Re: [Haskell-cafe] How can we detect and fix memory leak due to lazyness?

kyagrd:

On 8/7/06, Spencer Janssen wrote:

...

Forcing evaluation using (==) is a bit of a hack. Luckily, we have a better function to force evaluation: seq (which has type a -> b -> b). "seq x y" evaluates "x" to weak head normal form before returning "y".

Let's try another feature of Haskell to force evaluation: strict data fields. A ! in front of a field in a data declaration signifies strictness. In the example below, whenever we construct a value with TT, the second argument is evaluated.

\begin{code} data TT a b = TT a !b \end{code}

Perhaps your instances will work correctly with this data declaration?

Surely I've tried that.

Unfortunately seq and the strict data declaration is not helpful in general. They are only helpful on base values such as Int or Bool. What they do is just making sure that it is not a thunk. That is if it was a list it would just evaluate to see the cons cell but no further.

Someone wrote a deepSeq module for forcing deep evaluation, which is like doing self equality strictness hack like x==x. However, we should be able to locate what is the source of the memory leak to apply such strictness tricks.

The key is to profile. Compile the code, with optimisations on, with -prof -auto-all, then run the resulting program with +RTS -p -RTS. This will identify costly and timely functions. You can then refine the search further with {-# SCC "line1" #-} pragmas, next to expressoins you want to check the cost of. -- Don