wdamron / cassowary-js-refactor

Cassowary/JS, better, faster, future-ready

Cassowary JS

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 repo hosts an improved version of Greg Badros's port of the Cassowary hierarchial constraint toolkit to JavaScript.

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 Solver? Say What?

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"

Getting Started

This repo pulls in other Git repositories through submodules. After cloning the repo, run:

$ git submodule init
$ git submodule update
...

To run the tests, point your thorougly modern browser at tests/unittests.html or demos/quad/quaddemo.html. To run from the command line, first, see if they already run without any extra work (they should on most Mac or Linux boxes):

$ cd tests
$ ./run.sh
...

If you get an error like:

$ ./run.sh 
FAILED: No JavaScript Runtime Found! Please install Java or the V8 Shell (d8) and add them to your $PATH

Check out a copy of V8 and building the latest debugging shell:

scons snapshot=on console=readline objectprint=on d8

This make take a while and yes, it requires Scons (which implies a Python dependency) and a sane C++ compiler in your PATH. Now add the directory with the d8 executable to your PATH for running the unit tests.

Then invoke the command-line test runner from inside the test directory:

$ cd tests
$ ./run.sh
...
done adding 63 constraints [500 attempted, 0 exceptions]
time = 0.021
done adding 63 constraints [500 attempted, 0 exceptions]
time = 0.023
Editing vars with indices 70, 56
about to start resolves
done resolves -- now ending edits
total time = 0.325

  number of constraints:             100
  number of solvers:                  10
  numbers of resolves:                50
  tests:                               1
  time to add (ms):                   23
  time to edit (ms):                   5
  time to resolve (ms):               62
  time to edit (ms):                   2
  add time per solver (ms):        0.023
  edit time per solver (ms):        0.25
  resolve time per resolve (ms):   0.124
  time to end edits per solver (ms): 0.1

If you would like to avoid building V8 but have Java installed, a copy of Rhino (js.jar) is included in this repo. run.sh should detect that d8 is not in your path and attempt to fall back to using Rhino to run the tests. If it's not working out of the box, check to ensure that Java is installed and in your PATH.

Supported Runtimes

This refactoring currently runs in:

• Chrome (and Chrome Frame)
• Firefox 9+
• Opera 11+
• Safari 5+
• IE 9+
• Command-line:
  • 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.

Configuration

// Log general debugging information
c.debug = [ false || true ]; // default false
// Detailed logging
c.trace = [ false || true ]; // default false
// Verbose logging
c.verbose = [ false || true ]; // default false
// Logging of tableau additions
c.traceAdded = [ false || true ]; // default false
// Logging of ...?
c.GC = [ false || true ]; // default false