ENH: Support setting the backend per-test. (Fixes #35) #38
Conversation
dopplershift
force-pushed
the
set-backend
branch
from
54b9ab2 to
09757a7
Compare
| baseline_remote = baseline_dir.startswith('http') | ||
|
|
||
| with plt.style.context(style): | ||
| with plt.style.context(style), switch_backend(backend): |
There was a problem hiding this comment.
@dopplershift looks like this is causing issues with Python 2.6. Does Matplotlib still support Python 2.6?
There was a problem hiding this comment.
Only on the 1.4 series--so I guess the question is when does pytest-mpl drop 1.4?
There was a problem hiding this comment.
Now 2.0 is out, I'd be fine with dropping 1.4 (I normally like keeping compatibility with the latest two versions)
dopplershift
force-pushed
the
set-backend
branch
from
09757a7 to
36f9890
Compare
|
I've gone ahead and removed all traces of Python 2.6 and Matplotlib 1.4--I'm hoping that's what you were looking for. Rebased on latest master. |
dopplershift
force-pushed
the
set-backend
branch
2 times, most recently
from
cb9ff8e to
f4a588e
Compare
|
This isn't introducing any new test failures. Apparently, something's not quite right with testing against matplotlib 2.0. |
|
@dopplershift - can you rebase on the latest master? (I've merged in changes that should fix Travis) |
dopplershift
force-pushed
the
set-backend
branch
from
f4a588e to
63caacc
Compare
|
Linux tests are failing because conda can't find packages for numpy 1.9 and matplotlib 1.5. Thoughts? |
|
Anything more you'd like to see here? |
.travis.yml
Outdated
| - PYTHON_VERSION=3.3 MATPLOTLIB_VERSION=1.4 NUMPY_VERSION=1.9 | ||
| - PYTHON_VERSION=3.4 MATPLOTLIB_VERSION=1.4 NUMPY_VERSION=1.9 | ||
| - PYTHON_VERSION=3.3 MATPLOTLIB_VERSION=1.5 NUMPY_VERSION=1.9 | ||
| - PYTHON_VERSION=3.4 MATPLOTLIB_VERSION=1.5 NUMPY_VERSION=1.9 |
There was a problem hiding this comment.
@dopplershift - try not pinnng the Numpy version here
dopplershift
force-pushed
the
set-backend
branch
from
63caacc to
a97e80f
Compare
|
No conda packages for Python 3.3 for matplotlib 1.5. Willing to drop Python 3.3? We'd still have the last 3 versions of the 3.x series. |
|
@dopplershift - You can always move mpl from |
|
True, but probably best to make a firm decision on 3.3 support, or at least testing for it. |
|
Does Matplotlib support 3.3? |
|
As far back as 1.5, matplotlib is >= 3.4: http://matplotlib.org/1.5.3/users/installing.html |
|
Ok then yes let's drop 3.3 |
|
All testing passing now. |
|
Are we waiting on anything here? |
|
(on vacation but will review/merge on my return!) |
|
Since #47, plt and matplotlib are only available locally, so there's a logical merge conflict in this PR. Since matplotlib is only imported locally now, the user at least has the option to import matplotlib globally and set the backend once as a fixture, prior to it being imported by pytest-mpl. I also wonder if we can more generally allow the user to specify arguments for |
|
@dopplershift - this looks good, but as @jnothman pointed out this won't work as-is due to #47, so could you rebase and make sure that the tests pass locally? |
This works by using matplotlib's "experimental" support for switching backends, and should only be used to switch between image backends.
Matplotlib only has 1.4 wheels for that, which we're removing here.
Conda doesn't have packages for that combination.
Matplotlib >= 1.5 only supports Python >= 3.4, so given we've dropped matplotlib 1.4, it doesn't make sense to support 3.3.
dopplershift
force-pushed
the
set-backend
branch
from
dbf4653 to
9f73275
Compare
|
Rebased and tests pass locally. |
|
Looks like we're good here. |
|
Thanks @dopplershift! |
4 participants
This works by using matplotlib's "experimental" support for switching
backends, and should only be used to switch between image backends.