Skip to content
Closed
Show file tree
Hide file tree
Changes from 5 commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 8 additions & 0 deletions datafusion/spark/src/function/datetime/mod.rs
Original file line number Diff line number Diff line change
Expand Up @@ -27,6 +27,7 @@ pub mod last_day;
pub mod make_dt_interval;
pub mod make_interval;
pub mod next_day;
pub mod quarter;
pub mod time_trunc;
pub mod to_utc_timestamp;
pub mod trunc;
Expand Down Expand Up @@ -72,6 +73,7 @@ make_udf_function!(
unix_seconds,
unix::SparkUnixTimestamp::seconds
);
make_udf_function!(quarter::SparkQuarter, quarter);

pub mod expr_fn {
use datafusion_functions::export_functions;
Expand Down Expand Up @@ -179,6 +181,11 @@ pub mod expr_fn {
"Returns the number of seconds since epoch (1970-01-01 00:00:00 UTC) for the given timestamp `ts`.",
ts
));
export_functions!((
quarter,
"Returns the quarter of the year for date, in the range 1 to 4.",
arg1
));
}

pub fn functions() -> Vec<Arc<ScalarUDF>> {
Expand All @@ -204,5 +211,6 @@ pub fn functions() -> Vec<Arc<ScalarUDF>> {
unix_micros(),
unix_millis(),
unix_seconds(),
quarter(),
]
}
111 changes: 111 additions & 0 deletions datafusion/spark/src/function/datetime/quarter.rs
Original file line number Diff line number Diff line change
@@ -0,0 +1,111 @@
// Licensed to the Apache Software Foundation (ASF) under one
// or more contributor license agreements. See the NOTICE file
// distributed with this work for additional information
// regarding copyright ownership. The ASF licenses this file
// to you under the Apache License, Version 2.0 (the
// "License"); you may not use this file except in compliance
// with the License. You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing,
// software distributed under the License is distributed on an
// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
// KIND, either express or implied. See the License for the
// specific language governing permissions and limitations
// under the License.

use arrow::array::{ArrayRef, AsArray, Int32Array};
use arrow::datatypes::{DataType, Date32Type, Field, FieldRef};
use chrono::Datelike;
use datafusion::logical_expr::{ColumnarValue, Signature, Volatility};
use datafusion_common::utils::take_function_args;
use datafusion_common::{Result, ScalarValue, internal_err};
use datafusion_expr::{ReturnFieldArgs, ScalarFunctionArgs, ScalarUDFImpl};
use std::any::Any;
use std::sync::Arc;

#[derive(Debug, PartialEq, Eq, Hash)]
pub struct SparkQuarter {
signature: Signature,
}

impl Default for SparkQuarter {
fn default() -> Self {
Self::new()
}
}

impl SparkQuarter {
pub fn new() -> Self {
Self {
signature: Signature::exact(vec![DataType::Date32], Volatility::Immutable),
Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think this is the main thing we should fix before merging. Right now the UDF is registered with an exact Date32 signature, which means we no longer preserve Spark’s documented call shape for quarter.

Spark’s SQL docs show SELECT quarter('2016-08-31'); returning 3, and this SLT file used to carry that example before it was replaced with explicit ::DATE casts. With the current signature, we only validate the casted form and could end up rejecting the plain string-literal case that Spark accepts.

Could we switch this to a coercible signature, or possibly just route through the existing date_part('quarter', ...) behavior, and add coverage for the uncasted query?

}
}
}

impl ScalarUDFImpl for SparkQuarter {
fn as_any(&self) -> &dyn Any {
self
}

fn name(&self) -> &str {
"quarter"
}

fn signature(&self) -> &Signature {
&self.signature
}

fn return_type(&self, _arg_types: &[DataType]) -> Result<DataType> {
internal_err!("return_field_from_args should be used instead")
}

fn return_field_from_args(&self, args: ReturnFieldArgs) -> Result<FieldRef> {
Ok(Arc::new(Field::new(
self.name(),
DataType::Int32,
args.arg_fields[0].is_nullable(),
Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think the return-field nullability needs to be loosened here.

Right now return_field_from_args mirrors the input field nullability, but the new string path can produce NULL even when the input is non-null. This patch adds cases like quarter('abc'::string) and quarter(''::string) returning NULL, so quarter(non_null_utf8_col) would still be advertised as Int32 NOT NULL even though execution can yield nulls.

That looks like a schema contract bug. It also differs from existing Spark helpers like next_day, which force nullable output when invalid strings can map to NULL.

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks. fixed

)))
}

fn invoke_with_args(&self, args: ScalarFunctionArgs) -> Result<ColumnarValue> {
let [arg] = take_function_args("quarter", args.args)?;
Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Small suggestion here: this seems to repeat the same scalar/array Date32 dispatching pattern we already have in other datetime helpers, while datafusion_functions::datetime::date_part() already supports "quarter".

Would it make sense to delegate to that existing implementation instead? It feels like that would help keep coercion rules, null handling, and any future date-part behavior aligned in one place.

match arg {
ColumnarValue::Scalar(ScalarValue::Date32(days)) => {
if let Some(days) = days {
Ok(ColumnarValue::Scalar(ScalarValue::Int32(Some(
spark_quarter(days)?,
))))
} else {
Ok(ColumnarValue::Scalar(ScalarValue::Int32(None)))
}
}
ColumnarValue::Array(array) => {
let result = match array.data_type() {
DataType::Date32 => {
let result: Int32Array = array
.as_primitive::<Date32Type>()
.try_unary(spark_quarter)?
.with_data_type(DataType::Int32);
Ok(Arc::new(result) as ArrayRef)
}
other => {
internal_err!(
"Unsupported data type {other:?} for Spark function `quarter`"
)
}
}?;
Ok(ColumnarValue::Array(result))
}
other => {
internal_err!("Unsupported arg {other:?} for Spark function `quarter")
}
}
}
}

fn spark_quarter(days: i32) -> Result<i32> {
let quarter = Date32Type::to_naive_date_opt(days).unwrap().quarter();
Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

One thing that made me a little nervous here is the unwrap(). That introduces a panic path if a malformed Date32 value ever makes it down to this helper.

last_day handles the same kind of conversion by returning an explicit error instead, which feels a bit safer for a public UDF. Could we do the same here so bad inputs show up as query errors rather than aborting execution?

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I tried to rework it

Ok(quarter as i32)
}
44 changes: 34 additions & 10 deletions datafusion/sqllogictest/test_files/spark/datetime/quarter.slt
Original file line number Diff line number Diff line change
Expand Up @@ -15,13 +15,37 @@
# specific language governing permissions and limitations
# under the License.

# This file was originally created by a porting script from:
# https://github.com/lakehq/sail/tree/43b6ed8221de5c4c4adbedbb267ae1351158b43c/crates/sail-spark-connect/tests/gold_data/function
# This file is part of the implementation of the datafusion-spark function library.
# For more information, please see:
# https://github.com/apache/datafusion/issues/15914

## Original Query: SELECT quarter('2016-08-31');
## PySpark 3.5.5 Result: {'quarter(2016-08-31)': 3, 'typeof(quarter(2016-08-31))': 'int', 'typeof(2016-08-31)': 'string'}
#query
#SELECT quarter('2016-08-31'::string);
query I
Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The added coverage for DATE and TIMESTAMP inputs looks good 👍

That said, we’re still missing the specific Spark regression case that was called out earlier: SELECT quarter('2016-08-31');

Since the implementation still doesn’t accept plain string literals, not having this exact case in the SLT means the mismatch isn’t being caught.

It would be great to add this test back in so we lock in the expected Spark behavior and prevent regressions once the coercion issue is fixed.

SELECT quarter('2009-01-12'::DATE);
----
1

query I
SELECT quarter('1970-01-01'::DATE);
----
1

query I
SELECT quarter('1870-01-01'::DATE);
----
1

query I
SELECT quarter('2011-04-21'::DATE);
----
2

query I
SELECT quarter('2024-08-14'::DATE);
----
3

query I
SELECT quarter('2016-12-12'::DATE);
----
4

query I
SELECT quarter(NULL::DATE);
----
NULL