引導式測驗 · 硬 等級
如何使用 AI 測驗重構遺留程式碼
透過建立基準、進行小的變更和測試邊緣情況,使用 AI 實踐行為保留重構。
問題 1 的 8
A legacy function has no written specification. What should a team capture before changing its structure?
預覽每個問題8 問題
當您參加測驗時,就會出現答案和解釋。
- A legacy function has no written specification. What should a team capture before changing its structure?
- Current observable behavior with tests or representative fixtures
- Only the model’s summary of what the function should do
- A shorter replacement implementation
- The number of lines removed
- An AI proposal both extracts a helper and changes how malformed input is handled. What is the safest review plan?
- Merge both because the new code is cleaner
- Delete the malformed-input tests
- Assume the behavior change is required by refactoring
- Separate the structural refactor from the behavior change and review each on its own
- Why is a model-generated test not automatically proof that a refactor is safe?
- Tests cannot be run after refactoring
- It may encode the same mistaken assumption as the generated implementation
- AI-generated tests always fail
- A test suite makes behavior irrelevant
- A refactor changes iteration order for returned records, and a downstream client depends on that order. Which risk did the change expose?
- Only the function’s internal naming changed
- The source file became easier to navigate
- An observable behavior or compatibility contract may have changed
- The tests now run faster
- Which work pattern best limits risk when asking AI to refactor an unfamiliar module?
- One small transformation, full diff review and focused tests before the next step
- A complete rewrite followed by one final test run
- Accept every suggestion and review only changed filenames
- Combine feature work and cleanup in one opaque patch
- A test calls a live external service whose response changes over time. What can make the characterization test more dependable?
- Increase the number of retries until it passes
- Remove all assertions
- Ask AI to predict the service’s next response
- Use a deterministic test double or recorded fixture for the external interaction
- After helper extraction, the focused tests pass. What should happen before merging?
- Assume every possible behavior is preserved
- Run broader checks and inspect compatibility, side effects and representative edge cases
- Skip review because the diff is small
- Delete tests that document old behavior
- A reviewer finds that “cleaned up” code now logs less information on failures. What is the relevant next question?
- Whether the assistant used a modern syntax feature
- Whether the refactor reduced line count
- Whether logging is an observable operational behavior the system needs to preserve
- Whether comments can be generated automatically