Consolidate on go.yaml.in/yaml/v3

[davidzhao]

Why

protocol compiles two copies of the same yaml library:

module	pulled in by
gopkg.in/yaml.v3	auth/provider.go, utils/configutil/observer.go, livekit/types.go
go.yaml.in/yaml/v3	transitively, via buf.build/go/protoyaml

go.yaml.in/yaml is the project's new canonical home for the same codebase, so this switches our own imports over and leaves one copy in the build.

The part that needed care

RoomConfiguration, RoomEgress and RoomAgent implement a custom UnmarshalYAML(value *yaml.Node). That's an interface, and a decoder only recognizes it when the Node parameter is its own type.

Just re-pointing the import would have been a silent, runtime-only break. Every one of these still decodes with gopkg.in/yaml.v3:

livekit/pkg/config/config.go:231
  RoomConfigurations map[string]*livekit.RoomConfiguration `yaml:"room_configurations,omitempty"`

…plus egress, ingress, sip and cloud. Their decoders would stop seeing an Unmarshaler and fall back to reflective struct decoding of a protobuf message — no compile error anywhere, room_configurations fields such as min_playout_delay just quietly come back zero.

What this does instead

The three methods now use the older func-based unmarshal signature, which both packages honor because it names no yaml types at all:

// gopkg.in/yaml.v3@v3.0.1/yaml.go:40  and  go.yaml.in/yaml/v3@v3.0.4/yaml.go:39
type obsoleteUnmarshaler interface {
	UnmarshalYAML(unmarshal func(interface{}) error) error
}

So downstream repos keep working unchanged, whichever yaml package they import — no coordinated migration needed. MarshalYAML() (interface{}, error) was already yaml-package-neutral and is untouched.

Verification

• TestUnmarshallRoomConfigurationBothYAMLPackages decodes the same document with both packages and asserts a fully populated message.
• Confirmed the test has teeth: restoring the *yaml.Node signature makes the gopkg.in/yaml.v3 subtest fail with expected: 0x2a, actual: 0x0 on min_playout_delay — precisely the silent-misparse mode above.
• Full go test ./... passes.
• End-to-end against the real consumer: built livekit-server's pkg/config against this branch and parsed a config containing room.room_configurations through config.NewConfig. All fields (min_playout_delay: 42, nested egress.room.room_name, agents[]) parse correctly while the server still imports gopkg.in/yaml.v3.
• Confirmed the payoff downstream: with this branch replace'd into server-sdk-go, its compiled dependency set drops to just go.yaml.in/yaml/v3 + protoyaml; gopkg.in/yaml.v3 is gone.

Note

gopkg.in/yaml.v3 stays in go.mod — the regression test imports it deliberately, since testing the both-decoders property requires both decoders. It is test-only and no longer in the compiled dependency set, so consumers don't build it.

🤖 Generated with Claude Code

[changeset-bot]

🦋 Changeset detected

Latest commit: a192275

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 0 packages

When changesets are added to this PR, you'll see the packages that this PR includes changesets for and the associated semver types

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR