A matplotlib backend based on HTML5 Canvas.
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:
- Pure Python
- Uses mod_pywebsocket to provide multi-browser support through multiple websocket standards
- Requires up-to-date web browser with Canvas and WebSocket support (since the start of 2013 the latest versions of all major browsers should work out of the box - see Wiki page for more details)
- 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
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, 2.6 or 2.7 (2.4 should also be OK, but has not been tested)
- matplotlib 0.99.1.1 or newer (will be automatically installed if you are going the easy_install mplh5canvas route)
- pywebsocket 0.6 or newer (will be automatically installed if you are going the easy_install mplh5canvas route)
If you want to make use of easy_install as suggested in the instructions below, you should install setuptools. Alternatively you can use pip instead of easy_install by installing the pip Python package.
The web browser you use to display the plots must support Canvas and WebSocket (see http://caniuse.com/#feat=websockets for a summary of current WebSocket support). The status quo for the major browsers in July 2013 is:
Chrome 9 and later works out of the box and is the target browser (both desktop and mobile versions)
Safari 5 and later works out of the box (both Mac OS X and iOS versions)
Firefox 4 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 to 10 is not supported due to a custom MozWebSocket class
Firefox 11 and later works out of the box
Opera 11 to 12.02 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
Opera 12.10 and later works out of the box
Internet Explorer 10 and later should work out of the box but is untested
Opera Mini and Android Browser have no WebSocket support yet and are therefore not supported
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:
This will be installed along with mplh5canvas if you follow the instructions below. However, you can install it yourself:
Since this package is available on PyPI the simplest way to install it is to do:
Alternatively, install mod_pywebsocket yourself, download the latest mplh5canvas tarball (or check out the source code) from the Google Code page 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).
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.
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:
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.