feat: change Expr OuterReferenceColumn to Box type for reducing expr struct size #16771
+69
−36
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
github-actions
bot
added
sql
zhuqi-lucas
changed the title
alamb
reviewed
Jul 14, 2025
There was a problem hiding this comment.
Thanks @zhuqi-lucas -- this looks quite nice
I think it is an API change so I will mark the PR as such and I think it is a good improvement
However, given it is an API change I think we should
- Wait until DataFusion 50.0.0 (let's not make any more changes in DataFusion 49 that we are starting to prepare for)
- Update the upgrading guide in https://datafusion.apache.org/library-user-guide/upgrading.html to explain to people hw to change things
| @@ -4009,12 +4009,12 @@ logical_plan | |||
| 09)------------Unnest: lists[__unnest_placeholder(generate_series(Int64(1),outer_ref(t1.t1_int)))|depth=1] structs[] | |||
| 10)--------------Projection: generate_series(Int64(1), CAST(outer_ref(t1.t1_int) AS Int64)) AS __unnest_placeholder(generate_series(Int64(1),outer_ref(t1.t1_int))) | |||
| 11)----------------EmptyRelation | |||
| physical_plan_error This feature is not implemented: Physical plan does not support logical expression OuterReferenceColumn(UInt32, Column { relation: Some(Bare { table: "t1" }), name: "t1_int" }) | |||
| physical_plan_error This feature is not implemented: Physical plan does not support logical expression OuterReferenceColumn((UInt32, Column { relation: Some(Bare { table: "t1" }), name: "t1_int" })) | |||
There was a problem hiding this comment.
do you know why the plan changes? I don't think the extra () really adds much -- can we restore the original?
There was a problem hiding this comment.
I changed it to OuterReference in latest PR, so it will use OuterReference instead of this change now, thanks!
There was a problem hiding this comment.
Next step, we may can add debug/fmt for OuterReference to restore the original behaviour.
There was a problem hiding this comment.
Restore all slt in latest PR, no diff from main branch.
| @@ -3818,7 +3816,7 @@ mod test { | |||
| // If this test fails when you change `Expr`, please try | |||
| // `Box`ing the fields to make `Expr` smaller | |||
| // See https://github.com/apache/datafusion/issues/16199 for details | |||
| assert_eq!(size_of::<Expr>(), 128); | |||
| assert_eq!(size_of::<Expr>(), 112); | |||
There was a problem hiding this comment.
This change saves 16 bytes per Expr. Nice.
I will run some planning benchmarks and see if we can see any difference
There was a problem hiding this comment.
Thank you @alamb , i am doing more amazing experiment, try to reduce from 128 to 80, so we can save 48 bytes per Expr!
datafusion/expr/src/expr.rs
Outdated
| @@ -279,7 +279,7 @@ use sqlparser::ast::{ | |||
| #[derive(Clone, PartialEq, PartialOrd, Eq, Debug, Hash)] | |||
| pub enum Expr { | |||
| /// An expression with a specific name. | |||
| Alias(Alias), | |||
| Alias(Box<Alias>), | |||
There was a problem hiding this comment.
Another change that might be a bit less impactful would be to box the fields of Alias instead
So Alias {
expr: Box
..
}
It may be just as bad / worse though
datafusion/expr/src/expr.rs
Outdated
| @@ -362,7 +362,7 @@ pub enum Expr { | |||
| Placeholder(Placeholder), | |||
| /// A placeholder which holds a reference to a qualified field | |||
| /// in the outer query, used for correlated sub queries. | |||
| OuterReferenceColumn(DataType, Column), | |||
| OuterReferenceColumn(Box<(DataType, Column)>), | |||
There was a problem hiding this comment.
If we are going to make this change anyways, can we also pull this into a named struct rather than a unnamed tuple
like
struct OuterReference {
// fields here
}
enum Expr {
...
OuterReferenceColumn(OuterReference),
...
}There was a problem hiding this comment.
Thank you @alamb for this good suggestion, addressed in latest PR.
|
🤖 |
|
🤖: Benchmark completed Details
|
Thank you @alamb for review and i agree that we can do this for Datafusion 50.0.0, and since we have enough time, i want to do more improvement, try to reduce Expr more, from size 128 to 80, i am in progress now. |
zhuqi-lucas
changed the title
|
Updated: Successfully changed the size from 128 to 80 in latest PR. |
|
🤖 |
|
🤖: Benchmark completed Details
|
|
🤖 |
|
🤖: Benchmark completed Details
|
Thank you @alamb for benchmark, interesting, it seems the performance decrease from 112 size to 80, i will investigate it, if i can't find the root cause, i will revert to 112 bytes first! |
|
Strange, i can't reproduce this in my local: with_param_values_many_columns 1.00 129.8±0.74µs ? ?/sec 1.06 137.2±0.79µs |
kosiew
reviewed
Jul 16, 2025
Comment on lines
260
to
262
| Expr::Alias(alias_box) => { | ||
| // alias_box: Box<Alias> | ||
| let Alias { expr, name, .. } = *alias_box; |
There was a problem hiding this comment.
nit - use boxed_alias instead, for consistency
There was a problem hiding this comment.
Thank you @kosiew , addressed in latest PR.
Comment on lines
140
to
142
| if let Expr::Alias(alias_box) = expr { | ||
| // alias_box: &Box<Alias>, so alias_box.as_ref() is &Alias | ||
| let alias: &Alias = alias_box.as_ref(); |
There was a problem hiding this comment.
same as above - boxed_alias
datafusion/expr/src/expr_schema.rs
Outdated
Comment on lines
249
to
250
| Expr::Alias(alias_box) => { | ||
| let Alias { expr, .. } = alias_box.as_ref(); |
datafusion/expr/src/utils.rs
Outdated
| @@ -992,7 +996,7 @@ pub fn iter_conjunction_owned(expr: Expr) -> impl Iterator<Item = Expr> { | |||
| stack.push(*right); | |||
| stack.push(*left); | |||
| } | |||
| Expr::Alias(Alias { expr, .. }) => stack.push(*expr), | |||
| Expr::Alias(alias_box) => stack.push(*alias_box.expr), | |||
Thank you @kosiew for review and suggestion, addressed in latest PR. |
|
After opening the DF50.0.0 release issue, you can add it to the list |
I will rerun |
|
🤖 |
|
🤖: Benchmark completed Details
|
Thank you @alamb , almost all improvement for most cases, but still the last one has %4 regression, and i can't reproduce in my local Mac. |
15 tasks
Thank you @xudong963 , added it in #16799 (comment) |
|
My guess is that some of the new slowdown / less predictability is due to many more |
Thank you @alamb , this is a very good point, let me try this and to see the benchmark result! |
zhuqi-lucas
force-pushed
the
reduce_expr_size
branch
from
42af92a to
998950a
Compare
github-actions
bot
removed
physical-expr
@alamb Updated in latest PR, only small change for OuterReferenceColumn now. |
zhuqi-lucas
changed the title
|
🤖 |
|
🤖: Benchmark completed Details
|
The result is bad, no performance gain from only OuterReferenceColumn changing to Box, so i guess the improvement coming from Alias changing to Box? 🤔 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Which issue does this PR close?
Continue to reduce the Expr struct size.
Rationale for this change
Continue to reduce the Expr struct size.
What changes are included in this PR?
Continue to reduce the Expr struct size.
Are these changes tested?
Yes
The size reduce from 128 to 112
Updated:
Continue reduce to 80 now!
Are there any user-facing changes?
No