As of a recent [commit](https://github.com/rspec/rspec-expectations/blame/79582c2c160ac7b3a58d3d128012b022c63d6174/lib/rspec/matchers/built_in/raise_error.rb#L38) to rspec-expectations, Rspec::Matchers::BuiltIn::RaiseError strictly requires a Proc rather than something that quacks like one.

Reverts half of be0b142

@ronen: it looks like what happened here is that does_not_match#super 
will pass whatever it is given into `matches?` and as a result, the 
matcher was never ever being passed an actual ::Given::Failure. This 
caused all of the (already red) negative cases to remain red even after
the patch.

Jim had published 2.12.0 as the lowest supported RSpec version, but 
under that version, several important tests fail. I've updated the 
lowest supported version to 2.14.0 (the earliest version at which all
tests pass)

see #2

Explanation:
http://myronmars.to/n/dev-blog/2014/05/notable-changes-in-rspec-3

Rspec3 support

This change is very basic: if ActiveSupport::TestCase is defined, then
require minitest-given under an artificial circumstance in which 
Minitest::Spec is aliased to ActiveSupport::TestCase (so that TestCase
absorbs all of the monkey-patching normally applied to Minitest::Spec).

This will be a breaking change to anyone currently using Minitest::Spec
directly from a test suite that loads Rails, but it's not clear to me
that this is a common (if possible?) use case. If that use case matters,
it may be necessary to force-load all the files again after un-aliasing
the Minitest::Spec constant.

lib/minitest-given was changed so that Bundler's auto-require would pick
up the same intelligence that `require 'minitest/given'` will now have.

Support minitest-given use in Rails tests

The build will fail for older versions of RSpec 2+, but the lib will continue to support them

`value` is defined by Minitest::Spec and will
error when a `let` is created with that name

I hate effectively branching in tests, but this is
necessary to make the build pass under RSpec 2.x, 
as I forgot all the same tests are run against both

Fixes the build

…ls so it works for both at the same time.

that Given works with ActiveSupport::TestCase.

Keeping blocks to one line despite length.

It looks like Jim's idea with `_Gvn_before` was to have each framework
quack with its own implementation. This commit keeps the Minitest 
before hacks on its side of the fence (and the RSpec before extension
that simply delegates to RSpec before remains in place).

Make rspec-given work for both TestCase and Minitest::Spec together.

Updated code that compares the version of JRUBY to '1.7.5' to determine
if natural assertions are supported. Fixes #2.

warning: assigned but unused variable - ex

fix warning

Add natural assertions support for JRuby 1.7.10 and higher

RSpec changed how it detects source lines to print in
rspec/rspec-core#2088, which necessitates us configuring
`project_source_dirs` for our specs. (This should not affect
end-users).

In addition, we now print multi-line code snippets as of
rspec/rspec-core#2083. Multiline snippets are not printed
on the same line as `Failure/Error:` so our regex has to
be updated to accommodate a line break.

Having the if-else branch wrap the entire describe confused me, because
I couldn't tell whether the metadata support we offer differs 
substantially between RSpec 2 & 3 (it conflated the fact that RSpec 3
added default-true metadata, which 2 lacks with the fact the `example`
DSL was removed from RSpec 3)

I decided to make the default-to-true metadata example more explicit, as
well, including a slightly more rugged rspec_3_or_later? helper available
to the rest of the unit test sutie.

Bump travis to test against the latest rubby

…older versions of JRuby

This closes #15.

Because this is a publicly available method (even though zero people 
likely need/know it exists), I'm going to do a minor version bump over
it.

Since the messages you get on non-Ripper Rubies are borderline worthless
I'm not comfortable saying that natural assertions are supported per se,
just agree that it's no longer enough for rspec-given to blow up.

This also lets us use a meaningfully named constant we own to branch
around the ripper/sorcerer stuff instead of doing defined? checks on
ripper/sorcerer.

Additionally, having two LoadErrors was duplicative and would result in
two very similar warnings hitting a person's console.

This reverts commit 35d2f82.

Because we're publishing multiple gems, using rspec-given from the git repo does
not quite work correctly or make sense.

`Then { }` shouldn't fail normally (and doesn't), but the way we 
ascertain that is by verifying via ripper that the statement is a 
:void_stmt.

I don't want to get into the business of manually parsing Ruby with
regex to guess whether I'm looking at a void assertion statement, so
I'm just going to add to the warning that these will also fail without
Ripper.

Transferred from jimweirich#54

Grammar tweaks to readme

Fixes #30

Thanks @schadenfred!

Fixes #31