gtk 2 problem with Gramps in Linux Mint 18 (beta) and ubuntu 16.04

[Enno Borgsteede <ennoborg gmail com>]

Hello,

I'm a Gramps developer, who still works with Gramps version 3.4.9, 
because of its data model. Gramps is a genealogy program written in 
Python. The version I work with is based on GTK+ 2, using the PyGTK wrapper.

If I understand things right, this version of Gramps uses GtkTreeModel 
for display, but not GtkTreeStore, because that would mean keeping a 
whole list of person data in memory, which is not always possible. In 
other words, it has its own store, which reads data from the database 
when needed.

This version has worked well for me, up to Linux Mint 17.3, but when I 
tried the Mint 18 beta, I ran into two problems. These problems also 
show up in ubuntu 16.04, but not in 15.10. The Gramps version that I use 
is always the same, installed from the same .deb file, or the same 
branch on GitHub.

Problem #1 is that when I work a list of persons grouped by surname, 
like John A and Jane B, expanding the 2nd surname, which is at the 
bottom in the displayed surname/person detail tree (two levels), I get a 
Python exception, saying that the node (surname) index is out of bounds. 
In this example surname nodes have indexes 0 and 1, and when I expand 
the 2nd node, I can see that our store's on_get_iter method is called 
with path (2,), which is obviously out of bounds, because the only valid 
paths are (0,0) and (1,0). I can see that after that, on_get_iter is 
also called with path (1,0), so after acknowledging the error dialog, I 
can still see the expanded B surname node.

With Mint 17.3, everything is OK, but since that version of Mint is 
based on ubuntu 14.04, I also tried this on ubuntu 15.10, where things 
are also perfectly well.

Problem #2 doesn't show up in a two person tree, but it is visible in my 
real tree, with more than 8000 surnames. When I expand a surname node 
near the end of the alphabet in that, say at path (8000,), I can see 
that on_get_iter() is not just called with that path, as is normal in 
Linuxes older than Mint 18/ubuntu 16.04, but that a full tree walk is 
done, starting at path (0,), ending at (8000,), or whatever node I 
clicked on. Compared to older Linuxes, that is 8000 or more useless 
calls. This phenomenon does not show when I click a surname near the 
start of the alphabet, like my own.

I found both issues by adding a print line to Gramps' Python code, and 
it might be obvious that this print line causes a bit of a delay when 
testing problem #2. But even without it, it is quite obvious that some 
sort of full tree walk is done, because even then I see a considerable 
delay when I expand a surname group that sits near the end of the 
alphabet, and also when I edit a person in that area, which triggers a 
check for a possible rebuilding of the display tree. These delays do not 
show under Mint 17.3, or ubuntu 15.10, and when I add the print line to 
my code on those Linux versions, I can indeed see that the number of 
on_get_iter() calls is much less. In general, I only see such calls when 
the mouse pointer hovers over a display line, or when a click causes a 
focus change. In Mint 18, or ubuntu 16.04, I also see that on_get_iter() 
is called for every person in a surname group when I expand that, where 
it is only called for a few in Mint 17.3 or ubuntu 15.10, most probably 
associated with mouse hovers (hints).

I'm mentioning Mint 18 and ubuntu 16.04, because they have the same 
(ubuntu) code base, but a different desktop (cinnamon vs. unity). The 
problem also shows in ubuntu GNOME, so I expect it's a GNOME issue, and 
with the little experience that I have, I suspect it's GTK. I also do 
that, because in Gramps' detailed log with module info, I see different 
version numbers for gtk, 2.24.28 vs. 2.24.30, and no differences for 
PyGTK or PyGObject. Python versions and Linux kernels differ too, but I 
can not imagine that they are causing this, because GTK is much closer 
to our code.

Since the diff between the 2.24.28 and 2.24.30 provides no clues in this 
area, i.e. no obvious changes in get_iter like calls, I'm at Witt's end. 
Any ideas, anyone?

thank you,

Enno