Fix Discord screen share stopping after about a second
Symptom, logs, cause and the restart workaround all match xdg-desktop-portal-hyprland#418 verbatim, but the record is now obsolete as written. That issue was closed by upstream commit c46162255e00 ('core: fix loop hangup detection', 2026-07-24), which shipped in XDPH v1.4.1 (2026-07-29). Arch extra currently carries 1.4.1-1. Telling a user to permanently pin force_shm (which the XDPH wiki explicitly documents as the slower path, especially at high resolutions) when a plain system upgrade fixes the bug outright is the wrong first move. The force_shm advice itself is valid (the option exists, and the Omarchy caveat about editing the existing screencopy block is correct) but belongs after the upgrade, scoped to genuine multi-GPU DMA-BUF allocation failures.
force_shm = true sends every frame through shared memory instead of a GPU buffer. At 4K/high refresh this raises CPU use noticeably. Only set it if DMA-BUF allocation is actually failing in the logs.
Symptom
Screen share starts and immediately stops after about a second, and the list of shareable windows is frozen: opening or closing windows no longer updates it. Portal log shows: [WARN] [pipewire] Asked for a wl_shm buffer which is legacy., [WARN] [pw] DMA-BUF allocation failed, falling back to SHM, [LOG] [sc] Incompatible formats, renegotiate stream, [ERR] [screencopy] tried scheduling on already scheduled cb (type 1).
Cause
XDPH's screencopy session state was left corrupt after a previous share ended (buffer renegotiation between DMA-BUF and SHM raced and double-scheduled a frame callback), so the portal kept running but every subsequent ScreenCast session died immediately. Reported against xdg-desktop-portal-hyprland 1.4.0 with Hyprland 0.56. **This was fixed upstream** by commit c46162255e00 ('core: fix loop hangup detection', 2026-07-24), released in XDPH v1.4.1 (2026-07-29), which Arch extra now carries, so on a current system this is an out-of-date XDPH rather than a live bug.
Fix
First: this exact bug (xdg-desktop-portal-hyprland#418) was fixed upstream and the fix is in XDPH 1.4.1, which is already in Arch extra. Upgrade before doing anything else:
sudo pacman -Syu
pacman -Q xdg-desktop-portal-hyprland # want 1.4.1 or newer
Then restart the backend once:
systemctl --user restart xdg-desktop-portal-hyprland.service
That restart is also the immediate recovery if you hit the frozen-source-list state on an older build. No logout needed.
Only if you are on 1.4.1+ and DMA-BUF allocation still genuinely fails (real multi-GPU / hybrid laptops) fall back to shared memory in ~/.config/hypr/xdph.conf:
screencopy {
max_fps = 60
force_shm = true
}
force_shm is a real XDPH option, but SHM is slower than DMA-BUF (noticeably so at high resolution), so treat it as a workaround and not a default. Omarchy ships its own ~/.config/hypr/xdph.conf containing a screencopy { } block with allow_token_by_default and custom_picker_binary. Add force_shm inside that existing block rather than creating a second one. Restart the portal again afterwards.
Verify
Start a share, stop it, and start a second one. The second share must survive past a few seconds and the source list must refresh when you open a new window.