or-pattern narrowed type is too restrictive

[alrz]

Currently all of these patterns only check for int and > 0

        if (o is (long or int) and > 0) // (!(o is long) && o is int && (int)o > 0)
        if (o is (byte or int) and > 0) // (!(o is byte) && o is int && (int)o > 0)
        if (o is (int or long) and > 0) // (o is int && (int)o > 0)
        if (o is (int or byte) and > 0) // (o is int && (int)o > 0)

Flipping types changes the codegen due to implicit conversions, but effectively all of the checks are wrong. (sharplab)

I would expect we generate the code equivalent to o is A and > 0 or B and > 0 regardless of the order of types (we will emit > twice, one per each type).

Any pattern that is sensitive to the input type would be affected, this includes list patterns.

        if (list is (char[] or string) and { '1', '2', '3' })

Note that the list pattern codegen for each type would be different, but per current spec, the narrowed type here is object and the code won't compile.

The alternative is that we repeat the pattern for each type.

        if (list is char[] { '1', '2', '3' } or string { '1', '2', '3' })
        if (o is byte and > 0 or int and > 0)