[PATCH] AST: Mangle nullptr properly in template args for the Itanium ABI

[David Majnemer]

On Sat, Sep 7, 2013 at 10:35 AM, Richard Smith <richard at metafoo.co.uk>wrote:

>
> On 7 Sep 2013 02:39, "David Majnemer" <david.majnemer at gmail.com> wrote:
> >
> > Hi rsmith, rjmccall, eli.friedman, doug.gregor,
> >
> > We would previously treat nullptr literals as we were mangling the for
> > the null pointer value of the template parameter's type instead.
> >
> > This fixes PR17141.
> >
> > http://llvm-reviews.chandlerc.com/D1625
> >
> > Files:
> >   lib/AST/ItaniumMangle.cpp
> >   test/CodeGenCXX/mangle-nullptr-arg.cpp
> >   test/CodeGenCXX/visibility.cpp
> >
> > Index: lib/AST/ItaniumMangle.cpp
> > ===================================================================
> > --- lib/AST/ItaniumMangle.cpp
> > +++ lib/AST/ItaniumMangle.cpp
> > @@ -3337,10 +3337,8 @@
> >      break;
> >    }
> >    case TemplateArgument::NullPtr: {
> > -    //  <expr-primary> ::= L <type> 0 E
> > -    Out << 'L';
> > -    mangleType(A.getNullPtrType());
> > -    Out << "0E";
> > +    //  <expr-primary> ::= L <nullptr type> E
> > +    Out << "LDnE";
> >      break;
> >    }
> >    case TemplateArgument::Pack: {
> > Index: test/CodeGenCXX/mangle-nullptr-arg.cpp
> > ===================================================================
> > --- test/CodeGenCXX/mangle-nullptr-arg.cpp
> > +++ test/CodeGenCXX/mangle-nullptr-arg.cpp
> > @@ -2,15 +2,15 @@
> >
> >  template<int *ip> struct IP {};
> >
> > -// CHECK-LABEL: define void @_Z5test12IPILPi0EE
> > +// CHECK-LABEL: define void @_Z5test12IPILDnEE
> >  void test1(IP<nullptr>) {}
>
> The old mangling was correct here.
>
> >  struct X{ };
> >  template<int X::*pm> struct PM {};
> >
> > -// CHECK-LABEL: define void @_Z5test22PMILM1Xi0EE
> > +// CHECK-LABEL: define void @_Z5test22PMILDnEE
> >  void test2(PM<nullptr>) { }
>
> And here. You should only change the mangling of a null pointer of type
> std::nullptr_t.
>
I considered only changing the mangling for std::nullptr_t type arguments
but the following wording in the standard discouraged me:

"The pointer literal nullptr is encoded as "LDnE"."

I don't see where in the ABI we get the latitude to mangle this as-if it
were (param-type)0. That said there are obvious problems with giving all of
them the "LDnE" mangling. I will only use the different mangling for
nullptr_t template parameters.

Also, maybe this ABI break should be listed in the release notes.
>
Will do.

> Does GCC implement this rule yet?
>
They do not but icc 13.0.1 does.


> > -// CHECK-LABEL: define void @_Z5test316DependentTypePtrIPiLS0_0EE
> > +// CHECK-LABEL: define void @_Z5test316DependentTypePtrIPiLDnEE
> >  template<typename T, T x> struct DependentTypePtr {};
> >  void test3(DependentTypePtr<int*,nullptr>) { }
> > Index: test/CodeGenCXX/visibility.cpp
> > ===================================================================
> > --- test/CodeGenCXX/visibility.cpp
> > +++ test/CodeGenCXX/visibility.cpp
> > @@ -971,8 +971,8 @@
> >    void f() {
> >      zed<nullptr>();
> >    }
> > -  // CHECK-LABEL: define internal void
> @_ZN6test523zedILPNS_12_GLOBAL__N_13fooE0EEEvv
> > -  // CHECK-HIDDEN-LABEL: define internal void
> @_ZN6test523zedILPNS_12_GLOBAL__N_13fooE0EEEvv
> > +  // CHECK-LABEL: define internal void @_ZN6test523zedILDnEEEvv
> > +  // CHECK-HIDDEN-LABEL: define internal void @_ZN6test523zedILDnEEEvv
> >  }
> >
> >  namespace test53 {
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20130907/24ce863d/attachment.html>