Improve test coverage and coverage measurement

.github/workflows/test.yml

@@ -114,7 +114,7 @@ jobs:
 
       - name: 🏃🏻‍➡️ Test LKPY
         run: |
-          mise run test -- --coverage --accel-coverage -- --log-file=test.log
+          mise run test -- --coverage --accel-coverage -- --log-level=debug
 
       - name: 📤 Upload coverage to CodeCov
         uses: codecov/codecov-action@v7.0.0

.woodpecker/dataset-tests.yml

@@ -11,7 +11,7 @@ steps:
     image: codeberg.org/mdekstrand/woodpecker-mise:v1-f44-dev
     pull: true
     settings:
-      task: "ci:prepare ::: test -v --coverage -m realdata ::: test-cli tests/cli/test-data-convert.sh"
+      task: "ci:data-tests"
       mise-env: ci
 
   - name: upload code coverage

Cargo.toml

@@ -41,6 +41,9 @@ numpy = "^0.29"
 [dev-dependencies]
 ntest = "^0.9"
 
+[lints.rust]
+unexpected_cfgs = { level = "warn", check-cfg = ["cfg(coverage_nightly)"] }
+
 [profile.dev]
 opt-level = 2
 debug-assertions = true

conftest.py

@@ -16,6 +16,7 @@
 from hypothesis import settings
 from pytest import fixture, register_assert_rewrite, skip
 
+from lenskit.logging._accel import update_log_level  # ruff: ignore[import-private-name]
 from lenskit.parallel import ensure_parallel_init
 from lenskit.random import init_global_rng
 
@@ -105,3 +106,4 @@ def pytest_collection_modifyitems(items):
 
 settings.register_profile("default", deadline=1000)
 ensure_parallel_init()
+update_log_level()

mise/tasks/ci/data-tests.sh

@@ -0,0 +1,18 @@
+#!/bin/zsh
+#MISE description="Run data tests in CI"
+#MISE depends=["ci:prepare"]
+#USAGE flag "-v --verbose" help="Enable verbose logging."
+
+. "$MISE_PROJECT_ROOT/mise/task-functions.sh"
+
+echo-run mise run test -- -v --coverage -m realdata
+if (($?)); then
+    die "tests failed"
+fi
+
+echo-run mise run test-cli --coverage --cov-append tests/cli/test-data-convert.sh
+if (($?)); then
+    die "CLI tests failed"
+fi
+
+mise run coverage:export || die "coverage export failed"

mise/tasks/coverage/export.sh

@@ -0,0 +1,5 @@
+#!/usr/bin/env zsh
+#MISE description="Export test coverage data."
+#MISE wait_for=["test", "test-cli"]
+
+coverage xml

src/accel/arrow/lists.rs

@@ -28,9 +28,9 @@ impl ExtractListArray for LargeListArray {
         if let Some(arr) = any.downcast_ref::<LargeListArray>() {
             Some(arr.clone())
         } else if let Some(arr) = any.downcast_ref::<ListArray>() {
-            info!("converting type {}", arr.data_type());
+            debug!("converting type {}", arr.data_type());
             let (field, offsets, values, nulls) = arr.clone().into_parts();
-            info!("field: {}", field);
+            debug!("field: {}", field);
             // convert offsets into Int64
             let offsets: Vec<_> = offsets.iter().map(|o| *o as i64).collect();
             let offsets = ScalarBuffer::from(offsets);

src/accel/lib.rs

@@ -3,7 +3,7 @@
 // Copyright (C) 2023-2026 Drexel University.
 // Licensed under the MIT license, see LICENSE.md for details.
 // SPDX-License-Identifier: MIT
-
+#![cfg_attr(coverage_nightly, feature(coverage_attribute))]
 use pyo3::{exceptions::PyRuntimeError, prelude::*};
 
 mod als;

src/lenskit/_accel/data.pyi

@@ -37,8 +37,16 @@ def dense_cooc(
     *,
     diagonal: bool = True,
 ) -> AccelTask[np.ndarray[tuple[int, int], np.dtype[np.float32]]]: ...
-def scatter_array(dst: _A, idx: pa.Array, src: _A) -> _A: ...
-def scatter_array_empty(dst_size: int, idx: pa.Array, src: _A) -> _A: ...
+def scatter_array(dst: _A, idx: pa.Array, src: _A) -> _A:
+    """
+    Create a new array merging a base array with the scattered content of a second array.
+    """
+
+def scatter_array_empty(dst_size: int, idx: pa.Array, src: _A) -> _A:
+    """
+    Scatter array elements into a new, otherwise-empty array.
+    """
+
 def sample_negatives(
     coords: CoordinateTable,
     rows: np.ndarray[tuple[int], np.dtype[np.int32]],

src/lenskit/operations.py

@@ -54,7 +54,7 @@ def recommend(
     if items is not None and not isinstance(items, ItemList):
         items = ItemList(items)
     res = pipeline.run(node, query=query, n=n, items=items, _profile=profiler)
-    if not isinstance(res, ItemList):
+    if not isinstance(res, ItemList):  # pragma: nocover
         raise TypeError("recommender pipeline did not return an item list")
 
     return res
@@ -93,7 +93,7 @@ def score(
     if items is not None and not isinstance(items, ItemList):
         items = ItemList(items)
     res = pipeline.run(node, query=query, items=items, _profile=profiler)
-    if not isinstance(res, ItemList):
+    if not isinstance(res, ItemList):  # pragma: nocover
         raise TypeError("scorer pipeline did not return an item list")
 
     return res

src/lenskit/testing/__init__.py

@@ -18,7 +18,7 @@
 
 import hypothesis.strategies as st
 
-from ._arrays import coo_arrays, scored_lists, sparse_arrays, sparse_tensors
+from ._arrays import coo_arrays, scored_lists, sparse_arrays
 from ._components import BasicComponentTests, ScorerTests
 from ._movielens import (
     DemoRecs,
@@ -52,7 +52,6 @@
     "scored_lists",
     "set_env_var",
     "sparse_arrays",
-    "sparse_tensors",
 ]
 
 

src/lenskit/testing/_arrays.py

@@ -18,7 +18,6 @@
 from hypothesis import assume
 
 from lenskit.data import ItemList
-from lenskit.math.sparse import torch_sparse_from_scipy
 
 
 @st.composite
@@ -76,17 +75,6 @@ def sparse_arrays(draw, *, layout="csr", **kwargs):
             raise ValueError(f"invalid layout {layout}")
 
 
-@st.composite
-def sparse_tensors(draw, *, layout="csr", **kwargs):
-    if isinstance(layout, list):
-        layout = st.sampled_from(layout)
-    if isinstance(layout, st.SearchStrategy):
-        layout = draw(layout)
-
-    M: sps.coo_array = draw(coo_arrays(**kwargs))
-    return torch_sparse_from_scipy(M, layout)  # type: ignore
-
-
 @st.composite
 def scored_lists(
     draw: st.DrawFn,

src/lenskit/torch.py

@@ -27,25 +27,6 @@ def wrapper(*args, **kwargs):
     return wrapper
 
 
-def sparse_row(mat: torch.Tensor, row: int) -> torch.Tensor:
-    """
-    Get a row of a sparse (CSR) tensor.  This is needed because indexing a
-    tensor does not work in inference mode.
-    """
-
-    assert mat.is_sparse_csr
-
-    cri = mat.crow_indices()
-    sp = cri[row]
-    ep = cri[row + 1]
-
-    cs = mat.col_indices()
-    vs = mat.values()
-    return torch.sparse_coo_tensor(
-        indices=cs[sp:ep].reshape(1, -1), values=vs[sp:ep], size=mat.shape[1:]
-    )
-
-
 def safe_tensor(array) -> torch.Tensor:
     """
     Safely convert an array into a NumPy tensor.  This includes copying it to

tests/accel/test_accel_arrow.py

@@ -0,0 +1,14 @@
+"""
+Test Arrow utility functions in the accelerator.
+"""
+
+import numpy as np
+import pyarrow as pa
+
+from lenskit import _accel
+
+
+def test_array_type():
+    arr = pa.array(np.arange(10, dtype=np.int32))
+    t = _accel.arrow_type(arr)
+    assert t == "Int32"

tests/accel/test_argsort.py

@@ -10,7 +10,7 @@
 import hypothesis.extra.numpy as nph
 import hypothesis.strategies as st
 from hypothesis import given
-from pytest import mark
+from pytest import mark, raises
 
 from lenskit._accel import data
 
@@ -38,7 +38,7 @@ def test_sort_floats(arr):
 
 @given(
     nph.arrays(
-        nph.integer_dtypes(endianness="="),
+        st.one_of(nph.integer_dtypes(endianness="="), nph.unsigned_integer_dtypes(endianness="=")),
         nph.array_shapes(max_dims=1),
         elements={"allow_nan": False, "allow_infinity": False},
     )
@@ -209,3 +209,9 @@ def test_topn_any_float(arr, n):
         mask[np.isnan(arr)] = False
         nopes = arr[mask]
         assert np.all(nopes <= np.min(items))
+
+
+def test_topn_rejects_strings():
+    strings = pa.array(["a", "b", "c", "x", "9", "0", "3", "@"])
+    with raises(TypeError):
+        data.argtopn(strings, 5)

tests/accel/test_pyon.py

@@ -0,0 +1,59 @@
+from lenskit import _accel
+
+
+def test_parse_json():
+    json = """
+    {"name": "FOOBIE BLETCH", "tags": ["foo", "bar"]}
+    """
+    obj = _accel.data.pyon_loads(json)
+    assert obj == {"name": "FOOBIE BLETCH", "tags": ["foo", "bar"]}
+
+
+def test_parse_single_quotes():
+    json = """
+    {'name': 'FOOBIE BLETCH', 'tags': ['foo', 'bar'], 'count': 7, 'active': false, 'value': 1.0}
+    """
+    obj = _accel.data.pyon_loads(json)
+    assert obj == {
+        "name": "FOOBIE BLETCH",
+        "tags": ["foo", "bar"],
+        "count": 7,
+        "active": False,
+        "value": 1.0,
+    }
+
+
+def test_parse_escape_dquote():
+    json = r"""{"name": "foo\"bob"}"""
+    obj = _accel.data.pyon_loads(json)
+    assert obj == {"name": 'foo"bob'}
+
+
+def test_parse_escape_squote():
+    json = r"""{"name": 'pe\'taq'}"""
+    obj = _accel.data.pyon_loads(json)
+    assert obj == {"name": "pe'taq"}
+
+
+def test_parse_escape_tab():
+    json = r"""{"name": '\t'}"""
+    obj = _accel.data.pyon_loads(json)
+    assert obj == {"name": "\t"}
+
+
+def test_parse_escape_lf():
+    json = r"""{"name": '\n'}"""
+    obj = _accel.data.pyon_loads(json)
+    assert obj == {"name": "\n"}
+
+
+def test_parse_escape_cr():
+    json = r"""{"name": '\r'}"""
+    obj = _accel.data.pyon_loads(json)
+    assert obj == {"name": "\r"}
+
+
+def test_parse_escape_unicode():
+    json = r"""{"name": '\u2230'}"""
+    obj = _accel.data.pyon_loads(json)
+    assert obj == {"name": "\u2230"}

tests/accel/test_scatter.py

@@ -12,6 +12,7 @@
 import hypothesis.extra.numpy as nph
 import hypothesis.strategies as st
 from hypothesis import given
+from pytest import raises
 
 from lenskit._accel import data
 
@@ -21,6 +22,7 @@
     nph.arrays(
         st.one_of(
             nph.integer_dtypes(endianness="="),
+            nph.unsigned_integer_dtypes(endianness="="),
             nph.floating_dtypes(endianness="=", sizes=(16, 32, 64)),
         ),
         nph.array_shapes(max_dims=1),
@@ -81,3 +83,15 @@ def test_scatter_dst_size(hd: st.DataObject, size, idx_t: np.dtype):
     arr = arr_a.to_numpy(zero_copy_only=False)
 
     assert np.array_equal(arr[idx], src, equal_nan=True)
+
+
+def test_scatter_rejects_strings():
+    strings = pa.array(["a", "b", "c", "x", "9", "0", "3", "@"])
+    with raises(TypeError):
+        data.scatter_array(strings, pa.array([2, 7, 0]), strings)
+
+
+def test_scatter_empty_rejects_strings():
+    strings = pa.array(["a", "b", "c", "x", "9", "0", "3", "@"])
+    with raises(TypeError):
+        data.scatter_array_empty(100, pa.array([2, 7, 0]), strings)

tests/cli/test-data-convert.sh

@@ -12,3 +12,19 @@ if [[ -f data/australian_users_items.json.gz ]]; then
 else
     skip 3
 fi
+
+if [[ -f data/anonymous-msweb.data.gz ]]; then
+    run-lenskit data convert --ms-web data/anonymous-msweb.data.gz "$TEST_WORK/msweb"
+    require -d "$TEST_WORK/msweb"
+    require -f "$TEST_WORK/msweb/schema.json"
+else
+    skip 3
+fi
+
+if [[ -f data/az23/Video_Games.csv.gz ]]; then
+    run-lenskit data convert --amazon data/az23/Video_Games.csv.gz "$TEST_WORK/az23-games"
+    require -d "$TEST_WORK/az23-games"
+    require -f "$TEST_WORK/az23-games/schema.json"
+else
+    skip 3
+fi