RFC: g++ compiler flag usage

RFC: g++ compiler flag usage

[Mws]

[-- Attachment #1: Type: text/plain, Size: 696 bytes --]

hi,

as i am using -Werror and also -Wold-style-cast and many more flags,
i do run into some problems when compiling code,
that uses e.g. the FD_SET(int, fd_set*) makro.

of course, i can ommit the -Wold-style-cast or the -Werror flag, and everything will compile fine,
just showing up some warnings. but, i feel very uncomfortable with that situation.

isn't there a possibility e.g to have a 

#ignore Wflags
FD_SET(...,...);
#endignore

that would be helpfull for more people than only me, i think.

having different compiler flags depending on what _can_ be checked within a project is much more
confusing imho.

any opinions? or did i miss something?

regards
marcel

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: RFC: g++ compiler flag usage

[Eljay Love-Jensen]

Hi Marcel,

FD_SET is a C macro function that happens to work in C++.

>any opinions? or did i miss something?

Yes:  #undef the FD_SET macro function, and create the C++ equivalent that complies to the rigorous settings you are using.

Alternatively, use FD_SET in C code, and think compile your C++ code against your C code thunk layer.

HTH,
--Eljay

Re: RFC: g++ compiler flag usage

[Mws]

[-- Attachment #1: Type: text/plain, Size: 2011 bytes --]

On Monday 06 June 2005 13:56, Eljay Love-Jensen wrote:
> Hi Marcel,
> 
> FD_SET is a C macro function that happens to work in C++.
> 
> >any opinions? or did i miss something?
> 
> Yes:  #undef the FD_SET macro function, and create the C++ equivalent that complies to the rigorous settings you are using.
> 
> Alternatively, use FD_SET in C code, and think compile your C++ code against your C code thunk layer.


yes it works, but not with the -W flags i mentioned.

e.g. compilation will fail on x86 but not on ppc

(given i as fd and rfds as fd_set)

FD_SET(i, &rfds);

is expanded on x86 architecture to 

__asm__ __volatile__ ("btsl %1,%0" : "=m" ((&rfds)->fds_bits)[((i) / (8 * sizeof (__fd_mask)))]) : "r" (((int) (i)) % (8 * sizeof (__fd_mask))) : "cc","memory");

and  on mpc/ppc architecture to

(((&rfds)->fds_bits)[((i->first) / (8 * sizeof (__fd_mask)))] |= ((__fd_mask) 1 << ((i->first) % (8 * sizeof (__fd_mask)))));



as i want to have ( as that seems to be possible without using supporting warning flags ) platform independ code, 
i do not like the alternative you offered me, cause i would have to redesign a given wheel :)

would you like to read a 

#ifdef X86VERSION
	__asm__ __volatile__ ("btsl %1,%0" : "=m" ((&rfds)->fds_bits)[((i) / (8 * sizeof (__fd_mask)))]) : "r" ((static_cast<int>(i)) % (8 * sizeof (__fd_mask))) : "cc","memory");
#endif
#ifdef PPCVERSION
	(((&rfds)->fds_bits)[((i->first) / (8 * sizeof (__fd_mask)))] |= ((__fd_mask) 1 << ((i->first) % (8 * sizeof (__fd_mask)))));
#endif
#ifdef MIPSVERSION
	......
#endif

i think the advantages is that you just can just write 

FD_SET(i, &rfds);

instead having those constructs mentioned above.

i do not want to do a compilers work, so i think if there isn't a applicable of FD_SET(...,...) in c++ and you
are forced to use c functions - the compiler should be 
a) able to detect it on his own, or
b) let the coder declare critic sections

regards
marcel






[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: RFC: g++ compiler flag usage

[Eljay Love-Jensen]

Hi Marcel,

>i do not want to do a compilers work...

C++ in general (or GCC's g++ in particular) do not provide FD_SET.

FD_SET is provided by the system.

>...so i think if there isn't a applicable of FD_SET(...,...) in c++ and you are forced to use c functions - the compiler should be
a) able to detect it on his own, or
b) let the coder declare critic sections

Since the compiler does not provide FD_SET, I do not think (a) is a reasonable expectation.

I've already provided you with a suggestion for how to utilize FD_SET in a C file, and then call that extern "C" routine in your C++ program.  That C thunk layers acts as a critical section, for your (b) expectation.

If you really want to use C++, then you could put the FD_SET in a C++ source file that is compiled with different flags from the rest of your source code.  That separate source file on its own acts as a critical section.

Sincerely,
--Eljay