unifont-intro.txt

INTRODUCTION
------------
This document describes the process of using the GNU Unifont utilities
to create a font.  The steps described in "Using Graphical Tools" are
more or less the steps that I (Paul Hardy) followed to add thousands
of glyphs to the GNU Unifont, except that I didn't have the luxury of
just typing "make" to make a new font while adding those glyphs.

I streamlined the font build process after I was done drawing the
Unicode 5.1 glyphs.

I know that plain ASCII text is *so* last millenium, especially for
a package related to Unicode.  Yet ASCII can be read with anything;
hence this format.

DISCLAIMER: Donald Knuth warned in his Metafont book that if someone
started designing type, they would never again be able to look at
a page of text normally and just read its content.  There is a
point of no return beyond which a serious font designer begins
looking at how individual letters in a font on a page are drawn,
and how they might be improved.  Be warned!


-- Paul Hardy <unifoundry@unifoundry.com>  July 2008


UNICODE
-------
Unicode is an international standard to encode all the world's
scripts with one universal scheme.  Unicode is the default encoding
for web pages and is gaining popularity in many other applications.
To learn more about Unicode, look at code charts, and see the
latest developments, check out

     http://unicode.org

The GNU Unifont follows the Unicode encoding scheme.  Unicode
defines the numeric value of a character, but does not define
one particular font.  There can be (and are) many fonts that
support a subset of Unifont characters.

In 1998, Roman Czyborra observed that there was still no font,
free or commercial, with complete Unicode coverage.  He envisioned
a low-quality bitmapped font as an easy way to produce a font
that covered much of the Unifont standard.


GNU UNIFONT STRUCTURE
---------------------
The GNU Unifont is a dual-width pixel font.  Roman Czyborra
began this font in 1998 with a goal of having one glyph
rendered for each visible character in the Unicode Basic
Multilingual Plane (Plane 0, the first 65,536 characters).
His original writing on this is at http://czyborra.com/unifont/,
but as of this writing (July 2008) the website has been down
for a while.

(Note that the term "character" is used very loosely here for
simplicity; the Unicode Standard has a stricter definition
of what constitutes a character.)

The font is dual-width.  Each character is 16 pixels tall, and
either 8 or 16 pixels wide.  The characters are stored in a
unique .hex file format invented by Roman Czyborra as a convenient
way of giving each character exactly a one line specification.


HEX FILE FORMAT
---------------
By convention, files containing the GNU Unifont native font format
have the extension ".hex".  Their format is extremely simple, consisting
of two fields separated with a colon (":") and ending with a newline.

The first field is the Unicode code point, in hexadecimal.  This is
a four digit hexadecimal number for all Plane 0 fonts.  Hexadecimal
digits are (in order) 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, A, B, C, D, E,
and F.  The Unicode Standard uses a hexadecimal number to assign
each character a location.  These locations are called "code points"
and their range is 0 through 10FFFF, inclusive.

The range 0 through FFFF, inclusive, is called the Basic Multilingual
Plane (BMP), or Plane 0.  This plane contains most characters used by
all modern scripts of the world.

Because the original goal of the GNU Unifont was to cover the entire
Unicode Basic Multilingual Plane, a four digit hexadecimal number
suffices.  In the future, I plan to allow the utilities to go beyond
the BMP now that coverage of the Unicode BMP is finally complete.

The C programs allow up to an eight digit hexadecimal code point,
but not all the utilities have been thoroughly tested beyond the BMP.

The second field is a string of hexadecimal digits.  There are 32
digits for a character that is 8 pixels wide, and 64 digits for a
character that is 16 pixels wide.

The good news is you don't have to worry about these long digit
strings.  Roman Czyborra wrote a utility, "hexdraw", to convert
.hex fonts to a form that can be edited with a plain text editor,
then converted back into .hex format.

Paul Hardy wrote two utilities to do the same thing except with
bitmapped graphics images for editing with a graphics editor:
"hex2bmp" converts a block of 256 characters into a graphics
file, and "bmp2hex" converts such a graphics file back into
.hex format.  These bitmaps display the 256 characters in a block
arranged in a 16 by 16 character grid.


USING GRAPHICAL TOOLS
---------------------
Let's use an example.  Suppose you want to modify the Coptic letters
in the range U+2C80..U+2CFF ("U+" is Unicode shorthand).  These
letters are in the upper half of the block U+2C00..U+2CFF.  The
Unicode utilities in this package refer to this as "page" 2C.
("Page" is not a Unicode term -- it is just a term unique to this
package to refer to a block of 256 code points/characters).

So the tasks will be:

     1) Convert .hex version of the page 2C range as a 16 by 16
        bitmapped grid.
     2) Modify the bitmap in any graphics editor, being careful
        to re-save it as a Windows Bitmap (.bmp) or Wireless
        Bitmap file when finished.
     3) Convert the modified bitmap back into a .hex font file.
     4) Merge the results with the original unifont.hex file.
     5) Run "unidup" on the resulting file to guard against
        duplicate character definitions.
     6) Create the new bitmapped version of the font.
     7) Check the compiled font for duplicates.
     8) If there are duplicates, remove them and go back to
        Step 5.
     9) Create the new TrueType version or other versions of the font.

Step 1: Convert the .hex range into a bitmap grid.
        Assuming our font file is named "unifont.hex", type

             unihex2bmp -p2C -f <unifont.hex >uni2C.bmp

        Using the "-f" flag will flip the grid so the glyphs
        appear ordered the way they are in the Unicode Standard
        and all Unicode code charts, running from top to bottom,
        left to right.  (If this isn't clear, create two bitmaps,
        one created with the "-f" option and one without and
        compare the results.)  In the future this "flipped" order
        might become the normal order, but the software has been
        well tested the way it is now and a change would require
        more testing.

Step 2: Modify uni2C.bmp with your favorite graphics editor.

Step 3: Convert the edited .bmp file back into .hex format:

             unibmp2hex <uni2C.bmp >uni2C.hex

        Note that "unibmp2hex" will automagically detect whether
        or not a bitmap is flipped, so there is no need for a
        "-f" flag.

Step 4: Merge the results with the original "unifont.hex" file.
        This requires several sub-steps:

        - Edit the original "unifont.hex" file and delete the
          lines that begin with "2C".
        - Insert the "uni2C.hex" file into "unifont.hex", either
          with a text editor such as vi or emacs, or with a
          Unix command such as:

               cat uni2C.hex unifont.hex | sort >new-unifont.hex

          This second option (using "sort") is preferred, because
          "unidup" (in Step 5) might miss duplicate code points
          if your final result isn't in proper order.

Step 5: Make sure there are no duplicates with "unidup":

             unidup < unifont.hex

        or

             unidup < new-unifont.hex

        depending on the name of your final font file.  If there
        is no output, your modified font contains no duplicates.

        This editing is best done on an input .hex file, such as
        "rc-base.hex" (Roman Czyborra's base glyphs).  If this
        editing is done on a final file (one of the .hex files
        in the font/compiled directory), the checkered boxes
        representing unassigned code points must be erased before
        converting the .bmp file back into .hex.  Otherwise, 
        duplicates will be created in the new compiled font
        that will be created in the font/compiled directory.

Step 6: Create the new bitmapped version of the font.  In the
        font/ directory, type

             make hex

Step 7: Check the compiled font for duplicates.  Change to the
        font/compiled directory and run

             unidup < mynewfontfile.hex

        for whatever font file you created.

Step 8: If there are duplicates, remove them in the font/ directory
        and go back to Step 5.

Step 9: Create the new TrueType version of the font and all other
        bitmapped versions.  From the font/ directory, type

             make distclean && make

        Then be prepared to wait a loooong time.  On a modern Pentium
        processor, this can take an hour, consume at least 256 Megabytes
        of virtual memory (RAM), and consume just under 250 Megabytes
        of disk space, so plan accordingly.

        To only create a BDF font, in the font/ directory just type

             make bdf

        To only create a BDF and PCF font, in the font/ directory type

             make pcf

        Creating a BDF font is the first step in creating a PCF font.
        BDF fonts can be created just with the tools in this package.
        PCF fonts are created by running "bdftopcf" on the BDF font.
        TrueType fonts require FontForge.


USING HEXDRAW
-------------
Roman Czyborra's original utility to edit glyphs is the "hexdraw"
Perl script.  Using the same script as above, Coptic, here are the
steps for modifying "unifont.hex" using "hexdraw".

First, realize that the GNU Unifont has tens of thousands of glyphs
(characters, using the term character loosely).  In this example,
out of the tens of thousands of glyphs, we want to modify the range
U+2C80..U+2CFF (only 128 glyphs).

The range U+2C80..U+2CFF could be extracted from unifont.hex by
using the "grep" utility to look for lines beginning with "2C8"
through "2CF", or that range could be isolated by copying "unifont.hex"
into another file, and deleting all lines except the desired range.

The following steps will probably minimize typographical errors,
but they aren't the only way.

     1) "grep" the desired block of 256 glyphs and convert them
        into a text representation for editing.
     2) Edit the block with a text editor, such as vi or emacs.
     3) Convert the edited text file back into .hex format.
     4) Delete the edited block range from the original font file.
     5) Merge the two .hex files into one file.
     6) Check for duplicates with "unidup".
     7) Generate new fonts as described in the Using Graphical
        Tools section.

Step 1: Extract the desired block with grep:

             grep "^2C" unifont.hex | hexdraw > uni2C.txt

Step 2: Edit "uni2C.txt" with a text editor.

Step 3: Convert the text file back into .hex format:

             hexdraw < uni2C.txt > uni2C.hex

Step 4: Delete the lines in the original "unifont.hex" file that
        begin with "2C".

Step 5: Merge the two files:

             cat unifont.hex uni2C.hex | sort > new-unifont.hex

        or use Roman's "hexmerge" utility:

             hexmerge unifont.hex uni2C.hex > new-unifont.hex

Step 6: Check for duplicates:

             unidup < new-unifont.hex

         Of course, remove any reported duplicates.

Step 7: Build the font as in the Using Graphical Tools section.
        This can be as simple as typing

             make

         in the font/ directory.

I (Paul Hardy) had only used hexdraw in the very beginning of my
work on the GNU Unifont.  Once I got my graphics programs working,
I ignored it for months.  Then I wanted to go through all of the
Yi Syllables and Yi Radicals -- over 1000 characters -- to fine-tune
their horizontal alignment after I drew them.  hexdraw turned out
to be the perfect tool for this.  By printing hyphens ("-") as
place holders where a pixel is off, it allowed me to verify space
to the left and right of each character.


CHECKING COVERAGE
-----------------
There should never be duplicates in a .hex file.  If there are, remove
them before the .hex font is turned into a .bdf or other font.  The
notes above include making liberal use of "unidup" to avoid such a
situation.

The "unipagecount" program will print a hexadecimal number of code
points that have coverage within each 256 code point block.  The
hexadecimal number will therefore range from 0 (no coverage) to
100 (= 256 decimal; full coverage).  If a number is ever more than
100 hexadecimal, there's an extra character (glyph) for that page.

To further look at the coverage within just one 256 code point
page (using page 2C, with Coptic, as our example) use

     unipagecount -p2C < unifont.hex

Note that the "page number" can use upper- or lower-case letters:
"-p2C" or "-p2c" will both work.  That will print a 16 x 16 grid
of U+2C00..U+2CFF.

Using the "-l" flag with "unipagecount" will produce an HTML
table with links to corresponding graphics images (see how
this works in the "./font/compiled" directory after running
"make").  The background color of the cells will range from
red (for 0% complete in that 256 code point block) to orange
(a little coverage) to yellow (more coverage) to green
(complete coverage).  If a cell looks light red or pink,
the corresponding code page might accidentally have duplicate
characters.  Verify that with "unidup".

To see the coverage of each Unicode script, copy the file (from
the top level directory) "./dat/coverage.dat" into the same
directory as the "unifont.hex" file you're examining.  Then run

     unicoverage < unifont.hex > coverage.out

This will give you all the scripts within the Unicode Basic
Multilingual Plane, in order, with a percentage (0.0% through
100.0%) of each script's coverage.


INSTALLING FONTS ON UNIX/LINUX
------------------------------
Compress PCF fonts using "gzip -9 fontname.pcf".  Copy the resulting
"fontname.pcf.gz" file to /usr/share/fonts/X11/misc/ or place in a
local font directory if your windowing software supports that (for
example, ~/.fonts/).

Copy TrueType fonts to /usr/share/fonts/truetype uncompressed or place
in your local font directory.  Note: on some versions of Unix, such as
Solaris, the name of the TrueType directory might be TrueType and
might be under /usr/share/fonts/X11 -- examine the system fonts
directories, then modify the font "make install" rule accordingly.

On most flavors of Linux with the latest "xset" utility (including
the latest Debian and Red Hat), the following command should allow
you to start using the font immediately:

     xset fp rehash

The safest way to make sure the system knows about the new fonts will
be to restart the X Window System or even reboot.


INSTALLING FONTS ON MICROSOFT WINDOWS
-------------------------------------
Copy the TrueType (*.ttf) font into your C:\WINDOWS\Fonts folder.


INSTALLING FONTS ON MACS
------------------------
Copy the TrueType (*.ttf) font into the fonts folder in the Finder
folder.


CREATING A BRAND NEW FONT
-------------------------
These tools will only produce glyphs that are 16 pixels tall, and
8 or 16 pixels wide.  This current release is furthermore only fully
tested over the Unicode Basic Multilingual Plane (code points from
0000 through FFFF, with none above that range -- this will change
in the future).  The tools can be used to create and manipulate
any bitmapped font within those constraints.

To create a brand new font (or even to add a new script to the GNU
Unifont in the future), plan out the basic dimensions of the characters:

     - How far above the lowest pixel will the baseline appear (in
       other words, how many rows are necessary for descenders)?
     - How many pixels must be left on top and bottom for accents
       (in other words, what will the height of capital letters be)?
     - Must glyphs be centered, left-aligned, or right-aligned?
     - For a Latin font, what will the "x-height" be (the height
       of a lower-case "x" and related letters, such as "n" and "r")?

Look over the entire script and plan out a grid that is consistent
for the entire script.

GNU Unifont characters for the most part leave the right-most column
of pixels blank if possible for right-to-left scripts.  Centering is
done around the fourth pixel (if a glyph is eight pixels wide) or
around the eighth pixel (if a glyph is 16 pixels wide).

Consistent cap heights, x-heights, descender depths, and centering
will produce a better looking font.

Experimenting and (above all) having fun with these utilities is
the best way to learn.


UPDATES TO UNICODE
------------------
The Unicode Standard is under constant update.  Unicode 5.1,
the basis for this release of the GNU Unifont package, was
released in April 2008.  At the time of its release, new
character and script additions were already being planned.

Many of these additions are planned for the  Unicode Basic
Multilingual Plane (the range covered by the current release
of the GNU Unifont).

If a currently unused code point is assigned to a character
in the future, the font can be updated as follows:

     1) Delete the code point's entry from "blanks.hex",
        as that code point will no longer have a blank
        filler glyph.
     2) Determine which existing .hex file should have the
        newly defined character:

           - rc-base.hex contains most scripts
           - rc-cjk.hex and wqy-cjk.hex contain most ideographs;
             their ranges are the same
           - rc-hangul.hex contains the Hangul Syllables block

     3) Update the appropriate .hex file.  If the new character
        is in a range covered by rc-cjk.hex and wqy-cjk.hex,
        add the new character to both files.
     4) Consider if "./dat/coverage.dat" has to be updated.
        Follow its existing format to insert a new script,
        being sure to change any previously unassigned ranges
        to reflect the new script.
     5) Make a new .hex version of the font, and verify that
        you didn't introduce any duplicates.
     6) Run "unipagecount" and/or "unicoverage" and verify
        that you have not mistakenly deleted any existing
        characters.

Enjoy!