Instead of dropping and recreating the database we now drop all the
tables and then recreate them.  This is because in our staging server I
currently have just one database available and I can't drop and recreate
it to run tests in it.

Additionally eliminated the update_votes.sql script - the triggers
themselves take care of votes being calculated.

This way the stdout can be directed to file and the file will only
contain SQL that can be fed into MySQL.

Also added error message when connecting to mongodb fails.

Fixed hard-coded offset on overview panel scroll into view.

As a first step in expanding FunctionAst to do full detection of
various return types, not just whether we're returning this or
something else.

We're now able to determine if the execution can flow through a control
structure or will the control structure always result in a return statement.

Forgot to update the @version in Options class and already pushed the
gem into rubygems.org.  So no other option than to release a next minor
release.

A regression introduced by refactoring the code from JsDuck::Class to
JsDuck::Lint.

Full match variable in ruby is $& not $0.

No need for it to be inside App.  The configuring of Logger already
happens inside Options, so lets have the others there too.

It was implemented long-long time ago and I doubt it's of much use.
Even if somebody uses it, the same information can be retrieved from
--export=full.

Now the App class really can be called simple.

Finally the App class is now relatively simple.

The class was still using the old JsonDuck instead of Util::Json.

Split the huge method into lots of smaller routines.

Move all the messy parsing logic away from already overcomplicated App
class.  Now the App is left with pretty much just one method that
represents the flow of the entire app.  It's still too complicated
though.

This new class encapsulates the formatting of all classes.

Also, while formatting, we now create a new class_formatter for each
class that we process.  That's because otherwise the list of images
detected in one class will be added to the images detected from the next
one, and so on and on...  This wasn't really a problem because we
stripped away the duplicate images anyway, but it wasn't really the
right way to do it.

- Constructor now takes just opts and relations is passed in separately
  through accessor.
- Remove old leftover: @images variable.

No reason at all for it to be part of DocFormatter.

Cancel out the min-width set in stylesheet.
Add <meta> to use device width as default.

Although our simple implementation worked also just fine, the builtin
Singleton has some more checks built into it - just good to rely on
something that's well tried and tested.

Using PHP Mobile_Detect library to do the device detection:
http://code.google.com/p/php-mobile-detect/

This mainly changes how we serve the front page - otherwise mobile is
treated exactly the same as print pages.

So that when the docs are loaded over file: protocol we'll still load
the fonts over http:, not attempting to load them from local filesystem.

Caused by the fact that IE8 doesn't allow the reserved keyword 'enum' to
be used inside a member expression, like so:

    someObj.enum

So I had to rewrite it like so:

    someObj["enum"]

IE9 and other browsers didn't have a problem with that.

Enhance Util::Json.parse to accept an options parameter which Esprima
class needs to pass.  Now everything accesses the JSON gem through
JsDuck::Util::Json wrapper.

Results in much simpler code without all the self.method_name uses.

Use it to abstract away the singleton logic in Logger.

This way splitting the assets into subgroups will help to reduce the
amount of information shown in ThumbList display.