Repository navigation
Different line styles between PNG and PDF exports. #6588
Description
Is the line style supposed to depend on the export backend?
MPL 1.5.1 is from Anaconda and MPL 2.x is from Github, both on Linux (CentOS 7)
With MPL 1.5.1, I get the same line styles between both PDF and PNG exports:
- PNG @ 100 DPI:
- PDF @ None DPI: MPL_ver_1.5.1_None.pdf
while with master version, they are different between PDF and PNG exports:
- PNG @ 100 DPI:
- PDF @ None DPI: MPL_ver_2.0.0b1.post1628+g1aba7c3_None.pdf
NB: the problem seems to be independent from the DPI value: I get similar line styles between both PNG exports with 100 DPI and 300 DPI.
- PNG @ 300 DPI:
The example snippet is
import matplotlib
import matplotlib.pyplot as plt
import numpy as np
test_case = matplotlib.__version__
fig, ax = plt.subplots(num=test_case, figsize=(4.8, 3.2))
ax.set_title("MPL version: {0:s}".format(test_case))
xx = np.array([0.0, 1.0])
ax.plot(xx, 1 * xx, ls='dashed', label='default lw')
ax.plot(xx, 2 * xx, ls='dashed', lw=2, label='lw=2')
ax.plot(xx, 3 * xx, ls='dashed', lw=3, label='lw=3')
ax.legend(loc='upper left')
plt.show()
for ext, dpi in (('pdf', None), ('png', 100), ('png', 300)):
fig.savefig("MPL_ver_{v:s}_{d}.{e:s}".format(v=test_case, d=dpi, e=ext),
dpi=dpi)My guess would that I may be somehow related to the new behavior introduced in #5926, but I didn't have time to have a look yet.
Activity
tacaswell commentedon Jun 15, 2016
That is also my guess, it looks like the scaling is not being passed through to the pdf backend correctly.
attn @jkseppan
afvincent commentedon Jun 15, 2016
I tested the same example with exports in EPS and SVG, and they look fine (i.e. like the PNG ones) even with MPL from the 2.x master branch. So it seems that the problem is indeed with the PDF backend.
jkseppan commentedon Jun 15, 2016
It seems #5926 changed the interface of GraphicsContext toward subclasses in a way that's a bit unfortunate for how the pdf backend used it
Implement dash pattern scaling in pdf
efiring commentedon Jun 15, 2016
This problem also shows up in interpolation_none_vs_nearest.py in the gallery, with #6584 applied (and presumably without it). The automatic selection of the number of ticks works for the agg output but not for the pdf.
efiring commentedon Jun 15, 2016
I will make a separate issue for the tick number problem.
FIX: do dash scaling early
FIX: do dash scaling early