wacrm/supabase/migrations/022_contact_phone_dedup.sql

-- ============================================================
-- 022_contact_phone_dedup
--
-- Prevent the same phone number from becoming multiple contacts
-- within one account (issue #212).
--
-- Until now `contacts.phone` had only a non-unique index, phone was
-- stored un-normalized ("+1 555-123-4567" vs "15551234567" are
-- distinct strings), and only the WhatsApp webhook de-duped. Manual
-- create and CSV import inserted freely, fragmenting conversations,
-- deals, and tags across duplicate rows.
--
-- This migration, in order:
--   1. adds a generated `phone_normalized` column (digits-only,
--      mirroring the app's normalizePhone) that can never drift;
--   2. merges existing duplicates into the oldest row, re-pointing
--      all child records first so nothing is lost;
--   3. adds a UNIQUE index on (account_id, phone_normalized) — the
--      authoritative guarantee that covers every write path.
--
-- Idempotent. **No data loss** — duplicate rows are merged, not
-- dropped: child rows (conversations, messages, deals, notes, tags,
-- custom values, broadcast recipients, automation/flow records) are
-- re-pointed to the surviving (oldest) contact before deletion.
-- ============================================================

-- 1) Normalized phone — STORED generated column, kept in lockstep
--    with `phone` by Postgres. Matches normalizePhone()
--    (src/lib/whatsapp/phone-utils.ts): strip every non-digit.
ALTER TABLE contacts
  ADD COLUMN IF NOT EXISTS phone_normalized TEXT
  GENERATED ALWAYS AS (regexp_replace(phone, '\D', '', 'g')) STORED;

-- 2) One-time (re-runnable) merge of existing duplicates.
--    SECURITY DEFINER so it can re-point rows across tables
--    regardless of the caller's RLS; it only ever collapses exact
--    normalized duplicates within the same account.
CREATE OR REPLACE FUNCTION public.merge_duplicate_contacts()
RETURNS INTEGER
LANGUAGE plpgsql
SECURITY DEFINER
SET search_path = public
AS $$
DECLARE
  v_group   RECORD;
  v_survivor UUID;
  v_losers   UUID[];
  v_merged   INTEGER := 0;
BEGIN
  FOR v_group IN
    SELECT account_id,
           phone_normalized,
           array_agg(id ORDER BY created_at ASC, id ASC) AS ids
    FROM contacts
    WHERE phone_normalized <> ''
    GROUP BY account_id, phone_normalized
    HAVING count(*) > 1
  LOOP
    v_survivor := v_group.ids[1];
    v_losers   := v_group.ids[2:array_length(v_group.ids, 1)];

    -- Plain re-point: these tables have no contact-scoped unique
    -- constraint. `conversations` is ON DELETE CASCADE, so this
    -- re-point is what saves its rows (and their messages) from
    -- being deleted with the loser contact.
    UPDATE conversations                 SET contact_id = v_survivor WHERE contact_id = ANY(v_losers);
    UPDATE contact_notes                 SET contact_id = v_survivor WHERE contact_id = ANY(v_losers);
    UPDATE deals                         SET contact_id = v_survivor WHERE contact_id = ANY(v_losers);
    UPDATE broadcast_recipients          SET contact_id = v_survivor WHERE contact_id = ANY(v_losers);
    UPDATE automation_logs               SET contact_id = v_survivor WHERE contact_id = ANY(v_losers);
    UPDATE automation_pending_executions SET contact_id = v_survivor WHERE contact_id = ANY(v_losers);

    -- Conflict-guarded re-point for UNIQUE(contact_id, tag_id):
    -- move only tags the survivor doesn't already have, drop the rest.
    UPDATE contact_tags ct SET contact_id = v_survivor
      WHERE ct.contact_id = ANY(v_losers)
        AND NOT EXISTS (
          SELECT 1 FROM contact_tags s
          WHERE s.contact_id = v_survivor AND s.tag_id = ct.tag_id
        );
    DELETE FROM contact_tags WHERE contact_id = ANY(v_losers);

    -- Same guard for UNIQUE(contact_id, custom_field_id). Survivor's
    -- own value wins on conflict.
    UPDATE contact_custom_values cv SET contact_id = v_survivor
      WHERE cv.contact_id = ANY(v_losers)
        AND NOT EXISTS (
          SELECT 1 FROM contact_custom_values s
          WHERE s.contact_id = v_survivor AND s.custom_field_id = cv.custom_field_id
        );
    DELETE FROM contact_custom_values WHERE contact_id = ANY(v_losers);

    -- flow_runs has a partial UNIQUE on active runs per contact.
    -- Re-point only NON-active runs (exempt from the partial index)
    -- to preserve history; any active loser run is left to be
    -- NULLed by its FK's ON DELETE SET NULL when the loser is
    -- removed below — avoids colliding with the survivor's active run.
    UPDATE flow_runs SET contact_id = v_survivor
      WHERE contact_id = ANY(v_losers) AND status <> 'active';

    DELETE FROM contacts WHERE id = ANY(v_losers);

    v_merged := v_merged + COALESCE(array_length(v_losers, 1), 0);
  END LOOP;

  RETURN v_merged;
END;
$$;

ALTER FUNCTION public.merge_duplicate_contacts() OWNER TO postgres;
REVOKE ALL ON FUNCTION public.merge_duplicate_contacts() FROM PUBLIC;

-- Collapse whatever duplicates exist right now.
SELECT public.merge_duplicate_contacts();

-- 3) Authoritative guarantee. Partial index defends against any
--    empty normalized value (phone is NOT NULL, but belt-and-braces).
CREATE UNIQUE INDEX IF NOT EXISTS idx_contacts_account_phone_normalized
  ON contacts (account_id, phone_normalized)
  WHERE phone_normalized <> '';