v1.4.10

Clips records on desktops that never could, and the interface a headset is addressed on is worked out from the hardware instead of typed by hand.

Added

• Clips records where the desktop has no ScreenCast portal. It built one pipeline whose source was a portal session, so XFCE, MATE, Cinnamon and plain window managers — anything using xdg-desktop-portal-gtk, which does not implement that interface — could not record at all. X11 is now captured directly there, with the same encoder, the same frame rate and the same per-channel audio tracks. Nothing to configure: the portal is asked whether it exists rather than inferred from the session type, because X11 under GNOME or KDE has a working one and must keep using it. (#214)
• Bug reports say how Clips can capture this screen — the desktop, which portal backends are installed, whether ScreenCast answers, and which GStreamer elements and encoders exist. Only when Clips is switched on. There is also scripts/diagnose-clips-capture.py for anyone who cannot open the window to produce a report in the first place. (#215)

Changed

• A headset's control interface is found from what it declares, not from a number written by hand. SteelSeries' own device specifications address it the way Windows does — by HID usage page — and never carry the interface number Linux needs, so every profile's was transcribed, and several were wrong. The number is now read back from the device's report descriptor. A profile that already names the right interface is untouched, so nothing that works today changes. (#213)

Fixed

• The Arctis Pro GameDAC has its own profile. It shared one with the Arctis 7 dongle and inherited all of it: an interface that does not exist on this device, so every command failed twice a second and its status came back empty; a battery request, which a mains-powered DAC has none of; ChatMix on the wrong command; and the wrong sidetone opcode. Read from SteelSeries' specification for this product rather than carried over by name. (#213)
• The ChatMix dial no longer skips whole ranges. The read loop waited on every interface at once, so the slowest set the pace for all of them: the dial pushes its frames on one while another blocks for a full second on a polled status frame. Reported on a Nova 7 Gen 2, where the wheel jumped from 100 to 53 rather than moving through the values. (#211)
• The Nova 5's dial interface, lost when its profile was copied. Upstream lists two interfaces for it and ASM's profile has carried one since the day it was written, so the wheel worked only when the runtime detection happened to catch a frame, and stopped whenever that cache was cleared.
• An application listed once per tick. Two sources feed the row of applications under each channel and they were drawn in sequence, the first clearing the card and the second appending to it. Once the first learned to skip the rebuild when nothing had changed, the second kept appending to a card nobody had cleared — a game running under Proton gained a copy of itself twice a second, and closing it left them behind.