The warning type :link_private is now gone.

It made sense to warn about such links when we had no private members included
to docs. But now we can freely link to private ones too.

The important thing is for public members to be documented.

--ignore-global now only toggles between showing global class or hiding it.

--warnings=-global turns warnings about global members on/off regardless of
the previous option.

Printing of the warnings is also no more duplicated in both lint.rb and app.rb.

Class#all_members and Class#all_local_members no more filter out private members.
Small refactoring inside Class to avoid confusing use of all_members as local
variable.

No more does it return only public members.

Copy-paste colors from: http://yuilibrary.com/yui/docs/api/

Avoids double-negations to have show:{public:true} than hide:{public:false}.

Filtering class members now also applies the filter to menus.
For this the field is now aligned right - otherwise it'll jump on
every character typed to the field as the member counts change.

So we can now toggle public/protected/private separately.

A bit experimental for now, but seems to work well.

:private is injected to :meta hash when data is exported for use in Docs app.
Inside there it's treated as just another meta-tag.

The fourth argument to compare is an optional file listing
class#method names to ignore. Like so:

    Ext.Element#fly
    Ext.lib.Container#beforeremove

Would be better to get completely rid of the hard-coded offset.

For now the buttons have small text labels - have to come up with some icons.

Instead of two boolean settings, now saving one Object setting containing
two keys.

Hiding of accessors now also works in dropdown menus.

And the code is shorter & cleaner.

When private class happened to have a child node, the class wasn't
grayed out.

Replace ASCII duck with video in README.

Previously this code:

    /** Some docs */
    doSomething(1, 2, 3);

would have resulted in property "doSomething" being detected.

Now it detects no name, which results in warning of missing name,
and that's what we want.

This way when we switch to full JavaScript parser, the tests won't
fail because of syntax errors.

This speeds up source writing at least 2x on multi-processor system.

A strange case where IE9 behaves worse than IE7 and 8.

Implementing this immediately discovered several bugs in ExtJS docs.

Previously the orphans got added multiple times to their class,
resulting in bogus ordering of them.

Previously all @cfgs beginning with uppercase letter were treated
as classes, because a class definition can also contain @cfg tags.
Now we only treat it as class definition when there is more than
one @cfg inside one doc-comment.

Created Docs.ClassRegistry for doing the mapping.

All keywords are now treated as different token types, so we don't
confuse function and "function".

For each member we now show a list of all the members that it overrides
(if there are any). No more needs one wonder if the method happens to
override a method in some parent class or mixin - the information is
right the in the docs.

This implementation mostly works. However I have to refactor it around a
bit to ensure it doesn't misbehave when parallel-processing is used.

The error-handling code had gone out-of sync with the rest. The
Index controller had been renamed, and so error handling failed.

Now there is a separate Failure controller.

In addition to Classes and Guides, error messages are now also shown
when loading of videos or examples fails.