studio-engine 0.139.0 → 0.140.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: a9937bbeb00f70902af7404d9ded16f10216e7123dba4d929eef0d04fe8c1f80
-  data.tar.gz: ff26fdac225b730cc611af898fe0fc786ee968545a5ee674f89131cfe183f035
+  metadata.gz: 607b184e22c3660009dfc7e342922ca3bdae6b13a7ca7ed32dc618957dfc819a
+  data.tar.gz: 26a6741ca01d65b37d6f16a663f35c6d253f87254af05fd41836d23e2a5ce48b
 SHA512:
-  metadata.gz: d3721562277466c4af36db5835ad05ad5fba414d4a091c9a447787f95d5d8bbe8ae053b5ff0a0563422a732169130aedd77e2bf5f721e6026b099e99172b7e6f
-  data.tar.gz: 2b2405866b62361954dc69c30d66b5dbcd56536ef40b9e26d5b64094d3ed7a07bfb36cda26483ae872cc370f95a1f7f69a7a79dd4c24c4744c1d2ce1a59a8504
+  metadata.gz: 1f6a8450e4862aea8e8f80c214353442eafcf0c56bbcd0d78848dd291fab92ebc95ed41b836b01ba11a6601a66a652cf06153ef97f7e7c4f31910603457572d1
+  data.tar.gz: 24cdd966d0cad48ba9efb03fe6d54c2ec801051c7ccf23f104411fe3a4cf4eac7ea7155b453fb308ba5625be4bd30d7945c83bc2d822026ae4f6b3de194b2f76

data/CHANGELOG.md

@@ -4,6 +4,13 @@ The format is [Keep a Changelog](https://keepachangelog.com/en/1.1.0/). This pro
 
 ## Unreleased
 
+## 0.140.0 — 2026-11-09
+
+### Fixed
+
+- **`AddUsersRoleKnownCheck` no longer sets `users.role` NOT NULL, and writes its CHECK in the stable form.** The 0.138.0 migration's NOT NULL changed `t.string "role", default: "viewer"` in six consumers' schema dumps, which the release's schema-dump guard refuses, and an old dyno inserting a user without a role would fail during a rollback. It still fills NULL roles with `viewer` and adds CHECK `users_role_known`, now spelled `(role)::text = ANY (ARRAY['viewer'::character varying::text, ...])` on Postgres, so a dump after `db:migrate` equals the dump after `db:schema:load`. The migration was edited in place: 0.138.0 published it, but no app had installed it. Install it with `bin/rails studio_engine:install:migrations`; NOT NULL comes in a later migration.
+- **`RequireSlugsOnEngineTables` no longer makes `slug` NOT NULL.** 0.138.0's migration set `slug` NOT NULL on `error_logs`, `theme_settings`, `studio_geo_settings` and `studio_site_identities`. Heroku runs the migration while old dynos still serve, and 0.137.0's `ErrorLog.capture!` inserts a NULL slug and sets it after, so every error write raised until the restart and again after a `heroku rollback`. The migration (edited in place; no host had installed 0.138.0) still backfills NULL and blank slugs, de-dupes, and adds the unique index, which allows many NULLs. Its `down` now changes nothing. A host adopts in one release: `bin/rails studio_engine:install:migrations && bin/rails db:migrate`, then commit `db/schema.rb`; the two-release note under 0.138.0 no longer applies. NOT NULL comes in a later engine migration, in the release after every host runs the Sluggable `ErrorLog`.
+
 ## 0.139.0 — 2026-11-08
 
 ### Added

data/app/models/error_log.rb

@@ -1,8 +1,9 @@
 class ErrorLog < ApplicationRecord
-  # `slug` is NOT NULL and unique (RequireSlugsOnEngineTables), so every insert
-  # carries one, including a host's bare `ErrorLog.create!`. Sluggable writes a
-  # unique placeholder before the insert and settles it to `error-log-<id>` right
-  # after, inside the create, once the id exists.
+  # `slug` is unique (RequireSlugsOnEngineTables) and becomes NOT NULL a release
+  # later, so every insert carries one, including a host's bare
+  # `ErrorLog.create!`. Sluggable writes a unique placeholder before the insert
+  # and settles it to `error-log-<id>` right after, inside the create, once the
+  # id exists.
   include Sluggable
 
   belongs_to :target, polymorphic: true, optional: true

data/db/migrate/20261006120001_ensure_error_logs_table.rb

@@ -15,8 +15,9 @@
 #
 # On every consumer that existed when this shipped, the second branch adds
 # nothing. Studio::HostSchema checks the same columns at boot. The slug
-# standard (NOT NULL, unique) is the later RequireSlugsOnEngineTables
-# (20261108160000), which tightens the column on every host.
+# standard starts with the later RequireSlugsOnEngineTables (20261108160000),
+# which backfills and uniquely indexes the column on every host; NOT NULL
+# follows a release later, once every host runs the Sluggable ErrorLog.
 class EnsureErrorLogsTable < ActiveRecord::Migration[7.2]
   def up
     unless table_exists?(:error_logs)

data/db/migrate/20261006120002_ensure_theme_settings_table.rb

@@ -6,8 +6,8 @@
 # hand-copied migration omitted it, so a theme save raises on any host built
 # from that doc (cyvasse, mcritchie-industries and moms-app when this shipped).
 # The slug is "theme-<app_name>", derived on save. Rows this left with a NULL
-# slug are filled, and the column made NOT NULL and unique, by the later
-# RequireSlugsOnEngineTables (20261108160000).
+# slug are filled, and the column made unique, by the later
+# RequireSlugsOnEngineTables (20261108160000); NOT NULL comes a release after.
 #
 # A baseline, not a rebuild:
 #

data/db/migrate/20261108160000_require_slugs_on_engine_tables.rb

@@ -1,5 +1,5 @@
-# The engine's own settings and log tables take the hub's slug standard:
-# `slug` NOT NULL with a unique index, on error_logs, theme_settings,
+# The engine's own settings and log tables take the first half of the hub's slug
+# standard: every slug filled and a unique index on it, on error_logs, theme_settings,
 # studio_geo_settings and studio_site_identities (database audit 2026-11-07,
 # finding 6). Every host inherits the tables, so one migration reaches them all.
 #
@@ -15,13 +15,21 @@
 #      error_logs) may hold a slug twice; every copy after the lowest id takes
 #      `-<id>` after it. No table here has slug children, so nothing else
 #      references the old value.
-#   3. NOT NULL, then the unique index on `slug` unless one is there. A
-#      non-unique slug index is replaced by the unique one.
+#   3. The unique index on `slug` unless one is there. A non-unique slug index
+#      is replaced by the unique one.
+#
+# NOT NULL is deliberately NOT here. The release sweep installs engine
+# migrations in the same release as the gem bump, and Heroku runs them while the
+# old dynos still serve: the shipped ErrorLog.capture! inserts a NULL slug and
+# sets it after, so a NOT NULL would make every error write raise until the
+# restart, and again after any `heroku rollback`. The unique index allows many
+# NULLs, so old code keeps working. NOT NULL lands in a later engine migration,
+# in the release after every host runs the Sluggable ErrorLog this release ships.
 #
 # Production on 2026-11-08 (every app with a database, read-only counts): no
 # NULL or blank slug, no duplicate, so steps 1 and 2 change no row there. The
-# largest table is the hub's error_logs at 606 rows, so the NOT NULL scan and
-# the index build are short. On Postgres each ALTER waits at most LOCK_TIMEOUT
+# largest table is the hub's error_logs at 606 rows, so the index build is
+# short. On Postgres each ALTER waits at most LOCK_TIMEOUT
 # for its lock; a migrate that cannot get it fails whole, changes nothing, and
 # runs again on the next release.
 class RequireSlugsOnEngineTables < ActiveRecord::Migration[7.2]
@@ -43,21 +51,14 @@ class RequireSlugsOnEngineTables < ActiveRecord::Migration[7.2]
 
       backfill(table)
       dedupe(table)
-      change_column_null table, :slug, false
       unique_slug_index(table)
     end
   end
 
-  # Only the NOT NULL is undone. The unique index stays: most hosts had it before
-  # this migration, and a down cannot tell which index it added. Backfilled slugs
+  # Nothing to undo. The unique index stays: most hosts had it before this
+  # migration, and a down cannot tell which index it added. Backfilled slugs
   # stay too; a slug is a valid value either way.
-  def down
-    TABLES.each do |table|
-      next unless table_exists?(table) && column_exists?(table, :slug)
-
-      change_column_null table, :slug, true
-    end
-  end
+  def down; end
 
   private
 

data/db/migrate/20261108170000_add_users_role_known_check.rb

@@ -16,7 +16,14 @@
 #     nothing written. A stray role is a product call (Turf's "user" meant
 #     viewer); this migration does not guess it.
 #
-# Then it fills a NULL role with "viewer", sets NOT NULL, and adds the CHECK.
+# Then it fills a NULL role with "viewer" and adds the CHECK.
+#
+# It sets NO NOT NULL. A dyno still on the old code during a deploy or a
+# rollback may insert a user without a role; the column's default ("viewer")
+# fills it, and a NULL passes a CHECK, so that write keeps working. NOT NULL is
+# a later migration, once every host runs Studio::Account (the same reasoning as
+# the engine-tables slug migration, engine-slug-not-null-later). Edited after
+# 0.138.0 published it: no app had installed it yet.
 #
 # The list is SPELLED OUT, not read from Studio::Roles: a migration is a record
 # that must still run years from now, and a live constant may move. Widening the
@@ -40,15 +47,13 @@ class AddUsersRoleKnownCheck < ActiveRecord::Migration[7.2]
     end
 
     execute "UPDATE users SET role = 'viewer' WHERE role IS NULL"
-    change_column_null :users, :role, false
-    add_check_constraint :users, "role IN (#{allowed})", name: NAME
+    add_check_constraint :users, expression, name: NAME
   end
 
   # Removes the CHECK only when it is this migration's: one that names every
   # engine role. A host's own users_role_known, which `up` left in place, is
-  # narrower (the hub's and Turf's name viewer and admin only) and stays. NOT
-  # NULL stays too: `up` cannot record whether the host had it already, and a
-  # NULL role is the one value no host wants back.
+  # narrower (the hub's and Turf's name viewer and admin only) and stays. The
+  # NULL roles `up` filled stay "viewer".
   def down
     return unless string_role?
 
@@ -60,6 +65,21 @@ class AddUsersRoleKnownCheck < ActiveRecord::Migration[7.2]
 
   private
 
+  # On Postgres, the spelling Postgres itself settles on once a schema.rb is
+  # loaded (`'x'::character varying::text`), so a host's schema dump after this
+  # migrate equals the dump after a load, byte for byte. An `IN (...)` list
+  # dumps as `ARRAY[...]::text[]` first and changes on the next load. Anywhere
+  # else (the engine's SQLite test host), the plain IN list. The same rule as
+  # 20261108160001_add_engine_state_checks.rb.
+  def expression
+    return "role IN (#{ROLES.map { |role| connection.quote(role) }.join(', ')})" unless postgres?
+
+    array = ROLES.map { |role| "#{connection.quote(role)}::character varying::text" }.join(", ")
+    "(role)::text = ANY (ARRAY[#{array}])"
+  end
+
+  def postgres? = connection.adapter_name.downcase.include?("postgres")
+
   def string_role?
     return false unless table_exists?(:users)
 

data/lib/studio/version.rb

@@ -1,3 +1,3 @@
 module Studio
-  VERSION = "0.139.0"
+  VERSION = "0.140.0"
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: studio-engine
 version: !ruby/object:Gem::Version
-  version: 0.139.0
+  version: 0.140.0
 platform: ruby
 authors:
 - Alex McRitchie