int to long performance hit?

[benjchristensen]

PR #1443 changing request(int n) to request(long n) appears to have caused a non-trivial performance hit in the JMH tests.

Before:

Benchmark                                    (size)   Mode   Samples        Score  Score error    Units
r.o.OperatorMergePerf.merge1SyncStreamOfN         1  thrpt         5  5246591.736   240800.312    ops/s
r.o.OperatorMergePerf.merge1SyncStreamOfN      1000  thrpt         5    57309.810     3333.567    ops/s
r.o.OperatorMergePerf.merge1SyncStreamOfN   1000000  thrpt         5       60.163        1.654    ops/s

After:

Benchmark                                    (size)   Mode   Samples        Score  Score error    Units
r.o.OperatorMergePerf.merge1SyncStreamOfN         1  thrpt         5  4940791.384   317585.017    ops/s
r.o.OperatorMergePerf.merge1SyncStreamOfN      1000  thrpt         5    27655.680     2216.176    ops/s
r.o.OperatorMergePerf.merge1SyncStreamOfN   1000000  thrpt         5       30.091        1.015    ops/s

[benjchristensen]

Found it ... it was changing from -1 to Long.MAX_VALUE and one of the fast-path conditionals was missed (it was checking for > 0 rather than == -1 so was missed.