Bug - glob exceptions from base config do not get overwritten by more specific local ones

I have a base config with has an exception with globs to activate specific flake8 plugins for test files only and deactivate other rules:

...
[tool.flakeheaven.exceptions."**/test_*.py"]
flake8-aaa = ["+*"]
flake8-bandit = ["-S101"]

source

In a project I set this file as base and add another exception which is specific to a single file:

[tool.flakeheaven.exceptions."tests/test_rstcheck.py"]
flake8-aaa = ["-AAA01"]

But I still get multiple errors from this file like this:

tests/test_rstcheck.py:503:5: AAA01 no Act block found in test [flake8-aaa]

EDIT: I use the latest release version. EDIT2: On master this is also the case. 6138285ed808c334c9cb175e2a1f97dfb3cb679a

[orsinium]

I looked a bit more into it and found when it was changed: https://github.com/life4/flakehell/pull/104

So, yeah, my initial guess is correct. I implemented it to return the first match and then there was merging logic added which basically inverted the priorities. Now you can either revert the change or reverse the whole logic inside. Namely:

1. The reverse=True should be dropped, so longer exceptions go last (I tried to do it nginx-style for regular rules).
2. The globs should go before paths.

Shame on me for bad test coverage. Flakehell was my first experiment on “deliver fast, test later if people start using it”. TBH, in flake8-ruler I just dropped exceptions logic, “keep it simple”.

[Cielquan]

Then IMO this should be changed. Lets make an experiment:

Setup

Lets assume we have the following file structure.

File tree:

root
└─ dir_1
   └─ dir_1_1
      └─ file_1.py
   └─ dir_1_2
      └─ file_2.py

This means we have two files:

• root/dir_1/dir_1_1/file_1.py
• root/dir_1/dir_1_2/file_2.py

Example prefix

When we have the following exceptions:

[tool.flakeheaven.exceptions."root/dir_1"]
flake8-aaa = ["+AAA01"]

[tool.flakeheaven.exceptions."root/dir_1/dir_1_2"]
flake8-aaa = ["-AAA01"]

This would result in +AAA01 for both files, because the upper exception has a shorter prefix. If I would see this config without knowledge I would assume that file_2.py has -AAA01 and file_1.py has +AAA01, would you not?

Example glob

When we have the following exceptions:

[tool.flakeheaven.exceptions."*/dir1"]
flake8-aaa = ["+AAA01"]

[tool.flakeheaven.exceptions."*/dir_1/dir_1_2"]
flake8-aaa = ["-AAA01"]

This would again result in +AAA01 for both files, because the upper exception has a shorter glob. If I would see this config without knowledge I would again assume that file_2.py has -AAA01 and file_1.py has +AAA01, would you not?

Example prefix vs glob

When we have the following exceptions:

[tool.flakeheaven.exceptions."*/dir1"]
flake8-aaa = ["+AAA01"]

[tool.flakeheaven.exceptions."root/dir_1/dir_1/dir_1_2"]
flake8-aaa = ["-AAA01"]

This would again result in +AAA01 for both files, because the upper exception has is a glob even so it matches less good for file_2.py. If I would see this config without knowledge I would again assume that file_2.py has -AAA01 and file_1.py has +AAA01, would you not?

Conclusion

So in all three cases the result is against intuition.

I think it would be better to take the longest matching prefix/glob instead of the shortest, because elsewise you cannot overwrite deeper nested exceptions.

I also think it is no good to let globs take precedence over prefixes by default.

If I have an error in my thought experiment please tell me.