[yanscholze]

Adds experimental GNOME Wayland support for modifier + file dragging. A GNOME Shell extension supplies pointer state through a private Unix socket, allowing Filee to detect gestures without /dev/input access or membership in the input group.

The extension positions the conversion wheel once per opening, allowing users to move onto a format. The integration reuses the existing gesture detector and respects configured modifiers, drag thresholds, Pause, scope and excluded applications. Existing X11 and KDE gesture paths remain separate.

Includes Linux packaging, extension installation/removal, connection status messages and setup instructions.

Implementation details, limitations and validation results are documented in docs/GNOME-WAYLAND.md.

Validation:

110 targeted application tests passed.

Core, CLI and platform test suites passed, with platform-specific skips.

Real Gio socket tests passed.

An isolated GNOME Shell 50 session successfully handled a native GTK4 Wayland file drag into Filee’s XWayland wheel and produced a PDF while preserving the source file.

X11 startup continued to load the existing input hook.

Limitations:

Manual testing in Nautilus, fractional scaling and multiple monitors remains pending.

Global selection shortcuts, hold gestures and Tab/wheel flipping during dragging are not provided by this integration.

The full suite was not entirely green: Setup and Engines failures are documented, and the lengthy application rendering batch was interrupted after approximately 15 minutes.

[KnifeLemon]

Obrigado pela contribuição, @yanscholze! Escrevo a revisão em inglês para os outros mantenedores, mas pode responder em português.

Thanks for this, it's a careful piece of work. Using the Shell for pointer state avoids both the input group and /dev/input, and the socket checks (0600, peer UID), the bounded writes and the "don't pick up a drag that's already held" rule all look sensible. I merged it with current main on Windows and ran dotnet format --verify-no-changes and the build (both clean). The App tests for the GNOME bridge, triggers, gesture detector and localization passed (114), as did Platform (54) and Core (510). node --check passes on the extension, and sh -n passes on both scripts. I can't run a GNOME session here, so the real-session part rests on your testing. Thanks for writing down exactly what you did and didn't test.

A few things before it can go in:

1. docs/GNOME-WAYLAND.md. Docs in this repo are in English, and this file reads as a work log. It mentions an earlier version that isn't in the repo, local artifacts/ folders and the results of one test run. The validation record fits better in the PR description. Could the doc become a short English page with setup, what works, limits and troubleshooting? It also goes into the Linux package, so users will read it.

2. Overlap with #367 (native Wayland and Hyprland, being finished now). Both change LinuxPlatformServices.cs, installer/linux/INSTALL.txt and CONTRIBUTING.md, and git reports conflicts in all three. install.sh also conflicts with #360. #367 will probably go in first. After that, please check that _isDonut also matches the donut when Filee runs as a native Wayland client (FILEE_USE_WAYLAND=1), where the window has an app id instead of an X11 WM class.

3. Updates. When Filee is updated, either in the app or with the upgrade mode #360 adds, nothing touches ~/.local/share/gnome-shell/extensions/filee-donut@filee.sh, so the old extension stays. Also, install.sh refuses to run when that folder already exists, even if it carries our own .filee-package marker. Could install replace a folder that has our marker? And could the app say when the extension's protocol version doesn't match?

4. Smaller:

• Error = "GNOME pointer integration requires XDG_RUNTIME_DIR." reaches the Triggers page in English. A key, or the log only, would be better.
• The 16 ms timer runs for the whole session even when Filee isn't connected. It does almost nothing then, but it could start on incoming and stop when the connection closes.
• README: only the English and Portuguese READMEs changed, and there are 14. The experimental KDE path isn't in the READMEs either, so it may be simplest to leave the READMEs out and keep the details in INSTALL.txt and the doc.

Translations: the two new strings are English in 12 of the 14 language files. That's fine, we'll translate them ourselves.

CI hasn't run on this PR yet (first-time contributor), and we'll start it. Thanks again!

To get this into the next release, we'll push the remaining changes to your branch ourselves today (maintainer edits are on). Have a look at the commits when they land, and tell us if anything looks off.

[KnifeLemon]

@yanscholze I pushed the remaining changes to your branch, as promised. #367 and #360 are on main now, so the branch is merged with main and only the GNOME files differ. Have a look, and tell us if anything looks off.

1. Doc: your new docs/GNOME-WAYLAND.md was already the short user page we hoped for. I only added an "Updates" section, the log location, and what happens under native Wayland (FILEE_USE_WAYLAND=1).
2. feat(linux): add native Wayland support and generalize compositor gestures #367 overlap: the conflicts in LinuxPlatformServices.cs, INSTALL.txt, CONTRIBUTING.md and install.sh are resolved. Avalonia's Wayland backend doesn't set an app id, so under native Wayland the donut window has neither a WM class nor an app id. The extension now finds the donut by its title plus one of: WM class / app id, the PID of the Filee connected to the socket, or the process name.
3. Updates: install.sh replaces an extension folder that carries our .filee-package marker and still refuses one without it. --upgrade (which the in-app updater runs) refreshes our extension, staging the new copy first so a failure leaves the old one in place. If the extension isn't there yet on GNOME, it's installed but not enabled. The protocol version is checked on every line, and a mismatch shows on the Triggers page. To make that possible, I took the version out of the socket name (pointer-v1.sock → pointer.sock) and put it in a PROTOCOL_VERSION constant: a version change now shows as "versions differ" instead of "extension not running".
4. Smaller: the XDG_RUNTIME_DIR error goes to the log only. The 16 ms timer starts on a connection and stops when it closes, and placing the donut has its own short timer. The README changes are left out.

Translations for the GNOME strings are in all 14 languages (your Portuguese is kept). I also updated the general Wayland notice, which still said "KDE Plasma only" after #367 added Hyprland. The GNOME notice also no longer shows when FILEE_WAYLAND_GESTURES=0.

Build, dotnet format, the App tests for GNOME, triggers, gestures and localization (122), Platform tests, node --check and sh -n pass here. I can't run a GNOME session, so if you can try the branch on yours once more (an upgrade over an installed copy too), that would be great. Obrigado!

[yanscholze]

Tested the latest changes on Fedora GNOME Wayland.

Everything is working correctly on my side:

• dotnet build succeeds.
• Tested the new in-place upgrade over my existing Filee installation — it completed successfully and preserved the installation correctly.
• The GNOME extension was updated to v2, remained enabled, and loaded successfully after signing out/in.
• Ctrl + drag from Nautilus correctly opens the donut at the cursor and conversion works as expected.
• Also tested with FILEE_USE_WAYLAND=1; Ctrl + drag, donut positioning, and conversion all worked correctly with native Wayland as well.

I didn't notice any regressions or unexpected behavior during testing.

Looks good on Fedora GNOME Wayland from my side. Thanks for the changes!