fixed typos/import ordering and removed the commit that add the color cycler in rcsetup (as that breaks tests).

I think this should go in (despite @WeatherGod 's protests) to fix the test and we open a new issue to discuss how the default color is handled.

Looks like this still have issues on python 2.7

yeah, those 2.7 failures are non-trivial. Something is seriously wrong.

I'll build and test this pr for v2.7 to see if I can figure it out

I can reproduce the two hatch validation errors, and one of the two image comparison failures. The plot for the linestyle cycle test is coming up blank... no lines, no legend. I can't reproduce the markerstyle test failure, though. Makes me wonder if something non-deterministic is happening?

I got other obligations for today, so I don't have more time to pursue this until tonight.

The python2 issue was just using the wrong string type.

It makes me happy to find string/unicode issues that only show up on LPy 😄 .

Given that this is passing tests now, I am going to go ahead and merge it so that testing of other PRs will no longer be broken. Work can proceed towards solving #6380.

I merged this into master.

I am still a little concerned about those non-deterministic errors. Could it be a problem with the caching mechanism?

@WeatherGod, I thought the remaining Travis error (frequent but intermittent) was just the doc build error we have been seeing recently, which is a failure of Travis to download basemap. However, the attempt by @jenshnielsen to fix that with #6379 failed Appveyor last night on 2.7, apparently because it didn't find latex.

Nope. As I noted earlier:

and one of the two image comparison failures. The plot for the linestyle cycle test is coming up blank... no lines, no legend. I can't reproduce the markerstyle test failure, though. Makes me wonder if something non-deterministic is happening?

Now, the damnest thing is that I just tried out the branch with the last fix in there (which should have nothing to do with the remaining failure), and it passes -- multiple times. I don't get it.

I think the problem is that there is a leak in decorators exception
handling. Because we are running the tests in parallel exactly which test
gets broken is a bit random.

On Sun, May 8, 2016, 13:39 Benjamin Root notifications@github.com wrote:

Nope. As I noted earlier:

and one of the two image comparison failures. The plot for the linestyle
cycle test is coming up blank... no lines, no legend. I can't reproduce the
markerstyle test failure, though. Makes me wonder if something
non-deterministic is happening?

Now, the damnest thing is that I just tried out the branch with the last
fix in there (which should have nothing to do with the remaining failure),
and it passes -- multiple times. I don't get it.

—
You are receiving this because you authored the thread.
Reply to this email directly or view it on GitHub
#6374 (comment)

perhaps, but it isn't limited to parallel execution. Yesterday, I ran only test_cycles.py directly and got that one failure in addition to the two hatch property validation errors. Nothing was parallel. There may be a leak, but it is likely within the same process.