Need a temp-focus option, similar to the temp-raise one.

(In reply to comment #0)
> Is it possible to make fluxbox move the focus temporarily to the
> currently-raised window (if the config contains <raise>yes</raise>) when doing

Fluxbox? Wrong bugzilla :P

(In reply to comment #2)
> (In reply to comment #0)
> > Is it possible to make fluxbox move the focus temporarily to the
> > currently-raised window (if the config contains <raise>yes</raise>) when doing
> 
> Fluxbox? Wrong bugzilla :P

Sorry sorry sorry. Freud seems to be in my head.

I actually meant openbox of course :)

Created attachment 2542 [details]
The patch implementing the feature for cyclewindows.c

I had a few free days at last to implement the feature by myself.

I added an option "focus" (similar to "raise") in "NextWindow" action controlling whether the active window gets the focus while switching between windows using Alt-Tab.

Unlike stacking_temp_raise(), we cannot just reorder focus_order contents since such reorderings influence the way other parts of openbox works. For example, "NextWindow" algorithm uses this list via focus_cycle() and my early implementations led me to strange bugs so I decided just not to modify the list.

The solution is to implement focus_temp_set_client() (analog of stacking_temp_raise()) which moves "visual" focus to the current active window and focus_restore() (analog of stacking_restore()) returning the focus back to the original focus_client, focus_client and focus_order are unused. Besides, when switching between windows we disable focus_set_client() execution.

So, the questions:
1. Does the patch look correct or did I miss anything in openbox sources?
2. Can this be commited to mainline?
3. Is the similar option required for directionalwindows.c?

Any comments on the patch?

I think this needs to be done without calling client_focus().  I don't want to actually move focus until it's done.

Setting _net_active_window will get it showing in other apps.
It needs to have the frames repaint so need a way to say to the frame "display this focus state".

Frame currently has a bool focused but we don't want to change its actual state.

In my first though it would take something like this added to Frame..
bool override_fake_focused - TRUE when we display focus from fake_focused instead of focused.
bool fake_focused - TRUE if the window should fake being focused, FALSE for unfocused.

The modified version of the patch. Changes:
1. A very annoying hard-to-reproduce issue with focusing incorrect windows fixed;
2. The patch is updated to openbox 3.5.0;
3. An unnecessary focus_restore() call removed.

Created attachment 3148 [details]
Newer patch version

> I think this needs to be done without calling client_focus().  I don't want to
> actually move focus until it's done.

You are right. Just calling client_focus() was a stupid idea and caused strange and very annoying effects mentioned in the previous message. I discovered that it was due to sending notification when "temporarily" setting the input focus. Calling client_focus() without a notify event seems to fix the issue for me.

> Setting _net_active_window will get it showing in other apps.
> It needs to have the frames repaint so need a way to say to the frame "display
this focus state".

Sorry, I don't fully understand what you mean. May be it's because my X11 programming knowledge are not too deep.

If you mean we don't have to call XSetInputFocus() then you're wrong. I tried that and just setting _net_active_window doesn't make openbox to work as expected (though I looked at tint2 sources and it seems it just checks NET_ACTIVE_WINDOW). I guess it's just insufficient to just set the NET_* and we also have to somehow notify the X that we have set it, and I guess XSetInputFocus does that for us.

> Frame currently has a bool focused but we don't want to change its actual state.
Will this affect only openbox or will it affect external programs such as panels (like tint2 or bmpanel), docks, etc... as well?

(In reply to comment #9)
> > I think this needs to be done without calling client_focus().  I don't want to
> > actually move focus until it's done.
> 
> You are right. Just calling client_focus() was a stupid idea and caused strange
> and very annoying effects mentioned in the previous message. I discovered that
> it was due to sending notification when "temporarily" setting the input focus.
> Calling client_focus() without a notify event seems to fix the issue for me.
> 
> > Setting _net_active_window will get it showing in other apps.
> > It needs to have the frames repaint so need a way to say to the frame "display
> this focus state".
> 
> Sorry, I don't fully understand what you mean. May be it's because my X11
> programming knowledge are not too deep.
> 
> If you mean we don't have to call XSetInputFocus() then you're wrong. I tried
> that and just setting _net_active_window doesn't make openbox to work as
> expected (though I looked at tint2 sources and it seems it just checks
> NET_ACTIVE_WINDOW). I guess it's just insufficient to just set the NET_* and we
> also have to somehow notify the X that we have set it, and I guess
> XSetInputFocus does that for us.

It really should be enough to set the NET_ACTIVE_WINDOW property to change what panels are showing as focused.

Doing XSetInputFocus is going cause race conditions with apps setting focus. We can't tell when we receive the focus change if it came from us or from another app. So then, how do we know to change the focus order or not?

We should just make the window's frame look focused by setting internal variables on it, and set NET_ACTIVE_WINDOW for panels to match it.

> > Frame currently has a bool focused but we don't want to change its actual state.
> Will this affect only openbox or will it affect external programs such as
> panels (like tint2 or bmpanel), docks, etc... as well?