PROTON-2287 Improve Symbol decoding cache

[franz1981]

https://issues.apache.org/jira/browse/PROTON-2287

[franz1981]

This is the improvement related:

master

Benchmark                                             Mode  Cnt    Score     Error   Units
SymbolsBenchmark.decode                               avgt    5  283.784 ±   4.618   ns/op

Benchmark                                                                Mode  Cnt    Score   Error  Units
CompositeReadableBufferBenchmark.hashCodeMultipleArrayFullSpan           avgt   10  770.841 ± 6.285  ns/op
CompositeReadableBufferBenchmark.hashCodeMultipleArraySpanInSingleArray  avgt   10  113.202 ± 0.330  ns/op
CompositeReadableBufferBenchmark.hashCodeSingArraySpanInSingleArray      avgt   10  114.003 ± 1.756  ns/op
CompositeReadableBufferBenchmark.hashCodeSingleArrayFullSpan             avgt   10  581.803 ± 2.284  ns/op

Benchmark                                                                     (chunks)  (direct)  (size)  Mode  Cnt    Score   Error  Units
CompositeReadableBufferEqualsBenchmark.equalsToByteBufferReader                      1     false       8  avgt   10   25.528 ± 0.038  ns/op
CompositeReadableBufferEqualsBenchmark.equalsToByteBufferReader                      1     false      16  avgt   10   33.414 ± 0.117  ns/op
CompositeReadableBufferEqualsBenchmark.equalsToByteBufferReader                      1     false      64  avgt   10   61.316 ± 2.427  ns/op
CompositeReadableBufferEqualsBenchmark.equalsToByteBufferReader                      1      true       8  avgt   10   26.567 ± 1.156  ns/op
CompositeReadableBufferEqualsBenchmark.equalsToByteBufferReader                      1      true      16  avgt   10   40.832 ± 0.672  ns/op
CompositeReadableBufferEqualsBenchmark.equalsToByteBufferReader                      1      true      64  avgt   10  125.804 ± 0.590  ns/op
CompositeReadableBufferEqualsBenchmark.equalsToWithSingleArraySubsetOfBuffer         1     false       8  avgt   10   27.091 ± 0.156  ns/op
CompositeReadableBufferEqualsBenchmark.equalsToWithSingleArraySubsetOfBuffer         1     false      16  avgt   10   29.060 ± 0.032  ns/op
CompositeReadableBufferEqualsBenchmark.equalsToWithSingleArraySubsetOfBuffer         1     false      64  avgt   10   60.711 ± 3.901  ns/op
CompositeReadableBufferEqualsBenchmark.equalsToWithSingleArraySubsetOfBuffer         1      true       8  avgt   10   29.594 ± 0.262  ns/op
CompositeReadableBufferEqualsBenchmark.equalsToWithSingleArraySubsetOfBuffer         1      true      16  avgt   10   45.407 ± 1.230  ns/op
CompositeReadableBufferEqualsBenchmark.equalsToWithSingleArraySubsetOfBuffer         1      true      64  avgt   10  149.863 ± 0.794  ns/op

on this pr:

Benchmark                       Mode  Cnt    Score   Error  Units
SymbolsBenchmark.decode         avgt   20  197.239 ± 2.010  ns/op

Benchmark                                                                Mode  Cnt    Score   Error  Units
CompositeReadableBufferBenchmark.hashCodeMultipleArrayFullSpan           avgt   10  768.331 ± 2.098  ns/op
CompositeReadableBufferBenchmark.hashCodeMultipleArraySpanInSingleArray  avgt   10   59.398 ± 0.385  ns/op
CompositeReadableBufferBenchmark.hashCodeSingArraySpanInSingleArray      avgt   10   61.704 ± 4.174  ns/op
CompositeReadableBufferBenchmark.hashCodeSingleArrayFullSpan             avgt   10  279.393 ± 2.857  ns/op

Benchmark                                                                                         (chunks)  (direct)  (size)  Mode  Cnt    Score    Error   Units
CompositeReadableBufferEqualsBenchmark.equalsToByteBufferReader                                          1     false       8  avgt   10   19.090 ±  0.548   ns/op
CompositeReadableBufferEqualsBenchmark.equalsToByteBufferReader                                          1     false      16  avgt   10   24.537 ±  0.175   ns/op
CompositeReadableBufferEqualsBenchmark.equalsToByteBufferReader                                          1     false      64  avgt   10   57.523 ±  0.647   ns/op
CompositeReadableBufferEqualsBenchmark.equalsToByteBufferReader                                          1      true       8  avgt   10   22.165 ±  0.068   ns/op
CompositeReadableBufferEqualsBenchmark.equalsToByteBufferReader                                          1      true      16  avgt   10   35.275 ±  0.153   ns/op
CompositeReadableBufferEqualsBenchmark.equalsToByteBufferReader                                          1      true      64  avgt   10   82.179 ±  0.196   ns/op
CompositeReadableBufferEqualsBenchmark.equalsToWithSingleArraySubsetOfBuffer                             1     false       8  avgt   10   22.076 ±  0.080   ns/op
CompositeReadableBufferEqualsBenchmark.equalsToWithSingleArraySubsetOfBuffer                             1     false      16  avgt   10   26.474 ±  0.194   ns/op
CompositeReadableBufferEqualsBenchmark.equalsToWithSingleArraySubsetOfBuffer                             1     false      64  avgt   10   59.139 ±  0.410   ns/op
CompositeReadableBufferEqualsBenchmark.equalsToWithSingleArraySubsetOfBuffer                             1      true       8  avgt   10   25.968 ±  0.120   ns/op
CompositeReadableBufferEqualsBenchmark.equalsToWithSingleArraySubsetOfBuffer                             1      true      16  avgt   10   35.986 ±  0.130   ns/op
CompositeReadableBufferEqualsBenchmark.equalsToWithSingleArraySubsetOfBuffer                             1      true      64  avgt   10  100.742 ±  0.395   ns/op

[franz1981]

The changes on this PR (especially those that save creating a slice while decoding should benefit decoding in general :)
I'm now testing this with and end 2 end test using QPID JMS and Artemis 👍

[franz1981]

I've noticed a thing, due to the changes on decoding ie not creating the slice anymore:
I'm getting a huge amount of StringType$1.decode unseen before, see

in violet

[franz1981]

@gemmellr I've tried an end 2 end test but I'm still struggling to evaluate the perf improvement (that's strange, but not unexpected, according to our chats). I'm investigating what's happening ;)