Re: [PATCH] attributes: target_clone expects a string argument

[Martin Sebor via Gcc-patches]

On 5/13/21 6:05 AM, Martin Liška wrote:

Hello.

The change is about error handling.

Ready to be installed?
Thanks,
Martin

     PR middle-end/100504

gcc/c-family/ChangeLog:

     * c-attribs.c (handle_target_clones_attribute): Expect a string
     argument to target_clone argument.

gcc/testsuite/ChangeLog:

     * gcc.target/i386/pr100504.c: New test.
---
  gcc/c-family/c-attribs.c                 | 6 ++++++
  gcc/testsuite/gcc.target/i386/pr100504.c | 7 +++++++
  2 files changed, 13 insertions(+)
  create mode 100644 gcc/testsuite/gcc.target/i386/pr100504.c

diff --git a/gcc/c-family/c-attribs.c b/gcc/c-family/c-attribs.c
index c1f652d1dc9..9905ee56947 100644
--- a/gcc/c-family/c-attribs.c
+++ b/gcc/c-family/c-attribs.c

@@ -5300,6 +5300,12 @@ handle_target_clones_attribute (tree *node, tree name, tree ARG_UNUSED (args),

             "with %qs attribute", name, "target");
        *no_add_attrs = true;
      }
+      else if (TREE_CODE (TREE_VALUE (args)) != STRING_CST)
+    {
+      error ("%qE attribute argument not a string constant", name);
+      *no_add_attrs = true;
+    }

Since errors are higher priority than warnings I'd suggest making
this the first check, before the warnings above, (and adding a test
to verify that that's how it works).

Martin

+
        else
        /* Do not inline functions with multiple clone targets.  */
      DECL_UNINLINABLE (*node) = 1;

diff --git a/gcc/testsuite/gcc.target/i386/pr100504.c b/gcc/testsuite/gcc.target/i386/pr100504.c

new file mode 100644
index 00000000000..2910dfb948b
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr100504.c
@@ -0,0 +1,7 @@
+/* PR middle-end/100504 */
+/* { dg-do compile } */
+
+__attribute__((target_clones(0)))
+foo()

+{ /* { dg-error ".target_clones. attribute argument not a string constant" } */

+}