Bug report for ghci: using Word64

[Julian Seward (Intl Vendor)]

We never really implemented full 64-bit support in the interpreter.
We had a scheme for doing that, and some infrastructure was put in
post 5.02.1, but connecting it all up for 5.02.2 (released today?)
was considered too disruptive for a patchlevel release, so I guess
it will only see the light of day when it is made to work in the
head (current development branch).  + to tell the truth I think
we just forgot all about it.  Oops!

The interpreter (ghci) will have problems if you ask it to compile
code which 

* deals directly with unboxed 64 bit ints   (Int64#)
* passes them between compiled and interpreted code

Note (perhaps critically) that boxed 64-bit Ints (Int64)
are harmless, provided that the bytecode generator never
has to see them unboxedly.  So you may be able to workaround
by cramming all your low-level 64-bit stuff into a single
module or group thereof, which you compile with ghc,
and then load into ghci, with the rest of your modules 
interpreted.  This will only work if the compiled modules
don't have Int64# in the signatures of any exported fns,
*and* no unfoldings which might surreptitiously expose
Int64#s are exported either.  Ensuring the latter is 
difficult at best.

J


| -----Original Message-----
| From: Laura McKinney [mailto:laura at galois.com] 
| Sent: Monday, January 07, 2002 4:47 PM
| To: glasgow-haskell-bugs at haskell.org
| Subject: Bug report for ghci: using Word64
| 
| 
| We are using Word64 and encountered the following problem 
| upon loading into 
| ghci:
| 
| ghc-5.02.1: panic! (the `impossible' happened, GHC version 5.02.1):
| 	ByteCodeGen.mkUnpackCode LW_
| 
| Program works under ghc.
| 
| Laura McKinney
| 
| _______________________________________________
| Glasgow-haskell-bugs mailing list
| Glasgow-haskell-bugs at haskell.org
| http://www.haskell.org/mailman/listinfo/glasgow-haskell-bugs
|