Undo the Const < (uint)span.Length hacks in the BCL

[SingleAccretion]

With #40180 having been merged for some time (many thanks to @nathan-moore!), the if (Const < (uint)span.Length) pattern and its permutations can now be get rid of in the BCL. A quick regex search found the following places:

System.Private.CoreLib\src\System\Buffers\Text\Utf8Parser\Utf8Parser.Boolean.cs:44:    if (4 < (uint)source.Length)
System.Private.CoreLib\src\System\Text\Unicode\Utf8.cs:105:                            if (2 >= (uint)destination.Length)
System.Private.CoreLib\src\System\Text\EncoderFallback.cs:137:                         if (1 < (uint)chars.Length)
System.Private.CoreLib\src\System\Text\Rune.cs:319:                                    if (1 < (uint)source.Length)
System.Private.CoreLib\src\System\Text\Rune.cs:820:                                    if (1 >= (uint)input.Length)

I have manually verified that the codegen for them and the equivalent "natural" versions is the same on the main branch (even without the casts).

cc @GrabYourPitchforks

Do we know of any other places that may have similar things lying around?

[benaadams]

Do we know of any other places that may have similar things lying around?

https://github.com/dotnet/aspnetcore

[jkotas]

Nice!

[GrabYourPitchforks]

Huh, I'm still getting the bad codegen in sharplab (see here). Did the JIT fix miss the 5.0 RTM cutoff?

That said, the change LGTM. But we do need to be mindful that we're not going to regress performance if we change code that's being cross-compiled for many different runtimes: some of which have the fix and some of which don't. It's relevant here because just the other day I started an internal thread asking if we should make an OOB System.Text.Rune package.

[GrabYourPitchforks]

LGTM, but see comment at #49450 (comment).

[benaadams]

Huh, I'm still getting the bad codegen in sharplab (see here). Did the JIT fix miss the 5.0 RTM cutoff?

#40180 was milestoned 6.0 and the last improvement was merged 28 days ago #43568 so I assume so?

[benaadams]

Using @EgorBo's Disasmo with current dotnet/runtime; it looks like the only difference between the two is jbe vs jle

[GrabYourPitchforks]

Argh, I completely forgot about Egor's Disasmo tool! Thank you for the timely reminder. 🚀

[GrabYourPitchforks]

Networking unit test failures are described at #48758 (comment).