Skip to content
Snippets Groups Projects

negative syntax

Last edited by simonmar@microsoft.com

get rid of unary '-' operator

the unary '-' is a very odd case in the report and causes a lot of confusion.

to fix it we can

  • make '-' part of the lexical syntax of numbers. so a '-' followed by digits is interpreted as a negative number
  • get rid of all other special treatment of '-'
  • people can use 'negate' if they want to negate non literals.
  • the removal of n + k patterns will make this more attractive

Cons

  • This does introduce a syntactic oddity, namely that -1 is different from - 1 though syntax highlighting editors could make this distinction obvious.

  • It would no longer be possible to write x-1 to mean "x minus 1", instead it would mean "x applied to the literal -1", which is counter-intuitive. Also, x - 1 with spaces would have the "x minus 1" meaning.

  • In Haskell 98, '-' is explicitly excluded from right sections, so:

Prelude> :t (- 1)
(- 1) :: (Num a) => a
Prelude> :t (1 -)
(1 -) :: (Num a) => a -> a

If we removed special treatment of '-', then these become symmetrical again. However it seems unintuitive that (-x) is a right section of '-', but (-1) is the literal -1.

Another motivation for lexing - as part of a numeric literal (or changing its precedence), from 2007-10-10 #haskell:

[09:33] < quicksilver> am I the only person who finds mod's behaviour with
                       negative numbers stupid
[09:34] < matthew-_> try rem then
[09:34] < quicksilver> > (-4) `mod` 16
[09:34] < quicksilver> > (-4) `rem` 16
[09:34] < lambdabot>  12
[09:34] < lambdabot>  -4
[09:34] < quicksilver> that's odd, that's not what my ghci did
[09:34] < quicksilver> doh
[09:34] < quicksilver> bracketing error :P
[09:35] < quicksilver> mod does do what I expect, if I put the negative number
                       in bracets
[09:35] < quicksilver> I wouldn't expect 'unary minus' to have lower precedence
                       that `mod` though
[09:35] < sieni> > -4 `rem` 16
[09:35] < lambdabot>  -4
[09:35] < sieni> oops
[09:35] < sieni> > -4 `mod` 16
[09:35] < lambdabot>  -4
[09:35] < quicksilver> sieni: yeah, that's what I did
[09:36] < sieni> quicksilver: yeah I know, that has bitten me too :-)
[09:36] < quicksilver> I always expect unary minus to bind tight
[09:37] < osfameron> yeah, haskell's unary minus is surprisingly loose

haskell-src-exts, haskell-src, and HLint have been using a unary minus that binds tighter than any other infix operator, and have never encountered any problems with it: http://www.haskell.org/pipermail/haskell-prime/2009-August/002955.html.