On 29/07/2018 02:04, Jeremy Kloth wrote:
On Sat, Jul 28, 2018 at 6:43 PM Brett Cannon <brett@python.org> wrote:
If Windows doesn't clean up its temp directory on a regular basis then that doesn't suggest to me not to use tempfile, but instead that the use of tempfile still needs to clean up after itself. And if there is a lacking feature in tempfile then we should add it instead of a avoiding the module.
Mind you, this is mentioned in the confines of the test harness where just about anything can happen (and usually does!). Something that cannot be coded against using just tempfile is cleanup on process abort. The per-process-directory approach handles this case.
I would think it is desired to have no leftovers after running the test harness (especially in regards to the buildbots).
Now, I'm not sure the exact cause of all of the leftovers in the TEMP directory, but it is definitely something that is currently happening (and shouldn't be). It is not exactly the easiest of tasks to track the file usage of every test in the test suite. It is certainly easier to replace usages of os.unlink with test.support.unlink within the test suite.
In the interests of trying to keep a focus to the changes I'm making, I propose to start again by, as you suggest, making use of test.support.unlink where it's not currently used. From the evidence I don't believe that will solve every problem I'm seeing but it should certainly reduce them. I do there there's mileage in a wider change to revamp the test suite's naming and cleanup of temporary files but I'm very wary of trying to undertake what would undoubtedly be a sprawling and probably contentious change. TJG