Solveur de suggestions (OR-Tools CP-SAT) — ne place jamais personne

Nouveau module `backend/app/solver/` (mêmes conventions que `validation/` : models purs, engine pur testé, loader seul lien avec la base). Pour chaque créneau encore sous son effectif idéal, calcule une liste ordonnée de bénévoles candidats — c'est la coordination qui affecte, dans l'écran habituel ; le solveur fait juste remonter ces candidats en tête de liste (voir le câblage frontend à suivre). - Seule contrainte dure : aucun chevauchement de créneaux pour un même bénévole (règle SHIFT_OVERLAP), à la fois contre l'existant et entre deux suggestions du même calcul. L'effectif idéal est un plafond dur. - Tout le reste est un terme de l'objectif, pas un blocage : priorité forte aux créneaux sous l'effectif minimal, bonus vœu de poste, malus disponibilité manquante (réutilise `validation.engine.uncovered_minutes`), cumul de 8 h et repos (contre les affectations déjà décidées), répartition de la charge. - `POST /editions/{id}/solver/suggestions` (COORDINATION/référent) : recalcule à la demande, ne persiste jamais d'affectation. Dépendance `ortools`. 18 nouveaux tests (moteur + API), 172 tests au vert, `ruff` et `alembic check` verts. Testé sur le jeu de démo (120 bénévoles, 100 créneaux) : ~0,3 s. CLAUDE.md : nouvelle section 4 (solveur), renumérotation 5-10, compétences retirées de « hors périmètre », journal étape 15. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017uUsRg7XzopkCPWimwg9ZM
2026-09-11 08:16:17 +02:00
parent 36d108b47d
commit 803db5e9a9
12 changed files with 818 additions and 11 deletions
@@ -0,0 +1,148 @@
+"""Tests du solveur (suggestions, jamais d'affectation automatique)."""
+
+import datetime as dt
+
+from app.solver.engine import solve
+from app.solver.models import SolverInput, SolverShift, SolverVolunteer
+from app.validation.models import AvailabilityWindow, ShiftInfo
+
+
+def _dt(day: int, hour: int, minute: int = 0) -> dt.datetime:
+    return dt.datetime(2026, 7, day, hour, minute)
+
+
+def _shift(
+    sid: str = "s1",
+    *,
+    position_id: str = "p1",
+    start: dt.datetime | None = None,
+    end: dt.datetime | None = None,
+    min_staff=1,
+    ideal_staff=1,
+    already_assigned=0,
+) -> SolverShift:
+    return SolverShift(
+        id=sid,
+        position_id=position_id,
+        starts_at=start or _dt(10, 14),
+        ends_at=end or _dt(10, 18),
+        min_staff=min_staff,
+        ideal_staff=ideal_staff,
+        already_assigned=already_assigned,
+    )
+
+
+def _volunteer(
+    vid: str,
+    *,
+    availability=(),
+    wished_position_ids=frozenset(),
+    existing_shifts=(),
+    existing_minutes=0,
+) -> SolverVolunteer:
+    return SolverVolunteer(
+        id=vid,
+        availability=tuple(availability),
+        wished_position_ids=frozenset(wished_position_ids),
+        existing_shifts=tuple(existing_shifts),
+        existing_minutes=existing_minutes,
+    )
+
+
+def _full_day(day: int) -> AvailabilityWindow:
+    return AvailabilityWindow(dt.date(2026, 7, day), dt.time(6, 0), dt.time(4, 0))
+
+
+def _by_shift(solution, shift_id: str) -> tuple[str, ...]:
+    for s in solution.suggestions:
+        if s.shift_id == shift_id:
+            return s.volunteer_ids
+    return ()
+
+
+def test_no_variables_returns_empty_solution():
+    data = SolverInput(shifts=(), volunteers=())
+    assert solve(data).suggestions == ()
+
+
+def test_fully_staffed_shift_gets_no_suggestion():
+    shift = _shift(ideal_staff=1, already_assigned=1)
+    volunteer = _volunteer("v1", availability=[_full_day(10)])
+    solution = solve(SolverInput(shifts=(shift,), volunteers=(volunteer,)))
+    assert solution.suggestions == ()
+
+
+def test_picks_the_only_free_volunteer_over_a_busy_one():
+    shift = _shift()
+    busy = _volunteer(
+        "busy",
+        availability=[_full_day(10)],
+        existing_shifts=[ShiftInfo(id="e", starts_at=_dt(10, 15), ends_at=_dt(10, 19))],
+    )
+    free = _volunteer("free", availability=[_full_day(10)])
+    solution = solve(SolverInput(shifts=(shift,), volunteers=(busy, free)))
+    assert _by_shift(solution, "s1") == ("free",)
+
+
+def test_does_not_exceed_ideal_staff_cap():
+    shift = _shift(ideal_staff=2, already_assigned=1)  # il ne manque qu'une place
+    a = _volunteer("a", availability=[_full_day(10)])
+    b = _volunteer("b", availability=[_full_day(10)])
+    solution = solve(SolverInput(shifts=(shift,), volunteers=(a, b)))
+    assert len(_by_shift(solution, "s1")) == 1
+
+
+def test_prefers_volunteer_with_matching_position_wish():
+    shift = _shift(position_id="regie")
+    plain = _volunteer("plain", availability=[_full_day(10)])
+    wisher = _volunteer(
+        "wisher", availability=[_full_day(10)], wished_position_ids={"regie"}
+    )
+    solution = solve(SolverInput(shifts=(shift,), volunteers=(plain, wisher)))
+    assert _by_shift(solution, "s1") == ("wisher",)
+
+
+def test_prefers_fully_available_volunteer_over_partially_available():
+    shift = _shift(start=_dt(10, 14), end=_dt(10, 18))
+    full = _volunteer("full", availability=[_full_day(10)])
+    partial = _volunteer(
+        "partial",
+        availability=[AvailabilityWindow(dt.date(2026, 7, 10), dt.time(14, 0), dt.time(16, 0))],
+    )
+    solution = solve(SolverInput(shifts=(shift,), volunteers=(full, partial)))
+    assert _by_shift(solution, "s1") == ("full",)
+
+
+def test_ranks_both_candidates_when_two_slots_are_open():
+    shift = _shift(ideal_staff=2)
+    full = _volunteer("full", availability=[_full_day(10)])
+    partial = _volunteer(
+        "partial",
+        availability=[AvailabilityWindow(dt.date(2026, 7, 10), dt.time(14, 0), dt.time(16, 0))],
+    )
+    solution = solve(SolverInput(shifts=(shift,), volunteers=(full, partial)))
+    assert _by_shift(solution, "s1") == ("full", "partial")
+
+
+def test_never_double_books_a_volunteer_across_two_new_suggestions():
+    # Deux créneaux qui se chevauchent, un seul bénévole disponible pour les
+    # deux : le solveur ne le propose qu'une fois.
+    shift_a = _shift("a", start=_dt(10, 14), end=_dt(10, 18))
+    shift_b = _shift("b", start=_dt(10, 16), end=_dt(10, 20))
+    only = _volunteer("only", availability=[_full_day(10)])
+    solution = solve(SolverInput(shifts=(shift_a, shift_b), volunteers=(only,)))
+    picked_a = _by_shift(solution, "a")
+    picked_b = _by_shift(solution, "b")
+    assert len(picked_a) + len(picked_b) == 1
+
+
+def test_avoids_volunteer_who_would_exceed_eight_hours_that_day():
+    shift = _shift(start=_dt(10, 14), end=_dt(10, 18))  # 4 h
+    tired = _volunteer(
+        "tired",
+        availability=[_full_day(10)],
+        existing_shifts=[ShiftInfo(id="e", starts_at=_dt(10, 6), ends_at=_dt(10, 12))],  # 6 h déjà
+    )
+    fresh = _volunteer("fresh", availability=[_full_day(10)])
+    solution = solve(SolverInput(shifts=(shift,), volunteers=(tired, fresh)))
+    assert _by_shift(solution, "s1") == ("fresh",)