Bug 104339 – Make it possible to use other types as well as class and struct as an argument of messages.

• Component: WebKit2
• Status: RESOLVED INVALID

[Dongseong Hwang]

This patch changes forward_declarations_and_headers function to include headers
for types in special_cases set, not to make forward declarations. We can use
typedef as an argument now. It means we can use RefPtr as an argument if we
define typedef from RefPtr.

This patch applies this special handling to CoordinatedLayerID and
CoordinatedImageBackingID.

[Dongseong Hwang]

Created attachment 178155 [details]
Patch

[Darin Adler]

Anders, as the creator and architect of the WebKit2 message system, is this the direction you’d like to go with this?

[Dongseong Hwang]

I make this change to use RefPtr as an argument.
You can see the example in https://bugs.webkit.org/attachment.cgi?id=177979&action=review

I want this change to match your direction :)

[Anders Carlsson]

(In reply to comment #2)
> Anders, as the creator and architect of the WebKit2 message system, is this the direction you’d like to go with this?

No. I think that we shouldn't use typedefs in .message.in files, because they can hide the real size of the underlying data. (This is why we always use explicitly sized types as message parameters).

In addition, I'd also like to completely get rid of hard-coded types in the message generator by extending the syntax where possible.

[Dongseong Hwang]

(In reply to comment #4)
> (In reply to comment #2)
> > Anders, as the creator and architect of the WebKit2 message system, is this the direction you’d like to go with this?
> 
> No. I think that we shouldn't use typedefs in .message.in files, because they can hide the real size of the underlying data. (This is why we always use explicitly sized types as message parameters).

I don't understand what you mean. We can not know the real size from forward declaration too, and to know the real size we should include headers.
The problem I had is the c++ does not support forward declaration of typedef.

btw, I agree that only POD struct is natural to use it as an argument of messages, but current default argument is class, not struct.
And I agree that you don't want each class to change messages.py

> In addition, I'd also like to completely get rid of hard-coded types in the message generator by extending the syntax where possible.

Currently, there are three special case including headers.
1. in the case that type name does not match to header name. for example, class FooRef in Foo.h
2. in the case that type is struct, as mentioned.
3. nested class.

Do you mean that all classes, which we want to use an argument of messages, must be in the same header, and the type must be class? although we must create a redundant class to wrap the type that we want to use an argument. If yes, I'll give up.