C++: __java_double expressions promote to float

Wed Apr 10 09:45:00 GMT 2002

Jason Merrill writes:
 > I told you in person this was OK, but reading over the function again, I
 > see that we don't need to add explicit __java_double support, we just need
 > to stop assuming that any REAL_TYPE not long double or double must be
 > float. The 'else' should be an 'if' like the others, as specified in the
 > standard. The default should be to arbitrarily choose one of the two
 > types. Or perhaps use type_for_mode.
2002年04月10日 Andrew Haley <aph@cambridge.redhat.com>
	* typeck.c (type_after_usual_arithmetic_conversions): 
	When two floating-point operands have the same precision: 
	 convert to float if one of the operands is float;
	 if neither operand is one of the standard types, return the type
	 of the first operand.
 
Index: typeck.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/cp/typeck.c,v
retrieving revision 1.393
diff -c -2 -p -r1.393 typeck.c
*** typeck.c	22 Mar 2002 22:03:04 -0000	1.393
--- typeck.c	10 Apr 2002 14:15:52 -0000
*************** type_after_usual_arithmetic_conversions 
*** 443,449 ****
 	return build_type_attribute_variant (double_type_node,
 					 attributes);
! else 
 	return build_type_attribute_variant (float_type_node,
 					 attributes);
 }
 }
--- 443,457 ----
 	return build_type_attribute_variant (double_type_node,
 					 attributes);
! if (same_type_p (TYPE_MAIN_VARIANT (t1), float_type_node)
! 	 || same_type_p (TYPE_MAIN_VARIANT (t2), float_type_node))
 	return build_type_attribute_variant (float_type_node,
 					 attributes);
+ 
+ /* Two floating-point types whose TYPE_MAIN_VARIANTs are none of
+ the standard C++ floating-point types. Logic earlier in this
+ function has already eliminated the possibility that
+ TYPE_PRECISION (t2) != TYPE_PRECISION (t1), so there's no
+ compelling reason to choose one or the other. */
+ return build_type_attribute_variant (t1, attributes);
 }
 }
 > Also, the function should check for the types actually being the same up
 > where it tests for different precisions; no need to get fancy in that case.
Pardon?
Andrew.