.pyc files in source directories

[RussTedrake]

Running bazel run //doc:serve_sphinx results in doc/sphinx_base.pyc being left in the source dir (for git to complain about).

Running tests in manipulation station now results in examples/manipulation_station/differential_ik.pyc being left in the source dir.

Although I'm tempted to add the *.pyc to .gitignore, there is probably a better solution?

[jwnimmer-tri]

I started seeing this too, and was annoyed. I only now just realized that my realpath call in #10969 is likely at fault.

[jwnimmer-tri]

Actually, maybe it was tickled by my test runner change, but it was a bug in #10632 as it turns out.

[jwnimmer-tri]

The sphinx pyc file thing has been a bug for a while, I think it was @EricCousineau-TRI 8502b61 which did it. I'll let him decide how to mitigate / ignore it.

[EricCousineau-TRI]

In the FWIW category, I always add *.pyc files to my global gitignore file.

But yeah, we shouldn't be spewing these out in the first place!

[EricCousineau-TRI]

Confirmed that this also happens with --config=python3; in this case, it pukes __pycache__ all over the workspace.

Seems to be related to https://bugs.python.org/issue6386 (though probably not due to this bug, as the author wants to do weird things with relative imports?)

Made a repro as well, here: EricCousineau-TRI/repro@c913515

$ ./symlink_junk.sh python2
...
+ python2 ./symlink/root.py
...
+ find . -name '*.pyc' -o -name __pycache__
./sub.pyc

$ ./symlink_junk.sh python3
...
+ python3 ./symlink/root.py
...
+ find . -name '*.pyc' -o -name __pycache__
./__pycache__
./__pycache__/sub.cpython-36.pyc

This is mentioned here: bazelbuild/bazel#793 (comment)
EDIT: More directly here: bazelbuild/bazel#7091
Since bazel run runs without sandboxing, it spews junk everywhere.

I don't feel like doing any Bazel gymnastics to suppress this right now, as it's due to really esoteric behavior on Python's side for doing weird crap with symlinks. (I don't like this feature, and also really don't like that Python permits importing relative modules at all...)

(My ideal idea of behavior is that executing the symlink just straight up falls over, so that way there's no weird context bleeding that's currently happening.
Failing that, I'd prefer that the compiled bytecode litter the symlink tree instead...)

Will just submit a PR to suppress in .gitignore; maybe later we can try suppressing bytecode, but meh for now?

EDIT: Disclaimer: Still love Python. Just hate some of the odd design choices for paths...

[EricCousineau-TRI]

Er, looking at #10997, will first try absolute importing (using the other implicit behavior I hate... auto-gen'd __init__.py files...)