Looking at this very quickly (and trying to remember what I did a year ago), it looks like I copied the code a few lines above _get_rgba_face which turns 'none' -> None.

I think the fix is to add isRGBA=True to the set_foreground calls, give me a bit to test.

Just a caution, without inspecting the code in question: the general rule in mpl is that None means use the default, and "none" means no color, so when I see the latter being converted to the former, it looks suspect.

I had the same thought as @efiring here. I think the path of least change here is (which restores the use of 'none'), but as @tacaswell has looked at this more deeply and more recently than I, I hesitate to turn it into a PR until we've looked at possibly better alternatives:

diff --git a/lib/matplotlib/lines.py b/lib/matplotlib/lines.py
index 42df262..751b24b 100644
--- a/lib/matplotlib/lines.py
+++ b/lib/matplotlib/lines.py
@@ -536,7 +536,8 @@ class Line2D(Artist):

         ln_color_rgba = self._get_rgba_ln_color()
         gc.set_foreground(ln_color_rgba)
-        gc.set_alpha(ln_color_rgba[3])
+        if ln_color_rgba != 'none':
+            gc.set_alpha(ln_color_rgba[3])

         gc.set_antialiased(self._antialiased)
         gc.set_linewidth(self._linewidth)
@@ -1004,7 +1005,7 @@ class Line2D(Artist):
     def _get_rgb_face(self, alt=False):
         facecolor = self._get_markerfacecolor(alt=alt)
         if is_string_like(facecolor) and facecolor.lower() == 'none':
-            rgbFace = None
+            rgbFace = 'none'
         else:
             rgbFace = colorConverter.to_rgb(facecolor)
         return rgbFace
@@ -1012,7 +1013,7 @@ class Line2D(Artist):
     def _get_rgba_face(self, alt=False):
         facecolor = self._get_markerfacecolor(alt=alt)
         if is_string_like(facecolor) and facecolor.lower() == 'none':
-            rgbaFace = None
+            rgbaFace = 'none'
         else:
             rgbaFace = colorConverter.to_rgba(facecolor, self._alpha)
         return rgbaFace
@@ -1020,7 +1021,7 @@ class Line2D(Artist):
     def _get_rgba_ln_color(self, alt=False):
         ln_color = self._color
         if is_string_like(ln_color) and ln_color.lower() == 'none':
-            rgba_ln = None
+            rgba_ln = 'none'
         else:
             rgba_ln = colorConverter.to_rgba(ln_color, self._alpha)
         return rgba_ln

Your suggestion would make rgba_ln indexable, and would silently work in subsequent code that indexes what it thinks is a 3 or 4 element tuple.

I started down that path last night as well, but it generates a bunch of duplicate code (both in LoC and at run time) as the logic for turning 'none' into 'no-color' is done in colorConverter.to_rgba. If the isRGBA flag is False, then set_foreground calls the to_rgba on whatever is passed in, which in all the cases but 'none' is redundant. By having _get_rgba_ln_color always return an RGBA and the use of the flag, this can be short-circuited.

I am wary of changing the behavior in _get_rgba_face as they were like that when I added _get_rgba_ln_color (and committed some cargo-cult style programming). My guess is that some of face-coloring logic uses None to mean 'no color' someplace down-stream.

closing this as #2761 has been merged. @jakevdp Can you confirm that this fixes your problem?

I'm trying to confirm, but when I import matplotlib after re-installing it, I get

Library not loaded: libfreetype.6.dylib
  Referenced from: /Users/jakevdp/anaconda/envs/mpltest/lib/python2.7/site-packages/matplotlib-1.4.x-py2.7-macosx-10.5-x86_64.egg/matplotlib/ft2font.so
  Reason: image not found

It's probably my fault, but I don't have time to dig into this right now. If the test script above works for you, then it probably solves my problem.