[tae898]

Stacked on #514 (base native-columnar-strings, so the diff shows only this change; GitHub retargets it to native-preview when #514 merges). It touches _native/database.py::_insert_chunk, the function #514 edits, and one #514 test that asserted the old fallback.

On the native wheel, insert_many sent a chunk through INSERT INTO t CONTENT :row for each row (arcadedb_n_exec_many) as soon as one value made json.dumps raise: a datetime, date, Decimal, bytes, a numpy scalar, a pandas NA/NaT. #514 does not change that: columnar.pack refuses date-times and Decimals, and _insert_chunk then fell to the statement path (its test test_values_json_cannot_carry_still_go_row_by_row_in_one_crossing asserted exactly that for a datetime). Json.java already decodes the tagged values {"\u0001": [kind, payload]} (T, D, n, b), so no native-image change is needed (backlog rows N-04, J-04, J-05).

What changes

• insert_many writes those values as tagged values inside the arcadedb_n_insert_rows text (json_bulk_dumps_tagged, same function and semantics as the JVM wheel's PR). A chunk with such a value loses the columnar entry (columnar.pack still refuses it) but stays on the bulk entry.
• A row with a value the text still cannot carry (NaN and Infinity, an integer beyond 64 bits, a date-time or Decimal inside a list or dict, a non-str key, a lone surrogate, a datetime.time) goes to the statement on its own: runs of carriable rows go to the bulk entry (columnar first, then the JSON text), runs of uncarriable rows go to arcadedb_n_exec_many in one crossing, in the order given.
• insert_many(parallel=True) already writes every type as tagged JSON (extras.insert_parallel) and is unchanged.
• Native: columnar bulk insert by address for strings, nulls and vectors (no JSON text) #514's test is split in two: a datetime now stays on the JSON entry (and the columnar entry is not used), and the old assertion (statement path, one crossing, no INSERT_ROWS) now uses a datetime inside a list, which is still uncarriable.
• A quirk found on the way, not changed here: Json.java decodes any one-key map keyed by the tag marker, on the statement path and the bulk entry alike, so a user map {"\u0001": [...]} is read as a tag in both. The test pins that the two paths agree.

Stored values

tests/test_native_mixed_chunk.py (native backend only, collected through conftest.py like #514's file) stores 17 value kinds (naive and UTC+9 datetime, date, Decimal with and without exponent, bytes, bytearray, np.int64/int8/uint32/float64/float32/bool_, None) through the statement path (_command_many with INSERT ... CONTENT :row, what the chunk used to take) and through insert_many, in a declared type (DATETIME, DATE, DECIMAL, BINARY, LONG, INTEGER, DOUBLE, FLOAT, BOOLEAN properties) and a schemaless one, and compares the stored value and its Python type. A spy on lib.call proves the entry (arcadedb_n_insert_rows once, arcadedb_n_exec_many never). It also covers pandas.Timestamp (nanoseconds kept, as the statement path keeps them; aware), pandas.NA/NaT and np.datetime64 (the statement refuses these; the bulk entry stores null / a naive datetime), a 10,000-row chunk with one datetime, one uncarriable row among 1,000 (one statement run, two bulk calls, stored rows equal the statement path's), insert order around statement runs, commit_every, and rollback. 14 of the 17 tests failed before the change.

Untimed 1M-row figure

One run each (not through ab.py, so no claim), fresh process, 1,000,000 rows of six properties with a Python datetime per row, row building inside the timed call, #514's native library, through quiet_build: insert_many took 15.2 s before (100 arcadedb_n_exec_many crossings, INSERT ... CONTENT per row) and 11.2 s after (100 arcadedb_n_insert_rows crossings). The same rows without the ts column take 6.1 s on the columnar entry (arcadedb_n_insert_colp), so a date-time column still costs the columnar entry; carrying date-times in columnar.pack (an epoch-microsecond int64 column) is the next step and is not in this PR.

Local checks

• Against the Native: columnar bulk insert by address for strings, nulls and vectors (no JSON text) #514 native library (native-lib/colstr, built from the commit whose Java sources equal Native: columnar bulk insert by address for strings, nulls and vectors (no JSON text) #514's head): test_native_mixed_chunk.py 17 passed; test_native_columnar_insert.py, test_native_columnar_pack.py, test_bulk_insert.py, test_bulk_json_paths.py, test_insert_columns.py, test_dataframe_ingest.py, test_conversion_fixes.py, test_core.py, test_graph_batch.py, test_async_executor.py, test_check_test_skips.py pass (267 passed, 20 skipped, with the one updated Native: columnar bulk insert by address for strings, nulls and vectors (no JSON text) #514 test)
• black, isort and the repository pre-commit hooks pass
• No native-image rebuild was needed (Python only); the full suite and the wheel build are left to CI

Docs changed: the insert_many docstring in _native/database.py, docs/api/database.md, docs/development/testing.md.

🤖 Generated with Claude Code

[tae898]

Closing here: the bindings moved to ArcadeData/arcadedb-python. The branch stays for the queued A/B insert-many-temporal-ab-laptop-20261025; if it passes, the change is ported there.