Question on the efficiency of binding record values to variables

Learn
1

Suppose I have the following record type and a constructed value of that type:

data Point = P
    { x :: Double
    , y :: Double
    }
point = P {x = 1, y = 2}

and I want to bind x and y values of this point to new variables px and py. Which of the following two versions is more efficient?

P {x = px, y = py} = point

or

px = x point
py = y point

So far I have always used the latter variant because I did not know the first variant. Am I correct in assuming that the first variant is more efficient since both values are bound at once?

2

I’d like to point out that, as well as using record syntax like this:

point = P {x = 1, y = 2}
P {x = px, y = py} = point

If you are assigning to or from all of the fields, you can also just do this:

point = P 1 2
P px py = point

I don’t know which compiles to more efficient output. I’d imagine that any differences between the versions would be easy to optimise out, making them all the same, but that’s just a guess.

1 Like
3

My guess is that it gets optimized by the compiler and no difference (in efficiency) will show up.

Consider this example:

point = P 3 4
a = point
b = a
P c d = b
e = P c d
f = e
putStrln $ show f

I believe it produces the same executable as

putStrln $ show (P 3 4)

Maybe some ghc/optimisation guru can chime in and confirm: there is no “execution” here

In your assumption “since both values are bound at once?”, you may overlook that binding may be a compile-time trick, not an execution-time, CPU-consuming operation

1 Like
4

As others have guessed (and I just checked), both versions are exactly the same at runtime. What’s more, the runtime cost is zero in both cases. :slight_smile: All you’re doing is creating different names for existing values — useful for humans, but unnecessary for machines. It’s not like any memory is getting passed around or anything.

5 Likes