[PATCH wayland] client: prevent using and creating proxies after an error occured

[Pekka Paalanen]

On Fri, 18 Jul 2014 17:56:36 +0200
Marek Chalupa <mchqwerty at gmail.com> wrote:

> We won't receive any response anyway...
> 
> This removes the last line of:
> 
> [2230782.435] wl_display at 1.error(wl_display at 1, 2, "no memory")
> wl_display at 1: error 2: no memory
> [2230782.534]  -> wl_display at 1.sync(new id wl_callback at 3)
> 
> and prevents from creating proxy on invalidated display object.
> ---
>  src/wayland-client.c | 6 ++++++
>  1 file changed, 6 insertions(+)
> 
> diff --git a/src/wayland-client.c b/src/wayland-client.c
> index ee2215d..2f52900 100644
> --- a/src/wayland-client.c
> +++ b/src/wayland-client.c
> @@ -312,6 +312,9 @@ wl_proxy_create(struct wl_proxy *factory, const struct wl_interface *interface)
>  	struct wl_display *display = factory->display;
>  	struct wl_proxy *proxy;
>  
> +	if (display->last_error)
> +		return NULL;
> +
>  	pthread_mutex_lock(&display->mutex);
>  	proxy = proxy_create(factory, interface);
>  	pthread_mutex_unlock(&display->mutex);
> @@ -523,6 +526,9 @@ wl_proxy_marshal_array_constructor(struct wl_proxy *proxy,
>  	struct wl_proxy *new_proxy = NULL;
>  	const struct wl_message *message;
>  
> +	if (proxy->display->last_error)
> +		return NULL;
> +
>  	pthread_mutex_lock(&proxy->display->mutex);
>  
>  	message = &proxy->object.interface->methods[opcode];

Hi,

I think last_error needs to be checked under the display mutex.

I'm also not so sure about this whole change, because I believe apps
expect the wl_proxy creation to succeed, and assume that if it fails,
it is due to memory allocation failure. The reponse to memory
allocation failure may be just abort(), while the response to a
protocol error could include saving current work to a temporary file.

AFAICS, the only way to get NULL right now is a memory allocation
error, and I think that things like glib just abort if malloc ever
fails, so it is conceivable Gtk and apps will do that too. Whether they
actually handle protocol errors differently, I don't know, but at least
a protocol error is much more recoverable than a malloc failure.

Creating more proxies after the wl_display is already disconnected
should not hurt, right?

I need a second opinion before taking this in. Am I making any sense?


Thanks,
pq