[Haskell-cafe] Where does η-equivalence stop?

Richard Eisenberg lists at richarde.dev
Tue Jun 28 18:57:14 UTC 2022

My best understanding is that eta-equivalence holds only in total languages, and even there without regard to performance. So you can hunt for eta-equivalence trouble anywhere that non-totality or performance comes into play.

For example, compare (A) `f (slowFunction 1000)` and (B) `\x -> f (slowFunction 1000) x`. If we map (A) over a list, then `slowFunction 1000` is computed once. If we map (B) over a list, then `slowFunction 1000` is computed for each element of the list.

Note that this example is very bare-bones: no extensions, no parametricity-busting `seq`, no compiler flags. All it relies on is a non-strict functional language.

I do think compiling this examples is a nice service -- thanks!

> On Jun 20, 2022, at 4:38 AM, Hécate <hecate at glitchbra.in> wrote:
> Hi Café!
> I was wondering if anyone knew of a centralised list of occurrences where GHC Haskell stops upholding η-equivalence. I am interested in both type changes and operational semantics.
> So far I've collected the following things:
> * Rank-2 types
> * Monomorphism restriction
> * The presence of seq
> * Non-pedantic bottoms
> One source is the GHC Manual¹, and Neil Mitchell pointed me to the list of bugs and limitations of HLint².
> If you have other examples, or explanations of the mechanisms at play here, I would be very interested, and intend to upstream those in the GHC manual.
> Cheers,
> Hécate
> [¹] https://downloads.haskell.org/~ghc/latest/docs/html/users_guide/bugs.html#expressions-and-patterns
> [²] https://github.com/ndmitchell/hlint#bugs-and-limitations
> -- 
> Hécate ✨
> 🐦: @TechnoEmpress
> IRC: Hecate
> WWW: https://glitchbra.in
> _______________________________________________
> Haskell-Cafe mailing list
> To (un)subscribe, modify options or view archives go to:
> http://mail.haskell.org/cgi-bin/mailman/listinfo/haskell-cafe
> Only members subscribed via the mailman list are allowed to post.

More information about the Haskell-Cafe mailing list