Strings are not the fastpath

[sdboyer]

This pulls over a commit from #7 , because strings are not the fastpath. This one change makes a crazy amount of difference:

With absolute numbers, that's:

benchmark                                  old ns/op     new ns/op     delta
BenchmarkCheckVersionRange-8               1987          22.1          -98.89%
BenchmarkCheckVersionUnion-8               2047          31.2          -98.48%
BenchmarkValidateVersionRange-8            1998          30.7          -98.46%
BenchmarkCheckVersionCaret-8               1000          17.7          -98.23%
BenchmarkCheckVersionTilde-8               1032          19.1          -98.15%
BenchmarkCheckVersionUnary-8               1012          19.8          -98.04%
BenchmarkValidateVersionCaret-8            1006          25.3          -97.49%
BenchmarkValidateVersionTilde-8            1023          27.6          -97.30%
BenchmarkValidateVersionUnary-8            1034          29.0          -97.20%
BenchmarkCheckVersionWildcard-8            1054          42.0          -96.02%
BenchmarkValidateVersionWildcard-8         1086          50.5          -95.35%
BenchmarkValidateVersionRangeFail-8        3278          1238          -62.23%
BenchmarkValidateVersionUnion-8            3340          1336          -60.00%
BenchmarkValidateVersionTildeFail-8        2294          1257          -45.20%
BenchmarkValidateVersionCaretFail-8        2288          1272          -44.41%
BenchmarkValidateVersionUnaryFail-8        2266          1274          -43.78%
BenchmarkValidateVersionUnionFail-8        4627          2610          -43.59%
BenchmarkValidateVersionWildcardFail-8     2325          1314          -43.48%

And the memory allocation:

BenchmarkCheckVersionUnary-8               10             0              -100.00%
BenchmarkCheckVersionTilde-8               10             0              -100.00%
BenchmarkCheckVersionCaret-8               10             0              -100.00%
BenchmarkCheckVersionWildcard-8            10             0              -100.00%
BenchmarkCheckVersionRange-8               20             0              -100.00%
BenchmarkCheckVersionUnion-8               20             0              -100.00%
BenchmarkValidateVersionUnary-8            10             0              -100.00%
BenchmarkValidateVersionUnaryFail-8        20             10             -50.00%
BenchmarkValidateVersionTilde-8            10             0              -100.00%
BenchmarkValidateVersionTildeFail-8        20             10             -50.00%
BenchmarkValidateVersionCaret-8            10             0              -100.00%
BenchmarkValidateVersionCaretFail-8        20             10             -50.00%
BenchmarkValidateVersionWildcard-8         10             0              -100.00%
BenchmarkValidateVersionWildcardFail-8     20             10             -50.00%
BenchmarkValidateVersionRange-8            20             0              -100.00%
BenchmarkValidateVersionRangeFail-8        30             10             -66.67%
BenchmarkValidateVersionUnion-8            30             10             -66.67%
BenchmarkValidateVersionUnionFail-8        40             20             -50.00%

So yeah, drop to zero allocs, plus a two order of magnitude speedup. I'll worry about optimizing the errors to not actually evaluate strings over in #7, but avoiding string evaluation is the single most important performance change we can make.

[mattfarina]

@sdboyer great catch. Benchmarks for the win!