possible real bug in 6.2.2

Dear GHC team,

My investigation of why a certain list field in a record
does not print lazily has lead to something like a bug report
for ghc-6.2.2.

The program example of kind  

    let f      = ...
        mbPair = ...
    in
    f $ case mbPair of Just _ -> error "Lemma found\n" 
                       _      -> ProofResult {field = "abc"}

never reaches the `Just' branch,
and replacing the `case' construct with  (ProofResult ...)
changes the whole result greatly.
The effect is due to a complex code for  mbPair, 
its intensive work with memory, etc.;
and the `case' replacement avoids the mbPair computation.

(can all this happen without a bug?)

I am trying to reduce the example. The program is too complex.
I need greatly a reliable  ghc-6.2.2.xx  tool.  

-----------------
Serge Mechveliani
mechvel@xxxxxxxx