Cassowary is an algorithm that computes flexible, responsive layouts quickly without resorting to piles of imperative code. Just describe the preferred relationships between values, noting which constraints are more important than others, and Cassowary figures out an optimal solution based on the current inputs. When the inputs or constraints change, Cassowary is particularly efficient at computing a new answer quickly based on the last-known solution. These properties together make it ideal for use in layout systems -- indeed, it's the algorithm at the center of Apple's new automatic layout system for Cocoa.
This version dramatically improves the performance of the original translation, removes external library dependencies, and improves hackability. The solver core can now be used inside web workers, at the command line, and directly in modern browsers.
For civil discussion of this port and constraint-based UIs, join the Overconstrained mailing list.
Constraint solvers are iterative algorithms that work towards ever more ideal solutions, often using some variant of Dantzig's simplex method. They are primarialy of interest in situations where it's possible to easily set up a set of rules which you would like a solution to adhere to, but when it is very difficult to consider all of the possible solutions yourself.
Cassowary and other hierarchial constraint toolkits add a unique mechanism for deciding between sets of rules that might conflict in determining which of a set of possible solutions are "better". By allowing constraint authors to specify weights for the constraints, the toolkit can decide in terms of stronger constraints over weaker ones, allowing for more optimal solutions. These sorts of situations arise all the time in UI programming; e.g.: "I'd like this to be it's natural width, but only if that's smaller than 600px, and never let it get smaller than 200px". Constraint solvers offer a way out of the primordial mess of nasty conditionals and brittle invalidations.
If all of this sounds like it's either deeply esoteric or painfully academic, you might start by boning up on what optimizers like this do and what they're good for. I recommend John W. Chinneck's "Practical Optimization: A Gentle Introduction" and the Cassowary paper that got me into all of this: "Constraint Cascading Style Sheets for the Web"
Cassowary is "distributed as an NPM package" and can be added as a dependency or used under node in the usual way as a result. Using Cassowary under node is as simple as:
// The entire API is exported by the cassowary objectvar c = ;var solver = ;var x = value: 167 ;var y = value: 2 ;var eq = x y;solver;// ...
The current low (sub 0.1) version number reflects the instability of the API. Also, note that the NPM pacakage includes no tests or demos. For those, clone the github repo.
To make an NPM package from sources, clone the github repo, follow the below
instructions for installing dependencies, and run
make dist. This is the same
process the maintainers use to package NPM releases.
This repo pulls in other Git repositories through submodules. After cloning the repo, run:
$ git submodule update --init...
To run the tests, point your thorougly modern browser at
Running tests from the command line requires Node. Once you've installed Node, run:
$ npm install...$ npm testcd tests; ../node_modules/.bin/mocha --reporter list test.js| 93 tests to run in 10 groups| ------------------------------------------------------------| GROUP "c" has 5 tests to run| _inc| own| extend| inherit| basicJSON| ------------------------------------------------------------| GROUP "c.Constraint" has 13 tests to run| equationFromExpression...
If you have a working
make, a Makefile is provided with a
test target that
does the same thing. The Makefile also provides a
make build target which
generates a new minified
bin/c.js binary out of the files in
requires Python and isn't something you should need to do manually as it's not
reqired to run tests or use the solver. The checked-in binary should always be
up-to-date (or at some checkpoint which is known-good), so use it in your
projects instead of the source versions.
This refactoring currently runs in:
- Chrome (and Chrome Frame)
- Firefox 9+
- Opera 11+
- Safari 5+
- IE 9+
- V8 (d8 shell)
- JSC (built into OS X)
- Rhino (Java) js.jar included in checkout
This is an unapolgetically modern reinterpretation optimized for size, low complexity, and speed. And litle else. No, it will not work on old versions of IE -- that's what Chrome Frame is for.
// Log general debugging informationc.debug = [ false || true ]; // default false// Detailed loggingc.trace = [ false || true ]; // default false// Verbose loggingc.verbose = [ false || true ]; // default false// Logging of tableau additionsc.traceAdded = [ false || true ]; // default false// Logging of ...?c.GC = [ false || true ]; // default false