“Considered Harmful” titles always make me think of “Considered Harmful” Essays Considered Harmful

I’m in favour of validating keyword args and do it myself (personally I use NimbleOptions) but that article hardcore handwaves through the entire premise of how they got there: how on earth did their tests not catch a mis-spelled option?? It sounds like they were passing the option but not actually asserting on its effects. Am I wrong or is there an obvious valid hiccup you can have here?

While it’s getting outside the realm of options, keywords also allow us to have order matter in the rarer situations where that’s useful:

from q in query,
  join: u in User,
  on: u.id == q.user_id,
  join: o in Org,
  on: o.id == u.org_id

I support everything that helps people make less mistakes and dynamic languages like Elixir don’t provide as much protection there.

So firstly, good job.

Secondly, the linked article does not sell the resulting library to me. Keyword.get and Map.get are something that many Elixir devs, myself included, consider an anti-pattern simply because they don’t discern between “I don’t have the key” and “I have the key but the value is nil” – in some situations this difference is meaningless but I’d bravely claim that in at least 80%, if not 90%, of the Elixir code I had to author or maintain that difference was in fact important but people ignored it and introduced bugs. So just by using functions like take and fetch you can replace most of the conveniences of this library – though granted, it would take more boilerplate so the library still looks compelling.

Thirdly, not comparing the new library with NimbleOptions is giving homework to the reader so I am going to skip it and just use the former instead.

As with everything else we use, it is considered harmful when it expresses the wrong intent.

I use Keyword.get/3 in situations where an option might be missing and I have a default value for it. For example:

Keyword.get(opts, :option, default_value)

Using it in cases where having that key missing is invalid logic, you are just propagating a bug further into the system. This is especially bad when you leave the default value to nil and propagate it further, hard to debug and trace down.

On a tangent, Perl5 has a // operator, which is like the logical or || but only treats undef as falsey. Useful for disambiguating nil/null/undefined/missing values from defined but false ones.