[Haskell-cafe] Baffled by Disk IO

[Matthew Bromberg]

I have run into a baffling distinction between the behavior of GHCi and 
the compiled binary from GHC.
I suspect it's something pretty stupid on my part.

I have the following test program
<test.hs>
import Matrix

main = let
    f1 = bRgauss 4 3
    f2 = bRgauss 3 2
    fu = f1 *. f2   
    in bsave "fumat" fu
----------------------------------
The type signature of bsave is

   
    -- | save a matrix to a file
    bsave :: String -> m -> IO()
    bsave str z = do
        k <- matindx z
        withCString str (\x -> bmsave x k)

and bmsave is a C routine that does a simple write to disk using fprintf. 
If I compile using GHC and run this as test.exe it does absolutely 
nothing, no file is actually saved.  In fact no file is saved even
if I try to print components of fu in the IO monad.

Whereas if I load it into GHCi and run main, everything works as 
expected.  Well sort of.  Apparently everytime I extract something out
of fu and use it in an IO monad it executes all the IO actions in fu, so 
it's easy to get unwanted behavior with incautious use.