Loading time benchmark is inaccurate

[alexeyten]

Library loading time depends on the order and list of modules that loaded in benchmark/loading-runner.js

If I leave only chalk, kleur/colors, colorette and picocolors then loading times are:

~/picocolors ⬢ v16 ₪ main*= $ node benchmarks/loading
  chalk          7.546 ms
  kleur/colors   2.442 ms
  colorette      2.746 ms
+ picocolors     0.819 ms
~/picocolors ⬢ v16 ₪ main*= $ node benchmarks/loading
  chalk          7.926 ms
  kleur/colors   2.532 ms
  colorette      2.900 ms
+ picocolors     0.867 ms
~/picocolors ⬢ v16 ₪ main*= $ node benchmarks/loading
  chalk          7.351 ms
  kleur/colors   2.344 ms
  colorette      2.743 ms
+ picocolors     0.811 ms

but simply adding nanocolors shaves off 0.1 ms:

~/picocolors ⬢ v16 ₪ main*= $ node benchmarks/loading
  chalk          7.178 ms
  kleur/colors   2.304 ms
  colorette      2.694 ms
  nanocolors     1.200 ms
+ picocolors     0.720 ms
~/picocolors ⬢ v16 ₪ main*= $ node benchmarks/loading
  chalk          7.174 ms
  kleur/colors   2.337 ms
  colorette      2.715 ms
  nanocolors     1.227 ms
+ picocolors     0.711 ms
~/picocolors ⬢ v16 ₪ main*= $ node benchmarks/loading
  chalk          6.862 ms
  kleur/colors   2.183 ms
  colorette      2.563 ms
  nanocolors     1.136 ms
+ picocolors     0.685 ms

and putting picocolors right after kleur/colors will double loading time (also see that colorette time drops)

~/picocolors ⬢ v16 ₪ main*= $ node benchmarks/loading
  chalk          6.831 ms
  kleur/colors   2.195 ms
+ picocolors     1.499 ms
  colorette      1.767 ms
  nanocolors     1.023 ms
~/picocolors ⬢ v16 ₪ main*= $ node benchmarks/loading
  chalk          6.688 ms
  kleur/colors   2.133 ms
+ picocolors     1.486 ms
  colorette      1.702 ms
  nanocolors     1.027 ms
~/picocolors ⬢ v16 ₪ main*= $ node benchmarks/loading
  chalk          6.720 ms
  kleur/colors   2.185 ms
+ picocolors     1.511 ms
  colorette      1.767 ms
  nanocolors     1.027 ms

[alexeyraspopov]

Thanks for testing it out and reporting the issue! I've opened #36 to possibly make this benchmark more reliable.