[contrib] Fix "custom_path" already exist bug

[huanmei9]

fix error when use custom_path parameter in tempdir funciton.

[Mousius]

Hi @huanmei9, just curious what the use case is here? Doesn't this allow for race conditions and pollution by re-using temporary directories?

[tkonolige]

I believe the correct way to do this is os.makedirs with exist_ok=True.

[junrushao]

Hmm I agree with @tkonolige. It is slightly better to use the existing API provided by python. Please make changes accordingly and then we get the PR merged :-) Thanks a lot!

[huanmei9]

Hi @tkonolige @junrushao1994 .
Thanks for your suggestions, I have changed the code according to your suggestions.

[huanmei9]

Hi, @leandron .
Sorry for the late reply.
My use case is to check the build fils generated by zephyr in MicroTVM, such as .map and .hex files.
In addition, regardless of the use case, the tempdir function provides a custom_path interface, so it should be robust enough.

[Mousius]

Hi @huanmei9,

I would suggest that the current implementation of custom_path is robust, consider a typical task which does the following:

1. Creates a temporary directory
2. Writes some files
3. Checks those files

With the current behaviour, if two tasks attempt to do this in parallel (such as parallel tests from pytest-xdist), you get the following:

1. Task A creates a temporary directory
2. Task A writes some files
3. Task B attempts to create a temporary directory - raises Exception that temporary directory already exists
4. Task A checks those files
   Task B has failed as it tries to pollute the existing temporary directory.

With this change, you can instead have:

1. Task A creates a temporary directory
2. Task A writes some files
3. Task B attempts to create a temporary directory - no error raised
4. Task B writes some files
5. Task A checks those files - raises assertion error
6. Task B checks those files
   Task A has now failed because of the pollution of the temporary directory and the error is also incorrectly attributed to Task A rather than Task B. This scenario only occurs intermittently and is therefore harder to debug.

Therefore, I would suggest that the current implementation of temp dir helps to prevent such concurrency bugs and raises the correct error to ensure we use a new fresh temp directory per invocation - what are your thoughts?

[huanmei9]

Hi, @Mousius .
The situation you describe is indeed possible, Thanks for your suggestion.
(●'◡'●)