Specs fail on MRI Ruby 2.1.0

[felixbuenemann]

[ 188/1376] ExtendTest#test_multiple_extends_with_multiple_extenders_and_single_target = 0.01 s
  4) Failure:
ExtendTest#test_multiple_extends_with_multiple_extenders_and_single_target [/Users/felix/dev/gems/sass/test/sass/extend_test.rb:111]:
<".foo.bar, .bar.baz, .baz.bang, .foo.bang {\n  a: b; }\n"> expected but was
<".foo.bar, .bar.baz, .bang.baz, .foo.bang {\n  a: b; }\n">.

[ 229/1376] ExtendTest#test_parent_and_sibling_extend = 0.01 s
  5) Failure:
ExtendTest#test_parent_and_sibling_extend [/Users/felix/dev/gems/sass/test/sass/extend_test.rb:1250]:
<".parent1 .parent2 .child1.child2, .parent2 .parent1 .child1.child2 {\n  c: d; }\n"> expected but was
<".parent1 .parent2 .child2.child1, .parent2 .parent1 .child2.child1 {\n  c: d; }\n">.

For some reason the order of the class names is not alphabetical.

[felixbuenemann]

It seems the order of items returned by uniq has changed in ruby 2.1.0 (see https://bugs.ruby-lang.org/issues/9340) and Sass::Selector::SimpleSequence#do_extend returns uniq without sort. Unfortunately using sort here breaks a lot of other tests, but replacing uniq with uniq {|p| p} fixes the specs, so this seems to be the exact issue.

[chriseppstein]

I added ruby 2.1 to the build matrix on master: bf4c1a0

[chriseppstein]

it seems like ruby fixed this for the next release, given that the output is semantically identical, it would be nice to simply fix the test assertions to stop caring about this difference.

[felixbuenemann]

I'll give it a try.

[chriseppstein]

@felixbuenemann Thanks!

[nex3]

I'm actually content not to make the tests resilient to this, especially if the old behavior is going to be preserved in a future release of 2.1.0. I'd like the Sass output to remain consistent between Ruby versions as much as possible, and I'm happy to have our tests catch cases where it isn't. In this case, for example, under the 2.0.0 behavior the selector's source order is preserved, which is nice.

[felixbuenemann]

OK, I'll close this.