A gio.task.Task represents and manages a cancellable ‘task’.
Asynchronous operations
The most common usage of gio.task.Task is as a gio.async_result.AsyncResult, to manage data during an asynchronous operation. You call gio.task.Task.new_ in the ‘start’ method, followed by gio.task.Task.setTaskData and the like if you need to keep some additional data associated with the task, and then pass the task object around through your asynchronous operation. Eventually, you will call a method such as gio.task.Task.returnPointer or gio.task.Task.returnError, which will save the value you give it and then invoke the task’s callback function in the thread-default main context (see glib.main_context.MainContext.pushThreadDefault) where it was created (waiting until the next iteration of the main loop first, if necessary). The caller will pass the gio.task.Task back to the operation’s finish function (as a gio.async_result.AsyncResult), and you can use gio.task.Task.propagatePointer or the like to extract the return value.
Using gio.task.Task requires the thread-default glib.main_context.MainContext from when the gio.task.Task was constructed to be running at least until the task has completed and its data has been freed.
If a gio.task.Task has been constructed and its callback set, it is an error to not call g_task_return_*() on it. GLib will warn at runtime if this happens (since 2.76).
Here is an example for using gio.task.Task as a gio.async_result.AsyncResult:
typedef struct {
CakeFrostingType frosting;
char *message;
} DecorationData;
static void
decoration_data_free (DecorationData *decoration)
{
g_free (decoration->message);
g_slice_free (DecorationData, decoration);
}
static void
baked_cb (Cake *cake,
gpointer user_data)
{
GTask *task = user_data;
DecorationData *decoration = g_task_get_task_data (task);
GError *error = NULL;
if (cake == NULL)
{
g_task_return_new_error (task, BAKER_ERROR, BAKER_ERROR_NO_FLOUR,
"Go to the supermarket");
g_object_unref (task);
return;
}
if (!cake_decorate (cake, decoration->frosting, decoration->message, &error))
{
g_object_unref (cake);
// g_task_return_error() takes ownership of error
g_task_return_error (task, error);
g_object_unref (task);
return;
}
g_task_return_pointer (task, cake, g_object_unref);
g_object_unref (task);
}
void
baker_bake_cake_async (Baker *self,
guint radius,
CakeFlavor flavor,
CakeFrostingType frosting,
const char *message,
GCancellable *cancellable,
GAsyncReadyCallback callback,
gpointer user_data)
{
GTask *task;
DecorationData *decoration;
Cake *cake;
task = g_task_new (self, cancellable, callback, user_data);
if (radius < 3)
{
g_task_return_new_error (task, BAKER_ERROR, BAKER_ERROR_TOO_SMALL,
"%ucm radius cakes are silly",
radius);
g_object_unref (task);
return;
}
cake = _baker_get_cached_cake (self, radius, flavor, frosting, message);
if (cake != NULL)
{
// _baker_get_cached_cake() returns a reffed cake
g_task_return_pointer (task, cake, g_object_unref);
g_object_unref (task);
return;
}
decoration = g_slice_new (DecorationData);
decoration->frosting = frosting;
decoration->message = g_strdup (message);
g_task_set_task_data (task, decoration, (GDestroyNotify) decoration_data_free);
_baker_begin_cake (self, radius, flavor, cancellable, baked_cb, task);
}
Cake *
baker_bake_cake_finish (Baker *self,
GAsyncResult *result,
GError **error)
{
g_return_val_if_fail (g_task_is_valid (result, self), NULL);
return g_task_propagate_pointer (G_TASK (result), error);
}
Chained asynchronous operations
gio.task.Task also tries to simplify asynchronous operations that internally chain together several smaller asynchronous operations. gio.task.Task.getCancellable, gio.task.Task.getContext, and gio.task.Task.getPriority allow you to get back the task’s gio.cancellable.Cancellable, glib.main_context.MainContext, and I/O priority when starting a new subtask, so you don’t have to keep track of them yourself. gio.task.Task.attachSource simplifies the case of waiting for a source to fire (automatically using the correct glib.main_context.MainContext and priority).
Here is an example for chained asynchronous operations:
typedef struct {
Cake *cake;
CakeFrostingType frosting;
char *message;
} BakingData;
static void
decoration_data_free (BakingData *bd)
{
if (bd->cake)
g_object_unref (bd->cake);
g_free (bd->message);
g_slice_free (BakingData, bd);
}
static void
decorated_cb (Cake *cake,
GAsyncResult *result,
gpointer user_data)
{
GTask *task = user_data;
GError *error = NULL;
if (!cake_decorate_finish (cake, result, &error))
{
g_object_unref (cake);
g_task_return_error (task, error);
g_object_unref (task);
return;
}
// baking_data_free() will drop its ref on the cake, so we have to
// take another here to give to the caller.
g_task_return_pointer (task, g_object_ref (cake), g_object_unref);
g_object_unref (task);
}
static gboolean
decorator_ready (gpointer user_data)
{
GTask *task = user_data;
BakingData *bd = g_task_get_task_data (task);
cake_decorate_async (bd->cake, bd->frosting, bd->message,
g_task_get_cancellable (task),
decorated_cb, task);
return G_SOURCE_REMOVE;
}
static void
baked_cb (Cake *cake,
gpointer user_data)
{
GTask *task = user_data;
BakingData *bd = g_task_get_task_data (task);
GError *error = NULL;
if (cake == NULL)
{
g_task_return_new_error (task, BAKER_ERROR, BAKER_ERROR_NO_FLOUR,
"Go to the supermarket");
g_object_unref (task);
return;
}
bd->cake = cake;
// Bail out now if the user has already cancelled
if (g_task_return_error_if_cancelled (task))
{
g_object_unref (task);
return;
}
if (cake_decorator_available (cake))
decorator_ready (task);
else
{
GSource *source;
source = cake_decorator_wait_source_new (cake);
// Attach @source to @task’s GMainContext and have it call
// decorator_ready() when it is ready.
g_task_attach_source (task, source, decorator_ready);
g_source_unref (source);
}
}
void
baker_bake_cake_async (Baker *self,
guint radius,
CakeFlavor flavor,
CakeFrostingType frosting,
const char *message,
gint priority,
GCancellable *cancellable,
GAsyncReadyCallback callback,
gpointer user_data)
{
GTask *task;
BakingData *bd;
task = g_task_new (self, cancellable, callback, user_data);
g_task_set_priority (task, priority);
bd = g_slice_new0 (BakingData);
bd->frosting = frosting;
bd->message = g_strdup (message);
g_task_set_task_data (task, bd, (GDestroyNotify) baking_data_free);
_baker_begin_cake (self, radius, flavor, cancellable, baked_cb, task);
}
Cake *
baker_bake_cake_finish (Baker *self,
GAsyncResult *result,
GError **error)
{
g_return_val_if_fail (g_task_is_valid (result, self), NULL);
return g_task_propagate_pointer (G_TASK (result), error);
}
Asynchronous operations from synchronous ones
You can use gio.task.Task.runInThread to turn a synchronous operation into an asynchronous one, by running it in a thread. When it completes, the result will be dispatched to the thread-default main context (see glib.main_context.MainContext.pushThreadDefault) where the gio.task.Task was created.
Running a task in a thread:
typedef struct {
guint radius;
CakeFlavor flavor;
CakeFrostingType frosting;
char *message;
} CakeData;
static void
cake_data_free (CakeData *cake_data)
{
g_free (cake_data->message);
g_slice_free (CakeData, cake_data);
}
static void
bake_cake_thread (GTask *task,
gpointer source_object,
gpointer task_data,
GCancellable *cancellable)
{
Baker *self = source_object;
CakeData *cake_data = task_data;
Cake *cake;
GError *error = NULL;
cake = bake_cake (baker, cake_data->radius, cake_data->flavor,
cake_data->frosting, cake_data->message,
cancellable, &error);
if (cake)
g_task_return_pointer (task, cake, g_object_unref);
else
g_task_return_error (task, error);
}
void
baker_bake_cake_async (Baker *self,
guint radius,
CakeFlavor flavor,
CakeFrostingType frosting,
const char *message,
GCancellable *cancellable,
GAsyncReadyCallback callback,
gpointer user_data)
{
CakeData *cake_data;
GTask *task;
cake_data = g_slice_new (CakeData);
cake_data->radius = radius;
cake_data->flavor = flavor;
cake_data->frosting = frosting;
cake_data->message = g_strdup (message);
task = g_task_new (self, cancellable, callback, user_data);
g_task_set_task_data (task, cake_data, (GDestroyNotify) cake_data_free);
g_task_run_in_thread (task, bake_cake_thread);
g_object_unref (task);
}
Cake *
baker_bake_cake_finish (Baker *self,
GAsyncResult *result,
GError **error)
{
g_return_val_if_fail (g_task_is_valid (result, self), NULL);
return g_task_propagate_pointer (G_TASK (result), error);
}
Adding cancellability to uncancellable tasks
Finally, gio.task.Task.runInThread and gio.task.Task.runInThreadSync can be used to turn an uncancellable operation into a cancellable one. If you call gio.task.Task.setReturnOnCancel, passing TRUE, then if the task’s gio.cancellable.Cancellable is cancelled, it will return control back to the caller immediately, while allowing the task thread to continue running in the background (and simply discarding its result when it finally does finish). Provided that the task thread is careful about how it uses locks and other externally-visible resources, this allows you to make ‘GLib-friendly’ asynchronous and cancellable synchronous variants of blocking APIs.
Cancelling a task:
static void
bake_cake_thread (GTask *task,
gpointer source_object,
gpointer task_data,
GCancellable *cancellable)
{
Baker *self = source_object;
CakeData *cake_data = task_data;
Cake *cake;
GError *error = NULL;
cake = bake_cake (baker, cake_data->radius, cake_data->flavor,
cake_data->frosting, cake_data->message,
&error);
if (error)
{
g_task_return_error (task, error);
return;
}
// If the task has already been cancelled, then we don’t want to add
// the cake to the cake cache. Likewise, we don’t want to have the
// task get cancelled in the middle of updating the cache.
// g_task_set_return_on_cancel() will return %TRUE here if it managed
// to disable return-on-cancel, or %FALSE if the task was cancelled
// before it could.
if (g_task_set_return_on_cancel (task, FALSE))
{
// If the caller cancels at this point, their
// GAsyncReadyCallback won’t be invoked until we return,
// so we don’t have to worry that this code will run at
// the same time as that code does. But if there were
// other functions that might look at the cake cache,
// then we’d probably need a GMutex here as well.
baker_add_cake_to_cache (baker, cake);
g_task_return_pointer (task, cake, g_object_unref);
}
}
void
baker_bake_cake_async (Baker *self,
guint radius,
CakeFlavor flavor,
CakeFrostingType frosting,
const char *message,
GCancellable *cancellable,
GAsyncReadyCallback callback,
gpointer user_data)
{
CakeData *cake_data;
GTask *task;
cake_data = g_slice_new (CakeData);
...
task = g_task_new (self, cancellable, callback, user_data);
g_task_set_task_data (task, cake_data, (GDestroyNotify) cake_data_free);
g_task_set_return_on_cancel (task, TRUE);
g_task_run_in_thread (task, bake_cake_thread);
}
Cake *
baker_bake_cake_sync (Baker *self,
guint radius,
CakeFlavor flavor,
CakeFrostingType frosting,
const char *message,
GCancellable *cancellable,
GError **error)
{
CakeData *cake_data;
GTask *task;
Cake *cake;
cake_data = g_slice_new (CakeData);
...
task = g_task_new (self, cancellable, NULL, NULL);
g_task_set_task_data (task, cake_data, (GDestroyNotify) cake_data_free);
g_task_set_return_on_cancel (task, TRUE);
g_task_run_in_thread_sync (task, bake_cake_thread);
cake = g_task_propagate_pointer (task, error);
g_object_unref (task);
return cake;
}
gio.task.Task’s API attempts to be simpler than gio.simple_async_result.SimpleAsyncResult’s in several ways:
- You can save task-specific data with gio.task.Task.setTaskData, and retrieve it later with gio.task.Task.getTaskData. This replaces the abuse of gio.simple_async_result.SimpleAsyncResult.setOpResGpointer for the same purpose with gio.simple_async_result.SimpleAsyncResult.
- In addition to the task data, gio.task.Task also keeps track of the priority, gio.cancellable.Cancellable, and glib.main_context.MainContext associated with the task, so tasks that consist of a chain of simpler asynchronous operations will have easy access to those values when starting each sub-task.
- gio.task.Task.returnErrorIfCancelled provides simplified handling for cancellation. In addition, cancellation overrides any other gio.task.Task return value by default, like gio.simple_async_result.SimpleAsyncResult does when gio.simple_async_result.SimpleAsyncResult.setCheckCancellable is called. (You can use gio.task.Task.setCheckCancellable to turn off that behavior.) On the other hand, gio.task.Task.runInThread guarantees that it will always run your
task_func, even if the task’s gio.cancellable.Cancellable is already cancelled before the task gets a chance to run; you can start your task_func with a gio.task.Task.returnErrorIfCancelled check if you need the old behavior. - The ‘return’ methods (eg, gio.task.Task.returnPointer) automatically cause the task to be ‘completed’ as well, and there is no need to worry about the ‘complete’ vs ‘complete in idle’ distinction. (gio.task.Task automatically figures out whether the task’s callback can be invoked directly, or if it needs to be sent to another glib.main_context.MainContext, or delayed until the next iteration of the current glib.main_context.MainContext.)
- The ‘finish’ functions for gio.task.Task based operations are generally much simpler than gio.simple_async_result.SimpleAsyncResult ones, normally consisting of only a single call to gio.task.Task.propagatePointer or the like. Since gio.task.Task.propagatePointer ‘steals’ the return value from the gio.task.Task, it is not necessary to juggle pointers around to prevent it from being freed twice.
- With gio.simple_async_result.SimpleAsyncResult, it was common to call gio.simple_async_result.SimpleAsyncResult.propagateError from the
_finish() wrapper function, and have virtual method implementations only deal with successful returns. This behavior is deprecated, because it makes it difficult for a subclass to chain to a parent class’s async methods. Instead, the wrapper function should just be a simple wrapper, and the virtual method should call an appropriate g_task_propagate_ function. Note that wrapper methods can now use gio.async_result.AsyncResult.legacyPropagateError to do old-style gio.simple_async_result.SimpleAsyncResult error-returning behavior, and gio.async_result.AsyncResult.isTagged to check if a result is tagged as having come from the _async() wrapper function (for ‘short-circuit’ results, such as when passing 0 to gio.input_stream.InputStream.readAsync).
Thread-safety considerations
Due to some infelicities in the API design, there is a thread-safety concern that users of gio.task.Task have to be aware of:
If the main thread drops its last reference to the source object or the task data before the task is finalized, then the finalizers of these objects may be called on the worker thread.
This is a problem if the finalizers use non-threadsafe API, and can lead to hard-to-debug crashes. Possible workarounds include:
- Clear task data in a signal handler for
notify::completed - Keep iterating a main context in the main thread and defer dropping the reference to the source object to that main context when the task is finalized