[PATCH] rtems: Improve RTEMS_NO_RETURN attribute

[Sebastian Huber]

On 06/10/2020 17:55, Joel Sherrill wrote:

>
>
> On Tue, Oct 6, 2020 at 10:41 AM Sebastian Huber 
> <sebastian.huber at embedded-brains.de 
> <mailto:sebastian.huber at embedded-brains.de>> wrote:
>
>     On 06/10/2020 17:34, Joel Sherrill wrote:
>
>     >
>     > On Tue, Oct 6, 2020 at 9:46 AM Sebastian Huber
>     > <sebastian.huber at embedded-brains.de
>     <mailto:sebastian.huber at embedded-brains.de>
>     > <mailto:sebastian.huber at embedded-brains.de
>     <mailto:sebastian.huber at embedded-brains.de>>> wrote:
>     >
>     >     On 06/10/2020 16:37, Joel Sherrill wrote:
>     >
>     >     > How does it break the API? It looks like a change of
>     annotation.
>     >     > Shouldn't be a big deal unless I am misunderstanding
>     something.
>     >
>     >     Please have a look at:
>     >
>     > https://lists.rtems.org/pipermail/devel/2020-October/062577.html
>     >
>     >
>     > Looks like basically the same code. What does this break? Why
>     the concern?
>
>     The C11/C++11 attribute syntax is less flexible compared to the GNU
>     attributes. For example:
>
>     echo "_Noreturn void ok(void);" > test.c
>
>     echo "void error(void) _Noreturn;" >> test.c
>
>     gcc test.c
>
>     test.c: In function ‘error’:
>     test.c:2:1: error: ‘_Noreturn’ in empty declaration
>       void error(void) _Noreturn;
>       ^~~~
>     test.c:2:18: warning: empty declaration
>       void error(void) _Noreturn;
>                        ^~~~~~~~~
>     test.c:2:1: error: expected ‘{’ at end of input
>       void error(void) _Noreturn;
>       ^~~~
>
>
> Yuck. I can see why they did the placement kind of. Makes it like const
> or static but it uses an odd keyword with leading _ and capital letter
> which is rather insane.
All new keywords in C are like this. You can include <stdnoreturn.h> to 
get something without the underscore.
>
> What's the positive value for doing this? Does llvm or some static
> analyser like it better?
The positive thing is being in line with the standard.
>
> C11 has improvements but it also has more than one thing I
> question if they really thought through. :(

Yes, I guess not every C compiler vendor liked the GNU attribute stuff. 
C++11 has a similar issue:

[[noreturn]] void ok(void);
void error(void) [[noreturn]];

g++ test.cc
test.cc:2:29: warning: attribute ignored [-Wattributes]
  void error(void) [[noreturn]];
                              ^
test.cc:2:29: note: an attribute that appertains to a type-specifier is 
ignored