Enthought Home home  |  www.enthought.com  |  www.scipy.org  |  installation
Enthought Projects

Chaco FAQ

Introduction  | FAQ | Architecture | Chaco Wiki | API (ETS 3.0.0) | API (ETS 2.7.1)

Does Chaco work with Qt? Tk? ‹insert favorite GUI toolkit here›

Nope, but it is built to support multiple backends. There was once a Tk backend, but that has languished and no one seems interested in working on it. There does seem to be moderate interest in a Qt backend. The actual work involved in creating a new backend for Chaco involves two parts: (1) getting something to draw on, and (2) getting events from the GUI toolkit.

Part (1) is a small but potentially annoying piece. The underlying drawing API that we've created is called Kiva, and in enthought/kiva/agg/src/win32/plat_support.i, we have a SWIG wrapper that creates an extension called PixelMap. This class is attached to the graphics context that is used by the rest of Enable and Chaco, and this class does the work (in C) of blitting the raster buffer into an actual on-screen window. For Win32, this is done using the windows SetDIBitsToDevice() call in agg_platform_specific.cpp:124. Of course, this implies that we can get the OS-dependent window handle from the underlying toolkit, and can thus pass it into this blitting function. WX allows us to get this. (Does QT? If it does, then this part should be fairly straightforward.)

Part (2) is actually the easy part. enthought/enable2/abstract_window.py:AbstractWindow defines several abstract methods that must be implemented by the underlying toolkit in order to get events correctly and translate them. See enable2/wx_backend/window.py for an example. This part really just involves some legwork and putting some dicts together to map things like event types and keycodes.

Where does the name "Chaco" come from?

It is named after Chaco Canyon, which had astronomical markings that served as an observatory for Native Americans. The original version of Chaco was built as part of a project for the Space Telescope Science Institute. This is also the origin of the name "Kiva" for our vector graphics layer that Chaco uses for rendering.

Why the "2" in Chaco2?

Starting in January of 2006, we refactored and reimplemented much of the core Chaco API. The effort has been named "chaco2", and lives in the enthought.chaco2 namespace. At this point, the original chaco package ("chaco classic") is in maintenance-only mode and all of our effort is going into building out Chaco2.

What are the pros and cons of Chaco vs matplotlib?

This question was posed on the enthought-dev mailing list.

Gaël Varoquaux's response:
On Fri, May 11, 2007 at 10:03:21PM +0900, Bill Baxter wrote:

> Just curious. What are the pros and cons of chaco vs matplotlib?

To me it seem the big pro of chaco is that it is much easier to use in a "programatic way" (I have no clue this means something in English). It is fully traited and rely quite a lot on inversion of control (sorry, I love this concept, so it has become my new buzz-word). You can make very nice object oriented interactive code.

Another nice aspect is that it is much faster than MPL.

The cons are that it is not as fully featured as MPL, that it does not has an as nice interactively useable functional interface (ie chaco.shell vs pylab) and that it is not as well documented and does not have the same huge community.

I would say that the codebase of chaco is nicer, but than if you are not developping interactive application, it is MPL is currently an option that is lickely to get you where you want to go quicker. Not that I wouldn't like to see chaco building up a bit more and becoming the reference.

Developers, if you want chaco to pick up momentum, give it a pylab-like interface (as close as you can to pylab) !

My 2 cents,

Peter Wang's response: (excerpt)
On May 11, 2007, at 8:03 AM, Bill Baxter wrote:

> Just curious. What are the pros and cons of chaco vs matplotlib?

You had to go and ask, didn't you? :) There are many more folks here who have used MPL more extensively than myself, so I'll defer the comparisons to them. (Gael, as always, thanks for your comments and feedback!) I can comment, however, on the key goals of Chaco.

Chaco is a plotting toolkit targeted towards developers for building interactive visualizations. You hook up pieces to build a plot that is then easy to inspect, interact with, add configuration UIs for (using Traits UI), etc. The layout of plot areas, the multiplicity and types of renderers within those windows, the appearance and locations of axes, etc. are all completely configurable since these are all first-class objects participating in a visual canvas. They can all receive mouse and keyboard events, and it's easy to subclass them (or attach tools to them) to achieve new kinds of behavior. We've tried to make all the plot renderers adhere to a standard interface, so that tools and interactors can easily inspect data and map between screen space and data space. Once these are all hooked up, you can swap out or update the data independently of the plots.

One of the downsides we had a for a while was that this rich set of objects required the programmer to put several different classes together just to make a basic plot. To solve this problem, we've assembled some higher-level classes that have the most common behaviors built-in by default, but which can still be easily customized or extended. It's clear to me that this is a good general approach to preserving flexibility while reducing verbosity.

At this point, Chaco is definitely capable of handling a large number of different plotting tasks, and a lot of them don't require too much typing or hacking skills. (Folks will probably require more documentation, however, but I'm working on that. :) I linked to the source for all of the screenshots in the gallery to demonstrate that you can do a lot of things with Chaco in a few dozen lines of code. (For instance, the audio spectrogram at the bottom of the gallery is just a little over 100 lines.)

Fundamentally, I like the Chaco model of plots as compositions of interactive components. This really helps me think about visualization apps in a modular way, and it "fits my head". (Of course, the fact that I wrote much of it might have something to do with that as well. ;) The goal is to have data-related operations clearly happen in one set of objects, the view layout and configuration happen in another, and the interaction controls fit neatly into a third. IMHO a good toolkit should help me design/ architect my application better, and we definitely aspire to make Chaco meet that criterion.

Finally, one major perk is that since Chaco is built completely on top of traits and its event-based component model, you can call edit_traits() on any visual component from within your app (or ipython) and get a live GUI that lets you tweak all of its various parameters in realtime. This applies to the axis, grid, renderers, etc. This seems so natural to me that I sometimes forget what an awesome feature it is. :)