Replacing data with newtype shaves off some numbers from the benchmark. Before (showing only mean), the lowest means out of three runs (the numbers for the first benchmark about parsing omitted, as they aren't changed):
mean 14.89 ms (14.86 ms .. 14.94 ms)
mean 6.550 ms (6.493 ms .. 6.700 ms)
mean 3.246 ms (3.204 ms .. 3.332 ms)
mean 16.88 ms (16.85 ms .. 16.93 ms)
mean 16.42 ms (16.30 ms .. 16.74 ms)
mean 15.28 ms (15.26 ms .. 15.30 ms)
After the change, the best out of three:
mean 14.09 ms (14.03 ms .. 14.24 ms)
mean 6.204 ms (6.148 ms .. 6.344 ms)
mean 3.059 ms (3.067 ms .. 3.202 ms)
mean 15.97 ms (16.13 ms .. 16.28 ms)
mean 14.98 ms (14.88 ms .. 15.40 ms)
mean 14.61 ms (14.62 ms .. 14.73 ms)
So, it's roughly 5-10% improvement.
Curiously, the other change, replacing the lambda with a ., also improves the run time and memory consumption of my own project (which uses hakyll, which uses pandoc, which uses skylighting). Overall, my own project runs faster by about 7% and consumes about 7% less peak RAM.
jgm
commented
1 year ago
Replacing
datawithnewtypeshaves off some numbers from the benchmark. Before (showing onlymean), the lowest means out of three runs (the numbers for the first benchmark about parsing omitted, as they aren't changed):After the change, the best out of three:
So, it's roughly 5-10% improvement.
Curiously, the other change, replacing the lambda with a
., also improves the run time and memory consumption of my own project (which uses hakyll, which uses pandoc, which uses skylighting). Overall, my own project runs faster by about 7% and consumes about 7% less peak RAM.