fix: adaptive tick significant figures so apply_extent ticks never collapse

_round_ticks adds significant figures (up to 10) until distinct ticks stay
distinct, so a milli-arcsecond extent far from zero no longer labels every
tick with the same value. Part of PyAutoLabs/PyAutoLens#789.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Author: Jammy2211

autoarray/plot/utils.py

@@ -1143,20 +1143,32 @@ def _inward_ticks(lo: float, hi: float, factor: float, n: int) -> np.ndarray:
     )
 
 
-def _round_ticks(values: np.ndarray, sig: int = 2) -> np.ndarray:
-    """Round *values* to *sig* significant figures.
+def _round_ticks(values: np.ndarray, sig: int = 2, max_sig: int = 10) -> np.ndarray:
+    """Round *values* to *sig* significant figures, adding figures until distinct ticks stay distinct.
+
+    Two significant figures suit an extent that spans its own magnitude (e.g. -2" to 2"), but on a
+    narrow extent far from zero (a few milli-arcseconds at 1.2") every tick rounds to the same value
+    and the labels collapse to one repeated number. The significant figures are therefore raised, up
+    to *max_sig*, until the rounded ticks are as distinct as the input ticks.
 
     After rounding, values smaller than 1e-10 of the overall tick scale are
     clamped to zero so that floating-point noise (e.g. 1e-16 centre ticks on
     symmetric extents) does not appear as scientific notation in labels.
     """
-    with np.errstate(divide="ignore", invalid="ignore"):
-        nonzero = np.where(values != 0, np.abs(values), 1.0)
-        mags = np.where(values != 0, 10 ** (sig - 1 - np.floor(np.log10(nonzero))), 1.0)
-    rounded = np.round(values * mags) / mags
-    scale = float(np.max(np.abs(rounded))) if len(rounded) > 0 else 1.0
-    if scale > 0:
-        rounded[np.abs(rounded) < scale * 1e-10] = 0.0
+    values = np.asarray(values, dtype=float)
+    n_distinct = len(np.unique(values))
+    for n_sig in range(sig, max(sig, max_sig) + 1):
+        with np.errstate(divide="ignore", invalid="ignore"):
+            nonzero = np.where(values != 0, np.abs(values), 1.0)
+            mags = np.where(
+                values != 0, 10 ** (n_sig - 1 - np.floor(np.log10(nonzero))), 1.0
+            )
+        rounded = np.round(values * mags) / mags
+        scale = float(np.max(np.abs(rounded))) if len(rounded) > 0 else 1.0
+        if scale > 0:
+            rounded[np.abs(rounded) < scale * 1e-10] = 0.0
+        if len(np.unique(rounded)) == n_distinct:
+            break
     return rounded
 
 

test_autoarray/plot/test_utils.py

@@ -619,3 +619,35 @@ def test__plot_regions__single_polygon_region_is_labelled_once():
         )
     finally:
         plt.close(figure)
+
+
+def test__round_ticks__two_significant_figures_by_default():
+    from autoarray.plot.utils import _round_ticks
+
+    assert list(_round_ticks(np.array([-1.875, 0.0, 1.875]))) == [-1.9, 0.0, 1.9]
+
+
+def test__round_ticks__adds_significant_figures_until_ticks_are_distinct():
+    from autoarray.plot.utils import _round_ticks
+
+    # A few milli-arcseconds at 1.2": two (and three) significant figures collapse every tick to 1.2.
+    ticks = _round_ticks(np.array([1.19655, 1.2, 1.20345]))
+
+    assert len(np.unique(ticks)) == 3
+    assert list(ticks) == [1.197, 1.2, 1.203]
+
+
+def test__apply_extent__narrow_extent_far_from_zero_has_distinct_tick_labels():
+    import matplotlib.pyplot as plt
+
+    from autoarray.plot.utils import apply_extent
+
+    fig, ax = plt.subplots()
+    apply_extent(ax, [1.1954, 1.2046, -0.8046, -0.7954])
+
+    xlabels = [label.get_text() for label in ax.get_xticklabels()]
+    ylabels = [label.get_text() for label in ax.get_yticklabels()]
+    plt.close(fig)
+
+    assert len(set(xlabels)) == len(xlabels) == 3
+    assert len(set(ylabels)) == len(ylabels) == 3