| Top | |
|
|
|
The purpose of the GdaThreadWrapper object is to execute functions in an isolated sub thread. As the GdaThreadWrapper is thread safe, one is able to isolate some code's execution is a private worker thread, and make a non thread safe code thread safe.
The downside of this is that the actual execution of the code will be slower as it requires threads to be synchronized.
The GdaThreadWrapper implements its own locking mechanism and can safely be used from multiple threads at once without needing further locking.
Each thread using a GdaThreadWrapper object can use it as if it was the only user: the GdaThreadWrapper will simply dispatch all the execution requests to its private worker thread and report the execution's status only to the thread which made the request.
The user can also specify a callback function to be called when an object emits a signal while being
used by the worker thread, see the gda_thread_wrapper_connect_raw() method.
The following diagram illustrates the conceptual working of the GdaThreadWrapper object: here two user threads are represented (assigned a red and green colors), both using a single GdaThreadWrapper, so in this diagram, 3 threads are present. The communication between the threads are handled by some GAsyncQueue objects (in a transparent way for the user, presented here only for illustration purposes). The queue represented in yellow is where jobs are pushed by each user thread (step 1), and popped by the worker thread (step 2). Once the worker thread has finished with a job, it stores the result along with the job and pushes it to the queue dedicated to the user thread (step 3) in this example the red queue (because the job was issued from the thread represented in red). The last step is when the user fetches the result (in its user thread), step 4.
If, when the worker thread is busy with a job, a signal is emitted, and if the user has set up a signal handler
using gda_thread_wrapper_connect_raw(),
then a "job as signal" is created by the worker thread and pushed to the user thread as illustrated
at the bottom of the diagram.
It's the user's responsability to regularly check if a submitted job has completed
(using gda_thread_wrapper_fetch_result()) or if a signal
has been emitted by an object while in the worker thread and needs to be handled by the GdaThreadWrapper
to call the associated callback (using gda_thread_wrapper_iterate() or automatically done when
calling gda_thread_wrapper_fetch_result()).
However, when a main loop is available, the GdaThreadWrapper can emit notifications through a
GIOChannel which can be intregated in the main loop. The GIOChannel can be created
using gda_thread_wrapper_get_io_channel().
GdaThreadWrapper *
gda_thread_wrapper_new (void);
Creates a new GdaThreadWrapper object
Since: 4.2
GIOChannel *
gda_thread_wrapper_get_io_channel (GdaThreadWrapper *wrapper);
Allow wrapper
to notify when an execution job is finished, by making its exec ID
readable through a new GIOChannel. This function is useful when the notification needs
to be included into a main loop. This also notifies that signals (emitted by objects in
wrapper
's internal thread) are available.
The returned GIOChannel will have something to read everytime an execution job is finished
for an execution job submitted from the calling thread. The user whould read GdaThreadNotification
structures from the channel and analyse its contents to call gda_thread_wrapper_iterate()
or gda_thread_wrapper_fetch_result().
Note1: the new communication channel will only be operational for jobs submitted after this
function returns, and for signals which have been connected after this function returns. A safe
practice is to call this function before the wrapper
object has been used.
Note2: this function will return the same GIOChannel everytime it's called from the same thread.
Note3: if the usage of the returned GIOChannel reveals an error, then g_io_channel_shutdown() and
g_io_channel_unref() should be called on the GIOChannel to let wrapper
know it should not use
that object anymore.
Since: 4.2.9
void
gda_thread_wrapper_unset_io_channel (GdaThreadWrapper *wrapper);
Does the opposite of gda_thread_wrapper_get_io_channel()
Since: 4.2.9
gpointer (*GdaThreadWrapperFunc) (gpointer arg,GError **error);
Specifies the type of function to be passed to gda_thread_wrapper_execute().
arg |
pointer to the data (which is the |
|
error |
a place to store errors |
guint gda_thread_wrapper_execute (GdaThreadWrapper *wrapper,GdaThreadWrapperFunc func,gpointer arg,GDestroyNotify arg_destroy_func,GError **error);
Make wrapper
execute the func
function with the arg
argument (along with a GError which is not error
)
in the sub thread managed by wrapper
. To execute a function which does not return anything,
use gda_thread_wrapper_execute_void().
This method returns immediately, and the caller then needs to use gda_thread_wrapper_fetch_result() to
check if the execution has finished and get the result.
Once func
's execution is finished, if arg
is not NULL, the arg_destroy_func
destruction function is called
on arg
. This call occurs in the thread calling gda_thread_wrapper_fetch_result().
If an error occurred in this function, then the arg_destroy_func
function is not called to free arg
.
wrapper |
a GdaThreadWrapper object |
|
func |
the function to execute, not |
|
arg |
argument to pass to |
[allow-none] |
arg_destroy_func |
function to be called when the execution has finished, to destroy |
[allow-none] |
error |
a place to store errors, for errors occurring in this method, not errors occurring while |
Since: 4.2
void (*GdaThreadWrapperVoidFunc) (gpointer arg,GError **error);
Specifies the type of function to be passed to gda_thread_wrapper_execute_void().
arg |
a pointer to the data (which is the |
|
error |
a place to store errors |
guint gda_thread_wrapper_execute_void (GdaThreadWrapper *wrapper,GdaThreadWrapperVoidFunc func,gpointer arg,GDestroyNotify arg_destroy_func,GError **error);
Make wrapper
execute the func
function with the arg
argument (along with a GError which is not error
)
in the sub thread managed by wrapper
. To execute a function which returns some pointer,
use gda_thread_wrapper_execute().
This method returns immediately. Calling gda_thread_wrapper_fetch_result() is not necessary as func
does not return any result. However, it may be necessary to call gda_thread_wrapper_iterate() to give wrapper
a
chance to execute the arg_destroy_func
function if not NULL (note that gda_thread_wrapper_iterate() is
called by gda_thread_wrapper_fetch_result() itself).
Once func
's execution is finished, if arg
is not NULL, the arg_destroy_func
destruction function is called
on arg
. This call occurs in the thread calling gda_thread_wrapper_fetch_result().
If an error occurred in this function, then the arg_destroy_func
function is not called to free arg
.
wrapper |
a GdaThreadWrapper object |
|
func |
the function to execute, not |
|
arg |
argument to pass to |
[allow-none] |
arg_destroy_func |
function to be called when the execution has finished, to destroy |
[allow-none] |
error |
a place to store errors, for errors occurring in this method, not errors occurring while |
Since: 4.2
gboolean gda_thread_wrapper_cancel (GdaThreadWrapper *wrapper,guint id);
Cancels a job not yet executed. This may fail for the following reasons:
the job id could not be found, either because it has already been treated or because
it does not exist or because it was created in another thread
the job id is currently being treated by the worker thread
wrapper |
a GdaThreadWrapper object |
|
id |
the ID of a job as returned by |
Since: 4.2
void gda_thread_wrapper_iterate (GdaThreadWrapper *wrapper,gboolean may_block);
This method gives wrapper
a chance to check if some functions to be executed have finished
for the calling thread. In this case it handles the execution result and
makes it ready to be processed using gda_thread_wrapper_fetch_result().
This method also allows wrapper
to handle signals which may have been emitted by objects
while in the worker thread, and call the callback function specified when gda_thread_wrapper_connect_raw()
was used.
If may_block
is TRUE, then it will block untill there is one finished execution
(functions returning void and signals are ignored regarding this argument).
Since: 4.2
gpointer gda_thread_wrapper_fetch_result (GdaThreadWrapper *wrapper,gboolean may_lock,guint exp_id,GError **error);
Use this method to check if the execution of a function is finished. The function's execution must have
been requested using gda_thread_wrapper_execute().
wrapper |
a GdaThreadWrapper object |
|
may_lock |
TRUE if this funct must lock the caller untill a result is available |
|
exp_id |
ID of the job for which a result is expected |
|
error |
a place to store errors, for errors which may have occurred during the execution, or |
Since: 4.2
gint
gda_thread_wrapper_get_waiting_size (GdaThreadWrapper *wrapper);
Use this method to query the number of functions which have been queued to be executed but which have not yet been executed.
Since: 4.2
void (*GdaThreadWrapperCallback) (GdaThreadWrapper *wrapper,gpointer instance,const gchar *signame,gint n_param_values,const GValue *param_values,gpointer gda_reserved,gpointer data);
Specifies the type of function to be passed to gda_thread_wrapper_connect_raw()
wrapper |
the GdaThreadWrapper |
|
instance |
a pointer to the instance which emitted the signal |
|
signame |
the name of the signal being emitted |
|
n_param_values |
number of GValue in |
|
param_values |
array of |
|
gda_reserved |
reserved |
|
data |
a pointer to the data (which is the |
gulong gda_thread_wrapper_connect_raw (GdaThreadWrapper *wrapper,gpointer instance,const gchar *sig_name,gboolean private_thread,gboolean private_job,GdaThreadWrapperCallback callback,gpointer data);
Connects a callback function to a signal for a particular object. The difference with g_signal_connect() and
similar functions are:
the callback argument is not a GCallback function, so the callback signature is not
dependent on the signal itself
the signal handler must not have to return any value
the callback function will be called asynchronously, the caller may need to use
gda_thread_wrapper_iterate() to get the notification
if private_job and private_thread control in which case the signal is propagated.
Also note that signal handling is done asynchronously: when emitted in the worker thread, it
will be "queued" to be processed in the user thread when it has the chance (when gda_thread_wrapper_iterate()
is called directly or indirectly). The side effect is that the callback function is usually
called long after the object emitting the signal has finished emitting it.
To disconnect a signal handler, don't use any of the g_signal_handler_*() functions but the
gda_thread_wrapper_disconnect() method.
wrapper |
a GdaThreadWrapper object |
|
instance |
the instance to connect to |
|
sig_name |
a string of the form "signal-name::detail" |
|
private_thread |
set to |
|
private_job |
set to |
|
callback |
a GdaThreadWrapperCallback function. |
[scope call] |
data |
data to pass to |
[closure] |
Since: 4.2
void gda_thread_wrapper_disconnect (GdaThreadWrapper *wrapper,gulong id);
Disconnects the emission of a signal, does the opposite of gda_thread_wrapper_connect_raw().
As soon as this method returns, the callback function set when gda_thread_wrapper_connect_raw()
was called will not be called anymore (even if the object has emitted the signal in the worker
thread and this signal has not been handled in the user thread).
wrapper |
a GdaThreadWrapper object |
|
id |
a handler ID, as returned by |
Since: 4.2
void gda_thread_wrapper_steal_signal (GdaThreadWrapper *wrapper,gulong id);
Requests that the signal which ID is id
(which has been obtained using gda_thread_wrapper_connect_raw())
be treated by the calling thread instead of by the thread in which gda_thread_wrapper_connect_raw()
was called.
Since: 4.2
typedef struct {
GdaThreadNotificationType type;
guint job_id;
} GdaThreadNotification;
A notification to be read through the GIOChannel which is returned by gda_thread_wrapper_get_io_channel(),
for example:
gboolean
wrapper_ioc_cb (GIOChannel *source, GIOCondition condition, gpointer data)
{
GIOStatus status;
gsize nread;
GdaThreadNotification notif;
if (condition & G_IO_IN) {
status = g_io_channel_read_chars (source, (gchar*) ¬if, sizeof (notif), &nread, NULL);
if ((status != G_IO_STATUS_NORMAL) || (nread != sizeof (notif)))
goto onerror;
switch (notif.type) {
case GDA_THREAD_NOTIFICATION_JOB:
check_for_wrapper_result (bcnc);
break;
case GDA_THREAD_NOTIFICATION_SIGNAL:
gda_thread_wrapper_iterate (bcnc->priv->wrapper, FALSE);
break;
default:
goto onerror;
break;
}
}
if (condition & (G_IO_ERR | G_IO_HUP | G_IO_NVAL))
goto onerror;
return TRUE; // keep callback
onerror:
g_io_channel_shutdown (bcnc->priv->ioc, FALSE, NULL);
return FALSE; // removed callback
}
{
[...]
GIOChannel *ioc;
ioc = gda_thread_wrapper_get_io_channel (wrapper);
if (!ioc)
[handle error]
else {
guint watch_id;
watch_id = g_io_add_watch (ioc, G_IO_IN | G_IO_HUP | G_IO_ERR | G_IO_NVAL,
(GIOFunc) wrapper_ioc_cb, NULL);
}
}
Defines the kind of notification which can be obtained when reading from te GIOChannel
returned by gda_thread_wrapper_get_io_channel().