Differentiate NativeTypedef structs that exist in the headers vs. only in metadata

[mikebattista]

Fixed #1533.

If we agree to just introduce a different attribute for metadata-only typedefs, we need to close on the name. I started with NoNativeTypedef. Another option might be MetadataTypedef.

Projections that want to use all the typedefs should just check for both attributes. Projections that want to stay true to the Win32 headers can unwrap MetadataTypedefs.

We could also merge the two attributes like Typedef(Native = true) and Typedef(Native = false).

[riverar]

I like the idea of a separate MetadataTypedefAttribute instead. This will allow projection authors to quickly filter out this attribute (if desired). The current implementation requires looking up every single TypedefAttribute to read/compare against its Native value.

[mikebattista]

Can you clarify? These MetadataTypedefs would still be applied to functions, so projections need to inspect an attribute and then unwrap values either way whether it's a different attribute or a single attribute with a property.

[riverar]

The .custom declaraction on, say, HANDLE, after this PR, may look something like:

.custom instance void [Windows.Win32.winmd]Windows.Win32.Foundation.Metadata.TypedefAttribute::.ctor(bool) = (
  01 00 01 00 00
)

A projection author wishing to ignore non-native typedefs will have fully parse the attribute, perhaps something like:

for each attribute in attributes
  if attribute == TypedefAttribute
    if attribute.valueType == bool && getBoolValue(attribute) != true
      continue;
    // ...

The proposed alternative eliminates a bunch of per-attribute work by simply using the type name:

for each attribute in attributes
  if attribute == MetadataTypedef
    continue;
  // ...

[mikebattista]

Ok. I'm fine with that to make the processing more efficient. Your earlier comment suggested maybe a different design altogether for the attribute and how it would be applied.

[riverar]

@mikebattista Oh didn't mean to imply that. Was just agreeing with your "Another option might be MetadataTypedef" 👍