Made serious progress also thanks to Erik Meijer's lectures. Good watch, maybe a hint. Haskell allows for several ways to write the same function. Which one of these would be best in terms of efficiency and readability?
sqr' = \x -> x * x
sqr'' x = x * x
sqr''' = (^2)
Between these two top-level definitions:
sqr' = \x -> x * x
sqr'' x = x * x
the second is pretty much universally preferred in Haskell programs. Search through almost any chunk of real-world Haskell code, and you will find many examples of the second but few of the first. Instead, "lambda abstraction" (ie, the \\x -> ...
syntax) is most often used for defining anonymous functions to pass as arguments to higher order functions.
There are a couple of reasons the second syntax is preferred. First, it's literally more concise and, from a readability perspective, incorporates fewer distinct syntactic elements (ie, juxtaposition and the =
operator, instead of juxtaposition, =
, \\
, and ->
). It also generalizes well to the common Haskell idiom of defining a function using multiple patterns:
factorial 0 = 1
factorial n | n > 0 = n * factorial (n-1)
To do this with the lambda syntax, you'd need to add an explicit case
construct, involving yet another set of syntactic elements.
Between:
sqr'' x = x * x
sqr''' = (^2)
or -- perhaps a fairer comparison -- between:
sqr'''' x = x^2
sqr''' = (^2)
it's more a matter of personal preference. Many Haskell programmers like the clean look of so-called point-free syntax, where larger functions are made up using higher-order functions and/or chains of composed functions without explicit arguments, like:
mostFrequentWord
= head . maximumBy (comparing length) . group . sort . words
and definitions like sqr'''
are more in line with this overall style.
In terms of differences in meaning between these forms, it's actually a little complicated. For obscure reasons having to do with things called "the monomorphism restriction" and "defaulting rules", if you took the following module:
module Square where
sqr' = \x -> x * x
sqr'' x = x * x
sqr''' = (^2)
and compiled it with ghc -O
, the definitions of sqr'
and sqr'''
would be equivalent -- both would be specialized to operate on the Integer
type and would generate exactly the same code. (Tested with GHC 8.0.2). In contrast, sqr''
remains polymorphic with signature Num a => a -> a
, meaning it can operate on any numeric type.
If you add top-level type signatures (good practice anyway!), like so:
module Square where
sqr', sqr'', sqr''' :: (Num a) => a -> a
sqr' = \x -> x * x
sqr'' x = x * x
sqr''' = (^2)
then they all generate exactly the same code. You can verify this yourself by peeking at the generated "core" (the intermediate Haskell-like language that the compiler creates as a midpoint in the compilation process) using:
ghc -O -ddump-simpl -dsuppress-all -fforce-recomp Square.hs
In the generated core, you'll see the definition:
sqr' = \ @ a_aBC $dNum_aLW x_arx -> * $dNum_aLW x_arx x_arx
which looks weird, but basically says, apply the *
operation for the appropriate Num
type to the arguments x_arx x_arx
. The generated code for the two variants:
sqr'' = sqr'
sqr''' = sqr'
shows that GHC sees no difference between them and sqr'
, and so there will be no semantic or performance difference.
The technical post webpages of this site follow the CC BY-SA 4.0 protocol. If you need to reprint, please indicate the site URL or the original address.Any question please contact:yoyou2525@163.com.