COMPAT: Support for MPL 1.5 #11145
COMPAT: Support for MPL 1.5 #11145
Conversation
|
u can add to the channel list (I think I showed this above) it will thrn pick up the latest version if it's not specified exactly |
jreback
added
Visualization
| self.bp_n_objects = 7 | ||
| else: | ||
| self.bp_n_objects = 8 | ||
| # Thier dev version is messed up, 0+unknown | ||
| if mpl_version[0] == '0' or mpl_version >= LooseVersion('1.5'): |
There was a problem hiding this comment.
prob best to put these in some functions (either in util.testing or top-level here)
TomAugspurger
force-pushed
the
mpl15
branch
from
f5106d8 to
0708996
Compare
| @@ -214,7 +214,8 @@ | |||
|
|
|||
| def mpl_style_cb(key): | |||
| import sys | |||
| from pandas.tools.plotting import mpl_stylesheet | |||
| from pandas.tools.plotting import mpl_stylesheet, _update_stylesheet | |||
| _update_stylesheet() | |||
There was a problem hiding this comment.
This is pretty hacky, but mpl_stylesheet is a dict defined in pandas.tools.plotting. I have my matplotlib version check in pandas.util.testing, but I can't run code from that in the main body of pandas.tools.plotting (I'm assuming a circular import).
There was a problem hiding this comment.
Bah, this doesn't work since you can't set the option multiple times.
TomAugspurger
force-pushed
the
mpl15
branch
3 times, most recently
from
6afb353 to
0e62a2c
Compare
| try: # mpl optional | ||
| import pandas.tseries.converter as conv | ||
| conv.register() # needs to override so set_xlim works with str/number | ||
| except ImportError: | ||
| pass | ||
|
|
||
| def _mpl_ge_1_5_0(): | ||
| import matplotlib |
There was a problem hiding this comment.
seems you already have this in utils.testing?
|
is this not tested on travis? I don't see the icon |
| @@ -72,6 +72,7 @@ bash miniconda.sh -b -p $HOME/miniconda || exit 1 | |||
|
|
|||
| conda config --set always_yes yes --set changeps1 no || exit 1 | |||
| conda update -q conda || exit 1 | |||
| conda config --add channels tacaswell || exit 1 | |||
There was a problem hiding this comment.
the conda-forge channel might be better to use here
TomAugspurger
force-pushed
the
mpl15
branch
from
0e62a2c to
6c095ed
Compare
|
Travis was flaky yesterday morning. I just pushed some changes, and it's picked up now. |
TomAugspurger
force-pushed
the
mpl15
branch
3 times, most recently
from
f88a7d6 to
4557b07
Compare
|
@jreback travis is green. Looks like https://travis-ci.org/pydata/pandas/jobs/81429896 and https://travis-ci.org/pydata/pandas/jobs/81429897 both ran with matplotlib RC1. |
|
make the 3.4 regular build not specify a matplotlib (in requirements-3.4.run). All of the slow tests are running (e.g. 2.7) just fine. But I don't think the fast ones actually are run by travis (in theory they are on 3.3, but that is not build ATM for 1.5.0rc1). |
MPL 1.5 seems to have changed the behavior of ax.scatter(x, y, c='green', color=['blue']) Since this doesn't really make sense anyway(?) I'm changing our code so that the user doesn't get into that state.
TomAugspurger
force-pushed
the
mpl15
branch
from
4557b07 to
bdd51ec
Compare
|
@jreback done. We'll want to followup and pin it at 1.5 once that's released. I'm assuming the |
|
thanks @TomAugspurger |
|
@TomAugspurger I don't think we plan to update conda forge from master, just tags. |
closes #11111 (nice issue number)
Testing with tox locally I get 1 failure using either matplotlib 1.4 or 1.5. I may have messed up my tox config though. Still need to cleanup a bunch of stuff, I just threw in changes to get each test passing as I went.
release notes et. al coming tonight or tomorrow. Are we making any changes to the build matrix to run a matplotlib 1.5?