👍 - looks good to me.

Ok, this looks VERY related to the PR you mentioned. You think this will be beneficial or does it interfere with the other patch?

Ok, this looks VERY related to the PR you mentioned.

This PR fixes a bug which #1185 does not. Namely: The figure dpi is reverted to 72 after using the MixedModeRasterizer. After that, there is an awful lot of overlap. I suggest that we focus on #1185 (with a keen eye from yourself) and once that is in, we look at implementing this fix.

How does that sound?

Just looking through the code, I don't think it will interfere with #1185 (or other backends).
@pwuertz, did you happen to check if the pdf output from the pgf backend is consistent with that from the pdf backend when rasterization is used?

The hard-coded dpi of 72 was because all the vector backend uses dpi of 72 internally. But, this seems not to be the case of pgf backend.

While the PR looks good to me, I will try some test sometime this weekends.

@leejjoon The test I included looks the same for backend_pdf and backend_pgf.

#2044 (which is a rebased #1185) is almost ready -- there's an unreproducible test failure there at the moment. But once that's done, maybe we could update this PR to be based on it? I'd like to get this in for 1.3.0 if at all possible.

Ok -- #2044 is in, so I think all that's left for you @pwuertz is to submit everything this does that it doesn't do. Thanks!

Rebased on new master, but a test for py2.6 failed, don't know if that is related to my PR. I can check that tomorrow.

Allright, the 2.6 test fail is not related to this PR, merging...

py2.6 now passes, but py3.3 throws an ImageComparisonFailure for matplotlib.tests.test_image.test_rasterize_dpi.test although the images seem to be identical. The diff image is completely black.