refactoring-patterns

Ensure tests exist and pass before refactoring, then make small incremental changes and run tests after each step. Commit working states frequently so any regression can be isolated and rolled back quickly.

Identify a cohesive code block, check which local variables it uses, create a new method with those as parameters, and replace the original code with a call. Run tests immediately after the extraction to confirm behavior is unchanged.

Use an IDE rename tool when possible, then search for dynamic references like getattr calls and string literals that tools miss. For public APIs, add the new name as an alias, deprecate the old one with warnings, and remove it after a grace period.

Replace conditionals with polymorphism when switch statements or type-checking branches grow across multiple locations. Create a subclass per branch and override a shared method so new types are added without modifying existing conditional logic.

Tests may miss dynamic references such as getattr calls, serialized data with old field names, environment variables, or database columns. Search for string references and check external contracts like APIs and configs before finalizing changes.

Refactor when methods exceed 20 lines, cyclomatic complexity exceeds 10, functions take more than 4 parameters, nesting goes deeper than 3 levels, or classes have more than 20 methods. Tools like radon can measure these metrics in Python.