[pulseaudio-discuss] [PATCH] Use local icon for zeroconf shared devices

[Tanu Kaskinen]

On Fri, 2016-08-26 at 18:22 +0200, Sylvain Baubeau wrote:
> systemd-hostnamed provides an icon for the machine it is running on.
> With this patch, module-zeroconf-publish uses this icon for the
> 'icon-name' attribute in the Avahi properties. module-zeroconf-discover
> passes this icon to module-tunnel using the module parameter
> {sink/source}_properties.
> 
> This allows to display a portable, desktop or phone instead of
> the generic sound card icon.

This also overrides the device icon when it's *not* a generic sound
card icon. But I guess this is in most cases better, since the good
icon names tend to be associated with ports, not sinks.

> @@ -243,6 +248,7 @@ static void resolver_cb(
>          pa_xfree(dname);
>          pa_xfree(args);
>          pa_xfree(if_suffix);
> +        pa_xfree(properties);

There are code paths that don't free the properties, causing memory
leaks.

> @@ -393,7 +399,7 @@ int pa__init(pa_module*m) {
>      u->avahi_poll = pa_avahi_poll_new(m->core->mainloop);
>  
>      if (!(u->client = avahi_client_new(u->avahi_poll, AVAHI_CLIENT_NO_FAIL, client_callback, u, &error))) {
> -        pa_log("pa_avahi_client_new() failed: %s", avahi_strerror(error));
> +        pa_log("avahi_client_new() failed: %s", avahi_strerror(error));

I'd prefer separate patches for unrelated fixes.

> +#define DBUS_HOSTNAME_ID "org.freedesktop.hostname1"

This is used as the interface name in D-Bus messages, so
DBUS_HOSTNAME_INTERFACE would be a better name.

Also, I would prefer "HOSTNAME_DBUS_" as the prefix, because these
strings are defined by systemd-hostnamed, not by D-Bus. (Sorry for
nitpicking...)

> +static char *get_icon_name(pa_module*m) {
> +    const char *interface = DBUS_HOSTNAME_ID;
> +    const char *property = DBUS_HOSTNAME_ICON_PROPERTY;
> +    char *icon_name;
> +    pa_dbus_connection *bus;
> +    DBusError error;
> +    DBusMessageIter args;
> +    DBusMessage *msg = NULL;
> +    DBusMessage *reply = NULL;
> +    DBusConnection *conn = NULL;
> +    DBusMessageIter sub;
> +
> +    if (!(bus = pa_dbus_bus_get(m->core, DBUS_BUS_SYSTEM, &error))) {
> +        pa_log("Failed to get session bus connection: %s", error.message);

System/session mismatch between the code and the error message.

> +        goto out;
> +    }
> +
> +    conn = pa_dbus_connection_get(bus);
> +
> +    msg = dbus_message_new_method_call(DBUS_HOSTNAME_ID,
> +                                       DBUS_HOSTNAME_PATH,
> +                                       "org.freedesktop.DBus.Properties",
> +                                       "Get");
> +    dbus_message_append_args(msg, DBUS_TYPE_STRING, &interface, DBUS_TYPE_STRING, &property, DBUS_TYPE_INVALID);
> +
> +    dbus_error_init(&error);
> +    if ((reply = dbus_connection_send_with_reply_and_block(conn, msg, -1, &error)) == NULL) {

I'd rather not do blocking IPC. If you move this to the Avahi thread,
that would be more tolerable (I'd rather not do even that, but this
module is a lost cause anyway, because the communication with Avahi
looks like it's done in a blocking manner). It looks like the
AVAHI_CLIENT_S_RUNNING handler in client_callback() would be a suitable
place to call get_icon_name().

> +        pa_log("Failed to send: %s:%s\n", error.name, error.message);
> +        dbus_error_free(&error);
> +        goto out;
> +    }
> +
> +    dbus_message_iter_init(reply, &args);
> +    if (dbus_message_iter_get_arg_type(&args) != DBUS_TYPE_VARIANT) {
> +        pa_log("Incorrect reply type\n");
> +        goto out;
> +    }
> +
> +    dbus_message_iter_recurse(&args, &sub);
> +    dbus_message_iter_get_basic(&sub, &icon_name);

We should check the variant type before assuming that it's a string.

-- 
Tanu