[ Cross-mailed to ghc-users, too, because this is not really a bug ]
Jan Kort wrote:
> It seems that FFI no longer accepts "ByteArray Int" in ghc 4.06
> (it does in ghc 4.04), it looks intentional ?
Even it is unintentional now, it will be omitted in future versions
when the new FFI design is finished.
> If so, is there another way to pass a String to a C function ?
> [...]
Of course! We (the fearless FFI task force) are not *that* evil. :-)
The current state of affairs is as follows:
* There has been consensus on the low-level part of the new FFI,
and 4.06 comes with a new module FFI (don't forget -syslib lang)
implementing the library part of it. There are no docs yet, but
I'll promise to fix that "soon". Until then:
"Use the source, Luke!"
* There will be some changes to the GHC/Hugs-specific modules
Foreign/Stable/Addr/Int/Word, some of which have already been
incorporated in 4.06 (but only the ones unlikely to break any
existing programs).
* We still have to do some design work for the higher-level libs,
which would be exactly the part needed for (un-)marshaling
strings.
More details can be found on
http://www.informatik.uni-muenchen.de/~Sven.Panne/haskell_libs/ffi.html
A *very* preliminary and incomplete version for the higher-level FFI
stuff can be found there (module Marshal). Manuel Chakravarty already
had some new ideas and improvements, which I was too lazy to merge into
Marshal (yet).
To get you going and have an easy cut'n'paste version for the common
task of (un-)marshaling strings, I've specialised some functions from
Marshal and extended your example with them:
---------------------------------------------------------
module Main where
import FFI
import Monad(zipWithM_)
foreign import ccall "c_test" h_test :: Addr -> IO ()
foreign import ccall ttyname :: Int -> IO Addr
callWithString :: (Addr -> IO a) -> String -> IO a
callWithString act str = do
let numElements = length str
buf <- mallocElems (head str) (numElements+1)
zipWithM_ (pokeElemOff buf) [ 0 .. ] str
pokeElemOff buf numElements '\0'
val <- act buf
free buf
return val
getStringFrom :: IO Addr -> IO String
getStringFrom act = do
buf <- act
let loop idx accu = do x <- peekElemOff buf idx
if x == '\0'
then return $ reverse accu
else loop (idx+1) (x:accu)
loop 0 []
main :: IO()
main = do
callWithString h_test "test123"
print =<< getStringFrom (ttyname 0)
---------------------------------------------------------
Now you know, why FFI is only a *low*-level interface... :-]
Cheers,
Sven
P.S.: Because of the readXYOffAddr bug in 4.06 discussed a few days
ago, you'd better compile without -O for now.
--
Sven Panne Tel.: +49/89/2178-2235
LMU, Institut fuer Informatik FAX : +49/89/2178-2211
LFE Programmier- und Modellierungssprachen Oettingenstr. 67
mailto:[EMAIL PROTECTED] D-80538 Muenchen
http://www.informatik.uni-muenchen.de/~Sven.Panne
