Clarify whether and when tmp_path is deleted after each use

[xuhdev]

Currently the document is unclear about whether and when tmp_path is deleted after each use. After I digged into the code, I still have no idea. I wish there could be some clarification on this topic in the document.

[RonnyPfannschmidt]

Pytest keeps around the last 3 test runs, we ought to make that more discoverable

[matthewhughes934]

Hi, I'm interested in helping here. Where would be best to note this so that it's more discoverable? It looks to be already noted here: https://github.com/pytest-dev/pytest/blob/master/doc/en/tmpdir.rst#the-default-base-temporary-directory, could this be moved to somewhere more visible?

[nicoddemus]

We might add a note to tmpdir and tmp_path in the reference docs.

[xuhdev]

Pytest keeps around the last 3 test runs, we ought to make that more discoverable

Thanks. When you say "the last 3 test runs," do you mean the last 3 test sessions, or the last 3 individual tests?

[RonnyPfannschmidt]

@xuhdev the last 3 sessions, when creating a new basetemp, pytest tries to keep the last 3 around to aid debugging

[hmaarrfk]

Would it be acceptable to add a flag where the tempfiles are delete if the test is successful? for large datasets that get created for tests, it is advantageous for us to be able to use a standard fixture, without worrying about running out of diskspace.

[nicoddemus]

Hi @hmaarrfk,

I think a configuration option would be acceptable, something like tmp_keep, which by default is 3 but could be set to 0 to accomplish what you mention. 👍

[RonnyPfannschmidt]

An option to generally keep only failed tests could be a big space safer

[nicoddemus]

An option to generally keep only failed tests could be a big space safer

Ahh indeed, I misinterpreted @hmaarrfk's request.

I agree keeping only the tmp directory of failed tests would make for a better option.

[matthewhughes934]

I agree keeping only the tmp directory of failed tests would make for a better option.

If this feature (or another from this thread) is desired, I'd be interested in working on it.

[RonnyPfannschmidt]

its a desired feature, lets open a issue to clarify the details and then it can be started

[nicoddemus]

Agreed, created: #8141

[nicoddemus]

Closing this as fixed by #8038