Handle empty dataframes in downsample_heatmap to prevent scipy crash

.github/workflows/unit-tests.yml

@@ -0,0 +1,37 @@
+name: Unit Tests
+
+# Runs the Python unit tests in tests/ (pytest). This is intentionally
+# separate from build-and-test.yml, whose "test" jobs are container/k8s
+# deployment smoke tests rather than pytest.
+
+on:
+  push:
+    branches: [develop]
+  pull_request:
+  workflow_dispatch:
+
+jobs:
+  pytest:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          # Match the runtime built in the Dockerfile (python=3.11).
+          python-version: "3.11"
+          cache: pip
+          cache-dependency-path: requirements.txt
+
+      - name: Install dependencies
+        run: |
+          python -m pip install --upgrade pip
+          # Pinned runtime deps (pyopenms is needed so ParameterManager imports
+          # cleanly at collection time) plus test-only deps. fakeredis backs the
+          # QueueManager/WorkflowManager tests, which pytest.importorskip it.
+          pip install -r requirements.txt
+          pip install pytest fakeredis
+
+      - name: Run unit tests
+        run: pytest tests/ -v

src/render/compression.py

@@ -51,6 +51,13 @@ def downsample_heatmap(data, max_datapoints=20000, rt_bins=400, mz_bins=50, logg
 
     # We need to collect here because scipy requires numpy arrays
     sorted_data = sorted_data.collect()
+
+    # scipy's binned_statistic_2d reduces over the inputs to derive bin edges
+    # and raises "zero-size array to reduction operation minimum" on empty
+    # input. With no peaks there is nothing to downsample, so return the input
+    # unchanged (same schema, zero rows).
+    if sorted_data.is_empty():
+        return data
 
     # Count peaks
     total_count = sorted_data.select(pl.count()).item()

tests/test_render_compression.py

@@ -0,0 +1,50 @@
+"""
+Tests for downsample_heatmap, the heatmap point-reduction helper.
+
+The deconvolved-heatmap panels are built per MS level, and a level with no
+peaks (e.g. an MS1-only run that still shows an MS2 panel) yields an empty
+frame. That empty frame used to reach scipy's binned_statistic_2d, which
+raises "zero-size array to reduction operation minimum which has no identity"
+while computing bin edges. downsample_heatmap must short-circuit on empty
+input and return an empty, schema-preserving frame instead of crashing.
+
+The helper is pure polars/numpy/scipy (no Streamlit), so it is unit-testable
+without booting the app.
+"""
+
+import os
+import sys
+
+sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+
+import polars as pl
+
+from src.render.compression import downsample_heatmap
+
+
+HEATMAP_SCHEMA = {"mass": pl.Float64, "rt": pl.Float64, "intensity": pl.Float64}
+
+
+def test_empty_dataframe_returns_empty_with_schema():
+    result = downsample_heatmap(pl.DataFrame(schema=HEATMAP_SCHEMA)).collect()
+    assert result.is_empty()
+    assert set(result.columns) >= {"mass", "rt", "intensity"}
+
+
+def test_empty_lazyframe_returns_empty_with_schema():
+    result = downsample_heatmap(pl.LazyFrame(schema=HEATMAP_SCHEMA)).collect()
+    assert result.is_empty()
+    assert set(result.columns) >= {"mass", "rt", "intensity"}
+
+
+def test_nonempty_input_passes_through_binning():
+    df = pl.DataFrame(
+        {
+            "mass": [100.0, 200.0, 300.0],
+            "rt": [1.0, 2.0, 3.0],
+            "intensity": [10.0, 20.0, 30.0],
+        }
+    )
+    result = downsample_heatmap(df).collect()
+    assert set(result.columns) >= {"mass", "rt", "intensity"}
+    assert result.height <= df.height