Re: [Vala] How to wrap select(2)'s fd_set?

["Sam Liddicott" <sam liddicott com>]

* David Keijser wrote, On 05/02/09 13:07:
> 2009/2/5 Michael 'Mickey' Lauer <mickey vanille-media de>:
>   
> > Am Friday 30 January 2009 14:26:51 schrieb Michael 'Mickey' Lauer:
> >     
> > > Am Friday 30 January 2009 08:30:43 schrieb Jürg Billeter:
> > >       
> > > > FdSet should be a struct, not a class, we do not need reference type
> > > > semantics here. This eliminates unnecessary heap allocation. FD_ZERO
> > > > should be bound as initialization method as it just zeroes the streuct.
> > > >         
> > > Understood.
> > >
> > >       
> > > > [CCode (cname = "fd_set", cheader_filename = "sys/select.h"]
> > > > public struct FdSet {
> > > >     [CCode (cname = "FD_ZERO")]
> > > >     public FdSet ();
> > > >     ...
> > > > }
> > > >         
> > > With this change, I get a strange C-compilation error:
> > >
> > > /local/pkg/fso/fso-gsm0710muxd/src/multiplexer.c: In
> > > Funktion »multiplexer_writefd«:
> > > /local/pkg/fso/fso-gsm0710muxd/src/multiplexer.c:324: Fehler: expected
> > > expression before »do«
> > > /local/pkg/fso/fso-gsm0710muxd/src/multiplexer.c:325: Fehler: expected
> > > expression before »do«
> > > /local/pkg/fso/fso-gsm0710muxd/src/multiplexer.c:326: Fehler: expected
> > > expression before »do«
> > >
> > > The respective generated code is:
> > >
> > > gboolean multiplexer_writefd (Multiplexer* self, const char* command, gint
> > > fd) {
> > >         fd_set _tmp0 = {0};
> > >         fd_set readfds;
> > >         fd_set _tmp1 = {0};
> > >         fd_set writefds;
> > >         fd_set _tmp2 = {0};
> > >         fd_set exceptfds;
> > >         struct timeval t = {(glong) 1, (glong) 0};
> > >         gint r;
> > >         gboolean _tmp3;
> > >         gssize bwritten;
> > >         g_return_val_if_fail (self != NULL, FALSE);
> > >         g_return_val_if_fail (command != NULL, FALSE);
> > >         readfds = (FD_ZERO (&_tmp0), _tmp0);
> > >         writefds = (FD_ZERO (&_tmp1), _tmp1);
> > >         exceptfds = (FD_ZERO (&_tmp2), _tmp2);
> > >         FD_SET (fd, &writefds);
> > > ...
> > >
> > > I have no idea where this 'do' comes from.
> > >       
> > Ok, I now found out:
> >
> > FD_ZERO is defined as __FD_ZERO which in turn is defined as:
> >
> > /* We don't use `memset' because this would require a prototype and
> >   the array isn't too big.  */
> > #define __FD_ZERO(s) \
> >  do { \
> >    unsigned int __i; \
> >    fd_set *__arr = (s); \
> >    for (__i = 0; __i < sizeof (fd_set) / sizeof (__fd_mask); ++__i) \
> >      __FDS_BITS (__arr)[__i] = 0; \
> >  } while (0)
> >
> > The valac-generated part
> >
> >     
> > > readfds = (FD_ZERO (&_tmp0), _tmp0);
> > >       
> > is a syntax error then.
> >
> > Anything we can do about that or just give in and leaving FdSet a full class?
> >     

use memset ?

> I have this vague memory of there being a way of marking a function as
> a macro in vapi. if not perhaps it should be considered adding it as
> macros are quite common.
>   
Try the gcc extension something like:
readfds = (__extension__ ({ FD_ZERO(&_tmp0)} ), _tmp);

I recall someone saying that generally vala code is compiled by gcc.

However I would like vala to be able to disentangle the long tuples and 
write some human looking code... although it is a nasty job as the 
re-written code can no longer be followed by variable declarations, but 
this is much simpler to read:

FD_ZERO(&_tmp0);
readfds = _tmp0;

I wonder why _tmp0 needed to be used at all - I would have guessed to 
stop readfds accessing an uninitialized struct, but that doesn't make 
sense if readfds starts out as an uninitialized struct...

Sam