QA

QA and tests

indexarr has a Go test suite for every package and an end-to-end smoke test that starts the real binary on a temporary database. The goal is to cover both the correctness of the internal functions and the HTTP behavior observable from Sonarr/Radarr and from the browser.

Commands

make test   # go test ./...
make vet    # go vet ./...
make qa     # vet + test + build + HTTP smoke test on a temporary instance
make e2e    # headless browser, field by field, on a temporary DB and port

make qa runs scripts/qa.sh: it builds the binary, starts it on 127.0.0.1 with a random port and a database in a temporary directory, checks the main pages, the mobile style sheet, the PWA manifest, the service worker and the Torznab t=caps response. At the end it stops the process and deletes the temporary files.

To use a fixed port:

INDEXARR_QA_PORT=18080 make qa

make e2e runs scripts/ui-e2e.py: it builds the binary, opens a temporary database on a random port, starts headless Chrome with the Chrome DevTools Protocol and a small local service that imitates FlareSolverr, qBittorrent and Sonarr. It changes every field of the forms (Settings, indexer edit, catalog, App, download clients, definitions, API keys, first start), saves, reloads and reads the value back both from the page and from the temporary database. It covers closed <details> too, the «remove value» checkbox, secret fields left empty, category/URL drop-downs, data-confirm, double click and confirm next to the button. Without Chrome it prints SKIP and exits 0; with UI_E2E_REQUIRE_CHROME=1 the absence of the browser is an error. make qa runs it when it finds Chrome.

Verification levels

1. Static — go vet ./... and compilation of all packages.
2. Unit/integration — go test ./... on the backup, config, downloadclient, feed, flaresolverr, ingest, logbuffer, notify, store, torznabapi, web packages.
3. HTTP integration — the internal/web tests use httptest and exercise handlers, CSRF, cookies, redirects and template rendering.
4. End-to-end smoke — scripts/qa.sh uses the compiled binary, so it also verifies the embedded assets and the real server startup.

What is covered

• Feeds and sources: RSS/Atom parsing, Torznab, MirCrew, HTML listing (ExtTo, TorrentGalaxy, CorsaroBlu), feed discovery, sanitization, redaction of secrets in logs and category mapping.
• Ingestion: de-duplication, release updates, backoff, auto-disable, scheduler, manual trigger and error handling.
• Store: schema and migrations, source CRUD, API keys, searches and filters, FTS5 index (triggers, rebuild on existing databases, fallback for short terms and query plan), blocklist, saved searches, backup, vacuum, integrity, de-duplication, maintenance actions and database report.
• Torznab: caps, search, tvsearch, movie, get, pagination, filters, authentication and rate limit.
• Web UI: dashboard, releases, detail, sources, quick add, indexer JSON export/import, settings, logs, maintenance, CSRF, optional Basic authentication, PWA and static assets.
• Logs: filter by level and by origin (indexarr, indexer, RSS, notifications), problem count, source/kind/host context on source errors, and filtered download log.
• Maintenance: integrity and foreign key check, analyze, WAL checkpoint, vacuum, reindex, cleanup by age/source, releases without links, hidden releases, run history, API counters and source counter recalculation.

Isolation

The tests do not use the network and do not touch the production database:

• every test opens a database in t.TempDir();
• remote sources are simulated with httptest.Server or fixtures;
• the smoke test uses an mktemp directory and terminates the process in the trap;
• no test requires external services or Docker; only make e2e uses a headless browser, and it skips if it is not installed.

CI

CI is local: make ci (scripts/ci-local.sh) requires a clean tree, writes the log to ci-logs/<commit>.log and ends with CI-LOCAL <commit> PASS or CI-LOCAL <commit> FAIL <passaggio>. It runs in order gofmt and go vet, definition validation, make qa, go test, go test -race, the build for all platforms, the test build for Windows and macOS and the browser test; the resource budget check (binary size and idle memory, error above 256 MiB) runs in the background during the tests. Every step prints its duration. All temporary files (test databases, go test binaries, builds for the other platforms, the budget server, the browser profile) are in a RAM folder under /dev/shm (CI_TMPFS to choose another one), deleted at the end even if a step fails; the Go build cache stays on disk. .github/workflows/qa.yml remains only as a manual start (workflow_dispatch) and does not run by itself.

Manual checklist for the UI

The automated tests verify that the mobile CSS is served, but the visual rendering must be checked by hand when the templates are changed:

• 390 px width: scrollable navigation, no horizontal overflow;
• tables turned into cards with readable labels;
• selection checkboxes and “Select all” reachable;
• 16 px inputs and buttons with an adequate touch area;
• pages /, /releases, /sources, /runs, /settings, /maintenance, /logs and release detail;
• light and dark theme;
• PWA installation and offline loading of the static assets.

Adding a test

• For database logic: internal/store/*_test.go, using newTestStore(t).
• For a handler or a page: internal/web/*_test.go, using newTestEnv(t), csrfFrom, postForm and get.
• For a parser or a client: local fixtures and httptest.Server.

Every new maintenance utility must have at least one test in the store and one in the web handler that verifies redirects and the effect on the database.

End-to-end simulation

Besides the package tests, sim/ contains a deterministic simulation with fake services (Torznab/Newznab, Sonarr, SABnzbd, NZBGet, Deluge, rTorrent and an HTML tracker) and assertions with a non-zero exit code. It runs locally with ./sim/run.sh or in Docker with ./sim/run-docker.sh, on the isolated ports 18484 and 19080. Details, assertions and reset in sim/README.md.