Table: append_python_compat Writer: Apache Paimon Python at commit 63ffd403648e1cf9848e32bb29a398c1b3b16ab1 Source: https://github.com/apache/paimon/tree/63ffd403648e1cf9848e32bb29a398c1b3b16ab1/paimon-python Python version: 3.11.15 Lance version: pylance 0.39.0 Arrow Python version: 19.0.1

This unpartitioned append-only table has no primary key. The Python Catalog and batch table-write/commit APIs wrote two rows per commit into two Lance files, with Avro manifests and two snapshots. All values were read back in Python after each commit; reordered projection was also verified after snapshot 2.

Schema (all fields nullable): id INT f_boolean BOOLEAN f_tinyint TINYINT f_smallint SMALLINT f_bigint BIGINT f_float FLOAT f_double DOUBLE f_string STRING f_binary BYTES f_date DATE f_time_0 TIME(0) f_time_3 TIME(3) f_ts_0 TIMESTAMP(0) f_ts_3 TIMESTAMP(3) f_ts_6 TIMESTAMP(6) f_ts_9 TIMESTAMP(9) f_decimal_1_0 DECIMAL(1, 0) f_decimal_18_2 DECIMAL(18, 2) f_decimal_19_2 DECIMAL(19, 2) f_decimal_38_18 DECIMAL(38, 18) f_array_int ARRAY f_array_array_int ARRAY<ARRAY> f_struct ROW<number INT, label STRING> f_vector VECTOR<FLOAT, 3>

Options: bucket = -1 file.format = lance manifest.format = avro metadata.stats-mode = none

Values below follow schema order; timestamps are UTC and binary values are hexadecimal. The second row has NULL in every field except id and the valid ROW parent with two NULL children.

Add: (1, true, -5, -1000, 10000000001, 1.25, -2.5, “value-1”, 0x0001027f, 1969-12-31, 00:00:00, 23:59:59.999, 1970-01-01 00:00:01, 1970-01-01 00:00:01.123, 1970-01-01 00:00:01.123456, 1970-01-01 00:00:01.123456789, -3, 1.25, 12345678901234567.89, 12345678901234567890.123456789012345678, [1, NULL, -1], [[1, NULL], NULL, []], (1, “row-1”), [1.0, -1.5, 0.25]) Add: (2, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, (NULL, NULL), NULL)

Commit - snapshot 1 (2 rows)

Add: (10, false, -5, -1000, 10000000010, 1.25, -2.5, "", 0x0001027f, 2024-10-04, 12:34:56, 00:00:01.123, 1970-01-01 00:00:01, 1970-01-01 00:00:01.123, 1970-01-01 00:00:01.123456, 1970-01-01 00:00:01.123456789, -3, 1.25, 12345678901234567.89, 12345678901234567890.123456789012345678, [], [[10, NULL], NULL, []], (10, “row-10”), [10.0, -1.5, 0.25]) Add: (11, true, -5, -1000, 10000000011, 1.25, -2.5, “value-11”, 0x0001027f, 2024-10-04, 23:59:59, 12:34:56.789, 1970-01-01 00:00:01, 1970-01-01 00:00:01.123, 1970-01-01 00:00:01.123456, 1970-01-01 00:00:01.123456789, -3, 1.25, 12345678901234567.89, 12345678901234567890.123456789012345678, [11, NULL, -11], [[11, NULL], NULL, []], (11, “row-11”), [11.0, -1.5, 0.25])

Commit - snapshot 2 (4 rows total)

Paimon Python maps every pyarrow time type to TIME(0), so f_time_3 was declared TIME(3) directly in the Paimon schema. Paimon Python writes both TIME columns to Lance as time32[ms] holding milliseconds of the day, with no field metadata; the declared precision comes from the table schema. Null and empty lists, null list elements, an empty string, binary zero bytes, null scalars, null ROW children and a null vector are included. ROW parents are always valid: this fixture does not claim null-parent ROW compatibility. MAP, local-zoned timestamps and other types absent from this schema are not covered. Statistics are disabled; this fixture makes no predicate-pruning claim.