ci: make clean Symfony bootstrap structurally checkable - #10
Open
phibkro wants to merge 12 commits into
Open
Conversation
`turbo build` was red at HEAD: config.ts reads import.meta.env for the Vite-bundled path, which plain tsc has no type for. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Vendored fork of a superseded Express/Drizzle/Postgres repo. Nothing consumed it — the SDK talks only to the Symfony server. Removed with it: the docker-compose api + postgres services, the CI postgres service container and drizzle-migrate step, the DATABASE_* passThroughEnv (api-only), the dev:api script, the pnpm-lock sync in pre-commit, and the README/CLAUDE.md entries. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Left over from the openapi-fetch client. The hand-written domain client does not reference it: tsconfig includes only src/, package.json ships only dist/, and no import or type reference points at it. Drops the oxlint ignore that existed solely for it. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The README still described the deleted openapi-fetch client (createClient(...).GET(path), createQueryApi, `turbo run generate`). Replace it with the actual hand-written domain-first Effect client: domain methods, dual promise/Effect export, SdkError hierarchy, and the Symfony adapter's date/status normalisation. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Receipt::$visualId is declared unique in the database but was generated as dechex(milliseconds), so every receipt built inside the same millisecond got the same id. Eight receipts constructed back to back produced one distinct id, which is why loading the fixtures aborted on receipt.visual_id. Appending four random bytes keeps the time-ordered hex prefix humans read in emails and receipt tables while making the value unique. No consumer parses the format -- the templates echo it and the SDK types it as a plain string. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
… data
Guard-parity items A1, A2 and A7 declared uniqueness invariants that a scan of
the production backup refutes. In each case the stated key would make valid
data unrepresentable, so the entity is corrected rather than the data:
school_capacity + department_id one school has capacity in one semester
under two departments (3 collisions -> 0)
team_membership + position_id one person holds two positions in one team
and semester (6 of 13 collisions are valid)
assistant_history + bolk a semester has two teaching blocks and an
assistant is placed once per bolk
The three constraint tests asserted the old keys, so they asserted rules the
domain does not have. Each now tests both directions: the duplicate that must
be rejected, and the case the old key would have wrongly rejected.
A2 and A7 still cannot be enforced in the database. A7 has 7 true duplicates
left; A2 needs a data-shape migration first, because 332 rows store "both
bolks" as a single 'Bolk 1, Bolk 2' row rather than two rows. The cleanup
queries, the split logic and the follow-up migration are written down in
docs/migrations-blocked-unique-constraints.md so the gap is visible instead of
silent.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…aints Guard-parity items A1, A2, A6 and A7 added four uniqueness invariants as ORM metadata only. Production is migration-driven, so none of them exist there: the test schema is built from metadata and enforced the constraints, while the database the code actually runs against did not. Only two of the four can be enforced today, so only those two are here: A6 receipt(visual_id) 0 duplicates, 0 nulls A1 school_capacity(school, semester, department) 0 duplicates preUp() re-runs the duplicate scan against the live database and aborts before any DDL if it is not clean. MySQL applies DDL non-transactionally, so a duplicate found while creating the second index would otherwise leave the database half-migrated; refusing to start is the only state that is safe to retry. The production scan is a reading of a backup, not a promise about the database this runs on. Verified against MariaDB 11.4 with the row shapes the production scan found: the guard aborts and creates nothing while a duplicate exists, ignores NULL key parts (which a unique index treats as distinct), applies both indexes once clean, rejects a real duplicate, still accepts the same school and semester in another department, and down() removes both. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…sible phpunit.xml.dist enumerated test directories one by one, so tests/App/ -- 30 files, 64 tests, written for the guard-parity batch -- belonged to no testsuite and `composer test` never discovered them. CI runs `composer test`, so reverting any of the code those tests cover broke nothing. The default suite is now declared by exclusion: every test under tests/ that a slower, more specific suite does not claim. Suite membership becomes a function of the filesystem, so a new directory cannot be orphaned by forgetting to list it. Discovery goes from 1096 to 1167 tests. PhpUnitSuiteCoverageTest guards the invariant directly, for the case where someone reverts to enumeration: it fails naming any *Test.php that no suite covers, and any declared directory that does not exist. Checked against the old config, where it fails and names all 30 orphaned files. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Catches a mapping that no longer describes anything real -- an index naming a column that does not exist, a relation that does not resolve. Runs without a database. It deliberately does not claim to be the metadata-vs-migrations gate. That gate is `doctrine:migrations:migrate` against an empty MySQL service followed by an empty `doctrine:migrations:diff`, and it cannot run today: 49 files in migrations/ import Doctrine Migrations 1.x's `Doctrine\DBAL\Migrations\ AbstractMigration`, which does not exist in the installed 3.x, so every `doctrine:migrations:*` command fatals before listing anything. The step comment records that so the check is not mistaken for more than it is. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Design spec
design-specs/0002-symfony-clean-checkout-bootstrap.mdJourney
A maintainer checks out the repository without cached PHP artifacts and runs the bounded Symfony bootstrap workflow. The workflow classifies the current PHPUnit, lint, and analysis results structurally instead of pinning transient result tuples.
Experience it
Evidence
1167/55/3/3156tuple as an acceptance constant..github/workflows/ci.yml.What is real
The bootstrap and structural result classification are real and locally verified. PHPUnit still has 2 failures, lint has 8 findings, and analysis has 1 finding; this PR does not claim a green Symfony suite, migration replay, deployment, or production acceptance.