I've been messing around with minitest and came across something very
strange. Does anyone have an explanation for what is going on here?
I think it must indicate subtle differences in blocks between 1.8 and
1.9.
require 'rubygems'
require 'minitest/spec'
···
#
# Including a module in a spec does not make constants
# available under ruby 1.8 but it does under ruby 1.9.
#
# This script requires the minitest gem to be installed:
#
# % gem install minitest
#
module A
module B
end
end
# This passes on ruby 1.9, but not on 1.8.
describe "MiniTest Includes" do
include A
it "should show inclusion works" do
B.must_equal A::B
end
end
# This passes on both. What's more, if you uncomment this
# spec, the assignment TRANSLATES to the "MiniTest Includes"
# spec and that spec will pass!
#
# describe "MiniTest Assigns" do
# B = A::B
#
# it "should show assignment works" do
# B.must_equal A::B
# end
# end
# Here is an example showing the assignment case works.
module X
module Y
end
end
describe "MiniTest Assigns X::Y" do
Y = X::Y
it "should show constant assignment works" do
Y.must_equal X::Y
end
end
As an additional note to this, assigning constants in a describe under
1.8 causes namespace pollution.
require 'rubygems'
require 'minitest/spec'
module X
module Y
end
end
describe "Namespace Pollution" do
Y = X::Y
end
puts Y # => 'X::Y' on 1.8, error on 1.9
It's not specific to minitest; it's a change in how constants are
resolved. I'm not sure of the exact formulation but what's happening
is that in a class_eval'd code block (which I think is what describe
does), 1.8 is looking for a constant scoped to the outer level, while
1.9 is looking for one that is scope to the class that's being
class_evaled.
You can see this also with a block for a new class:
module A
module B
end
end
Class.new do
include A
p B == A::B # true in 1.9, error in 1.8 (unknown constant B)
end
It's not specific to minitest; it's a change in how constants are
resolved. I'm not sure of the exact formulation but what's happening
is that in a class_eval'd code block (which I think is what describe
does), 1.8 is looking for a constant scoped to the outer level, while
1.9 is looking for one that is scope to the class that's being
class_evaled.
Interesting. I expected it would be something like this. Kinda makes
specs hard to use under 1.8 IMO. You're left with nesting describes
within a module, or using the full constant name.
module A
module B
end
describe "Nest in Module" do
it "should show nesting works" do
B.must_equal A::B
end
end
end
describe "Full Constant Name" do
it "should show full names works" do
end
end
It's not specific to minitest; it's a change in how constants are
resolved. I'm not sure of the exact formulation but what's happening
is that in a class_eval'd code block (which I think is what describe
does), 1.8 is looking for a constant scoped to the outer level, while
1.9 is looking for one that is scope to the class that's being
class_evaled.
Interesting. I expected it would be something like this. Kinda makes
specs hard to use under 1.8 IMO. You're left with nesting describes
within a module, or using the full constant name.
module A
module B
end
describe "Nest in Module" do
it "should show nesting works" do
B.must_equal A::B
end
end
end
describe "Full Constant Name" do
it "should show full names works" do
A::B.must_equal A::B
end
end
All it means is that you can't make an assertion in the spec that
conflicts with how the language is engineered, and that's always true
Here's a workaround that combines the include call with a const_get
call. Can still use 'describe'. Doesn't pollute namespace. Works on
1.8 and 1.9. Not too ugly:
require 'minitest/spec'
MiniTest::Unit.autorun
# put this little method in some kind of spec_helper.rb file
module Kernel
def use(const, subconst)
self.instance_eval( "include #{const}" )
const.const_get(subconst)
end
end
module A
module B
end
end
describe 'Includes' do
B = use A, 'B'
it 'shows inclusion working' do
B.must_equal A::B
end
end