chore(release): cut 0.34.0 #467

Merged

enricobuehler merged 2 commits from release/v0.34.0 into main 2026-08-31 00:25:05 +00:00

Conversation 0 Commits 2 Files Changed 9

+348 -40

enricobuehler commented 2026-08-30 23:46:57 +00:00

Owner

Copy Link

Copy Source

Version bump, CHANGELOG section, user notes and Play notes for v0.34.0. 140 commits since v0.33.0.

The eight files

The same bump surface as v0.33.0, plus one drift fix:

File	Why
Cargo.toml	[workspace.package] version → 0.34.0
Cargo.lock	cargo update --workspace; diff is 36/36 versions-only
api/openapi.json, docs-site/public/openapi.json	info.version stamp; the two are byte-identical
CHANGELOG.md	the v0.34.0 section, 199 lines
docs/releases/v0.34.0.md	the release body, seeded verbatim by CI
docs/releases/whatsnew/v0.34.0.txt	Play notes, 493/500 chars, unique
scripts/ci/docs-undocumented-env-baseline.txt	prunes one stale entry, below

Why minor, not patch

Not a renumber — what the tree gained. The quick-action ring and its editors on every shell, the on-screen virtual controller on Android and Apple touch devices, and the Steam Controller 2 passthrough leaving Linux for Apple, Android and Windows hosts.

⚠️ The embedder trap: the ABI moves twice, and one step widens a struct

Every recent release has been able to say "nothing versioned moves". This one cannot.

ABI_VERSION goes 26 → 28:

• v27 — raw HID passthrough over the C surface. This widens PunktfunkHidOutput 19 → 85 bytes, the first deliberate widening this surface has made. The pre-v27 prefix layout is byte-identical and the tail is appended, so a binary built against a v26 header passes a 19-byte out-slot that a v28 core overruns. punktfunk_abi_version() equality is the guard, as it has always been: recompile, not relink.
• v28 — punktfunk_connection_host_caps2 + PUNKTFUNK_HOST_CAP2_TOUCH. Purely additive.

Second, quieter break: punktfunk_connection_next_hidout now surfaces HidOutput::HidRaw where it previously fell out as NoFrame. A puller that assumed every event was DualSense feedback must switch on kind.

Everything else versioned stands still — wire 2, driver 6, gamepad channel 3, plugin index 1, host event schema 1, gamescope pfhdr8, SDK 0.1.6, plugin-kit 0.4.4, 27 crate dirs / 39 members.

⭐ The OpenAPI caveat is gone, because the host builds on macOS again

Every release since v0.31.1 carried some version of "re-stamped, not regenerated — punktfunk-host does not build on macOS". #456 fixed the eleven compile holes, so mgmt::tests::openapi_document_is_complete_and_checked_in runs here now and the surface is verified rather than assumed. It normalises info.version on both sides, so a bump alone cannot trip it.

The spec was regenerated on main (3db1f53d, re-synced by a5593de1). Against v0.33.0 its only non-stamp diff is the disable_physical_monitors description, rewritten because the second selector it documented now runs by default. No path, operation or schema moved.

Fixed en route: a drift-gate warning nobody could see

PUNKTFUNK_STANDBY_SINK_KEEP sat in the undocumented-env baseline while #452 had documented it. check-docs-drift.sh printed baseline entries no longer undocumented on every run and still exited 0, so it was invisible in CI. Pruned here.

Generalisable: read that gate's stdout, not only its exit code.

Crediting the contributors

Two people outside the team wrote code in this cut, and the notes name both.

• Keith Porcaro (#425, 5 commits) — the whole Steam Controller 2 push beyond Linux: the shared wire tables and DEVTYPE_TRITON, the Windows pf_triton virtual-pad backend, raw HID over the client ABI (this is what moves the ABI to 27), Apple CoreBluetooth capture, and the Android frozen-timestamp IMU gate. Five pieces, four platforms.
• luxus (#439, #441, 2 commits) — the install script's intent-based defaults with its new check-install-defaults.sh gate, and a tightening of the v0.33.0 changelog to house style.

Worth flagging as a change of convention: every prior ## Thanks section credited reporters only — v0.31.3 even refers to "a contributor's diagnosis" without a name. This one names code authors first, then keeps the reporter list.

⚠️ Packager- and operator-visible

pf_gamepad.inx gains pf_triton, an eighth hardware id. An existing install carries the seven-id package, so the Windows SC2 backend does nothing until punktfunk-host.exe driver install --gamepad. This is in the notes' ## Before you update.

Gates, all run on the Mac

• cargo fmt --all --check
• cargo test -p punktfunk-core --lib --features quic — 511 passed, 0 failed (was 508 at v0.33.0)
• C ABI harness — PASS, abi_version=28, the cheap oracle for the row that moved
• the OpenAPI staleness test — PASS (new here; see above)
• check-docs-drift.sh, check-docs-links.sh
• cargo audit — 2 allowed warnings (audiopus_sys, paste), no yanked crates
• both openapi copies cmp-identical; Play notes 493/500 and unique vs all 17

⚠️ cargo audit's first run errored with couldn't check if the package is yanked: request could not be completed in the allotted timeframe. That is a registry network timeout, not a finding — but it is the check that caught yanked chacha20 at v0.33.0, so it was re-run until it completed rather than accepted as a pass.

Still owed after merge

1. Tag and push — docs/releases/README.md step 2. Tag message is drafted.
2. Wait for every platform green. Merge-sha green does not predict tag green; the tag fires runs whose publish steps never execute on a PR.
3. Diff the tag build's asset name set against v0.33.0's 80, version substring normalised, before announcing.
4. Dispatch announce with the tag.
5. Not run here: the ISO 24495 auditor on the notes (that skill never self-invokes), and the punktfunk-website / CMS pass that step 1 also asks for.

Version bump, CHANGELOG section, user notes and Play notes for **v0.34.0**. 140 commits since `v0.33.0`. ## The eight files The same bump surface as v0.33.0, plus one drift fix: | File | Why | |---|---| | `Cargo.toml` | `[workspace.package] version` → 0.34.0 | | `Cargo.lock` | `cargo update --workspace`; diff is **36/36 versions-only** | | `api/openapi.json`, `docs-site/public/openapi.json` | `info.version` stamp; the two are byte-identical | | `CHANGELOG.md` | the v0.34.0 section, 199 lines | | `docs/releases/v0.34.0.md` | the release body, seeded verbatim by CI | | `docs/releases/whatsnew/v0.34.0.txt` | Play notes, **493/500 chars**, unique | | `scripts/ci/docs-undocumented-env-baseline.txt` | prunes one stale entry, below | ## Why minor, not patch Not a renumber — what the tree gained. The quick-action ring and its editors on every shell, the on-screen virtual controller on Android and Apple touch devices, and the Steam Controller 2 passthrough leaving Linux for Apple, Android and Windows hosts. ## ⚠️ The embedder trap: the ABI moves **twice**, and one step widens a struct Every recent release has been able to say "nothing versioned moves". This one cannot. `ABI_VERSION` goes **26 → 28**: * **v27** — raw HID passthrough over the C surface. This **widens `PunktfunkHidOutput` 19 → 85 bytes**, the first deliberate widening this surface has made. The pre-v27 prefix layout is byte-identical and the tail is appended, so a binary built against a v26 header passes a 19-byte out-slot that a v28 core overruns. `punktfunk_abi_version()` equality is the guard, as it has always been: **recompile, not relink.** * **v28** — `punktfunk_connection_host_caps2` + `PUNKTFUNK_HOST_CAP2_TOUCH`. Purely additive. Second, quieter break: `punktfunk_connection_next_hidout` now *surfaces* `HidOutput::HidRaw` where it previously fell out as `NoFrame`. A puller that assumed every event was DualSense feedback must switch on `kind`. Everything else versioned stands still — wire 2, driver 6, gamepad channel 3, plugin index 1, host event schema 1, gamescope `pfhdr8`, SDK 0.1.6, plugin-kit 0.4.4, 27 crate dirs / 39 members. ## ⭐ The OpenAPI caveat is gone, because the host builds on macOS again Every release since v0.31.1 carried some version of *"re-stamped, not regenerated — `punktfunk-host` does not build on macOS"*. #456 fixed the eleven compile holes, so `mgmt::tests::openapi_document_is_complete_and_checked_in` **runs here now** and the surface is verified rather than assumed. It normalises `info.version` on both sides, so a bump alone cannot trip it. The spec was regenerated on main (`3db1f53d`, re-synced by `a5593de1`). Against `v0.33.0` its only non-stamp diff is the `disable_physical_monitors` description, rewritten because the second selector it documented now runs by default. No path, operation or schema moved. ## Fixed en route: a drift-gate warning nobody could see `PUNKTFUNK_STANDBY_SINK_KEEP` sat in the undocumented-env baseline while #452 had documented it. `check-docs-drift.sh` printed `baseline entries no longer undocumented` on every run **and still exited 0**, so it was invisible in CI. Pruned here. Generalisable: read that gate's stdout, not only its exit code. ## Crediting the contributors Two people outside the team wrote code in this cut, and the notes name both. * **Keith Porcaro** (#425, 5 commits) — the whole Steam Controller 2 push beyond Linux: the shared wire tables and `DEVTYPE_TRITON`, the Windows `pf_triton` virtual-pad backend, raw HID over the client ABI (**this is what moves the ABI to 27**), Apple CoreBluetooth capture, and the Android frozen-timestamp IMU gate. Five pieces, four platforms. * **luxus** (#439, #441, 2 commits) — the install script's intent-based defaults with its new `check-install-defaults.sh` gate, and a tightening of the v0.33.0 changelog to house style. Worth flagging as a change of convention: every prior `## Thanks` section credited **reporters** only — v0.31.3 even refers to "a contributor's diagnosis" without a name. This one names code authors first, then keeps the reporter list. ## ⚠️ Packager- and operator-visible `pf_gamepad.inx` gains **`pf_triton`, an eighth hardware id**. An existing install carries the seven-id package, so the Windows SC2 backend does nothing until `punktfunk-host.exe driver install --gamepad`. This is in the notes' `## Before you update`. ## Gates, all run on the Mac * `cargo fmt --all --check` * `cargo test -p punktfunk-core --lib --features quic` — **511 passed, 0 failed** (was 508 at v0.33.0) * C ABI harness — **PASS, `abi_version=28`**, the cheap oracle for the row that moved * the OpenAPI staleness test — **PASS** (new here; see above) * `check-docs-drift.sh`, `check-docs-links.sh` * `cargo audit` — 2 allowed warnings (`audiopus_sys`, `paste`), no yanked crates * both openapi copies `cmp`-identical; Play notes 493/500 and unique vs all 17 ⚠️ `cargo audit`'s **first** run errored with `couldn't check if the package is yanked: request could not be completed in the allotted timeframe`. That is a registry network timeout, not a finding — but it is the check that caught yanked `chacha20` at v0.33.0, so it was re-run until it completed rather than accepted as a pass. ## Still owed after merge 1. Tag and push — `docs/releases/README.md` step 2. Tag message is drafted. 2. Wait for **every** platform green. Merge-sha green does not predict tag green; the tag fires runs whose publish steps never execute on a PR. 3. Diff the tag build's asset name set against v0.33.0's 80, version substring normalised, **before** announcing. 4. Dispatch `announce` with the tag. 5. Not run here: the ISO 24495 auditor on the notes (that skill never self-invokes), and the punktfunk-website / CMS pass that step 1 also asks for.

enricobuehler added 1 commit 2026-08-30 23:47:01 +00:00

chore(release): cut 0.34.0 ... 2531207f04

140 commits since v0.33.0. The release is a control surface for everyone
who streams to a screen they hold: a six-button ring opened by a
two-finger twist, editable by using it rather than by reading a list,
plus an on-screen controller on phones and tablets.

The Steam Controller 2 passthrough leaves Linux. It now captures from
Apple and Android devices and reaches a Windows host, still presented as
itself rather than flattened into a generic pad.

This is not a drop-in for embedders. The C ABI moves 26 to 28, and v27
widens PunktfunkHidOutput from 19 to 85 bytes -- a recompile, not a
relink, with the version equality check as the guard. Every other
version number stands still: wire 2, driver 6, gamepad channel 3.
Windows packagers additionally need the pf_gamepad package rebuilt, for
its eighth hardware id.

api/openapi.json is regenerated rather than re-stamped this time, and
the surface is verified rather than assumed: punktfunk-host compiles on
macOS again, so the staleness test runs here. Every prior release
carried a caveat for exactly that reason.

Also prunes PUNKTFUNK_STANDBY_SINK_KEEP from the undocumented-env
baseline, which the drift gate has warned about since #452 documented
it while still exiting 0.

Gates on the Mac: cargo fmt --all --check, cargo test -p punktfunk-core
--lib --features quic (511 passed, 0 failed), the C ABI harness
(abi_version=28), the openapi staleness test, check-docs-drift.sh,
check-docs-links.sh, cargo audit (2 allowed warnings, no yanked
crates), both openapi copies byte-identical, and Play notes at 493/500
chars and unique.

enricobuehler added 1 commit 2026-08-30 23:59:57 +00:00

fix(client/windows): allow the host tile's eighth argument ... 7f69100d39

The Windows clippy job has failed on main since a04c8cee: giving the
avatar its OS mark added `os` to `host_tile`, taking it to eight
arguments against clippy's threshold of seven, and `-D warnings` makes
that an error. The job never runs on a Mac or a Linux runner, so the
break reached main green.

Take the allow, as the two neighbours at this limit already do. All
three call sites pass a different mix of the optional tail, so a props
struct would only move the argument list into each of them.

`edit_profile_modal` and `described_overridable` sit at exactly seven
and are untouched.

enricobuehler merged commit b7c29329e5 into main 2026-08-31 00:25:05 +00:00

Sign in to join this conversation.

Reviewers

No Reviewers

Labels

No items

No labels

Milestone

No items

No Milestone

Projects

Clear projects

No projects

Assignees

Clear assignees

FernandoMX FlipperDan HitFrostbite bluelightspecial enricobuehler (Enrico Bühler) kszlim luxus nannku neoney sonar

No Assignees

1 Participants

Notifications

Subscribe

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: unom/punktfunk#467