Re: time code progress

[Rich Felker <dalias@aerifal.cx>]

On Wed, Jul 17, 2013 at 03:39:35PM +0200, Jens Gustedt wrote:
> Am Mittwoch, den 17.07.2013, 15:19 +0200 schrieb Szabolcs Nagy:
> > note that the problem is not that __VA_ARGS__ is empty
> > (it's not, contrary to what i might implied), but that
> > if n becomes 0 (== __VA_ARGS__ expands to one argument),
> > then there is no more arguments in the __SYSCALL_NARGS_X
> > call to substitute for '...', so a simple fix would be
> > 
> > #define __SYSCALL_NARGS(...) __SYSCALL_NARGS_X(__VA_ARGS__,7,6,5,4,3,2,1,0, tralala)
> 
> right, you always have the name of the syscall as first argument
> 
> > but i'm not sure if this should be fixed (this is internal
> > code and i think there are no 0 argument syscalls)
> > 
> > i just wanted to record how i found the close without fd issue
> > (which shows that some kind of type checking for syscall
> > arguments would help libc hacking.. but that's non-trivial
> > to do)
> 
> if there are really no 0 argument syscalls
> 
> #define __SYSCALL_NARGS(...) __SYSCALL_NARGS_X(__VA_ARGS__,7,6,5,4,3,2,1, tralali, tralala)

There are, e.g. getpid. So I'm not sure why this issue has never come
up before. I'm guessing you have a newer gcc that added a new warning
for it; my gcc does not seem to warn.

Anyway, I thought of the 0-arg issue back when I came up with these
macros, and it was resolved by always having the syscall number. But I
failed to realize there would be no arguments for the ... slot of
__SYSCALL_NARGS_X.

> implement. I could imagine how to check for the number of arguments of
> particular syscalls. Type checking would be more difficult, and would
> probably need some maintenance.

This might be more easily achieved with a grep recipe...

Rich