Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!zaphod.mps.ohio-state.edu!usc!orion.oac.uci.edu!uci-ics!gateway
From: rfg@paris.ics.uci.edu (Ronald Guilmette)
Newsgroups: comp.lang.c++
Subject: cfront 2.0 bug 900210_01
Message-ID: <25DA62F0.22077@paris.ics.uci.edu>
Date: 15 Feb 90 08:06:08 GMT
Organization: UC Irvine Department of ICS
Lines: 41

// cfront 2.0 bug 900210_01

// The following c++ code causes cfront to generate C code that will be
// rejected by any sane C compiler.

// The problem seems to be that cfront is generating a conditional
// expression using operator ?: where the condition being checked
// is merely whether or not the object has been allocated successfully
// by "new".  If not, the entire conditional expression yields the value
// of the final "expression" in the constructor (in this case, a member
// function pointer value which is represented in C as a structure)
// whereas if the allocation was unsuccessful, the value 0 is yielded.

// For the following case, the value yielded by the *entire* conditional
// expression is never even used in the generated code, but the type
// mismatch between the second and third operands of the ?: trinary operator
// (in the generated code) cases the entire ?: expression to be invalid.

// Cfront knows that this could be a problem, and uses a special trick
// to avoid the problem when and if the final expression in the constructor
// body yields a user-declared structure.  Unfortunately, it forgets to do
// this also for member function pointers (also represented as structs in the
// generated C code).

struct struct0 {
};

typedef (struct0::*MFP)();

MFP global_mfp;

struct B {
  B (MFP mfp) { global_mfp = mfp; }
};

void foo()
{
  new B (global_mfp);
}

int main () { return 0; }