Changed the title of the 3.5 chapter from "Using the change Method" to
"When to Use the change Method" since the chapter tells the migration
definitions supported by the change method. This fix the duplicate ID in the
document.

I've also changed some of the titles to respect the Guides guidelines.

erb tag wasn't closed well

In chapter 2.4 there was an underscore at the start of the attr_readonly method name,
this underscore was interpreted as the opening of the <em> tag but it wasn't closed.

Worst, the underscore in self.class.serialized_attributes (end of paragraph)
was interpreted as the closing of the <em> tag and not as the name of a method.

The app generator is not generally run under bundler, but the Bundler
constant is used here.

In particular you cannot create --dev apps without this.

This reverts commit fc80ebb1.

Such changes aren't allowed in docrails.

ActionView decoupling and sprockets-rails tests fail

Method invalid_asset_host! was delegated to controller but sprockets
compile assets in their own scope without controller. And if we set asset_host
with second parameter it should raise error through invalid_asset_host!.
But since controller is nil it cannot be reached.

Option to run `bundle install` after generating new appplication was
added recently to rails. Since introduction, it contained a subtle bug
that caused it to use `Gemfile` from current directory (if it exists)
rather than from generated directory. This also accidentaly caused
railties tests to work without any problems - after generating test app
it just used `Gemfile` from the repository, rather than the one in
generated app. After fixing the bug mentioned above, this of course
broke. The easiest way to bypass that is to not generate a `Gemfile` for
test application - with such setup Bundler will just use first available
`Gemfile` in one of the parent directories.

The cause of the previous revert was bug in bundler that made it hard to
make railties test work. Fix for bundler was recently pushed to github,
so now we can safely get back to the original commit.

This reverts commit 0f5cc34a.

SessionHash isn't a thing, and tenses are wrong.

This is purely a performance optimisation.

See https://gist.github.com/3552829

This is a performance/GC optimisation.

In theory, this could be optimised by the implementation (last time I
checked, this would have no effect on JRuby). But in practise, this make
attribute access faster.

We don't need separate @class_to_pool and @connection_pool hashes.

* Loop rather than recurse in retrieve_connection_pool
* Key the hash by class rather than class name. This avoids creating
  unnecessary strings.

This reverts commit 7256cb53, reversing
changes made to 6ebe22c3.

Reason: A test was failing.

This reverts commit e4b33b08.

https://github.com/rails/rails/pull/7452#issuecomment-8094302

Fixed little mistake in the docs

the example for contraint( :ip => XXXX) has an invalid regex

The example is about a Person model, say that "params[:model] would usually be a
hash of all the attributes for the person to create" could be confusing.
I changed to params[:person] since the example is about a Person creation.

Added a bullet for :enable_starttls_auto in the smtp_settings table

Since it uses request and response methods

in section 4.1 acts_as_yaffle_test was incorrectly failing due to assertion incorrectly using symbol instead of string

Fix build about TemplateDigestorTest.