Cop to require comment justifying # rubocop:disable

[bessey]

I think its pretty unavoidable that for non-stylistic cops, there will be >0 false positives. In such instances we can disable the cop, and so long as we aren't regularly disabling the cop instead of fixing the issue, this isn't a bad thing.

I believe its commonly consider best practice to justify the disabling of a cop, but AFAIK there is no programmatic way to enforce this. Justifications might not always be good of course, but by forcing it to be written down we can increase the odds that a peer reviewer identifies the flawed logic.

Has such functionality been discussed / implemented before? I could not find evidence of that, but please
do share them if you are.

Failing that, I propose introducing a formal syntax for justifying, and a cop to enforce the presence of a justification

# RUBOCOP: This batch is too big to run validations, and there are none on this attribute
relation.in_batches.update_all(verified: true) # rubocop:disable Rails/SkipsModelValidations

Or a slightly nicer syntax (that relies on a non-existent Rubocop feature) might be

# rubocop:disable:next Rails/SkipsModelValidations -- This batch is too big to run validations, 
# and there are none on this attribute
relation.in_batches.update_all(verified: true)

For existing codebases, there would need to be a way to note the absence of an existing justification when the cop was enabled, e.g.

# RUBOCOP: Disabled before Lint/JustifyDisable was enabled
relation.in_batches.update_all(verified: true) # rubocop:disable Rails/SkipsModelValidations

or

# rubocop:disable:next Rails/SkipsModelValidations -- Disabled before Lint/JustifyDisable was enabled
relation.in_batches.update_all(verified: true)