enricobuehler
cb4690f216
Three fixes to the same mistake: deciding what the Vulkan Video backend can do
from a table in our heads rather than from the driver, and routing everything
that didn't fit to libav VAAPI — where a session loses real RFI recovery and
the cursor blend for no reason the hardware asked for.
**Capability probe, per codec AND depth.** `probe_encode_support`'s "is there
an encode queue" boolean becomes `VulkanEncodeCaps { supported, eight_bit,
ten_bit }`, answered by `vkGetPhysicalDeviceVideoCapabilitiesKHR` against the
very profile chain the session open builds. So the dispatcher's prediction
cannot disagree with reality: a capable device keeps the Vulkan path, an
incapable one routes to VAAPI BEFORE burning a failed open, and the
cursor-blend mirror stays honest for free. This is the shape the direct-SDK
NVENC path already uses for its codec GUIDs.
**AV1 10-bit.** It was excluded on a guess about driver coverage; now the
device answers. `color_config()` carries `high_bitdepth` + the BT.2020/PQ CICP
triplet in both the `StdVideoAV1ColorConfig` and the sequence-header OBU we
bit-pack ourselves — they must stay identical or the driver's frame OBUs parse
against a header we didn't write. `high_bitdepth` sits BEFORE the CICP bytes,
so getting it wrong doesn't just mislabel the depth, it puts every following
field one bit out of phase; the new test reads the packed bits back.
**Zero-CSC RGB-direct in HDR.** The EFC probe assumed BT.709 and BGRA. It now
asks for the model this session's colourimetry needs (`MODEL_YCBCR_2020` for
10-bit — the extension has always had it) and for the CAPTURED format as an
encode-source format, and the session create-info selects the matching model.
An HDR session with no pointer to composite therefore hands the captured
buffer straight to the fixed-function front end and runs no host CSC at all.
Sessions that DO composite a pointer keep the compute CSC, unchanged: the EFC
cannot blend, and that rule outranks everything.
Also: `can_encode_10bit` on AMD/Intel now reports the union of VAAPI's and
Vulkan Video's answers instead of VAAPI's alone. `open_amd_intel` tries Vulkan
first and falls back, so either one being able to encode Main10 makes the
session 10-bit-capable — answering `false` because only one of them said yes
stranded encodable HDR sessions at 8 bits.
Release notes
One file per stable release: docs/releases/vX.Y.Z.md. Its contents become the Gitea release
body verbatim and are the source of the Discord #releases announcement.
Why this exists
Releases used to be created by CI with an empty body; the notes were pasted in by hand afterward. That left a window where the release — and anything announcing it — carried no notes. Now the notes are authored before the tag is pushed, as part of the version bump, so the release is born complete and the announcement always has something to say.
The flow
- Write the notes. Add
docs/releases/vX.Y.Z.mdin the same commit (or PR) as the version bump. CopyTEMPLATE.mdand fill it in. This file is the single source of truth for the body. - Tag & push.
git tag -a vX.Y.Z … && git push origin vX.Y.Zfans out to the build workflows. Whichever one wins the create race seeds the release body from this file (scripts/ci/gitea-release.sh→ensure_release, and its PowerShell twin). The release page shows the notes immediately. - Wait for green. Let every platform's CI finish and go green.
- Announce. Dispatch the
announceworkflow (.gitea/workflows/announce.yml) with the tag. It re-asserts this file over the live release (so any late edit wins) and posts an embed to Discord#releases. Pressing "go" is the quality gate — a half-built release is never announced. Stable-only; a-rctag is refused unlessallow_prerelease=true.
Editing the notes after the tag is fine: update this file, then re-run step 4 (or PATCH the body via the API) — the announce step always re-syncs from the file, so the file stays authoritative even across a tag re-point.
Canary / -rc builds have no file here on purpose: they get no curated body and are not
announced.
Voice & format
Write for the people who USE Punktfunk to stream their games and desktops — not for the people who
build it. A non-engineer should finish knowing what's new and whether it affects them; an engineer
should never be confused or forced to decode internals. (See any recent vX.Y.Z.md for the target.)
- Lead with the benefit. Each entry = what the user can now do, what now works, or what stopped going wrong — in their words. Implementation is not the story.
- No internal vocabulary in the body. No protocol/message names, code type names, hex codes or hardware IDs, crate/component names, or API symbols. Translate any essential detail to plain language. Name things users recognize (iPad, Apple Pencil, Steam Deck, Android TV, the Windows sign-in screen) — not subsystems.
- Group as New / Improved / Fixed, each a bold one-line lead-in + a tight plain explanation.
Skimmable. The lead-in text before the first
##is what the Discord announcement shows, so make it a real, plain-language summary. - Be specific and honest — no vague "various improvements"; a reader should know exactly what changed.
- Compatibility line up top, in plain terms: can they update one side at a time? does their existing setup keep working? No version numbers in the lead.
- All protocol / ABI / driver / embedder detail goes in ONE
## Under the hood (for developers)section at the very bottom — the only place internal names and version numbers belong, clearly optional. The old dense engineering style survives only there.
The short annotated-tag message stays separate and short (a headline + a paragraph); it is the tag object's message, not this file.