re: Write a simple but impactful script VIEW POST

TOP OF THREAD FULL DISCUSSION
re: Serious question - How do you test randomness of an output?

One way is by controlling the seeder if possible, seeding it with same value every time should give you the same value every time.

There are some languages where you can instruct the test suite to run a test e.g. a hundred times and if those don't break the code you can be fairly confident (though not sure) that the code does what's intended. If your code takes input these tests usually test boundary conditions, maybe max and min values and a number of random values. If I remember correctly I saw it when playing around with Haskell.

I'd check the things you can know. Such as, if you ran the same method twice then it should result in two arrays of the same size that contain all the same things, but that aren't equal, i.e. the order is different.

I tried another thing recently where I needed to create a random string as output. I used dependency injection for the randomiser function with a default to the built in language version. Then in my test, I used a deterministic object for the randomiser function, such that I knew what the output was. I wasn't implementing the actual randomiser function, so testing it is not necessary, but testing that I'd call the right method on it and it returned the thing I expected satisfied me at the time.

Just some ideas.

i.e. the order is different.

But wouldn't this fail at times?

f.e. I have 3 elements in my array, [1, 2, 3]. Now the shuffle method only has six outcomes here, so there is a real possibility that the tests might randomly fail even if the code is correct.

Dependency injection seems like a better approach if you want to test it, but it makes code not as nice to read.

Trade-offs I suppose.

Aye, that's true. I was kind of thinking about it in terms of generating an array of 10000 items!

Writing testable code is a trade off, but probably worthwhile in the long run. Dependency injection itself makes for more and more flexible code without a great deal of extra cognitive overload, so I am a big fan of it at the moment.

This was a great talk from Sandy Metz that drove towards this point for a different reason too. I recommend a watch!

code of conduct - report abuse