RE: [Patch, ObjC/C++] Recognize attributes on class, category and protocol declarations.

["Nicola Pero" <nicola dot pero at meta-innovation dot com>]

I haven't looked at the details, but the patch looks very cool :-)

I wonder about the warning messages though.  They mention "ObjC Version 1",
but I'm not even sure I know myself what "ObjC Version 1" is ;-)

I would suggest we use something more neutral (for now), such as

 class attributes are ignored (not supported by the compiler)

We'll need to classify and document the Objective-C features as we're going to be
adding all the new ones, so we may want to revisit the message once we have done
some classification (which should probably go in the gcc/doc/standards.texi).

Thanks

-----Original Message-----
From: "IainS" <developer@sandoe-acoustics.co.uk>
Sent: Sunday, 26 September, 2010 16:04
To: "GCC Patches" <gcc-patches@gcc.gnu.org>
Cc: "Mike Stump" <mrs@gcc.gnu.org>, "Joseph S. Myers" <joseph@codesourcery.com>, "Jason Merrill" <jason@redhat.com>, "Nicola Pero" <nicola.pero@meta-innovation.com>
Subject: [Patch, ObjC/C++] Recognize attributes on class, category and protocol declarations.

Hi,

The patch implements the parsing for the prefix attributes on class,  
category and protocol in both ObjC and ObjC++ FEs and hands off the  
result to the shared code in objc-act (which simply raises a warning  
that the attribute is currently ignored).

This is part of the support needed for (eventually) ObjC V1 and (as  
soon as possible) to parse Application/Foundation headers on NeXT  
platforms >= Darwin9.

Nicola came up with the approach and started the patch off - I tweaked  
the c parts and added the c++ , objc-act and testsuite bits.

We feel that this approach is useful in that:
  (a) the remainder of the implementation can be local to  objc/  and..
  (b) instead of a fatal syntax error, we now have a warning which may  
often be ignored safely on m32 NeXT (the main use in system headers is  
to warn of deprecated interfaces that become unavailable at m64).

The source locations shown in diagnostics could be better, but that's  
for a different patch.

Tested on i686-darwin9, x86_64-unknown-linux-gnu, and on a cross to  
mipsisa64-elf

We need approval for c, c-family, c++ and objc.
OK to apply?

Iain

gcc/c-family:

	* c-common.h (objc_start_class_interface): Adjust prototype.
	(objc_start_category_interface): Likewise.
	(objc_start_protocol): Likewise.
	* stub-objc.c (objc_start_protocol): Adjust for extra argument.
	(objc_start_class_interface): Likewise.
	(objc_start_category_interface): Likewise.

gcc/objc:

	* objc-act.c (objc_start_class_interface): Handle and ignore  
attributes.
	(objc_start_category_interface): Likewise.
	(objc_start_protocol): Likewise.

gcc/cp:

	* parser.c (objc_at_keyword_follows_attributes): New.
	(cp_parser_declaration): Parse prefix attributes for ObjC++.
	(cp_parser_objc_protocol_declaration): Handle attributes.
	(cp_parser_objc_class_interface): Likewise.
	(cp_parser_objc_declaration): Likewise.

gcc:

	* c-parser.c (c_parser_objc_class_definition): Adjust prototype.
	(c_parser_objc_protocol_definition): Likewise.
	(c_parser_external_declaration): Provide dummy attribute arguments.
	(c_parser_declaration_or_fndef): Parse prefix attributes for ObjC.
	(c_parser_objc_class_definition): Handle attributes.
	(c_parser_objc_protocol_definition): Likewise.

gcc/testsuite:

	* objc.dg/attributes: New.
	* objc.dg/attributes/attributes.exp: New.
	* objc.dg/attributes/class-attribute-1.m: New.
	* objc.dg/attributes/class-attribute-2.m: New
	* objc.dg/attributes/categ-attribute-1.m: New
	* objc.dg/attributes/categ-attribute-2.m: New
	* objc.dg/attributes/proto-attribute-1.m: New

	* obj-c++.dg/attributes: New.
	* obj-c++.dg/attributes/attributes.exp: New
	* obj-c++.dg/attributes/class-attribute-1.mm: New
	* obj-c++.dg/attributes/class-attribute-2.mm: New
	* obj-c++.dg/attributes/categ-attribute-1.mm: New
	* obj-c++.dg/attributes/categ-attribute-2.mm: New
	* obj-c++.dg/attributes/proto-attribute-1.mm: New