Move to modular lodash

[mwaldstein]

• Removed use of chain as it required (effectively) all of lodash
• Switched to native Function.bind for the same reason.

With similar change to graphlib, dagrejs/graphlib#86

• dagre.js: 636K => 317K
• dagre.min.js: 554K => 278K

[mwaldstein]

Some notes -

This is more invasive than changes to graphlib or dagre-d3. The reason is that 2 lodash functions effectively ended up pulling in or needing all of lodash, negating the exercise. These functions were used in limited places, so they were written out.

_.chain

Permits composing functions rather than nesting - e.g.

var result = ._chain(input).function1().function2().value()

vs

var result = function2(function1(input))

used in lib/order/cross-count.js and lib/order/resolve-conflicts.js

In both cases there were only 2 functions in the chain, so while nesting is less attractive, it is not illegible.

No risk.

_.bind

The lodash bind is fairly powerful with a lot of options. dagre only uses it to set a scope for 'this' in lib/position/bk.js

There is a low risk with this change. lodash uses Function.prototype.apply to create its bind which has been supported in browsers forever. I made the change to use the native Function.prototype.bind which does not go back as far - mainly not supported by ie8 (compatibility)

There are options to add a polyfill if ie8 support is desired.

[mwaldstein]

Follow-up - graphlib is already using Function.prototype.bind, so ie8 is already non-functional.

[lutzroeder]

The dagre.js built via mwaldstein:master seems to be larger than dagre:master. Any ideas?

[mwaldstein]

I would imagine it is because of using the unmodified graphlib.

Without also making this change dagrejs/graphlib#86, you end up still with the full lodash since graphlib gets fully bundled in.