Fix histplot shrink with non-discrete bins (#2477)

The code for shifting the shrunken bars assumed that discrete binning
was in effect. This is probably the only situation where shrinking
really makes sense, but there was no prevention or warning of getting
an innacurate result when using it with continuous bins.

Fixes #2476

(cherry picked from commit 6183f1e)

doc/releases/v0.12.0.txt

@@ -29,6 +29,11 @@ v0.12.0 (Unreleased)
 - |Fix| In :func:`histplot` and :func:`kdeplot`, fixed a bug where the `multiple` was ignored when `hue` was provided as a vector without a name (:pr:`2462`).
 
  - |Defaults| In :func:`displot`, the default alpha value now adjusts to a provided `multiple` parameter even when `hue` is not assigned (:pr:`2462`).
+- |Fix| In :func:`histplot`, fixed a bug where using `shrink` with non-discrete bins shifted bar positions inaccurately (:pr:`2477`).
+
+- |Fix| In :func:`displot`, fixed a bug where `common_norm` was ignored when `kind="hist"` and faceting was used without assigning `hue` (:pr:`2468`).
+
+- |Defaults| In :func:`displot`, the default alpha value now adjusts to a provided `multiple` parameter even when `hue` is not assigned (:pr:`2462`).
 
 - Made `scipy` an optional dependency and added `pip install seaborn[all]` as a method for ensuring the availability of compatible `scipy` and `statsmodels` libraries at install time. This has a few minor implications for existing code, which are explained in the Github pull request (:pr:`2398`).
 

seaborn/distributions.py

@@ -446,9 +446,12 @@ def plot_univariate_histogram(
                 edges = np.power(10, edges)
 
             # Pack the histogram data and metadata together
+            orig_widths = np.diff(edges)
+            widths = shrink * orig_widths
+            edges = edges[:-1] + (1 - shrink) / 2 * orig_widths
             index = pd.MultiIndex.from_arrays([
-                pd.Index(edges[:-1], name="edges"),
-                pd.Index(np.diff(edges) * shrink, name="widths"),
+                pd.Index(edges, name="edges"),
+                pd.Index(widths, name="widths"),
             ])
             hist = pd.Series(heights, index=index, name="heights")
 
@@ -551,9 +554,8 @@ def plot_univariate_histogram(
                 # Use matplotlib bar plotting
 
                 plot_func = ax.bar if self.data_variable == "x" else ax.barh
-                move = .5 * (1 - shrink)
                 artists = plot_func(
-                    hist["edges"] + move,
+                    hist["edges"],
                     hist["heights"] - bottom,
                     hist["widths"],
                     bottom,

seaborn/tests/test_distributions.py

@@ -1520,10 +1520,21 @@ def test_weights_with_auto_bins(self, long_df):
 
     def test_shrink(self, long_df):
 
+        f, (ax1, ax2) = plt.subplots(2)
+
         bw = 2
-        shrink = .5
-        ax = histplot(long_df, x="x", binwidth=bw, shrink=shrink)
-        assert ax.patches[0].get_width() == bw * shrink
+        shrink = .4
+
+        histplot(long_df, x="x", binwidth=bw, ax=ax1)
+        histplot(long_df, x="x", binwidth=bw, shrink=shrink, ax=ax2)
+
+        for p1, p2 in zip(ax1.patches, ax2.patches):
+
+            w1, w2 = p1.get_width(), p2.get_width()
+            assert w2 == pytest.approx(shrink * w1)
+
+            x1, x2 = p1.get_x(), p2.get_x()
+            assert (x2 + w2 / 2) == pytest.approx(x1 + w1 / 2)
 
     def test_log_scale_explicit(self, rng):