Re: Native/Extended layout

[Matthias Clasen <matthias clasen gmail com>]

On Mon, Dec 14, 2009 at 10:36 AM, Johannes Schmid <jhs jsschmid de> wrote:

>
> The new "native-layout" branch is thought to be a starting point for
> anybody wanting to work on this. It is up-to-date with master and the
> tests/testextendedlayout is kind of working (at least better than with
> current master).

So, I took a look at this, and tried to port over some of the
container support from the older extended-layout branch (bin, scrolled
window, plug/socket, treeview, cellview, cellrenderertext).

However, things don't quite work, and I think there are some
fundamental issues. E.g. Matthias decided to implement

get_desired_size (GtkRequisition *minimal,
                          GtkRequisition *natural)

instead of Havocs

get_desired_width (int *minimal, int *natural)
get_desired_height (int *minimal, int *natural)

and gtksizegroup.c:do_size_request has the assertions

g_assert (widget->requisition.width <=
              aux_info->natural_size.width);
g_assert (widget->requisition.height <=
              aux_info->natural_size.height);

But that doesn't quite make sense. With width-for-height, you would
expect that shrinking the width grows the height. So there may not be
a desired size that is larger than the minimal size in both
dimensions, and the exact minimal size may depend on whether you are
trying to minimize the width or the height.

And after implementing extended layout for the containers, most labels
in testgtk and testextendedlayout come up with much too small
allocations.

Clearly, some debugging needed.

However, I don't really have the time to sink into this, and it is not
clear to me if it would not be better to go back to Havocs original
api proposal.

Matthias