From 3b65566b57d7575d8523cb3c2bd287e8baf7dfcc Mon Sep 17 00:00:00 2001
From: Robert Kern <rkern@enthought.com>
Date: Fri, 19 Sep 2014 10:07:21 +0100
Subject: [PATCH] BUG: Avoid a numpy crash seen when using a data source of a
 unicode array.

---
 CHANGES.txt                              |  1 +
 chaco/array_data_source.py               | 17 ++++++++++++++---
 chaco/tests/arraydatasource_test_case.py |  8 ++++++++
 3 files changed, 23 insertions(+), 3 deletions(-)

diff --git a/CHANGES.txt b/CHANGES.txt
index 8bd8083af..ab3e66bcd 100644
--- a/CHANGES.txt
+++ b/CHANGES.txt
@@ -21,6 +21,7 @@ Fixes
    PR#101 (PR#175).
  * Make the `alpha` attribute of scatter plots work as intended (PR#164).
  * Resume running test for empty container that has been skipped (PR#190).
+ * Avoid a numpy crash when using a data source of a unicode array (PR#213).
 
 Release 4.4.1
 -------------
diff --git a/chaco/array_data_source.py b/chaco/array_data_source.py
index 1905bcfd9..372f437cf 100644
--- a/chaco/array_data_source.py
+++ b/chaco/array_data_source.py
@@ -1,7 +1,8 @@
 """ Defines the ArrayDataSource class."""
 
 # Major library imports
-from numpy import array, isfinite, ones, nanargmin, nanargmax, ndarray
+from numpy import array, isfinite, ones, ndarray
+import numpy as np
 
 # Enthought library imports
 from traits.api import Any, Constant, Int, Tuple
@@ -19,7 +20,12 @@ def bounded_nanargmin(arr):
     # Different versions of numpy behave differently in the all-NaN case, so we
     # catch this condition in two different ways.
     try:
-        min = nanargmin(arr)
+        if np.issubdtype(arr.dtype, np.floating):
+            min = np.nanargmin(arr)
+        elif np.issubdtype(arr.dtype, np.number):
+            min = np.argmin(arr)
+        else:
+            min = 0
     except ValueError:
         return 0
     if isfinite(min):
@@ -33,7 +39,12 @@ def bounded_nanargmax(arr):
     If all NaNs, return -1.
     """
     try:
-        max = nanargmax(arr)
+        if np.issubdtype(arr.dtype, np.floating):
+            max = np.nanargmax(arr)
+        elif np.issubdtype(arr.dtype, np.number):
+            max = np.argmax(arr)
+        else:
+            max = -1
     except ValueError:
         return -1
     if isfinite(max):
diff --git a/chaco/tests/arraydatasource_test_case.py b/chaco/tests/arraydatasource_test_case.py
index da7f14dae..1ca889637 100644
--- a/chaco/tests/arraydatasource_test_case.py
+++ b/chaco/tests/arraydatasource_test_case.py
@@ -5,6 +5,8 @@
 import unittest
 
 from numpy import arange, array, allclose, empty, isnan, nan
+import numpy as np
+
 from chaco.api import ArrayDataSource, PointDataSource
 
 
@@ -51,6 +53,12 @@ def test_bounds_all_nans(self):
         self.assertTrue(isnan(bounds[0]))
         self.assertTrue(isnan(bounds[1]))
 
+    def test_bounds_non_numeric(self):
+        myarray = np.array([u'abc', u'foo', u'bar', u'def'], dtype=unicode)
+        sd = ArrayDataSource(myarray)
+        bounds = sd.get_bounds()
+        self.assertEqual(bounds, (u'abc', u'def'))
+
 
 class PointDataTestCase(unittest.TestCase):
     # Since PointData is mostly the same as ScalarData, the key things to