You are writing a library. An evil genius steals all your library’s code somehow (you can’t retrieve it) and offers you a choice. You can either recover all your library’s implementation code, or all of your library’s tests, but you may not have both. What do you recover?
You are writing an application. An evil genius steals all your app’s code somehow (you can’t retrieve it) and offers you a choice. You can either recover all your app’s implementation code, or all of your app’s tests, but you may not have both. What do you recover?
Can you guess my answers?
This ofc depends on circumstances but other things equal, my answer is this:
https://twitter.com/LauraPilkingt13/status/1465509235832274947
Let’s take React for example. If we lose the entire source, we can reconstruct it (and probably in a better way) by starting from scratch and using the tests to guide and verify the implementation. I know we can do it because we literally did this once: engineering.fb.com/2017/09/26/web…
On the other hand, if we lose all tests for React, we’re screwed in a major way. We’d lose the specification about how our APIs work, edge cases, weird things to look out for, unintuitive consequences, bugs we already fixed, and so on. And any further change is scary and unsafe.
Whereas for the app, usually tests cover only a tiny percentage of what the app *is*. I’ll take a working login form over a test verifying a login form works any day! The implementation is the hard part, while good tests look trivial and obvious (“click here, get this”).
Might be interesting to note that React’s tests are written with public API (or at least, external entry points). There are no unit tests for individual modules. This is intentional because otherwise tests don’t help with rewrites. Unit tests can’t guide bigger-unit rewrites.
Another aspect. As a library developer I mostly know how it works already. Not all pieces but decently enough to figure it out. Whereas when I’m working in an app, I know a relatively small % of it. So I wouldn’t know where to even start reconstructing many pieces.
Also with an app, I can mostly write/guess tests without deep knowledge. Because I know what the app *should do*. Pressing log in should log me in. Pressing Buy should add to cart. Etc. Whereas library scenarios aren’t always obvious due to how generic and varied usage can be.
Interesting that people Disagree over this
• • •
Missing some Tweet in this thread? You can try to
force a refresh
