Refactor tests

[ap--]

from #96

Refactor tests and fixtures for improved isolation. Make it easier and safer to allow for tests to be run live. Reduce inadvertent test interaction and flakiness. Eventually try to use UPath's great test suite directly from other fsspec filesystem projects to improve coverage.

@brl0 that would be great. Maybe we could coordinate with fsspec/filesystem_spec#651 ?

Did you run into specific issues when running the current test suite? Should we add some test order randomization to reveal issues regarding isolation?

[brl0]

@ap--,

First, thanks for pulling these out into issues for discussion, I appreciate your diligence.

Maybe we could coordinate with fsspec/filesystem_spec#651 ?

Good point... I've been sitting on some of these changes for a while, and it has taken a bit more effort to wrap them up than I thought it would, but I was planning to ping that issue once I had something for others to look at.

Did you run into specific issues when running the current test suite?

The most recent issue is that certain flaky tests failed in CI on Windows with python 3.10, but passed after two retries. There were other issues, but this was initially more of a back-ground tinkering effort, so I had not been diligent about noting them.

Should we add some test order randomization to reveal issues regarding isolation?

Definitely, and the changes I have been planning to submit include using pytest-random-order.

When I prepare to submit those changes, I'll try to ping back here if I notice any particularly remarkable issues.

[ap--]

I'm really looking forward to this. Do you remember (or can you reproduce) which tests causes problems on windows? Was it limited to a specific filesystem?

[ap--]

It seems tests are now pretty reliable, since the following PRs were merged:

• 123

• 122

• 109

• 107

• 104

• 103

Let's reopen a new issue in case we run into issues in the future.