Before you write code
Before you dive right into the implementation details of your rule, it's good to ask yourself the following questions first.
Do you want a linter or rewrite?
- Rewrites can automatically fix problems but can only address problems that have un-ambiguous solutions.
- Linters highlight problematic code without providing a fix, which means they require the user to fix the problem. Linters are not limited to problems that have unambiguous solutions so linters can address a larger problem space than rewrites.
Is your rule syntactic or semantic?
- Syntactic rules are simple to run because they don't require compiling input sources beforehand but syntactic rules can only do limited code analysisj since they don't have access to compiler information such as symbols and types.
- Semantic rules are are slower and more complicated to run since they require compilation but on the other hand semantic rules are able to perform more advanced code analysis since they have access to compiler information such as symbols and types.
What is an acceptable false-positive rate?
- If you are doing a one-off migration script for your company codebase then it might be OK to manually fix trickier corner cases.
- If your linter detects serious production bugs it's maybe OK that it has a few false positives. If your linter reports low-importance issues like "method name should start with lowercase character" then false positives are not acceptable.
What diff does the rewrite produce?
Before implementing a rule, it's good to manually migrate/refactor a few examples first. Manual refactoring is great to discover corner cases and estimate how complicated the rule needs to be.
Who will use your rule?
The target audience/users of your rule can impact the implementation the rule. If you are the only end-user of the rule, then you can maybe take shortcuts and worry less about rare corner cases that may be easier to fix manually. If your rule is intended to be used by the entire Scala community, then you might want to be more careful with corner cases.
What codebase does your rule target?
Is your rule specific to a particular codebase or is the rule intended to be used for any arbitrary project? It's easier to validate your a rule if it will only run on a single codebase. You may not even need tests, since the codebase is the only test. If your rule is intended to be used in any random codebase, you need to put more effort into handling corner cases. In general, the smaller the target domain of your rule, the easier it is to implement a rule.
How often will your rule run?
Are you writing a one-time migration script or will your rule run on every pull request? A rule that runs on every pull request should ideally have some unit tests and be documented so that other people can help maintain the rule.