fix: harden Task 2 schema quality boundaries

harness/tests/test_schema_fk_annotations.py

@@ -1,10 +1,13 @@
 # ruff: noqa: DTZ001
+import warnings
 from datetime import datetime
 
+import pytest
 import yaml
 from typer.testing import CliRunner
 
 from tht.cli import app
+from tht.cli.schema_cmd import check_schema_data
 from tht.mschema.merge import find_orphans
 from tht.mschema.models import (
     Annotations,
@@ -163,6 +166,10 @@ def test_suggest_fks_skips_generic_and_ambiguous_pks(tmp_path):
     assert res.exit_code == 0, res.output
     assert "nessuna FK da suggerire" in res.output  # id generico, cod_x ambigua
     assert "cod_x" in res.output  # segnalata come ambigua saltata
+    exact = CliRunner().invoke(
+        app, ["schema", "suggest-fks", "--json", "-c", str(cfg)]
+    )
+    assert yaml.safe_load(exact.stdout)["ambiguous_columns"] == ["cod_x"]
 
     # --assume disambigua la PK multi-proprietario
     res2 = CliRunner().invoke(
@@ -205,6 +212,18 @@ def test_suggest_fks_from_sql_mines_joins(tmp_path):
     assert "ref_table: dim_patient" in res.output
 
 
+def test_suggest_fks_reports_staged_file_without_mined_joins(tmp_path):
+    cfg = _write_workspace(tmp_path)
+    staged = tmp_path / "approved"
+    staged.mkdir()
+    (staged / "no-joins.sql").write_text("SELECT 1")
+    response = CliRunner().invoke(
+        app, ["schema", "suggest-fks", "-c", str(cfg), "--from-sql", str(staged)]
+    )
+    assert response.exit_code == 0, response.output
+    assert "Minati 0 equi-join da 1 file SQL" in response.output
+
+
 def test_suggest_fks_write_merges_and_is_idempotent(tmp_path):
     cfg = _write_workspace(tmp_path)
     ann_path = tmp_path / "artifacts" / "mschema" / "annotations.yaml"
@@ -435,3 +454,59 @@ def test_schema_human_suggest_missing_config_has_original_error(tmp_path):
     assert response.exit_code == 1
     assert response.output
     assert "ERRORE:" in response.output
+
+
+def test_human_config_keeps_legacy_warning_but_json_suppresses_it(tmp_path):
+    cfg = _write_workspace(tmp_path)
+    with pytest.warns(FutureWarning, match="legacy workspace resource keys"):
+        check_schema_data(cfg)
+    with warnings.catch_warnings(record=True) as caught:
+        warnings.simplefilter("always")
+        check_schema_data(cfg, suppress_legacy_warning=True)
+    assert not [item for item in caught if issubclass(item.category, FutureWarning)]
+
+
+def test_staged_sql_enumeration_stops_at_count_sentinel(tmp_path, monkeypatch):
+    from tht.cli.schema_cmd import _MachineSchemaError, _staged_sql_files
+
+    staged = tmp_path / "staged"
+    staged.mkdir()
+    files = [staged / f"q{i:03d}.sql" for i in range(100)]
+    for path in files:
+        path.write_text("SELECT 1")
+    yielded = 0
+
+    def bounded_rglob(_self, _pattern):
+        nonlocal yielded
+        for path in files:
+            yielded += 1
+            yield path
+
+    monkeypatch.setattr(type(staged), "rglob", bounded_rglob)
+    with pytest.raises(_MachineSchemaError, match="staged_sql_too_many"):
+        _staged_sql_files([staged])
+    assert yielded == 33
+
+
+def test_staged_sql_oversize_read_is_bounded(tmp_path, monkeypatch):
+    from tht.cli.schema_cmd import _MachineSchemaError, _read_staged_sql
+
+    staged = tmp_path / "oversize.sql"
+    staged.write_bytes(b"unused")
+    reads = []
+
+    class BoundedReader:
+        def __enter__(self):
+            return self
+
+        def __exit__(self, *_args):
+            return False
+
+        def read(self, size):
+            reads.append(size)
+            return b"x" * size
+
+    monkeypatch.setattr(type(staged), "open", lambda *_args, **_kwargs: BoundedReader())
+    with pytest.raises(_MachineSchemaError, match="staged_sql_too_large"):
+        _read_staged_sql([staged])
+    assert reads == [(1 << 20) + 1]