Try utf-8, the system’s default encoding, and latin1 when reading egg_info-files to avoid UnicodeDecodeErrors at the installation of a package.

1 files changed, 26 insertions, 3 deletions

diff --git a/pip/req.py b/pip/req.py
index bf6d66636..f05aeeb66 100644
--- a/pip/req.py
+++ b/pip/req.py
@@ -1,6 +1,7 @@
 from email.parser import FeedParser
 import os
 import imp
+import locale
 import pkg_resources
 import re
 import sys
@@ -35,6 +36,30 @@ import pip.wheel
 from pip.wheel import move_wheel_files, Wheel, wheel_ext
 
 
+def read_text_file(filename):
+    """Return the contents of *filename*.
+
+    Try to decode the file contents with utf-8, the preffered system encoding
+    (e.g., cp1252 on some Windows machines) and latin1, in that order. Decoding
+    a byte string with latin1 will never raise an error. In the worst case, the
+    returned string will contain some garbage characters.
+
+    """
+    with open(filename, 'rb') as fp:
+        data = fp.read()
+
+    encodings = ['utf-8', locale.getpreferredencoding(False), 'latin1']
+    for enc in encodings:
+        try:
+            data = data.decode(enc)
+        except UnicodeDecodeError:
+            continue
+        break
+
+    assert type(data) != bytes  # Latin1 should have worked.
+    return data
+
+
 class InstallRequirement(object):
 
     def __init__(self, req, comes_from, source_dir=None, editable=False,
@@ -322,9 +347,7 @@ exec(compile(getattr(tokenize, 'open', open)(__file__).read().replace('\\r\\n',
         filename = self.egg_info_path(filename)
         if not os.path.exists(filename):
             return None
-        fp = open(filename, 'r')
-        data = fp.read()
-        fp.close()
+        data = read_text_file(filename)
         return data
 
     def egg_info_path(self, filename):