Hi,
I noticed that a vanilla Boot app spends some time in StringSequence.startsWith or respectively StringSequence.charAt:
This PR optimizes startsWith in cases where the length doesn't differ from the source length, which essentially means that StringSequence is "equal" to the source. In those cases, we can use startsWith on the actual source and by that avoid the repeated boundary checks inside charAt. A little benchmark of old vs. new approach shows the following results.
Benchmark Mode Cnt Score Error Units
MyBenchmark.testNew avgt 10 6,270 ± 0,255 ns/op
MyBenchmark.testNew:·gc.alloc.rate avgt 10 ≈ 10⁻⁴ MB/sec
MyBenchmark.testNew:·gc.alloc.rate.norm avgt 10 ≈ 10⁻⁶ B/op
MyBenchmark.testNew:·gc.count avgt 10 ≈ 0 counts
MyBenchmark.testOld avgt 10 9,029 ± 0,330 ns/op
MyBenchmark.testOld:·gc.alloc.rate avgt 10 ≈ 10⁻⁴ MB/sec
MyBenchmark.testOld:·gc.alloc.rate.norm avgt 10 ≈ 10⁻⁶ B/op
MyBenchmark.testOld:·gc.count avgt 10 ≈ 0 counts
Let me know what you think. Cheers, Christoph
Comment From: philwebb
Nice!
Comment From: philwebb
Moving the parameter type to String opens the door for even more optimizations I think. @dreis2211 and @wilkinsona I'd appreciate it if you could look at commit 4a8492d428be5c659acc007e3b6f1462ff9fe645 and see if it makes sense.
Comment From: philwebb
Reopening to ensure we get at least one other pair of eyes on this before we release.
Comment From: dreis2211
Looks good to me. In reality it won't change much though, because there are actually no other cases in the code where the length differs from the original string (apart from tests). So it's a good polishing commit, but no real performance improvement anymore ;)
Comment From: wilkinsona
Thanks, @dreis2211. FWIW, it looks good to me too.
Comment From: philwebb
Thanks for the reviews!