How does Haskell tail recursion work?

Question

I wrote this snippet of code and I assume len is tail-recursive, but a stack overflow still occurs. What is wrong?

myLength :: [a] -> Integer         


        

Answer 1

The simplest solution to your problem is turning on optimization.

I have your source in a file called tail.hs.

jmg$ ghc --make tail.hs -fforce-recomp
[1 of 1] Compiling Main             ( tail.hs, tail.o )
Linking tail ...
jmg$ ./tail 
Stack space overflow: current size 8388608 bytes.
Use `+RTS -Ksize -RTS' to increase it.
girard:haskell jmg$ ghc -O --make tail.hs -fforce-recomp
[1 of 1] Compiling Main             ( tail.hs, tail.o )
Linking tail ...
jmg$ ./tail 
10000000
jmg$ 

@Hynek -Pichi- Vychodil The tests above were done on Mac OS X Snow Leopard 64bit with a GHC 7 and GHC 6.12.1, each in a 32 bit version. After you're downvote, I repeated this experiment on Ubuntu Linux with the following result:

jmg@girard:/tmp$ cat length.hs
myLength :: [a] -> Integer

myLength xs = len xs 0
    where len [] l = l
          len (x:xs) l = len xs (l+1)

main = print $ myLength [1..10000000]

jmg@girard:/tmp$ ghc --version
The Glorious Glasgow Haskell Compilation System, version 6.12.1
jmg@girard:/tmp$ uname -a
Linux girard 2.6.35-24-generic #42-Ubuntu SMP Thu Dec 2 02:41:37 UTC 2010 x86_64 GNU/Linux
jmg@girard:/tmp$ ghc --make length.hs -fforce-recomp
[1 of 1] Compiling Main             ( length.hs, length.o )
Linking length ...
jmg@girard:/tmp$ time ./length 
Stack space overflow: current size 8388608 bytes.
Use `+RTS -Ksize -RTS' to increase it.

real    0m1.359s
user    0m1.140s
sys 0m0.210s
jmg@girard:/tmp$ ghc -O --make length.hs -fforce-recomp
[1 of 1] Compiling Main             ( length.hs, length.o )
Linking length ...
jmg@girard:/tmp$ time ./length 
10000000

real    0m0.268s
user    0m0.260s
sys 0m0.000s
jmg@girard:/tmp$ 

So, if you're interested we can continue to find out what is the reason, that this fails for you. I guess, GHC HQ, would accept it as a bug, if such a straight forward recursion over lists is not optimized into an efficient loop in this case.