Currently, when I have a number of Task instances where I want to perform aggregate operations across all of them (waiting for all of them to complete, for any of them to complete, etc.), I like to use the static WhenAll, WhenAny, etc. methods.

I also like to use ValueTask<T> as well for areas where I don't need the heap allocation that a Task<T> brings.

That said, I can't currently use WhenAll, WhenAny, etc. because they take sequences of Task<T>. While there are workarounds, all of them fall short in some way:

• I could call AsTask, but then that allocates a Task<T> and call the existing methods on Tasks but the allocations is the situation I was trying to avoid in the first place.
• I could write my own, but the coreclr implementation is quite robust, and I'd certainly miss situations that are coded for in there.

This specific example is for ValueTask<T> but it really applies to any awaitable type.

I also think we'd need a new signature for WhenAny; the current implementation returns a Task, and the pattern to figure out which task finished is to do an equality comparison on the reference returned. I don't really agree that checking the result as ValueTask<T> does is the right way (imagine waiting on signals for true/false results from multiple votes, checking equality on the result won't work here).