[cfe-dev] CXX11 attribute spelling and LateParsed=1

[Javier López via cfe-dev]

I have an update on this, but I forgot to post it here.

The fix turned out to be simple.  Given that the attribute subject in the
Attr.td file is
  let Subjects = SubjectsList<[Function]>

The only thing to do is to move the type attribute to the attribute list of
the declarator.  That is accomplished by the following lines:

In lib/Sema/SemaType.cpp, function processTypeAttrs():

  case AttributeList::AT_TheAttribute:
    moveAttrFromListToList(attr, state.getCurrentAttrListRef(),
                           state.getDeclarator().getAttrListRef());
    break;

Regards.

El vie, 20-10-2017 a las 13:25 +0200, Javier López escribió:
> El mar, 10-10-2017 a las 22:48 -0600, Eric Fiselier escribió:
> > 
> > On Mon, Oct 9, 2017 at 7:47 PM, Richard Smith via cfe-dev <cfe-dev at lists.l
> > lvm.org> wrote:
> > > On 6 October 2017 at 07:36, Javier López Gómez via cfe-dev <cfe-dev at list
> > > s.llvm.org> wrote:
> > > > Hi,
> > > > 
> > > > I'm new to Clang hacking. Recently -prior to addition of a new
> > > > attribute-, I tried adding a C++11 spelling of the DiagnoseIf
> > > > attribute in the file
> > > > `tools/clang/include/clang/Basic/Attr.td':
> > > > 
> > > >   let Spellings = [GNU<"diagnose_if">,
> > > >                    CXX11<"", "diagnose_if", 201603>];
> > > > 
> > > > but the parser complains if the expression contains references to
> > > > function arguments, even if `LateParsed = 1' is specified, i.e. this
> > > > works
> > > > 
> > > >   int foo(int arg) __attribute__((diagnose_if(arg == 0, "text",
> > > > "warning")));
> > > > 
> > > > but this will not:
> > > > 
> > > >   [[diagnose_if(arg == 0, "text", "warning")]] int foo(int arg);
> > > > 
> > > > 
> > > > Am I missing something? Any help will be much appreciated.
> > > 
> > > There is nowhere that a C++11-syntax attribute can be written where it
> > > appertains to the function entity and the function parameter names are
> > > in scope. The two locations where attributes can be applied to a name
> > > introduced by a declarator are:
> > > 
> > >   [[here]] int f [[and_here]] (int x);
> > > 
> > > ... and neither of them follows the "x" parameter's declaration. After
> > > some discussion with Aaron, we think the best way forward is to recast
> > > such attributes as attributes on the function *type* (that perhaps don't
> > > change the canonical type, but do affect how the function declarator is
> > > interpreted as forming a function in some way). That'd mean you'd put
> > > the attributes:
> > > 
> > >   int f(int x) [[here]];
> > > 
> > > ... which conveniently is a place where the function parameters are in
> > > scope. The downside is that we don't have parsing support for attributes
> > > that appertain to function types yet. That would need to be added.
> > > 
> > 
> > I've been working on a fix for this; But if you want to hack on this, I'm
> > happy to yield.
> 
> Yes, I will be working on this for the next days. Thanks.
> 
> >  
> > > _______________________________________________
> > > cfe-dev mailing list
> > > cfe-dev at lists.llvm.org
> > > http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev
> > > 
> 
> 
-- 
Javier López
2.2.B.08 ARCOS lab, Computer Architecture Group
University Carlos III of Madrid
e-mail: jalopezg at inf.uc3m.es
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20171221/62d27616/attachment.html>