mplh5canvas 0.6

A matplotlib backend based on HTML5 Canvas.

This provides a web-delivered interactive matplotlib backend using HTML5 technologies including WebSocket and the Canvas element.

Our main goal is to have a backend that is consistent across multiple platforms, has few installation dependencies, is easy and fast to animate, and retains compatibility with current matplotlib usage scenarios.

Installation instructions can be found below or on the project’s Wiki page. The short answer:

easy_install mplh5canvas

Features

• Pure Python

• Uses mod_pywebsocket to provide multi browser support through multiple websocket standards

• Requires web browser with Canvas and WebSocket support (Chrome 4+, Safari 5+ (OSX and IOS) work out of the box. Opera 11+ works after enabling WebSockets in preferences. Firefox is not supported and Internet Explorer will never be supported.)

• Designed with animation and interactivity in mind (resizable, zoomable, clickable regions, etc)

• Simple plots (e.g. a 2048-point line plot) can be animated at around 60 frames per second

• Allows proper remote access to plots

• Allows multiple concurrent access to plots

• Thumbnail window allows quick cycling between plots on a single page

Screenshot

Requirements

We have tried to keep this module as free of dependencies as possible in order to open up the widest possible installation base. We do however depend on the excellent pywebsocket code for handling our browser communications. This allows us to better track the WebSocket standard as it continues to evolve.

The current base system requirements prior to installation are:

• Python 2.5 or newer (2.4 should also be OK, but has not been tested)

• Matplotlib 0.99.1.1 or newer

The web browser you use to display the plots must support Canvas and WebSockets (see http://caniuse.com/#feat=websockets for a summary of current WebSocket support). The status quo for the main browsers in May 2011 is:

• Chrome 9.0+ works out of the box and is the target browser

• Safari 5.0+ works out of the box (both OSX and IOS versions)

• Firefox 4.0 works after unblocking WebSocket support:

  • Browse to the about:config preferences page and promise to be careful

  • Type websocket in the filter to find the right option

  • Double-click on network.websocket.override-security-block to set it to true

• Firefox 5.0+ is not supported due to a custom websocket class.

• Opera 11.0+ works after unblocking WebSocket support:

  • Browse to the Preference Editor at the opera:config page

  • Type websocket in the “Quick find” search field to zoom in on the right option (“Enable WebSockets” under User Prefs)

  • Check the tickbox and click on the Save button

• Internet Explorer is not supported

Netifaces

It is surprisingly difficult to make a good guess of the IP address of a user’s primary network interface across a range of operating systems. The code uses socket.gethostbyname by default, which does a reasonable job but is completely borked if you have VMware installed.

If available it will use the netifaces module which generally does a better job. It is recommended that you install this by running:

easy_install netifaces

mod_pywebsocket

This will be installed along with mplh5canvas if you follow the instructions below. However you can install it yourself:

easy_install mod_pywebsocket

The PyPI page may be out of data as we maintain this ourselves. The master repository for mod_pywebsocket is pywebsocket.

Installation

Since this package is available on PyPI the simplest way to install it is to do:

easy_install mplh5canvas

Alternatively, download the latest tarball or check out the source code from Google Code and do:

python setup.py install

It is assumed that you have the proper permissions to install Python packages on your system (if not, you can make use of virtualenv instead).

Testing

We provide a number of example scripts for initial testing. Surprisingly these are found in the examples directory of the source code.

The script names are self-explanatory. The URL of the management server should be printed out by the script, and if a web browser is installed and available a new tab should be opened in your browser. If not, then just copy and paste the management URL into a browser window.

Conformance Testing

To try and reach a level of reasonable conformance we have a crude test suite that will run against a directory of matplotlib examples and produce a web page for side-by-side comparison:

cd tests
python test.py -d <matplotlib source tree>/lib/mpl_examples/pylab_examples

This produces output files in the tests/output directory. The file test.html when viewed in a web browser will show the mplh5canvas implementation alongside a PNG and SVG for each file in the target directory.

Be aware that these test results can be pretty massive and may well lead to browser instability. You can run on a restricted set of tests by using a wildcard parameter (see test.py --help).

There is also the option of rendering each canvas on the page to a PNG for easier side-by-side comparison. To do this, run:

python rec.py

once you have a completed test run. At the bottom of the test.html page click the “Connect” button. Then click the “Put Images to Server” button. The rec.py instance should indicate a variety of files being written to disk. Then open the test_rendered.html page which will have a side-by-side column of PNGs.