[bugfix] declare copy-namespaces no-inherit breaking element constructors

[joewiz]

Summary

declare copy-namespaces preserve, no-inherit was incorrectly applying the no-inherit flag during element constructor evaluation, causing two distinct bugs:

1. Name resolution failure: getURIForPrefix() skips inheritedInScopeNamespaces when inheritNamespaces=false, so a child constructor <b/> inside <e xmlns="http://example.com/"> could not see the inherited default namespace, and a prefixed name like <ns:c/> whose prefix was declared on an enclosing element threw XPTY0004.

2. in-scope-prefixes() returning incomplete results: Nested direct constructors like <e3>{<e2>{<e1/>}</e2>}</e3> — calling in-scope-prefixes(<e1>) only returned <e1>'s own prefix, missing namespace2 and namespace3 declared by the enclosing constructors.

Per XQuery 3.1 §3.9.3.4, no-inherit governs how namespaces propagate from copied source nodes (existing XDM nodes placed into constructors via {$var}) — it must not affect namespace resolution for element constructors themselves.

Also fixes the real-world regression in #2182 where declare copy-namespaces preserve, no-inherit caused XPath path steps over constructed elements to silently return empty sequences.

Companion to PR #6219 (serializer xmlns="" fix).

What Changed

ElementConstructor.java — Fix A (name resolution)

Temporarily restores inheritNamespaces=true while resolving the element's QName via QName.parse(), so that prefix-to-URI lookups consult the inherited namespace context. Restored in a finally block.

FunInScopePrefixes.java — Fix B (ancestor traversal)

Always traverse the in-memory node's ancestor chain when collecting namespace prefixes, removing the coarse inheritNamespaces() switch that previously blocked traversal for all no-inherit queries. Updated cleanup to remove all entries with an empty URI (namespace undeclarations).

EnclosedExpr.java — Fix C (no-inherit copy semantics)

When no-inherit is active and an enclosed expression {...} places a pre-existing element node (a variable reference) into the outer element, a new NoInheritCopyReceiver intercepts the copy event stream and injects namespace undeclarations (xmlns:prefix="") onto the root of the copy for every ancestor namespace binding not already declared by that root. This neutralizes ancestor traversal for those nodes so in-scope-prefixes() returns only the node's own context.

Pre-existing nodes are distinguished from direct constructors by capturing the MemTreeBuilder allocated during the enclosed expression's evaluation (via new peekDocumentBuilder()) and checking whether each result node belongs to that builder's document.

XQueryContext.java / ModuleContext.java: added peekDocumentBuilder() to inspect the current builder without disturbing the context stack.

Spec References

• XQuery 3.1 §3.9.3.4 — Copy Namespaces Declaration

XQTS Before/After (prod-CopyNamespacesDecl, XQ 3.1)

Test	Before	After
K2-CopyNamespacesProlog-4	FAIL	PASS
K2-CopyNamespacesProlog-5	FAIL	PASS
K2-CopyNamespacesProlog-9 (direct constructor half)	FAIL	PASS
copynamespace-2	FAIL	PASS
K2-CopyNamespacesProlog-9 (variable copy half)	PASS	PASS ✓
copynamespace-3 (no-inherit, variable copy)	PASS	PASS ✓
copynamespace-5 (no-inherit, variable copy + FLWOR sort)	PASS	PASS ✓

35/35 passing in the prod-CopyNamespacesDecl test set.

CI Changes (second commit)

This PR also cherry-picks CI improvements from PR #6186 and extends them:

• exist-parent/pom.xml: Added forkedProcessTimeoutInSeconds=600 to surefire and failsafe plugins, preventing hung test JVMs from blocking CI indefinitely (e.g. DeadlockIT, MoveResourceTest).
• .github/workflows/ci-test.yml: Added --offline to the license:check step (eliminates 7+ min timeouts from unreachable repos; the develop cache has all needed artifacts). Added wagon HTTP timeout flags via step-level MAVEN_OPTS to prevent Maven Build hangs caused by stalled connections to repo.exist-db.org and repo.evolvedbinary.com.

Test Plan

• XQTS prod-CopyNamespacesDecl: 35/35 pass (4 were failing before, none regressed)
• JUnit full suite: 6,542 tests, 0 failures, 0 errors
• Manual: declare copy-namespaces preserve, no-inherit; <e xmlns="http://example.com/">{ <b/> }</e> → <b> in http://example.com/ (implied by K2-CopyNamespacesProlog-4 ✓)
• Manual: Fragment selection skips namespace declaration for xsi:type value #2182 (@daliboris) — $items/t:item/cc:collection returns result instead of empty sequence (implied by copynamespace-2 + Fix A ✓)

🤖 Generated with Claude Code

[duncdrum]

more namespace fun, one comment in file.

[duncdrum]

dead code?

[line-o]

Looks like it, I would be in favour of removing this.

[line-o]

Why is this method named _peek_DocumentBuilder?
This looks like a standard getter to me

[line-o]

Do these extra checks have a significant impact on element construction performance even if copy-namespaces no-inherit is not set?

[line-o]

I would prefer to populate innerBuilder lazily on first access instead of always regardless if it will ever be used.
Then it also becomes clear that only its document is needed to check for preExisting nodes.

[joewiz]

[This response was co-authored with Claude Code. -Joe]

Good catches, addressed in 2d9e248:

• Removed the stale commented-out block in ElementConstructor (dead since the in-scope namespace API was modernized)
• Renamed peekDocumentBuilder() → getCurrentDocumentBuilder() in XQueryContext, ModuleContext, and the single call site in EnclosedExpr
• Made innerBuilder capture lazy: moved noInherit computation before the try block so getCurrentDocumentBuilder() is only called when copy-namespaces no-inherit is active — the hot path (default inherit mode) now has zero overhead from this method

On the performance question: with the lazy fix, the overhead in the default (inherit) case is exactly one context.inheritNamespaces() call per EnclosedExpr evaluation — just a field read on XQueryContext. The per-item check at line 168 (noInherit && ancestorNS != null) short-circuits immediately on the first boolean. I don't think this warrants a microbenchmark, but happy to add one if you'd like to see the numbers.