Any idea what caused this to start failing?

No clue...this is really weird now it is failing with an EISDIR error

It looks like this fix was applied to one particular copy operation, but perhaps not to all of them (in the loop below).

I'd expect it to be doing a clean_arena before running each test anyway; I wonder why that's not happening.

It is the rmtree that I added that is barfing though...we never make it to that loop

Traceback (most recent call last):
  File "/home/travis/.micropython/lib/unittest.py", line 213, in run_class
  File "tests/api_tests.py", line 95, in test_medium_motor
  File "tests/fake-sys/populate_arena.py", line 27, in populate_arena
  File "/home/travis/.micropython/lib/shutil.py", line 8, in rmtree
  File "/home/travis/.micropython/lib/os/__init__.py", line 105, in unlink
  File "/home/travis/.micropython/lib/os/__init__.py", line 72, in check_error
OSError: [Errno 21] EISDIR

Traceback above from https://travis-ci.org/ev3dev/ev3dev-lang-python/jobs/537604548 for ev3dev/ev3dev-lang-python#641

My suggestion would be to figure out why clean_arena doesn't actually appear to be deleting everything. In theory, the whole folder should be empty at the beginning of each run.

@WasabiFan do we have docs somewhere on how to run the travis CI system locally?

There used to be a Travis CI documentation page on this, but I can't find it now. The simple answer is that, if you want to perfectly replicate the build environment, you can run their Docker container with their system setup scripts. There's an SO replica of the instructions here: https://stackoverflow.com/questions/21053657/how-to-run-travis-ci-locally

If you aren't able to/interested in running the full setup in Docker, my next recommendation would be to find a clean environment (perhaps a straight VM) and install uPy+dependencies from scratch using the install script we use on Travis and see if the versions it installs are broken. I didn't see any relevant changes in micropython-lib history (or, actually, any at all), but maybe it's another module.

@WasabiFan progress...I can get the tests to fail on my laptop if I use micropython v1.9.4 (I was using v1.10)

ok the delta from v1.9.4 to v1.10 is in os.walk. It is broken in 1.9.4 and thinks that the directories are files so when we try to os.unlink them it barfs. In v1.10 os.walk correctly IDs them as directories.

submitted PR #15 will continue discussion there