Add keyboard shortcut feature

[ehamiter]

This adds the ability to assign keyboard shortcuts via the Cosmic keyboard custom settings to activate, search, select, and copy from the UI.

The command to add is:

cosmic-ext-applet-clipboard-manager --toggle

and you can assign it to any shortcut. This will show/hide the clipboard manager, and allow you to type to search, use the arrow keys to select an item, and enter to copy and close the interface. Escape also works to exit without selecting anything.

Full disclosure: I am not an expert in Rust; I'm scratching my own itch to add this feature, so there may be a more suitable way of doing this.

---

• closes Toggle with keyboard shortcut and navigate using arrows #158

[wiiznokes]

Hi, thanks for this pr. This is a good work arround while waiting for the flatpak portal implementation.

I have some questions/suggestions. I understand you're not familiar with rust, so you can ask me if you don't understand something.

• Did you test your code? Does it works?
• get_signal_file_path is duplicated
• You seems to write the timestamp but never actually use it in the code
• It would be nice if these change be in one module, e.i, the functions to read/write the signal file, and the subscription
• The subscription should be a file watcher, especially if we are doing io every 100ms
• why you define a shortcut in the app code? If i understand correctly, it should only be in the cosmic-comp config

[wiiznokes]

Note that the approach looks good otherwise

[ehamiter]

Thanks @wiiznokes -- I believe I addressed everything, please let me know if not.

Tested fully on my local machine-- able to activate, start typing to search, use arrow keys to select, and enter to populate clipboard with selection and close the UI. Escape also exits the UI without altering the clipboard.

[wiiznokes]

So i wanted to make some minor modif before merging the pr (see this branch https://github.com/cosmic-utils/clipboard-manager/tree/pr160-modif), but i've realised that the popup doesn't show near the applet icon now.
This is unfortunate. Futhermore, i think the layer surface stuff was meant to be removed from cosmic-panel, so idt it's a good idea to start using this.

Maybe @wash2 have some insight about this problem 🙏
tldr: We want to show the popup programatically. The popup doesn't show up when using get_popup, but does when using get_layer_surface

[ehamiter]

Yep, I noticed the positioning as well- I think it's due to how Wayland treats the layers, so I'm not sure how to align it with the applet icon if it goes that route.

Whatever you think is best-- I have a local copy that works for me, so I'm happy to do anything else needed to get the functionality available to others.

[OmarAhmed-A]

I LOVEE this, is it possible to have the selection ENTER trigger the paste directly?

This would help someone transitioning from windows with my muscle memory.

[OmarAhmed-A]

There's an issue where having more items in history than length and using down arrow to scroll doesn't actually scroll them into view.

[OmarAhmed-A]

There's an issue where having more items in history than length and using down arrow to scroll doesn't actually scroll them into view.

fixed in 8a94319
#177

[mbchristoff]

Been testing this for a week. functions fine on pop-os 24.04.
Only found out that from the first time it gets toggled it will use 200% cpu untill it is killed, am I the only one with that problem?
It seems to loop around this small snippet from strace:

madvise(0x7d308802b000, 4096, MADV_DONTNEED) = 0
madvise(0x7d3088059000, 4096, MADV_DONTNEED) = 0
madvise(0x7d3088064000, 4096, MADV_DONTNEED) = 0
madvise(0x7d3088041000, 4096, MADV_DONTNEED) = 0
madvise(0x7d308803d000, 4096, MADV_DONTNEED) = 0
madvise(0x7d308805f000, 8192, MADV_DONTNEED) = 0
madvise(0x7d3088053000, 12288, MADV_DONTNEED) = 0
madvise(0x7d3088066000, 12288, MADV_DONTNEED) = 0
write(11, "\1\0\0\0\0\0\0\0", 8)        = 8
write(11, "\1\0\0\0\0\0\0\0", 8)        = 8
write(11, "\1\0\0\0\0\0\0\0", 8)        = 8
write(11, "\1\0\0\0\0\0\0\0", 8)        = 8
timerfd_settime(6, 0, {it_interval={tv_sec=0, tv_nsec=0}, it_value={tv_sec=0, tv_nsec=0}}, {it_interval={tv_sec=0, tv_nsec=0}, it_value={tv_sec=0, tv_nsec=0}}) = 0
epoll_ctl(4, EPOLL_CTL_MOD, 6, {events=EPOLLIN|EPOLLPRI|EPOLLERR|EPOLLHUP|EPOLLONESHOT, data={u32=4294967295, u64=18446744073709551615}}) = 0
epoll_pwait(4, [{events=EPOLLIN, data={u32=0, u64=4294967296}}], 1024, -1, NULL, 1000000000) = 1
read(5, 0x7fffb4abbc98, 8)              = -1 EAGAIN (Resource temporarily unavailable)
epoll_ctl(4, EPOLL_CTL_MOD, 5, {events=EPOLLIN|EPOLLPRI|EPOLLERR|EPOLLHUP|EPOLLONESHOT, data={u32=4294967295, u64=18446744073709551615}}) = 0
read(8, "\2\0\0\0\0\0\0\0", 8)          = 8
write(11, "\1\0\0\0\0\0\0\0", 8)        = 8
write(11, "\1\0\0\0\0\0\0\0", 8)        = 8
write(11, "\1\0\0\0\0\0\0\0", 8)        = 8
openat(AT_FDCWD, "/run/user/1000/cosmic-clipboard-manager-toggle", O_RDONLY|O_CLOEXEC) = 24
statx(24, "", AT_STATX_SYNC_AS_STAT|AT_EMPTY_PATH, STATX_ALL, {stx_mask=STATX_ALL|STATX_MNT_ID, stx_attributes=0, stx_mode=S_IFREG|0664, stx_size=13, ...}) = 0
read(24, "1765188137517", 13)           = 13
read(24, "", 32)                        = 0
close(24)                               = 0
madvise(0x56928c431000, 4096, MADV_DONTNEED) = 0
madvise(0x56928c43a000, 4096, MADV_DONTNEED) = 0
madvise(0x56928c44c000, 4096, MADV_DONTNEED) = 0
madvise(0x56928c475000, 4096, MADV_DONTNEED) = 0
madvise(0x56928c451000, 4096, MADV_DONTNEED) = 0
madvise(0x56928c504000, 4096, MADV_DONTNEED) = 0

Could it be that it's missing a pause somewhere?

[wiiznokes]

i'm not sure but i think the line that block us to use a regular xdg shell popup is this one: https://github.com/pop-os/cosmic-panel/blob/8eb8a1b6305213ec7402cb2ec24bef6b501b978a/cosmic-panel-bin/src/xdg_shell_wrapper/server/handlers/xdg_shell.rs#L45

Maybe we could add a check to bypass this if the client is not sandboxed

[GonzRon]

sorry, i use this now and want to pull in the latest greatest changes so excuse me for butting in but hopefully this can help move this along:

The 200% CPU bug is: The toggle file watcher subscription spinning in a tight loop. Look at the strace cycle:

openat("/run/user/1000/cosmic-clipboard-manager-toggle") → open file
read("1765188137517") → read timestamp
close(24) → close file
write(11, "\1\0\0\0\0\0\0\0") ×4 → wake eventfd (trigger re-render)
epoll_pwait(...) → wait for events
→ immediately wakes up again → repeat forever

The problem is a feedback loop in the toggle subscription:

1. The file watcher sees /run/user/1000/cosmic-clipboard-manager-toggle exists
2. It reads the timestamp, fires a toggle event
3. The toggle event causes a UI update, which writes wake events to the eventfd (the 4x write(11, ...))
4. The UI re-render cycles back to the subscription, which checks the file again
5. The file still has the same timestamp — but nothing marks it as "already handled"
6. So it fires the toggle event again → goto 3

The fix needs to be in the toggle subscription logic itself — either:

• Delete the file after reading it (so the next poll finds nothing)
• Track the last-seen timestamp and ignore if unchanged
• Replace the file-polling mechanism entirely with a Unix socket or D-Bus signal that's inherently one-shot

This is a bug in PR #160's toggle implementation, not in the clipboard data path that PR #170 touches.

[Lcstyle]

I decided to go more with a dbus approach -> https://github.com/Lcstyle/clipboard-manager/tree/feat/dbus-toggle for a oneshot event based trigger rather than polling - reason being :

Three crash waves — all SIGABRT (Rust panic). The panic message from the coredump:

called Result::unwrap() on an Err value: PoisonError { .. }
thread 'main' panicked: panic in a destructor during cleanup

And the root cause of the CPU loop is in our ipc.rs. The notify watcher on /run/user/1000/ is problematic — that directory is extremely busy. But the critical bug is here:

loop {
rx.recv().await; // ← returns None instantly if sender drops
output.send(AppMsg::CheckSignalFile).await.ok();
}

If the notify watcher thread dies for any reason, rx.recv() returns None instantly and we never check the return value — creating an infinite tight loop flooding the app with CheckSignalFile messages. That's the 200% CPU,
eventually leading to the PoisonError and abort.

The notify-based approach is fundamentally fragile for /run/user/1000/. Let me replace it with a simple polling approach — one stat() call every 250ms uses negligible CPU and can't possibly loop.