CSD and window resizing, once again

I'm referring back to
http://listengine.tuxfamily.org/lists.tuxfamily.org/sawfish/2015/07/msg00007.html

I've not worked on coding window managers so I'm not confident about 
much of the following, but anyway, here goes...

1) The problem I spoke of -- namely GTK CSD windows wanting to 
resize themselves arbitrarily when handled under Sawfish -- seems 
almost a Heisenbug: when I'm just going about my business it seems 
to happen a lot, but when I stop to ask, under exactly what 
conditions does this occur?, the problem mostly goes away. I take 
this to mean that at some subconscious level I must be using the 
mouse a bit differently when I'm thinking about what exactly it's 
doing. But I'm not a specially sloppy mouse user and I maintain that 
this is a real problem.

2) At a "phenomenological" level, it seems you can tell something 
has gone wrong if the cursor turns into a "+" shape: this seems to 
mean that a CSD window is now going to resize itself when you next 
move the mouse (even if this doesn't make any sense, given what 
you've clicked on to date).

3) In the sawfish C code, I took a look at button_press() in 
events.c, and the most relevant part seems to be the call to 
find_frame_part_by_window(). Here's my understanding: if the button 
press relates to a "frame_part" (i.e. part of a window frame) then 
it's something that Sawfish should act on. So I put in a print to 
stderr when find_frame_part_by_window() returned non-NULL. Most of 
the time this triggered under the right conditions (e.g. when I 
click somewhere on a Sawfish window-frame), but... it sometimes 
triggered when I clicked on a control within a GTK CSD window.

4) It seems to me that find_frame_part_by_window(), as called from 
the button_press event, should NEVER return non-NULL if you're 
clicking within a CSD window, since no part of that window 
corresponds to part of a Sawfish-handled window frame.

5) It therefore seems that there's some conditionality missing from 
the Sawfish function find_frame_part_by_window(). Now, as regards 
what exactly that conditionality ought to be, I'm afraid I run out 
of gas. Except that I think it should involve detecting CSD via mwm 
hints. So far as I can tell, there's no mechanism in the Sawfish 
code to detect mwm hints, and I suspect that's the problem.

-- 
Allin Cottrell
Department of Economics
Wake Forest University

-- 
Sawfish ML