Add support for common type of int and double?.

[gafter]

See #881

As requested by LDM about a year ago

[gafter]

@MadsTorgersen @dotnet/ldm Could someone please look at this and verify that it is as expected?

[alrz]

There's a prototype available over at alrz/roslyn@645c343.

[alrz]

@gafter Can you please take a look at the above commit to see if it's using the right approach. Thanks.

[gafter]

I will look at it when back from vacation on Monday.

[alrz]

HashSet can handle a null value, in the next iteration I'm going to use null to signify a null literal in the BestType algorithm. btw should this change include target-typing for ?: and ?? expressions? If so I'd be happy to take a stab at it.

[gafter]

@alrz I don't think the approach you are taking addresses the problem described in the title of this issue. See also #881

If we target-type ?? and ?: as we plan to, there will be less motivation for making this change.

[alrz]

It's buried in the diff https://github.com/alrz/roslyn/blob/645c343114b9eec2c069242d231cfce95af1e40c/src/Compilers/CSharp/Test/Semantic/Semantics/NullableEnhancedCommonTypeTests.cs

I think the compiler yields the correct result for both int?/double => double? and double/null => double? however, as I saw in that prototype (at least the way I wrote it), doing either of cases just does not fit into the current way of how things are done, we need to make a lot of exceptions and whatnot.

If we target-type ?? and ?: as we plan to, there will be less motivation for making this change.

makes sense. target-typing of those would have the same effect. 👍