Fix relationship name-to-post resolution

[sirreal]

Three bugs in lib/class-relationships.php broke the name→slug→post-ID resolution that builds p2p connections:

1. names_to_slugs() never detects fully qualified names. strpos( '\\', $name ) has haystack and needle swapped, so $fully_qualified is always false and a \-prefixed call is still resolved against the current namespace first. PHP semantics say a fully qualified name resolves only globally; now it does. This deliberately changes candidate order for \-qualified calls made inside a namespace.

2. get_ids_for_slugs() connects every matching scope. The loop says "stop searching the chain" but uses continue, so when both the namespace-scoped and the global candidate resolve to posts, both are connected — duplicate/spurious p2p connections. Now the first (most specific) match wins.

3. An empty slug map connects items to post ID 1. When no posts of a target type were imported, the raw array-of-candidate-arrays was left in $relationships, and the connection loop's intval( array, 10 ) coerces each to 1 — connecting items to whatever post has ID 1. The candidates are now cleared so nothing reaches the connection loop. This is the TODO why might this be empty? test class-IXR.php spot.

The new tests fail on master (3 of 5; the other two pin existing correct behavior) and pass with the fix; the full suite passes.

Observed while here, left alone: the methods branches of the connection loop pass 'data' => current_time( 'mysql' ) where the functions branches pass 'date' => — looks like a typo, but changing stored p2p meta is out of scope for this fix.

Found by the multi-agent review during #262; extracted as a standalone change.

🤖 Generated with Claude Code

[sirreal]

Note

This is an agentic review, generated by Claude Code at the repository owner's request.

Ready to land. Three genuine bugs, three minimal fixes, tests that actually fail on master.

Verified

• CI green on 7.4 and 8.4; mergeable, clean, based on master.
• Bug 1 — swapped strpos() arguments. strpos( '\\', $name ) searches a one-character haystack for $name; it returns false for every realistic input, so $fully_qualified was permanently false and \Foo\bar() called inside namespace Baz still got Baz\Foo\bar as its first candidate. strpos( $name, '\\' ) === 0 is the right test, and it matches the resolution rules the method's own docblock describes.
• Bug 2 — continue where break was meant. The inner loop walks the candidate chain most-specific-first; continue advances to the next candidate and keeps matching, so a name that exists both namespace-scoped and globally produced two entries in $slugs_with_ids and therefore two p2p connections. The comment on the line already said "stop searching the chain". break exits the inner foreach only, leaving the outer foreach ( $slugs as ... ) intact — correct scope.
• Bug 3 — post ID 1. Confirmed by reading the second loop: on master, when $this->slugs_to_ids[ $to_type ] is empty the continue leaves $this->relationships[...][ $to_type ] holding the raw array-of-candidate-arrays. The connection loop then does intval( $to_id, 10 ) on an array, which is 1 for any non-empty array, passes the 0 != $to_id guard, and connects the item to whatever post holds ID 1. Assigning array() before the continue is the right minimal fix — the connection loop iterates nothing, and nothing else reads that key. Note the assignment is safe against the surrounding iteration: $to_types comes from (array) @$this->relationships[ $from_type ], a by-value copy, and the non-empty branch already writes the same key.
• Bugs 1 and 2 interact, in the right direction. Together they mean a \-qualified call inside a namespace now resolves to exactly one target — the global one — instead of possibly connecting to a namespace-local post as well.
• RED/GREEN structure is as described: 3 of 5 fail on master. test_names_to_slugs_fully_qualified, test_get_ids_for_slugs_first_matching_scope_wins, and test_ending_import_with_empty_slug_map_makes_no_connections (its first assertion — the raw candidate arrays survive on master) are RED; test_names_to_slugs_unqualified and test_get_ids_for_slugs_unknown_slug_is_ignored pin already-correct behavior. No tautologies. The public visibility of $slugs_to_ids / $relationships makes the direct-injection setup legitimate rather than a hack.
• PHP 7.4-clean.

Worth noting

• Confirmed the 'data' vs 'date' p2p connection meta typo is noted in the description and left alone — the functions branches pass 'date' => current_time( 'mysql' ) while all three methods branches pass 'data' =>. Agreed it does not belong in this PR; it does deserve its own issue, since fixing it changes stored connection meta and anything reading date off a methods_to_* connection has been reading nothing.
• This is importer/relationship code, so the corpus-diff check in #284 will never cover it — the corpus diff compares exporter output only. These unit tests are the entire safety net for this file, which is exactly why adding the first tests/phpunit/tests/relationships.php matters.
• Minor doc drift: names_to_slugs()'s @return still says "starting with the context of the namespace, and falling back to the global namespace", which is now only true for unqualified names. The prose above it already explains the \ rule; a one-line touch-up would close the gap.