fix(preprocess): harden S3 and real Compose jobs

harness/tests/test_s3_evidence_source.py

@@ -15,12 +15,12 @@ class Client:
     def __init__(self): self.body = Body(b"hello")
     def get_paginator(self, name): return self
     def paginate(self, **kwargs):
-        yield {"Contents": [{"Key": "clinical/a.md", "ETag": '"abc"', "VersionId": "v1",
+        yield {"Contents": [{"Key": "clinical/a.md", "ETag": '"abc"',
                               "Size": 5, "LastModified": datetime(2026, 1, 1, tzinfo=UTC)}]}
     def get_object(self, **kwargs):
-        assert kwargs == {"Bucket": "evidence", "Key": "clinical/a.md", "VersionId": "v1"}
+        assert kwargs == {"Bucket": "evidence", "Key": "clinical/a.md"}
         return {"Body": self.body, "ContentLength": 5, "ContentType": "text/markdown",
-                "ETag": '"abc"', "VersionId": "v1"}
+                "ETag": '"abc"'}
 
 
 def test_s3_canonical_uri_version_fingerprint_and_closed_body():
@@ -29,7 +29,7 @@ def test_s3_canonical_uri_version_fingerprint_and_closed_body():
     source = S3EvidenceSource(bucket="evidence", prefix="clinical/", client=client)
     item = next(iter(source.discover()))
     assert item.uri == "s3://evidence/clinical/a.md"
-    assert item.fingerprint == "s3-version:v1"
+    assert item.fingerprint.startswith("etag:")
     assert source.acquire(item).content == b"hello"
     assert client.body.closed
 
@@ -43,10 +43,55 @@ def test_s3_etag_fallback_and_bounds():
 
 def test_s3_rejects_private_or_insecure_endpoint_without_explicit_opt_in():
     from tht.adapters.evidence.s3 import S3EvidenceSource
-    with pytest.raises(ValueError, match="private"):
+    with pytest.raises(ValueError, match="trusted"):
         S3EvidenceSource(bucket="evidence", endpoint_url="https://127.0.0.1:9000", client=Client())
     with pytest.raises(ValueError, match="HTTPS"):
         S3EvidenceSource(bucket="evidence", endpoint_url="http://s3.example.test", client=Client())
+    source = S3EvidenceSource(bucket="evidence", endpoint_url="http://127.0.0.1:9000",
+                              trusted_endpoint=True, allow_private_endpoint=True,
+                              allow_insecure_endpoint=True, client=Client())
+    assert source is not None
+
+
+@pytest.mark.parametrize("bucket", ["UPPER", "bad_bucket", "-start", "end-", "a..b"])
+def test_s3_rejects_invalid_bucket_names(bucket):
+    from tht.adapters.evidence.s3 import S3EvidenceSource
+    with pytest.raises(ValueError, match="bucket"):
+        S3EvidenceSource(bucket=bucket, client=Client())
+
+
+def test_s3_rejects_endpoint_query_path_fragment_and_untrusted_custom_host():
+    from tht.adapters.evidence.s3 import S3EvidenceSource
+    for endpoint in ("https://s3.example.test/path", "https://s3.example.test/?x=1",
+                     "https://s3.example.test/#x"):
+        with pytest.raises(ValueError, match="root"):
+            S3EvidenceSource(bucket="evidence", endpoint_url=endpoint,
+                             trusted_endpoint=True, client=Client())
+    with pytest.raises(ValueError, match="trusted"):
+        S3EvidenceSource(bucket="evidence", endpoint_url="https://s3.example.test", client=Client())
+
+
+def test_s3_rejects_out_of_prefix_key_and_missing_validator():
+    from tht.adapters.evidence.s3 import S3EvidenceSource
+    client = Client()
+    client.paginate = lambda **kwargs: iter([{"Contents": [{"Key": "other/a.md", "ETag": '"x"'}]}])
+    with pytest.raises(EvidenceSourceError):
+        list(S3EvidenceSource(bucket="evidence", prefix="clinical/", client=client).discover())
+
+
+def test_s3_acquire_rejects_exact_etag_drift_and_closes_body():
+    from tht.adapters.evidence.s3 import S3EvidenceSource
+    client = Client()
+    source = S3EvidenceSource(bucket="evidence", client=client)
+    item = next(iter(source.discover()))
+    client.get_object = lambda **kwargs: {"Body": client.body, "ContentLength": 5,
+                                          "ETag": '"changed"'}
+    with pytest.raises(EvidenceSourceError):
+        source.acquire(item)
+    assert client.body.closed
+    client.paginate = lambda **kwargs: iter([{"Contents": [{"Key": "clinical/a.md"}]}])
+    with pytest.raises(EvidenceSourceError):
+        list(S3EvidenceSource(bucket="evidence", prefix="clinical/", client=client).discover())
 
 
 def test_s3_size_limit_closes_body():