[Haskell-cafe] Guards (Was: Some random newbie questions)

Jon Cast jcast at ou.edu
Fri Jan 7 21:46:43 EST 2005

```Henning Thielemann <lemming at henning-thielemann.de> wrote:
<snip>

> What about dropping Guards? :-) Are they necessary? Do they lead to

do notation, and Haskell lacks a direct analogy to Lisp's cond.

case () of
() | p1 -> e1
| p2 -> e2
...

works beautifully as a replacement.  Also, GHC's pattern guards are a
nice feature, and frequently seem clearer than case.  Compare, e.g.,

parseCmd ln
| Left err <- parse cmd "Commands" ln
= BadCmd \$ unwords \$ lines \$ show err
| Right x <- parse cmd "Commands" ln
= x

parseCmd ln = case parse cmd "Commands" ln of
Left err -> BadCmd \$ unwords \$ lines \$ show err
Right x  -> x

The trade-off: using pattern guards makes it harder to verify (and
ensure) that the exact same expression is being matched against; using
case makes it harder to see exactly what is being matched against.

Furthermore, guards are an extension of pattern matching, which means
you can write code like this:

xn !! n     | n < 0  = error "Prelude.(!!): Negative index"
[] !! n              = error "Prelude.(!!): Index overflow"
(x:xn) !! n | n == 0 = x
(x:xn) !! n          = xn !! (n - 1)

Exactly one equation for each edge in the control-flow graph, which is
nice and not easily done (I'm not sure it's even possible) without
guards.

Pattern guards are also nice for implementing â€˜viewsâ€™:

-- | Convert an 'XMLData' into an equivalent application of
-- 'Balanced', if possible.  In any case, return an equivalent data
-- structure.
balance (Balanced es) = Balanced es
balance (LeftLeaning (LeftBalanced e:es))
| Balanced es' <- balance (LeftLeaning es)
= Balanced (e:es')
balance (LeftLeaning []) = Balanced []
balance (RightLeaning [("", "", es)]) = Balanced es
balance (RightLeaning []) = Balanced []
balance e = e

Where XMLData can store a (nearly) arbitrary fragment of an XML
document.  The problem being solved by the pattern guard in the second
equation is that the data type is ambiguous; there is more than one way
to represent a â€˜balancedâ€™ XML fragment (that is, the concatenation of a
sequence of well-formed XML fragments and CDATA sections).  This
function attempts to coerce the data structure passed in into a
canonical representation; it succeeds if the data is in fact balanced
and fails otherwise.  The pattern guard illustrates how to use this
function as a replacement for pattern matching on Balanced, to catch all
cases where the argument is in fact balanced (we can't use it in this
case as a replacement for the first equation, since that create an
infinite loop, but in other functions we could).

I'm sure there are uses I'm forgetting, but I think that's enough.

> Do they lead to more efficient code? I could perfectly live without
> them up to now.

Well, I could never do without them.

Jonathan Cast
```