Re: [gpsd-dev] Implementing pselect() in gpsctl (bug #35765) - what is wrong with my understanding of code?

[Eric S. Raymond]

Andrew Evdokimov <address@hidden>:
> Hi all,
> 
> I'm currently working on implementing pselect() for gpsctl (see bug #35765) 
> but have faced some difficulties understanding how this program currently 
> works. gpsctl has the get_packet() function that is supposed to return a 
> packet mask when gpsd_poll() will return one with ONLINE_SET set. Current 
> code is as follows (only significant lines are left):
> 
> {
>     static fd_set rfds;
>     gps_mask_t fieldmask;
>     struct timeval tv;
> 
>     FD_ZERO(&rfds);
>     for (;;) {
>       FD_CLR(session->gpsdata.gps_fd, &rfds);
> ...
>       tv.tv_sec = 2;
>       tv.tv_usec = 0;
>       errno = 0;
>       if (select(session->gpsdata.gps_fd + 1, &rfds, NULL, NULL, &tv) == -1) {
>           if (errno == EINTR || !FD_ISSET(session->gpsdata.gps_fd, &rfds))
>               continue;
>           gpsd_report(LOG_ERROR, "select %s\n", strerror(errno));
>           exit(2);
>       }
> 
>       fieldmask = gpsd_poll(session);
> ...
>       if ((fieldmask &~ ONLINE_SET)!=0)
>           return fieldmask;
>     }
> }
> 
> So before entering a for-loop we're zeroing fd_set rfds (OK with
> this), then in a loop we're removing 'session->gpsdata.gps_fd' file
> descriptor from already-empty fd_set and then calling select() with
> 2-seconds timeout that will wait until some of our file descriptors
> will become ready for IO. As soon as rfds does not contain any file
> descriptors the select() call will timeout in 2 seconds passing
> control to gpsd_poll() etc. In other words, select() here is just a
> 2-second sleep. gpsd_poll() does not use our rfds thus at the end of
> a loop we have an empty fd_set which will be cleared of
> session->gpsdata.gps_fd at next iteration etc...
>
> Where am I wrong? I'm new to this project and may not fully
>  understand some internal logic, so please help. Thanks in advance!

Sorry about the belated response.  I think I see what confused you.
The FD_ISSET macro doesn't mutate rfds, just tests to see if 
session->gpsdata.gps_fd is in it.
-- 
                <a href="http://www.catb.org/~esr/";>Eric S. Raymond</a>