Skip to content

release: 3.0.0-rc3 - #180

Merged
raspbeguy merged 1 commit into
mainfrom
release/v3.0.0-rc3
Aug 13, 2026
Merged

release: 3.0.0-rc3#180
raspbeguy merged 1 commit into
mainfrom
release/v3.0.0-rc3

Conversation

@raspbeguy

Copy link
Copy Markdown
Member

Sixteen commits have landed since rc2 was tagged, one of them breaking and five of them fixing defects in the published schema, so rc2 soaked a wire surface that no longer exists. This cuts rc3 rather than promoting it.

A changelog retcon, fixed here

Every entry written since rc2 had been appended inside the [3.0.0-rc2] section, which was tagged and shipped on 2026-08-12. Each PR inserted above the rc1 heading, which put it at the end of rc2's block rather than in a new one, and the section had accumulated four separate ### Fixed headings along the way.

Everything merged after the tag now sits under [3.0.0-rc3], and [3.0.0-rc2] is back to exactly the two Security entries git show v3.0.0-rc2:CHANGELOG.md records. Headings are consolidated to one of each, in Keep a Changelog order.

Three changes had no entry at all

0ef117b, b2ad46c and 5bb4065 never touched the CHANGELOG, so the release notes were missing them. Two are user-visible and are now written up: the query-string fallbacks becoming declared spec parameters, and a revoked token's rate-limit bucket being reaped. The third only added a docs caveat saying allowed_cidrs is IPv4-only, which the IPv6 work in this same cycle removed again, so it needs no entry of its own.

That leaves 13 entries in rc3, against 16 commits: two CI-only changes and the superseded doc caveat carry none.

What is in rc3

One breaking change, refuse a request field the resource does not declare. Five published-schema defects, all found by validating live bodies against the document rather than the document against itself: TokenMetadata's unsatisfiable required set, 105 response properties forbidding the null they return, request schemas rejecting the round-trip bodies the server accepts, a conditional requiring a masked field, and a type admitting a null its own enum refused. Two write-path fixes, empty bodies and default materialisation. IPv6 allowlisting. A new conformance test and two new lint gates.

Was it verified against a real device, or only by CI? Both, throughout. Merged main was last verified on hardware with 28/28 live checks and a clean conformance sweep in both directions, 52 responses and 96 round-trip bodies, zero violations. make lint, 1230 unit tests, openapi-check, coverage and gate-selftest all pass on this commit.

Does this change the wire surface? It is the release commit for a major RC that does. The semver-relevant content is the breaking change above; everything else is additive or a document correction.

Per docs/release-process.md an RC publishes to GitHub Releases and never to the apk feed. No tag is created by this PR.

@raspbeguy
raspbeguy merged commit 469d90f into main Aug 13, 2026
6 checks passed
@raspbeguy
raspbeguy deleted the release/v3.0.0-rc3 branch August 13, 2026 05:10
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant