img/Halant-Regular-O-good-Q-well-hinted-12px.png)

### Unset Direction of Points

> *\[*\ font‑idx\ *\]*\ \ glyph‑id\ \ *`n`\[`odir`\]*\ \ points\

Parameter `nodir` (or '`n`') sets the 'out' direction of the following
points to 'no direction', this is, neither left nor right.  If the specified
direction is identical to what ttfautohint computes, nothing special
happens.  Otherwise, ttfautohint no longer considers those points as part of
horizontal segments, thus treating them as ['weak'](#grid-fitting) points.

Modifying or adding segments doesn't directly modify the outlines; it only
influences the hinting process.

### Delta Exceptions

> *\[*\ font‑idx\ *\]*\ \ glyph‑id\ \ *`t`\[`ouch`\]|`p`\[`oint`\]*\ \ points\ \ *\[*\ *`x`\[`shift`\]*\ x‑shift\ *\]*\ \ *\[*\ *`y`\[`shift`\]*\ y‑shift\ *\]*\ \ *`@`*\ \ ppems\

The mutually exclusive parameters `touch` and `point` (which can be
abbreviated as '`t`' and '`p`', respectively) make ttfautohint apply delta
exceptions for the given points, shifting them by the given values.  Delta
exceptions entered with `touch` are applied before the final 'IUP'
(*interpolate untouched points*) instructions in a glyph's bytecode,
exceptions entered with `point` after 'IUP' (please consult Greg Hitchcock's
[ClearType Whitepaper] for more on pre-IUP and post-IUP delta hints).
Additionally, the `touch` parameter makes the bytecode *touch* the affected
points; such points are no longer affected by 'IUP' at all.  Note that in
ClearType mode all deltas along the x\ axis are discarded, and deltas along
the y\ axis are only executed for touched points.  As a consequence,
vertical delta exceptions entered with `point` should not be used in
ClearType mode.^[Unfortunately, there is a bug in FreeType prior to version
2.5.4 (released in December 2014) that completely disables vertical delta
exceptions if subpixel hinting is activated.  For this reason you should
expect that the `touch` parameter fails on older GNU/Linux distributions.]

*ppems*, similar to *points*, are number ranges, see '[x Height Snapping
Exceptions](#x-height-snapping-exceptions)' for the syntax.

*x‑shift* and *y‑shift* represent real numbers that get rounded to multiples
of 1/8 pixels.  The entries for `xshift` ('`x`') and `yshift` ('`y`') are
optional; if missing, the corresponding value is set to zero.  If both
values are zero, the delta exception entry is ignored as a whole.

Values for *x‑shift* and *y‑shift* must be in the range [−1.0;1.0].  Values
for *ppems* must be in the range [6;53].  Values for *points* are limited by
the number of points in the glyph.

Note that only character '`.`' is recognized as a decimal point, and a
thousands separator is not accepted.

As an example for delta instructions, let's assume that you want to shift
points 2, 3, and\ 4 in glyph 'Aacute' at ppem sizes 12 and\ 13 by a vertical
amount of 0.25 pixels.  This corresponds to the line

```
    Aacute  touch 2-4  yshift 0.25  @ 12, 13
```

in a control instructions file.  Since we use `touch` and not `point`,
points 2, 3, and\ 4 are no longer subject to the final 'IUP' instruction,
which interpolates weak, untouched point positions between strong, touched
ones, cf.  the description
[here](https://developer.apple.com/fonts/TrueType-Reference-Manual/RM05/Chap5.html#IUP).