That is why we have 4 functions per module: the join function is called by any
process, and is in charge of creating the process-wide globals. If it is the
first time a process uses the module, it also calls the init function, in
-charge of initializing the world-wide globals. We have the symetric functions
+charge of initializing the world-wide globals. We have the symmetric functions
leave, called by any process not using the module anymore, and exit, in charge
of freeing the world-wide globals when no process use it anymore.
create process-wide globals. The fact that indeed the calling sequence goes
from join to init and not the other side is just an implementation bias ;)
-Then again, we want these functionnalities to be quick. We want to access
+Then again, we want these functionalities to be quick. We want to access
the process-wide globals by providing their rank in a dynar, we don't want to
-search them in a dictionnary. This is especially true in the module
+search them in a dictionary. This is especially true in the module
implementation, where each functions is likely to require them to work. The
position could be a stored in a global variable only visible from the module
-implementation.
+implementation.
The need for an array in which to store the globals does not hold for
world-wide globals: only one instance of them can exist in the same unix naming
tables. This is why we have the ID field in the module structure: it points
exactly to the implementation side global.
-Yeah, I know. All this is not that clear. But at least, writing this helped me
+Yeah, I know. All this is not that clear. But at least, writing this helped me
to design that crap ;)
*/
allow modules initializing other modules while tracking dependencies
properly and leave() only when needed. This would allow dynamic module
loading/unloading */
-
+
int *p_id; /* where the module stores the libdata ID (a global somewhere), to tweak it on need */
- void_f_void_t *init_f; /* First time the module is referenced. */
- void_f_void_t *exit_f; /* When last process referencing it stops doing so. */
- void_f_pvoid_t *join_f; /* Called by each process in initialization phase (init_f called once for all processes) */
- void_f_pvoid_t *leave_f; /* Called by each process in finalization phase. Should free moddata passed */
+ void_f_void_t init_f; /* First time the module is referenced. */
+ void_f_void_t exit_f; /* When last process referencing it stops doing so. */
+ void_f_pvoid_t join_f; /* Called by each process in initialization phase (init_f called once for all processes) */
+ void_f_pvoid_t leave_f; /* Called by each process in finalization phase. Should free moddata passed */
} s_gras_module_t, *gras_module_t;
static xbt_set_t _gras_modules = NULL; /* content: s_gras_module_t */
static void gras_module_freep(void *p) {
free( ((gras_module_t)p) ->name);
+ free(p);
}
* @param name: name of the module, of course (beware of dupplicates!)
* @param datasize: the size of your data, ie of the state this module has on each process
* @param ID: address of a global you use as parameter to gras_module_data_by_id
- * @param init_f: function called the first time a module gets by a process of the naming space.
+ * @param init_f: function called the first time a module gets by a process of the naming space.
* A classical use is to declare some messages the module uses, as well as the initialization
* of module constants (accross processes boundaries in SG).
* @param exit_f: function called when the last process of this naming space unref this module.
- * @param join_f: function called each time a process references the module.
+ * @param join_f: function called each time a process references the module.
* It is passed the moddata already malloced, and should initialize the fields as it wants.
* It can also attach some callbacks to the module messages.
* @param leave_f: function called each time a process unrefs the module.
*/
void gras_module_add(const char *name, unsigned int datasize, int *ID,
- void_f_void_t *init_f, void_f_void_t *exit_f,
- void_f_pvoid_t *join_f, void_f_pvoid_t *leave_f) {
+ void_f_void_t init_f, void_f_void_t exit_f,
+ void_f_pvoid_t join_f, void_f_pvoid_t leave_f) {
gras_module_t mod=NULL;
xbt_ex_t e;
volatile int found = 0;
}
if (found) {
- xbt_assert1(mod->init_f == init_f,
+ xbt_assert1(mod->init_f == init_f,
"Module %s reregistered with a different init_f!", name);
xbt_assert1(mod->exit_f == exit_f,
"Module %s reregistered with a different exit_f!", name);
"Module %s reregistered with a different join_f!", name);
xbt_assert1(mod->leave_f == leave_f,
"Module %s reregistered with a different leave_f!", name);
- xbt_assert1(mod->datasize == datasize,
+ xbt_assert1(mod->datasize == datasize,
"Module %s reregistered with a different datasize!", name);
xbt_assert1(mod->p_id == ID,
"Module %s reregistered with a different p_id field!", name);
-
+
DEBUG1("Module %s already registered. Ignoring re-registration",name);
return;
}
mod = xbt_new(s_gras_module_t, 1);
mod->name = xbt_strdup(name);
mod->name_len = strlen(name);
-
+
mod->datasize = datasize;
mod->p_id = ID;
mod->init_f = init_f;
mod->join_f = join_f;
mod->leave_f = leave_f;
mod->refcount = 0;
-
+
*mod->p_id = xbt_set_length(_gras_modules);
-
+
xbt_set_add(_gras_modules,(void*)mod,gras_module_freep);
}
gras_procdata_t *pd;
void *moddata;
gras_module_t mod = (gras_module_t)xbt_set_get_by_name(_gras_modules, name);
-
+
VERB2("Join to module %s (%p)",name,mod);
-
+
/* NEW */
if (mod->refcount == 0) {
VERB1("Init module %s",name);
(*mod->exit_f)();
- /* Don't remove the module for real, sets don't allow to
+ /* Don't remove the module for real, sets don't allow to
free(mod->name);
free(mod);
}
}
-
+
void *gras_moddata_by_id(unsigned int ID) {
gras_procdata_t *pd=gras_procdata_get();
void *p;
