Created attachment 1248855 [details]
request_iter_lines_test.py

The attached test should not report any errors if run as:

$ python3 -m unittest request_iter_lines_test.py

but instead, depending on where the delimiter is in the
input, it will incorrectly add an empty entry.

  Failures can be corrected by this quick/test patch, for
example, on fedora 25 python 3.5 this quick patch fixes all
failures in the test case when there is a delimiter:
"""
--- /usr/lib/python3.5/site-packages/requests/models.py.orig	2017-02-01 16:04:02.117318286 -0200
+++ /usr/lib/python3.5/site-packages/requests/models.py	2017-02-01 16:53:15.219410425 -0200
@@ -707,16 +707,29 @@
 
             if pending is not None:
                 chunk = pending + chunk
+                pending = None
 
             if delimiter:
+                wrap = False
+                for i in range(1, len(delimiter) + 1):
+                    if chunk.endswith(delimiter[:i]):
+                        wrap = True
+                        break
+                if wrap:
+                    if pending is not None:
+                        pending = pending + chunk
+                    else:
+                        pending = chunk
+                    continue
                 lines = chunk.split(delimiter)
             else:
                 lines = chunk.splitlines()
 
             if lines and lines[-1] and chunk and lines[-1][-1] == chunk[-1]:
-                pending = lines.pop()
-            else:
-                pending = None
+                if pending is not None:
+                    pending = pending + lines.pop()
+                else:
+                    pending = lines.pop()
 
             for line in lines:
                 yield line
"""

but the test case fails when a delimiter is not specified, lines are
separated by '\r\n' and the text chunk ends in the middle of the
sequence, that is, ends in '\r' and next chunk starts with '\n'.
So, chunk.splitlines() possibly needs to override TextIOWrapper
logic in this condition.