[tae898]

Row N-09 of the bindings backlog: on the native wheel, a record id was pasted into the SQL text, so each traversal step and each lookup_by_rid was a new statement text. The id now goes over as a parameter (the RID the codec already tags as R).

Statement cache finding: the claim holds. The engine's parsed-statement cache is StatementCache over SQLLiteralParameterizer/LiteralParameterizer, a SegmentedLRUCache<String, Lookup> keyed by the statement text (LiteralParameterizer.java:65, :134), sized by arcadedb.sqlStatementCache, default 300 (GlobalConfiguration.java:765). The literal extraction that lets texts share an entry only takes integer, floating point and non-empty string tokens (SQLLiteralParameterizer.java:112, isCandidate) and keeps every literal in a FROM target (class comment and classify), so SELECT FROM #1:0 and SELECT FROM #1:1 are two entries. The native image reaches the same path: ArcadeNative.java:258-262 calls db.query and db.command, which end in SQLQueryEngine.parseParameterized (SQLQueryEngine.java:285). Checked on ArcadeData/arcadedb main at 26f0bf8. So 25 records meant 25 texts, and a loop over more than 300 records evicts the statements the application re-runs.

What changed (Python only, no image rebuild): NativeDatabase.lookup_by_rid (SELECT FROM ?), and in records.py the Vertex.get_out_edges/get_in_edges/get_both_edges, Vertex.new_edge, Edge.get_in/get_out, save() (UPDATE :rid MERGE :row) and delete() (DELETE FROM :rid). The id validation and the error messages are unchanged, and a text without the leading # is still accepted as before. Docs: the Parameters section of guide/core/queries.md and lookup_by_rid in api/database.md.

Untimed counts (test_native_statement_texts.py, 25 records): lookup_by_rid gave 25 distinct texts before and gives 1 after (the test failed on exactly this before the change); get_out_edges("Knows"), get_in_edges() and get_both_edges() over the 24 linked vertices are 72 calls and now 3 texts, Edge.get_in/get_out over 24 edges are 48 calls and now 2 texts, save() and delete() are one text each, and new_edge is two texts (with and without CONTENT) per label instead of one per pair of records. Before the change every one of these carried the record id in its text (the second test failed on CREATE EDGE Likes FROM #1:0 TO #1:1). No timing was run.

Local checks, all against the image of native-lib/nnext (the engine of the current native-preview) with ARCADEDB_BACKEND=native: the two new tests fail before the change and pass after; test_native_statement_texts.py, test_graph_api.py, test_graph.py, test_core.py, test_resultset.py, test_graph_batch.py, test_schema.py, test_cypher.py, test_parameter_binding.py, test_restore_sql.py: 228 passed, 26 skipped (the NOTYET skips), 0 failed. The new behaviour tests in test_graph_api.py (traversal over a 12-vertex chain, save and delete by id, new_edge with and without properties, the refusals of lookup_by_rid) pass on the JPype wheel too: test_graph.py, test_core.py, test_graph_api.py, 64 passed. Pre-commit (black, isort) ran and its changes are in the commit.

Overlap with the open native PRs, checked with git merge-tree against each head: independent of #520, #521, #523, #525 (which also edits records.py, database.py and codec.py but in other hunks) and #526. Two adjacent-line conflicts, no semantic overlap: #514 in the import block of _native/database.py (it adds from .bulk import insert_rows_columnar next to my from .codec import RID) and #522 in tests/conftest.py (both add a collect_ignore.append line at the same place). Whichever merges second needs a trivial resolution; I did not stack on any of them.

🤖 Generated with Claude Code

[tae898]

Closing here: the bindings moved to ArcadeData/arcadedb-python. This change has no counterpart there yet and no queued A/B. If it is wanted, it goes there as its own PR: it is now a line in the tracking issue ArcadeData/arcadedb-python#19. The branch stays.