Wayland-native Caja in spatial mode does NOT remember window positions

[mycophobia]

Expected behaviour
When opening a previously opened folder in spatial mode ("Open each folder in its own window"), the resulting window should have the position, size, view state, and scroll position that it had when it was last closed. This is how it works in Xorg and Xwayland.

Actual behaviour
When opening a previously opened folder in spatial mode ("Open each folder in its own window"), the resulting window has the previous size, view state, and scroll position, but NOT the actual window position it had when it last closed. The resulting position is likely based on the compositor default (in labwc, it appears to be the center of the screen.) This completely defeats spatial orientation.

For myself, in practice, I'll be on Xorg for another couple of years since I'll ultimately be using Debian Stable's MATE and all of its components, so for me this isn't immediately an issue. Furthermore, I don't believe Xwayland is going away anytime soon, so as long as Caja can run as an Xwayland client then this will indefinitely be less of an issue.

Nonetheless, as it stands, Caja as a native Wayland client is broken in spatial mode. I am unsure what options there are in Wayland compositors to get and set window positions, but should there be one, Caja should leverage it when running as a Wayland client.

Steps to reproduce the behaviour
In spatial mode, open a folder and move the window somewhere. Close the window and then reopen the folder.

MATE general version
N/A; I'm testing caja-master by itself

Package version
1.29.0

Linux Distribution
Debian 13

[lukefromdc]

In Wayland it is almost impossible to do this, and it is entirely impossible to do it in a compositor-agnostic manner. Apps under wayland are normally denied the ability to set their own window positions. It is possible under wlroots to anchor a window to the screen edges and a spacing, this still requires knowing where the window whose position is to be remembered is relative to those edges. Wlroots can only set not get position relative to a screen edge so far as I know, so this may be impossible unless wayland devs loosen up on some of that security theater. The theory behind this of course is to keep an app from using an invisible (alpha of zero) window to clickjack another window. Example "mymaliciousbanker" rendering invisibly over your bank's login window on Firefox. Real world the authors of the "banker" would target Windows, iOS, and Android but probably ignore x11 and wayland as desktop Linux is such a small part of the online world and considered a more difficult target for little return. Lots of Linux attacks do exist-targetting servers that normally run headless altogether, not the end users,

[mycophobia]

Is it possible in wlroots to get a position relative to a parent window (say, the desktop for example)?

[lukefromdc]

Not that I know of, pretty sure this is SET but not GET. I myself don't know a way of doing this. If someone finds a way I can test it but I do not have the time to research this myself.

[lukefromdc]

If you need this capability you should probably use the x11 session. The wayland session does work but so far it has NOT been possible to port over all features. The main reason for writing it is insurance: to allow MATE to be installed even if distros decide to dump x11. We started seeing talk of distros not including x11 in their default installs two years ago.

[mycophobia]

Yeah, like I said as long as Xwayland sticks around and Caja is able to run as an X client then it's close to a non-issue. Just thinking ahead...