Re: [RFC 00/17] Merge event loop implementations

[Tom Tromey <tom@tromey.com>]

>>>>> "Eli" == Eli Zaretskii <eliz@gnu.org> writes:

Eli> The problem here is that Windows' implementation of 'select' works
Eli> only on sockets, and wants HSOCKET handles instead of file
Eli> descriptors.  If you feed it a file descriptor, it won't work, even if
Eli> the descriptor is for a socket.

Thanks.

I am still not sure what to do.  Move the mingw-hdep select code to
common?  Or is it better to reintroduce gdb_fildes_t and use it
everywhere?  The former seems simpler I suppose, but the way that the
mingw-hdep select implementation relies on readline gives me pause.
Though perhaps that code will be removed in the readline upgrade, when I
get back to that?  So one idea might be to try to land that first.

Eli> I cannot really tell, by reading the patches, what code would
Eli> gdbserver on Windows use after this series, so I don't know whether it
Eli> will be broken or not.  But one thing I do see is that mingw-hdep.c
Eli> doesn't support waiting on write descriptors, whereas gdbserver's
Eli> event-loop.c seems to support that.

Both event loops claim to support it, but in practice it isn't actually
exposed via the API -- the only exposed API is add_file_handler, which
doesn't allow for requesting a write notification.  All the write stuff
is dead code.

One other thing I forgot to mention is that maybe it would be nice to
just remove gdb's event loop entirely in favor of something like
libevent.  Though of course it is a pain to introduce a new dependency.

Tom