[PATCH wayland v2] client: Fully flush during blocking dispatch

Jonas Ådahl jadahl at gmail.com
Thu Jan 14 17:42:40 PST 2016


On Thu, Jan 14, 2016 at 04:59:48PM -0600, Derek Foreman wrote:
> On 11/01/16 10:31 PM, Jonas Ådahl wrote:
> > wl_display_flush() may fail with EAGAIN which means that not all data
> > waiting in the buffer has been flushed. We later block until there
> > data to read, which could mean that we block on input from the
> > compositor without having sent out all data from the client. Avoid this
> > by fully flushing the socket before starting to wait.
> 
> Is this because wl_connection_flush() uses sendmsg with MSG_DONTWAIT?
> 
> Can we figure out a way to get "bool blocking" down to
> wl_connection_flush and avoid the poll thing entirely?  Maybe have a
> wl_display_blocking_flush() or whatever...
> 
> Wouldn't even need to be external API if only wl_display_dispatch_queue
> calls it...

Part of the reason for this series is to make
wl_display_dispatch_queue() use the public prepare-read API in order to
provide a proper example of how the API is supposed to be used, so I
don't think we should use a private wl_display_blocking_flush() function
here for that reason. Adding more API for a blocking flush() I'm not
sure is worth the trouble.


Jonas

> 
> Thanks,
> Derek
> 
> > Signed-off-by: Jonas Ådahl <jadahl at gmail.com>
> > ---
> >  src/wayland-client.c | 39 +++++++++++++++++++++++++++++----------
> >  1 file changed, 29 insertions(+), 10 deletions(-)
> > 
> > diff --git a/src/wayland-client.c b/src/wayland-client.c
> > index 8bf6124..f47e395 100644
> > --- a/src/wayland-client.c
> > +++ b/src/wayland-client.c
> > @@ -1442,6 +1442,21 @@ wl_display_cancel_read(struct wl_display *display)
> >  	pthread_mutex_unlock(&display->mutex);
> >  }
> >  
> > +static int
> > +wl_display_poll(struct wl_display *display, short int events)
> > +{
> > +	int ret;
> > +	struct pollfd pfd[2];
> > +
> > +	pfd[0].fd = display->fd;
> > +	pfd[0].events = events;
> > +	do {
> > +		ret = poll(pfd, 1, -1);
> > +	} while (ret == -1 && errno == EINTR);
> > +
> > +	return ret;
> > +}
> > +
> >  /** Dispatch events in an event queue
> >   *
> >   * \param display The display context object
> > @@ -1485,27 +1500,31 @@ WL_EXPORT int
> >  wl_display_dispatch_queue(struct wl_display *display,
> >  			  struct wl_event_queue *queue)
> >  {
> > -	struct pollfd pfd[2];
> >  	int ret;
> >  
> >  	if (wl_display_prepare_read_queue(display, queue) == -1)
> >  		return wl_display_dispatch_queue_pending(display, queue);
> >  
> > +	while (true) {
> > +		ret = wl_display_flush(display);
> > +		if (ret == -1 && errno == EAGAIN) {
> > +			if (wl_display_poll(display, POLLOUT) == -1) {
> > +				wl_display_cancel_read(display);
> > +				return -1;
> > +			}
> > +		} else {
> > +			break;
> > +		}
> > +	}
> > +
> >  	/* Don't stop if flushing hits an EPIPE; continue so we can read any
> >  	 * protocol error that may have triggered it. */
> > -	ret = wl_display_flush(display);
> > -	if (ret < 0 && errno != EAGAIN && errno != EPIPE) {
> > +	if (ret < 0 && errno != EPIPE) {
> >  		wl_display_cancel_read(display);
> >  		return -1;
> >  	}
> >  
> > -	pfd[0].fd = display->fd;
> > -	pfd[0].events = POLLIN;
> > -	do {
> > -		ret = poll(pfd, 1, -1);
> > -	} while (ret == -1 && errno == EINTR);
> > -
> > -	if (ret == -1) {
> > +	if (wl_display_poll(display, POLLIN) == -1) {
> >  		wl_display_cancel_read(display);
> >  		return -1;
> >  	}
> > 
> 


More information about the wayland-devel mailing list