From bcae47ee791de99aa14a0dee1e92f460913154d6 Mon Sep 17 00:00:00 2001
From: "Joshua D. Drake" <jd@commandprompt.com>
Date: Wed, 14 Oct 2026 15:32:28 +0000
Subject: [PATCH 1/4] test: arrow_import's argument and unique-violation arms
 assert their SQLSTATE, and a failure keeps its name (#1341)

expect_error accepted any error, and recorded a failure as "<label> (expected
error)", so a red never reached the name the ledger keeps. The three arms now
compare 42809, 42804 and 23505 under their own names.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VjTxfQJj5pnii9wzmPP9Ho
---
 test/arrow_import.sh | 23 ++++++++++-------------
 1 file changed, 10 insertions(+), 13 deletions(-)

diff --git a/test/arrow_import.sh b/test/arrow_import.sh
index 33601c7f0..3e0bc0e3c 100755
--- a/test/arrow_import.sh
+++ b/test/arrow_import.sh
@@ -114,15 +114,6 @@ fifo_release() { exec 9<>"$1" 2>/dev/null; exec 9>&- 2>/dev/null; }
 have_pyarrow=1
 python3 -c 'import pyarrow' 2>/dev/null || have_pyarrow=0
 
-expect_error() {
-	local label="$1" sql="$2"
-	if psql_run "$sql" >/dev/null 2>&1; then
-		check "$label (expected error)" "succeeded" "error"
-	else
-		check "$label" "error" "error"
-	fi
-}
-
 ARROW="$PGC_WORKDIR/rt.arrows"
 
 # --- 1. round trip through pgColumnar's own writer ---------------------------
@@ -332,10 +323,16 @@ fi
 
 # --- 4. error cases ---------------------------------------------------------
 echo "-- argument validation"
+# BY SQLSTATE (#1341). These accepted any error at all, and a check that failed
+# was recorded under a different name, "<label> (expected error)", so its red
+# never reached the name the ledger keeps. 42809 is wrong_object_type, 42804
+# datatype_mismatch; an empty state is a success.
 psql_run "CREATE TABLE ri_heap (a bigint, b float8, c text) USING heap;"
-expect_error "reject non-columnar target" "SELECT pgcolumnar.import_arrow('ri_heap', '$ARROW');"
+check "reject non-columnar target" \
+	"$(sqlstate_or_hang "SELECT pgcolumnar.import_arrow('ri_heap', '$ARROW')")" "42809"
 psql_run "CREATE TABLE ri_wrong (a int) USING pgcolumnar;"
-expect_error "reject column-count mismatch" "SELECT pgcolumnar.import_arrow('ri_wrong', '$ARROW');"
+check "reject column-count mismatch" \
+	"$(sqlstate_or_hang "SELECT pgcolumnar.import_arrow('ri_wrong', '$ARROW')")" "42804"
 
 if [ "$have_pyarrow" = 1 ]; then
 	DICTF="$PGC_WORKDIR/dict.arrows"
@@ -790,8 +787,8 @@ check "import into an indexed table: a point lookup finds its row" \
 psql_run "CREATE TABLE ix_uq (id int, v text) USING pgcolumnar;"
 psql_run "CREATE UNIQUE INDEX ix_uq_id ON ix_uq (id);"
 psql_run "SELECT pgcolumnar.import_arrow('ix_uq', '$IXFILE');"
-expect_error "importing the same keys twice raises unique_violation" \
-	"SELECT pgcolumnar.import_arrow('ix_uq', '$IXFILE');"
+check "importing the same keys twice raises unique_violation" \
+	"$(sqlstate_or_hang "SELECT pgcolumnar.import_arrow('ix_uq', '$IXFILE')")" "23505"
 check "the failed second import left the row count alone" \
 	"$(q 'SELECT count(*) FROM ix_uq;')" "3000"
 

From c9365013c551bf73cbc60415c2c28d7fdf0ad352 Mon Sep 17 00:00:00 2001
From: "Joshua D. Drake" <jd@commandprompt.com>
Date: Wed, 14 Oct 2026 15:32:28 +0000
Subject: [PATCH 2/4] test: arrow_import gets its pytest twin, every check by
 the shell's name (#1341)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VjTxfQJj5pnii9wzmPP9Ho
---
 test/pytest/TESTS.md                |  16 +
 test/pytest/expected_tests.txt      |   4 +-
 test/pytest/test_arrow_import.py    | 433 ++++++++++++++++++++++++++++
 test/pytest/test_compare_to_bash.py |   1 +
 4 files changed, 453 insertions(+), 1 deletion(-)
 create mode 100644 test/pytest/test_arrow_import.py

diff --git a/test/pytest/TESTS.md b/test/pytest/TESTS.md
index 6c1cbb09b..7816fec76 100644
--- a/test/pytest/TESTS.md
+++ b/test/pytest/TESTS.md
@@ -153,6 +153,7 @@ behaviour, the source of that number is named.
 - [105. test_import_export_privilege.py: import and export check the table's privilege, serial and parallel](#105-test_import_export_privilegepy-import-and-export-check-the-tables-privilege-serial-and-parallel)
 - [106. test_every_suite_job_installs_the_extensions.py: every job that runs the suites installs the extension packages](#106-test_every_suite_job_installs_the_extensionspy-every-job-that-runs-the-suites-installs-the-extension-packages)
 - [107. test_fdw_plan_privilege.py: planning a foreign table reads no server file for a role the scan would refuse](#107-test_fdw_plan_privilegepy-planning-a-foreign-table-reads-no-server-file-for-a-role-the-scan-would-refuse)
+- [108. test_arrow_import.py: import_arrow round trips, reads a foreign file, and refuses what it cannot decode](#108-test_arrow_importpy-import_arrow-round-trips-reads-a-foreign-file-and-refuses-what-it-cannot-decode)
 
 ## 1. How to read a test in here
 
@@ -6859,3 +6860,18 @@ The pytest twin of `test/fdw_plan_privilege.sh`. A planner estimate is visible t
 | test | what it holds |
 | --- | --- |
 | `test_fdw_planning_reads_no_server_file_for_an_unprivileged_role` | a role without `pg_read_server_files` gets the flat estimate from both wrappers and plans a missing Iceberg path the same as a real one, while the scan still refuses it (42501); a privileged role gets the manifest count and the size estimate (controls) |
+
+## 108. test_arrow_import.py: import_arrow round trips, reads a foreign file, and refuses what it cannot decode
+
+The pytest twin of `test/arrow_import.sh` (ported for #1341), every check by the shell's name. A 16-column table survives export and import exactly (a set hash on both sides), non-finite dates and NaN numerics arrive as null, and a non-columnar target and a column-count mismatch are refused by SQLSTATE. A pyarrow-written file imports with its values; nanosecond values are floored to the microsecond and the loss is reported once, counted in VALUES (two columns, so values and rows differ), never refused; a dictionary batch and a compressed body are refused by SQLSTATE and message; every Arrow date, time and timestamp carrier and unit decodes to the instant it holds or is refused 22008, a file whose declared type does not match its column is refused 42804, and nested units are honoured. An import maintains plain, unique (23505 on a second import) and partial indexes, and a FIFO is refused rather than hung. The four pyarrow tests decline by name without pyarrow, and fail where `PGC_REQUIRE` names it.
+
+### Every test
+
+| test | what it holds |
+| --- | --- |
+| `test_a_round_trip_and_the_documented_lossy_mapping` | export and import counts, the identical round trip, the non-finite mapping, and the two argument refusals |
+| `test_a_file_pyarrow_wrote` | a foreign file's row count and values |
+| `test_nanosecond_narrowing_is_reported_never_refused` | every row kept, the loss counted in values (8, then 12 through time64 and a list), no report on the two controls, and the floor |
+| `test_the_refusals_and_every_temporal_carrier` | the dictionary and compressed refusals, every carrier and unit, the 22008 guards, the 42804 type matches with their controls, an empty target after a refusal, and nested units |
+| `test_an_import_maintains_the_tables_indexes` | the plan really uses the index, index and sequential scans agree, the unique violation and its row count, and the partial index |
+| `test_a_fifo_is_refused_not_a_hang` | XX001 within the wall clock, and the backend still up |
diff --git a/test/pytest/expected_tests.txt b/test/pytest/expected_tests.txt
index a78088982..0376bc038 100644
--- a/test/pytest/expected_tests.txt
+++ b/test/pytest/expected_tests.txt
@@ -647,4 +647,6 @@ guard_tests 433
 #
 # cluster_tests 542 -> 543 for test_fdw_plan_privilege.py, one test, the twin of
 # test/fdw_plan_privilege.sh, by collection over the cluster files on this branch.
-cluster_tests 543
+# cluster_tests 543 -> 549 for #1341's test_arrow_import.py, the twin of arrow_import.sh:
+# six tests, by collection over the cluster files on this tree.
+cluster_tests 549
diff --git a/test/pytest/test_arrow_import.py b/test/pytest/test_arrow_import.py
new file mode 100644
index 000000000..e203455ac
--- /dev/null
+++ b/test/pytest/test_arrow_import.py
@@ -0,0 +1,433 @@
+"""pgcolumnar.import_arrow: the pytest twin of `test/arrow_import.sh` (#1341). Every
+check carries the shell's name.
+
+The reverse of export_arrow: a round trip through our own writer against a
+differential oracle, files a foreign producer (pyarrow) wrote, the documented lossy
+mappings, and the refusals. The four tests that need a pyarrow-written file decline
+by name when pyarrow is absent, as their first outcome, and fail where PGC_REQUIRE
+names it; the four that need no pyarrow always run, which is the shell's #1159
+lesson (a suite that runs fewer checks reads exactly like one that has fewer).
+
+Refusals are asserted by SQLSTATE, and where the shell does, by message too. Every
+file lives in `server_dir`, because the backend opens it.
+"""
+import decimal
+import os
+import re
+import struct
+import threading
+
+import pytest
+
+
+def _one(conn, sql):
+    with conn.cursor() as cur:
+        cur.execute(sql)
+        row = cur.fetchone()
+    return row[0] if row else None
+
+
+def _run(conn, *sqls):
+    with conn.cursor() as cur:
+        for s in sqls:
+            cur.execute(s)
+
+
+def _state(conn, sql):
+    """-> the SQLSTATE `sql` raised, or '00000'."""
+    import psycopg
+    try:
+        _run(conn, sql)
+        return "00000"
+    except psycopg.Error as e:
+        return e.sqlstate or "?"
+
+
+def _refusal(conn, sql):
+    """-> 'SQLSTATE message' of the error `sql` raised, or 'ok'."""
+    import psycopg
+    try:
+        _run(conn, sql)
+        return "ok"
+    except psycopg.Error as e:
+        return f"{e.sqlstate} {e.diag.message_primary}"
+
+
+def _notices(conn, sql):
+    """-> the messages the server sent while running `sql`, errors included."""
+    import psycopg
+    got = []
+
+    def keep(diag):
+        got.append(diag.message_primary or "")
+
+    conn.add_notice_handler(keep)
+    try:
+        _run(conn, sql)
+    except psycopg.Error as e:
+        got.append(e.diag.message_primary or "")
+    finally:
+        conn.remove_notice_handler(keep)
+    return got
+
+
+def _set_hash(conn, query):
+    return _one(conn, "SELECT coalesce(md5(string_agg(t, chr(10) ORDER BY t)), 'EMPTY')"
+                      f" FROM (SELECT _row::text AS t FROM ({query}) _row) _s")
+
+
+def _write_stream(pa, path, table, options=None):
+    import pyarrow.ipc as ipc
+    with ipc.new_stream(pa.OSFile(str(path), "wb"), table.schema, options=options) as w:
+        w.write_table(table)
+
+
+def test_a_round_trip_and_the_documented_lossy_mapping(pgc_conn, server_dir, expect):
+    conn = pgc_conn
+    arrow = server_dir / "rt.arrows"
+    # 16 columns: the maximum export_arrow supports.
+    _run(conn,
+         "CREATE TABLE ri_src (id int, c int8, d float4, e float8, f bool, g text, h bytea, dt date,"
+         " tm time, ts timestamp, tz timestamptz, u uuid, num numeric(20,4), numun numeric, j json,"
+         " jb jsonb) USING pgcolumnar",
+         "INSERT INTO ri_src SELECT s, s::int8*7, (s::float4/3), (s::float8/9), (s%2=0), 'row'||s,"
+         " decode(lpad(to_hex(s),4,'0'),'hex'), DATE '2000-01-01' + s,"
+         " TIME '00:00:00' + (s || ' seconds')::interval,"
+         " TIMESTAMP '2001-01-01' + (s||' minutes')::interval,"
+         " TIMESTAMPTZ '2001-01-01 00:00:00+00' + (s||' minutes')::interval,"
+         " ('00000000-0000-0000-0000-' || lpad(to_hex(s),12,'0'))::uuid,"
+         " (s::numeric/100)::numeric(20,4), (s::numeric/7), json_build_object('n', s),"
+         " jsonb_build_object('n', s) FROM generate_series(1, 3000) s",
+         # explicit NULLs and float NaN/Inf, which round trip exactly
+         "INSERT INTO ri_src (id, d, e) VALUES (900001, NULL, NULL),"
+         " (900002, 'NaN'::float4, 'Infinity'::float8), (900003, '-Infinity'::float4, 'NaN'::float8)")
+    src_rows = _one(conn, "SELECT count(*) FROM ri_src")
+    expect.num(_one(conn, f"SELECT pgcolumnar.export_arrow('ri_src', '{arrow}')"), src_rows, "export rows")
+    _run(conn, "CREATE TABLE ri_dst (LIKE ri_src) USING pgcolumnar")
+    expect.num(_one(conn, f"SELECT pgcolumnar.import_arrow('ri_dst', '{arrow}')"), src_rows, "import rows")
+    expect.text(_set_hash(conn, "SELECT * FROM ri_dst"), _set_hash(conn, "SELECT * FROM ri_src"),
+                "round trip identical")
+
+    # non-finite date/timestamp and NaN numeric export as null
+    lossy = server_dir / "lossy.arrows"
+    _run(conn, "CREATE TABLE ri_nf (id int, dt date, ts timestamp, num numeric(10,2)) USING pgcolumnar",
+         "INSERT INTO ri_nf VALUES (1, 'infinity', '-infinity', 'NaN'), (2, '2020-01-01', '2020-01-01 00:00', 1.25)",
+         f"SELECT pgcolumnar.export_arrow('ri_nf', '{lossy}')",
+         "CREATE TABLE ri_nf2 (LIKE ri_nf) USING pgcolumnar",
+         f"SELECT pgcolumnar.import_arrow('ri_nf2', '{lossy}')")
+    expect.num(_one(conn, "SELECT count(*) FROM ri_nf2 WHERE id=1 AND dt IS NULL AND ts IS NULL AND num IS NULL"),
+               1, "non-finite imported as null")
+    expect.num(_one(conn, "SELECT count(*) FROM ri_nf2 WHERE id=2 AND dt='2020-01-01' AND num=1.25"), 1,
+               "finite row preserved")
+
+    # argument validation, by SQLSTATE: 42809 wrong_object_type, 42804 datatype_mismatch
+    _run(conn, "CREATE TABLE ri_heap (a bigint, b float8, c text) USING heap",
+         "CREATE TABLE ri_wrong (a int) USING pgcolumnar")
+    expect.text(_state(conn, f"SELECT pgcolumnar.import_arrow('ri_heap', '{arrow}')"), "42809",
+                "reject non-columnar target")
+    expect.text(_state(conn, f"SELECT pgcolumnar.import_arrow('ri_wrong', '{arrow}')"), "42804",
+                "reject column-count mismatch")
+
+
+def test_a_file_pyarrow_wrote(pgc_conn, server_dir, expect):
+    pa = expect.require_module("pyarrow", name="pyarrow import rows")
+    if pa is None:
+        return
+    conn = pgc_conn
+    pyf = server_dir / "foreign.arrows"
+    _write_stream(pa, pyf, pa.table({
+        "a": pa.array([1, 2, None, 4], pa.int64()),
+        "b": pa.array([1.5, None, 3.5, 4.5], pa.float64()),
+        "c": pa.array(["x", "y", "z", None], pa.string())}))
+    _run(conn, "CREATE TABLE ri_for (a bigint, b float8, c text) USING pgcolumnar")
+    expect.num(_one(conn, f"SELECT pgcolumnar.import_arrow('ri_for', '{pyf}')"), 4, "pyarrow import rows")
+    expect.text(_one(conn, "SELECT string_agg(coalesce(a::text,'-')||'/'||coalesce(b::text,'-')||'/'||"
+                           "coalesce(c,'-'), ',' ORDER BY a NULLS LAST) FROM ri_for"),
+                "1/1.5/x,2/-/y,4/4.5/-,-/3.5/z", "pyarrow values")
+
+
+def test_nanosecond_narrowing_is_reported_never_refused(pgc_conn, server_dir, expect):
+    pa = expect.require_module("pyarrow", name="a lossy nanosecond file imports every row")
+    if pa is None:
+        return
+    conn = pgc_conn
+    s2000, noon = 946684800, 12 * 3600
+
+    def ns(off):
+        return pa.array([(s2000 + i) * 10**9 + off for i in range(4)], pa.timestamp("ns"))
+
+    def us():
+        return pa.array([(s2000 + i) * 10**6 for i in range(4)], pa.timestamp("us"))
+
+    # TWO temporal columns over four rows, so VALUES that lose digits (8) differ from ROWS (4).
+    _write_stream(pa, server_dir / "ns_lossy.arrows", pa.table({"ts": ns(123), "ts2": ns(123)}))
+    _write_stream(pa, server_dir / "ns_exact.arrows", pa.table({"ts": ns(0), "ts2": ns(0)}))
+    _write_stream(pa, server_dir / "us_plain.arrows", pa.table({"ts": us(), "ts2": us()}))
+    # the time call site and the list recursion: 4 + 8 = 12 values, all 123ns past a boundary
+    tm = pa.array([(noon + i) * 10**9 + 123 for i in range(4)], pa.time64("ns"))
+    lst = pa.array([[(s2000 + i) * 10**9 + 123, (s2000 + i + 1) * 10**9 + 123] for i in range(4)],
+                   pa.list_(pa.timestamp("ns")))
+    _write_stream(pa, server_dir / "ns_deep.arrows", pa.table({"tm": tm, "lst": lst}))
+    for t in ("ri_nsl", "ri_nse", "ri_usp"):
+        _run(conn, f"CREATE TABLE {t} (ts timestamp, ts2 timestamp) USING pgcolumnar")
+    nsl = _notices(conn, f"SELECT pgcolumnar.import_arrow('ri_nsl', '{server_dir / 'ns_lossy.arrows'}')")
+    nse = _notices(conn, f"SELECT pgcolumnar.import_arrow('ri_nse', '{server_dir / 'ns_exact.arrows'}')")
+    usp = _notices(conn, f"SELECT pgcolumnar.import_arrow('ri_usp', '{server_dir / 'us_plain.arrows'}')")
+
+    expect.num(_one(conn, "SELECT count(*) FROM ri_nsl"), 4, "a lossy nanosecond file imports every row")
+    expect.num(_one(conn, "SELECT count(*) FROM ri_nse"), 4, "so does a microsecond-exact nanosecond file")
+    expect.num(_one(conn, "SELECT count(*) FROM ri_usp"), 4, "and a microsecond file")
+    expect.num(sum("columnar.import_arrow: 8 values lost sub-microsecond precision" in m for m in nsl), 1,
+               "the narrowing reports the number of VALUES that lost digits, not rows")
+    expect.num(sum("import_arrow: 4 values lost" in m for m in nsl), 0, "and does not report the row count instead")
+
+    _run(conn, "CREATE TABLE ri_nsd (tm time, lst timestamp[]) USING pgcolumnar")
+    nsd = _notices(conn, f"SELECT pgcolumnar.import_arrow('ri_nsd', '{server_dir / 'ns_deep.arrows'}')")
+    expect.num(sum("columnar.import_arrow: 12 values lost sub-microsecond precision" in m for m in nsd), 1,
+               "a time64 column and a nested list are counted too, not just a top-level timestamp")
+    expect.num(_one(conn, "SELECT count(*) FROM ri_nsd"), 4, "and that file imports every row as well")
+
+    expect.num(sum("sub-microsecond" in m.lower() for m in nse), 0,
+               "control: an ns file on microsecond boundaries reports nothing")
+    expect.num(sum("sub-microsecond" in m.lower() for m in usp), 0, "control: a microsecond file reports nothing")
+    expect.text(_one(conn, "SELECT string_agg(ts::text, ',' ORDER BY ts) FROM ri_nsl"),
+                "2000-01-01 00:00:00,2000-01-01 00:00:01,2000-01-01 00:00:02,2000-01-01 00:00:03",
+                "the narrowed values are floored to the microsecond")
+
+
+def _temporal_fixtures(pa, out):
+    """The shell's fixture files, written the same ways: raw-byte packing for the
+    temporal carriers and pa.array for the #881 parameter cases."""
+    s2000, noon = 946684800, 12 * 3600
+    cases = {
+        "tu_date32": (pa.date32(), s2000 // 86400), "tu_date64": (pa.date64(), s2000 * 1000),
+        "tu_ts_s": (pa.timestamp("s"), s2000), "tu_ts_ms": (pa.timestamp("ms"), s2000 * 1000),
+        "tu_ts_us": (pa.timestamp("us"), s2000 * 1000000), "tu_ts_ns": (pa.timestamp("ns"), s2000 * 1000000000),
+        "tu_t64_us": (pa.time64("us"), noon * 1000000), "tu_t64_ns": (pa.time64("ns"), noon * 1000000000),
+        "tu_t32_s": (pa.time32("s"), noon), "tu_t32_ms": (pa.time32("ms"), noon * 1000),
+        "tu_ts_s_ovf": (pa.timestamp("s"), 2**63 - 1), "tu_ts_ms_far": (pa.timestamp("ms"), -300000000000000),
+        "tu_d64_frac": (pa.date64(), 86400001), "tu_d64_neg": (pa.date64(), -1),
+        "tu_t64_ns_ovf": (pa.time64("ns"), 25 * 3600 * 10**9),
+        "tu_ts_ns_trunc": (pa.timestamp("ns"), s2000 * 10**9 + 1500),
+        "tu_t64_ns_neg": (pa.time64("ns"), -500), "tu_ts_ns_neg": (pa.timestamp("ns"), -1500),
+        "tu_d32_far": (pa.date32(), 2**31 - 1), "tu_d32_past": (pa.date32(), -2**31),
+        "tu_d64_far": (pa.date64(), 2**62), "tu_t64_plain": (pa.time64("us"), noon * 10**6),
+    }
+    for name, (typ, v) in cases.items():
+        raw = struct.pack("<i" if typ.bit_width == 32 else "<q", v)
+        arr = pa.Array.from_buffers(typ, 1, [None, pa.py_buffer(raw)])
+        _write_stream(pa, out / f"{name}.arrows", pa.table({"v": arr}))
+    extra = {
+        "tu_u64": pa.array([2**63 + 5], pa.uint64()), "tu_i32": pa.array([7], pa.int32()),
+        "tu_i64": pa.array([7], pa.int64()), "tu_f32": pa.array([1.5], pa.float32()),
+        "tu_f64": pa.array([1.5], pa.float64()),
+        "tu_d102": pa.array([decimal.Decimal("1.25")], pa.decimal128(10, 2)),
+        "tu_d204": pa.array([decimal.Decimal("1.25")], pa.decimal128(20, 4)),
+        "tu_fsb32": pa.array([b"0123456789abcdef0123456789abcdef"], pa.binary(32)),
+        "tu_fsb16": pa.array([b"0123456789abcdef"], pa.binary(16)),
+        "tu_int64_ts": pa.array([946684800 * 10**6], pa.int64()),
+        # three date32 rows: a 12-byte buffer a stride-8 decoder runs off the end of
+        "tu_d32_3": pa.array([10957, 10958, 10959], pa.date32()),
+    }
+    for name, arr in extra.items():
+        _write_stream(pa, out / f"{name}.arrows", pa.table({"v": arr}))
+    ns = s2000 * 10**9
+    _write_stream(pa, out / "tu_list_ns.arrows", pa.table({"v": pa.array([[ns]], pa.list_(pa.timestamp("ns")))}))
+    st = pa.array([{"a": 7, "b": ns}], pa.struct([("a", pa.int32()), ("b", pa.timestamp("ns"))]))
+    _write_stream(pa, out / "tu_struct_ns.arrows", pa.table({"v": st}))
+
+
+def test_the_refusals_and_every_temporal_carrier(pgc_conn, server_dir, expect):
+    pa = expect.require_module("pyarrow", name="reject dictionary-encoded file")
+    if pa is None:
+        return
+    import pyarrow.ipc as ipc
+    conn = pgc_conn
+
+    _write_stream(pa, server_dir / "dict.arrows", pa.table({"c": pa.array(["a", "b", "a", "b"]).dictionary_encode()}))
+    _run(conn, "CREATE TABLE ri_d (c text) USING pgcolumnar")
+    expect.text(_refusal(conn, f"SELECT pgcolumnar.import_arrow('ri_d', '{server_dir / 'dict.arrows'}')"),
+                "0A000 columnar.import_arrow does not support dictionary-encoded Arrow files",
+                "reject dictionary-encoded file")
+
+    codec = next((c for c in ("zstd", "lz4") if pa.Codec.is_available(c)), None)
+    zipf = server_dir / "compressed.arrows"
+    if codec:
+        _write_stream(pa, zipf, pa.table({"a": pa.array(range(1000), pa.int64())}),
+                      options=ipc.IpcWriteOptions(compression=codec))
+    _run(conn, "CREATE TABLE ri_z (a bigint) USING pgcolumnar")
+    expect.text("codec" if codec in ("zstd", "lz4") else f"none ({codec})", "codec",
+                "premise: this pyarrow writes a compressed body (zstd or lz4)")
+    expect.text(_refusal(conn, f"SELECT pgcolumnar.import_arrow('ri_z', '{zipf}')")
+                + f"|{_one(conn, 'SELECT count(*) FROM ri_z')}",
+                "0A000 columnar.import_arrow does not support compressed Arrow bodies|0",
+                "reject a compressed record-batch body")
+
+    _temporal_fixtures(pa, server_dir)
+
+    def tu_value(fixture, pgtype):
+        import psycopg
+        _run(conn, "DROP TABLE IF EXISTS tu_t", f"CREATE TABLE tu_t (v {pgtype}) USING pgcolumnar")
+        try:
+            _run(conn, f"SELECT pgcolumnar.import_arrow('tu_t', '{server_dir / (fixture + '.arrows')}')")
+        except psycopg.Error:
+            return "IMPORT-FAILED"
+        return _one(conn, "SELECT v::text FROM tu_t LIMIT 1") or ""
+
+    def tu_state(fixture, pgtype):
+        _run(conn, "DROP TABLE IF EXISTS tu_t", f"CREATE TABLE tu_t (v {pgtype}) USING pgcolumnar")
+        return _state(conn, f"SELECT pgcolumnar.import_arrow('tu_t', '{server_dir / (fixture + '.arrows')}')")
+
+    # PREMISES: our own exporter's units round-trip, or every value arm could pass
+    # on a reader that rejects everything.
+    expect.text(tu_value("tu_ts_us", "timestamp"), "2000-01-01 00:00:00",
+                "premise: the exporter's own timestamp unit still imports correctly")
+    expect.text(tu_value("tu_t64_us", "time"), "12:00:00", "premise: and its own time unit does too")
+    expect.text(tu_value("tu_date32", "date"), "2000-01-01",
+                "premise: and a date32 carrier, which shares Arrow's tag 8 with date64")
+    expect.text(tu_value("tu_date64", "date"), "2000-01-01", "a date64 carrier decodes to the date it holds (#864)")
+    expect.text(tu_value("tu_ts_s", "timestamp"), "2000-01-01 00:00:00",
+                "a timestamp in seconds decodes to the instant it holds (#865)")
+    expect.text(tu_value("tu_ts_ms", "timestamp"), "2000-01-01 00:00:00",
+                "a timestamp in milliseconds decodes to the instant it holds (#865)")
+    expect.text(tu_value("tu_ts_ns", "timestamp"), "2000-01-01 00:00:00",
+                "a timestamp in nanoseconds decodes to the instant it holds (#865)")
+    expect.text(tu_value("tu_t64_ns", "time"), "12:00:00", "a time64 in nanoseconds decodes to the time it holds (#865)")
+    expect.text(tu_value("tu_t32_s", "time"), "12:00:00", "a time32 in seconds decodes to the time it holds (#865)")
+    expect.text(tu_value("tu_t32_ms", "time"), "12:00:00",
+                "a time32 in milliseconds decodes to the time it holds (#865)")
+
+    # the guards the unit fix introduced; 22008 is datetime_field_overflow
+    expect.text(tu_state("tu_ts_us", "timestamp"), "00000",
+                "premise: a well-formed import reports 00000, so the probe reads success")
+    expect.text(tu_state("tu_ts_s_ovf", "timestamp"), "22008", "a second count that overflows on scaling is refused, not wrapped")
+    expect.text(tu_state("tu_ts_ms_far", "timestamp"), "22008", "a timestamp before PostgreSQL's range is refused")
+    expect.text(tu_state("tu_t64_ns_ovf", "time"), "22008", "a time past midnight is refused")
+    expect.text(tu_value("tu_d64_frac", "date"), "1970-01-02", "a date64 instant mid-day reports the day it falls in")
+    expect.text(tu_value("tu_d64_neg", "date"), "1969-12-31", "a date64 instant before the epoch floors rather than truncating")
+    expect.text(tu_value("tu_ts_ns_trunc", "timestamp"), "2000-01-01 00:00:00.000001",
+                "a nanosecond timestamp truncates to microseconds")
+    expect.text(tu_value("tu_ts_ns_neg", "timestamp"), "1969-12-31 23:59:59.999998",
+                "a nanosecond timestamp before the epoch floors rather than truncating")
+    expect.text(tu_state("tu_t64_ns_neg", "time"), "22008", "a negative nanosecond time is refused, not narrowed into midnight")
+    expect.text(tu_state("tu_d32_far", "date"), "22008", "a date32 past PostgreSQL's last date is refused")
+    expect.text(tu_state("tu_d32_past", "date"), "22008", "a date32 before PostgreSQL's first date is refused")
+    expect.text(tu_state("tu_d64_far", "date"), "22008", "a date64 past PostgreSQL's last date is refused")
+
+    # the file's declared type must match the target's
+    targets = ("bigint", "uuid", "time", "numeric(20,4)", "timestamp")
+    expect.num(len(targets), 5, "premise: the date32 sweep covers five target types")
+    for target in targets:
+        expect.text(tu_state("tu_d32_3", target), "42804",
+                    f"a date32 file is refused for a {target} column, not read past its buffer")
+    expect.text(tu_state("tu_t64_plain", "date"), "42804", "a time64 file is refused for a date column, not stored as year 687342")
+    expect.text(tu_state("tu_ts_us", "date"), "42804", "a timestamp file is refused for a date column")
+    expect.text(tu_state("tu_date64", "timestamp"), "42804", "a date64 file is refused for a timestamp column")
+
+    # the file's own Int/Float/Decimal parameters (#881)
+    expect.text(tu_state("tu_u64", "bigint"), "42804", "a uint64 file is refused for bigint (#881)")
+    expect.text(tu_state("tu_i64", "int"), "42804", "an int64 file is refused for a 4-byte int (#881)")
+    expect.text(tu_state("tu_i32", "bigint"), "42804", "an int32 file is refused for bigint (#881)")
+    expect.text(tu_state("tu_f32", "float8"), "42804", "a float32 file is refused for float8 (#881)")
+    expect.text(tu_state("tu_i64", "bigint"), "00000", "control: a matching int64 file imports into bigint")
+    expect.text(tu_state("tu_i32", "int"), "00000", "control: a matching int32 file imports into int")
+    expect.text(tu_state("tu_f64", "float8"), "00000", "control: a matching float64 file imports into float8")
+    expect.text(tu_state("tu_f32", "float4"), "00000", "control: a matching float32 file imports into float4")
+    expect.text(tu_state("tu_fsb32", "uuid"), "42804", "a wider fixed-size binary is refused for uuid (#881)")
+    expect.text(tu_state("tu_fsb16", "uuid"), "00000", "control: a 16-byte fixed-size binary imports into uuid")
+    expect.text(tu_state("tu_d102", "numeric(20,4)"), "42804", "a decimal at another scale is refused (#881)")
+    expect.text(tu_state("tu_d204", "numeric(20,4)"), "00000", "control: a matching decimal imports")
+    tu_state("tu_u64", "bigint")
+    expect.num(_one(conn, "SELECT count(*) FROM tu_t"), 0, "and a refused import leaves the target empty")
+    expect.text(tu_state("tu_int64_ts", "timestamp"), "00000",
+                "control: a non-temporal tag is not caught by the temporal refusal")
+
+    # the unit on a nested field
+    expect.text(tu_value("tu_list_ns", "timestamp[]"), '{"2000-01-01 00:00:00"}',
+                "a list element's nanosecond unit is honoured, not just the top level")
+    _run(conn, "DROP TYPE IF EXISTS tu_st CASCADE", "CREATE TYPE tu_st AS (a int, b timestamp)")
+    expect.text(tu_value("tu_struct_ns", "tu_st"), '(7,"2000-01-01 00:00:00")',
+                "a struct field's nanosecond unit is honoured at child index 1")
+
+
+def test_an_import_maintains_the_tables_indexes(pgc_conn, server_dir, expect):
+    """#153: table_tuple_insert writes the row and nothing else, so an importer must
+    maintain indexes itself. The two access paths are compared on one predicate."""
+    conn = pgc_conn
+    ixfile = server_dir / "ix_roundtrip.arrows"
+    _run(conn, "CREATE TABLE ix_src (id int, v text) USING pgcolumnar",
+         "INSERT INTO ix_src SELECT g, 'v' || g FROM generate_series(1, 3000) g",
+         f"SELECT pgcolumnar.export_arrow('ix_src', '{ixfile}')",
+         "CREATE TABLE ix_tgt (id int, v text) USING pgcolumnar",
+         "CREATE INDEX ix_tgt_id ON ix_tgt (id)",
+         f"SELECT pgcolumnar.import_arrow('ix_tgt', '{ixfile}')")
+    # enable_seqscan alone leaves the columnar custom scan cheapest, so the index
+    # was never consulted; enable_custom_scan takes that path out, and the plan is asserted.
+    idx_on = ("SET enable_seqscan = off", "SET pgcolumnar.enable_custom_scan = off")
+    idx_off = ("RESET enable_seqscan", "RESET pgcolumnar.enable_custom_scan")
+    seq_on = ("SET enable_indexscan = off", "SET enable_bitmapscan = off", "SET enable_indexonlyscan = off")
+    seq_off = ("RESET enable_indexscan", "RESET enable_bitmapscan", "RESET enable_indexonlyscan")
+
+    def under(on, off, sql):
+        _run(conn, *on)
+        try:
+            with conn.cursor() as cur:
+                cur.execute(sql)
+                return cur.fetchall()
+        finally:
+            _run(conn, *off)
+
+    q100 = "SELECT count(*) FROM ix_tgt WHERE id BETWEEN 100 AND 199"
+    plan = "\n".join(r[0] for r in under(idx_on, idx_off, f"EXPLAIN (COSTS OFF) {q100}"))
+    expect.text("yes" if re.search(r"index.*scan", plan, re.I) else "no", "yes",
+                "the index-scan check really uses the index")
+    idx_n = under(idx_on, idx_off, q100)[0][0]
+    seq_n = under(seq_on, seq_off, q100)[0][0]
+    expect.num(idx_n, seq_n, "import into an indexed table: both scans agree")
+    expect.num(seq_n, 100, "import into an indexed table: the rows are there")
+    expect.num(under(idx_on, idx_off, "SELECT count(*) FROM ix_tgt WHERE id = 2222")[0][0], 1,
+               "import into an indexed table: a point lookup finds its row")
+
+    _run(conn, "CREATE TABLE ix_uq (id int, v text) USING pgcolumnar",
+         "CREATE UNIQUE INDEX ix_uq_id ON ix_uq (id)",
+         f"SELECT pgcolumnar.import_arrow('ix_uq', '{ixfile}')")
+    expect.text(_state(conn, f"SELECT pgcolumnar.import_arrow('ix_uq', '{ixfile}')"), "23505",
+                "importing the same keys twice raises unique_violation")
+    expect.num(_one(conn, "SELECT count(*) FROM ix_uq"), 3000, "the failed second import left the row count alone")
+
+    _run(conn, "CREATE TABLE ix_part (id int, v text) USING pgcolumnar",
+         "CREATE INDEX ix_part_id ON ix_part (id) WHERE id <= 500",
+         f"SELECT pgcolumnar.import_arrow('ix_part', '{ixfile}')")
+    expect.num(under(idx_on, idx_off, "SELECT count(*) FROM ix_part WHERE id BETWEEN 100 AND 199")[0][0], 100,
+               "partial index covers its rows after import")
+
+
+def test_a_fifo_is_refused_not_a_hang(pgc_conn, pgc_cluster, server_dir, expect):
+    """#686/#644: fopen("rb") on a FIFO blocks in open(2), and SA_RESTART carries the
+    block through a cancel, so only a wall clock can tell a hang. The call runs on a
+    connection of its own in a thread; past 5 s it is a HANG, and the FIFO is then
+    opened here so the backend can return."""
+    import psycopg
+    conn = pgc_conn
+    schema = _one(conn, "SELECT current_schema()")
+    _run(conn, "CREATE TABLE ia_fifo (a int) USING pgcolumnar")
+    fifo = server_dir / "ia.fifo"
+    os.mkfifo(fifo)
+    os.chmod(fifo, 0o666)
+    got = {}
+
+    def call():
+        with psycopg.connect(pgc_cluster.dsn(), autocommit=True) as c:
+            c.execute(f'SET search_path TO "{schema}", pgcolumnar, public')
+            got["state"] = _state(c, f"SELECT pgcolumnar.import_arrow('ia_fifo', '{fifo}')")
+
+    t = threading.Thread(target=call)
+    t.start()
+    t.join(5)
+    if t.is_alive():
+        fd = os.open(fifo, os.O_RDWR | os.O_NONBLOCK)
+        os.close(fd)
+        t.join(30)
+        got["state"] = "HANG"
+    expect.text(got.get("state", "no result"), "XX001", "import_arrow on a FIFO is refused, not a hang (XX001)")
+    expect.num(_one(conn, "SELECT 1"), 1, "backend still up after the import_arrow FIFO refusal")
diff --git a/test/pytest/test_compare_to_bash.py b/test/pytest/test_compare_to_bash.py
index 6948ebdc7..84cf640a9 100644
--- a/test/pytest/test_compare_to_bash.py
+++ b/test/pytest/test_compare_to_bash.py
@@ -91,6 +91,7 @@
             "analyze_function",
             "analyze_reltuples",
             "approx_aggregates",
+            "arrow_import",
             "autovacuum_expire",
             "base_scan_io",
             "catalog_delete_index",

From 6ea3e0682fa7452df9c8488068e0e27e8e713d56 Mon Sep 17 00:00:00 2001
From: "Joshua D. Drake" <jd@commandprompt.com>
Date: Wed, 14 Oct 2026 15:32:28 +0000
Subject: [PATCH 3/4] test: arrow_import's two argument arms are dated by their
 checks removed (#1341)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VjTxfQJj5pnii9wzmPP9Ho
---
 test/check_ledger.tsv        |  4 ++--
 test/check_ledger_budget.txt | 16 +++++++++++++++-
 2 files changed, 17 insertions(+), 3 deletions(-)

diff --git a/test/check_ledger.tsv b/test/check_ledger.tsv
index af03b6af9..d1833895f 100644
--- a/test/check_ledger.tsv
+++ b/test/check_ledger.tsv
@@ -119,9 +119,9 @@ arrow_import	arrow_import	premise: this pyarrow writes a compressed body (zstd o
 arrow_import	arrow_import	pyarrow import rows	15;16;17;18;19	never	-
 arrow_import	arrow_import	pyarrow values	15;16;17;18;19	never	-
 arrow_import	arrow_import	reject a compressed record-batch body	15;16;17;18;19	2026-10-14	import_arrow without its compressed-body refusal
-arrow_import	arrow_import	reject column-count mismatch	15;16;17;18;19	never	-
+arrow_import	arrow_import	reject column-count mismatch	15;16;17;18;19	2026-10-14	import_arrow without its column-count check
 arrow_import	arrow_import	reject dictionary-encoded file	15;16;17;18;19	2026-10-14	import_arrow without its dictionary-batch refusal
-arrow_import	arrow_import	reject non-columnar target	15;16;17;18;19	never	-
+arrow_import	arrow_import	reject non-columnar target	15;16;17;18;19	2026-10-14	import_arrow without its columnar-relation check
 arrow_import	arrow_import	round trip identical	15;16;17;18;19	never	-
 arrow_import	arrow_import	so does a microsecond-exact nanosecond file	15;16;17;18;19	never	-
 arrow_import	arrow_import	the failed second import left the row count alone	15;16;17;18;19	never	-
diff --git a/test/check_ledger_budget.txt b/test/check_ledger_budget.txt
index e1faa6895..e18ddaee9 100644
--- a/test/check_ledger_budget.txt
+++ b/test/check_ledger_budget.txt
@@ -2002,4 +2002,18 @@ suites_not_covered 242
 #     never         1925 + 8 - 3 = 1930
 #     dated          580 + 3     =  583
 #     1930 + 583 == 2513
-checks_never_observed_red 1930
+#
+# #1341, arrow_import's pytest twin, and its argument arms by SQLSTATE (OffgridwithJD):
+# 1930 -> 1928. No new rows. "reject non-columnar target" and "reject column-count
+# mismatch" accepted any error, and a failure was recorded as "<name> (expected
+# error)", so a red never reached the row. They now assert 42809 and 42804, and each is
+# dated by its check removed: without the relation check the import fails 42804
+# instead, which main's suite accepts; without the count check a 16-column file
+# imports into a 1-column table, which main's suite reports under the other name.
+# Five-major runs: 77/77 on 15-19, the twin 6 passed with pyarrow required.
+#
+#     ledger rows   2513     = 2513
+#     never         1930 - 2     = 1928
+#     dated          583 + 2     = 585
+#     1928 + 585 == 2513
+checks_never_observed_red 1928

From fadd320fe0df7d4d14c09e17bf6d4eb648d6ee9f Mon Sep 17 00:00:00 2001
From: "Joshua D. Drake" <jd@commandprompt.com>
Date: Wed, 14 Oct 2026 19:45:44 +0000
Subject: [PATCH 4/4] test: the arrow_import twin reaches ipc through the
 module require_module returns (#1395 review)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VjTxfQJj5pnii9wzmPP9Ho
---
 test/pytest/test_arrow_import.py | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)

diff --git a/test/pytest/test_arrow_import.py b/test/pytest/test_arrow_import.py
index e203455ac..907d53ca9 100644
--- a/test/pytest/test_arrow_import.py
+++ b/test/pytest/test_arrow_import.py
@@ -77,8 +77,7 @@ def _set_hash(conn, query):
 
 
 def _write_stream(pa, path, table, options=None):
-    import pyarrow.ipc as ipc
-    with ipc.new_stream(pa.OSFile(str(path), "wb"), table.schema, options=options) as w:
+    with pa.ipc.new_stream(pa.OSFile(str(path), "wb"), table.schema, options=options) as w:
         w.write_table(table)
 
 
@@ -241,7 +240,6 @@ def test_the_refusals_and_every_temporal_carrier(pgc_conn, server_dir, expect):
     pa = expect.require_module("pyarrow", name="reject dictionary-encoded file")
     if pa is None:
         return
-    import pyarrow.ipc as ipc
     conn = pgc_conn
 
     _write_stream(pa, server_dir / "dict.arrows", pa.table({"c": pa.array(["a", "b", "a", "b"]).dictionary_encode()}))
@@ -254,7 +252,7 @@ def test_the_refusals_and_every_temporal_carrier(pgc_conn, server_dir, expect):
     zipf = server_dir / "compressed.arrows"
     if codec:
         _write_stream(pa, zipf, pa.table({"a": pa.array(range(1000), pa.int64())}),
-                      options=ipc.IpcWriteOptions(compression=codec))
+                      options=pa.ipc.IpcWriteOptions(compression=codec))
     _run(conn, "CREATE TABLE ri_z (a bigint) USING pgcolumnar")
     expect.text("codec" if codec in ("zstd", "lz4") else f"none ({codec})", "codec",
                 "premise: this pyarrow writes a compressed body (zstd or lz4)")