Still writing a restore tool :(

Still writing a restore tool :(
Last two hours were spent debugging an interesting problem with TreeView in PyGTK. It was too slow to parse and add all files from a backup snapshot to the tree view at once (not to mention that it took 35 Mb of RAM :P), so I decided to load the tree as needed - I would add the children of a node only when this node gets expanded. So I happily wrote a handler to 'row-expanded' event that does just that - adds some children to the newly expanded node.
Note: as the node cannot expand if it doesn't have some children already, I also add a dummy child to all directory nodes
Then the problem came up - once I enabled my handler, the nodes would not expand anymore: the expansion handles were there, I could click on them and see the CPU being chewed away by the parsing of the 6 Mb nodelist, but nothing changed in the interface - even the dummy node didn't come up.
That confused not only me, but also #pygtk people. I wrote a 15 line simple script to replicate the problem, but everything worked fine there :O. At this point I started commenting stuff out at random and found out that breaking the link between treestore and treeview (recommended in docs to avoid excessive updates) resets the expansion state. Doh.
But it was not the end yet. After that I noticed that the nodes didn't expand on the first try, but only on the second. 8) After some mental mummbo-jummbo I came to an idea that proved to be dumb, but correct. Prepare for a gem boys and girls - if, in the process of execution of row-expanded handler, at at least one point the expanding node has no children (like when you have removed the dummy node, but still haven't added the real ones) - the expansion doesn't happen!
Two bugs^Wfeatures with the same effect. Oh, the fun of debugging never stops :D