Add tests for new Random API and for the distinct unseeded path

[danmoseley]

Tests for changes in dotnet/runtime#47390

cc @colgreen

[danmoseley]

BenchmarkDotNet=v0.12.1.1466-nightly, OS=Windows 10.0.19042
Intel Core i7-10510U CPU 1.80GHz, 1 CPU, 8 logical and 4 physical cores
.NET SDK=6.0.100-preview.1.21075.9
[Host] : .NET 6.0.0 (6.0.21.7403), X64 RyuJIT
Job-CFOLDV : .NET 6.0.0 (6.0.21.7403), X64 RyuJIT

PowerPlanMode=00000000-0000-0000-0000-000000000000 Arguments=/p:DebugType=portable IterationTime=250.0000 ms
MaxIterationCount=20 MinIterationCount=15 WarmupCount=1

Method	Mean	Error	StdDev	Median	Min	Max	Gen 0	Gen 1	Gen 2	Allocated
ctor	266.672 ns	34.0699 ns	39.2349 ns	269.363 ns	203.979 ns	351.375 ns	0.0163	-	-	72 B
ctor_seeded	1,180.033 ns	202.4371 ns	233.1268 ns	1,171.310 ns	806.414 ns	1,609.254 ns	0.0727	-	-	312 B
Next_int	31.807 ns	1.5263 ns	1.6965 ns	31.997 ns	29.147 ns	35.128 ns	-	-	-	-
Next_int_unseeded	12.383 ns	0.7647 ns	0.7853 ns	12.044 ns	11.532 ns	14.451 ns	-	-	-	-
Next_int_int	15.260 ns	0.8519 ns	0.9115 ns	15.455 ns	13.209 ns	16.281 ns	-	-	-	-
Next_int_int_unseeded	13.295 ns	0.5646 ns	0.6275 ns	12.951 ns	12.849 ns	15.017 ns	-	-	-	-
Next_long	148.044 ns	2.2936 ns	2.1455 ns	147.844 ns	145.846 ns	151.795 ns	-	-	-	-
Next_long_unseeded	8.163 ns	0.0936 ns	0.0782 ns	8.126 ns	8.096 ns	8.374 ns	-	-	-	-
Next_long_long	227.569 ns	53.0941 ns	61.1432 ns	193.907 ns	177.535 ns	342.688 ns	-	-	-	-
Next_long_long_unseeded	16.942 ns	0.6433 ns	0.7151 ns	16.899 ns	15.933 ns	18.531 ns	-	-	-	-
NextBytes	9,092.806 ns	697.9947 ns	746.8465 ns	8,789.110 ns	8,379.180 ns	11,109.769 ns	-	-	-	-
NextBytes_unseeded	353.442 ns	6.5098 ns	6.3935 ns	350.555 ns	347.158 ns	370.248 ns	-	-	-	-
NextDouble	14.138 ns	0.2560 ns	0.2138 ns	14.213 ns	13.725 ns	14.430 ns	-	-	-	-
NextDouble_unseeded	3.877 ns	0.1161 ns	0.1140 ns	3.890 ns	3.720 ns	4.119 ns	-	-	-	-

[colgreen]

Couple of nits:

• You flipped from ctor_seeded to Next_int_ unseeded :)
• There is a NextSingle() method now too.

[danmoseley]

You flipped from ctor_seeded to Next_int_ unseeded :)

Yes, this is because I want to avoid changing benchmark names, so we don't break history. And unfortunately the ctor was the exception in that it measured unseeded. (@adamsitnik I assume there isn't an attribute that would allow me to have consistent method names while still being recorded under the old test name...)

There is a NextSingle() method now too.

I'll add.

[danmoseley]

OK, should be done now.

I added a bunch of #if's for new API since it seemed more readable in this case than separate files which is what we mostly seem to do in this repo.

[danmoseley]

@billwert some CI legs are failing with Not found PathtoPublish: /__w/1/s/artifacts/log/x64_release/5.0.1xx_roslyn .. or Not found PathtoPublish: D:\a\1\s\artifacts\log\x86_release\5.0.1xx_micro .. known? Can I ignore them?

[billwert]

It looks like the actual failure was that installation of the SDK was not reliable this afternoon. (For example). These are probably fine to ignore. I've got a work item to fix this up.

[danmoseley]

@stephentoub any more feedback ? As noted above, I can change to 3 partial classes, but I think this is one of those cases where #if actually makes it much clearer what's going on.

[danmoseley]

Thanks @billwert.. would retrying help or not?

[stephentoub]

As noted above, I can change to 3 partial classes, but I think this is one of those cases where #if actually makes it much clearer what's going on.

I'll defer to @adamsitnik. I prefer the partial files approach, but if the preferred scheme for this repo is ifdefs, that's fine.

[danmoseley]

Merging with CI failures due to missing SDK on some machines, noted by @billwert

@adamsitnik if you believe partial classes would be clearer, LMK and I'll put up a new PR.

[adamsitnik]

I assume there isn't an attribute that would allow me to have consistent method names while still being recorded under the old test name.

You are right. Perhaps we should create such an attribute?

I added a bunch of #if's for new API since it seemed more readable in this case than separate files which is what we mostly seem to do in this repo.

Great!

@danmosemsft big thanks for adding the benchmarks!

@colgreen very nice improvements!