Re: [PATCH v3 2/4] ethdev: move error checking macros to header

[Adrien Mazarguil <adrien.mazarguil@6wind.com>]

On Wed, Nov 04, 2015 at 02:10:50PM +0000, Bruce Richardson wrote:
> On Wed, Nov 04, 2015 at 11:24:18AM +0100, Adrien Mazarguil wrote:
> > On Wed, Nov 04, 2015 at 02:19:36AM +0100, Thomas Monjalon wrote:
> > > 2015-11-03 12:00, Bruce Richardson:
> > > > Move the function ptr and port id checking macros to the header file, so
> > > > that they can be used in the static inline functions there. In doxygen
> > > > comments, mark them as for internal use only.
> > > [...]
> > > > +/**
> > > > + * @internal
> > > > + *  Macro to print a message if in debugging mode
> > > > + */
> > > > +#ifdef RTE_LIBRTE_ETHDEV_DEBUG
> > > > +#define RTE_PMD_DEBUG_TRACE(fmt, args...) \
> > > > +		RTE_LOG(ERR, PMD, "%s: " fmt, __func__, ## args)
> > > > +#else
> > > > +#define RTE_PMD_DEBUG_TRACE(fmt, args...)
> > > > +#endif
> > > 
> > > It does not compile because Mellanox drivers are pedantic:
> > > 
> > > In file included from /home/thomas/projects/dpdk/dpdk/drivers/net/mlx4/mlx4.c:78:0:
> > > /home/thomas/projects/dpdk/dpdk/x86_64-native-linuxapp-gcc-shared-next/include/rte_ethdev.h: At top level:
> > > /home/thomas/projects/dpdk/dpdk/x86_64-native-linuxapp-gcc-shared-next/include/rte_ethdev.h:933:38: error: ISO C does not permit named variadic macros [-Werror=variadic-macros]
> > >  #define RTE_PMD_DEBUG_TRACE(fmt, args...) \
> > 
> > I suggest something like the following definitions as a pedantic-proof and
> > standard compliant method (one drawback being that it cannot be done with a
> > single macro), see PMD_DRV_LOG() in drivers/net/mlx5/mlx5_utils.h which also
> > automatically appends a line feed:
> > 
> >  #ifdef RTE_LIBRTE_ETHDEV_DEBUG
> > 
> >  #define STRIP(a, b) a
> >  #define OPAREN (
> >  #define CPAREN )
> >  #define COMMA ,
> > 
> >  #define RTE_PMD_DEBUG_TRACE(...) \
> >          RTE_PMD_DEBUG_TRACE_(__VA_ARGS__ STRIP OPAREN, CPAREN)
> > 
> >  #define RTE_PMD_DEBUG_TRACE_(fmt, ...) \
> >          RTE_PMD_DEBUG_TRACE__(fmt COMMA __func__, __VA_ARGS__)
> > 
> >  #define RTE_PMD_DEBUG_TRACE__(...) \
> >          RTE_LOG(ERR, PMD, "%s: " __VA_ARGS__)
> > 
> >  #else /* RTE_LIBRTE_ETHDEV_DEBUG */
> > 
> >  #define RTE_PMD_DEBUG_TRACE(...)
> > 
> >  #endif /* RTE_LIBRTE_ETHDEV_DEBUG */
> > 
> > STRIP() and other helper macros are used to manage the dangling comma issue
> > when __VA_ARGS__ is empty as in the first call below:
> > 
> >  RTE_PMD_DEBUG_TRACE("foo\n");
> >  RTE_PMD_DEBUG_TRACE("foo %u\n", 42);
> > 
> > -- 
> > Adrien Mazarguil
> > 6WIND
> 
> Is this really the best way around this? It looks like a lot more complicated
> than the original solution. Do we really need to support the -pedantic flag
> in our header files?

I know you didn't ask me directly but I think we should, at least for
exposed/installed headers. What we do internally does not matter, but we
cannot prevent user applications from adding -pedantic if they want to.

The above solution is one I suggest, perhaps it can be done in a different
manner if you think it's too complicated, although it's difficult using
macros only.

-- 
Adrien Mazarguil
6WIND