[Bug c++/35056] [4.3 Regression] ICE in copy_to_mode_reg, at explow.c:621

["aoliva at gcc dot gnu dot org" <gcc-bugzilla@gcc.gnu.org>]

------- Comment #16 from aoliva at gcc dot gnu dot org  2008-02-05 03:35 -------
Jakub, build_identity_conv is correct, at least in this case.  In C++, the
bitfieldness :-) of a variable is not to be taken into account for purposes of
overload resolution.  So, when tfrom != from, this means it is an identity
conversion, but one that C++ represents internally using the variant bitfield
type, which leads to this sort of problem.

I'm a bit concerned about adding the type conversion to build_target_expr(),
because of the comment above force_target_expr():
/* Like the above function, but without the checking.  This function should
   only be used by code which is deliberately trying to subvert the type
   system, such as call_builtin_trap.  */

Forcing a conversion might break the deliberate expectations mentioned there,
although I can't see that such expectations are still present in the current
code.  There are only two callers of force_target_expr(), and I don't see that
either has anything to do with subverting the type system.

Anyhow, I think it might be more appropriate to add the type conversion to
build_target_expr_with_type(), which most of C++ uses anyway, and where the
desired type is explicitly specified.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35056