Add test for LogFormatter with very big numbers

doronbehar · doronbehar · commit 4f8dc9210b6c · 2024-10-30T18:28:50.000+02:00
Reported at #27609 . Values < 1e300 that trigger this test failure might be found, but it should be a bit better not to be on the edge of this issue.
diff --git a/lib/matplotlib/tests/test_ticker.py b/lib/matplotlib/tests/test_ticker.py
@@ -1843,6 +1843,27 @@ def test_small_range_loglocator(numticks):
         assert (np.diff(np.log10(ll.tick_values(6, 150))) == 1).all()
 
 
+# https://github.com/matplotlib/matplotlib/pull/27609
+# TODO: This test currently fails, as expected and it needs to be fixed...
+# To do this completely correctly, we should figure out what are our limits
+# for when do integers are interpreted eventually as infinities and why,
+# and perhaps always work with np.float128 to increase accuracy as much as
+# possible. Eitherway, we should document what is our limit and how it is
+# related to the accuracy of np.float{64,128} and Python's native float
+# which seems to be the same as np.float64 for this purpose...
+def test_LogFormatter_almost_inf():
+    fig, ax = plt.subplots()
+    # TODO: Figure out why 1e400 won't fail, but will make the ax.plot describe
+    # 1e400 as inf (and hence not print it, and hence not fail).
+    ax.plot([1,2], [1, 1e300])
+    ax.set_yscale("log")
+    fig.draw_without_rendering()
+    almost_inf = ax.get_lines()[0].get_ydata()[1]
+    # TODO/WIP: remove of course when the above TODOs are fixed. Perhaps assert
+    # something with ydata...
+    assert True == False
+
+
 def test_NullFormatter():
     formatter = mticker.NullFormatter()
     assert formatter(1.0) == ''
