Realistic test data: factory_boy fixtures + curated prod media subset

[jonfroehlich]

Context

Spinoff from #1268, which was originally bundling REST API + test-data sync. The actual driver was getting realistic content into localhost so UI/template changes can be reviewed against prod-like data. This issue tracks that work as a standalone effort with no API dependency.

Background

Current test infrastructure (landed in #1267) uses DatabaseTestCase + small make_person / make_publication / make_news_item helpers in website/tests.py. These work for unit/integration tests but:

• Don't compose well for the M2M / SortedManyToManyField / ProjectRole-through graph.
• Don't include real media files, so image_cropping / easy_thumbnails / PDF preview code paths only ever see empty file fields.
• Don't produce a coherent dev environment for visual review.

Recommendation — split into two layers

Layer 1: factory_boy + Faker (Django standard)

• Add factory_boy and Faker to requirements.txt. De-facto Django standard for ~12 years; preferred over model_bakery here because explicit factories are easier to debug with our complex relationship graph.

• Create website/tests/factories.py with PersonFactory, PublicationFactory, ProjectFactory, ProjectRoleFactory, TalkFactory, PosterFactory, VideoFactory, NewsItemFactory, AwardFactory. Pattern:

  class PublicationFactory(DjangoModelFactory):
      class Meta:
          model = Publication
      title = factory.Faker("sentence", nb_words=8)
      date = factory.Faker("date_between", start_date="-3y", end_date="today")
      pdf_file = factory.django.FileField(from_path=SEED_PDF_PATH)
  
      @factory.post_generation
      def authors(self, create, extracted, **kwargs):
          if create:
              self.authors.set(extracted or PersonFactory.create_batch(3))

• Refactor existing make_person / make_publication / make_news_item helpers in website/tests.py to thin wrappers that delegate to the new factories (keeps existing tests green).

• Add website/tests/seed_media/ with ~5 hand-curated representative files (one real PDF, a few JPGs at different aspect ratios, a project logo). Factory FileFields point at these so every test exercises real thumbnail / PDF / image-cropping code paths.

Layer 2: prod media subset + seed_dev_data command

• Add python manage.py seed_dev_data that uses the factories to build a coherent graph (~5 projects, ~15 people, ~20 pubs with cross-links, ~5 talks/posters, news items, awards). Idempotent. Run after docker-compose up for a realistic-shaped site with zero prod dependency.
• For prod-realistic media (when you actually want to see what real content looks like): add scripts/pull_prod_subset.sh that rsyncs a curated subset from recycle.

Prod media size analysis (2026-06-14)

Total media tree at /cse/web/research/makelab/www/media on recycle: 56 GB / 13,130 files.

Folder	Size
talks/	50 GB
publications/	1.7 GB
banner/	1.3 GB
person/	837 MB
posters/	719 MB
projects/	611 MB
news/	738 MB
uploads/	323 MB
others	~120 MB

Whole-tree rsync is impractical (talks alone is 50 GB, mostly slide PDFs/videos). But person/ + projects/ + publications/ + posters/ ≈ 3.9 GB is the visual core of the site and is realistic to mirror.

Optional refinement: pair the rsync with a dumpdata JSON of the matching DB rows so the local DB references real filenames. Gitignore both outputs.

Suggested first PR slice

1. Add factory_boy + Faker to requirements.txt.
2. Create website/tests/factories.py with core factories.
3. Add website/tests/seed_media/ with curated files.
4. Migrate existing make_* helpers to delegate to factories.

Subsequent PRs: seed_dev_data management command, then pull_prod_subset.sh.

Out of scope (parked in #1268)

REST API for external consumers. Conversation concluded the only known consumer is Jon's academic page ("Recent Pubs" list), which can be served by a single JsonResponse view rather than a full DRF API.

[jonfroehlich]

Status (2026-06-17): the suggested first PR slice is up as #1319 — factory_boy/Faker added, website/tests/factories.py with the core factories, and the DatabaseTestCase.make_* helpers rewritten to delegate to them (suite stays green, 201→211). Two deliberate choices documented there: filesystem-light factories (tiny in-memory files; curated seed_media/ deferred) and no auto-authors (artifacts authorless unless authors=[...] passed, to preserve ~200 existing call sites).

Remaining for this issue (each its own PR):

• seed_dev_data management command building a coherent graph (~5 projects / ~15 people / ~20 pubs / talks / news / awards), idempotent.
• Curated website/tests/seed_media/ (~5 real files) so seed/visual-review paths exercise real thumbnail/crop/PDF code.
• scripts/pull_prod_subset.sh (rsync the ~3.9 GB visual core: person/projects/publications/posters) — waits on a DB snapshot from CSE IT.