Pre-process Build

[bagnell]

Adds RequireJS pragmas to conditionally include code that throws DeveloperErrors like:

//>>includeStart('tag', condition);
if (/*condition*/) {
    new DeveloperError('something went wrong.');
}
//>>includeEnd('tag');

tag can be any string describing the block, but the tags in each includeStart/includeEnd must be the same. condition can be any javascript that is evaluated at build time. pragmas.debug will be true for debug builds or false for release builds.

Adds two new build targets, combineRelease and minifyRelease, which are the same as combine and minify except they defined pragmas.debug to be false.

Adds pragma statements like above to every function in Context, Cartesian*, Matrix* and Quaternion that will be removed when pragmas.debug is false.

The performance was improved by ~2-4% for the most frequently called functions. The minified size decreased by ~0.02 MB.

[shunter]

There's a tab in here. Sorry. The Eclipse formatter should be safe to run on the whole file.

[shunter]

Actually, this file uses tabs exclusively, I forgot. In any case, the formatter should be fine.

[shunter]

You should add the <arg value="pragmas.debug=${pragmas.debug}" /> option to the r.js commands that create the workers too.

[bagnell]

@shunter I wasn't sure what else should have it. Everything that uses r.js?

[shunter]

Well, the places where we use r.js to minify CSS doesn't need it. Otherwise, yeah. I think for buildCesiumViewer we can always do debug=false.

[pjcozzi]

Add the new build targets to the Contributor's Guide.

[pjcozzi]

We need links in index.html to be able to run with these?

Also are we required to use the combined version? I thought we were going to setup an AMD version just like code coverage?

[pjcozzi]

I thought we had to make changes to the tests so they still pass in release?

[pjcozzi]

I stepped through Scene.render and added pragmas for most the developer errors, include a few that where still called per-command.

[pjcozzi]

The performance was improved by ~2-4% for the most frequently called functions.

I don't know what this means. We microbenchmarked individual functions? We care about performance in practice for scenes with lots of draw commands.

Compare both FPS and the profiles for:

• Rambler model - http://localhost:8080/Apps/CesiumViewer2/index.html?model=Gallery/model/rambler/Rambler.json. I expect 0.5-2% off the profile.
• Terrain Sandcastle example - Mount Everest.

[pjcozzi]

@shunter can do the merge once this is ready.

[bagnell]

@shunter This is ready for another review.

[bagnell]

This also addresses #1294.

[pjcozzi]

Any update here?

[shunter]

As discussed offline, I fixed a few problems and added info to CHANGES. Merging.