perf: Use row-encoding for multiple key group by (#15392)

crates/polars-core/src/chunked_array/object/mod.rs

@@ -33,6 +33,14 @@ pub trait PolarsObjectSafe: Any + Debug + Send + Sync + Display {
     fn as_any(&self) -> &dyn Any;
 
     fn to_boxed(&self) -> Box<dyn PolarsObjectSafe>;
+
+    fn equal(&self, other: &dyn PolarsObjectSafe) -> bool;
+}
+
+impl PartialEq for &dyn PolarsObjectSafe {
+    fn eq(&self, other: &Self) -> bool {
+        self.equal(*other)
+    }
 }
 
 /// Values need to implement this so that they can be stored into a Series and DataFrame
@@ -55,6 +63,13 @@ impl<T: PolarsObject> PolarsObjectSafe for T {
     fn to_boxed(&self) -> Box<dyn PolarsObjectSafe> {
         Box::new(self.clone())
     }
+
+    fn equal(&self, other: &dyn PolarsObjectSafe) -> bool {
+        let Some(other) = other.as_any().downcast_ref::<T>() else {
+            return false;
+        };
+        self == other
+    }
 }
 
 pub type ObjectValueIter<'a, T> = std::slice::Iter<'a, T>;

crates/polars-core/src/chunked_array/ops/sort/arg_sort_multiple.rs

@@ -3,7 +3,6 @@ use polars_row::{convert_columns, RowsEncoded, SortField};
 use polars_utils::iter::EnumerateIdxTrait;
 
 use super::*;
-#[cfg(feature = "dtype-struct")]
 use crate::utils::_split_offsets;
 
 pub(crate) fn args_validate<T: PolarsDataType>(
@@ -88,8 +87,7 @@ pub fn _get_rows_encoded_compat_array(by: &Series) -> PolarsResult<ArrayRef> {
     Ok(out)
 }
 
-#[cfg(feature = "dtype-struct")]
-pub(crate) fn encode_rows_vertical(by: &[Series]) -> PolarsResult<BinaryOffsetChunked> {
+pub(crate) fn encode_rows_vertical_par_default(by: &[Series]) -> PolarsResult<BinaryOffsetChunked> {
     let n_threads = POOL.current_num_threads();
     let len = by[0].len();
     let splits = _split_offsets(len, n_threads);
@@ -108,6 +106,12 @@ pub(crate) fn encode_rows_vertical(by: &[Series]) -> PolarsResult<BinaryOffsetCh
     Ok(BinaryOffsetChunked::from_chunk_iter("", chunks?))
 }
 
+pub(crate) fn encode_rows_default(by: &[Series]) -> PolarsResult<BinaryOffsetChunked> {
+    let descending = vec![false; by.len()];
+    let rows = _get_rows_encoded(by, &descending, false)?;
+    Ok(BinaryOffsetChunked::with_chunk("", rows.into_array()))
+}
+
 pub fn _get_rows_encoded(
     by: &[Series],
     descending: &[bool],

crates/polars-core/src/chunked_array/ops/sort/mod.rs

@@ -15,9 +15,7 @@ use rayon::prelude::*;
 pub use slice::*;
 
 use crate::prelude::compare_inner::TotalOrdInner;
-#[cfg(feature = "dtype-struct")]
-use crate::prelude::sort::arg_sort_multiple::_get_rows_encoded_ca;
-use crate::prelude::sort::arg_sort_multiple::{arg_sort_multiple_impl, args_validate};
+use crate::prelude::sort::arg_sort_multiple::*;
 use crate::prelude::*;
 use crate::series::IsSorted;
 use crate::utils::NoNull;

crates/polars-core/src/datatypes/any_value.rs

@@ -946,6 +946,8 @@ impl AnyValue<'_> {
                 // 1.2 at scale 1, and 1.20 at scale 2, are not equal.
                 *v_l == *v_r && *scale_l == *scale_r
             },
+            #[cfg(feature = "object")]
+            (Object(l), Object(r)) => l == r,
             _ => false,
         }
     }

crates/polars-core/src/datatypes/dtype.rs

@@ -241,6 +241,21 @@ impl DataType {
         matches!(self, DataType::Binary)
     }
 
+    pub fn is_object(&self) -> bool {
+        #[cfg(feature = "object")]
+        {
+            matches!(self, DataType::Object(_, _))
+        }
+        #[cfg(not(feature = "object"))]
+        {
+            false
+        }
+    }
+
+    pub fn is_null(&self) -> bool {
+        matches!(self, DataType::Null)
+    }
+
     pub fn contains_views(&self) -> bool {
         use DataType::*;
         match self {