Fix wrong order of classes when reordering or adding classes to the front

[liquidnya]

An implementation with 2 test cases which should fix #926.
I also replaced set_attribute and remove_attribute with their stdweb equivalents.
Unfortunately the iterator of indexmap::set::IndexSet is not fused, this might create some overhead.
This fix might introduce more updates (add, remove) to the class_list.

[jstarry]

Thanks for the quick turn around on this PR @LiquidBlock!

What do you think about using set_attribute("class", className) instead of using patches to update classList? We would only need to set that attribute if the two indexsets are different

[liquidnya]

What do you think about using set_attribute("class", className) instead of using patches to update classList? We would only need to set that attribute if the two indexsets are different

That would propably result in faster and less code. I'll give it a try.

[liquidnya]

There is a comment in the code at https://github.com/yewstack/yew/blob/master/src/virtual_dom/vtag.rs#L158-L160.

     /// Adds attribute to a virtual node. Not every attribute works when
     /// it set as attribute. We use workarounds for:
     /// `class`, `type/kind`, `value` and `checked`.

Does this mean, that there might be (or have been) complications with set_attribute("class", className)?

[jstarry]

Nice, did your tests catch this? Pretty odd that IndexSet's equality implementation isn't transitive

[liquidnya]

My tests did not catch this, but the vtag implementation did not use the PartialEq trait of Classes.
Instead it compared the inner classes.set, but since diff_classes(...) also needs to check if the classes changed, i moved the ordered equal check to the PartialEq trait of Classes.

[jstarry]

Does this mean, that there might be (or have been) complications with set_attribute("class", className)?

Good catch, I wasn't aware of the inconsistency of browser support for setAttribute("class", ...). Looks like the ideal solution would be if stdweb supported setting the className property on elements. Unfortunately, it doesn't :/

As far as I can tell, it's only IE that doesn't work well with setAttribute. IE support isn't a priority right now so let's move forward this simpler approach.

[liquidnya]

Good catch, I wasn't aware of the inconsistency of browser support for setAttribute("class", ...). Looks like the ideal solution would be if stdweb supported setting the className property on elements. Unfortunately, it doesn't :/

The absent of className in stdweb is why i first tried to fix the bug with remove and add patches.
What do you think about using className within an js!(...) workaround like
kind (https://github.com/yewstack/yew/blob/master/src/virtual_dom/vtag.rs#L314) and checked (https://github.com/yewstack/yew/blob/master/src/virtual_dom/vtag.rs#L463) do?

[jstarry]

What do you think about using className within an js!(...) workaround

I think that's a great idea