In  issue 680549  we limited invalidation of type selectors to the selectors where there were no other simple selectors for which we had an invalidation set. So, for a stylesheet like this:

  div {}
  #id span {}

we would invalidate all divs, but only invalidate spans which are descendants of #id through scheduling the invalidation set for #id.

Still, we schedule invalidation sets both for ".a" and ".b" if we have a selector ".a .b" where we could have scheduled for ".a" only, avoiding invalidation of ".b" not inside ".a".

For selectors with more than one combinators like ".a #b .c" it might be better to schedule invalidations for "#b" instead of ".a". Also, if we have a selector ".a .b .c" and no elements in the document has a class ".b", not scheduling a set for ".a" could be an optimization.

See https://codereview.chromium.org/2683373003/ which was done before the changes for type selector and custom pseudo optimizations were done for  issue 680549 .