From b9def8ab84772ef604734f407fa3b8db26d76d30 Mon Sep 17 00:00:00 2001
From: Robsdedude <rouven.bauer@neo4j.com>
Date: Thu, 24 Jul 2025 14:52:36 +0200
Subject: [PATCH 1/2] Optimize packing of `bytearray`

By special-casing `bytearray`, we can avoid an allocation and complete extra
copy of the data when packing it. This speeds up packing by roughly 1/3.
---
 src/codec/packstream/v1/pack.rs | 63 +++++++++++++++++++++++++++------
 1 file changed, 53 insertions(+), 10 deletions(-)

diff --git a/src/codec/packstream/v1/pack.rs b/src/codec/packstream/v1/pack.rs
index 5f89b6a..a0ba339 100644
--- a/src/codec/packstream/v1/pack.rs
+++ b/src/codec/packstream/v1/pack.rs
@@ -18,8 +18,9 @@ use std::sync::atomic::{AtomicBool, Ordering};
 
 use pyo3::exceptions::{PyImportError, PyOverflowError, PyTypeError, PyValueError};
 use pyo3::prelude::*;
+use pyo3::sync::with_critical_section;
 use pyo3::sync::GILOnceCell;
-use pyo3::types::{PyBytes, PyDict, PyString, PyType};
+use pyo3::types::{PyByteArray, PyBytes, PyDict, PyString, PyTuple, PyType};
 use pyo3::{intern, IntoPyObjectExt};
 
 use super::super::Structure;
@@ -43,6 +44,35 @@ struct TypeMappings {
 
 impl TypeMappings {
     fn new(locals: &Bound<PyDict>) -> PyResult<Self> {
+        /// Remove some byte types from an iterable of types.
+        /// Types removed are `bytes`, `bytearray`, as those are handled specially in `pack`.
+        /// If the filtering fails for any reason, it returns the original input.
+        fn filter_bytes_types(types: Bound<PyAny>) -> Bound<PyAny> {
+            fn inner<'py>(types: &Bound<'py, PyAny>) -> PyResult<Bound<'py, PyAny>> {
+                fn is_of_known_bytes_types(typ: &Bound<PyType>) -> PyResult<bool> {
+                    Ok(typ.is_subclass_of::<PyBytes>()? || typ.is_subclass_of::<PyByteArray>()?)
+                }
+
+                let py = types.py();
+                let types = types
+                    .try_iter()?
+                    .filter(|typ| {
+                        let Ok(typ) = typ else {
+                            return true;
+                        };
+                        let Ok(typ) = typ.downcast::<PyType>() else {
+                            return true;
+                        };
+                        is_of_known_bytes_types(typ).map(|b| !b).unwrap_or(true)
+                    })
+                    .collect::<Result<Vec<_>, _>>()?;
+
+                Ok(PyTuple::new(py, types)?.into_any())
+            }
+
+            inner(&types).unwrap_or(types)
+        }
+
         let py = locals.py();
         Ok(Self {
             none_values: locals
@@ -87,12 +117,15 @@ impl TypeMappings {
                     PyErr::new::<PyValueError, _>("Type mappings are missing MAPPING_TYPES.")
                 })?
                 .into_py_any(py)?,
-            bytes_types: locals
-                .get_item("BYTES_TYPES")?
-                .ok_or_else(|| {
-                    PyErr::new::<PyValueError, _>("Type mappings are missing BYTES_TYPES.")
-                })?
-                .into_py_any(py)?,
+            bytes_types: filter_bytes_types(
+                locals
+                    .get_item("BYTES_TYPES")?
+                    .ok_or_else(|| {
+                        PyErr::new::<PyValueError, _>("Type mappings are missing BYTES_TYPES.")
+                    })?
+                    .into_bound_py_any(py)?,
+            )
+            .unbind(),
         })
     }
 }
@@ -180,8 +213,18 @@ impl<'a> PackStreamEncoder<'a> {
             return self.write_string(value.extract::<&str>()?);
         }
 
-        if value.is_instance(self.type_mappings.bytes_types.bind(py))? {
-            return self.write_bytes(value.extract::<Cow<[u8]>>()?);
+        if let Ok(value) = value.downcast::<PyBytes>() {
+            return self.write_bytes(value.as_bytes());
+        } else if let Ok(value) = value.downcast::<PyByteArray>() {
+            return with_critical_section(value, || {
+                // SAFETY:
+                //  * we're holding the GIL/are attached to the Python interpreter
+                //  * we're using a critical section to ensure exclusive access to the byte array
+                //  * we don't interact with the interpreter/PyO3 APIs while reading the bytes
+                unsafe { self.write_bytes(value.as_bytes()) }
+            });
+        } else if value.is_instance(self.type_mappings.bytes_types.bind(py))? {
+            return self.write_bytes(&value.extract::<Cow<[u8]>>()?);
         }
 
         if value.is_instance(self.type_mappings.sequence_types.bind(py))? {
@@ -278,7 +321,7 @@ impl<'a> PackStreamEncoder<'a> {
         Ok(())
     }
 
-    fn write_bytes(&mut self, b: Cow<[u8]>) -> PyResult<()> {
+    fn write_bytes(&mut self, b: &[u8]) -> PyResult<()> {
         let size = Self::usize_to_u64(b.len())?;
         if size <= 255 {
             self.buffer.extend(&[BYTES_8]);

From ec93d272d50b9dc6bc0bd92da67a603a35747b0e Mon Sep 17 00:00:00 2001
From: Robsdedude <rouven.bauer@neo4j.com>
Date: Fri, 25 Jul 2025 09:18:02 +0200
Subject: [PATCH 2/2] Add changelog entry

---
 changelog.d/51.improve.md | 3 +++
 1 file changed, 3 insertions(+)
 create mode 100644 changelog.d/51.improve.md

diff --git a/changelog.d/51.improve.md b/changelog.d/51.improve.md
new file mode 100644
index 0000000..d45b238
--- /dev/null
+++ b/changelog.d/51.improve.md
@@ -0,0 +1,3 @@
+Optimize packing of `bytearray`<ISSUES_LIST>.
+By special-casing `bytearray`, we can avoid an allocation and complete extra copy of the data when packing it.
+This speeds up packing of `bytearray`s by roughly 1/3.