This is in the latest master as of 1184f78

Hmm... Can't reproduce.

Do you have anything customized in your matplotlibrc? Which backend is this?

I suspect you're on the MacOsx backend. First, try Agg instead. Second, try passing "snap=True" or "snap=False" to both of them. I suspect one or the other method is ignoring that flag.

I tried using the Agg backend (I guess it was using MacOS X) but that doesn't help. Setting snap=False in plot does help, though I don't quite get what you do. Setting snap in LineCollection has no effect.

Here's what I get with snap=False in plot:

Also, I made sure to remove my .matplotlib directory first to ensure no custom configuration.

Ok -- are you sure this is the Agg backend? What does

import matplotlib
matplotlib.get_backend()

give?

The full script is:

import matplotlib
matplotlib.use('Agg')
print(matplotlib.get_backend())

from matplotlib import pyplot as plt
from matplotlib.collections import LineCollection

x1, x2, y1, y2 = -6.87194525904,-6.87194525904,0.0446252911239,0.0478759487178

fig = plt.figure(figsize=(10,8))
ax = fig.add_subplot(1,1,1)

lc = LineCollection([([[x1, y1], [x2, y2]])], color='green')
ax.add_collection(lc)        
ax.plot([x1, x2], [y1, y2], color='red')

fig.savefig('offset.png')

and the output is:

$ python bug.py 
agg

I was finally able to reproduce this -- I had forgotten I had a custom dpi set for some other testing. I think the attached resolves the (long standing) issue.

Works perfectly - thanks!!

It would seem that this merge has had an impact on the behaviour of pcolor that becomes evident when a comparison between pcolor and pcolormesh is carried out. This can be seen in the image below, where 10 random values are plotted onto the same axes, using pcolor for the left column and pcolormesh for the right (and generated using the version of matplotlib as at the point where this commit was merged):

But setting snap=False on the pcolor call solves this:

However this problem does not show up in an earlier version of matplotlib (namely v1.2.1). Here the pcolor call has no snap keyword set yet the problem does not appear:

The code used to generate these three images is as below (but note that snap=False is added to the pcolor call for the second image):

import matplotlib
import matplotlib.pyplot as plt
import numpy as np
import numpy.random as rnd

MPL_VERSION = matplotlib.__version__

nx, ny = 1, 10
x0, x1 = [0, 1], [1, 2]
y = np.arange(ny+1)
X0, Y = np.meshgrid(x0, y)
X1, Y = np.meshgrid(x1, y)
rnd.seed(5)
z = rnd.randint(20, size=(ny, 1)).astype(np.float32)

fig = plt.figure(dpi=45)
pcr_ax = plt.subplot(111)
plt.pcolor(X0, Y, z)
plt.pcolormesh(X1, Y, z)
pcr_ax.grid(True)
pcr_ax.set_xlim([0, nx+1])
pcr_ax.set_ylim([0, ny])
plt.title('pcolor - pcolormesh (using MPL v{})'.format(MPL_VERSION))

plt.savefig('comparison_mpl_v{}_snap.png'.format(MPL_VERSION),
            dpi=45)
plt.show()

@dkillick: Is your proposal that we should be setting snap=False as default to pcolor then?

@pelson the above was designed just to highlight the problem, not to also recommend a solution.

#2726 turns of snap in pcolor, and it does revert your example to the old behavior.