From bacf7f87c5cc76f56de0d925cde2bbbcde800818 Mon Sep 17 00:00:00 2001
From: Andrew Lamb <andrew@nerdnetworks.org>
Date: Tue, 15 Apr 2025 14:13:35 -0400
Subject: [PATCH 1/2] Add slt tests for
 datafusion.execution.parquet.coerce_int96 setting

---
 .../sqllogictest/test_files/parquet.slt       | 76 +++++++++++++++++++
 1 file changed, 76 insertions(+)

diff --git a/datafusion/sqllogictest/test_files/parquet.slt b/datafusion/sqllogictest/test_files/parquet.slt
index 2970b2effb3e..5ff340bb8260 100644
--- a/datafusion/sqllogictest/test_files/parquet.slt
+++ b/datafusion/sqllogictest/test_files/parquet.slt
@@ -629,3 +629,79 @@ physical_plan
 
 statement ok
 drop table foo
+
+
+# Tests for int96 timestamps written by spark
+# See https://github.com/apache/datafusion/issues/9981
+
+statement ok
+CREATE EXTERNAL TABLE int96_from_spark
+STORED AS PARQUET
+LOCATION '../../parquet-testing/data/int96_from_spark.parquet';
+
+# by default the value is read as nanosecond precision
+query TTT
+describe int96_from_spark
+----
+a Timestamp(Nanosecond, None) YES
+
+# Note that the values are read as nanosecond precision
+query P
+select * from int96_from_spark
+----
+2024-01-01T20:34:56.123456
+2024-01-01T01:00:00
+1816-03-29T08:56:08.066277376
+2024-12-30T23:00:00
+NULL
+1815-11-08T16:01:01.191053312
+
+statement ok
+drop table int96_from_spark;
+
+# Enable coercion of int96 to microseconds
+statement ok
+set datafusion.execution.parquet.coerce_int96 = ms
+
+statement ok
+CREATE EXTERNAL TABLE int96_from_spark
+STORED AS PARQUET
+LOCATION '../../parquet-testing/data/int96_from_spark.parquet';
+
+# The value should be read as MICROSECOND precision
+# see https://github.com/apache/datafusion/issues/15721
+query TTT
+describe int96_from_spark
+----
+a Timestamp(Nanosecond, None) YES
+
+# Per https://github.com/apache/parquet-testing/blob/6e851ddd768d6af741c7b15dc594874399fc3cff/data/int96_from_spark.md?plain=1#L37
+# these values should be
+#
+# Some("2024-01-01T12:34:56.123456"),
+# Some("2024-01-01T01:00:00Z"),
+# Some("9999-12-31T01:00:00-02:00"),
+# Some("2024-12-31T01:00:00+02:00"),
+# None,
+# Some("290000-12-31T01:00:00+02:00"))
+#
+# However, printing the large dates (9999-12-31 and 290000-12-31) is not supported by
+# arrow yet
+#
+# See https://github.com/apache/arrow-rs/issues/7287
+query P
+select * from int96_from_spark
+----
+2024-01-01T20:34:56.123
+2024-01-01T01:00:00
+NULL
+2024-12-30T23:00:00
+NULL
+NULL
+
+# Cleanup / reset default setting
+statement ok
+drop table int96_from_spark;
+
+statement ok
+set datafusion.execution.parquet.coerce_int96 = ns

From 6508ec350dc425ea22394fe002fe70a0ec40dd4c Mon Sep 17 00:00:00 2001
From: Andrew Lamb <andrew@nerdnetworks.org>
Date: Mon, 28 Apr 2025 11:51:36 -0400
Subject: [PATCH 2/2] tweak

---
 datafusion/sqllogictest/test_files/parquet.slt | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)

diff --git a/datafusion/sqllogictest/test_files/parquet.slt b/datafusion/sqllogictest/test_files/parquet.slt
index 42a7382935e3..abc6fdab3c8a 100644
--- a/datafusion/sqllogictest/test_files/parquet.slt
+++ b/datafusion/sqllogictest/test_files/parquet.slt
@@ -668,10 +668,9 @@ CREATE EXTERNAL TABLE int96_from_spark
 STORED AS PARQUET
 LOCATION '../../parquet-testing/data/int96_from_spark.parquet';
 
-# The value should be read as MICROSECOND precision
-# see https://github.com/apache/datafusion/issues/15721
+# Print schema
 query TTT
-describe int96_from_spark
+describe int96_from_spark;
 ----
 a Timestamp(Millisecond, None) YES
 
@@ -704,4 +703,4 @@ statement ok
 drop table int96_from_spark;
 
 statement ok
-set datafusion.execution.parquet.coerce_int96 = ns
+set datafusion.execution.parquet.coerce_int96 = ns;