Sorry Daniel, i've seen your answer just after posting my reply. But i can't write !t1f <- gettime, nor t1f <- !gettime; because it's an IO ? Your method is a good help form me. Thanks, Didier Daniel Fischer a écrit :

Am Samstag 13 März 2010 21:20:58 schrieb legajid:

...

Hi, my following code should show time before executing listeprem, then time after execution.

import System.Time gettime :: IO ClockTime gettime = getClockTime

main=do t1d <- gettime

This gets the time, now.

...

let t1=last (listeprem 5000)

This doesn't do the calculation, it only creates a thunk what to calculate should t1 be demanded, takes a few nanoseconds

...

t1f <- gettime

This gets the time, a few nanoseconds later

...

putStrLn ("Methode 1 : " ++ show t1d) putStrLn (" " ++ show t1)

Now t1 is demanded, and only now it's calculated.

...

putStrLn (" " ++ show t1f)

Looking at the screen, t1d is displayed then, after a few seconds, t1 and t1f. But, t1d and t1f are equal. It seems like if t1d and t1f where calculated at start of procedure, before we need calculating t1 for putStrLn.

Exactly.

...

How can i have t1f evaluated after t1, so i can calculate time elapsed for calculation of t1?

Depends on what sort of value t1 is. If it's a simple value like an Integer (as I think it is here, derniere nombre premier avant 5000, 5000-th prime?),

{-# LANGUAGE BangPatterns #-} ... let !t1 = last (listeprem 5000) t1f <- gettime

or

let t1 = last (listeprem 5000) t1f <- t1 `seq` gettime

will do. For small values, printing takes negligible time, so

t1d <- gettime putStrLn (" " ++ show t1f) t1f <- gettime

is another option. But getClockTime isn't particularly exact, I'd suggest using System.CPUTime.getCPUTime

...

Thanks, Didier

I tried !t1f <- gettime and i got an error message : premier.hs:92:1: Illegal bang-pattern (use -XBangPatterns) Failed, modules loaded: none. t1f <- !gettime gives : premier.hs:92:8: parse error on input `!' Failed, modules loaded: none. Daniel Fischer a écrit :

Am Samstag 13 März 2010 22:02:41 schrieb legajid:

...

Sorry Daniel, i've seen your answer just after posting my reply.

Thought so. Happens to me often enough :)

...

But i can't write !t1f <- gettime,

You can, and it's much like

let !t1 = ...

for the pure calculation, it evaluates t1f to the outermost constructor. If deeper evaluation is needed, like for pure values, you have to use the appropriate forcing strategy.

!a <- mx

is less important for IO than for other monads, though, since IO imposes a fair amount of sequencing. In particular, the sequencing of IO makes

!t1f <- gettime

superfluous, since all that could be delayed in gettime is the transformation of the timestamp retrieved from the system clock to the ClockTime datatype, the timestamp is retrieved at the exact point in the IO-sequence.

...

nor t1f <- !gettime;

I'm not sure whether you can write that, but it wouldn't be what you want anyway. If that's legal BangPattern syntax, it says "evaluate gettime to the outermost constructor", which is far less than we need to evaluate gettime to bind its result to a name.

...

because it's an IO ? Your method is a good help form me.

Thanks, Didier