Optimize ContextPairs nested name splitting

[aalsanie]

ContextPairs builds nested pairs by splitting each name on ".". The current implementation uses String.split("\\."), which is regex-based and allocates an array and list per name.

Here I replaced the regex split with a manual dot delimited scan that reduces allocations and avoids regex overhead while preserving existing behavior.

A test has been added.

[wilkinsona]

Thanks for the PR. Unfortunately, I'm not sure that we should merge it as it adds a significant amount of complexity. To consider merging it, I think we'd need to see evidence that the change makes a notable difference to real-world performance.

[aalsanie]

@wilkinsona Thanks for the feedback, I went and isolated both implementations into the same class and executed them under identical JVM conditions.

java SplitPerf

regex: 664 ms (66.42357 ns/op) manual: 478 ms (47.80414 ns/op) blackhole=0 

java -XX:StartFlightRecording=filename=split.jfr,settings=profile -cp . SplitPerf

regex: 706 ms (70.6994 ns/op) manual: 487 ms (48.79678 ns/op) blackhole=0 

Running 10M iterations shows a consistent ~30% reduction in execution time,, the reduced allocation rate also lowers GC pressure.

[wilkinsona]

Thanks. That's not quite what I meant by real-world and is more of a micro benchmark. To justify the extra complexity, I think we'd need to see an improvement in the performance of structured logging.

[aalsanie]

Fair enough, I’ll put together a small boot app with structured logging end-to-end and compare baseline.