Sprint review: URL templates from a spec may only name arguments; form requests send their fields

http2py/py2request.py

@@ -200,6 +200,11 @@ def mk_request_function(
 
     original_spec = method_spec
     method_spec = method_spec.copy()  # make a copy
+    if "url_template" in method_spec:
+        check_url_template(
+            method_spec["url_template"],
+            query_arg_names=_ensure_list(method_spec.get("query_arg_names", [])),
+        )
     method_spec["input_trans"] = method_spec.get("input_trans", None) or {}
 
     request_kwargs = method_spec.get("request_kwargs", {}).copy()
@@ -285,7 +290,7 @@ def request_func(*args, **kwargs):
                 _request_kwargs["data"] = json.dumps(json_kwargs)
             elif content_type == BINARY_CONTENT_TYPE:
                 _request_kwargs["data"] = binary_dumps(kwargs)
-        elif content_type == FORM_CONTENT_TYPE:
+        if content_type == FORM_CONTENT_TYPE:  # (it used to be sent with GET only)
             fields = json.dumps(json_kwargs).encode("utf-8")
             binaries = {k: (f"{k}.bin", v) for k, v in binary_kwargs.items()}
             files_data = dict(binaries, __fields=fields)
@@ -502,6 +507,44 @@ def _difference_conserving_order(a, b):
     return [x for x in a if x not in b]
 
 
+def _reads_into_arguments(field_name: str) -> bool:
+    """Whether a ``str.format`` field looks up an attribute or item of an argument;
+    any other name (``user-id``, ``0``...) only selects an argument."""
+    return "." in field_name or "[" in field_name
+
+
+def check_url_template(url_template: str, *, query_arg_names=()) -> str:
+    """Return ``url_template`` if no placeholder in it, nor any of the
+    ``query_arg_names`` (appended as ``?name={name}`` when calling), reads into an
+    argument.
+
+    A template is filled with ``str.format(**arguments)``, and it may come from a spec
+    fetched from a server. ``{a.__class__}`` or ``{a[0]}`` would read attributes or
+    items of the caller's arguments into the URL, so they are refused. Other names,
+    such as OpenAPI's hyphenated ``{user-id}``, are fine.
+
+    >>> check_url_template('http://h/u/{user-id}?q={q}')
+    'http://h/u/{user-id}?q={q}'
+    >>> check_url_template('http://h/{a.__init__.__globals__}')
+    Traceback (most recent call last):
+      ...
+    ValueError: URL template placeholder '{a.__init__.__globals__}' reads into an argument
+    """
+    fields = [
+        (field_name, format_spec)
+        for _, field_name, format_spec, _ in str_formatter.parse(url_template)
+        if field_name is not None
+    ]
+    fields += [(name, "") for name in query_arg_names]
+    for field_name, format_spec in fields:
+        if _reads_into_arguments(field_name) or "{" in (format_spec or ""):
+            raise ValueError(
+                f"URL template placeholder {'{' + field_name + '}'!r} reads into an "
+                "argument"
+            )
+    return url_template
+
+
 def extract_format_names(string_template):
     """
     Extracts the names of the fields in a string template, returning a list of

tests/test_response_decoding.py

@@ -0,0 +1,166 @@
+"""How responses are decoded (``http2py.decorators``) and how a client's requests are
+built from a spec it may have fetched from a server (``mk_request_function``)."""
+
+import pickle
+import warnings
+
+import pytest
+
+from http2py.decorators import (
+    BinaryResponseNotDecoded,
+    check_loads,
+    handle_binary_resp,
+    unsafe_pickle_loads,
+)
+from http2py.py2request import mk_request_function
+
+
+class Resp:
+    def __init__(self, content=b"", *, status_code=200, text="", reason="OK"):
+        self.content, self.status_code, self.text, self.reason = (
+            content,
+            status_code,
+            text,
+            reason,
+        )
+
+
+def _identity(x):
+    return x
+
+
+def test_binary_response_needs_an_explicit_decoder():
+    with pytest.raises(BinaryResponseNotDecoded):
+        handle_binary_resp(_identity)(Resp(pickle.dumps(1)))
+    assert handle_binary_resp(_identity, loads=bytes)(Resp(b"raw")) == b"raw"
+    trusted = handle_binary_resp(_identity, loads=unsafe_pickle_loads)
+    assert trusted(Resp(pickle.dumps({"a": 1}))) == {"a": 1}
+
+
+def test_check_loads():
+    check_loads(None)
+    check_loads(bytes)
+    with pytest.raises(TypeError):
+        check_loads("pickle.loads")
+    with pytest.warns(UserWarning, match="unsafe_pickle_loads"):
+        check_loads(pickle.loads)
+    with warnings.catch_warnings():
+        warnings.simplefilter("error")
+        check_loads(unsafe_pickle_loads)  # the explicit opt-in does not warn
+
+
+def test_non_200_binary_response_raises_the_mapped_error():
+    from i2.errors import NotFoundError
+
+    with pytest.raises(NotFoundError):
+        handle_binary_resp(_identity, loads=bytes)(Resp(status_code=404, text="no"))
+
+
+def _captured_request(method_spec, **call_kwargs):
+    """The (method, url, request kwargs) a request function would dispatch."""
+    calls = []
+
+    def dispatch(method, url, **kwargs):
+        calls.append((method, url, kwargs))
+        return Resp(b"{}")
+
+    func = mk_request_function(method_spec, dispatch=dispatch)
+    func(None, **call_kwargs)
+    return calls[0]
+
+
+def test_binary_request_bodies_use_binary_dumps():
+    spec = {
+        "method": "POST",
+        "url_template": "http://h/f",
+        "body_arg_names": ["x"],
+        "content_type": "application/octet-stream",
+        "output_trans": _identity,
+    }
+    _, _, kwargs = _captured_request(spec, x=1)
+    assert pickle.loads(kwargs["data"]) == {"x": 1}
+    calls = []
+    func = mk_request_function(
+        spec,
+        dispatch=lambda m, u, **kw: calls.append(kw) or Resp(),
+        binary_dumps=lambda d: b"custom",
+    )
+    func(None, x=1)
+    assert calls[0]["data"] == b"custom"
+
+
+def test_path_args_are_url_encoded():
+    spec = {
+        "method": "GET",
+        "url_template": "http://h/u/{uid}",
+        "path_arg_names": ["uid"],
+        "output_trans": _identity,
+    }
+    _, url, _ = _captured_request(spec, uid="a/b c")
+    assert url == "http://h/u/a%2Fb%20c"
+
+
+@pytest.mark.parametrize(
+    "template",
+    [
+        "http://h/{x.__class__}",
+        "http://h/{x[0]}",
+        "http://h/{x:{x.__class__}}",
+        "http://h/{x!r:{y}}",
+    ],
+)
+def test_url_templates_may_only_name_arguments(template):
+    """A spec served by a remote must not read attributes of the caller's
+    arguments into the URL."""
+    spec = {"method": "GET", "url_template": template, "path_arg_names": ["x"]}
+    with pytest.raises(ValueError, match="reads into an argument"):
+        mk_request_function(spec)
+
+
+def test_query_arg_names_may_only_name_arguments():
+    spec = {
+        "method": "GET",
+        "url_template": "http://h/x",
+        "query_arg_names": ["obj", "obj.token"],
+    }
+    with pytest.raises(ValueError, match="reads into an argument"):
+        mk_request_function(spec)
+
+
+def test_hyphenated_placeholders_still_work():
+    spec = {"method": "GET", "url_template": "http://h/u/{user-id}"}
+    calls = []
+    func = mk_request_function(
+        spec, dispatch=lambda m, u, **kw: calls.append(u) or Resp(b"")
+    )
+    func(None, **{"user-id": "42"})
+    assert calls == ["http://h/u/42"]
+
+
+def test_spec_with_attribute_placeholder_is_refused_by_the_client():
+    import copy
+    import sys
+    from pathlib import Path
+
+    sys.path.insert(0, str(Path(__file__).parent))
+    from spec_fixture import SPEC
+
+    from http2py import HttpClient
+
+    spec = copy.deepcopy(SPEC)
+    spec["paths"]["/x/{uid.__class__.__mro__}"] = spec["paths"].pop("/u/{uid}/p/{pid}")
+    with pytest.raises(ValueError, match="reads into an argument"):
+        HttpClient(spec, session_state={"session": object(), "refresh_inputs": {}})
+
+
+@pytest.mark.parametrize("method", ["GET", "POST"])
+def test_form_requests_send_their_fields_whatever_the_method(method):
+    spec = {
+        "method": method,
+        "url_template": "http://h/f",
+        "body_arg_names": ["x", "blob"],
+        "content_type": "multipart/form-data",
+        "output_trans": _identity,
+    }
+    _, _, kwargs = _captured_request(spec, x=1, blob=b"\x00")
+    assert set(kwargs["files"]) == {"blob", "__fields"}

tests/test_route_namespace.py

@@ -83,6 +83,7 @@ def test_cli_calls_the_route_not_the_client_attribute(monkeypatch):
     spec = _spec_naming_plain("ensure_login")
     with pytest.warns(UserWarning):
         route = _client(spec).routes["ensure_login"]
+
     class Sent(Exception):
         pass
 
@@ -162,3 +163,14 @@ def test_unusable_operation_id_falls_back_to_method_and_path(operation_id):
     spec = _spec_without_names()
     spec["paths"]["/plain"]["get"]["operationId"] = operation_id
     assert "get_plain" in _client(spec).routes
+
+
+def test_a_second_route_of_the_same_name_replaces_the_first_with_a_warning():
+    spec = copy.deepcopy(SPEC)
+    other = copy.deepcopy(spec["paths"]["/plain"])
+    spec["paths"]["/plain2"] = other
+    with pytest.warns(UserWarning, match="Two routes are named"):
+        client = _client(spec)
+    name = other["get"]["x-method_name"]
+    assert client.routes[name].method_spec["url_template"].endswith("/plain2")
+    assert getattr(client, name) is client.routes[name]