Optimize replace_first, replace_last, and truncate

[danielmpetrov]

Leverage the string.Concat with ReadOnlySpan<char> - see https://learn.microsoft.com/en-us/dotnet/api/system.string.concat?view=netcore-3.1#system-string-concat(system-readonlyspan((system-char))-system-readonlyspan((system-char))), to remove extra allocations. Since the overload was added in .NET Core 3.0, the optimization can only be applied thereafter (unless we shim the method).

The following scenarios were benchmarked for replace_first and replace_last

{{ "a a a a" | replace_first: "a", "b" }}
{{ "c c c c" | replace_first: "a", "b" }}
{{ "a a a a" | replace_last: "a", "b" }}
{{ "c c c c" | replace_last: "a", "b" }}

BenchmarkDotNet=v0.13.2, OS=Windows 10 (10.0.19044.2130/21H2/November2021Update)
AMD Ryzen 5 2600X, 1 CPU, 12 logical and 6 physical cores
.NET SDK=6.0.402
  [Host]     : .NET 6.0.10 (6.0.1022.47605), X64 RyuJIT AVX2
  DefaultJob : .NET 6.0.10 (6.0.1022.47605), X64 RyuJIT AVX2

Method	Input	Mean	Error	StdDev	Ratio	RatioSD	Gen0	Allocated	Alloc Ratio
ReplaceFirst_Span	a a a a	45.70 ns	0.727 ns	0.645 ns	1.00	0.00	0.0172	72 B	1.00
ReplaceFirst_String	a a a a	86.56 ns	1.418 ns	1.184 ns	1.89	0.04	0.0267	112 B	1.56
ReplaceFirst_Span	c c c c	16.55 ns	0.301 ns	0.282 ns	1.00	0.00	-	-	NA
ReplaceFirst_String	c c c c	55.64 ns	0.437 ns	0.387 ns	3.36	0.06	-	-	NA
ReplaceLast_Span	a a a a	43.55 ns	0.542 ns	0.507 ns	1.00	0.00	0.0172	72 B	1.00
ReplaceLast_String	a a a a	84.02 ns	1.271 ns	1.127 ns	1.93	0.02	0.0267	112 B	1.56
ReplaceLast_Span	c c c c	15.98 ns	0.153 ns	0.143 ns	1.00	0.00	-	-	NA
ReplaceLast_String	c c c c	62.40 ns	0.708 ns	0.662 ns	3.91	0.06	-	-	NA

The following scenarios were benchmarked for truncate

{{ "The cat came back the very next day" | truncate: 13 }}

BenchmarkDotNet=v0.13.2, OS=Windows 10 (10.0.19044.2130/21H2/November2021Update)
AMD Ryzen 5 2600X, 1 CPU, 12 logical and 6 physical cores
.NET SDK=6.0.402
  [Host]     : .NET 6.0.10 (6.0.1022.47605), X64 RyuJIT AVX2
  DefaultJob : .NET 6.0.10 (6.0.1022.47605), X64 RyuJIT AVX2

Method	Input	Mean	Error	StdDev	Ratio	RatioSD	Gen0	Allocated	Alloc Ratio
Truncate_New	The c(...)t day [35]	44.13 ns	0.890 ns	0.832 ns	1.00	0.00	0.0191	80 B	1.00
Truncate_Old	The c(...)t day [35]	52.49 ns	0.471 ns	0.418 ns	1.19	0.03	0.0306	128 B	1.60

[hishamco]

Could you please add the benchmark to the PR?

[danielmpetrov]

That's a bit of a problem, since the benchmark project is not multi targeted, I copy-pasted the implementation twice just to evaluate it. The code is not good to be merged.

[danielmpetrov]

I cleaned up the benchmarks and pushed them. I think they're good to verify the benefits, but should probably not merge them into the main branch.

[sebastienros]

I don't see the point of keeping these benchmarks. And I don't like the fact that the old implementation is in them. My preference is to not have them here. You used them to prove the new implementation is better, good job, that's sufficient. I don't expect someone else to reuse them. We could have a set of benchmarks for all current implementations the day we start tracking it.

[hishamco]

Of course he add them to prove it's better ;), if that's sufficient @danielmpetrov please remove them, then we could merge

Thanks a lot for your contribution