`rack.version` was bumped in https://github.com/rack/rack/commit/28e77c710ac2ddb35c436d78d72e28f28f964446

The current implementation of `variants=` don't allow a resetting to nil, wich is the default value.

This results in the following code smell:
```ruby
case request.user_agent
when /iPhone/
  request.variants = :phone
when /iPad/
  request.variants = :ipad
end
```

With the ability to reset variants to nil, it could be:
```ruby
request.variants = case request.user_agent
when /iPhone/
  :phone
when /iPad/
  :ipad
end
```

When an `around_action` does not `yield`, then the corresponding action is
*never* executed and the `after_` actions are *never* invoked.

The value returned by the `around_action` does not have any impact on this:
an `around_action` can "return" `true`, `false`, or `"pizza"`, but as long
as `yield` is not invoked, the corresponding action and after callbacks are
not executed.

The test suite for `ActionController::Callbacks` currently includes separate
tests to distinguish the cases in which a non-yielding `around_actions` returns
`true` or `false`.

In my opinion, having such tests is misleading, giving the impression that the
returned value might have some sort of impact, while it does not. At least
that's the impression I got when I read those tests.

For completeness, the tests were introduced 7 years ago by @NZKoz in e80fabbb.

There is no need to subtract one from the path_params size when there is
no format parameter because it is not present in the path_params array.

Fixes #17819.

As suggested in #16299([1]), this method should be a new public API for
retrieving unfiltered parameters from `ActionController::Parameters`
object, given that `Parameters#to_hash` will no longer work in Rails
5.0+ as we stop inheriting `Parameters` from `Hash`.

[1]: https://github.com/rails/rails/pull/16299#issuecomment-50220919

This fixes a regression in 4.2.0 from 4.1.8.

https://github.com/rails/rails/pull/17823 fixed a similar regression regarding _explicitly_ named routes for a mounted Rack app, but there was another regression for the default value.

With a route like:

    Rails.application.routes.draw do
      mount Mountable::Web, at: 'some_route'
    end

The "Prefix" column of rake routes gives the following:

- 4.1.8:         mountable_web
- 4.2.0.beta1-4: [nothing]
- 4.2.0.rc1:     [nothing]
- 4.2.0.rc2:     some_route   <- regression

This fixes the default to go back to being based off the name of the class like the docs specify: https://github.com/rails/rails/blob/785d04e3109f69d0b9b9f4732179592f0ef04e52/actionpack/lib/action_dispatch/routing/mapper.rb#L558-L560

Explicitly named routes still work correctly per https://github.com/rails/rails/pull/17823:

    Rails.application.routes.draw do
      mount Mountable::Web, at: 'some_route', as: 'named'
    end

- 4.1.8:         named
- 4.2.0.beta1-4: [nothing]
- 4.2.0.rc1:     [nothing]
- 4.2.0.rc2:     named

Some `require 'openssl'` statements were surrounded by `rescue` blocks to deal with Ruby versions that did not support `OpenSSL::Digest::SHA1` or `OpenSSL::PKCS5`.

[As @jeremy explains](https://github.com/rails/rails/commit/a6a0904fcb12b876469c48b1c885aadafe9188cf#commitcomment-8826666) in the original commit:

> If jruby didn't have jruby-openssl gem, the require wouldn't work. Not sure whether either of these are still relevant today.

According to the [release notes for JRuby 1.7.13](http://www.jruby.org/2014/06/24/jruby-1-7-13.html):

> jruby-openssl 0.9.5 bundled

which means the above `rescue` block is not needed anymore.

All the Ruby versions supported by the current version of Rails provide those OpenSSL libraries, so Travis CI should also be happy by removing the `rescue` blocks.

---

Just to confirm, with JRuby:

    $ ruby --version #=> jruby 1.7.16.1 (1.9.3p392) 2014-10-28 4e93f31 on Java HotSpot(TM) 64-Bit Server VM 1.8.0_20-b26 +jit [darwin-x86_64]
    $ irb
    irb(main):001:0> require 'openssl' #=> true
    irb(main):002:0> OpenSSL::Digest::SHA1 #=> OpenSSL::Digest::SHA1
    irb(main):003:0> OpenSSL::PKCS5 # => OpenSSL::PKCS5

And with Ruby 2.1:

    $ ruby --version #=> ruby 2.1.2p95 (2014-05-08 revision 45877) [x86_64-darwin13.0]
    $ irb
    irb(main):001:0> require 'openssl' #=> true
    irb(main):002:0> OpenSSL::Digest::SHA1 #=> OpenSSL::Digest::SHA1
    irb(main):003:0> OpenSSL::PKCS5 #=> OpenSSL::PKCS5

Avoid logic in ERB and use helpers

`_generate_paths_by_default` should always be private.

See https://github.com/rails/rails/commit/9b15828b5c347395b42066a588c88e5eb4e72279#commitcomment-8764492

Fixes https://github.com/rails/rails/issues/16958

[Byron Bischoff & Melanie Gilman]

This will help you debug missing template errors, especially if they
come from a programmatic template selection. Thanks to @dhh for
suggesting that.

As a bonus, also show request and response info on the routing error
page for consistency.

Unlike ShowExceptions or PublicExceptions, ParamsParser shouldn't
transform exceptions like Interrupt and NoMemoryError into ParserError.

Reference #17453

These are currently working "by accident" because `match_route` does not check
that the name is valid.

Reference #17453

[Godfrey Chan, Washington Luiz]

Fixes https://github.com/rails/rails/issues/17714.

This reverts commit f93df528, reversing
changes made to a455e3f4.

Conflicts:
	actionpack/lib/action_controller/test_case.rb
	actionview/lib/action_view/test_case.rb

When calling assigns(:...) or cookies(:...), template assertions would be reset, which they obviously shouldn't be.

It returns multiple source extracts since 1ed264bc. Also cleaned its
result structure, as we no longer need the file in a code extract.

Those three can be nil when exception backtrace is nil. This happens and
that forced a couple of nil guards in the code. I'm proposing to make
those always return an array, even on nil backtrace.

Closes #17586

We added a deprecation warning for these cases in aa1fadd4, so these are now
causing deprecation warnings in the test output. AFAICT, in these two cases, the
option is not integral to the purpose of the test, so they can be safely removed

Follow up to 212057b9. Since that commit, we need to pass the `route_name`
explicitly. This is one of the left-over cases that was not handled in that
commit, which was causing `use_route` to be ignored in functional tests.

Fixes: https://github.com/rails/rails/issues/16814

In cases where this option is set to `true`, the option is redundant and can
be safely removed; otherwise, the corresponding `*_url` helper should be
used instead.

Fixes #17294.

See also #17363.

[Dan Olson, Godfrey Chan]

See comment in this patch for the rationale.

References #16468