Regression in Regex construction (and thereby DateFormat) performance on 1.7

[Sacha0]

               _
   _       _ _(_)_     |  Documentation: https://docs.julialang.org
  (_)     | (_) (_)    |
   _ _   _| |_  __ _   |  Type "?" for help, "]?" for Pkg help.
  | | | | | | |/ _` |  |
  | | |_| | | | (_| |  |  Version 1.6.5 (2021-12-19)
 _/ |\__'_|_|_|\__'_|  |
|__/                   |

julia> using Dates

julia> using BenchmarkTools

julia> letters = String(collect(keys(Dates.CONVERSION_SPECIFIERS)))
"MIHdEepmSUYyus"

julia> pattern = "(?<!\\\\)([\\Q$letters\\E])\\1*"
"(?<!\\\\)([\\QMIHdEepmSUYyus\\E])\\1*"

julia> @benchmark Regex($pattern)
BenchmarkTools.Trial: 10000 samples with 6 evaluations.
 Range (min … max):  4.743 μs … 42.846 μs  ┊ GC (min … max): 0.00% … 0.00%
 Time  (median):     4.925 μs              ┊ GC (median):    0.00%
 Time  (mean ± σ):   5.273 μs ±  1.469 μs  ┊ GC (mean ± σ):  0.00% ± 0.00%

  ▇█▅▃▃▁▄▆▅▂▁                                                ▂
  ███████████▇▆▆▆▅▄▅▆▆▄▅▄▃▃▄▃▄▃▄▅▃▃▅▃▃▄▅▄▄▃▄▁▄▄▄▄▄▃▄▅▅▅▅▅▆▇▆ █
  4.74 μs      Histogram: log(frequency) by time     11.4 μs <

 Memory estimate: 32 bytes, allocs estimate: 1.

while

               _
   _       _ _(_)_     |  Documentation: https://docs.julialang.org
  (_)     | (_) (_)    |
   _ _   _| |_  __ _   |  Type "?" for help, "]?" for Pkg help.
  | | | | | | |/ _` |  |
  | | |_| | | | (_| |  |  Version 1.7.0 (2021-11-30)
 _/ |\__'_|_|_|\__'_|  |
|__/                   |

julia> using Dates

julia> using BenchmarkTools

julia> letters = String(collect(keys(Dates.CONVERSION_SPECIFIERS)))
"MIHdEepmSUYyus"

julia> pattern = "(?<!\\\\)([\\Q$letters\\E])\\1*"
"(?<!\\\\)([\\QMIHdEepmSUYyus\\E])\\1*"

julia> @benchmark Regex($pattern)
BenchmarkTools.Trial: 10000 samples with 5 evaluations.
 Range (min … max):   5.517 μs … 202.124 μs  ┊ GC (min … max): 0.00% … 0.00%
 Time  (median):     16.872 μs               ┊ GC (median):    0.00%
 Time  (mean ± σ):   20.625 μs ±  13.905 μs  ┊ GC (mean ± σ):  0.00% ± 0.00%

  ▁   █▆▃▁▂   ▄▅
  █▇▃▅██████▂▄██▇▆▄▃▅▃▄▃▅▄▅▆▄▃▂▂▃▃▃▃▃▃▃▃▃▂▂▂▂▂▂▃▃▂▂▂▂▂▁▁▁▁▁▁▁▁ ▃
  5.52 μs         Histogram: frequency by time         57.9 μs <

 Memory estimate: 32 bytes, allocs estimate: 1.

The particular Regex comes from the body of DateFormat(f::AbstractString, locale::DateLocale=ENGLISH)

julia/stdlib/Dates/src/io.jl

Lines 382 to 383 in 28b47b8

	letters = String(collect(keys(CONVERSION_SPECIFIERS)))
	for m in eachmatch(Regex("(?<!\\\\)([\\Q$letters\\E])\\1*"), f)

through with the regression was observed.

Profiling a function that runs the Regex constructor call above many times in a loop reveals that expansion of sljit_malloc_exec accounts for the difference in time. The first profile below is on 1.6, and the second on 1.7. (Note that for the 1.7 profile, the constructor was called about a quarter the number of times as for the 1.6 profile, yielding a similar number of samples.)

[maleadt]

Bisected to d336d48, #39524, ref #25997. So I guess this is on purpose. cc @vtjnash

[JeffBezanson]

I thought DateFormat(...) is not really expected to be super fast. But, it would be really nice to cache the Regex it uses somehow. It might be enough to quickly check whether anything has been added to CONVERSION_SPECIFIERS.

[Sacha0]

@JeffBezanson meanwhile #43647 😄

[kpamnany]

This one seems as well resolved as it's going to be, right @Sacha0?