Regarding differences with DIPY, I agree that it's very strange.

Maybe you can initialize the temporary directory in the same place that we put data? op.join(os.expanduser('~'), '.dipy')

I just remove the concurrency lines to investiguate, if this is the source maybe changing the folder is the easiest solution and still safe (since we will know what is the source)

Hopefully, tomorrow this will be solved. Getting close!

it would be great to reproduce the error here.

I do not believe that the user will try to find the folder and delete manually. Actually, there are some chances that even manually, he will not be able to delete the folder. He will have to kill the terminal and then be able to delete the folder.

I think the file do not close because he still have an active reference somewhere. In the case of DIPY, sft reference is still active when you close, and you still use sft which have a reference to the file. (due to memory map)

it might be some design issue, I am not sure. I did not check your test here but do you have such an example.

1. load trx
2. create a reference
3. close TRX
4. still doing something with the reference.
5. check if TRX was closed or still open because of the reference.

@skoudoro I just added test_close_tmp_file to do explicitly what you said and it is still passing.

When loading the trx zip file, it creates a tmp folder, then I create a SFT, close the TRX and then I

• Verify the folder does not exist
• Check the SFT
• Call sft.to_vox()
• Check the SFT again

Do you want to add a test with a context manager here, to see if that's somehow messing things up?

I am not sure how to do that since I create the tmp directory in the class, and I want to keep it active outside of the scope of the __init__ and pass the trx around freely.

I think your test is not complete. In your test.

• you should remove tmp_dir = deepcopy(trx._uncompressed_folder_handle). (this is something you do not do in DIPY)
• after sft.to_vox() you should load again (call trx = tmm.load(path)). This second load should not fail. This is what is happening is DIPY, the second load is failing because it believe that the file is still open / use

Sorry, I missed the second loading on my previous description

I added a second loading (of the same trx), checking the data, then closing it again.

I also made the deepcopy of the tmp_dir.name (just a str) clearer so it is safer (just a string)

@arokem @skoudoro my new test is much more direct. test_close_tmp_files

Earlier I was checking if the folder stopped existing, now I am using psutil to actually check if there are any opened files by the process.

And they are passing (on github, on my linux & windows). So I think its all working as intended, but all of these new tests are very good to have!

(even if I have no idea why it is not working in Dipy)