REFL003 incorrectly assumes that .GetType() returns only the type seen at compile-time

[jnm2]

0.1.16-dev

void M(IEnumerable<object> objectImplementingIEnumerable)
{
    objectImplementingIEnumerable.GetType().GetMethod("Add");
}

REFL003 The type System.Collections.Generic.IEnumerable<object> does not have a member named Add.

The analyzer should bail if it sees a GetType() on an expression of a virtual type since it can know nothing about the expected concrete type which may be internal to an external lib.

Additionally, if it sees a GetType() on a nonvirtual type, maybe it should bail anyway because it will be reporting #169 in all these cases anyway.

---

Actual code, in case you wonder why you would do this:

        public static IReadOnlyCollection<Assembly> GetAssemblies(this ITypeDiscoveryService typeDiscoveryService)
        {
            var serviceType = typeDiscoveryService.GetType();
            if (serviceType.FullName == "Microsoft.VisualStudio.Design.VSTypeResolutionService") // Achieve instantaneous load, vs around a second's freeze
            {
                var normalEntries = (IEnumerable<object>)serviceType.GetField("_normalEntries", BindingFlags.Instance | BindingFlags.NonPublic).GetValue(typeDiscoveryService);
                // ...

[JohanLarsson]

We should at least change it to warn REFL009. Maybe REFL009 should be default info?

[jnm2]

Another problem that will be resolved once GetType bails when it can't prove the concrete type:

obj.GetType().Assembly.GetType("Foo")

REFL037 The type does not exist

Yes it does; I've just checked obj.GetType().FullName.

[jnm2]

REFL009 would be fine, but I would probably disable it because it says more about the IDE's metadata settings and the limitations of static analysis than about the types I'm using.