https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120453
Iain Sandoe <iains at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |jason at gcc dot gnu.org
--- Comment #9 from Iain Sandoe <iains at gcc dot gnu.org> ---
OK so the reason this fails is because check_return_expr() concludes that we
cannot do NRVO and we try to do the bad conversion.
There are two reasons that it makes this conclusion.
1. We actually need a "DECL_RAMP_P()" - that is easy to fix (i have a fix).
#define DECL_RAMP_P(NODE) \
DECL_COROUTINE_P (NODE) && !DECL_RAMP_FN (NODE)
2. named_return_value_okay_p is false
because want_nrvo_p() is false
because aggregate_value_p () is false
because TYPE_EMPTY_P() is true for the return type.
Now ISTM that we can elide the conversion in the case that TYPE_EMPTY_P() ==
true (assuming that can_do_nrvo_p() would be true otherwise).
so... updating the condition to:
if (DECL_RAMP_P (current_function_decl)
&& (named_return_value_okay_p
|| (can_do_nrvo_p (bare_retval, functype)
&& TYPE_EMPTY_P (functype))))
converted = true;
@jason - does this seem reasonable, or are there yet other cases we should
cater for?
