Re: roland/systemtap branch

[Paul Eggert <eggert at cs dot ucla dot edu>]

On 05/24/2012 03:18 PM, Carlos O'Donell wrote:
> Two thumbs up from me.

I also think this sort of patch has a high reward-to-risk ratio
and would like to see something like it go in.

Re this:

   #  define LIBC_PROBE(name, n, ...)             DUMMY_PROBE##n (__VA_ARGS__)
   ...
   /* This silliness lets us evaluate all the arguments for each arity
      of probe.  My kingdom for a real macro system.  */

   # define DUMMY_PROBE0()                do {} while (0)
   # define DUMMY_PROBE1(a1)              do {} while ((void) (a1), 0)
   # define DUMMY_PROBE2(a1, a2)          do {} while ((void) (a1), \
                                                       (void) (a2), 0)
   # define DUMMY_PROBE3(a1, a2, a3)      do {} while ((void) (a1), \
                                                       (void) (a2), \
                                                       (void) (a3), 0)
   ... and so on for DUMMY_PROBE4 etc. ...

How about something like the following instead?  It's still silly, but
arguably it's less silly, and it's indubitably shorter.

#define LIBC_PROBE(name, n, ...)                                        \
  do {                                                                  \
    _Bool probe_args[] = { 0, ## __VA_ARGS__ };                         \
    _Bool verify_arg_count[sizeof probe_args == n + 1 ? 1 : -1];        \
    (void) verify_arg_count;                                            \
  } while (0)

I tried to rewrite it to avoid the need for 'n' entirely.  I can do this
for the dummy case, but not for the USE_STAP_PROBE case as the <sys/sdt.h>
stuff insists on a literal decimal constant for 'n'; so I stopped with the
above.