Numpy.array elementwise comparison to None very slow in backend_pdf.py

[Gerenuk]

When using the PDF backend with

fig.savefig(filename, format="PDF", bbox_inches="tight")

it may end up comparing None to a numpy.array which can be very slow if the array is large. This happens at line 2264 in backend_pdf.py in GraphicsContextPdf.delta() with the line

different = bool(ours != theirs)

It happens that ours is None and theirs a numpy.array. Due to the (new?) numpy behaviour using element-wise comparison here, this steps turns out to be very slow.
(matplotlib 1.4.0, numpy 1.9.0)

[jenshnielsen]

I think this has been fixed on master already by #3718 but perhaps it is worth back porting it to 1.4.3?

[tacaswell]

Yeah, I agree that it should be back-ported, doing it now.

[tacaswell]

back ported as 431edb7