[yt-dev] Issue #1052: Field access tests intermittently fail under python3 (yt_analysis/yt)

[Nathan Goldbaum]

New issue 1052: Field access tests intermittently fail under python3
https://bitbucket.org/yt_analysis/yt/issues/1052/field-access-tests-intermittently-fail

Nathan Goldbaum:

There seems to be an issue with the way the field access tests work under python3.

This can (only sometimes, unfortunately) be reproduced by running `nosetests fields` and waiting until the `gas` field tests run. The first failure will usually be for accessing `('gas', 'averaged_density')`.

Some hints:

* only happens for field access tests with `nprocs != 1`. This doesn't actually use multiple processors, instead the stream dataset is segmented into multiple grids, so it does have something to do with the selection system in nontrivial grid hierarchies.

* Might have something to do with the `__hash__` function implementation for selector objects. The definition is in `selection_routines.pyx`.

* Might have something to do with hash randomization in python 3. I was able to get the errors to stop happening after doing `PYTHONHASHSEED=0` before running nose.

* Happens intermittently.  I was able to reproducibly trigger this earlier today but am now no longer able to trigger it at all.

I'm honestly mystified. I'm going ahead and creating an issue in case my findings are useful for others who want to investigate.

Unfortunately this blocks turning on the test bot for python3 CI.