Segfault with @extends and :not

[xzyfer]

Originally reported by @dryoma in sass/node-sass#1465

foo {
  bar: baz;
}

@mixin link() {
  a:not(.btn) {
    color: red;
  }
}

foo {
  @include link;
  @extend .btn;
  @include link;
}

Ruby Sass

foo {
  bar: baz; }

foo a:not(.btn):not(foo) {
  color: red; }
foo a:not(.btn):not(foo) {
  color: red; }

LibSass 3.3.3+

[1]    51362 segmentation fault  /Users/michael/Projects/Sass/sassc/bin/sassc test.scss

LibSass 3.3.2

foo {
  bar: baz; }

foo a:not(.btn) {
  color: red; }

foo a:not(.btn) {
  color: red; }

[mgreter]

I got Internal Error: deque::_M_new_elements_at_front on the first run (now getting segfault too)

[mgreter]

The segfault was easy to fix as most of the time (another missing check).
It was triggered by a recent refactor of mine and the results look "promising".

libsass 3.5.5+

foo {
  bar: baz; }
foo a:not(.btn, foo) {
  color: red; }
foo a:not(.btn, foo) {
  color: red; }

libsass 3.3.2 (sassmeister)

foo {
  bar: baz; }
foo a:not(.btn) {
  color: red; }
foo a:not(.btn) {
  color: red; }

expected

foo {
  bar: baz; }
foo a:not(.btn):not(foo) {
  color: red; }
foo a:not(.btn):not(foo) {
  color: red; }

So it was silently dropping the :not(foo) selector, which means it could potentially break the expected css (if compared to ruby sass). Therefore I consider the now invalid css that is output more accurate, since it basically also shows that we now have the right selectors, but not in the correct form yet. And I'm pretty sure that is a duplicate of #1960. I guess we simply need to expand wrapped selectors that contain a list (maybe restricted to some specific type).

[mgreter]

Continued in #2031