Fix document order in REXML::XPathParser.sort

[naitoh]

REXML::XPathParser.sort keys each node on its index under each ancestor, but two different nodes could end up with the same key, and ties are then broken arbitrarily by an unstable sort.

The walk stopped at the root element, so the root and everything outside it -- comments and PIs at document level -- all keyed on an empty array:

    <!--c1--><?pi1?><root/><!--c2--><?pi2?>

//comment() | //processing-instruction()
  was: c1, c2, pi1, pi2   want: c1, pi1, c2, pi2

Walk up to the document instead, so those nodes are ordered against each other.

An attribute borrowed the key of the element carrying it, so the element and its attributes tied:

    <root><a x="1" y="2"><b/></a></root>

//a/@* | //a | //a/*
  was: @x, @y, <a>, <b>   want: <a>, @x, @y, <b>

Extend an attribute's key past its element's with ATTRIBUTE_POSITION. A child index is never negative, so -1 lands the attributes after <a> and ahead of <b>, which is where document order wants them.

The attributes of one element tied with each other too, which went unnoticed because a small enough sort leaves its input alone:

<a z="1" m="2" b="3"/>       //a/@*  ->  z, m, b
<a z="1" ... 26 attributes/> //a/@*  ->  scrambled

XPath 1.0 leaves their relative order implementation dependent, but document order is a total ordering, so it has to be decided. Key them on where they were written, which is the order the small case already appeared to have. Finding that out means walking the attribute list, so index the whole list at once and remember it for the rest of the sort; asking per attribute would make sorting quadratic in the number of attributes an element carries, which is something a document gets to choose.

The ancestor walk becomes a method of its own, since both branches of the key need it. It, attribute_position and ATTRIBUTE_POSITION are private rather than public names marked :nodoc:, sort being the only part callers outside the class use.

[Copilot]

Pull request overview

This PR fixes REXML::XPathParser.sort so XPath results are reliably returned in true document order, including for document-level nodes (comments/PIs outside the root) and for attributes (which must be totally ordered even though XPath 1.0 leaves attribute order implementation-dependent).

Changes:

• Extend the ancestor-walk key generation up to the document node so document-level siblings are ordered correctly.
• Disambiguate attribute ordering by extending an attribute’s sort key past its owning element’s key using a dedicated attribute position marker and a per-element attribute index cache.
• Add targeted regression tests covering top-level nodes, descendant-or-self from the document, element-vs-attribute ordering, attribute list ordering, and mixed child node types.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
lib/rexml/xpath_parser.rb	Reworks XPathParser.sort keying to include document-level ordering and stable attribute ordering via cached attribute positions.
test/xpath/test_base.rb	Adds regression tests to validate document order for top-level nodes, attributes, and mixed child content.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[naitoh]

I have addressed the issues pointed out by PR
Please review it again.

[Copilot]

Pull request overview

Copilot reviewed 2 out of 2 changed files in this pull request and generated no new comments.

[kou]

Thanks.