BUG: fix autoscale_view with log scale and margins

LogLocator.view_limits was always clipping vmin to minpos;
now it sets vmin to minpos only if vmin <= 0.

Author: efiring

lib/matplotlib/tests/test_axes.py

@@ -4667,6 +4667,21 @@ def test_fillbetween_cycle():
         assert tuple(cc.get_edgecolors().squeeze()) == tuple(edge_target)
 
 
+@cleanup
+def test_log_margins():
+    plt.rcParams['axes.autolimit_mode'] = 'data'
+    fig, ax = plt.subplots()
+    margin = 0.05
+    ax.set_xmargin(margin)
+    ax.semilogx([1, 10], [1, 10])
+    xlim0, xlim1 = ax.get_xlim()
+    transform = ax.xaxis.get_transform()
+    xlim0t, xlim1t = transform.transform([xlim0, xlim1])
+    x0t, x1t = transform.transform([1, 10])
+    delta = (x1t - x0t) * margin
+    assert_allclose([xlim0t + delta, xlim1t - delta], [x0t, x1t])
+
+
 if __name__ == '__main__':
     import nose
     import sys

lib/matplotlib/ticker.py

@@ -1912,7 +1912,7 @@ def view_limits(self, vmin, vmax):
                 "Data has no positive values, and therefore can not be "
                 "log-scaled.")
 
-        if vmin <= minpos:
+        if vmin <= 0:
             vmin = minpos
 
         if rcParams['axes.autolimit_mode'] == 'round_numbers':