Hacker News Reader

Comment by rgoulter | Hacker News Reader

A "snapshot test" like this would save me time if I wanted to check that the same input yields the same output for some module. -- I can see wanting that if I'm adding new code which doesn't change the old part of the module much, or if I'm changing the implementation (but not the interface) of the module.

But, yes, in many cases it's not a good automated check.

View on HN

I like to "make the change easy, then make the easy change". This looks like 2 commits: 1st a refactor to make code more straightforward around the area I'm about to change, and 2nd to actually make the change. Ideally, the 2nd should be a small, local, easy to reason about change, enabled by code cleanup in the 1st.

The 1st commit in this sequence is a pure refactor, and definitionally should change no behavior. The "snapshot test" described sounds perfect for this, especially in unfamiliar code. Ideally, I'd go even further, and have a compiler prove for me that my refactor produced a perfectly equivalent program from a black-box perspective. Snapshot testing is great because it gets pretty close, very cheaply, whereas the full program equivalence problem is impossible.

Very simple - I really like that, thanks for sharing.

When the snapshots are somewhat human readable this technique is also very useful for validating changes.

Hook up the last known good snapshot and the new output to your favorite diff UI and you'll get a pretty good idea about the actual effects of the code changes. I feel more confident signing off a new "known good" based on reviewing a full document diff than after iterating a set of meticiously hand-crafted expectations alongside the production code until convergence.

Reply on news.ycombinator.com