Support __is_standard_layout and __is_literal_type for gcc compatibility

[llvmbot]

Bugzilla Link	9584
Resolution	FIXED
Resolved on	May 20, 2011 10:04
Version	trunk
OS	All
Blocks	#9955
Reporter	LLVM Bugzilla Contributor
CC	@chandlerc,@DougGregor

Extended Description

$ cat test2.cc
void foo() {
__is_standard_layout(int);
}
$ ./clang++ -std=gnu++0x -c test2.cc
test2.cc:2:29: error: expected '(' for function-style cast or type construction
__is_standard_layout(int);
~~~^
1 error generated.
$ g++-mp-4.6 -std=gnu++0x -c test2.cc
$

[llvmbot]

Also:

$ cat test2.cc
void foo() {
__is_literal_type(int);
}
$ ./clang++ -std=gnu++0x -c test2.cc
test2.cc:2:26: error: expected '(' for function-style cast or type construction
__is_literal_type(int);
~~~^
1 error generated.
$ g++-mp-4.6 -std=gnu++0x -c test2.cc
$

[chandlerc]

Clang already has partial support for '__is_literal' rather than '__is_litearl_type'.

We need to arrive at some consensus as to what the name should be. As soon as I know that, I'll fix the name (if need be) and finish the implementation.

Jeffrey, could you take this up w/ GCC folks and get a prompt response? Certainly, '__is_literal' fits more naturally with the other names:

__is_trivial
__is_trivially_copyable
__is_pod
__is_abstract

and so on, no '_type' suffix. So I'm inclined to suggest GCC change its name to '__is_literal' if possible...

[chandlerc]

Actually, none of these type traits you've requested Clang add for libstdc++ are documented in GCC's own documentation: http://gcc.gnu.org/onlinedocs/gcc/Type-Traits.html

(Nor are they mentioned in the latest 4.6.0 copy of that document)

I'm going to implement these based on the assumption that their intent is to precisely model the definition in the standard, but it'd be really good to get these documented by GCC if their standard library is going to start depending on them.

I'll also work on adding them to Clang's documentation once everything is implemented and tested.

[chandlerc]

r130078 and r130079 implement the rest of __is_literal (and support for GCC's spelling of __is_literal_type) to the extent possible in Clang currently.

[llvmbot]

In C++0x, there's a type trait defined in [meta.unary.prop] named "is_literal_type", and none named "is_literal", so I suspect Clang's __ trait should also be named __is_literal_type. IIRC, there was some discussion in the committee about whether to end this trait with "_type", so I don't think it was an accident.

I'll file a gcc bug about the missing documentation.

[llvmbot]

Filed http://gcc.gnu.org/#48748 .

[DougGregor]

These type traits are supported in Clang trunk.