DB Tsai created SPARK-60704:
-------------------------------
Summary: Avoid repeated Decimal type lookup for explicit decimal
local input in Spark Connect Rust
Key: SPARK-60704
URL: https://issues.apache.org/jira/browse/SPARK-60704
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
After SPARK-60697/PR133 eliminates DecimalTuple construction, combined pandas
decimal input remains15.53ms vsPython5.64ms and tuple13.70ms vs8.72ms
for10000x3 values. Native decimal_to_value still imports decimal and gets
Decimal for every cell. Evaluate a narrow per-construction resolved-type path
for explicit DecimalType schemas and exact Decimal/None cells, avoiding
repeated module/type lookup and unrelated generic probes. Preserve canonical
string/scale, exact cells, subclass hooks and fallback/error ordering; do not
introduce global cached Python objects, class-name heuristics or unsafe state.
Validate any acceptance guard against mixed values, module rebinding,
subclasses and exceptions. Benchmark incrementally with identical full PR133 on
both frozen sides and unchanged harness; publish only meaningful ROI with
correct independent IPC/schema values and unchanged guards. Scientific parser
changes SPARK-60696 must be identical on both optional correctness sides. 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]
