looks like a bugfix to me

attn @jkseppan as this names the PDF backend specifically

It seems to me that since the pdf backend doesn't define points_to_pixels, dpi_cor would be one and this would have no effect with that backend. This is @leejjoon's code and @NelleV added the FIXME comment, maybe they can comment on the pull request?

@jkseppan I am not 100% sure this the right fix, but it does fix my problem and doesn't brake tests.

Both backend_bases and backend_agg, which afaik renders the image even in the pdf backend, implement points_to_pixels.

To explain what it fixes: Have a look at
http://matplotlib.org/examples/pylab_examples/demo_annotation_box.html
The png, hires-png and pdf version look all different, which should not be the case
if dpi_cor is set to True.

I wonder if this is connected to #2831, #3364, #3918, #4375?

attn @mdboom, @WeatherGod

Hmm, quite possibly. I'll try out some of the SCCEs in those issues and
report back.

On Tue, Sep 22, 2015 at 8:08 AM, Jouni K. Seppänen <notifications@github.com

wrote:

I wonder if this is connected to #2831
#2831, #3364
#3364, #3918
#3918, #4375
#4375?

attn @mdboom https://github.com/mdboom, @WeatherGod
https://github.com/WeatherGod

—
Reply to this email directly or view it on GitHub
#5112 (comment)
.

none of the examples in those issues got fixed by this.

@tacaswell I don't see any good reason not to merge this and backport it to 2.x, assuming it fixes the blatant error in our gallery as @Tillsten notes above.

The only Travis failure is in the Xcode build:

Error: libpng-1.6.23 already installed
To install this version, first `brew unlink libpng`

Lots of package installations occurred after this, but then the build apparently stalled and timed out.

Back-ported to v2.x as 2fe9e32. Thank you, @Tillsten!