[PATCH 6/8] libqmi-glib, device: port qmi_device_release_client to use GTask
Aleksander Morgado
aleksander at aleksander.es
Tue Apr 18 13:58:36 UTC 2017
Hey Ben,
See below.
On 12/04/17 22:15, Ben Chan wrote:
> ---
> src/libqmi-glib/qmi-device.c | 56 ++++++++++++++------------------------------
> 1 file changed, 18 insertions(+), 38 deletions(-)
>
> diff --git a/src/libqmi-glib/qmi-device.c b/src/libqmi-glib/qmi-device.c
> index 108b605..bc9480c 100644
> --- a/src/libqmi-glib/qmi-device.c
> +++ b/src/libqmi-glib/qmi-device.c
> @@ -1120,32 +1120,18 @@ qmi_device_allocate_client (QmiDevice *self,
> /*****************************************************************************/
> /* Release client */
>
> -typedef struct {
> - QmiClient *client;
> - GSimpleAsyncResult *result;
> -} ReleaseClientContext;
> -
> -static void
> -release_client_context_complete_and_free (ReleaseClientContext *ctx)
> -{
> - g_simple_async_result_complete_in_idle (ctx->result);
> - g_object_unref (ctx->result);
> - g_object_unref (ctx->client);
> - g_slice_free (ReleaseClientContext, ctx);
> -}
> -
> gboolean
> qmi_device_release_client_finish (QmiDevice *self,
> GAsyncResult *res,
> GError **error)
> {
> - return !g_simple_async_result_propagate_error (G_SIMPLE_ASYNC_RESULT (res), error);
> + return g_task_propagate_boolean (G_TASK (res), error);
> }
>
> static void
> client_ctl_release_cid_ready (QmiClientCtl *client_ctl,
> GAsyncResult *res,
> - ReleaseClientContext *ctx)
> + GTask *task)
> {
> GError *error = NULL;
> QmiMessageCtlReleaseCidOutput *output;
> @@ -1156,21 +1142,21 @@ client_ctl_release_cid_ready (QmiClientCtl *client_ctl,
> /* Check result of the async operation */
> output = qmi_client_ctl_release_cid_finish (client_ctl, res, &error);
> if (!output) {
> - g_simple_async_result_take_error (ctx->result, error);
> - release_client_context_complete_and_free (ctx);
> + g_task_return_error (task, error);
> + g_object_unref (task);
> return;
> }
>
> /* Check result of the QMI operation */
> if (!qmi_message_ctl_release_cid_output_get_result (output, &error)) {
> - g_simple_async_result_take_error (ctx->result, error);
> - release_client_context_complete_and_free (ctx);
> + g_task_return_error (task, error);
> + g_object_unref (task);
> qmi_message_ctl_release_cid_output_unref (output);
> return;
> }
>
> - g_simple_async_result_set_op_res_gboolean (ctx->result, TRUE);
> - release_client_context_complete_and_free (ctx);
> + g_task_return_boolean (task, TRUE);
> + g_object_unref (task);
> qmi_message_ctl_release_cid_output_unref (output);
> }
>
> @@ -1183,7 +1169,7 @@ qmi_device_release_client (QmiDevice *self,
> GAsyncReadyCallback callback,
> gpointer user_data)
> {
> - ReleaseClientContext *ctx;
> + GTask *task;
> QmiService service;
> guint8 cid;
> gchar *flags_str;
> @@ -1206,21 +1192,15 @@ qmi_device_release_client (QmiDevice *self,
>
> /* NOTE! The operation must not take a reference to self, or we won't be
> * able to use it implicitly from our dispose() */
This previous comment is misleading and obsolete, and we should remove
it. There is no explicit CID releasing done during dispose(), and
therefore we can (must I would say) take a reference to self in the GTask.
> -
> - ctx = g_slice_new0 (ReleaseClientContext);
> - ctx->client = g_object_ref (client);
> - ctx->result = g_simple_async_result_new (G_OBJECT (self),
> - callback,
> - user_data,
> - qmi_device_release_client);
> + task = g_task_new (client, cancellable, callback, user_data);
>
The source_object for the GTask *must* be 'self', not 'client'. This is
because during the task completion, the first argument given to the
finish() function will be this object and that would be a QmiDevice.
Now, this is really not required as the pointer is anyway not used in
the finish() method, but we should keep it consistent.
And if we do this, beware, because a bit further down in the logic there
is a unregister_client() call that will remove the internal reference to
the client (and therefore may be the last one), and just after that we
do a g_object_set() on the client object. This worked always before
because there was a client reference in the Context; but now there
isn't. We cannot remove the g_object_set() call, as we don't know
whether the caller holds a full reference or not, so the correct thing
to do would be to make sure there is an explicit valid reference to the
client object around those calls; or just set a full client reference as
task data.
> /* Do not try to release an already released client */
> if (cid == QMI_CID_NONE) {
> - g_simple_async_result_set_error (ctx->result,
> - QMI_CORE_ERROR,
> - QMI_CORE_ERROR_INVALID_ARGS,
> - "Client is already released");
> - release_client_context_complete_and_free (ctx);
> + g_task_return_new_error (task,
> + QMI_CORE_ERROR,
> + QMI_CORE_ERROR_INVALID_ARGS,
> + "Client is already released");
> + g_object_unref (task);
> return;
> }
>
> @@ -1252,15 +1232,15 @@ qmi_device_release_client (QmiDevice *self,
> timeout,
> cancellable,
> (GAsyncReadyCallback)client_ctl_release_cid_ready,
> - ctx);
> + task);
>
> qmi_message_ctl_release_cid_input_unref (input);
> return;
> }
>
> /* No need to release the CID, so just done */
> - g_simple_async_result_set_op_res_gboolean (ctx->result, TRUE);
> - release_client_context_complete_and_free (ctx);
> + g_task_return_boolean (task, TRUE);
> + g_object_unref (task);
> return;
> }
>
>
--
Aleksander
https://aleksander.es
More information about the libqmi-devel
mailing list