fix(parser): handle N3 verb edge cases

[jeswr]

Summary

• keep the N3 has, is, and of verbs from preempting longer valid prefixed names, including across stream chunks
• scope inverse-predicate state to the active predicate-object list: preserve it across commas and reset it across semicolons and annotation boundaries
• apply inverse direction consistently when closing blank-node property lists and formulas, and when constructing triple terms and reified triples
• add regression coverage for the cases identified in the Copilot review of #694

This is a follow-up to #694.

Verification

• 6,876 tests pass
• 100% statement, branch, function, and line coverage
• ESLint passes
• Node, IIFE browser, and ESM browser builds pass

[Copilot]

🟢 Approval recommended

The changes are cohesive, direction handling is consistently applied via a single quad creation path, and the added tests directly cover the reported edge cases and nesting/punctuation boundaries.

Pull request overview

This PR tightens N3 parsing/tokenization around the has, is, and of verbs to avoid keyword preemption of longer prefixed names (including across stream chunks) and to correctly scope/apply inverse-predicate state across punctuation and nested constructs (blank node property lists, formulas, triple terms, and reified triples), with regression tests added for the edge cases.

Changes:

• Add lexer logic to prefer longer prefixed names over N3 verb keywords when the verb is a prefix-start (including split chunk scenarios), while still recognizing numeric-boundary cases like has1.
• Refine parser inverse-predicate state handling to persist across commas, reset across semicolons/annotation boundaries, and apply inversion consistently when closing nested scopes and constructing triple terms / reified triples.
• Add regression coverage for inverse-predicate scoping and verb/prefix tokenization edge cases.

File summaries

File	Description
src/N3Lexer.js	Adds _matchN3Verb to avoid verb keyword tokenization when a longer prefixed name is possible, including across stream chunks.
src/N3Parser.js	Centralizes direction-aware quad creation/emission and resets/scopes inverse-predicate state across punctuation and nested constructs.
test/N3Lexer-test.js	Adds lexer regression tests for numeric boundaries and keyword-like prefixes (including chunk-splitting).
test/N3Parser-test.js	Adds parser regression tests for inverse scoping across , / ;, blank node property lists, formulas, triple terms, reification, and annotations.

Review details

• Files reviewed: 4/4 changed files
• Comments generated: 0
• Review effort level: Lite

---

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

[jeswr]

Performance follow-up for the inverse-handling concern:

I benchmarked main (ffb71f9) against this branch with Node 24.19.0. The harness alternated base/candidate order, used four fresh-process pairs with 20 samples per process, and included A/A (main versus main) controls. Parser workloads streamed 30,000 quads per sample; the isolated _emit workload created 2,000,000 quads per sample.

The initial implementation routed every _emit through _createQuad and showed a 5–8% paired-median slowdown in isolated emission. Commit 99d1e11 now installs the direction-aware _createQuad and _emit implementations only on N3 parser instances. For non-N3 formats, ordinary _emit remains the same direct factory call as main; blank-node and formula tails use a four-argument alias to that same method. This centralizes inverse swapping without putting inverse checks or a fifth argument on non-N3 emission paths.

Latest results (negative values are faster):

Workload	Aggregate median	Paired median	A/A paired control
Isolated _emit	-1.9%	-0.7%	+0.1%
Turtle	-4.3%	-2.9%	+4.9%
N-Triples	+0.7%	+3.4%	+2.3%
Turtle with blank-node property lists	-4.8%	-4.7%	+3.6%
Turtle with RDF 1.2 triple terms	-0.1%	-2.2%	—

There is no measured non-N3 regression beyond the process/JIT/GC noise visible in the A/A controls. The only positive aggregate result is N-Triples at +0.7%; its paired result is 1.1 percentage points above its +2.3% A/A control, while its ordinary emission code is source-identical to main.

Verification on the revised branch: 6,876 tests pass with 100% statement/branch/function/line coverage; ESLint and the Node, IIFE, and ESM builds pass.

[github-actions]

🎉 This PR is included in version 2.7.3 🎉

The release is available on:

• npm package (@latest dist-tag)
• GitHub release

Your semantic-release bot 📦🚀