From Fedora Project Wiki

Revision as of 00:43, 20 September 2011 by Dmalcolm (talk | contribs) (add link to bug 739766)


GCC Python Plugins

Summary

I have created a GCC plugin that embeds Python within GCC. Interested developers should be able to use this to easily extend GCC by writing Python scripts, such as adding new warnings for a library that they work on, without needing to use C.

Owner

  • Email: dmalcolm@redhat.com

Current status

  • Targeted release: Fedora 16
  • Last updated: 2011-09-19
  • Percentage of completion: 95%

Packaged and built within Fedora as gcc-python-plugin-0.3-1.fc16 (review was RHBZ #725570)

The latest tarball release I've made is 0.6, and there's as updated build to that version awaiting testing, to go into the F16 tree: https://admin.fedoraproject.org/updates/gcc-python-plugin-0.6-2.fc16

(this fixes some segfaults seen on i686 (fixed in 0.4), along with numerous other improvements).

However, that build currently fails to work, with this error: https://bugzilla.redhat.com/show_bug.cgi?id=739766


Detailed Description

The plugins will allow interested developers to extend GCC in Python 2 and Python 3, without needing to deal with the C internals of GCC.

Example of potential uses include adding new compiler warnings based on:

  • GTK's reference-counting semantics
  • locking in the Linux kernel
  • signal-safety in APIs.
  • malloc/free usage (e.g. use-after-free bugs)
  • array boundary checks

I aim to use the feature to add compile-time checking of the source code to all CPython extension modules in Fedora, as described in Features/StaticAnalysisOfCPythonExtensions

Other ideas include:

  • visualizations of code structure
  • semantic grepping of source code (e.g. "find all invocations of C++ methods named 'draw' that use the default params")

There will be four GCC plugins:

  • a plugin embedding Python 2 (actually dynamically linked against libpython2.7.so)
  • a plugin embedding python 3 (actually dynamically linked against libpython3.2.so)
  • as above, but for the debug builds of Python 2 and Python 3, rather than the optimized builds.

The "upstream" for the plugins is this Fedora-hosted project: https://fedorahosted.org/gcc-python-plugin/

Benefit to Fedora

Fedora is already a very attractive platform for software development. By making it very easy to extend GCC, the plugin makes Fedora even more compelling to developers. The plugin code was written by a Fedora developer, and is a "Fedora Hosted" project.

Scope

This involves:

  • writing the plugin
  • packaging it
  • documenting it
  • keeping it in sync with gcc

How To Test

The plugin already has a testsuite, which "make" runs, and is run when the rpm is built (within the %check)

The plugin ships with some sample scripts (within the -docs subpackage, below /usr/share/doc/gcc-python-plugin-docs-*/examples)

  • show-gimple.py
  • show-ssa.py

which halt the compilation at each function, showing a graphviz-rendered diagram of that function (in the "gimple" representation, or the gimple representation after transforming to Static Single Assigment form). An interested developer can invoke these scripts by replacing "gcc" with "gcc-with-python2 PATH_TO_SCRIPT":

gcc-with-python2 /usr/share/doc/gcc-python-plugin-docs-*/examples/show-gimple.py main.c

User Experience

Non-technical end-users of Fedora should see no difference.

Developers will be able to invoke Python scripts whilst running GCC by using a script that invokes gcc with the plugin:

gcc-with-python2 PATH_TO_SCRIPT.py rest of regular gcc args

for a Python 2 script and:

gcc-with-python3 PATH_TO_SCRIPT.py rest of regular gcc args

Alternatively, this is equivalent:

 gcc -fplugin=python2 -fplugin-arg-python2-script=PATH_TO_SCRIPT.py

for a Python 2 script and:

 gcc -fplugin=python3 -fplugin-arg-python3-script=PATH_TO_SCRIPT.py 

for a Python 3 script.

Dependencies

This will require working closely with the gcc maintainer. I anticipate needing to rebuild the plugin each time that gcc is rebuilt.

Contingency Plan

Documentation

API documentation for the plugin can be seen at http://readthedocs.org/docs/gcc-python-plugin/en/latest/index.html

Release Notes

  • GCC plugins that embed Python are now available, enabling developers to more easily hook into GCC's inner workings (e.g. to add new compiler warnings). These were written by Fedora contributor David Malcolm.

Comments and Discussion