libvirt/docs/strategy.html.in

147 lines
6.7 KiB
XML

<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE html>
<html xmlns="http://www.w3.org/1999/xhtml">
<body>
<h1>Project Strategy</h1>
<p>
This document attempts to outline the libvirt project strategy for
the near future. Think of this as a high level vision or to-do list
setting the direction for the project and its developers to take.
</p>
<h2>Language consolidation</h2>
<p>
At time of writing libvirt uses the following languages:
</p>
<dl>
<dt>C</dt>
<dd>The core libvirt library, daemons, and helper tools are all written
in the C language.</dd>
<dt>Python</dt>
<dd>Various supporting build/test scripts are written in Python, with
compatibility for Python 2 and 3.</dd>
<dt>Perl</dt>
<dd>Various supporting build/test scripts are written in Perl. It is
also used for many syntax-check inline rules</dd>
<dt>Shell</dt>
<dd><code>configure</code>, generated by autoconf, is a shell script.
Shell is also used for some simple build/test scripts. At runtime
libvirt avoids shell except when using SSH tunnels to a remote
host</dd>
<dt>XSLT</dt>
<dd>The website uses XSLT for its templating system. The API
documentation is also autogenerated from an XML description
using XSLT</dd>
<dt>HTML</dt>
<dd>The website documentation is all written in plain HTML. Some HTML
is also auto-generated for API documentation</dd>
<dt>M4</dt>
<dd>The autoconf <code>configure</code> script uses a large number of
M4 macros to generate its content</dd>
<dt>make</dt>
<dd>The core build system uses the traditional GNU make recipes</dd>
<dt>automake</dt>
<dd>The make recipes use automake's language extensions which are
then turned into regular make rules</dd>
<dt>awk/sed</dt>
<dd>A number of the syntax-check inline rules involve use of awk/sed
scripts</dd>
<dt>POD</dt>
<dd>The command line manual pages are typically written in Perl's POD
format, and converted to troff</dd>
</dl>
<p>
The wide range of languages used present a knowledge burden for
developers involved in libvirt, especially when there are multiple
languages all used in the same problem spaces. This is most notable
in the build system which uses a combination of shell, M4, make,
automake, awk, sed, Perl and Python, with debugging requiring
understanding of the interactions between many languages. The
popularity of Perl has declined, while Python has become
more popular. This directly influences the amount and quality of
contributions that can be expected for programs written in the
respective languages.
</p>
<p>
The C language has served libvirt well over the years, but its age shows
giving rise to limitations which negatively impact the project in terms
of code quality, reliability, and efficiency of development. Most notably
its lack of memory safety means that many code bugs become trivially
exploitable security flaws or denial of service. The lack of a high
level portable runtime results in a lot of effort being spent to
ensure cross platform portability. The modern languages Rust and Go
provide viable options for low level systems programming, in a way that
is not practical with other common languages such as Python and Java.
There is thus a desire to make use of either Rust or Go, or a combination
of both, to incrementally replace existing use of C, and also for
greenfield development.
</p>
<p>
With this in mind the libvirt project has set a vision for language
usage in the future:
</p>
<dl>
<dt>C</dt>
<dd>Large parts of the core libvirt library, daemons, and helper tools
will continue to make use in the C language. Integration of other
languages will be an incremental, targetted process where they can
bring the greatest benefit.</dd>
<dt>Rust / Go</dt>
<dd>Parts of the core libvirt library, daemons and helper tools are to
leverage Rust or Go or both to replace C.</dd>
<dt>Meson</dt>
<dd>The core build system is to be written in Meson.</dd>
<dt>Python</dt>
<dd>Various supporting build/test scripts are written in Python 3
compatible mode only.</dd>
<dt>reStructuredText</dt>
<dd>The website and command man pages are to be written in RST, using
Sphinx as the engine to convert to end user formats like HTML, troff,
etc</dd>
</dl>
<p>
Some notable points from the above. Whether the core library / daemons
will use Rust or Go internally is still to be decided based on more
detailed evaluation to identify the best fit. The need to link and embed
this functionality in other processes has complex interactions both at a
technical and non-technical level. For standalone helper tools, either
language is viable, but there are fewer concerns around interactions with
other in-process code from 3rd parties. Thus a different decision may be
made for daemons/libraries vs tools. Any rewrite proposed for existing
functionality will have to weigh up the benefits of the new code,
against the risk of introducing regressions with respect to the previous
code.
</p>
<p>
The Meson build system is written in Python 3. This directly informs the
choice of Python 3 as the language for all supporting build scripts,
re-inforcing the other benefits of Python over Perl, Shell, M4,
automake, etc. There is no intention to support Python 2 given Meson's
requirement for Python 3.
</p>
<p>
Using the RST format for documentation allows for the use of XSLT to be
eliminated from the build process. RST and the Sphinx toolkit are widely
used, as seen by the huge repository of content on
<a href="https://readthedocs.org/">Read The Docs</a>.
The ability to embed raw HTML in the RST docs will greatly facilitate its
adoption, avoiding the need for a big bang conversion of existing content.
Given the desire to eliminate Perl usage, replacing the use of POD
documentation for manual pages is an obvious followup task. RST is the
obvious choice to achieve alignment with the website, allowing the man
pages to be easily published online with other docs. It is further
anticipated that the current API docs generator which uses XSLT to
convert the XML API description would be converted to something which
generates RST using Python instead of XSLT.
</p>
</body>
</html>