Re: always poll, even without poll(2)

Il giorno lun, 22/12/2008 alle 23.16 -0500, James K. Lowden ha scritto:
> Frediano Ziglio wrote:
> > 
> > Mmmm... I would put fakepoll.h in include/replacements directory. Also
> > there is similar function named tds_REPLACED so perhaps it would be
> > better to call this function tds_poll for coherence. Also for
> > coherence in include/replacements.h I used defines like
> > 
> > #define poll(a,b,c) tds_poll(a,b,c)
> > 
> > if poll is not defined. You could include
> > include/replacements/fakepoll.h in include/replacements.h or just
> > include <replacements/fakepoll.h> when used.  
> 
> Will do.  
> 
> I think this is wrong, though:
> 
> > Sat Dec 20 20:11:46 CET 2008
> > 	* src/replacements/fakepoll.c: 
> >		fix win32 portability issue
> 
> http://msdn.microsoft.com/en-us/library/ms740141(VS.85).aspx
> 
> Microsoft says they *ignore* nfds (parameter 1 of select(2)).  So it's OK
> to pass any value to Win32's select() function.  
> 
> More important, we should check against FD_SETSIZE even on Win32: 
> 
> 	"The variable FD_SETSIZE determines the maximum number of descriptors in
> a set. (The default value of FD_SETSIZE is 64, which can be modified by
> defining FD_SETSIZE to another value before including Winsock2.h.)"  
> 
> but the count, not the maxfd, is what matters.  
> 
> I updated fakepoll.c according to that theory.  I left in the maxfd
> computation because preprocessor directives make code hard to read and
> understand.  The fewer we have, the better, even if it means a few extra
> cycles.  
> 

I saw the commit

What do you think about

@@ -72,37 +72,38 @@
        /*
         * Transcribe flags from the poll set to the fd sets.
         * Also ensure we don't exceed select's capabilities.
         */
        for (p = fds; p < endp; p++) {
                /* Negative fd checks nothing and always reports zero */
                if (p->fd < 0) {
                        continue;
                }

+#if defined(WIN32)
+               /* Win32 cares about the number of descriptors, not the highest one. */
+               ++maxfd;
+#else
                if (p->fd > maxfd)
                        maxfd = p->fd;
+#endif

                /* POLLERR is never set coming in; poll(2) always reports errors */
                /* But don't report if we're not listening to anything at all. */
                if (p->events & POLLIN)
                        FD_SET(p->fd, &fdsr);
                if (p->events & POLLOUT)
                        FD_SET(p->fd, &fdsw);
                if (p->events != 0)
                        FD_SET(p->fd, &fdsp);
        }

-#if defined(WIN32)
-       /* Win32 cares about the number of descriptors, not the highest one. */
-       maxfd = nfds;
-#endif
        /*
         * If any FD is too big for select(2), we need to return an error.
         * Which one, though, is debatable.  There's no defined errno for
         * this for poll(2) because it's an "impossible" condition;
         * there's no such thing as "too many" FD's to check.
         * select(2) returns EINVAL, and so do we.
         * EFAULT might be better.
         */
        if (maxfd > FD_SETSIZE) {
                assert(FD_SETSIZE > 0);


Frediano