Re: [PATCH v3 2/3] qapi: Do not generate empty enum

[Eric Blake]

On Tue, Mar 21, 2023 at 03:19:28PM +0000, Daniel P. Berrangé wrote:
> On Tue, Mar 21, 2023 at 03:31:56PM +0100, Philippe Mathieu-Daudé wrote:
> > On 16/3/23 15:57, Markus Armbruster wrote:
> > > Daniel P. Berrangé <berrange@redhat.com> writes:
> > > 
> > > > On Thu, Mar 16, 2023 at 01:31:04PM +0100, Markus Armbruster wrote:
> > > > > Philippe Mathieu-Daudé <philmd@linaro.org> writes:
> > > > > 
> > > > > > Per the C++ standard, empty enum are ill-formed. Do not generate
> > > 
> > > The C standard.  The C++ standard doesn't apply here :)
> > 
> > I can't find how empty enums are considered by the C standard...
> 
> The C standard doesn't really matter either.
> 
> What we actually care about is whether GCC and CLang consider the
> empty enums to be permissible or not. or to put it another way...
> if it compiles, ship it :-)

But it doesn't compile:

$ cat foo.c
typedef enum Blah {
} Blah;
int main(void) {
  Blah b = 0;
}
$ gcc -o foo -Wall foo.c
foo.c:2:1: error: empty enum is invalid
    2 | } Blah;
      | ^
foo.c: In function ‘main’:
foo.c:4:5: error: unknown type name ‘Blah’; use ‘enum’ keyword to refer to the 
type
    4 |     Blah b = 0;
      |     ^~~~
      |     enum 
foo.c:4:10: warning: unused variable ‘b’ [-Wunused-variable]
    4 |     Blah b = 0;
      |          ^

So we _do_ need to avoid creating an enum with all members optional in
the configuration where all options are disabled, if we want that
configuration to compile.  Or require that all QAPI enums have at
least one non-optional member.

-- 
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org