Omit tagName from nth-child selector as it is redundant

[eoghanmurray]

This is another way of asking the following question which I asked in #3

I'm wondering why #some_id div:nth-child(2) a is preferred over #some_id :nth-child(2) a in the existing code? That div appears to be redundant in the examples I've checked.

[eoghanmurray]

I'm publishing this PR to see if the automatic tests fail (they won't run for me on my local machine)

[antonmedv]

So from the tests. Changes to the next:

- script:nth-child(8)
+ body > :nth-child(8)

- .cell:nth-child(1) > p:nth-child(3)
+ .-gutters > :nth-child(1) > :nth-child(3)

So this change will not make selectors shorter in all cases. Sometimes the selector will become longer.

More interesting solution will be to add support for :nth-of-type.

[eoghanmurray]

Ah okay, I've got the tests running locally so can have a better play around with it.
I've changed it to only omit tagNames, but the longer selectors in the examples you thrown up by the tests that you highlighted still exist.
I think we'd need to see what :nth-of-type looks like first as the two could interact.

On a more semantic note, I think relying on the intersection of position and tagName, while it might produce a shorter selector is very brittle.

E.g.

<span
<a href
<a href   <-- a:nth-child(3)

If a span is later inserted (in a different version of the page, or as modifications are made to the page), then we are now referencing a different element:

<span
<span class="new
<a href   <-- a:nth-child(3)
<a href   

So applying this PR is maybe a judgement call either in favour of robustness or brevity, would you agree?

[antonmedv]

So applying this PR is maybe a judgement call either in favour of robustness or brevity, would you agree?

True. I'm not sure what path is better yet.

One concern I have: imagine selector on

.foo :nth-child(0), later some other tag added. The selector is still valid but points to a wrong element. But .foo div:nth-child(0) will break (better to fail fast).

[eoghanmurray]

The selector is still valid but points to a wrong element

Yes, but you are acknowledging that redundancy is good. There are plenty of other times when finder doesn't produce the required level of redundancy for the situation (e.g. my example above).

I've some further ideas with which to update this PR, but for now have produced #70 as the next step, so no point in continuing this discussion if that one is unacceptable!

[eoghanmurray]

I've some parts of nth-of-type already implemented and I'll include it in this PR shortly