Switch from cask to brews

Update releaser flow

v0.1.10 - update releaser flow

v0.19 - adds homebrew tap

Fix optional fields with database defaults.

v0.1.7

Bug Fixes

• Fix M2M link/unlink/list queries on Postgres. to match Bob's RawQuery.
• Fix validation on optional enum fields.** Columns with both a CHECK constraint and a DEFAULT value (e.g. status TEXT DEFAULT 'draft' CHECK (status IN ('draft', 'published'))) generated a oneof= validation
  tag without omitempty.

Internal

• Remove gin framework scaffolding from router and config.
• Simplify config internals. Replace cached map lookups with direct slice scans.
• Remove unused ForeignKey.OnDelete and ForeignKey.OnUpdate fields.
• Add router generator tests (11 tests covering CRUD, FK nested routes, M2M routes, operation disabling, endpoint overrides).
• Add IR tests for IsReadOnly, ValidationTag, UpdateValidationTag, IsFilterable, and SupportsRangeOps.

What's New

Many-to-Many Relationships

Junction tables are now detected automatically and generate link/unlink endpoints on both sides of the relationship.

Given a schema:

CREATE TABLE post_tags (
  post_id UUID NOT NULL REFERENCES posts(id) ON DELETE CASCADE,
  tag_id  UUID NOT NULL REFERENCES tags(id)  ON DELETE CASCADE,
  PRIMARY KEY (post_id, tag_id)
);

kiln generates:

POST   /api/v1/posts/{id}/tags           link a tag to a post
DELETE /api/v1/posts/{id}/tags/{tagId}   unlink a tag from a post
GET    /api/v1/posts/{id}/tags           list tags linked to a post

And the reverse on tags. Link operations are idempotent (ON CONFLICT DO NOTHING). Junction tables with extra columns (e.g. created_at) are detected and stored for future use.

Disable per table with disable: [link, unlink] in overrides.

Filter and Sort Validation

Invalid filter values now return 400 errors instead of being silently ignored:

GET /api/v1/users?created_at=garbage
→ 400 {"error": "invalid value for created_at: expected RFC3339 format"}

GET /api/v1/users?sort=nonexistent
→ 400 {"error": "invalid sort field: nonexistent"}

kiln doctor Command

New diagnostic command that checks project health without modifying files:

kiln doctor

  Config
    ✓ kiln.yaml loaded
    ✓ driver: postgres
    ✓ output: ./generated

  Schema
    ✓ 4 tables found (users, posts, comments, tags)
    ✓ junction: post_tags (posts ↔ tags)

  Generated Files
    ✓ generated/models/users.go
    ⚠ generated/store/users.go (user-modified)
    ✗ generated/models/old_table.go (stale - table not in schema)

  Overrides
    ✓ users: endpoint=members, hidden_fields=[password_hash]

  Project
    ✓ go.mod found

  No issues found

Store Accepts bob.Executor

Store constructors now accept bob.Executor instead of bob.DB. Both bob.DB and bob.Tx satisfy this interface, so you can wrap store calls in transactions without modifying generated code:

tx, _ := sqlDB.BeginTx(ctx, nil)
bobTx := bob.NewTx(tx)

userStore := store.NewUserStore(bobTx)
postStore := store.NewPostStore(bobTx)

user, _ := userStore.Create(ctx, userReq)
postReq.UserID = user.ID
postStore.Create(ctx, postReq)

tx.Commit()

Pagination Fix

Count query now runs before the data fetch in all list methods (List, ListByParent, ListLinked). This ensures total is always >= the number of returned rows, which is the safe direction for pagination UIs.

Bob Plugin Documentation

Full guide for using kiln as a bob plugin: README section, dedicated docs page at docs/guides/bob-plugin.md, and a working example at examples/bob-plugin/ with gen/main.go.

Breaking Changes

IsReadOnly for Primary Key Columns

Primary key columns are now only auto-readonly if they have a database default (e.g. gen_random_uuid()). PK columns without defaults are writable. This affects composite PK tables if you add support for them in the future, but has no effect on single-PK tables with auto-generated IDs (the common case).

Store Constructor Signature

Generated store constructors changed from NewUserStore(db bob.DB) to NewUserStore(db bob.Executor). Existing code compiles without changes since bob.DB satisfies bob.Executor.

Internal Changes

• IR refactored from PrimaryKey *Column to PrimaryKeys []*Column
• Junction table detection moved from adapter to a classification pass after FK wiring
• Removed dead code: plugin/adapter.go, TestsConfig, unused IR methods
• Adapter test for composite PK tables updated (now included in schema, not skipped)

New Documentation

• Many-to-Many - junction table detection and M2M endpoints
• API Evolution - versioning, deprecation, breaking changes
• Escape Hatches - transactions, custom handlers, mixing generated and hand-written code
• Bob Plugin Mode - using kiln as a bob plugin

Fix unused import bug in handler