From: Pete B. <pb...@gm...> - 2010-12-02 00:38:35
|
On 2010.12.02 00:09, Peter Stuge wrote: >> Now, properly including poll.h without breaking things is going to >> require some time, > > Actually autoconf makes it trivial. But I don't think this is the > problem for Cygwin. > > Do we want to use Cygwin's poll()? I think not - I'm not sure if it > will do what we want, or perform as well as poll_windows. I would be > happy with poll_windows also for Cygwin. What do you say? I think you're right, we're not using cygwin's poll in the first place, so that's why we have the issue. I kind of stopped my initial investigation of that issue when I got puzzled about the fact that there wasn't a special case for the inclusion of poll_windows.h in Makefile.am, and I should have made the link by then. So the issue then is that configure does detect that cygwin has poll, and nfds_t, but we've made sure we didn't use poll. That's actually why I had to define nfds_t in poll_windows.h (which the commit removed). So now we have to add a special case for cywgin. We could undef POLL_NFDS_TYPE and redef it to unsigned int in poll_windows.h, but that would look quite ugly. Better do it in configure.ac and get it defined right in the first place... Regards, /Pete |