Add CollectionsMarshal ref assemblies

[benaadams]

Resolves dotnet/corefx#31597

/cc @tannergooding @ahsonkhan

[ahsonkhan]

nit: add partial

We should also try to auto-generate this if possible.

@safern might be able to help provide guidance here on how for local build of coreclr (outside of https://github.com/dotnet/corefx/blob/master/Documentation/coding-guidelines/updating-ref-source.md)

[stephentoub]

Thanks for addingthis.

What about tests?

Also, presumably there are places we should be using this in corefx? I don't love adding API we can't/don't ourselves use.

[benaadams]

What about tests?

Added

Also, presumably there are places we should be using this in corefx?

Yes though; it will be easier to address that when corefx/coreclr are in sync (post merge).

As an aside this

foreach (T item in CollectionsMarshal.AsSpan(list))
{
    // ...
}

would be faster than this

foreach (T item in list)
{
    // ...
}

for larger lists and same for for loops (as the later has additional bounds checks and enumerator; while the former does not); not sure that would be a desirable change to make?

[benaadams]

Applied feedback

[ahsonkhan]

Looks good.

[safern]

CoreCLR was consumed yesterday. Let's re-run CI.

[safern]

/azp run corefx-ci

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[benaadams]

 The type or namespace name 'List<>' does not exist in the namespace 'System.Collections.Generic' (are you missing an assembly reference?)

Need to check what I should include in the ref assembly

[safern]

Thanks @benaadams now that this is merge are you planning on updating corefx to use this API?

[benaadams]

are you planning on updating corefx to use this API?

Yes, however I have quite a few open PRs that need cleaning up and also am currently short on free time, so it may be a while.

One nice feature (and where InteropServices fits) is getting a pointer and length from a List<T> to then pass into a native call.

As an aside, @stephentoub would this be a dubious performance optimisation? #41306 (comment)

[stephentoub]

As an aside, @stephentoub would this be a dubious performance optimisation? #41306 (comment)

I'd expect the IL to be better, and the generated asm. Is there any situation/condition where it would result in worse code or perf in a microbenchmark (maybe because of the null check if that doesn't get elided)? If not, then it seems reasonable to employ in places where we think there's a reasonable gain to be had, where we expect to be enumerating long lists with minimal work done per element. Where did you have in mind?

[benaadams]

Added a few examples in coreclr dotnet/coreclr#27032