improve from dictionary -> categorical (#3996)

polars/polars-core/src/chunked_array/logical/categorical/from.rs

@@ -1,4 +1,5 @@
 use super::*;
+use crate::use_string_cache;
 use arrow::array::DictionaryArray;
 use arrow::datatypes::IntegerType;
 use polars_arrow::compute::cast::cast;
@@ -80,3 +81,30 @@ impl From<&CategoricalChunked> for DictionaryArray<i64> {
         }
     }
 }
+
+impl CategoricalChunked {
+    /// # Safety
+    /// The caller must ensure that index values in the `keys` are in within bounds of the `values` length.
+    pub(crate) unsafe fn from_keys_and_values(
+        name: &str,
+        keys: &PrimitiveArray<u32>,
+        values: &Utf8Array<i64>,
+    ) -> Self {
+        if use_string_cache() {
+            // todo!
+            // we probably can make this faster as we have more information.
+            let mut builder = CategoricalChunkedBuilder::new(name, keys.len());
+            let iter = keys
+                .into_iter()
+                .map(|opt_key| opt_key.map(|k| values.value_unchecked(*k as usize)));
+            builder.drain_iter(iter);
+            builder.finish()
+        } else {
+            CategoricalChunked::from_chunks_original(
+                name,
+                vec![Box::new(keys.clone())],
+                RevMapping::Local(values.clone()),
+            )
+        }
+    }
+}

polars/polars-core/src/series/from.rs

@@ -247,12 +247,9 @@ impl Series {
                 let keys = keys.as_any().downcast_ref::<PrimitiveArray<u32>>().unwrap();
                 let values = values.as_any().downcast_ref::<Utf8Array<i64>>().unwrap();
 
-                let mut builder = CategoricalChunkedBuilder::new(name, keys.len());
-                let iter = keys
-                    .into_iter()
-                    .map(|opt_key| opt_key.map(|k| values.value_unchecked(*k as usize)));
-                builder.drain_iter(iter);
-                Ok(builder.finish().into_series())
+                // Safety
+                // the invariants of an Arrow Dictionary guarantee the keys are in bounds
+                Ok(CategoricalChunked::from_keys_and_values(name, keys, values).into_series())
             }
             #[cfg(not(feature = "dtype-u8"))]
             ArrowDataType::LargeBinary | ArrowDataType::Binary => {