From: Patrick Palka <ppalka@redhat.com>
To: gcc-patches@gcc.gnu.org
Cc: jason@redhat.com, Patrick Palka <ppalka@redhat.com>
Subject: [PATCH 1/2] c++: improve "NTTP argument considered unused" fix [PR53164, PR105848]
Date: Thu, 23 Mar 2023 17:18:02 -0400 [thread overview]
Message-ID: <20230323211803.396326-1-ppalka@redhat.com> (raw)
r13-995-g733a792a2b2e16 worked around the problem of FUNCTION_DECL
template arguments not always getting marked as odr-used by redundantly
calling mark_used on the substituted ADDR_EXPR callee of a CALL_EXPR.
This is just a narrow workaround however, since using a FUNCTION_DECL as
a template argument alone should constitutes an odr-use; we shouldn't
need to subsequently e.g. call the function or take its address.
This patch fixes this in a more general way at template specialization
time by walking the template arguments of the specialization and calling
mark_used on all entities used within. As before, the call to mark_used
as it worst a no-op, but it compensates for the situation where we end up
forming a specialization from a template context in which mark_used is
inhibited. Another approach would be to call mark_used whenever we
substitute a TEMPLATE_PARM_INDEX, but that would result in many more
redundant calls to mark_used compared to this approach.
Bootstrapped and regtested on x86_64-pc-linux-gnu, does this look OK for
trunk?
PR c++/53164
PR c++/105848
gcc/cp/ChangeLog:
* pt.cc (instantiate_class_template): Call
mark_template_arguments_used.
(tsubst_copy_and_build) <case CALL_EXPR>: Revert r13-995 change.
(mark_template_arguments_used): Define.
(instantiate_template): Call mark_template_arguments_used.
gcc/testsuite/ChangeLog:
* g++.dg/template/fn-ptr3a.C: New test.
* g++.dg/template/fn-ptr4.C: New test.
---
gcc/cp/pt.cc | 51 ++++++++++++++++--------
gcc/testsuite/g++.dg/template/fn-ptr3a.C | 25 ++++++++++++
gcc/testsuite/g++.dg/template/fn-ptr4.C | 14 +++++++
3 files changed, 74 insertions(+), 16 deletions(-)
create mode 100644 gcc/testsuite/g++.dg/template/fn-ptr3a.C
create mode 100644 gcc/testsuite/g++.dg/template/fn-ptr4.C
diff --git a/gcc/cp/pt.cc b/gcc/cp/pt.cc
index 7e4a8de0c8b..9b3cc33331c 100644
--- a/gcc/cp/pt.cc
+++ b/gcc/cp/pt.cc
@@ -220,6 +220,7 @@ static tree make_argument_pack (tree);
static tree enclosing_instantiation_of (tree tctx);
static void instantiate_body (tree pattern, tree args, tree d, bool nested);
static tree maybe_dependent_member_ref (tree, tree, tsubst_flags_t, tree);
+static void mark_template_arguments_used (tree);
/* Make the current scope suitable for access checking when we are
processing T. T can be FUNCTION_DECL for instantiated function
@@ -12142,6 +12143,9 @@ instantiate_class_template (tree type)
cp_unevaluated_operand = 0;
c_inhibit_evaluation_warnings = 0;
}
+
+ mark_template_arguments_used (INNERMOST_TEMPLATE_ARGS (args));
+
/* Use #pragma pack from the template context. */
saved_maximum_field_alignment = maximum_field_alignment;
maximum_field_alignment = TYPE_PRECISION (pattern);
@@ -21173,22 +21177,10 @@ tsubst_copy_and_build (tree t,
}
/* Remember that there was a reference to this entity. */
- if (function != NULL_TREE)
- {
- tree inner = function;
- if (TREE_CODE (inner) == ADDR_EXPR
- && TREE_CODE (TREE_OPERAND (inner, 0)) == FUNCTION_DECL)
- /* We should already have called mark_used when taking the
- address of this function, but do so again anyway to make
- sure it's odr-used: at worst this is a no-op, but if we
- obtained this FUNCTION_DECL as part of ahead-of-time overload
- resolution then that call to mark_used wouldn't have marked it
- odr-used yet (53164). */
- inner = TREE_OPERAND (inner, 0);
- if (DECL_P (inner)
- && !mark_used (inner, complain) && !(complain & tf_error))
- RETURN (error_mark_node);
- }
+ if (function != NULL_TREE
+ && DECL_P (function)
+ && !mark_used (function, complain) && !(complain & tf_error))
+ RETURN (error_mark_node);
if (!maybe_fold_fn_template_args (function, complain))
return error_mark_node;
@@ -21883,6 +21875,31 @@ check_instantiated_args (tree tmpl, tree args, tsubst_flags_t complain)
return result;
}
+/* Call mark_used on each entity within the template arguments ARGS of some
+ template specialization, to ensure that each such entity is considered
+ odr-used regardless of whether the specialization was first formed in a
+ template context.
+
+ This function assumes push_to_top_level has been called beforehand, and
+ that processing_template_decl has been set iff the template arguments
+ are dependent. */
+
+static void
+mark_template_arguments_used (tree args)
+{
+ gcc_checking_assert (TMPL_ARGS_DEPTH (args) == 1);
+
+ if (processing_template_decl)
+ return;
+
+ auto mark_used_r = [](tree *tp, int *, void *) {
+ if (DECL_P (*tp))
+ mark_used (*tp, tf_none);
+ return NULL_TREE;
+ };
+ cp_walk_tree_without_duplicates (&args, mark_used_r, nullptr);
+}
+
/* We're out of SFINAE context now, so generate diagnostics for the access
errors we saw earlier when instantiating D from TMPL and ARGS. */
@@ -22012,6 +22029,8 @@ instantiate_template (tree tmpl, tree orig_args, tsubst_flags_t complain)
push_nested_class (ctx);
}
+ mark_template_arguments_used (INNERMOST_TEMPLATE_ARGS (targ_ptr));
+
tree pattern = DECL_TEMPLATE_RESULT (gen_tmpl);
fndecl = NULL_TREE;
diff --git a/gcc/testsuite/g++.dg/template/fn-ptr3a.C b/gcc/testsuite/g++.dg/template/fn-ptr3a.C
new file mode 100644
index 00000000000..7456be5d51f
--- /dev/null
+++ b/gcc/testsuite/g++.dg/template/fn-ptr3a.C
@@ -0,0 +1,25 @@
+// PR c++/53164
+// PR c++/105848
+// A stricter version of fn-ptr3.C that verifies using f as a template
+// argument alone constitutes an odr-use.
+
+template<class T>
+void f(T) { T::fail; } // { dg-error "fail" }
+
+template<void (*P)(int)>
+struct A {
+ static void wrap();
+};
+
+template<void (&P)(char)>
+void wrap();
+
+template<int>
+void g() {
+ A<f>::wrap(); // { dg-message "required from here" }
+ wrap<f>(); // { dg-message "required from here" }
+}
+
+int main() {
+ g<0>();
+}
diff --git a/gcc/testsuite/g++.dg/template/fn-ptr4.C b/gcc/testsuite/g++.dg/template/fn-ptr4.C
new file mode 100644
index 00000000000..e7425ba96cc
--- /dev/null
+++ b/gcc/testsuite/g++.dg/template/fn-ptr4.C
@@ -0,0 +1,14 @@
+// { dg-do compile }
+
+template<void (*P)()>
+void wrap() {
+ P(); // OK, despite A::g not being accessible from foo.
+}
+
+struct A {
+ static void f() {
+ wrap<A::g>();
+ }
+private:
+ static void g();
+};
--
2.40.0.130.g27d43aaaf5
next reply other threads:[~2023-03-23 21:18 UTC|newest]
Thread overview: 13+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-03-23 21:18 Patrick Palka [this message]
2023-03-23 21:18 ` [PATCH 2/2] c++: duplicate "use of deleted fn" diagnostic [PR106880] Patrick Palka
2023-03-30 2:58 ` Jason Merrill
2023-03-23 21:26 ` [PATCH 1/2] c++: improve "NTTP argument considered unused" fix [PR53164, PR105848] Patrick Palka
2023-03-27 13:30 ` Patrick Palka
2023-03-29 21:36 ` Jason Merrill
2023-03-30 18:53 ` Patrick Palka
2023-03-30 21:39 ` Jason Merrill
2023-03-31 15:25 ` Patrick Palka
2023-03-31 15:55 ` Patrick Palka
2023-03-31 17:28 ` Jason Merrill
2023-03-31 19:06 ` Patrick Palka
2023-04-01 3:50 ` Jason Merrill
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20230323211803.396326-1-ppalka@redhat.com \
--to=ppalka@redhat.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=jason@redhat.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).