This is a set of tools to simplify creation of htf fonts for tex4ht.
cd `kpsewhich -var-value TEXMFHOME`
mkdir -p tex/latex
cd tex/latex
git clone https://github.com/michal-h21/htfgen.git
cd htfgen
tex makeenc.tex
chmod +x lstexenc
chmod +x tfmtochars
chmod +x htfgen
chmod +x dvitohtf
chmod +x scanfdfile
chmod +x make-t1-htf
# Should almost always be set, but you never know.
[ -z "$PWD" ] && pwd="$(pwd)"
ln -s $PWD/lstexenc /usr/local/bin/lstexenc
ln -s $PWD/tfmtochars /usr/local/bin/tfmtochars
ln -s $PWD/htfgen /usr/local/bin/htfgen
ln -s $PWD/dvitohtf /usr/local/bin/dvitohtf
ln -s $PWD/scanfdfile /usr/local/bin/scanfdfile
ln -s $PWD/make-t1-htf /usr/local/bin/make-t1-htf
mktexlsr # this is optional, only if your distro cannot find installed Lua filesObjective of Htfgen is to automatize the creation of the HTF font mapping files. These files are used by tex4ht to map character codes in the DVI files to Unicode.
There are two main scripts: scanfdfile and dvitohtf. The first one searches for
declared fonts in the FD files, the other generates literate TeX file for the HTF
generation. Sample usage is as follows:
cat /usr/local/texlive/2021/texmf-dist/tex/latex/ebgaramond/*.fd | scanfdfile | dvitohtf > ebgaramont-htf.tex
etex ebgaramont-htf.tex
This will create HTF files for all detected fonts defined in FD files for EB Garamond.
dvitohtf can also generate HTF files for missing fonts in the DVI file.
So if tex4ht reports missing HTF files, it can be used directly on the DVI
file:
dvitohtf sample.dvi > missing.tex etex missing.tex
dvitohtf supports both virtual and tfm fonts. It looks for virtual fonts
first, the tfm file is used only when no vf is found. It looks for all
fonts referenced in the virtual font and tries to look for corresponding .enc
files in pdftex.map. The .enc files contain glyph lists, which are then mapped
to Unicode.
It also parses the .pfb file for font family name and tries to detect style
(italic, bold, small caps) from the font full name saved in the .pfb file.
It computes hashes for the font tables, so duplicate font tables aren't written, the fonts with same characters just link to the first used font.
If no .enc file is found, then the font cannot be supported. There can be
also missing mappings between glyphs and Unicode. The missing mappings are
reported in the TeX file. Htfgen contains large mapping files, but some fonts
just use some custom glyphs which doesn't have Unicode equivalent. For example
Q_u ligatures etc. In this case the mapping must be added by hand to
glyphlists/glyphlist-fixes.txt.
It works reasonably well for fonts generated by Fontinst, because they usually
use standard glyph names, contains .enc files, etc. For complex virtual fonts,
especially math, it fails. HTF files for such fonts still needs to be created
by hand.
Use the listfonts.lua script to find .fd files required by a LaTeX package like times or bookman
texlua listfonts.lua times avant bookman | xargs cat | scanfdfile | dvitohtf
It is possible to update the literate source file using the
update_literate_sources.lua script. It takes list of .htf file as an
argument and reads the source from standard input.
texlua update_literate_sources.lua `ls *.htf` < tex4ht-fonts-noncjk.tex > tex4ht-fonts-noncjk-updated.tex
It is useful for updating already existing fonts in the sources.
If all what you need is to add support for normal text font encoded in T1
fontenc, you may try to use the make-t1-htf script.
Lets say that you have following file:
\documentclass{article}
\usepackage[T1]{fontenc}
\usepackage[utf8]{inputenc}
%\usepackage{dejavu}
\renewcommand{\familydefault}{\sfdefault}
\usepackage[defaultsans]{droidsans}
\newcommand\sample{Příliš žluťoučký kůň úpěl ďábelské ódy. \% ``Nazdar světě''}
\begin{document}
\sample
\textit{\sample}
\textbf{\sample}
\textsc{\sample}
\texttt{\sample}
\end{document}
Then you will get lot of warnings in the terminal output about missing .htf
files:
--- warning --- Couldn't find font `DroidSans-t1.htf' (char codes: 0--255)
because T1 encoded fonts should produce identical output, we don't need to
create full .htf file with char table, but we can create file which points to
some existing full .htf file in T1 encoding. This file has such form:
.lm-ec
dot on the first line indicated that this points to file lm-ec.htf, which is
.htf file for Latin Modern in T1 encoding.
We can add also some CSS information to the file, with this instruction:
.lm-ec
htfcss: DroidSans-Bold-t1 font-weight:bold; font-family: 'Droid Sans', sans-serif
first entry after htfcss is the font name, which is also used as CSS class.
Then CSS instructions follows.
To simplify creation of this light .htf file, make-t1-htf script is
provided. The usage is following:
usage: maket1htf fontname [style] [family]
available styles are rm for normal text, it -- italic, b -- bold and sc -- small-caps
The sample .htf file was created with command:
$ make-t1-htf DroidSans-Bold-t1 b "'Droid Sans', sans-serif"
First, we need to figure out, whether tested font is virtual. Virtual fonts aren't supported at the moment.
lstexenc fontname
for tfm font, we get something like:
$ lstexenc eccc1000
eccc1000 tfm EXTENDED TEX FONT ENCODING - LATIN
for virtual font, the outpus is like this:
$ lstexenc ntxmia
ntxmia vf FONTSPECIFIC
txmia tfm FONTSPECIFIC
txsyc tfm FONTSPECIFIC
txr vf TEX TEXT
rtxptmr tfm TEXBASE1ENCODING
rtxr tfm FONTSPECIFIC
ntxexb tfm UNSPECIFIED
rtxmio tfm FONTSPECIFIC
zxlr-8r tfm TEXBASE1ENCODING
ptmr8r tfm TEXBASE1ENCODING
zxxrl7z tfm ADOBESTANDARDENCODING
first entry on each line is a font name, second font type and third is font encoding. Font encoding information provided by fonts is unreliable!
If font type is tfm, we can process to another step, which is translation of
8-bit characters to unicode characters using glyph info encoded in enc or
afm files.
tfmtochars script is used to generate tsv file with char number, glyph name
and hex unicode value.
$ tfmtochars fontname [enc]
tfmtochars tries to find enc file in map files provided by pdftex, you
need to provide the name only in the case the enc file is not configured for
given font.
$ tfmtochars eccc1000
0 grave 0060
1 acute 00B4
2 circumflex 02C6
...
253 yacute 00FD
254 thorn 00FE
255 germandbls 00DF
for some fonts, enc file can't be found:
$ tfmtochars zxxrl7z
Cannot load enc file for txmia
You may try to find a enc file by hand (Google?). Common enc names can be
found in fontname guide.
From lstexenc zxxrl7z we know that this font uses ADOBESTANDARDENCODING, which is coded as 8a:
$ tfmtochars zxxrl7z 8a
we should control, whether the output is correct (nil values instead
of hexadecimal codes are suspicious). We may use
$ tfmtochars zxxrl7z 8a | grep nil
to test that. If the output is correct, we need to save the output to the
tsv file, with command:
$ tfmtochars zxxrl7z 8a > zxxrl7z.zsv
now we can generate the htf file with htfgen
$ htfgen zxxrl7z.tsv > zxxrl7z.htf
To test the htf file, you can use testfont.lua script:
$ texlua testfont.lua zxxrl7z
and compare the resulting pdf and html files. Note that occasionally some
characters in either pdf or html files are missing, I am not sure why.
Math fonts with calligraphic features such as fraktur or scripts doesn't
reflect these styles in glyph names, so it is impossible to automatically
reflect these features. See unicode math symbols
for unicode values of these features. You may need to correct unicode hex codes
in the tsv file by hand. Any ideas how to solve that automatically will be
really appreciated.
We may parse afm files for glyph names, when no encoding is found or provided.
Not all fonts does have afm files, though.
My idea is following:
- first we need to make
tsvfiles for all referenced fonts - then make new
tsvwith referenced characters from particulartsvfiles - sometimes character in the virtual font is composed from more characters from one or more referenced fonts. It is unlikely that this can be solved automatically