Date and DateTime parsing slower than 1.7 and much slower than MRI

[mjc]

string = "2016-02-03T17:44:04Z"

Benchmark.ips do |x|
  x.config warmup: 90 if RUBY_PLATFORM == 'java'
  x.report("Date#parse") { Date.parse(string) }
  x.report("DateTime#parse") { DateTime.parse(string) }
end

1.7.23:

          Date#parse     23.823k (± 4.3%) i/s -    119.444k
      DateTime#parse     17.959k (±10.8%) i/s -     89.750k

9.0.5.0:

          Date#parse     14.201k (±11.8%) i/s -     70.884k
      DateTime#parse     11.330k (±10.1%) i/s -     57.246k

MRI 2.3.0

          Date#parse     61.387k (± 4.6%) i/s -    308.966k
      DateTime#parse     60.220k (± 6.5%) i/s -    301.658k

[kares]

9K still doing noticeably worse than 1.7, invented a new benchmark script :

require 'benchmark'
require 'date'

string = "2016-02-03T17:44:04Z"

Benchmark.bmbm do |x|
  x.report("Date#parse") { 100_000.times{ Date.parse(string) } }
  x.report("DateTime#parse") { 100_000.times { DateTime.parse(string) } }
end

2.3.3

Rehearsal --------------------------------------------------
Date#parse       2.170000   0.000000   2.170000 (  2.173531)
DateTime#parse   2.230000   0.000000   2.230000 (  2.238035)
----------------------------------------- total: 4.400000sec

                     user     system      total        real
Date#parse       2.250000   0.000000   2.250000 (  2.244404)
DateTime#parse   2.300000   0.000000   2.300000 (  2.295370)

1.7.25

Rehearsal --------------------------------------------------
Date#parse      11.970000   0.040000  12.010000 (  6.722000)
DateTime#parse  12.220000   0.020000  12.240000 (  8.419000)
---------------------------------------- total: 24.250000sec

                     user     system      total        real
Date#parse       5.010000   0.010000   5.020000 (  4.920000)
DateTime#parse   7.040000   0.010000   7.050000 (  6.988000)

9.1.7.0

Rehearsal --------------------------------------------------
Date#parse      16.450000   0.120000  16.570000 ( 10.357322)
DateTime#parse  17.600000   0.050000  17.650000 ( 11.852176)
---------------------------------------- total: 34.220000sec

                     user     system      total        real
Date#parse      10.110000   0.020000  10.130000 (  9.630812)
DateTime#parse  10.350000   0.010000  10.360000 ( 10.310710)

[headius]

More work needed here than we have time for in 9.1.8.0, but I want to work on this for 9.2. I'll re-target.

Note that I had some patches in #1662 that seemed to help, and there's a bit of Java integration happening here that we don't optimize as well as we could (e.g. indy does not bind directly but still goes through reflection).

[kares]

believe work at #4635 (+ more commits already on master) resolves this to much acceptable numbers