Determining the window frame extents on Linux

[rombust]

It seems that the "X11Window::update_frame_extents() " function is broken on various Linux window managers

The update_frame_extents() function calculates the size of the window decorations for top, right, bottom and left.

For example this is broken in the Cinnamon Window Manager (known as Muffin)

This is used for setting absolute window positions on the screen by using the clan::DisplayWindowDescription client_area flag's

This used to work in the distant past, however times have changed with what is actually supported on the Linux window manager

Currently we use "_NET_REQUEST_FRAME_EXTENTS" for unmapped windows. Various WM's have this unimplemented. Since the spec says " The Window Manager MUST respond by estimating the prospective frame extents". The word "estimate" means it cannot be relied on, even if it's actually implemented.

For mapped windows, I tried waiting for "_NET_FRAME_EXTENTS" event and using XGetWindowProperty to retrieve the extent , and it still returns 0,0,0,0. Even if this worked, it would be a hack.

I remember looking at this problem many moons ago. In my opinion, we shouldn't pretend it works, and remove the "update_frame_extents" function.

To summarize, Essentially, there is no fully reliable way via X11 to measure the true visible frame for compositing WMs.

Note, in Wayland you also cannot know the frame size before showing the window.

[dpjudas]

X11 and Wayland both have in common that if you follow the core spec to the letter then all you can reliably do is present a window like it is 1989 (*). With Wayland you can't even reliably activate a window, set cursors, support modern PC hidpi displays that have 150% scale factors, etc. It all requires extensions to Wayland to do and you just have to assume they are around.

What really matters is this question: does the mainstream window managers support the extension or not. Then fall back to semi-broken behavior if a WM doesn't support it. Returning zero for the extents is perfectly acceptable for that scenario IHMO.

There are situations where you want to position a window at a certain location including its border. And there are situations where you want the size and position to be the content area. Of course if nothing mainstream today supports the extension anymore we just have to accept that Linux won't let you position a window as precisely as other platforms will.

Speaking of Cinnamon - it would be nice if they could fix their Wayland implementation. :)

*) With Wayland you can't even do that. You can do a kiosk application at best.

[dpjudas]

I took a quick look at Muffin's source code and it does implement _NET_REQUEST_FRAME_EXTENTS: https://github.com/linuxmint/muffin/blob/master/src/x11/events.c#L948

In what way is it broken?

[rombust]

So muffin does support it. Next week, I will try creating a minimal example (not using clanlib) to figure out what's going on.

[rombust]

I have updated how ClanLib X11 handles frame extents.

It "should" work on more WM's now. Well it works here on Linux Mint with the examples: Basic2D, MultiWindow, UI/HelloWorld (including the popup windows) and Tests/Display/WinPos

The were some issues. " _NET_REQUEST_FRAME_EXTENTS " requests the "_NET_FRAME_EXTENTS" property to be filled. But there is no guarantee that a PropertyNotify event is sent when the property is actually set by the WM. Thus using "XCheckTypedWindowEvent" failed when waiting for the change to occur.

When the Window is mapped, we now capture "_NET_FRAME_EXTENTS" PropertyNotify event in the message handler: X11Window::process_message()

For windows that don't have a frame (popup windows - using override redirect). We ignore the frame extents. Their size were being reported including a frame, instead of 0,0,0,0.

It was a challenge to get the code "working". I tried AI generated code, but that really didn't help. It had various mistakes and contradictions.

[dpjudas]

I don't understand why the event isn't firing. Muffin's code calls XChangeProperty with PropModeReplace as a response to _NET_REQUEST_FRAME_EXTENTS.

According to Xlib's documentation calling this function generates a PropertyNotify event on the window. It does not mention any situation where it isn't generated.

I only use AI for situations where I need a solution for the beaten path where I know there's a hundred examples out there in the wild it can loot from. The finer details of X is not one of those situations as only a handful of developers for toolkits actually use X directly.