This is an automated email from the ASF dual-hosted git repository.
Cole-Greer pushed a commit to branch 3.7-dev
in repository https://gitbox.apache.org/repos/asf/tinkerpop.git
The following commit(s) were added to refs/heads/3.7-dev by this push:
new 5efeb43d43 Speed up Python GraphBinary deserialization (#3504)
5efeb43d43 is described below
commit 5efeb43d437070c6bdb09e17e099f3833fea2a5a
Author: kirill-stepanishin <[email protected]>
AuthorDate: Tue Jul 14 18:34:26 2026 -0700
Speed up Python GraphBinary deserialization (#3504)
The GraphBinary reader built a `DataType` enum member from the type byte
for every object it decoded. That per-object enum construction heavily degrades
deserialization performance on large result sets.
The reader now builds a `{type code: deserializer}` lookup table once up
front and dispatches on the raw integer instead, avoiding per-object enum
construction. Behavior is unchanged: an unknown type code still raises
`ValueError("... is not a valid DataType")`.
Assisted-by: Claude Code:claude-opus-4-8
---
.../python/gremlin_python/structure/io/graphbinaryV1.py | 14 ++++++++++++--
1 file changed, 12 insertions(+), 2 deletions(-)
diff --git
a/gremlin-python/src/main/python/gremlin_python/structure/io/graphbinaryV1.py
b/gremlin-python/src/main/python/gremlin_python/structure/io/graphbinaryV1.py
index 39a0c3a678..14c73e48a3 100644
---
a/gremlin-python/src/main/python/gremlin_python/structure/io/graphbinaryV1.py
+++
b/gremlin-python/src/main/python/gremlin_python/structure/io/graphbinaryV1.py
@@ -118,6 +118,9 @@ class DataType(Enum):
NULL_BYTES = [DataType.null.value, 0x01]
+# null type code as a plain int, so the per-read null check skips the aenum
lookup
+_NULL = DataType.null.value
+
def _make_packer(format_string):
packer = struct.Struct(format_string)
@@ -187,6 +190,9 @@ class GraphBinaryReader(object):
self.deserializers = _deserializers.copy()
if deserializer_map:
self.deserializers.update(deserializer_map)
+ # Mirror of self.deserializers keyed by int type code instead of
DataType.
+ # Avoids the per-read DataType(bt) call, whose aenum construction
negatively affects performance on large results.
+ self._deserializer_by_type_code = {dt.value: des.objectify for dt, des
in self.deserializers.items()}
def read_object(self, b):
if isinstance(b, bytearray):
@@ -197,11 +203,15 @@ class GraphBinaryReader(object):
def to_object(self, buff, data_type=None, nullable=True):
if data_type is None:
bt = uint8_unpack(buff.read(1))
- if bt == DataType.null.value:
+ if bt == _NULL:
if nullable:
buff.read(1)
return None
- return self.deserializers[DataType(bt)].objectify(buff, self,
nullable)
+ try:
+ objectify = self._deserializer_by_type_code[bt]
+ except KeyError:
+ raise ValueError("%r is not a valid DataType" % bt) from None
+ return objectify(buff, self, nullable)
else:
return self.deserializers[data_type].objectify(buff, self,
nullable)
