Re: gtk_tree_view so slow

[Maxim Reznik <max1 mbank com ua>]

Kristian Rietveld wrote:
> On Wed, 2003-08-27 at 08:58, Maxim Reznik wrote:
>
> > The problem I encountered is gtk_tree_view scans
> > whole tree_model when I create the form.
> That's the validation step, where all rows are being measured.
>
> > It takes too much time if there are many rows in table.
> Yes, it does for tables with more than 10000 rows.
Even worse. It takes about 6 seconds for 1000 rows.
Under M$ Windows it takes longer (23 sec).
By the way why gtk_tree_view slower under M$ Windows?
It seems it very depend on number of collumn in the table.
Table with 16 columns is so heavy to scroll.

> > Is there a way to prevent such full scan?
> There will be a way to "disable" validation in gtk+ 2.4, which will
> prevent a full scan. Until then there's not much you can do.
I considered follow ways to speed up:

1) I found gtk_tree_view_build_tree where there is loop over all rows.
I make my gtk_tree_model_iter_next trivial. It just increment row index.
All real job for positioning in the table is done in get_value.
It doesn't help much perhaps because of validation phase.

2) Next I realised get_value is called for each cell in table.
I thought I could manage it by pretending there are only 30 rows in the
table. Then I waited while user scroll down to the last row and
emitted "insert_row" signal for next 30 rows.

It doesn't help either. I think it's because validation calls
get_value(column=1, row=30) triggering my "insert_row" signal.
Singnal then triggers next validation process. And so on.

Is there a way to distinguish validation process from user
scrolling?
Could I rely on fact than first call get_value(column=1, row=30) is
validation phase and all next is visualization?

Is there any other tricky way to speed up before 2.4 version is out?

>         -Kris
Thank you Kris very much for your help.

Maxim

PS Excuse me pure English.