v0.2.2

Requires the migration and the core everybody actually runs.

This module pinned migration v0.1.24 while the published package was v0.2.0, and api v0.6.0 against v0.6.3.

MVS lifts both in a real application, so no consumer noticed. The cost is ours: GOWORK=off go test ./... is the pass this repo treats as the real signal, precisely because a workspace masks a wrong requirement — and it was validating this module against a migration a minor behind and a core three patches behind.

migration v0.2.0 is breaking in principle, so the bump was checked rather than assumed. Two kinds of generated identifier gain the table name they lacked: a multi-column Index and a single-column UniqueKey. The DSL's output was executed under both versions and diffed, on both dialects — on Postgres only the multi-column index name moves, because UniqueKey renders as an anonymous UNIQUE (col) there and the server names the constraint itself; on MySQL both move. Nothing in this module builds schema with the DSL, so nothing here changes.

api v0.6.1 serves /build/ as well as /public/build/, v0.6.2 adds LOG_QUIET so a worker or a scheduler can ask for its own voice back, and v0.6.3 is this same migration pin one level down.

v0.2.1

Requires the core everybody actually runs.

This module pinned api v0.2.0 while the published core was v0.6.0 — four minors of drift, and the same for auth at v0.3.1 against v0.6.0.

A real application never noticed, because MVS lifts api to the highest requirement in the graph. The cost was ours: GOWORK=off go test ./... is the pass this repo treats as the real signal, precisely because a workspace masks a wrong requirement — and it was validating this module against a core three or four minors old. It also means a fresh go get of this module brought api v0.2.0 with it unless the consumer pinned the core themselves.

api v0.6.0 is a breaking change — App.FileSystem() and Context.disk moved onto interfaces declared in api/app, which gets the AWS and Google Cloud SDKs out of the core — and it is a pin change here: nothing outside api and media named the concrete types. Verified by building and testing every module standalone after the bump.

Where miniredis appears it is unified at v2.38.0, which was skewed against tasker.

This repository also has a CI workflow for the first time. The three that existed were written as though the monorepo were one repository, filtering on paths that do not exist in a per-module checkout, so none had ever run.

v0.2.0 — api v0.2.0

Requires api v0.2.0, up from v0.1.3. No source change.

api v0.2.0 replaced the framework's two form binders with one and removed httpin. This module's go.mod still named httpin as an indirect requirement, inherited from the api version it was last tidied against. Nothing here imported the package, but a dependency's go.mod is read even when nothing in it is built — so every consumer still resolved httpin in its module graph, and go list -m all still answered that the framework depends on it.

The go directive also moves from 1.25.0 to 1.27, matching every other module — an uncommitted edit this repo has carried since September.

A minor rather than a patch bump because the dependency floor moves: anything depending on this module now resolves api v0.2.0, which is itself a breaking release.

Verified with GOWORK=off, where api resolves from the proxy rather than from ./api in the workspace — the masking that lets a published module ship broken.

v0.1.0

Full Changelog: https://github.com/lemmego/templ/commits/v0.1.0