Pattern compilation should not be included in elapsed time

[h12w]

A compiled regular expression is almost always used rather than recompiling each time.
And the measurement should be an average of multiple times.
A benchmark is only useful if it is close enough to real applications.

[mariomka]

Pattern compilation is a necessary part of regex process and not always patterns can be reused. I'm sorry but the benchmark can’t be good for all scenarios.

Anyway, I have tested it again and the results are similar:

Including pattern compilation:

C PCRE2 | 24.59 | 23.55 | 5.55 | 53.69
Rust | 29.02 | 27.97 | 6.28 | 63.28
PHP | 53.38 | 49.97 | 5.33 | 108.67
Javascript | 78.66 | 63.92 | 2.01 | 144.59
D ldc | 142.25 | 135.63 | 4.79 | 282.67
D dmd | 200.00 | 195.35 | 5.70 | 401.06
Perl | 239.81 | 168.08 | 44.65 | 452.54
Crystal | 332.21 | 272.89 | 26.84 | 631.94
Ruby | 338.32 | 290.29 | 51.11 | 679.72
Python PyPy | 204.90 | 171.47 | 323.83 | 700.20
Java | 375.97 | 460.51 | 288.50 | 1124.98
Kotlin | 389.48 | 460.31 | 289.03 | 1138.82
Python 2 | 366.29 | 286.07 | 498.07 | 1150.44
Python 3 | 568.25 | 416.31 | 489.55 | 1474.11
Go | 412.98 | 404.22 | 703.53 | 1520.73
C# .Net Core | 1908.27 | 1654.43 | 109.74 | 3672.44
C# Mono | 2355.97 | 2025.43 | 147.90 | 4529.30

Excluding pattern compilation (if apply):

C PCRE2 | 23.84 | 23.53 | 5.82 | 53.19
Rust | 27.30 | 28.67 | 6.14 | 62.11
PHP | 53.39 | 48.43 | 5.26 | 107.08
Javascript | 77.64 | 62.83 | 1.92 | 142.39
D ldc | 143.50 | 137.39 | 5.08 | 285.97
D dmd | 200.42 | 195.75 | 5.46 | 401.63
Perl | 239.37 | 165.62 | 44.55 | 449.54
Crystal | 335.80 | 274.04 | 26.54 | 636.38
Ruby | 335.38 | 294.44 | 52.37 | 682.19
Python PyPy | 203.62 | 170.96 | 326.98 | 701.57
Java | 375.25 | 453.16 | 290.51 | 1118.91
Python 2 | 361.41 | 275.90 | 492.16 | 1129.46
Kotlin | 385.70 | 465.36 | 287.34 | 1138.40
Python 3 | 548.91 | 403.32 | 481.29 | 1433.52
Go | 412.31 | 405.62 | 700.31 | 1518.25
C# .Net Core | 1907.12 | 1658.22 | 109.74 | 3675.08
C# Mono | 2447.18 | 2096.15 | 156.32 | 4699.65

By the way, you can check the run-benchmarks.php to see that measurement is the average of ten executions, maybe execute more times is more accurate but I think is enough.

[h12w]

I have tested myself and I agree with you. The result does not make significant difference whether including pattern compilation or not.