mostlyright-data 0.19.7__tar.gz → 0.19.8__tar.gz

{mostlyright_data-0.19.7 → mostlyright_data-0.19.8}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: mostlyright-data
-Version: 0.19.7
+Version: 0.19.8
 Summary: Mostly Right hosted CLI for reviewed datasets
 Project-URL: Homepage, https://mostlyright.md/
 Project-URL: Documentation, https://mostlyright.md/docs/guides/cli/

{mostlyright_data-0.19.7 → mostlyright_data-0.19.8}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "mostlyright-data"
-version = "0.19.7"
+version = "0.19.8"
 description = "Mostly Right hosted CLI for reviewed datasets"
 readme = "README.md"
 requires-python = ">=3.13"

{mostlyright_data-0.19.7 → mostlyright_data-0.19.8}/src/mostlyright/data_harness/thin/download.py

@@ -88,7 +88,15 @@ class DownloadedArtifact:
 
 
 def _required_headers(session: Mapping[str, Any]) -> dict[str, str]:
-    """The headers the signature covers. Sending fewer, or more, invalidates it."""
+    """The headers Studio signed, replayed exactly as it stated them.
+
+    Omitting one, or altering a value, is what invalidates the signature; this builds the outgoing
+    set from the session and from nothing else, so neither can happen here. It is NOT an exact
+    header set for the request as a whole -- :class:`~mostlyright.data_harness.thin.transport.
+    UrlLibTransport` adds this client's identity to every call it makes, signed transfers
+    included, and those transfers succeed. What a given storage backend tolerates beyond the
+    signature is that backend's business, and nothing in this lane may start depending on it.
+    """
 
     declared = session.get("required_headers")
     if declared is None:

{mostlyright_data-0.19.7 → mostlyright_data-0.19.8}/src/mostlyright/data_harness/thin/runs.py

@@ -46,6 +46,31 @@ from mostlyright.data_harness.thin.transport import (
 #: lane's half of that pin, and ``tests/test_thin_client.py`` holds the two equal.
 PINNED_CONTRACT_VERSION = "3.0.0"
 
+#: The header this client names itself under on every Studio call, and the name it gives.
+#:
+#: ⚠ IT IS A COUNTER SELECTOR AND NOTHING ELSE. Studio cannot tell a call from here apart from a
+#: call its own dashboard makes server-side -- both present a Cloud-minted token of the same
+#: credential class, minted by the same function -- so the caller states which one it is and
+#: Studio reads it to decide what to COUNT. It authorizes nothing, no answer changes because it
+#: was sent, and Studio is free to ignore it. Anyone can send it and anyone can withhold it, which
+#: makes the number it feeds a floor with noise in it rather than a measurement of anything.
+#:
+#: Stated per Studio call rather than in the transport, which is the other place every request
+#: passes through and the tempting one, because ``with_user_agent`` is already applied there. The
+#: reason is not that a signed transfer would break under one more header: the transport adds a
+#: user agent to those too, outside the ``required_headers`` the session names, and they succeed.
+#:
+#: The reason is who each header is for. That transport speaks to two kinds of origin -- our own
+#: services, and whatever host a signed transfer URL names, which ``signed_transfer_url`` checks
+#: for a scheme and for the absence of an embedded credential while asserting nothing about whose
+#: storage it is. An identity is owed to every origin a client calls, which is why the user agent
+#: is applied to all of them. A name that means something only to the service keeping the counter
+#: is owed to that service alone, and on the transport it would ride to a third party's storage on
+#: every download, where nothing can count it and nobody asked for it. ``_call`` builds its URL
+#: from ``session.studio_base_url``, so it is the layer that knows the answer is Studio.
+CLIENT_HEADER = "X-Mostlyright-Client"
+CLIENT_NAME = "mr-data"
+
 #: The routes. Kept as one table because it is the thing a reviewer checks against the contract.
 START_RUN_PATH = "/v3/runs"
 #: The same collection, read rather than added to. Written out rather than aliased to
@@ -211,7 +236,11 @@ class StudioApiClient:
         shape: type = dict,
         response_headers: dict[str, str] | None = None,
     ) -> Any:
-        headers = {"Accept": "application/json", **self._session.authorization()}
+        headers = {
+            "Accept": "application/json",
+            CLIENT_HEADER: CLIENT_NAME,
+            **self._session.authorization(),
+        }
         if extra_headers:
             headers.update(extra_headers)
         raw_body: bytes | None = None
@@ -505,6 +534,8 @@ def _studio_refusal(status: int, parsed: Any, path: str) -> ThinLaneError:
 
 __all__ = [
     "ARTIFACT_DOWNLOAD_SESSION_PATH",
+    "CLIENT_HEADER",
+    "CLIENT_NAME",
     "GET_RECIPE_PROPOSAL_PATH",
     "GET_RUN_PATH",
     "LIST_RUNS_PATH",

{mostlyright_data-0.19.7 → mostlyright_data-0.19.8}/src/mostlyright/data_harness/thin/stream.py

@@ -34,7 +34,11 @@ from mostlyright.data_harness.thin.progress_vocabulary import (
     PROGRESS_EVENT_FACTS,
     PROGRESS_SCHEMA_VERSION,
 )
-from mostlyright.data_harness.thin.runs import RUN_EVENTS_STREAM_PATH
+from mostlyright.data_harness.thin.runs import (
+    CLIENT_HEADER,
+    CLIENT_NAME,
+    RUN_EVENTS_STREAM_PATH,
+)
 from mostlyright.data_harness.thin.session import StudioSession
 from mostlyright.data_harness.thin.transport import ThinLaneError, Transport, UrlLibTransport
 
@@ -389,7 +393,11 @@ class RunEventStream:
                 return True
             return False
 
-        headers = {"Accept": "text/event-stream", **self._session.authorization()}
+        headers = {
+            "Accept": "text/event-stream",
+            CLIENT_HEADER: CLIENT_NAME,
+            **self._session.authorization(),
+        }
         # Resume in Studio's own cursor space when we have one. `from_seq` is only ever the
         # starting position of a watch that has not yet seen a frame.
         if progress.last_event_id: