On Wed, Aug 17, 2016 at 11:26:51AM +0200, Markus Armbruster wrote:
> Peter Xu <pet...@redhat.com> writes:
>
> > On Tue, Aug 16, 2016 at 08:17:08PM +0800, Peter Xu wrote:
> >> > >
> >> > > Also, please make it a function, not a macro:
> >> > >
> >> > > void error_report_fatal(const char *fmt, ...)
> >> > > {
> >> > > va_list ap;
> >> > >
> >> > > va_start(ap, fmt);
> >> > > error_vreport(fmt, ap);
> >> > > va_end(ap);
> >> > > exit(1);
> >> > > }
> >
> > Marcel (and reviewers),
> >
> > Now if we are having both error_report_fatal() and
> > error_report_abort(), we'll write error_report() three times if we all
> > take them as functions.
>
> Yes, but all we duplicate is the usual var-arg boilerplate.
>
> > How about we still use macro this time but leverage error_setg()
> > macro as mentioned by Fam, like:
> >
> > +#define error_report_fatal(...) error_setg(&error_fatal, __VA_ARGS__)
> > +#define error_report_abort(...) error_setg(&error_abort, __VA_ARGS__)
> >
> > In this case, we avoided calling exit() directly in the macro, and is
> > much cleaner than writting error_report() content for three times.
>
> I'm afraid that destroys the layering.
>
> Currently, Error objects (util/error.c) use error reporting
> (util/qemu-error.c), but not vice versa. Their headers
> (include/qapi/error.h and include/qemu/error-report.h) are idependent.
> I like it that way.
I see. Then let me keep them functions.
Thanks,
-- peterx
