:has() selector should match against all descendants, not just direct children

[pivotal-benchmark]

From http://api.jquery.com/has-selector/:

The expression $('div:has(p)') matches a <div> if a <p> exists anywhere among its descendants, not just as a direct child.

So I would expect:

Nokogiri::CSS.xpath_for("#outer:has(#inner)")  # => ["//*[@id = 'outer' and .//*[@id = 'inner']]"]

But, instead:

Nokogiri::CSS.xpath_for("#outer:has(#inner)")  # => ["//*[@id = 'outer' and *[@id = 'inner']]"]

which doesn't match:

<div id="outer">
  <div>
    <div id="inner"/>
  </div>
</div>

(I think that's the right way to match all descendants, but I'm not totally solid on XPath. It works in my testing.)

[flavorjones]

On it

[flavorjones]

Just to be clear, to anyone following this issue: ":has()" does indeed match all descendents for most queries. Used in combination with an element id ("#inner"), though (and perhaps one or two other types of selectors), the xpath doesn't work.

It's not clear to me yet whether this is an xpath interpreter bug in libxml, or whether we simply need to restructure the xpath query in these cases. Need to investigate more.

[mgenereu]

Got bit by this. Want me to write test cases?

[flavorjones]

Sure, test cases of the sort found in https://github.com/tenderlove/nokogiri/blob/master/test/css/test_parser.rb#L61-64 would be great -- pass the CSS and assert on the generated XPath query string.

The point being, we need more thorough coverage of what the :has operator should generate for different (more complex) CSS queries.

Thanks!

[Phrogz]

This test line is wrong:

assert_xpath  "//a[b]", @parser.parse("a:has(b)")

It should be:

assert_xpath  "//a[.//b]", @parser.parse("a:has(b)")

Similarly, the next line:

assert_xpath  "//a[b/c]", @parser.parse("a:has(b > c)")

should instead be:

assert_xpath  "//a[.//b/c]", @parser.parse("a:has(b > c)")