Re: GtkListStore and Threads problems

["Lorena Salinas" <lsalinas invap com ar>]

Here is a graphic to show where (I suppose) the problem is:


Time-->
T0       T1          T2                                T3            T4
                         gtk_main() is blocked
------- |----------|-----------------------|-------------|----->
          |               |                                   | 
|
          |               | gtk_list_store_set()     |                    |
          |               |                                   | 
|
          |              V                                 V 
|
          |   gdk_threads_enter()   gdk_threads_leave()    |
          | 
|
          | 
|
         V 
V
  gtk_tree_model_get()                        gtk_tree_model_get()
  (gtk_main() start)                              (gtk_main()continue)

At time:

T1: gtk_main() starts to read the 5 columns of row 1 from the model shared 
with
   the working thread (using gtk_tree_model_get())

T2: the working thread locks GTK main thread
   gtk_main() is blocked in the middle of the loop to read the 5 columns

Note: (is this possible??? I'm not sure how gtk_threads_enter() interacts 
with main thread)

   Lets suppose it has read only 2 columns when the working thread locks 
it.
   Then the working thread upgrades some columns of row 1
   (using gtk_list_store_set() which calls gtk_sequence_* functions), so
   the iter (row 1) changes

T3: working thread releases the lock

T4: gtk_main() continues reading the other 3 columns from the "updated" 
model
   (using gtk_tree_model_get()) and it complains because the iter has 
changed!!!



Regards,
Lorena

----- Original Message ----- 
From: "Peter Bloomfield" <peterbloomfield bellsouth net>
To: "Lorena Salinas" <lsalinas invap com ar>
Cc: <gtk-app-devel-list gnome org>
Sent: Wednesday, June 15, 2005 1:45 PM
Subject: Re: GtkListStore and Threads problems


> On 06/15/2005 10:11:53 AM Wed, Lorena Salinas wrote:
> > Here is part of the function...
> >
> > valid = gtk_tree_model_get_iter_first (GTK_TREE_MODEL
> > (my_model),
> > &iter);
> >
> > while (valid) {
> >
> > gtk_tree_model_get (GTK_TREE_MODEL (my_model), &iter,
> >   MYMODEL_COL_STATUS, &status,
> > -1);
> >
> > /* do some long calculations.... */
> >
> > gdk_threads_enter(); /* get GTK thread lock */
> >
> > gtk_list_store_set (GTK_LIST_STORE (my_model), &iter,
> > MYMODEL_COL_STATUS,STATUS_NORMAL,
> > -1);
> >
> > gdk_threads_leave(); /* release GTK thread lock */
> >
> > valid = gtk_tree_model_iter_next (GTK_TREE_MODEL (my_model),
> > &iter);
> >
> > }
> >
> >
> > This is the only place where I update the model...
> > Is it valid to assume that iter is the same, that it hasn't
> > changed?
>
> The iter shouldn't have changed, and if the model isn't changed  the iter 
> should still be valid.  But your original errors were  coming from 
> gtk_sequence_* functions, which are called from 
> gtk_list_store_iter_next() as well as (possibly)  gtk_list_store_set(), 
> and you call gtk_list_store_iter_next()  without holding the lock.
>
> You could organize the loop around a path instead of an iter, and  then 
> when you really need an iter:
> - grab the lock;
> - initialize the iter;
> - use the iter to get or set;
> - drop the lock;
> and get on with those really long calculations.
>
> Peter