7 Commits

Include renames

Author	SHA1	Message	Date
harry	795118c200	feat: record at the source's size, only once capture works ... Encoders and the segment output are created when the capture source first has a picture, at its own size (portrait monitors record portrait) unless a resolution is configured, so nothing is written while waiting for a source. SPEC.md records both decisions; validation docs cover how sessions should end when the game closes.	2026-10-08 15:34:42 +02:00
harry	afc54b363e	fix: time out a stalled portal and sample motion across segments ... Hyprland's portal sometimes creates the session and never answers source selection, before any picker appears. Recording now fails if the portal has not reached its picker within 10 s; time in the picker is not limited. The motion check uses the segment with the most motion, since capture can start late.	2026-10-08 15:16:52 +02:00
harry	06377de707	fix: fail on refused portal requests and catch black recordings ... The portal reports refusals only in OBS's log; a denied, cancelled or failed source selection now fails with an actionable reason instead of waiting forever. check --expect-motion rejects black or frozen footage, and the hardware script checks only runs that recorded, tests restore tokens in their own step, and treats closing a window as informational: on Hyprland the stream keeps its last frame, so sessions must end on game exit.	2026-10-08 15:01:25 +02:00
harry	9178a303ca	test: add Linux rig validation suite ... Runs encoder verification, Wayland and X11 capture with segment and remux checks, and five failure scenarios that must end in an honest failed status: source lost, window closed, unwritable segments, low disk space and a portal that grants no source.	2026-10-08 14:29:28 +02:00
harry	444758f9ec	feat: measure footage across segments and test X11 capture ... Matroska output can stay buffered until a segment closes, so progress counts bytes across all segments and the stall limit scales with segment length. Unknown capture source types now fail instead of recording an empty placeholder. The rig gains an Xwayland mode for Xcomposite window capture.	2026-10-08 14:25:23 +02:00
harry	263e7a964e	feat: detect lost sources, stalled writes and low disk space ... PipeWire capture keeps its size and last frame after its stream stops, so the recorder follows the stream state OBS logs. Recording fails when the segment on disk stops growing (OBS kept counting bytes into a dead muxer and missed a full disk) and stops while enough space remains to finalise the last segment. Failures stay visible after the output stops. Adds a check command that verifies keyframe-aligned segments and a remux across segment boundaries, and gives the rig desktop audio.	2026-10-08 14:19:57 +02:00
harry	f2eb7139da	feat: add OBS recording engine and validation probe ... clipbin-obs starts the pinned libobs on a dedicated thread, opens exactly the bundled plugin allowlist, and records official capture sources through one hardware encoder into keyframe-aligned Matroska segments. Status reaches Recording only once the source has a picture and the segment grows on disk. clipbin-probe lists and verifies hardware encoders and records from a real capture source, printing status, health and segment events for validation.	2026-10-08 14:12:06 +02:00