test: add mask padding integration test

scripts/imaging/convolution.py

@@ -166,3 +166,112 @@
 residuals = via_fft - via_real_space
 
 print(f"Mapping Matrix Max residual = {residuals.max()}")
+
+
+"""
+__Mask Padding__
+
+When the mask is close to the image edge and the PSF kernel footprint extends
+beyond the boundary, the blurring mask is automatically padded. This test
+verifies that the padded convolution produces the same result as an equivalent
+centred configuration that requires no padding.
+
+We simulate a compact lens+source on two grids:
+  - A large 51x51 grid with the model centred → no padding needed.
+  - A small 21x21 grid with the model offset near the edge → padding triggered.
+
+Both should give identical log-likelihoods and chi_squared ≈ 0.
+"""
+
+import warnings
+
+pixel_scales = 0.2
+
+psf_pad = al.Convolver.from_gaussian(
+    shape_native=(11, 11), pixel_scales=pixel_scales, sigma=0.75, normalize=True
+)
+
+lens_centred = al.Galaxy(
+    redshift=0.5,
+    light=al.lp.Sersic(
+        centre=(0.0, 0.0), intensity=0.1,
+        effective_radius=0.3, sersic_index=2.0,
+    ),
+    mass=al.mp.Isothermal(centre=(0.0, 0.0), einstein_radius=1.0),
+)
+source_centred = al.Galaxy(
+    redshift=1.0,
+    light=al.lp.Exponential(
+        centre=(0.0, 0.0), intensity=0.3, effective_radius=0.2,
+    ),
+)
+tracer_centred = al.Tracer(galaxies=[lens_centred, source_centred])
+
+sim_pad = al.SimulatorImaging(
+    exposure_time=300.0, psf=psf_pad, add_poisson_noise_to_data=False,
+)
+
+# --- Centred fit on a large grid: no padding needed ---
+grid_large = al.Grid2D.uniform(
+    shape_native=(51, 51), pixel_scales=pixel_scales, over_sample_size=1,
+)
+dataset_centred = sim_pad.via_tracer_from(tracer=tracer_centred, grid=grid_large)
+dataset_centred.noise_map = al.Array2D.ones(
+    shape_native=(51, 51), pixel_scales=pixel_scales,
+)
+mask_centred = al.Mask2D.circular(
+    shape_native=(51, 51), pixel_scales=pixel_scales,
+    radius=0.6, centre=(0.0, 0.0),
+)
+masked_centred = dataset_centred.apply_mask(mask=mask_centred)
+fit_centred = al.FitImaging(dataset=masked_centred, tracer=tracer_centred)
+
+# --- Off-centre fit on a small grid: triggers padding ---
+offset = (0.0, 1.2)
+lens_off = al.Galaxy(
+    redshift=0.5,
+    light=al.lp.Sersic(
+        centre=offset, intensity=0.1,
+        effective_radius=0.3, sersic_index=2.0,
+    ),
+    mass=al.mp.Isothermal(centre=offset, einstein_radius=1.0),
+)
+source_off = al.Galaxy(
+    redshift=1.0,
+    light=al.lp.Exponential(
+        centre=offset, intensity=0.3, effective_radius=0.2,
+    ),
+)
+tracer_off = al.Tracer(galaxies=[lens_off, source_off])
+
+grid_small = al.Grid2D.uniform(
+    shape_native=(21, 21), pixel_scales=pixel_scales, over_sample_size=1,
+)
+dataset_off = sim_pad.via_tracer_from(tracer=tracer_off, grid=grid_small)
+dataset_off.noise_map = al.Array2D.ones(
+    shape_native=(21, 21), pixel_scales=pixel_scales,
+)
+mask_off = al.Mask2D.circular(
+    shape_native=(21, 21), pixel_scales=pixel_scales,
+    radius=0.6, centre=offset,
+)
+
+with warnings.catch_warnings(record=True) as w:
+    warnings.simplefilter("always")
+    masked_off = dataset_off.apply_mask(mask=mask_off)
+    fit_off = al.FitImaging(dataset=masked_off, tracer=tracer_off)
+    padding_occurred = any("Mask padded" in str(x.message) for x in w)
+
+assert padding_occurred, "Expected mask padding to be triggered for the off-centre mask"
+assert fit_centred.chi_squared < 1e-4, f"Centred chi_squared too large: {fit_centred.chi_squared}"
+assert fit_off.chi_squared < 1e-4, f"Off-centre chi_squared too large: {fit_off.chi_squared}"
+
+likelihood_diff = abs(fit_centred.log_likelihood - fit_off.log_likelihood)
+assert likelihood_diff < 1e-4, (
+    f"Padded and non-padded log-likelihoods differ by {likelihood_diff}"
+)
+
+print(f"\nMask padding test PASSED")
+print(f"  Centred  log_likelihood = {fit_centred.log_likelihood:.8f}  chi_squared = {fit_centred.chi_squared:.8f}")
+print(f"  Padded   log_likelihood = {fit_off.log_likelihood:.8f}  chi_squared = {fit_off.chi_squared:.8f}")
+print(f"  Difference = {likelihood_diff:.2e}")