Skip to content

Latest commit

 

History

History
186 lines (135 loc) · 5.97 KB

Standard-Plugins-Documentation.rst

File metadata and controls

186 lines (135 loc) · 5.97 KB
Raw

Nuitka Standard Plugins Documentation

Background: Nuitka Plugins

Plugins are a feature to modify the way how Nuitka compiles Python programs in extremely flexible ways.

Plugins can automatically include data files and additional shared libraries, import modules which are not detectable by source code analysis, modify or extend the to-be-compiled source code, gather statistics, change Nuitka's parameter defaults and much more.

Any number of plugins may be used in each compilation.

Plugins come in two variants: standard plugins and user plugins.

User plugins are not part of the Nuitka package: they must be provided otherwise. To use them in a compilation, Nuitka must be able to find them using their path / filename. If user plugins are specified, Nuitka will activate them before it activates any of its standard plugins.

Standard plugins are part of the Nuitka package and thus always available.

Nuitka also differentiates between "mandatory" and "optional" standard plugins.

Mandatory standard plugins are always enabled and "invisible" to the user. Their behaviour cannot be influenced other than by modifying them.

Optional standard plugins must be enabled via the command line parameter --plugin-enable-=name, with an identifying string name. Even when not enabled however, optional standard plugins can detect, whether their use might have been "forgotten" and issue an appropriate warning.

Where appropriate, the behaviour of optional standard plugins (like with user plugins) can be controlled via options (see "Using Plugin Options").

A Word of Caution

Almost all optional standard plugins are relevant for standalone mode only. Specifying all the right plugins is up to the user and critical for success: for example. if you are using package numpy and forget to activate that plugin, then your compile will

  • end with no error, but a warning about missing numpy support,
  • not generate a working binary.
  • user plugins are able to programmatically enable optional standard plugins, the reverse is not possible. The user must know the requirements of his script and specify all appropriate optional standard plugins, including any required options (see below).
  • There is currently no way to automatically react to interdependencies. For example, when compiling a script using the tensorflow package in standalone mode, you must enable (at least) both, the tensorflow and the numpy plugin.
  • Like every compiler, Nuitka cannot always decide, whether a script will actually execute an import statement. This knowledge must be provided by you via specifying plugins.

In this repository folder you find help to address the above points of caution. These tools provide runtime information of your program to the Nuitka compiler, such that all required plugins are activated automatically, and only used packages are included.

List of Optional Standard Plugins

Create a list of available optional standard plugins giving their identifier together with a short description via --plugin-list:

       The following optional standard plugins are available in Nuitka
--------------------------------------------------------------------------------
dill-compat      Required by the dill module
eventlet         Required by the eventlet package
gevent           Required by the gevent package
multiprocessing  Required by Python's multiprocessing module
numpy            Required for numpy, scipy, pandas, matplotlib, etc.
pmw-freezer      Required by the Pmw package
pylint-warnings  Support PyLint / PyDev linting source markers
qt-plugins       Required by the PyQt and PySide packages
tensorflow       Required by the tensorflow package
tk-inter         Required by Python's Tk modules
torch            Required by the torch / torchvision packages

Optional Standard Plugins Documentation

dill-compat

  • Required by the dill module. Dill extends Python's pickle module for serializing and de-serializing objects.
  • Options: none.

eventlet

  • Required by the eventlet package. Eventlet is a concurrent networking library.
  • Options: none.

gevent

  • Required by the gevent package. Gevent is a coroutine-based Python networking library that uses greenlet to provide a high-level synchronous API.
  • Options: none.

numpy

  • Required for numpy, scipy, pandas, matplotlib, xarray, sklearn, skimage, and most other scientific packages.
  • Options: "scipy", "matplotlib" if used. E.g. --plugin-enable=numpy --include-scipy --include-matplotlib.

pmw-freezer

  • Required by the Pmw package. Pmw is a toolkit for building high-level compound widgets.
  • Options: none.

pylint-warnings

  • Support PyLint / PyDev linting source markers. Python static code analysis tools which help enforcing a coding standard.
  • Options: none

qt-plugins

  • Required by the PyQt and PySide GUI packages.
  • Options: "sensible", "styles", "qml", "xml", "all", where "sensible" means the default minimum set of Qt features.

tensorflow

  • Required by the tensorflow package. TensorFlow is an open source machine learning framework for everyone. Note that this package requires numpy and potentially many other packages.
  • Options: none.

tk-inter

  • Required by Python's Tk modules.
  • Options: none.

torch

  • Required by the torch and torchvision packages. Tensors and Dynamic neural networks in Python with strong GPU acceleration. Torchvision requires numpy.
  • Options: none.