fix(mcp): declare the HTTP transport's dependencies, and forbid its tests to skip
`starlette` and `uvicorn` arrive through the MCP SDK today -- measured,
both of them -- and relying on that is exactly the trap this project
already wrote down twice. `src/hhstream_mcp/pyproject.toml` declares
`anyio` rather than taking it for granted through the SDK because
`server.py` uses it by name; `http.py` imports `uvicorn` by name and sits
on Starlette's `Request` through the SDK's `TransportSecuritySettings`,
so both belong in the metadata on the same argument. An SDK release that
moved its HTTP transport behind an optional extra would otherwise break
`serve --http` with an ImportError naming somebody else's package.
`nix/mcp.nix` carries them for the same reason.
The worse half was in the gate. `tests/unit/test_mcp_http.py` opened
with `pytest.importorskip("starlette.testclient")`, which is the pattern
`checks/unit.nix`'s own comment argues against in as many words: "a file
that skipped when the SDK was absent would be green in exactly the build
where that mattered". It would not skip today -- `mcp` propagates
starlette and httpx -- but the day it did, 43 tests would leave the gate
and nothing would say so, because pytest exits 0 when every test was
skipped.
So the import is unconditional now, `checks/unit.nix` declares `starlette`
and `httpx` explicitly, and the check proves them importable before
pytest runs, beside the lines that already do that for `hhstream` and the
tool table.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KchcvnbfhhZhUHy83YL3Ya
This commit is contained in:
1 parent
d788d5723e
commit
8a0c9ca74d
4 files changed
+40
-9
No files matched your search
@@ -32,11 +32,20 @@ let
|
||||
# `mcp.types.Tool` objects is where a malformed schema or a name the spec
|
||||
# refuses would show up, and a file that skipped when the SDK was absent
|
||||
# would be green in exactly the build where that mattered.
|
||||
# `starlette` and `httpx` beside `mcp`, and for the same reason the comment
|
||||
# above gives: `tests/unit/test_mcp_http.py` drives the HTTP transport
|
||||
# through `starlette.testclient`, which needs `httpx`. Both arrive through
|
||||
# the SDK today -- measured -- and relying on that is the trap. An SDK
|
||||
# release that moved its HTTP transport behind an optional extra would make
|
||||
# 43 tests vanish from this gate, and a gate that loses tests silently is
|
||||
# worse than one that fails.
|
||||
pytestEnv = python3.withPackages (ps: [
|
||||
ps.pytest
|
||||
ps.pywayland
|
||||
ps.xkbcommon
|
||||
ps.mcp
|
||||
ps.starlette
|
||||
ps.httpx
|
||||
]);
|
||||
|
||||
# Only the suite itself, so an editor's droppings or a local .pytest_cache
|
||||
@@ -84,6 +93,10 @@ runCommand "hhstream-unit"
|
||||
# so here, as for `hhstream`, the import comes from what Nix installed, and a
|
||||
# packaging mistake is distinguishable from a test failure.
|
||||
python3 -c 'import hhstream_mcp.tools as t; print("mcp tool table:", len(t.TOOLS), "tools")'
|
||||
# The HTTP transport's stack, proved present here rather than discovered as a
|
||||
# skip in the report: `test_mcp_http.py` imports `starlette.testclient` at
|
||||
# module scope and is not allowed to skip.
|
||||
python3 -c 'import starlette.testclient, hhstream_mcp.http; print("http transport imports")'
|
||||
|
||||
set +e
|
||||
pytest suite \
|
||||
|
||||
@@ -41,6 +41,11 @@ python3.pkgs.buildPythonApplication {
|
||||
dependencies = [
|
||||
python3.pkgs.mcp
|
||||
python3.pkgs.anyio
|
||||
# The HTTP transport's own two, declared for the reason `anyio` is: the SDK
|
||||
# propagates both today, and `hhstream_mcp.http` imports them by name. See
|
||||
# the note in src/hhstream_mcp/pyproject.toml.
|
||||
python3.pkgs.starlette
|
||||
python3.pkgs.uvicorn
|
||||
# An application is not importable by a dependent until this converts it,
|
||||
# and `hhstream_mcp.invoke` imports `hhstream.cli` directly.
|
||||
(python3.pkgs.toPythonModule hhstream)
|
||||
|
||||
@@ -29,13 +29,23 @@ requires-python = ">=3.11"
|
||||
# SDK: `server.py` calls `anyio.run` and `anyio.to_thread` directly, and a
|
||||
# dependency used by name belongs in the metadata.
|
||||
#
|
||||
# `starlette` and `uvicorn` are declared for exactly that reason and no other.
|
||||
# The SDK pulls both in today -- its streamable-HTTP transport needs them -- so
|
||||
# leaving them out would work, which is the trap: `http.py` imports
|
||||
# `uvicorn` and the SDK's `TransportSecuritySettings` sits on Starlette's
|
||||
# `Request`, and an SDK release that moved either behind an optional extra
|
||||
# would break `serve --http` with an ImportError naming someone else's package.
|
||||
# They are only needed by the HTTP transport, which is why they are imported
|
||||
# inside the functions that use them rather than at module scope: the stdio
|
||||
# path must neither pay for them nor fail on them.
|
||||
#
|
||||
# `hhstream` is NOT listed, and it is not optional either: this package imports
|
||||
# `hhstream.cli`, `hhstreamd.protocol` and `hhstream_session` and does not work
|
||||
# without them. It is absent because `hhstream` is not a wheel anyone installs
|
||||
# from an index -- it is this repository's other derivation, and `nix/mcp.nix`
|
||||
# is where the two are joined. A dependency naming a version nobody can
|
||||
# resolve would be worse than the comment you are reading.
|
||||
dependencies = ["mcp>=1.9", "anyio>=4"]
|
||||
dependencies = ["mcp>=1.9", "anyio>=4", "starlette>=0.40", "uvicorn>=0.30"]
|
||||
|
||||
[project.scripts]
|
||||
# Stdio only. One process per client, spawned by the client; no port, no daemon.
|
||||
|
||||
@@ -1,8 +1,10 @@
|
||||
"""The HTTP transport: every way a request is refused, and the one way it is not.
|
||||
|
||||
This file imports Starlette as well as the SDK, for the reason
|
||||
``test_mcp_server.py`` imports the SDK unconditionally: the thing under test is
|
||||
the wiring, and only a real ASGI stack can judge it. The store itself is
|
||||
This file imports Starlette as well as the SDK, **unconditionally**, for the
|
||||
reason ``test_mcp_server.py`` imports the SDK that way: the thing under test is
|
||||
the wiring, only a real ASGI stack can judge it, and a file that skipped when
|
||||
its dependency was missing would be green in precisely the build where that
|
||||
mattered. The store itself is
|
||||
standard library and tested without any of this
|
||||
(``test_mcp_authorizations.py``).
|
||||
|
||||
@@ -29,11 +31,12 @@ from hhstream_mcp import http as http_transport
|
||||
from hhstream_mcp import tools
|
||||
from hhstream_mcp.invoke import Runner
|
||||
|
||||
starlette_testclient = pytest.importorskip(
|
||||
"starlette.testclient",
|
||||
reason="the HTTP transport's own dependency; `nix develop` has it",
|
||||
)
|
||||
TestClient = starlette_testclient.TestClient
|
||||
# Imported unconditionally, and not through `importorskip`, for the reason
|
||||
# `test_mcp_server.py` imports the SDK unconditionally: a file that skipped
|
||||
# when its dependency was absent would be green in exactly the build where
|
||||
# that mattered. `checks/unit.nix` declares `starlette` and `httpx` so this
|
||||
# cannot become a skip, and proves them importable before pytest runs.
|
||||
from starlette.testclient import TestClient
|
||||
|
||||
#: The test client's Host, plus what a loopback deployment derives.
|
||||
TEST_HOSTS = ["testserver", *http_transport.default_allowed_hosts("127.0.0.1", 9989)]
|
||||
|
||||
Reference in new issue
Block a user