V3 S1-S2 : corbeille + versions + extract générique

S1 filet (backend + mobile) : corbeille 30j (deleted_at réutilisé,
trash/restore récursifs, purge auto serveur), versions N=3 à chaque
overwrite, ops outbox trash/restore/purge (delete legacy mappé), Room
v19 (trashed_at), Deleters -> trash, écran Corbeille, walk guards.

S2 extract (backend) : package extract (règles FR stdlib, corpus
testé), resource_facts 000019 (origin auto/user), hook worker OCR +
backfill boot, GET /files/:id/facts (viewer+).

Contrat docs/api-v1.md étendu, V3.md (vision + plan).
This commit is contained in:
venture
2026-09-26 10:14:07 +02:00
parent a34220bf2d
commit 95b4cf79cf
75 changed files with 5220 additions and 119 deletions
@@ -0,0 +1 @@
DROP INDEX IF EXISTS idx_resources_trashed_active;
@@ -0,0 +1,7 @@
-- V3 corbeille : `deleted_at` est le timestamp de trash (restauration =
-- `deleted_at = NULL`, purge = hard delete après 30j via PurgeExpired).
-- Toutes les lectures filtrent déjà `deleted_at IS NULL` : les trashés sont
-- invisibles partout sauf via les requêtes dédiées (ListTrashed). Cet index
-- partiel accélère la corbeille (`GET /trash`) et la passe de purge.
CREATE INDEX IF NOT EXISTS idx_resources_trashed_active
ON resources(user_id, deleted_at) WHERE deleted_at IS NOT NULL;
@@ -0,0 +1 @@
DROP TABLE IF EXISTS resource_versions;
@@ -0,0 +1,19 @@
-- V3 versions : historique léger des octets écrasés (N=3, cf. V3.md P1).
-- À chaque overwrite (upload ciblé), l'ancienne copie physique est déplacée
-- sous `stored_path` (relatif à UPLOAD_DIR) et sa métadonnée figée ici.
-- Restaurer = recopier vers l'emplacement live (l'actuel est à son tour
-- versionné). `ON DELETE CASCADE` : purger la ressource purge ses versions.
CREATE TABLE resource_versions (
id TEXT PRIMARY KEY CHECK (id ~ '^[0-9a-f]{32}$'),
resource_id TEXT NOT NULL REFERENCES resources(resource_id) ON DELETE CASCADE,
version_no INT NOT NULL CHECK (version_no >= 1),
size_bytes BIGINT NOT NULL DEFAULT 0,
mime_type VARCHAR(255),
extension VARCHAR(64),
content_hash VARCHAR(128),
stored_path TEXT NOT NULL,
created_at TIMESTAMPTZ DEFAULT NOW(),
UNIQUE (resource_id, version_no)
);
CREATE INDEX idx_versions_resource ON resource_versions(resource_id);
@@ -0,0 +1 @@
DROP TABLE IF EXISTS resource_facts;
+23
View File
@@ -0,0 +1,23 @@
-- V3 extract générique (S2) : faits universels extraits du texte OCR
-- (dates, montants, émetteurs, références, type suggéré) — cf. V3.md §2.
-- `doc_type` n'est qu'un fait comme un autre (suggestion, jamais gate) ;
-- `origin` distingue l'auto (remplaçable à chaque passe) du `user` (S3,
-- correction en un tap — l'auto ne l'écrase jamais). Cascade : purger la
-- ressource purge ses faits.
CREATE TABLE resource_facts (
id TEXT PRIMARY KEY CHECK (id ~ '^[0-9a-f]{32}$'),
resource_id TEXT NOT NULL REFERENCES resources(resource_id) ON DELETE CASCADE,
fact_type TEXT NOT NULL CHECK (fact_type IN (
'date', 'due_date', 'period', 'amount', 'issuer', 'reference', 'doc_type'
)),
value_text TEXT,
value_num NUMERIC,
value_date DATE,
confidence REAL NOT NULL CHECK (confidence >= 0 AND confidence <= 1),
origin TEXT NOT NULL DEFAULT 'auto' CHECK (origin IN ('auto', 'user')),
created_at TIMESTAMPTZ DEFAULT NOW(),
updated_at TIMESTAMPTZ DEFAULT NOW()
);
CREATE INDEX idx_facts_resource ON resource_facts(resource_id);
CREATE INDEX idx_facts_type_value ON resource_facts(fact_type, value_text);
+53
View File
@@ -369,6 +369,59 @@ func TestMigrationsUpDown(t *testing.T) {
// --- 000016 : groupes + cible groupe sur shares -------------------------
assertTables(t, conn, "users", "devices", "resources", "operations", "ocr_jobs", "shares", "share_links", "storage_locations", "resource_placements", "groups", "group_members", "schema_migrations")
// --- 000017 : corbeille (index partiel deleted_at) ----------------------
var trashedIdx int
err = conn.QueryRow(`
SELECT COUNT(*) FROM pg_index i
JOIN pg_class t ON t.oid = i.indrelid
WHERE t.relname = 'resources' AND i.indisvalid
AND pg_get_indexdef(i.indexrelid) LIKE '%idx_resources_trashed_active%'`).Scan(&trashedIdx)
if err != nil {
t.Fatalf("000017 index: %v", err)
}
if trashedIdx == 0 {
t.Error("index idx_resources_trashed_active manquant après 000017")
}
// --- 000018 : versions --------------------------------------------------
assertTables(t, conn, "users", "devices", "resources", "operations", "ocr_jobs", "shares", "share_links", "storage_locations", "resource_placements", "groups", "group_members", "resource_versions", "schema_migrations")
assertHexCheck(t, conn, "resource_versions", "id")
var versionUniq int
err = conn.QueryRow(`
SELECT COUNT(*) FROM pg_index i
JOIN pg_class t ON t.oid = i.indrelid
WHERE t.relname = 'resource_versions' AND i.indisunique
AND ARRAY(SELECT a.attname FROM unnest(i.indkey) WITH ORDINALITY k(attnum, ord)
JOIN pg_attribute a ON a.attrelid = t.oid AND a.attnum = k.attnum
ORDER BY k.ord)::text[] = ARRAY['resource_id','version_no']`).Scan(&versionUniq)
if err != nil {
t.Fatalf("versions unique: %v", err)
}
if versionUniq == 0 {
t.Error("contrainte UNIQUE(resource_id, version_no) manquante sur resource_versions après 000018")
}
// --- 000019 : faits génériques -----------------------------------------
assertTables(t, conn, "users", "devices", "resources", "operations", "ocr_jobs", "shares", "share_links", "storage_locations", "resource_placements", "groups", "group_members", "resource_versions", "resource_facts", "schema_migrations")
assertHexCheck(t, conn, "resource_facts", "id")
var factTypeCheck int
err = conn.QueryRow(`
SELECT COUNT(*) FROM pg_constraint c
JOIN pg_class t ON t.oid = c.conrelid
WHERE t.relname = 'resource_facts'
AND pg_get_constraintdef(c.oid) LIKE '%doc_type%'
AND pg_get_constraintdef(c.oid) LIKE '%due_date%'`).Scan(&factTypeCheck)
if err != nil {
t.Fatalf("check resource_facts.fact_type: %v", err)
}
if factTypeCheck == 0 {
t.Error("CHECK fact_type manquant sur resource_facts après 000019")
}
if err := MigrateDownDatabase(url); err != nil {
t.Fatalf("migrate down: %v", err)
}