Re: Tracking down memory usage (leak?) in Pike program

On Wed, Jul 20, 2016 at 6:24 AM, Fredrik Hubinette <[email protected]> wrote:
> I notice that num_objects is WAY smaller in your newly started program.
> That means that your program leaks objects somehow.  I suspect that those
> objects also have some external data, such as UI elements or events
> associated with them, which would explain why you don't see all the data in
> object_bytes.

Ah, that would probably explain it. That almost certainly means it's
GTK2 objects that I'm leaking. Okay, so that's what I need to track
down...

> Of course, by "leaked" I mean that they are somehow still
> referenced, but not on the screen. This can happen if UI elements have
> callbacks to themselves for instance. It's possible that adding descruct()
> calls when you remove ui elements could resolve this problem.

Interesting. Okay. I do quite a bit of opening and closing windows
(including one notable one with roughly five million widgets on it -
it's a ten-tab Notebook for a character sheet, lots of GTK2.Entry
fields and stuff), so if those are getting leaked, that would
definitely explain it. Although my regular closewindow() function
(which gets called for the delete_event and any Close buttons and
stuff) does call mainwindow->destroy(), so I'd have thought that would
clean it all up. I'll slap in a destruct(mainwindow) after that and
see what happens.

> The next step is to walk all the objects in memory, get their programs and
> see how many instances you have of each program.

Debug.pp_object_usage() says this:

Object                            Clones   Memory                  Loc
___GTK2.Label                      23800     1.6M          pgtk.c:5249
___GTK2.GdkGC                      11318   795.8k          pgtk.c:6284

I've no idea how many *active* Labels there are, but it sure ain't
twenty three thousand, and GdkGCs ought to be extremely short-lived,
so I think I've found the leaks. A freshly-invoked Gypsum has 48
labels, quickly rising to 300 within the first minute of operation.

Thanks for the pointers! Now I can start tracking down where these
Labels are getting created and why. (I've a fairly good idea about the
GCs.)

ChrisA