DB Tsai created SPARK-60697:
-------------------------------
Summary: Avoid materializing DecimalTuple for exact Python Decimal
input conversion
Key: SPARK-60697
URL: https://issues.apache.org/jira/browse/SPARK-60697
Project: Spark
Issue Type: Improvement
Components: Connect
Affects Versions: connect-rust-4.3.0
Reporter: DB Tsai
Assignee: DB Tsai
Fix For: connect-rust-4.3.0
Client-only combined benchmark shows decimal pandas input23.80ms vsPython5.84ms
and tuple21.88ms vs8.89ms for10000x3 values. cProfile of10
constructions/300000cells spends0.043s of0.273s in DecimalTuple namedtuple
construction from decimal_to_value.as_tuple (about16percent). Binding already
obtains str(value); exact built-in Decimal scale can potentially be derived
from that canonical representation without materializing every coefficient
digit. Evaluate a narrow optimization preserving exact value, inferred/explicit
scale including trailing zeros, exponent notation, special values, Decimal
subclasses and exceptions. Keep subclass/as_tuple overrides on existing
fallback and do not bundle scientific parser changes tracked SPARK-60696.
Require separate frozen baseline/candidate client measurements, unchanged
harness and independent IPC/schema correctness including inferred inputs. No
JVM.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
