Optimized std.datetime.date.DateTime from string methods

[JackStouffer]

In the vein of #5413 and #5404

[dlang-bot]

Thanks for your pull request, @JackStouffer!

Bugzilla references

Your PR doesn't reference any Bugzilla issue.

If your PR contains non-trivial changes, please reference a Bugzilla issue or create a manual changelog.

[JackStouffer]

Speed results

# old
$ git checkout master
$ dmd/src/dmd -O -inline -release test.d && ./test
result	6 secs, 83 ms, 931 μs, and 1 hnsec

# new
$ git checkout DateTime-fromstring
$ dmd/src/dmd -O -inline -release test.d && ./test
result	1 sec, 746 ms, and 330 μs

code

import std.stdio;
import std.algorithm;
import std.conv;
import std.traits;
import std.datetime;

enum testCount = 5_000_000;


void main()
{
    auto result = to!Duration(benchmark!(() => DateTime.fromISOString("20100704T070612"))(testCount)[0]);

    writeln("result", "\t", result);
}

[jmdavis]

Does it work to use byCodeUnit here? Presumably, the speed-up is from getting rid of allocating the dstring, but you're introducing more auto-decoding in the process. If byCodeUnit doesn't work here at the moment (e.g. due to a function still requiring actual strings), then this is still clearly better, but I'd prefer to have byCodeUnit used if we can.

[JackStouffer]

@jmdavis None of the functions here use auto decoding. I've already removed it from Date.fromISOString and TimeOfDay.fromISOString, and countUntil with a single needle forwards to find, which already deals with auto-decoding.

[jmdavis]

I was thinking that strip and countUntil both did, but if they don't, then that's not an issue (though actually, I should probably know for strip, since I did a rewrite of that at one point - but I guess that it was long enough ago that I forgot).

[JackStouffer]

@jmdavis Thanks!