On 11-10-17 05:22 AM, Nick Treleaven wrote:
Hi Matthew, I'm a bit concerned about the changed ui_lookup_widget (and hookup) functions - these are in the plugin API and can be used independently from Glade. (plugin) API function behaviour should not normally change, but it seems the owner parameter is now ignored. If we are caching lookups this is probably wrong as the widget may get destroyed.
I haven't looked at the new implementation but perhaps you could help explain the changes.
Long story short: compatibility is the reason.
Typical short story long:
I didn't want to break all the existing code in core and plugins that were using ui_lookup_widget/ui_hookup_widget() functions, so I dropped the (now) pointless first parameter. There's no need to associate things with a parent/owner widget since all objects are required to have unique names in GtkBuilder (and the GHashTable).
I'm not 100% sure what you mean by caching lookups or widgets being destroyed. I'll try and explain how it works and hopefully we can get some more experienced developer's eyeballs on the code to review it (that's you!).
So the first function called is ui_init_builder() and what it does is to initialize the GtkBuilder so that it creates all of the GObjects. Then it iterates over all of the objects in the GtkBuilder and stores their pointers into a GHashTable (also ref'ing them) using their names as the key. This is basically a "drop-in" replacement for the old Glade 2 generated code. Replacing the hookup and lookup functions are the new ui_lookup_object() and ui_hookup_object() functions. The former just does a g_hash_table_lookup() to find the GObject with the associated key(name). The latter just inserts the object into the hash table using the name as the key (and ref'ing it).
It's my understanding that all objects that were hooked up were eternal until program close, though if this is not the case, we'll be leaking exactly 1 GObject per object that the user/plugin code destroys/unrefs since the GHashTable is holding a ref on it. It would be quite trivial to add a function to remove objects from the hash table, but existing code wouldn't be using this obviously. We could also not ref user-added object and only ref those from GtkBuilder if we wanted.
In a perfect world, we could drop the GHashTable and use GtkBuilder directly to track the objects, but it seems GtkBuilder can only add objects from an XML string. Actually, in a perfect world, we wouldn't be creating widgets outside of the GUI file at all and so we could use GtkBuilder to track all the objects.
So the basic plan I had thought up is to do these steps:
1. Convert the Glade 2 to Glade 3 (done) 2. Get rid of the old generated code with some compat code (done) 3. Separate out all the hard-coded GUI stuff and move it into the proper place, the GtkBuilder file (not even remotely done) 4. Stop adding new code that mixes UI and business logic (todo)
Only the first 2 are essential to do now to actually switch to GtkBuilder, the next two can be done gradually over time.
I hope this clears up somewhat how the GtkBuilder stuff is currently working. If you want to review the implementation, it's quite trivial really, you can get a pretty clear view of the meat of it by using:
$ git diff master..gtkbuilder src/ui_utils.c
That should show pretty much only the implementation and a few other minor changes that have happened since.
Do let me know if you have any more questions, and feel free to hack away on the code as usual. I don't think any more experienced developers have really actually reviewed the code in-depth, so I would be grateful if you get the time. It's something that will need to happen before we merge into master.
In the meanwhile, I'll keep plugging away on the odd little bugs I notice resulting from the changes.
Cheers, Matthew Brush