Django 4.2 → 5.2 LTS upgrade (spike — DO NOT MERGE without Raymond + Eric review) (#1081)

[rdhyee]

DRAFT — gated on Raymond AND Eric. Deliverable of the 5.2 spike (#1081). Full results + go/no-go analysis: UPGRADE_PLAN_5.2.md (Gluejar dir).

Headline

Executed the upgrade for real against the full test suite: 3 dependency bumps + a handful of small code fixes → Django 5.2.15 runs with ZERO new test failures vs the 4.2 baseline (identical 12-item failure set, all pre-existing/environmental). Everything the paper plan feared — vendored forks, django-selectable, django-registration — works unchanged.

Changes

• Deps: django → 5.2.15; django-sass-processor 0.8.2→1.4.2 (get_storage_class removed in 5.1); django-mptt 0.14→0.18 (index_together removed in 5.1); django-contrib-comments 2.0→2.2 (5.x migration-state drift).
• USE_TZ = False pinned — Django 5.0 flipped the default; regluit + its production data are naive-datetime. Preserves current behavior exactly; TZ-aware migration is a separate future project.
• Real bug fix — Campaign.clone(): filtered Transactions on an unsaved instance (pk=None); silent no-match on ≤4.2, hard error on 5.x. save() now precedes update_left() (identical semantics).
• Test fixes: assertRegexpMatches→assertRegex (removed in Python 3.12 — was already broken on prod's Python); Django 5.0 signal receivers are 3-tuples.
• Includes the #1202 STORAGES/USE_L10N prep (cherry-picked; land that PR first).

Verification (local, Docker MySQL 8 utf8mb3 + redis 7, Python 3.12.9)

• pip install with 5.2: zero conflicts.
• manage.py check: clean (only the known ckeditor.W001).
• makemigrations --check: clean. migrate over a 4.2 schema: applies only django_comments.0004 (an index).
• Full suite 4.2 vs 5.2: identical failure sets (12 pre-existing environmental failures both sides; celery tests pass both sides).
• Django 6.0 deprecation scan: mild (URLField scheme default; utcnow() — folds into the TZ project).

Before merge

1. Land Django 5.2 prep: STORAGES dict + drop USE_L10N (#1081) #1202. 2. Raymond + Eric review. 3. Deploy to test.unglue.it with -e run_pip=true and run the smoke gate (RUNBOOK_deploy_test_then_prod.md). 4. Soak, then release per runbook.

[rdhyee]

Codex review: LGTM (no high-confidence blockers). Notable points, all captured:

• USE_TZ=False pin confirmed correct and necessary.
• Campaign.clone() reorder: final persisted state equivalent; only a transient pre-update_left() row visible to concurrent readers/DB triggers (none in-repo). Non-blocking.
• Dep bumps: recommend a collectstatic/SCSS smoke on staging (sass-processor 1.x changed --use-storage behavior; repo doesn't use the removed names). comments 2.2 migrations touch real comment data (an index) — fine but real.
• Real catch: the rendered prod prod.py still sets DEFAULT_FILE_STORAGE → mutually-exclusive error on 4.2 the moment Django 5.2 prep: STORAGES dict + drop USE_L10N (#1081) #1202 deploys (reproduced locally). Fix filed: Remove DEFAULT_FILE_STORAGE from prod.py template — deploy-blocking conflict with regluit#1202 (STORAGES) EbookFoundation/regluit-provisioning#61 — must deploy in the same window as Django 5.2 prep: STORAGES dict + drop USE_L10N (#1081) #1202.