PLPLOT DOCUMENTATION IN THE DOCBOOK FORMAT
(Notes for contributors/developers)
Introduction
============
You will find in this directory tree the latest PLplot documentation in
DocBook 4.1 XML format. You will find in the current file
(README.developers) instructions for building the documentation from the
DocBook source files in this directory. The documentation that is built
includes man pages of the API as well as complete versions of the
documentation in info, HTML, and PDF forms.
Configuring and Building
========================
If you would like to help with improving the content, you will want to build
the documentation from the DocBook source so you can immediately see and
evaluate the results of your efforts.
Here is how. Simply add the cmake option -DBUILD_DOC=ON to your
normal cmake command-line options for the PLplot build. The resulting
cmake step looks for all the tools required for the documentation
build and generates a WARNING message and turns off all or some
component of the documentation build if any of those required tools
are missing. The tools searched for by our build system are onsgmls
(validation); db2x_texixml, db2x_xsltproc, and makeinfo (info);
XML::Parser and XML::DOM (man); xmlto (html); and dblatex and xelatex
(pdf).
After the cmake step using -DBUILD_DOC=ON you must run either
# Set JOBS to suitable value. 16 is correct for hardware such as the
# Ryzen 7 1700 system with 8 cpu cores and twice that number of
# hardware threads. But parallel builds are unreliable on Cygwin and
# MinGW-w64/MSYS2 as well if/when someone tries to build our DocBook
# documentation on that platform so for those two platforms you should
# always set JOBS=1.
JOBS=16
make -j$JOBS all >& all.out
or
make -j$JOBS build_docbook_documentation >& build_docbook_documentation.out
to build the DocBook documentation. (The former command builds all
aspects of PLplot including the documentation, while the latter
command is quicker because it just builds the documentation and its
PLplot prerequisites.) Furthermore, at build-time xmlto and dblatex
automatically check for required tools, and dblatex checks for missing
fonts when generating the pdf results. (Those required fonts are
chosen by doc/docbook/source/dblatex_stylesheet.xsl and are currently
FreeSans, FreeSerif, and FreeMono).
In sum, cmake WARNING messages with the -DBUILD_DOC=ON option _and_
build-time messages from xmlto and dblatex should be sufficient for you
to figure out what DocBook/XML related tools and fonts you need to
install in order to build the documentation.
N.B. As far as we know, the complete set of required tools is only
available on Linux and Cygwin platforms so you will need access to one of those
platforms to do a documentation build. It is possible as well that the required
tools are available for MinGW-w64/MSYS2, but so far nobody has reported on that
platform. For more information for the Linux and Cygwin cases,
see "Linux Specific" and "Cygwin Specific" below.
Linux Specific
==============
Installing the following packages on Debian should allow you
to build the documentation (this may not be a complete list,
but it should be close to complete since installation
of one Debian package also automatically installs all
Debian packages it depends on).
You can use, e.g., apt-file search bin/xmlto to discover
that
/usr/bin/onsgmls is contained in the "opensp" Debian package;
/usr/bin/db2x_texixml is contained in the "docbook2x" Debian package;
/usr/bin/db2x_xsltproc is contained in the "docbook2x" Debian package;
/usr/bin/makeinfo is contained in the "texinfo" Debian package;
/usr/lib/perl5/XML/Parser.pm (which implements XML::Parser) is contained in the "libxml-parser-perl" Debian package;
/usr/share/perl5/XML/DOM.pm (which implements XML::DOM) is contained in the "libxml-parser-perl" Debian package;
/usr/bin/xmlto is contained in the "xmlto" Debian package;
/usr/bin/dblatex is contained in the "dblatex" Debian package;
/usr/bin/xelatex is contained in the "texlive-xetex" Debian package;
/usr/share/texlive/texmf-dist/fonts/truetype/public/gnu-freefont/FreeSans.ttf is in the "texlive-fonts-extra" Debian package
/usr/share/texlive/texmf-dist/fonts/truetype/public/gnu-freefont/FreeSerif.ttf is in the "texlive-fonts-extra" Debian package
/usr/share/texlive/texmf-dist/fonts/truetype/public/gnu-freefont/FreeMono.ttf is in the "texlive-fonts-extra" Debian package
With these in place the documentation should build with the
appropriate cmake -DBUILD_DOC=ON option set on Debian and other
Debian-based distros such as Ubuntu.
N.B. For what it is worth, the deprecated (because you achieve better
UTF-8 results without this option which is equivalent to setting
-DDOCBOOK_DBLATEX_BACKEND=ON) -DDOCBOOK_DBLATEX_BACKEND=OFF cmake
option still works on Debian. However, there are substantial errors
to start until you address fairly obscure warning messages concerning
PassiveTeX and fop by installing the "xmltex" and "fop" packages. It
is notable that the "fop" package has large numbers of dependencies.
Those are, or course, automatically installed by Debian, but I don't
know the situation for other Linux distros.
PLplot developers have no recent experience with rpm-based Linux
distros, but Orion Poplawski (the Fedora packager for PLplot and a lot
of other free software) has recently had success building the PLplot
documentation on Fedora. We presume he used command-line rpm tools
similar to Debian's apt-file or else used generic rpm-finding software
such as to search rpm packages
for particular filenames similar to the above. Presumably a
similar situation would apply for other rpm-based distros such
as openSUSE.
Cygwin Specific
===============
Installing the following packages on Cygwin will allow you
to build the documentation (note this may include more
packages than strictly required, but this list definitely
works):
openjade, OpenSP, texlive-collection-htmlxml,
texlive-collection-genericrecommended,
texlive-collection-latexrecommended,
texlive-collection-fontsextra, docbook-dsssl, lynx,
ImageMagick, libxml2.
You must then make the gnu FreeFonts available to fontconfig. You can do
this by creating a file /etc/fonts/local.conf which should contain the
following:
/usr/share/texmf-dist/fonts/opentype
/usr/share/texmf-dist/fonts/truetype
/usr/share/texmf-dist/fonts/type1
Note also that if you wish you could copy the FreeFonts from the opentype
subdirectory of the folder listed above to anywhere else (e.g. ~/.fonts or
your Windows fonts directory) and add this to the local.conf file or you can
copy them to /usr/share/fonts/OTF in which case a local.conf file is not
needed.
Finally run fc-cache to update the fontconfig cache. Now the documentation
should build with the apropriate cmake -DBUILD_DOC=ON option set.
The DocBook Back-end Tool Chains
================================
PLplot uses a number of different applications to generate
PLplot documentation in various formats from our source DocBook XML files.
1. man pages.
Our man pages are generated with a configured home-brew perl script
bin/api2man.pl(.in). That script uses XML::DOM::Parser to parse the
combination of plplotdoc.xml (for entity definitions) and api.xml
(the subset of our DocBook source files which describes our core
library's API) to obtain the information used to generate the man
pages.
(Aside. The bin subdirectory also contained at one time api2text.pl which
was used by the non-swig version of the Octave bindings to generate
text documentation for the octave PLplot commands. That script has
now been removed and replaced by the closely related
api2swigdoc.pl which generates bindings/swig-support/swig_documentation.i
API documentation in a form which can be used by swig (for both
the Python and Octave bindings).
2. info pages.
Our info pages are generated by a combination of
bin/info-clean.pl --> db2x_xsltproc --> db2x_texixml --> makeinfo
where info-clean.pl is a home-brew perl script required (as far as I
can tell from reading comments in it) to get around some of the
db2x_xsltproc limitations. db2x_xsltproc and db2x_texixml are from
the docbook2x package.
3. Our HTML results are generated with xmlto and configured at run
time with a CSS stylesheet, see doc/docbook/src/stylesheet.css.xsl.in.
The xmlto application is actually a convenience script that uses
XML/XSL to generate the HTML results as opposed to the previous
SGML/DSSSL approach used to create the HTML results for plplot-5.9.9
and previous.
4. Our PDF results are generated with "dblatex --backend=xetex" using
a combination of XML/XSL and xelatex to generate the PDF results (in
contrast to the SGML/DSSL tools used for that job for plplot-5.9.9 and
previous, and the XML/XSL "xmlto --with-fop" method used with
PLplot-5.9.10 to do that job.
5. Generation of dvi and PostScript forms of our documentation is no longer available.
Validation
==========
Validation checks that the PLplot Docbook files in doc/docbook/src are
well-formed XML and correspond to the DocBook schema. You can perform
such validation by typing the
make validate
command at the top of the build tree. This make target is only available if
the PLplot CMake-based build system has found the onsgmls software
application (which is distributed as part of OpenSP) on your system. This
quick check works regardless of whether you decide to build the
documentation with -DBUILD_DOC=ON or not. Using "make validate" is
especially useful if you are just making a series of simple changes to the
files in doc/docbook/src, and you don't really feel it is necessary to check
every change by doing a complete documentation build.
In addition, the xmlto commands used to build the HTML and print
part of the documentation automatically include a validation step
with xmllint. It turns out that xmllint is more sensitive than onsgmls
to DocBook XML errors. On the other hand, onsgmls handles any
errors it finds with ease while xmllint tends to segfault when there
is a validation error. So use "make validate"
first to detect any obvious validation errors using onsgmls to avoid
situations where xmlto's call to xmllint will segfault.
Testing the documentation that has been built.
==============================================
The following tests are performed in the doc/docbook/src subdirectory _of the
build tree_. The given test commands are for the bash shell, and
plplot_version is an environment variable that has been set
to the current PLplot version. For the man pages, look carefully at
the style of the results. For the rest, look carefully at the style
of the api chapter and the style/results near Table-3.4 in the
advanced chapter. In particular look at how the following examples
render: the overline-underline example just prior to Table-3.4,
Table-3.4 of Greek symbols, Table-3.5 of the "Peace" word expressed in
various languages, and the mathematical symbols occurring in the
paragraph just after Table-3.5.
First, make sure all forms of the DocBook-generated documentation are built using
make -j$JOBS build_docbook_documentation.
Then here are the specific tests for each kind of that generated documentation.
1. man pages.
N.B. nroff is a wrapper for groff which is OK for ascii input but which does not
work correctly (since it doesn't have a -K option which it can pass on to groff)
with input utf8 manpage files so instead of
nroff -man |less
must instead use the following command
groff -K utf8 -T utf8 -mandoc |less
In particular to check out our two man pages which contain any UTF-8
symbols at all (in this case ≤ and ≥) execute the following command if
in this directory (doc/docbook in the build tree):
for MANPAGE in src/plot3dcl.3plplot src/plsurf3dl.3plplot ; do (groff -K utf8 -T utf8 -mandoc $MANPAGE|less) ; done
Those UTF-8 symbols ≤ and ≥ should render without issues (see discussion below), but if you
want to look at all our man pages as well, use
for MANPAGE in src/*.3plplot ; do (groff -K utf8 -T utf8 -mandoc $MANPAGE|less) ; done
So currently our man pages are a case of "so far so good" with regard
to UTF-8 system glyphs since at the present time our UTF-8 text in our
man pages is quite limited. For example, we don't know how well our
man backend and groff will do for exotic system glyphs such as the
many thousands of mathematical symbols normally available, and we also
don't know how that combination of software will render CTL scripts
(if we ever use those in the documentation of our API).
2. info pages.
If in this directory (doc/docbook in the build tree)
info src/plplotdoc.info
If in the top directory of the build tree
info doc/docbook/src/plplotdoc.info
I suggest searching explicitly for "≤", "≥", and "peace" to evaluate
how those first two glyphs, the glyphs mentioned in the "peace"
paragraph, and the additional mathematical glyphs mentioned in the
following paragraph are rendered. For Debian Testing I have found
that "konsole" (the KDE terminal emulation app that uses client-side
TrueType glyphs selected by font-config) renders all those unicode
glyphs well including rendering the CTL scripts for Hebrew, Arabic,
and Devanagari (the script for Hindi) "peace" words in the correct order,
while xterm (the legacy X terminal emulation app that
uses server side X fonts with no font-config glyph selection) is
missing a number of glyphs, renders the CTL language glyphs in
an incorrect order, and the results for the unicode glphys it
does recognize do not look as good as the results from konsole. So
the conclusion is for the purposes of this evaluation stick to a
terminal emulation app (such as konsole but NOT xterm) that uses X client-side
system unicode fonts and you should get perfect UTF-8
results with info.
3. web pages.
Browse with your favorite browser the file
in your build tree. To look especially for UTF-8 issues
browse , , and in the
range of lines near . I got
perfect UTF-8 results for this case including the glyph order for the
CTL scripts Hebrew, Arabic, and Devanagari (the script for Hindi).
4. pdf file
Browse with your favorite browser the
file that is generated in
your build tree. To look especially for UTF-8 issues search for
plot3dcl, plsurf3dl, and peace. In this case, a long-standing problem
is the Korean and Mandarin glyphs are missing, and the last-few Hindi
glyphs are rendered out of order.
The following tests of our swig documentation are done with python and octave
in separate directories:
1. Preliminaries
Use the cmake option -DADD_SPECIAL_CONSISTENCY_CHECKING=ON, check the cmake.out
file for any warnings about additional components of PLplot missing for this option, and build
the check_swig_documentation target. That target uses the api2swigdoc.pl perl
script to generate from api.xml a check file bindings/swig-support/swig_documentation.i_compare
in the build tree as the basis of a consistency check with bindings/swig-support/swig_documentation.i
in the source tree. Copy that check file to the source file to resolve
any of those inconsistencies. N.B. that source file is the basis
for the python and octave documentation checked below.
2. Check the python documentation
Build the test_python_svg target in order to build
all the required /prerequisites.
cd to examples/python in the build tree and invoke
the python interpreter
Execute the python3 (or python2 if appropriate) Python interpreter
Then for that Python environment execute the following commands:
>>> from plplot_python_start import *
>>> import plplot as w
>>> # list out all possible constants and commands for the plplot binding
>>> dir(w)
>>> # Check documentation strings for some of those commands
>>> print(w.plw3d.__doc__)
>>> print(w.plot3dcl.__doc__)
>>> print(w.plsurf3dl.__doc__)
>>> # etc
These documentation strings should be consistent with the latest
version of the api.xml file. Also the last two should show the expected
UTF-8 symbols ≤ and ≥.
3. Check the octave documentation
* Build the test_octave_svg target in order to build
all the required prerequisites.
* Determine the octavedir environment variable
SRC_EXAMPLES_PREFIX should point to the examples subdirectory of the source tree,
and EXAMPLES_PREFIX should point to the examples subdirectory of the build tree.
e.g.,
export SRC_EXAMPLES_PREFIX=/home/software/plplot/HEAD/plplot.git/examples
export EXAMPLES_PREFIX=/home/software/plplot/HEAD/build_dir/examples
export octavedir=\
"$SRC_EXAMPLES_PREFIX"/../bindings/octave/PLplot:\
"$SRC_EXAMPLES_PREFIX"/../bindings/octave/PLplot/support:\
"$SRC_EXAMPLES_PREFIX"/../bindings/octave/misc:\
"$SRC_EXAMPLES_PREFIX"/octave:\
"$EXAMPLES_PREFIX"/../bindings/octave:\
"$EXAMPLES_PREFIX"/../bindings/octave/PLplot
* Invoke the octave interpreter, load PLplot, and printout help strings for various
PLplot commands.
software@raven> octave -f -q
octave:1> warning("off","Octave:shadowed-function");
octave:2> addpath(getenv("octavedir"));
octave:3> plplot_stub;
octave:4> help plw3d
[...]
octave:5> help plot3dcl
[...]
octave:6> help plsurf3dl
[...]
These documentation strings should be consistent with the latest
version of the api.xml file. Also the last two should show the expected
UTF-8 symbols ≤ and ≥.
UTF-8 limitations of the current set of backend tools
=====================================================
The above set of DocBook backend tools has been chosen with the goal
of allowing essentially arbitrary UTF-8 strings into our DocBook
source. And the advanced.xml part of our DocBook source constitutes a
simple test of how close we have come to that goal by including the
"Peace" word in the 12 human languages expressed in example 24 and by
also including some UTF-8 forms of math symbols. Here are the current
results for how well those UTF-8 strings render for our various
backends.
Man and info results
Should be near-perfect if you are using a
terminal application (e.g., konsole, but NOT xterm) that uses
X client-side system unicode fonts.
HTML results
Should be perfect.
PDF results
These results are reasonably good. All math glyphs and all but the
glyphs occurring in the Korean and Mandarin "Peace" words are present.
I attribute the missing glyph issues to missing Korean and Mandarin
glyphs in the chosen FreeSans, FreeSerif, and FreeMono fonts (more
about that choice below). The order of the glyphs in the Hindi Peace
word is not correct (last two glyphs switched) which is a common
complex text layout (CTL) issue when using unsophisticated software to
render Hindi (Devangari script). However, the Peace word is laid out
in the correct (right-to-left) order for Arabic and Hebrew so there is
some CTL sophistication in the xelatex layout engine, and probably
hope that further layout issues such as the one for Hindi will be
fixed.
More on the reason for the missing PDF glyphs.
It was a huge step forward in the TeX world for the combination of
xelatex and the fontspec TeX package to make a very large number of
UTF-8 glyphs potentially available in pdf results. So I think the
above method of using dblatex with the xetex backend (which uses
xelatex and fontspec internally) is the best we can do to generate the
pdf form of our documentation as free as possible from unicode issues.
For example, the situation is much improved over the previous xmlto
method which severely limited the glyphs in the PDF form of our
results.
However, the fontspec package still has a fundamental limitation which
is you must use a specific rather than generic font for the sans,
serif, and mono cases. Specific fonts are never a good way to go for
documents containing different languages (such as the table of "Peace"
words I recently introduced into advanced.xml) since the user is
forced to compromise between font quality and comprehensiveness. For
example, I have tried using the unifont font (which gives complete
coverage of glyphs for the basic multilingual plane), and the
resulting pdf form of our documentation has no missing glyphs.
However, that font is of extremely poor quality (scaled,
low-resolution bitmapped, monospaced fonts) which makes our
documentation look ugly so I am not going to use it. Instead, I have
chosen (via doc/docbook/src/dblatex_stylesheet.xsl) to use FreeSans,
FreeSerif, and FreeMono fonts which have reasonable quality but which
have some missing glyphs (notably CJK glyphs that leave our Korean and
Mandarin "peace" words empty in the documentation).
The proper way to solve this issue (which would bring TeX completely
into the modern unicode world) would be to modify fontspec so that
certain given font names are considered generic, (e.g., the
"sans-serif", "serif", and "mono-space" font names could be adopted
for this purpose to follow what is done in the SVG world). For those
generic font names, the idea is the fontspec package would simply hand
off font selection to fontconfig which does a very nice job of
automatically selecting the ideal sans, serif, or mono system font to
provide a particular UTF-8 glyph. I have asked the fontspec
maintainer about this potential feature, and his reply indicated there
will be no quick fixes to xelatex/fontspec with regard to automatic
selection of fonts for the generic case, and we will be stuck for the
foreseeable future with the specific font approach. That approach
forces a compromise between complete glyph coverage and reasonable
quality. Thus, as a result of avoiding truly ugly fonts in our
generated pdf documentation there are missing CJK glyphs in the
results.
Installing the generated documentation (and the rest of the generated website) at plplot.org.
===============================================================================================
Follow the directions in README.Release_Manager_Cookbook. That file is
located in the top-level source tree for PLplot.