Add PlayStation Network as an import source

[Revisor01]

Summary

Adds PSN alongside the existing Xbox and Steam importers. Connect with an NPSSO token, pull played PS4/PS5 titles via psnawp, match them to IGDB and write cumulative playtime into Game.progress — the same shape as imports/xbox.py, which this is modelled on throughout.

On my own account: 134 played titles → 19 non-games filtered out → 115 games, all 115 matched to IGDB → 103 Game rows (12 titles are the same game under PS4 and PS5 SKUs, so their playtime is summed). Final Fantasy VII Rebirth lands at 4900 minutes, which is exactly what PSN reports.

Non-game filtering was the interesting part. PSN's category field doesn't separate apps from games — Netflix and Cyberpunk 2077 both come back as UNKNOWN. Entitlements looked promising but fail badly: God of War Ragnarök and ELDEN RING have no entitlement at all (PS Plus / disc), so they'd vanish. What does work is the store concept metadata — games carry genres, apps return an empty list. Verified against all 42 UNKNOWN titles on my account: 23 games, 19 apps, no misclassification, including Genshin Impact (F2P, correctly a game) and two VR short films (correctly not). A failed lookup keeps the title rather than dropping it, and skipped titles are surfaced as an import warning.

Sync frequency matches Xbox: daily or every two days, no polling. PSN's playtime is cumulative, so a missed run costs nothing.

Two things worth flagging before you find them yourself:

• psnawp has to be pinned to 2.0.0. 3.x needs pyrate-limiter>=4, our requests-ratelimiter==0.8.0 needs <3; 2.1/2.2 want requests-ratelimiter<0.8.0. 2.0.0 is the only version that resolves — and it pins attrs==23.1.0 exactly, which pulls attrs back from 26.1.0 project-wide. Dependents (aiohttp, jsonschema, referencing) all still satisfy their constraints and the full suite passes, but it is a three-year downgrade as a side effect of a game importer. If that is unacceptable, [tool.uv] override-dependencies = ["attrs==26.1.0"] overrides psnawp's pin — say the word and I'll add it.
• psnawp 2.0.0 sends requests with no timeout. A hanging PSN response would block a Celery worker indefinitely, so the client injects a 30s default at the request builder.

Two refactor commits precede the feature: the store-agnostic title normalisation is extracted from xbox.py into imports/title_matching.py, and the Xbox schedule helpers are generalised for console reuse — both behaviour-preserving, Xbox suite unchanged and green.

AI Assistance

I used Claude Code for this — claude-opus-5 for the code, claude-fable-5 for review. Two review rounds found real bugs (progress could be lowered by a partial IGDB failure or a zero-duration title; skipped titles were invisible; a status guard raced), all fixed here, each with a test I verified fails without the fix. The approach, the API probing and the live testing are mine.

Validation

SECRET=test-only scripts/test.sh integrations users app.tests.test_log_safety
  → Ran 1155 tests, OK
SECRET=test-only scripts/test.sh integrations.tests.imports.test_xbox
  → Ran 58 tests, OK (refactor is behaviour-preserving)
uv run --no-sync ruff check src
  → All checks passed!

Model change, so also: makemigrations --check clean, migrate --plan includes 0024_psnaccount, check_migration_hygiene --strict passes against latest.

Running on my own instance against the live PSN API — the screenshots are from that.

Contract Handoff

• Domain guide regeneration/check outcome: domain_vocabulary --check passes, no regeneration needed
• Verified OpenAPI regeneration outcome: regenerated, no diff — src/api/ is untouched
• Contract-test outcome: 57 tests pass (test_about, test_api_contracts, test_domain_vocabulary)

Human Review

• Pending human review.

Gstack QA

• Pending /gstack-qa.

Notes

The PlayStation tile logo is drawn as its own paths rather than shipping Sony's asset, matching how xbox-logo.svg is done here.

Two things I wasn't sure about and deliberately left out, both your call rather than mine:

Per-platform playtime. PSN reports a game's PS4 and PS5 editions under separate title IDs, and I sum them into one figure. Keeping them apart — per-platform hours plus a combined total — would be more useful, especially once you can play the same game on Xbox, Steam and PlayStation and want to see where the hours actually went. But that is a model change touching all three importers, not something to smuggle in with one of them. Happy to spec it out separately.

A live "currently playing" card. psnawp exposes presence, so showing what is running right now is possible. Sony sends no webhooks though, so it would need polling every ~30s — the first polling of its kind in Floppy, against an unofficial API whose own library warns about heavy use. That felt like an architecture decision rather than a feature I should just build, so I left it alone.

[Revisor01]

[ryan-winkler]

Review & Status Check

Great work on the PlayStation Network import source! The architecture cleanly matches the console import patterns established by the Xbox importer (#642), with solid rate-limiting, non-game concept filtering, and comprehensive test coverage across all 1,155 integration tests.

CI Status Note

The test failure in GitHub Actions was caused by an upstream SQLite parking test flake that has since been hardened on latest in #826. Rebasing your branch onto current latest will turn CI green.