Add support for Utf8View, Boolean, Date32/64, int32/64 for writing hive style partitions

datafusion/core/src/datasource/file_format/write/demux.rs

@@ -31,7 +31,11 @@ use arrow_array::builder::UInt64Builder;
 use arrow_array::cast::AsArray;
 use arrow_array::{downcast_dictionary_array, RecordBatch, StringArray, StructArray};
 use arrow_schema::{DataType, Schema};
-use datafusion_common::cast::as_string_array;
+use chrono::NaiveDate;
+use datafusion_common::cast::{
+    as_boolean_array, as_date32_array, as_date64_array, as_int32_array, as_int64_array,
+    as_string_array, as_string_view_array,
+};
 use datafusion_common::{exec_datafusion_err, DataFusionError};
 use datafusion_common_runtime::SpawnedTask;
 use datafusion_execution::TaskContext;
@@ -320,9 +324,11 @@ async fn hive_style_partitions_demuxer(
 fn compute_partition_keys_by_row<'a>(
     rb: &'a RecordBatch,
     partition_by: &'a [(String, DataType)],
-) -> Result<Vec<Vec<&'a str>>> {
+) -> Result<Vec<Vec<String>>> {
     let mut all_partition_values = vec![];
 
+    const EPOCH_DAYS_FROM_CE: i32 = 719_163;
+
     // For the purposes of writing partitioned data, we can rely on schema inference
     // to determine the type of the partition cols in order to provide a more ergonomic
     // UI which does not require specifying DataTypes manually. So, we ignore the
@@ -342,7 +348,55 @@ fn compute_partition_keys_by_row<'a>(
             DataType::Utf8 => {
                 let array = as_string_array(col_array)?;
                 for i in 0..rb.num_rows() {
-                    partition_values.push(array.value(i));
+                    partition_values.push(array.value(i).to_string());
+                }
+            }
+            DataType::Utf8View => {
+                let array = as_string_view_array(col_array)?;
+                for i in 0..rb.num_rows() {
+                    partition_values.push(array.value(i).to_string());
+                }
+            }
+            DataType::Boolean => {
+                let array = as_boolean_array(col_array)?;
+                for i in 0..rb.num_rows() {
+                    partition_values.push(array.value(i).to_string());
+                }
+            }
+            DataType::Date32 => {
+                let array = as_date32_array(col_array)?;
+                // ISO-8601/RFC3339 format - yyyy-mm-dd
+                let format = "%Y-%m-%d";
+                for i in 0..rb.num_rows() {
+                    let date = NaiveDate::from_num_days_from_ce_opt(
+                        EPOCH_DAYS_FROM_CE + array.value(i),
+                    )
+                    .unwrap();
+                    partition_values.push(date.format(format).to_string());
+                }
+            }
+            DataType::Date64 => {
+                let array = as_date64_array(col_array)?;
+                // ISO-8601/RFC3339 format - yyyy-mm-dd
+                let format = "%Y-%m-%d";
+                for i in 0..rb.num_rows() {
+                    let date = NaiveDate::from_num_days_from_ce_opt(
+                        EPOCH_DAYS_FROM_CE + (array.value(i) / 86_400_000) as i32,
+                    )
+                    .unwrap();
+                    partition_values.push(date.format(format).to_string());
+                }
+            }
+            DataType::Int32 => {
+                let array = as_int32_array(col_array)?;
+                for i in 0..rb.num_rows() {
+                    partition_values.push(array.value(i).to_string());
+                }
+            }
+            DataType::Int64 => {
+                let array = as_int64_array(col_array)?;
+                for i in 0..rb.num_rows() {
+                    partition_values.push(array.value(i).to_string());
                 }
             }
             DataType::Dictionary(_, _) => {
@@ -354,7 +408,7 @@ fn compute_partition_keys_by_row<'a>(
 
                         for val in array.values() {
                             partition_values.push(
-                                val.ok_or(exec_datafusion_err!("Cannot partition by null value for column {}", col))?
+                                val.ok_or(exec_datafusion_err!("Cannot partition by null value for column {}", col))?.to_string()
                             );
                         }
                     },
@@ -377,7 +431,7 @@ fn compute_partition_keys_by_row<'a>(
 
 fn compute_take_arrays(
     rb: &RecordBatch,
-    all_partition_values: Vec<Vec<&str>>,
+    all_partition_values: Vec<Vec<String>>,
 ) -> HashMap<Vec<String>, UInt64Builder> {
     let mut take_map = HashMap::new();
     for i in 0..rb.num_rows() {

datafusion/sqllogictest/test_files/copy.slt

@@ -108,6 +108,28 @@ select * from validate_partitioned_parquet_1_x order by column2;
 ----
 a
 
+# Copy to directory as partitioned files
+query I
+COPY (values (1::int, 2::bigint, 19968::date, arrow_cast(1725235200000, 'Date64'), false, 'x'),
+    (11::int, 22::bigint, 19969::date, arrow_cast(1725148800000, 'Date64'), true, 'y')
+)
+TO 'test_files/scratch/copy/partitioned_table5/' STORED AS parquet PARTITIONED BY (column1, column2, column3, column4, column5)
+OPTIONS ('format.compression' 'zstd(10)');
+----
+2
+
+# validate partitioning
+statement ok
+CREATE EXTERNAL TABLE validate_partitioned_parquet5 (column1 int, column2 bigint, column3 date, column4 date, column5 boolean, column6 varchar) STORED AS PARQUET
+LOCATION 'test_files/scratch/copy/partitioned_table5/' PARTITIONED BY (column1, column2, column3, column4, column5);
+
+query IIDDBT
+select column1, column2, column3, column4, column5, column6 from validate_partitioned_parquet5 order by column1,column2,column3,column4,column5;
+----
+1 2 2024-09-02 2024-09-02 false x
+11 22 2024-09-03 2024-09-01 true y
+
+
 statement ok
 create table test ("'test'" varchar, "'test2'" varchar, "'test3'" varchar);