_⚠️ Potential issue_ | _🟡 Minor_

[FBumann]

⚠️ Potential issue | 🟡 Minor

🧩 Analysis chain

🏁 Script executed:

#!/bin/bash
# Verify numpy behavior with np.isnan on integer arrays
python3 -c "
import numpy as np
try:
    result = np.isnan(np.array([1, 2, 3]))
    print(f'np.isnan on int64: {result}')
except TypeError as e:
    print(f'np.isnan on int64 raises: {e}')
try:
    result = np.isnan(np.array([1.0, 2.0, float('nan')]))
    print(f'np.isnan on float64: {result}')
except TypeError as e:
    print(f'np.isnan on float64 raises: {e}')
"

Repository: FBumann/linopy

Length of output: 176

---

🏁 Script executed:

# Check the as_dataarray function
rg -n "def as_dataarray" linopy/ -A 20

Repository: FBumann/linopy

Length of output: 1141

---

🏁 Script executed:

# Get context around lines 565-576 in linopy/model.py
head -580 linopy/model.py | tail -30

Repository: FBumann/linopy

Length of output: 1160

---

🏁 Script executed:

# Check if there are any type hints or validation for lower/upper parameters
rg -n "lower.*upper" linopy/model.py | head -20

Repository: FBumann/linopy

Length of output: 595

---

🏁 Script executed:

# Get the full as_dataarray implementation
sed -n '232,280p' linopy/common.py

Repository: FBumann/linopy

Length of output: 1722

---

🏁 Script executed:

# Find the method containing lines 565-576 in linopy/model.py
sed -n '500,580p' linopy/model.py | head -90

Repository: FBumann/linopy

Length of output: 2679

---

🏁 Script executed:

# Look for the complete function signature with type hints
sed -n '450,520p' linopy/model.py

Repository: FBumann/linopy

Length of output: 2707

---

🏁 Script executed:

# Find numpy_to_dataarray function
rg -n "def numpy_to_dataarray" linopy/ -A 20

Repository: FBumann/linopy

Length of output: 1176

---

🏁 Script executed:

# Check if there are any tests for integer bounds
rg -n "lower.*int\|upper.*int" linopy/tests/ -i | head -20

Repository: FBumann/linopy

Length of output: 111

---

🏁 Script executed:

# Search for tests that use integer bounds
rg -n "add_variables.*lower=\d\|add_variables.*upper=\d" linopy/tests/ | head -20

Repository: FBumann/linopy

Length of output: 111

---

🏁 Script executed:

# Get full numpy_to_dataarray implementation
sed -n '187,230p' linopy/common.py

Repository: FBumann/linopy

Length of output: 1500

---

🏁 Script executed:

# Find test files
find . -name "*test*.py" -type f | head -10

Repository: FBumann/linopy

Length of output: 340

---

🏁 Script executed:

# Check for any tests involving bounds and data types
find . -type f -name "*.py" -exec grep -l "add_variables.*lower.*int\|dtype" {} \; | head -10

Repository: FBumann/linopy

Length of output: 554

---

🏁 Script executed:

# Check test files for integer bounds usage
grep -r "add_variables" ./test/*.py | grep -E "lower.*[0-9]|upper.*[0-9]" | head -15

Repository: FBumann/linopy

Length of output: 1449

---

🏁 Script executed:

# Look for any NaN or isnan handling in tests
grep -r "isnan\|NaN" ./test/*.py | head -10

Repository: FBumann/linopy

Length of output: 1078

---

🏁 Script executed:

# Check the benchmark file that mentions auto_mask
cat ./benchmark/benchmark_auto_mask.py

Repository: FBumann/linopy

Length of output: 21270

---

🏁 Script executed:

# Check for any type hints on lower/upper parameters
grep -A 5 "def add_variables" linopy/model.py | head -20

Repository: FBumann/linopy

Length of output: 263

---

🏁 Script executed:

# Look for any documentation about bounds types
rg "lower.*bound\|upper.*bound" linopy/model.py | head -10

Repository: FBumann/linopy

Length of output: 40

---

🏁 Script executed:

# Check what happens when integer scalars are passed to xarray DataArray
rg -A 5 "class DataArray" linopy/ | head -20

Repository: FBumann/linopy

Length of output: 40

---

🏁 Script executed:

# Check how xarray DataArray handles integer inputs
cat > /tmp/test_dtype.py << 'EOF'
import sys
sys.path.insert(0, '/repo')

# Test with xarray directly (avoiding numpy import issues)
try:
    import xarray as xr
    import numpy as np
    
    # Test 1: Python int scalar
    da1 = xr.DataArray(5)
    print(f"DataArray from Python int 5: dtype={da1.dtype}")
    
    # Test 2: Python int list
    da2 = xr.DataArray([1, 2, 3])
    print(f"DataArray from Python list [1,2,3]: dtype={da2.dtype}")
    
except Exception as e:
    print(f"Import error: {e}")
    print("Will check source code instead")
EOF
python3 /tmp/test_dtype.py 2>&1 || echo "Test skipped - will use code analysis"

Repository: FBumann/linopy

Length of output: 127

---

🏁 Script executed:

# Search for any existing safeguards against NaN on integer arrays
rg -B 3 -A 3 "isnan\|isnull" linopy/model.py

Repository: FBumann/linopy

Length of output: 40

---

🏁 Script executed:

# Check if there are any actual tests that pass integer numpy arrays as bounds
rg "add_variables.*lower.*np\.array|upper.*np\.array" ./test/

Repository: FBumann/linopy

Length of output: 40

---

🏁 Script executed:

# Look for the type annotation imports to understand if there's type checking
head -50 linopy/model.py

Repository: FBumann/linopy

Length of output: 1229

---

Use pd.isnull() instead of np.isnan() for integer-safe NaN detection.

If a user passes integer-typed bounds (e.g., np.array([0, 1, 2])), np.isnan() raises TypeError since integer dtypes don't support NaN. The as_dataarray call preserves input dtypes, so integer arrays pass through unchanged. Replace np.isnan() with pd.isnull(), which handles all numeric dtypes safely:

auto_mask_values = ~pd.isnull(data.lower.values) & ~pd.isnull(data.upper.values)

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@linopy/model.py` around lines 565 - 576, The auto-masking uses np.isnan which
fails for integer-typed bounds; update the logic in the block handling
self.auto_mask (around auto_mask_values / auto_mask_arr) to use pandas
null-checking: replace uses of np.isnan(data.lower.values) and
np.isnan(data.upper.values) with pd.isnull(...) so integer dtypes are handled
safely, and ensure pandas (pd) is imported where model.py defines this code;
keep the rest of the mask combination logic with mask & auto_mask_arr / mask =
auto_mask_arr unchanged.

Originally posted by @coderabbitai[bot] in FBumann#22 (comment)