> > We're groping around in the dark here
>
> Sorry, yes we are. It is a bit cheeky of me to always punt and say
> "I'll bet it is a GHC problem" when I haven't even seen the bug myself
> on a machine whose configuration I know. Sorry about that.
No problem :)
> In fact, it turns out that you've uncovered a bug in the ffi backend
> of GreenCard. The ghc backend (and GHC itself) is fine though so I'd
> guess that Sigbjorn tested this release with the ghc backend but since
> then the Makefile (or something in the hslibs Makefile hierarchy)
> switched to using the ffi backend.
>
> I'll set about fixing this right away.
>
> In the meantime, adding this line:
>
> SRC_GC_OPTS += --target ghc
Actually the win32 library is pre-compiled into .hs files in the
repository (I can't remember why we did this, but I suspect it was to
remove the dependency on GreenCard when building GHC on Windows).
> to hslibs/win32/Makefile may fix the problem (but I haven't been able
> to test this myself).
>
> I'm attaching details for those who want them.
>
> --
> Alastair Reid
>
>
> [Explanation of how word8 is currently marshalled by
> GreenCard deleted]
>
> > On a big-endian machine I'd expect this to break. Does it?
>
> Digging deeper into my memory of GreenCard, the answer should be no
> and the could should be perfectly safe (the word "should" is
> key here).
>
> Greencard is supposed to turn each %fun into a pair of functions - one
> Haskell, one C. They are supposed to comminucate with each other
> through a common set of types which includes Word32 but excludes
> Word8. To send a Word8 from one to the other we:
>
> Convert the word8 to word32.
> Hand the word32 across the border.
> Convert the word32 back into a word8.
Ok, this makes sense now.
> This is what happens in the Hugs backend
>
> primFun(prim_foo_rgb)
> { BYTE arg1; BYTE arg2; BYTE arg3;DWORD res1;;
> arg1 = (BYTE)(hugs->getWord()); // THESE 3 LINES
> arg2 = (BYTE)(hugs->getWord()); // THESE 3 LINES
> arg3 = (BYTE)(hugs->getWord()); // THESE 3 LINES
> { res1 = RGB(arg1,arg2,arg3);
> hugs->putWord((DWORD)(res1));
> hugs_returnIO(1);
> }
> }
>
> and, through the magic of _casm_, in the ghc backend
>
> rgb :: BYTE -> BYTE -> BYTE -> COLORREF
> rgb gc_arg1 gc_arg2 gc_arg3 =
> unsafePerformIO(
> case ( word8ToWord32 gc_arg1) of { arg1 ->
> case ( word8ToWord32 gc_arg2) of { arg2 ->
> case ( word8ToWord32 gc_arg3) of { arg3 ->
> _casm_ ``do {BYTE arg1; BYTE arg2; BYTE arg3;DWORD res1;
> arg1 = (BYTE)%0; arg2 = (BYTE)%1; arg3 = (BYTE)%2;
> do { res1 = RGB(arg1,arg2,arg3);
>
> %r = (DWORD)(res1);} while(0);}
> while(0);'' arg1 arg2 arg3
> >>= \ res1 ->
> (return (res1))}}})
>
> but not in the ffi backend
>
> DWORD prim_rgb(BYTE arg1,BYTE arg2,BYTE arg3)
> { DWORD res1;
> do { res1 = RGB(arg1,arg2,arg3);
>
> return((DWORD)(res1));} while(0);
> }
Right, I see the problem now. Thanks for looking into it.
Cheers,
Simon
_______________________________________________
Glasgow-haskell-bugs mailing list
[EMAIL PROTECTED]
http://www.haskell.org/mailman/listinfo/glasgow-haskell-bugs
