[4bsd-f77 #43] F77 scrambles the cmplx() intrinsic function

[4.2 BSD f77 bug reports]

From: Donn Seeley <donn at utah-cs.arpa>

Subject: F77 scrambles the cmplx() intrinsic function
Index:	usr.bin/f77/src/f77pass1/expr.c 4.2BSD

Description:
	F77 mutilates expressions that use the intrinsic function
	cmplx() to convert two REAL or DOUBLE PRECISION values into a
	COMPLEX value.  In particular the imaginary part of the value
	is lost.  This bug was found by lanl-a!pam (whose real name I
	never seem to have discovered).

Repeat-By:
	Compile the following program from lanl-a!pam:

	----------------------------------------------------------------
	      complex a, b

	      a = (1.0, 2.0)
	      b = cmplx (abs(real(a)), abs(imag(a)))

	      print *, a, b

	      stop
	      end
	----------------------------------------------------------------

	When run it prints:

	----------------------------------------------------------------
	  (   1.00000,   2.00000)  (   1.00000,  0.)
	----------------------------------------------------------------

	It's supposed to print:

	----------------------------------------------------------------
	  (   1.00000,   2.00000)  (   1.00000,   2.00000)
	----------------------------------------------------------------

Fix:
	The problem is that fixexpr(), the routine which handles the
	manipulation and checking of expression trees, does not know
	about complex conversions with two operands.  This might be
	okay if it just left trees like this alone, but it has an
	optimization it wants to try and of course it amputates the
	imaginary part of the conversion when it rewrites the tree.  In
	general the compiler does not seem well equipped to deal with
	complex conversions with two operands -- I wouldn't be
	surprised if there were more bugs of this sort.  The following
	changes to fixexpr() and cktype() in expr.c stop the butchery,
	for at least the example program:

	----------------------------------------------------------------
	*** /tmp/,RCSt1029508	Mon Aug 20 20:48:32 1984
	--- expr.c	Sun Aug  5 23:06:34 1984
	***************
	*** 659,664
	  
	  
		case OPCONV:
			ptype = cktype(OPCONV, p->vtype, ltype);
			if(lp->tag==TEXPR && lp->exprblock.opcode==OPCOMMA)
				{

	--- 663,675 -----
	  
	  
		case OPCONV:
	+ 		if(ISCOMPLEX(p->vtype))
	+ 			{
	+ 			ptype = cktype(OPCONV, p->vtype, ltype);
	+ 			if(p->rightp)
	+ 				ptype = cktype(OPCONV, ptype, rtype);
	+ 			break;
	+ 			}
			ptype = cktype(OPCONV, p->vtype, ltype);
			if(lp->tag==TEXPR && lp->exprblock.opcode==OPCOMMA)
				{
	***************
	*** 1950,1955
			return(TYADDR);
	  
		case OPCONV:
			if(rt == 0)
				return(0);
			if(lt==TYCHAR && ISINT(rt) )

	--- 1961,1972 -----
			return(TYADDR);
	  
		case OPCONV:
	+ 		if(ISCOMPLEX(lt))
	+ 			{
	+ 			if(ISNUMERIC(rt))
	+ 				return(lt);
	+ 			ERR("impossible conversion")
	+ 			}
			if(rt == 0)
				return(0);
			if(lt==TYCHAR && ISINT(rt) )
	----------------------------------------------------------------

Donn Seeley    University of Utah CS Dept    donn at utah-cs.arpa
40 46' 6"N 111 50' 34"W    (801) 581-5668    decvax!utah-cs!donn