@extend loops output selectors in the wrong order

[xzyfer]

.foo {a: b; @extend .bar}
.bar {c: d; @extend .baz}
.baz {e: f; @extend .foo}

Ruby Sass

.foo, .baz, .bar { a: b; }
.bar, .foo, .baz { c: d; }
.baz, .bar, .foo { e: f; }

LibSass

.baz, .bar, .foo { a: b; }
.foo, .baz, .bar { c: d; }
.bar, .foo, .baz { e: f; }

Spec https://github.com/sass/sass-spec/tree/master/spec/libsass-todo-tests/180_test_basic_extend_loop
Spec https://github.com/sass/sass-spec/tree/master/spec/libsass-todo-tests/181_test_three_level_extend_loop

[mgreter]

Have seen this one already, IMO it only happens with circular extends ...

[xzyfer]

As pointed out by @andybenedict in #1539 this can actually cause broken code when attribute selectors are involved.

.foo{
  &[disabled]{
    display: none;
  }
}

.bar {
  @extend .foo;
}

Libsass

.foo[disabled], [disabled].bar {
  display: none;
}

[mgreter]

@chriseppstein maybe you have any pointers as to how the ordering works?

[mgreter]

This is another interestingly similar issue:

.thing[disabled] {
    @extend .thing;
    background: blue;
}

ruby sass:

.thing[disabled] {
  background: blue; }

libsass:

[disabled].thing {
  background: blue; }

[chriseppstein]

I'd really like to make sass-spec agnostic of non-semantic differences like this. The order doesn't matter and I don't think libsass should work very hard to achieve that level of consistency when there are bigger fish to fry.

[mgreter]

@xzyfer you pointed out that "by @andybenedict in #1539 this can actually cause broken code when attribute selectors are involved." Can you provide an example of this? I agree with @chriseppstein that it should not make any difference!?

[xzyfer]

You're both correct. I realised my mistake but hadn't removed the comment. This is why I removed the milestone as per #2174

[mgreter]

Thanks for the clarification, so it's basically a whitespace issue ...

[xzyfer]

Correct.

[xzyfer]

Once we have semantic equivalence in sass spec, as opposed to exact matching, this will be "fixed".

[mgreter]

Closing as dart-sass also exhibits inconsistent order with extend and doesn't seem to care.