Finish release warning cleanup (5.88.1)

CHANGELOG.md

@@ -7,6 +7,15 @@ between published tags; older pre-5.0 notes are retained below.
 
 For current interfaces, see the [consumer guide](docs/consumer-guide.md).
 
+## 5.88.1
+
+- Prefer the current predictor DataFrame API for cache fallbacks, retaining
+  compatibility with predictors that only implement the legacy API (#432).
+- Require GTFParse's released Polars deprecation fix and mhctools' predictor
+  output handle cleanup. Close a test fixture's input file explicitly.
+- Assert expected scientific diagnostics in tests and fail the suite on any
+  unexpected warning.
+
 ## 5.88.0
 
 - Import native Exacto peptide variants, primary structures and translations,

pyproject.toml

@@ -49,7 +49,7 @@ ignore = ["F821", "E501", "F841", "E731", "E741", "E722", "E721"]
 "test*/*" = ["F401"]
 
 [tool.pytest.ini_options]
-filterwarnings = ["error::FutureWarning"]
+filterwarnings = ["error"]
 markers = [
     "isovar: exercises the installed Isovar optional integration",
     "pirlygenes: exercises the installed PirlyGenes optional integration",

requirements.txt

@@ -1,8 +1,8 @@
 numpy>=2.0.0,<3.0.0
 pandas>=2.0.0
-mhctools>=3.44.26
+mhctools>=3.45.1
 varcode>=4.18.0
-gtfparse>=2.7.0,<3.0.0
+gtfparse>=2.8.1,<3.0.0
 mhcgnomes>=3.4.0
 pyensembl>=2.3.0,<3.0.0
 # PEP 440 versions and validated package names for release preflight

tests/test_cached_predictor.py

@@ -338,14 +338,36 @@ def _stamp(self, df):
             df["kind"] = "pMHC_affinity"
         return df
 
-    def predict_peptides_dataframe(self, peptides):
-        return self._stamp(super().predict_peptides_dataframe(peptides))
+    def predict_dataframe(self, peptides, sample_name="", n_flanks=None, c_flanks=None):
+        return self._stamp(super().predict_dataframe(
+            peptides, sample_name=sample_name, n_flanks=n_flanks, c_flanks=c_flanks))
 
     def predict_proteins_dataframe(self, name_to_seq):
         return self._stamp(super().predict_proteins_dataframe(name_to_seq))
 
 
 class TestFallback:
+    def test_legacy_only_fallback_still_fills_misses(self):
+        from types import SimpleNamespace
+
+        predictor = _matched_fallback(name="random", version="1.0")
+        fallback = SimpleNamespace(
+            predict_peptides_dataframe=predictor.predict_dataframe,
+            alleles=predictor.alleles,
+        )
+        cache = CachedPredictor.from_dataframe(_df([_row()]), fallback=fallback)
+        assert cache.predict_dataframe(["GILGFVFTL"]).peptide.tolist() == ["GILGFVFTL"]
+
+    def test_current_fallback_api_takes_precedence(self, monkeypatch):
+        fallback = _matched_fallback(name="random", version="1.0")
+
+        def deprecated(*args, **kwargs):
+            pytest.fail("Cache called the deprecated fallback API")
+
+        monkeypatch.setattr(fallback, "predict_peptides_dataframe", deprecated)
+        cache = CachedPredictor.from_dataframe(_df([_row()]), fallback=fallback)
+        assert cache.predict_dataframe(["GILGFVFTL"]).peptide.tolist() == ["GILGFVFTL"]
+
     def test_fallback_fills_miss(self):
         cache = CachedPredictor.from_dataframe(
             _df([_row()]),

tests/test_consumer_workflows.py

@@ -2866,11 +2866,21 @@ def table(value, method, version):
         sources = {"only": table(50., "original", "")}
     else:
         sources = {"only": table(50., "original", None).drop(columns="predictor_version")}
-    combined = combine_sources(sources, sample_name="p")
+    if case == "versioned":
+        with pytest.warns(UserWarning) as diagnostics:
+            combined = combine_sources(sources, sample_name="p")
+        assert [str(warning.message) for warning in diagnostics] == [
+            "Model 'original' has conflicting versions: '4.1b' vs '4.2'",
+            "Model 'original' has conflicting versions: '4.1b' vs ''",
+        ]
+        with pytest.warns(UserWarning, match="Multiple predictor versions"):
+            wide = combined.to_wide()
+    else:
+        combined = combine_sources(sources, sample_name="p")
+        wide = combined.to_wide()
     policy = dict(ascending=True, strata=["source_label"])
     before = rank_candidates(combined, "affinity.value", **policy)
     for suffix, writer, method, reader in DELIMITED_IO_TWINS:
-        wide = combined.to_wide()
         path = tmp_path / f"combined.{suffix}"
         if writer_style == "result":
             method(wide, path)

tests/test_io_pvacseq.py

@@ -35,7 +35,8 @@
 
 
 def _data_row_count(path):
-    return sum(1 for _ in open(path)) - 1
+    with open(path) as handle:
+        return sum(1 for _ in handle) - 1
 
 
 # ---------------------------------------------------------------------------

tests/test_peptide_view.py

@@ -513,10 +513,11 @@ def test_default_methods_settles_the_dependence():
     """An explicit default names one model; that model's mode applies."""
     df = _two_model_presentation_df()
 
-    scores = evaluate_scores(
-        df, peptide_view(Presentation.score), kind_support=_SPLIT_SUPPORT,
-        default_methods={"pMHC_presentation": "netmhcpan"},
-    )
+    with pytest.warns(UserWarning, match="not a joint multi-allele aggregate"):
+        scores = evaluate_scores(
+            df, peptide_view(Presentation.score), kind_support=_SPLIT_SUPPORT,
+            default_methods={"pMHC_presentation": "netmhcpan"},
+        )
 
     # netmhcpan is single_allele: best of 0.9 / 0.4 across the peptide.
     assert scores.tolist()[:2] == [0.9, 0.9]
@@ -526,10 +527,11 @@ def test_filter_auto_aggregation_binds_the_method_for_dependence():
     """apply_filter binds one method per iteration; the resolver must see it."""
     df = _two_model_presentation_df()
 
-    kept = apply_filter(
-        df, parse("peptide_view(el.score) >= 0.5"),
-        kind_support=_SPLIT_SUPPORT,
-    )
+    with pytest.warns(UserWarning, match="not a joint multi-allele aggregate"):
+        kept = apply_filter(
+            df, parse("peptide_view(el.score) >= 0.5"),
+            kind_support=_SPLIT_SUPPORT,
+        )
 
     assert len(kept) > 0
 

tests/test_ranking.py

@@ -863,7 +863,8 @@ def test_default_methods_processing_synonyms(key):
     ])
     from topiary.ranking import Processing
     ctx = EvalContext(df, default_methods={key: "mhcflurry"})
-    assert Processing.score.eval(ctx).iloc[0] == 0.75
+    with pytest.warns(UserWarning, match="peptide-level.*rows name alleles"):
+        assert Processing.score.eval(ctx).iloc[0] == 0.75
 
 
 def test_default_methods_accepts_multiple_synonyms_in_one_dict():
@@ -3393,7 +3394,8 @@ def test_a_filter_on_an_unpopulated_reserved_scope_keeps_nothing():
     from topiary import apply_filter
 
     df = _make_df(_UNSCOPED_ROW)
-    assert len(apply_filter(df, parse("shuffled.affinity < 500"))) == 0
+    with pytest.warns(UserWarning, match="which this frame does not have"):
+        assert len(apply_filter(df, parse("shuffled.affinity < 500"))) == 0
     # The same expression keeps the row once the producer supplies the column.
     supplied = _make_df([{**_UNSCOPED_ROW[0], "shuffled_value": 400.0}])
     assert len(apply_filter(supplied, parse("shuffled.affinity < 500"))) == 1

topiary/__init__.py

@@ -191,7 +191,7 @@
     encode_amino_acids,
 )
 
-__version__ = "5.88.0"
+__version__ = "5.88.1"
 
 __all__ = [
     "normalize_isovar_rna_support",

topiary/cached.py

@@ -877,10 +877,10 @@ def _fallback_resolve(self, peptides) -> None:
                 f"no fallback set.  Missed peptides: {missed_preview}{extra}."
             )
 
-        if hasattr(self.fallback, "predict_peptides_dataframe"):
-            fb_df = self.fallback.predict_peptides_dataframe(peptides)
-        elif hasattr(self.fallback, "predict_dataframe"):
+        if hasattr(self.fallback, "predict_dataframe"):
             fb_df = self.fallback.predict_dataframe(peptides)
+        elif hasattr(self.fallback, "predict_peptides_dataframe"):
+            fb_df = self.fallback.predict_peptides_dataframe(peptides)
         else:
             raise TypeError(
                 f"fallback does not implement predict_peptides_dataframe "