On Wed, Nov 28, 2018 at 07:35:42PM +0100, Paolo Carlini wrote:
> Hi,
>
> On 28/11/18 19:05, Marek Polacek wrote:
> > Paolo's recent commit changed
> >
> > - pedwarn (input_location, OPT_Wpedantic,
> > + pedwarn (declarator->id_loc, OPT_Wpedantic,
> >
> > but as this testcase shows, declarator might not always be present. So
> > let's brace ourselves for that.
>
> Thanks Marek. Just to clarify a bit more, ultimately the reason why I missed
> this case is that, when the declarator is null - for the testcase at issue,
> that is - we end up printing something of the form:
>
> warning: ISO C++ forbids flexible array member ‘type name’
>
> thus, all in all, I'm wondering if it wouldn't be a tad less cryptic to do
> something like the attached., which avoids printing "type name".
I think you're right, so this is another version:
2018-11-28 Marek Polacek <pola...@redhat.com>
PR c++/88222 - ICE with bit-field with invalid type.
* decl.c (grokdeclarator): Check if declarator is null.
* g++.dg/ext/flexary31.C: New test.
diff --git gcc/cp/decl.c gcc/cp/decl.c
index 1aaf51750ab..3734bfe39ac 100644
--- gcc/cp/decl.c
+++ gcc/cp/decl.c
@@ -12221,7 +12221,7 @@ grokdeclarator (const cp_declarator *declarator,
if (in_system_header_at (input_location))
/* Do not warn on flexible array members in system
headers because glibc uses them. */;
- else if (name)
+ else if (name && declarator)
pedwarn (declarator->id_loc, OPT_Wpedantic,
"ISO C++ forbids flexible array member %qs", name);
else
diff --git gcc/testsuite/g++.dg/ext/flexary31.C
gcc/testsuite/g++.dg/ext/flexary31.C
new file mode 100644
index 00000000000..90f8431a2a4
--- /dev/null
+++ gcc/testsuite/g++.dg/ext/flexary31.C
@@ -0,0 +1,8 @@
+// PR c++/88222
+// { dg-options -Wno-pedantic }
+
+typedef char a[];
+
+class S {
+ a : 4; // { dg-error "bit-field" }
+};
