fix date_part('isodow')

datafusion/functions/src/datetime/date_part.rs

@@ -80,7 +80,7 @@ use datafusion_macros::user_doc;
     - dow (day of the week where Sunday is 0)
     - doy (day of the year)
     - epoch (seconds since Unix epoch for timestamps/dates, total seconds for intervals)
-    - isodow (day of the week where Monday is 0)
+    - isodow (ISO 8601 day of the week where Monday is 1 and Sunday is 7)
 "#
     ),
     argument(
@@ -239,7 +239,18 @@ impl ScalarUDFImpl for DatePartFunc {
                 "qtr" | "quarter" => date_part(array.as_ref(), DatePart::Quarter)?,
                 "doy" => date_part(array.as_ref(), DatePart::DayOfYear)?,
                 "dow" => date_part(array.as_ref(), DatePart::DayOfWeekSunday0)?,
-                "isodow" => date_part(array.as_ref(), DatePart::DayOfWeekMonday0)?,
+                "isodow" => {
+                    // Postgres `isodow` is 1..=7 with Mon=1. Arrow's
+                    // `DayOfWeekMonday0` returns 0..=6 with Mon=0; shift by
+                    // +1 to match Postgres. TODO: switch to a future
+                    // `DatePart::DayOfWeekMonday1` upstream variant once it
+                    // exists, so this kernel-then-add becomes a single call.
+                    let zero_based =
+                        date_part(array.as_ref(), DatePart::DayOfWeekMonday0)?;
+                    let int_arr = as_int32_array(&zero_based)?;
+                    let one_based: Int32Array = int_arr.unary(|v| v + 1);
+                    Arc::new(one_based) as ArrayRef
+                }
                 "epoch" => epoch(array.as_ref())?,
                 _ => return exec_err!("Date part '{part}' not supported"),
             }

datafusion/spark/src/function/datetime/date_part.rs

@@ -125,8 +125,11 @@ impl ScalarUDFImpl for SparkDatePart {
         ));
 
         match part {
-            // Add 1 for day-of-week parts to convert 0-indexed to 1-indexed
-            "dow" | "isodow" => Ok(ExprSimplifyResult::Simplified(
+            // Spark's `dayofweek` is 1..=7 (Sun=1) but df's `dow` is 0..=6
+            // (Sun=0); shift by +1. df's `isodow` already returns the
+            // PG-correct 1..=7 (Mon=1), which matches Spark's
+            // `dayofweek_iso`/`dow_iso`, so no shift is needed there.
+            "dow" => Ok(ExprSimplifyResult::Simplified(
                 date_part_expr + Expr::Literal(ScalarValue::Int32(Some(1)), None),
             )),
             _ => Ok(ExprSimplifyResult::Simplified(date_part_expr)),

datafusion/sqllogictest/test_files/datetime/date_part.slt

@@ -1240,22 +1240,22 @@ true
 query I
 SELECT date_part('ISODOW', CAST('2000-01-01' AS DATE))
 ----
-5
+6
 
 query I
 SELECT EXTRACT(isodow FROM to_timestamp('2020-09-08T12:00:00+00:00'))
 ----
-1
+2
 
 query I
 SELECT EXTRACT("isodow" FROM to_timestamp('2020-09-08T12:00:00+00:00'))
 ----
-1
+2
 
 query I
 SELECT EXTRACT('isodow' FROM to_timestamp('2020-09-08T12:00:00+00:00'))
 ----
-1
+2
 
 ## Preimage tests
 

docs/source/user-guide/sql/scalar_functions.md

@@ -2573,7 +2573,7 @@ date_part(part, expression)
   - dow (day of the week where Sunday is 0)
   - doy (day of the year)
   - epoch (seconds since Unix epoch for timestamps/dates, total seconds for intervals)
-  - isodow (day of the week where Monday is 0)
+  - isodow (ISO 8601 day of the week where Monday is 1 and Sunday is 7)
 
 - **expression**: Time expression to operate on. Can be a constant, column, or function.