Flip WhatsApp URNs to the new scheme model (management command, BSUID contacts only)

[norkans7]

What

Adds a flip_whatsapp_urns management command (not a data migration) that flips existing contact URNs to the new scheme model, only for contacts that have a bsuid URN:

• whatsapp:<e164 digits> → tel:+<e164 digits> (the phone identity)
• bsuid:<CC.alphanumeric> → whatsapp:<CC.alphanumeric> (the business-scoped user id)

Scoping to contacts with a bsuid means a contact reachable only by a plain whatsapp: phone URN (no bsuid) is left untouched — we only migrate contacts that actually have a business-scoped id to take over the whatsapp: slot.

Dropping redundant whatsapp URNs

When a contact's all-digit whatsapp:<digits> URN would collide with a tel:+<digits> URN already on the same contact, that whatsapp URN is redundant (the BSUID is taking over the whatsapp: slot). Rather than leaving it behind, the command moves its Msg / Call / ChannelEvent references onto the surviving tel URN and deletes it. Collisions with a tel: URN on a different contact are left untouched.

Batching, concurrency & re-runnability

• Runs in id-ordered batches (cursor by URN id, so skipped collisions don't stall the loop).
• The collision lookup is scoped by org_id and identity; if a concurrent write inserts a colliding identity between the check and the update, that batch is logged and skipped (the command is safe to re-run and retries it).
• Bumps modified_on on flipped contacts so the search indexer picks them up.
• Safe to re-run: only all-digit whatsapp paths are flipped, and once a contact's bsuid is gone it's no longer processed.

URN model changes

• URN.validate / URN.normalize: the whatsapp scheme now holds either a phone number (legacy) or a business-scoped id (CC.alphanumeric), matching gocommon (Accept business-scoped user IDs as WhatsApp URNs gocommon#218). The BSUID pattern is shared via URN.BSUID_PATH_REGEX.
• URN.format: business-scoped whatsapp ids are shown as-is; legacy all-digit whatsapp paths still get friendly phone formatting.
• ContactURN gains CONTENT_RELATIONS + has_content() / reassign_content_to(), centralising "what references a URN" (previously hardcoded in Contact._full_release, now shared) so the command's reassign-then-delete uses the model.

Why a command instead of a migration

Lets ops run and observe the flip deliberately (and re-run it) rather than having it fire implicitly during deploy.

Tests

temba/contacts/management/commands/tests.py — both flip directions; the "no bsuid → not flipped" scoping; whatsapp→tel and bsuid→whatsapp collision skips; dropping a redundant whatsapp URN and reassigning its message reference onto the tel URN; a multi-batch id-cursor test (patched BATCH_SIZE) that would loop forever on a cursor regression; and a re-apply idempotency check. Plus updated test_urn.py for the new whatsapp validate/normalize/display semantics.

🤖 Generated with Claude Code

[claude]

Claude finished @norkans7's task in 4m 20s —— View job

---

PR Review

• Load PR context
• Read changed files
• Analyze migration correctness & idempotency
• Analyze URN model changes
• Check test coverage
• Post review

Summary

Overall this looks solid — the two-pass flip, digits_only=True guard on the whatsapp→tel pass, and per-batch collision skip are correct and make the migration re-runnable. The id-cursor batching correctly advances past skipped collisions so the loop can't get stuck. Tests cover the substantive cases (both flip directions, both collision shapes, non-digit whatsapp left alone, idempotency). URN model changes match gocommon.

A few things worth surfacing (all inline):

1. Concurrency during migration — atomic=False + no per-batch try/except means a concurrent insert of a colliding (org_id, identity) between the existing fetch and bulk_update will raise IntegrityError and abort the batch. Re-running recovers, but consider a maintenance window or per-batch error handling. Also the existing query could scope by org_id__in=… — currently it pulls matches across all orgs.
2. Display regression for collision-skipped legacy whatsapp URNs — removing the whatsapp branch from URN.format means any leftover whatsapp:<digits> row (collision case) will render as bare digits instead of the previous friendly phone format. Probably fine, but explicit.
3. URN.validate still accepts bare digits for whatsapp — necessary during migration, but leaves the API able to keep inserting legacy-shape rows indefinitely. Worth a follow-up to tighten once traffic stops writing the old form.
4. Batching test coverage — the migration test only exercises a single batch. Overriding BATCH_SIZE to 2 in a dedicated test would exercise the cursor/collision-advancement logic that's the trickiest part of _flip_scheme.

Nothing here is a merge blocker — no correctness bugs found in the flip logic itself, and the merge-order note (after #6709) is already called out in the PR body.
• branch

[norkans7]

Replaced by #6727