superfluous space char in marco

superfluous space char in marco

[Syrone Wong]

For include/net/if.h, we define it as

#define IFF_VOLATILE (IFF_LOOPBACK|IFF_POINTOPOINT|IFF_BROADCAST| \
        IFF_ECHO|IFF_MASTER|IFF_SLAVE|IFF_RUNNING|IFF_LOWER_UP|IFF_DORMANT)

while iproute2 4.10.0 define it as

#define IFF_VOLATILE (IFF_LOOPBACK|IFF_POINTOPOINT|IFF_BROADCAST|IFF_ECHO|\
IFF_MASTER|IFF_SLAVE|IFF_RUNNING|IFF_LOWER_UP|IFF_DORMANT)

in include/linux/if.h. The only difference is the space char after
IFF_BROADCAST.

This causes the redefinition of IFF_VOLATILE.

Warning when compiling iproute2 4.10.0:

In file included from ../include/linux/if_tunnel.h:5:0,
                 from iptunnel.c:24:
../include/linux/if.h:130:0: warning: "IFF_VOLATILE" redefined
 #define IFF_VOLATILE (IFF_LOOPBACK|IFF_POINTOPOINT|IFF_BROADCAST|IFF_ECHO|\

In file included from iptunnel.c:21:0:
/home/wong/github/lede-1/staging_dir/toolchain-arm_cortex-a9+vfpv3_gcc-6.3.0_musl_eabi/include/net/if.h:48:0:
note: this is the location of the previous definition
 #define IFF_VOLATILE (IFF_LOOPBACK|IFF_POINTOPOINT|IFF_BROADCAST| \


Best Regards,
Syrone Wong

Re: superfluous space char in marco

[Rich Felker]

On Sun, Apr 16, 2017 at 03:58:11PM +0800, Syrone Wong wrote:
> For include/net/if.h, we define it as
> 
> #define IFF_VOLATILE (IFF_LOOPBACK|IFF_POINTOPOINT|IFF_BROADCAST| \
>         IFF_ECHO|IFF_MASTER|IFF_SLAVE|IFF_RUNNING|IFF_LOWER_UP|IFF_DORMANT)
> 
> while iproute2 4.10.0 define it as
> 
> #define IFF_VOLATILE (IFF_LOOPBACK|IFF_POINTOPOINT|IFF_BROADCAST|IFF_ECHO|\
> IFF_MASTER|IFF_SLAVE|IFF_RUNNING|IFF_LOWER_UP|IFF_DORMANT)
> 
> in include/linux/if.h. The only difference is the space char after
> IFF_BROADCAST.

Do the other iproute2 include/linux/* headers actually work now? My
experience building it in the past has been that you have to rm -rf
that directory to get it to build at all since it contained outdated
and clashing versions of those headers. Maybe something has changed in
recent versions.

> This causes the redefinition of IFF_VOLATILE.
> 
> Warning when compiling iproute2 4.10.0:
> 
> In file included from ../include/linux/if_tunnel.h:5:0,
>                  from iptunnel.c:24:
> .../include/linux/if.h:130:0: warning: "IFF_VOLATILE" redefined
>  #define IFF_VOLATILE (IFF_LOOPBACK|IFF_POINTOPOINT|IFF_BROADCAST|IFF_ECHO|\
> 
> In file included from iptunnel.c:21:0:
> /home/wong/github/lede-1/staging_dir/toolchain-arm_cortex-a9+vfpv3_gcc-6.3.0_musl_eabi/include/net/if.h:48:0:
> note: this is the location of the previous definition
>  #define IFF_VOLATILE (IFF_LOOPBACK|IFF_POINTOPOINT|IFF_BROADCAST| \

This is odd and looks like a gcc bug. Preprocessor macros definitions
are sequences of PP tokens, not literal code strings, meaning that
whitespace is irrelevant as long as the resulting sequence of PP
tokens remains the same; it's a matching definition not a
redefinition.

Anyone else have thoughts on what's going on?

Rich

Re: superfluous space char in marco

[Alexander Monakov]

On Mon, 17 Apr 2017, Rich Felker wrote:
> This is odd and looks like a gcc bug. Preprocessor macros definitions
> are sequences of PP tokens, not literal code strings, meaning that
> whitespace is irrelevant as long as the resulting sequence of PP
> tokens remains the same; it's a matching definition not a
> redefinition.

No, according to C11 6.10.3 p1 presence of whitespace is significant (its
contents are not), and 6.10.3.5 example 6 corroborates that.  GCC, Clang,
ICC and MSVC on Godbolt all agree with that.

Alexander

Re: superfluous space char in marco

[Rich Felker]

On Mon, Apr 17, 2017 at 06:28:58PM +0300, Alexander Monakov wrote:
> On Mon, 17 Apr 2017, Rich Felker wrote:
> > This is odd and looks like a gcc bug. Preprocessor macros definitions
> > are sequences of PP tokens, not literal code strings, meaning that
> > whitespace is irrelevant as long as the resulting sequence of PP
> > tokens remains the same; it's a matching definition not a
> > redefinition.
> 
> No, according to C11 6.10.3 p1 presence of whitespace is significant (its
> contents are not), and 6.10.3.5 example 6 corroborates that.  GCC, Clang,
> ICC and MSVC on Godbolt all agree with that.

Thanks for clarifying! I had forgotten this detail. It probably makes
sense to change it, then.

Rich

Re: superfluous space char in marco

[Szabolcs Nagy]

* Rich Felker <dalias@libc.org> [2017-04-17 13:02:08 -0400]:
> On Mon, Apr 17, 2017 at 06:28:58PM +0300, Alexander Monakov wrote:
> > On Mon, 17 Apr 2017, Rich Felker wrote:
> > > This is odd and looks like a gcc bug. Preprocessor macros definitions
> > > are sequences of PP tokens, not literal code strings, meaning that
> > > whitespace is irrelevant as long as the resulting sequence of PP
> > > tokens remains the same; it's a matching definition not a
> > > redefinition.
> > 
> > No, according to C11 6.10.3 p1 presence of whitespace is significant (its
> > contents are not), and 6.10.3.5 example 6 corroborates that.  GCC, Clang,
> > ICC and MSVC on Godbolt all agree with that.
> 
> Thanks for clarifying! I had forgotten this detail. It probably makes
> sense to change it, then.

linux uapi header has whitespace befor IFF_MASTER
if we match the definition with something i think
it should be the linux header not iproute2

Re: superfluous space char in marco

[Rich Felker]

On Mon, Apr 17, 2017 at 07:29:19PM +0200, Szabolcs Nagy wrote:
> * Rich Felker <dalias@libc.org> [2017-04-17 13:02:08 -0400]:
> > On Mon, Apr 17, 2017 at 06:28:58PM +0300, Alexander Monakov wrote:
> > > On Mon, 17 Apr 2017, Rich Felker wrote:
> > > > This is odd and looks like a gcc bug. Preprocessor macros definitions
> > > > are sequences of PP tokens, not literal code strings, meaning that
> > > > whitespace is irrelevant as long as the resulting sequence of PP
> > > > tokens remains the same; it's a matching definition not a
> > > > redefinition.
> > > 
> > > No, according to C11 6.10.3 p1 presence of whitespace is significant (its
> > > contents are not), and 6.10.3.5 example 6 corroborates that.  GCC, Clang,
> > > ICC and MSVC on Godbolt all agree with that.
> > 
> > Thanks for clarifying! I had forgotten this detail. It probably makes
> > sense to change it, then.
> 
> linux uapi header has whitespace befor IFF_MASTER
> if we match the definition with something i think
> it should be the linux header not iproute2

Yes, the official Linux one, not iproute2's fork, is what we should
match, if anything.

Rich