Update unit tests to be more strict

Since we now have an easily reproducible and identical test database
behind the PHP API, we can be more strict about ensuring there are
no differences in the data. Only changed interfaces are now
accounted for in the unit tests.

tests/identical_test.py

@@ -34,32 +34,8 @@ def test_dataset_response_is_identical(dataset_id: int, api_client: FastAPI) ->
     if "div" in original:
         pytest.skip("A PHP error occurred on the test server.")
 
-    # In case the test environment is set up to communicate with a snapshot of
-    # the test server database, we expect some fields to be outdated:
-    assert int(original["description_version"]) >= int(new["description_version"])
-    if "tag" in original and "tag" in new:
-        # TODO: Ask Jan why some datasets don't have tags.
-        assert set(original["tag"]) >= set(new["tag"])
-
-    assert original["format"].lower() == new["format"]
-
-    # Sometimes the test server provides `parquet_url`s, sometimes not.
-    if "parquet_url" in original:
-        if original["format"] == "sparse_arff":
-            # https://github.com/openml/OpenML/issues/1189
-            # The test server incorrectly thinks there is an associated parquet file:
-            del original["parquet_url"]
-    elif "parquet_url" in new:
-        del new["parquet_url"]
-
-    # Format is tested in normalized form above, tags and description_version may
-    # be out of sync with the snapshot of the database that we use to generate
-    # new responses.
-    for field in ["description_version", "tag", "format"]:
-        if field in original:
-            del original[field]
-        if field in new:
-            del new[field]
+    # The new API has normalized `format` field:
+    original["format"] = original["format"].lower()
 
     # There is odd behavior in the live server that I don't want to recreate:
     # when the creator is a list of csv names, it can either be a str or a list
@@ -73,11 +49,6 @@ def test_dataset_response_is_identical(dataset_id: int, api_client: FastAPI) ->
     ):
         original["creator"] = [name.strip() for name in original["creator"].split(",")]
 
-    # For some reason, the TALLO dataset has multiple 'ignore attribute' but the
-    # live server is not able to parse that and provides no 'ignore attribute' field:
-    if dataset_id in range(8592, 8606):
-        del new["ignore_attribute"]
-
     # The remainder of the fields should be identical:
     assert original == new