On Wed, Nov 24, 2021 at 11:42:28PM +0100, Jakub Jelinek via Gcc-patches wrote:
> > I'm surprised the source_location::current handling would be needed; why do
> > calls to that function live long enough for us to walk into the ADDR_EXPR
> > here? Maybe we should fold it in cp_fold instead of cp_genericize_r.
I've already committed the patch, but perhaps we shouldn't do it in cp_fold
where it will be folded even for warnings etc. and the locations might not
be the final yet. This patch moves it to cp_fold_r so that it is done just
once for each function and just once for each static initializer.
Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
2021-11-27 Jakub Jelinek <ja...@redhat.com>
* cp-gimplify.c (cp_fold_r): Perform folding of
std::source_location::current() calls here...
(cp_fold): ... rather than here.
--- gcc/cp/cp-gimplify.c.jj 2021-11-26 10:10:53.227121384 +0100
+++ gcc/cp/cp-gimplify.c 2021-11-26 11:51:50.945204127 +0100
@@ -930,6 +930,13 @@ cp_fold_r (tree *stmt_p, int *walk_subtr
}
break;
+ case CALL_EXPR:
+ if (tree fndecl = cp_get_callee_fndecl_nofold (stmt))
+ if (DECL_IMMEDIATE_FUNCTION_P (fndecl)
+ && source_location_current_p (fndecl))
+ *stmt_p = stmt = cxx_constant_value (stmt);
+ break;
+
default:
break;
}
@@ -2672,14 +2679,6 @@ cp_fold (tree x)
int sv = optimize, nw = sv;
tree callee = get_callee_fndecl (x);
- if (tree fndecl = cp_get_callee_fndecl_nofold (x))
- if (DECL_IMMEDIATE_FUNCTION_P (fndecl)
- && source_location_current_p (fndecl))
- {
- x = cxx_constant_value (x);
- break;
- }
-
/* Some built-in function calls will be evaluated at compile-time in
fold (). Set optimize to 1 when folding __builtin_constant_p inside
a constexpr function so that fold_builtin_1 doesn't fold it to 0. */
Jakub
