From aa78eedb9c25ccef35792aab3c301341e344b6d1 Mon Sep 17 00:00:00 2001
From: alex-dembele <alexandredembele16@gmail.com>
Date: Fri, 2 Oct 2026 13:04:22 +0100
Subject: [PATCH] perf(db): index LOWER(email) for the case-insensitive account
 lookup (#876)

Signed-off-by: alex-dembele <alexandredembele16@gmail.com>
---
 docs/MIGRATIONS.md                            | 16 ++++++++++++++
 .../0068_users_email_lower_index.down.sql     |  1 +
 .../0068_users_email_lower_index.up.sql       | 22 +++++++++++++++++++
 3 files changed, 39 insertions(+)
 create mode 100644 migrations/0068_users_email_lower_index.down.sql
 create mode 100644 migrations/0068_users_email_lower_index.up.sql

diff --git a/docs/MIGRATIONS.md b/docs/MIGRATIONS.md
index 0598618c..63d46d89 100644
--- a/docs/MIGRATIONS.md
+++ b/docs/MIGRATIONS.md
@@ -78,3 +78,19 @@ vide : schéma de la version précédente, données de tous les âges et de tout
 NULL, lignes antérieures aux hooks), puis la séquence de boot (`PrepareForAutoMigrate`,
 AutoMigrate, couche SQL). Modèle : `backend/cmd/server/migration_0060_integration_test.go`.
 Procédure de montée de version et de retour arrière associée : `docs/runbooks/migration-0060.md`.
+
+**Adresses e-mail en double à la casse près** (`0068`, #876). `GetByEmail` compare
+`LOWER(email)` depuis #688, et `idx_users_email_lower` sert cette recherche. L'index n'est pas
+unique, parce que des comptes créés avant #687 peuvent exister en double à la casse près. Avant de
+proposer une contrainte unique, lister ces doublons (lecture seule) :
+```sql
+SELECT LOWER(email) AS adresse, COUNT(*) AS comptes,
+       array_agg(email ORDER BY created_at) AS graphies
+  FROM users
+ WHERE deleted_at IS NULL
+ GROUP BY LOWER(email)
+HAVING COUNT(*) > 1;
+```
+Sans ligne, la base est prête pour une contrainte unique. Sinon, la connexion atteint la ligne
+écrite exactement comme saisie, puis la plus ancienne : fusionner ou renommer les doublons est une
+décision de données, pas une migration.
diff --git a/migrations/0068_users_email_lower_index.down.sql b/migrations/0068_users_email_lower_index.down.sql
new file mode 100644
index 00000000..bb183c6e
--- /dev/null
+++ b/migrations/0068_users_email_lower_index.down.sql
@@ -0,0 +1 @@
+DROP INDEX IF EXISTS idx_users_email_lower;
diff --git a/migrations/0068_users_email_lower_index.up.sql b/migrations/0068_users_email_lower_index.up.sql
new file mode 100644
index 00000000..3c684bc9
--- /dev/null
+++ b/migrations/0068_users_email_lower_index.up.sql
@@ -0,0 +1,22 @@
+-- #876 — index the lookup login actually runs.
+--
+-- Since #688, GetByEmail matches LOWER(email) so the casing typed never decides
+-- which account is reached. The unique index on users.email is on the raw
+-- column and cannot serve that predicate, so every sign-in, sign-up, reset and
+-- invitation scanned users.
+--
+-- Deliberately NOT unique: rows written before #687 may hold case-variant
+-- duplicates, and a unique index would fail this migration, and the boot with
+-- it, on such a database. docs/MIGRATIONS.md has the query that lists them.
+--
+-- users is created by AutoMigrate, which runs before this layer; the guard only
+-- matters for a run against an empty database.
+
+DO $$
+BEGIN
+    IF to_regclass('users') IS NULL THEN
+        RETURN;
+    END IF;
+    CREATE INDEX IF NOT EXISTS idx_users_email_lower ON users (LOWER(email));
+END
+$$;