GtkTreeModelFilter

[Vikram Noel Ambrose <noel ambrose gmail com>]

Not sure if this is a bug or just another one of GTK's quirks.

When using a GtkTreeModelFilter to filter a GtkTreeStore, if a top level 
row node (parent?) is set to hidden (via returning FALSE in the 
visibility function or otherwise via the visibility column), it remains 
hidden even if a sub node (child) has been set visible. Thus in essence 
the filtered view hides the child even though it requested to be seen.

When a child has requested to be visible, the view must make sure all 
parents required to make the child visible, are indeed visible as well 
regardless of the parent's setting.

According to the docs, GtkTreeModelFilter provides two ways of 
"filtering" rows out of a view. Both ways make it very difficult to 
ensure the parent(s) are explicitly set to visible if only the child 
matches the filter expression.

Example: Filter based on textual string comparison of row text

- Filter Method 1.
If one chooses to employ the "visible function" method of filter, the 
function is called once for every row in classic tree depth fashion. ie.

P1
-> C1
-> C2
   -> GC1
   -> GC2
-> C3
P2
-> C1
-> C2
etc...

would result in "visible function" being called for 
P1,C1,C2,GC1,GC2,C3,P2,C1,C2 etc...

Hypothetically, if P1 does not match the search expression, return 
FALSE, its marked hidden. C1 comes round, and it matches the search 
expression, return TRUE, its set to visible. C2 is next and it doesn't 
match the expression, return FALSE, its marked hidden etc...

However when the view is drawn, C1 will not be visible. The view hides 
C1 because it's parent P1 was marked hidden.

How do you counter this behaviour? Even if you look up "parents" for the 
arbitrary nodes passed to the "visible function", how do you change 
their visibility statuses?

- Filter Method 2.
Using the "visible column" approach it becomes possible, but not very 
plausible.

One can call gtk_tree_model_foreach() and walk through all the 
G_TYPE_BOOLEAN fields of the corresponding rows and mark them TRUE/FALSE 
and look up parents of nodes (like C1) and mark them TRUE even if they 
did no match the search criteria. However looking up parents 
continuously like this takes a massive performance hit when you have 
large stores.

Thus both filter approaches are useful when only the parent node is 
basis for filter criteria. In a list store this is perfectly fine, but 
in a tree store it is definitely not.

Does anyone know how to get GtkTreeModelFilter to treat it's model more 
like a tree and less like a list?

thanks,

Vik.