Change `flatten` so it does only a level, not recursively #15160

delamarch3 · 2025-03-11T14:45:51Z

Which issue does this PR close?

Closes flatten should be single-step, not recursive #13757

Rationale for this change

Parity with the flatten implementation in duckdb.

What changes are included in this PR?

Remove the recursion in flatten_internal so that only the top level elements are flattened.

Are these changes tested?

Existing sqllogictests have been updated.

Are there any user-facing changes?

Yes, flatten no longer recursively flattens the array

Weijun-H · 2025-03-12T05:03:27Z

datafusion/functions-nested/src/flatten.rs

+                _ => arg_types[0].clone(),
+            },
+            LargeList(field) => match field.data_type() {
+                LargeList(field) => LargeList(Arc::clone(field)),


It deserves to return LargeList, if the nested data is List

Suggested change

LargeList(field) => LargeList(Arc::clone(field)),

List(field) | FixedSizeList(field, _) | LargeList(field) => {

LargeList(Arc::clone(field))

}

Ah ok, I'll need to make some changes to support this. Currently trying to flatten LargeList(List) will fail when casting the inner list to generic list with i64 offset (using the type parameter O)

findepi · 2025-03-12T15:50:16Z

datafusion/functions-nested/src/flatten.rs

@@ -77,7 +77,6 @@ impl Flatten {
    pub fn new() -> Self {
        Self {
            signature: Signature {
-                // TODO (https://github.com/apache/datafusion/issues/13757) flatten should be single-step, not recursive
                type_signature: TypeSignature::ArraySignature(
                    ArrayFunctionSignature::RecursiveArray,


Is this signature still applicable?

Maybe we should switch to Array, and mark RecursiveArray as deprecated?

What would the deprecated note be since, from what I understand, this was added in specifically for flatten to recursively coerce FixedLengthList to List. I'm wondering if any users would rely on that downstream

cht42 · 2025-04-02T06:10:17Z

Hey @delamarch3 are you still tracking merging this ? interested in this new behavior

delamarch3 · 2025-04-02T14:06:24Z

Hi @cht42, I'm just waiting for another review.

@findepi could you take another look at this when you get the chance please?

cht42 · 2025-04-09T09:13:44Z

Maybe @alamb or @Weijun-H can review ?

Weijun-H

Thanks @delamarch3 👍 , I think the pr looks good. Waiting for other comments.

cht42

Hey @delamarch3 , noticed that null handling is not managed properly in the current implementation. Might as well fix it here directly

cht42 · 2025-04-11T14:05:15Z

datafusion/functions-nested/src/flatten.rs

-            let list_arr = as_list_array(&array)?;
-            let flattened_array = flatten_internal::<i32>(list_arr.clone(), None)?;
-            Ok(Arc::new(flattened_array) as ArrayRef)
+            let (field, offsets, values, _) = as_list_array(&array)?.clone().into_parts();


Suggested change

let (field, offsets, values, _) = as_list_array(&array)?.clone().into_parts();

let (field, offsets, values, nulls) = as_list_array(&array)?.clone().into_parts();

cht42 · 2025-04-11T14:05:27Z

datafusion/functions-nested/src/flatten.rs

+                        inner_field,
+                        offsets,
+                        inner_values,
+                        None,


Suggested change

None,

nulls,

cht42 · 2025-04-11T14:05:41Z

datafusion/functions-nested/src/flatten.rs

-            let list_arr = as_large_list_array(&array)?;
-            let flattened_array = flatten_internal::<i64>(list_arr.clone(), None)?;
-            Ok(Arc::new(flattened_array) as ArrayRef)
+            let (field, offsets, values, _) =


Suggested change

let (field, offsets, values, _) =

let (field, offsets, values, nulls) =

cht42 · 2025-04-11T14:05:48Z

datafusion/functions-nested/src/flatten.rs

+                        inner_field,
+                        offsets,
+                        inner_values,
+                        None,


Suggested change

None,

nulls,

cht42 · 2025-04-11T14:05:55Z

datafusion/functions-nested/src/flatten.rs

+                    Ok(Arc::new(flattened_array) as ArrayRef)
+                }
+                LargeList(_) => {
+                    let (inner_field, inner_offsets, inner_values, _) =


Suggested change

let (inner_field, inner_offsets, inner_values, _) =

let (inner_field, inner_offsets, inner_values, nulls) =

cht42 · 2025-04-11T14:06:05Z

datafusion/functions-nested/src/flatten.rs

+                        inner_field,
+                        offsets,
+                        inner_values,
+                        None,


Suggested change

None,

nulls,

delamarch3 · 2025-04-12T09:58:48Z

Thanks for the reviews!

cht42 · 2025-04-16T07:53:01Z

@alamb can we get this merged please ?

alamb · 2025-04-17T16:53:59Z

Thanks @cht42 @delamarch3 and @Weijun-H -- I reviewed this PR and marked it as an API change. I think it looks good and does what is requested

Thanks!

* flatten array in a single step instead of recursive * clippy * update flatten type signature to Array * add fixed list to list coercion to flatten signature * support LargeList(List) and LargeList(FixedSizeList) in flatten * add test for LargeList(FixedSizeList) * handle nulls * uncomment flatten(NULL) test - it already works

flatten array in a single step instead of recursive

0417ac4

github-actions bot added the sqllogictest SQL Logic Tests (.slt) label Mar 11, 2025

clippy

8be0a0b

alamb requested a review from findepi March 11, 2025 18:56

Weijun-H reviewed Mar 12, 2025

View reviewed changes

findepi reviewed Mar 12, 2025

View reviewed changes

delamarch3 added 3 commits March 13, 2025 12:55

update flatten type signature to Array

25ee411

add fixed list to list coercion to flatten signature

2c151f4

support LargeList(List) and LargeList(FixedSizeList) in flatten

517334f

delamarch3 requested review from findepi and Weijun-H March 15, 2025 15:31

add test for LargeList(FixedSizeList)

77bacd1

Weijun-H approved these changes Apr 10, 2025

View reviewed changes

cht42 reviewed Apr 11, 2025

View reviewed changes

delamarch3 added 2 commits April 12, 2025 10:55

handle nulls

59bd238

uncomment flatten(NULL) test - it already works

9238722

alamb changed the title ~~Flatten array in a single step instead of recursive~~ Change flatten so it does only a level, not recursively Apr 17, 2025

alamb added the api change Changes the API exposed to users of the crate label Apr 17, 2025

alamb merged commit a0b0221 into apache:main Apr 17, 2025
28 checks passed

joroKr21 mentioned this pull request Apr 30, 2025

Saner handling of nulls inside arrays #15149

Merged

This was referenced Apr 30, 2025

Fix flatten scalar function when inner list is FixedSizeList #15898

Merged

Main is broken #15896

Closed

joroKr21 mentioned this pull request Apr 30, 2025

Handle FixedSizeList in flatten #15899

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Change `flatten` so it does only a level, not recursively #15160

Change `flatten` so it does only a level, not recursively #15160

delamarch3 commented Mar 11, 2025

Weijun-H Mar 12, 2025 •

edited

Loading

delamarch3 Mar 12, 2025

findepi Mar 12, 2025

delamarch3 Mar 13, 2025

cht42 commented Apr 2, 2025

delamarch3 commented Apr 2, 2025

cht42 commented Apr 9, 2025

Weijun-H left a comment

cht42 left a comment •

edited

Loading

cht42 Apr 11, 2025

cht42 Apr 11, 2025

cht42 Apr 11, 2025

cht42 Apr 11, 2025

cht42 Apr 11, 2025

cht42 Apr 11, 2025

delamarch3 commented Apr 12, 2025

cht42 commented Apr 16, 2025

alamb commented Apr 17, 2025

	let (field, offsets, values, _) = as_list_array(&array)?.clone().into_parts();
	let (field, offsets, values, nulls) = as_list_array(&array)?.clone().into_parts();

	let (field, offsets, values, _) =
	let (field, offsets, values, nulls) =

	let (inner_field, inner_offsets, inner_values, _) =
	let (inner_field, inner_offsets, inner_values, nulls) =

Change flatten so it does only a level, not recursively #15160

Change flatten so it does only a level, not recursively #15160

Conversation

delamarch3 commented Mar 11, 2025

Which issue does this PR close?

Rationale for this change

What changes are included in this PR?

Are these changes tested?

Are there any user-facing changes?

Weijun-H Mar 12, 2025 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

cht42 commented Apr 2, 2025

delamarch3 commented Apr 2, 2025

cht42 commented Apr 9, 2025

Weijun-H left a comment

Choose a reason for hiding this comment

cht42 left a comment • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

delamarch3 commented Apr 12, 2025

cht42 commented Apr 16, 2025

alamb commented Apr 17, 2025

Change `flatten` so it does only a level, not recursively #15160

Change `flatten` so it does only a level, not recursively #15160

Weijun-H Mar 12, 2025 •

edited

Loading

cht42 left a comment •

edited

Loading