Brandon.Si(mmons)

code / art / projects

In Defense of Partial Functions in the Haskell Prelude

…because I’m trying to blog more, and this sounds like a fun argument to try to make.

One of the most universally-maligned parts of Haskell is the inclusion of partial functions in its standard library called Prelude. These include head and tail which are undefined for the empty list:

head :: [a] -> a
head (a:_) = a
head [] = error "empty list"

It’s generally understood that the inclusion of these sorts of functions are a wart (that the type of head should be [a] -> Maybe a) that has motivated the proliferation of Prelude alternatives, few of which are used by anyone besides their authors (and fewer still have enthusiastic advocates).

I’ve heard a handful of allusions to tricky production bugs that involved some deep dive to find the source of a "*** Exception: Prelude.head: empty list", but personally I can recall only one instance of such a bug in the code I’ve worked on professionally and it was trivial to track down. I can’t recall flagging a use of head in a code review either, or raising an eyebrow at some use of the function in some library source I was perusing.

But most of the time the argument goes that partial functions should be removed for the sake of new users, who will become quickly disillusioned when their first function blows up with an exception. But you said haskell was safe!

It would be unfortunate if this caused a new user to give up, and maybe this is a real problem for the community, but here’s what I think really happens to most of us:

  • Your homework doesn’t work; this doesn’t matter.
  • You use Google and quickly learn that partial functions (will forever) exist, and that they’re bad
  • You ask yourself “Hm, come to think of it what did I expect to happen…?”

And so you learn an important lesson, early on and in the most painless way possible, you acquire a nose for inferring which functions must be partial, an appreciation for compiler warnings that help prevent accidentally-partial functions, etc.

Would I recommend designing a standard library around this weird sort of tough-love? Probably not, but I think the haskell library ecosystem and pedagogy have benefited from this wart.

The problems that get the most (and most passionate) attention are usually not the ones that are the most important, but the ones that are the most easily understood. I think in the proliferation of Preludes and the discussion around partial functions (and the fact that they haven’t been excised yet) we see evidence of both the Law of Triviality, and a healthy language pedagogy.

Comments