tests: consider running from isolated scratch cwd

[lread]

While working on #152, I found myself wanting to test the effect of operations like (fs/delete-tree "").

Would you be ok with always running tests from an isolated scratch current working directory so that unit tests can test destructive operations on the cwd?

An isolated mutable cwd for tests seems like a good idea in general, what do you think?

If so, I am happy to take a crack at a PR to explore the idea.

[borkdude]

ok!

[lread]

Thanks! I'm starting to experiment.

Initial thoughts:

1. A contributor will likely be annoyed if the cwd isn't set to the scratch dir and tests wipe the cwd. I'll start with a warning in the README and a test fixture to assert cwd is actually the test scratch dir.

2. I remember that babashka runs babashka/process tests, I don't think it runs babashka/fs tests... if it did it, it would have to run them in isolation where cwd was set to a scratch dir.

3. many babashka.fs tests currently use source files in the repo as test data. This is convenient, but perhaps a bit awkward. With a cwd test scratch dir, we could create specific pre-conditions (with specific files and specific dirs) in the test cwd scratch dir and decouple.

[borkdude]

Maybe not to over-engineer this. If this is just about (babashka.fs "") we could test that specific case by shelling out to bb or clojure in a different tmp directory.

[lread]

Yeah, thanks, sounds good.
The empty-string tests for #152 are what triggered this issue.
I'll scope the isolated scratch cwd to those tests in a new separate test ns for now.

This issue was exploring doing this in general, so I think it can be closed.

[lread]

Because having two test tasks (test and test-cwd) & 2 REPL launch tasks (dev and dev-cwd) is not ideal, re-opening this issue (as discussed on Slack).

I'm happy to do the work.