[GCC] PR23529 Sema part of attrbute abi_tag support

[DmitryPolukhin]

Summary

Original patch by Stefan Bühler http://reviews.llvm.org/D12834

Difference between original and this one:

• fixed all comments in original code review
• added more tests, all new diagnostics now covered by tests
• moved abi_tag on re-declaration checks to Sema::mergeDeclAttributes where they actually may work as designed
• clang-format + other stylistic changes

Mangle part will be sent for review as a separate patch.

[DmitryPolukhin]

Aaron, friendly ping, please take a look!

[aaron.ballman]

I'll definitely take a look as I have the chance. In WG21 meetings all
week, so it may likely be sometime next week.

~Aaron

[sberg]

lib/Sema/SemaDeclAttr.cpp

4497 ↗

(On Diff #48901)

needs to be SmallVector<StringRef, 4> with current trunk

[DmitryPolukhin]

Rebase

[aaron.ballman]

docs/ItaniumMangleAbiTags.rst

2 ↗

(On Diff #49617)

ABI tags (since it's an acronym). Elsewhere as well.

30 ↗

(On Diff #49617)

an <unqualified-name>

49 ↗

(On Diff #49617)

A namespace does not have any active tags.

50 ↗

(On Diff #49617)

Comma after "enum)".

52 ↗

(On Diff #49617)

Comma after "functions".

68 ↗

(On Diff #49617)

Code examples for these might be useful for the reader.

79 ↗

(On Diff #49617)

Separate into two sentences instead of using a semicolon.

"Also, for functions, all tags from the..."

83 ↗

(On Diff #49617)

Replace the ! with a .

Also, an example might be useful for the reader.

89 ↗

(On Diff #49617)

as an active tag.

include/clang/Basic/AttrDocs.td

1965 ↗

(On Diff #49617)

to modify the mangled name

1966 ↗

(On Diff #49617)

the ability to distinguish between different versions

1967 ↗

(On Diff #49617)

versions supported.

a newer version of a class could have a different set of data members

1968 ↗

(On Diff #49617)

a different size.

Comma after attribute.

1969 ↗

(On Diff #49617)

names

1970 ↗

(On Diff #49617)

Therefore, (note the comma)

the old mangled name

will use the new

include/clang/Basic/DiagnosticSemaKinds.td

2465 ↗

(On Diff #49617)

Comma after functions.

4179 ↗

(On Diff #49617)

in a redeclaration

[DmitryPolukhin]

• fixed comments
• rebase with resolving conflicts

[DmitryPolukhin]

Thank you for the comments!

docs/ItaniumMangleAbiTags.rst

83 ↗

(On Diff #49617)

I'm 100% sure what exactly Stefan meant here and what is intended GCC behavior vs observed behavior, for example:

namespace A {

inline namespace B __attribute__((abi_tag)) {
   struct C { int x; };
}

}

std::string foo() {

struct E {
   __attribute__((abi_tag("X")))
  static A::C bar() {
    // mangle as _ZZ3fooB5cxx11vEN1E3barB1XEv
    // i.e. foo[abi:cxx11]()::E::bar[abi:X]()
    // abi_tag 'B' is missing for bar in GCC 5.3.0
    return A::C();
  }
};
E::bar();

}

Perhaps we should skip this part.

[aaron.ballman]

LGTM, thank you!

cfe/trunk/docs/ItaniumMangleAbiTags.rst

+========
+ABI tags
+========
+
+Introduction
+============
+
+This text tries to describe gcc semantic for mangling "abi_tag" attributes
+described in https://gcc.gnu.org/onlinedocs/gcc/C_002b_002b-Attributes.html
+
+There is no guarantee the following rules are correct, complete or make sense
+in any way as they were determined empirically by experiments with gcc5.
+
+Declaration
+===========
+
+ABI tags are declared in an abi_tag attribute and can be applied to a
+function, variable, class or inline namespace declaration. The attribute takes
+one or more strings (called tags); the order does not matter.
+
+See https://gcc.gnu.org/onlinedocs/gcc/C_002b_002b-Attributes.html for
+details.
+
+Tags on an inline namespace are called "implicit tags", all other tags are
+"explicit tags".
+
+Mangling
+========
+
+All tags that are "active" on an <unqualified-name> are emitted after the
+<unqualified-name>, before <template-args> or <discriminator>, and are part of
+the same <substitution> the <unqualified-name> is.
+
+They are mangled as:
+
+    <abi-tags> ::= <abi-tag>*   # sort by name
+    <abi-tag> ::= B <tag source-name>
+
+Example:
+
+    __attribute__((abi_tag("test")))
+    void Func();
+
+    gets mangled as: _Z4FuncB4testv (prettified as `Func[abi:test]()`)
+
+Active tags
+===========
+
+A namespace does not have any active tags. For types (class / struct / union /
+enum), the explicit tags are the active tags.
+
+For variables and functions, the active tags are the explicit tags plus any
+"required tags" which are not in the "available tags" set:
+
+    derived-tags := (required-tags - available-tags)
+    active-tags := explicit-tags + derived-tags
+
+Required tags for a function
+============================
+
+If a function is used as a local scope for another name, and is part of
+another function as local scope, it doesn't have any required tags.
+
+If a function is used as a local scope for a guard variable name, it doesn't
+have any required tags.
+
+Otherwise the function requires any implicit or explicit tag used in the name
+for the return type.
+
+Example:
+    namespace A {
+      inline namespace B __attribute__((abi_tag)) {
+        struct C { int x; };
+      }
+    }
+
+    A::C foo();
+
+    gets mangled as: _Z3fooB1Bv (prettified as `foo[abi:B]()`)
+
+Required tags for a variable
+============================
+
+A variable requires any implicit or explicit tag used in its type.
+
+Available tags
+==============
+
+All tags used in the prefix and in the template arguments for a name are
+available. Also, for functions, all tags from the <bare-function-type>
+(which might include the return type for template functions) are available.
+
+For <local-name>s all active tags used in the local part (<function-
+encoding>) are available, but not implicit tags which were not active.
+
+Implicit and explicit tags used in the <unqualified-name> for a function (as
+in the type of a cast operator) are NOT available.
+
+Example: a cast operator to std::string (which is
+std::__cxx11::basic_string<...>) will use 'cxx11' as an active tag, as it is
+required from the return type `std::string` but not available.

cfe/trunk/include/clang/Basic/Attr.td

class IgnoredAttr : Attr {
   let SemaHandler = 0;
   let Documentation = [Undocumented];
 }
 
 //
 // Attributes begin here
 //
 
+def AbiTag : Attr {
+  let Spellings = [GCC<"abi_tag">];
+  let Args = [VariadicStringArgument<"Tags">];
+  let Subjects = SubjectList<[Struct, Var, Function, Namespace], ErrorDiag,
+      "ExpectedStructClassVariableFunctionOrInlineNamespace">;
+  let Documentation = [AbiTagsDocs];
+}
+
 def AddressSpace : TypeAttr {
   let Spellings = [GNU<"address_space">];
   let Args = [IntArgument<"AddressSpace">];
   let Documentation = [Undocumented];
 }
 
 def Alias : Attr {
   let Spellings = [GCC<"alias">];

cfe/trunk/include/clang/Basic/AttrDocs.td

 value does not point to storage of adequate size and alignment for a
 value of type ``T``.
 
 Making indirect results explicit in the signature allows C functions to
 directly construct objects into them without relying on language
 optimizations like C++'s named return value optimization (NRVO).
   }];
 }
+
+def AbiTagsDocs : Documentation {
+  let Content = [{
+The ``abi_tag`` attribute can be applied to a function, variable, class or
+inline namespace declaration to modify the mangled name of the entity. It gives
+the ability to distinguish between different versions of the same entity but
+with different ABI versions supported. For example, a newer version of a class
+could have a different set of data members and thus have a different size. Using
+the ``abi_tag`` attribute, it is possible to have different mangled names for
+a global variable of the class type. Therefor, the old code could keep using
+the old manged name and the new code will use the new mangled name with tags.
+  }];
+}

cfe/trunk/include/clang/Basic/DiagnosticSemaKinds.td

def warn_attribute_wrong_decl_type : Warning<
   "variables and typedefs|thread-local variables|"
   "variables and fields|variables, data members and tag types|"
   "types and namespaces|Objective-C interfaces|methods and properties|"
   "struct or union|struct, union or class|types|"
   "Objective-C instance methods|init methods of interface or class extension declarations|"
   "variables, functions and classes|Objective-C protocols|"
   "functions and global variables|structs, unions, and typedefs|structs and typedefs|"
   "interface or protocol declarations|kernel functions|non-K&R-style functions|"
-  "variables, enums, fields and typedefs|functions, methods, enums, and classes}1">,
+  "variables, enums, fields and typedefs|functions, methods, enums, and classes|"
+  "structs, classes, variables, functions, and inline namespaces}1">,
   InGroup<IgnoredAttributes>;
 def err_attribute_wrong_decl_type : Error<warn_attribute_wrong_decl_type.Text>;
 def warn_type_attribute_wrong_type : Warning<
   "'%0' only applies to %select{function|pointer|"
   "Objective-C object or block pointer}1 types; type here is %2">,
   InGroup<IgnoredAttributes>;
 def warn_incomplete_encoded_type : Warning<
   "encoding of %0 type is incomplete because %1 component has unknown encoding">,
def err_definition_of_implicitly_declared_member : Error<
   "operator|destructor|function}1">;
 def err_definition_of_explicitly_defaulted_member : Error<
   "definition of explicitly defaulted %select{default constructor|copy "
   "constructor|move constructor|copy assignment operator|move assignment "
   "operator|destructor}0">;
 def err_redefinition_extern_inline : Error<
   "redefinition of a 'extern inline' function %0 is not supported in "
   "%select{C99 mode|C++}1">;
+def warn_attr_abi_tag_namespace : Warning<
+  "'abi_tag' attribute on %select{non-inline|anonymous}0 namespace ignored">,
+  InGroup<IgnoredAttributes>;
+def err_abi_tag_on_redeclaration : Error<
+  "cannot add 'abi_tag' attribute in a redeclaration">;
+def err_new_abi_tag_on_redeclaration : Error<
+  "'abi_tag' %0 missing in original declaration">;
 
 def note_deleted_dtor_no_operator_delete : Note<
   "virtual destructor requires an unambiguous, accessible 'operator delete'">;
 def note_deleted_special_member_class_subobject : Note<
   "%select{default constructor|copy constructor|move constructor|"
   "copy assignment operator|move assignment operator|destructor}0 of "
   "%1 is implicitly deleted because "
   "%select{base class %3|%select{||||variant }4field %3}2 has "

cfe/trunk/include/clang/Sema/AttributeList.h

enum AttributeDeclKind {
   ExpectedObjectiveCProtocol,
   ExpectedFunctionGlobalVarMethodOrProperty,
   ExpectedStructOrUnionOrTypedef,
   ExpectedStructOrTypedef,
   ExpectedObjectiveCInterfaceOrProtocol,
   ExpectedKernelFunction,
   ExpectedFunctionWithProtoType,
   ExpectedVariableEnumFieldOrTypedef,
-  ExpectedFunctionMethodEnumOrClass
+  ExpectedFunctionMethodEnumOrClass,
+  ExpectedStructClassVariableFunctionOrInlineNamespace
 };
 
 }  // end namespace clang
 
 #endif

cfe/trunk/lib/Sema/SemaDecl.cpp

if (AsmLabelAttr *NewA = New->getAttr<AsmLabelAttr>()) {
     } else if (Old->isUsed()) {
       // This redeclaration adds an __asm__ label to a declaration that has
       // already been ODR-used.
       Diag(New->getLocation(), diag::err_late_asm_label_name)
         << isa<FunctionDecl>(Old) << New->getAttr<AsmLabelAttr>()->getRange();
     }
   }
 
+  // Re-declaration cannot add abi_tag's.
+  if (const auto *NewAbiTagAttr = New->getAttr<AbiTagAttr>()) {
+    if (const auto *OldAbiTagAttr = Old->getAttr<AbiTagAttr>()) {
+      for (const auto &NewTag : NewAbiTagAttr->tags()) {
+        if (std::find(OldAbiTagAttr->tags_begin(), OldAbiTagAttr->tags_end(),
+                      NewTag) == OldAbiTagAttr->tags_end()) {
+          Diag(NewAbiTagAttr->getLocation(),
+               diag::err_new_abi_tag_on_redeclaration)
+              << NewTag;
+          Diag(OldAbiTagAttr->getLocation(), diag::note_previous_declaration);
+        }
+      }
+    } else {
+      Diag(NewAbiTagAttr->getLocation(), diag::err_abi_tag_on_redeclaration);
+      Diag(Old->getLocation(), diag::note_previous_declaration);
+    }
+  }
+
   if (!Old->hasAttrs())
     return;
 
   bool foundAny = New->hasAttrs();
 
   // Ensure that any moving of objects within the allocated map is done before
   // we process them.
   if (!foundAny) New->setAttrs(AttrVec());

cfe/trunk/lib/Sema/SemaDeclAttr.cpp

static void handleDeclspecThreadAttr(Sema &S, Decl *D,
   if (VD->hasLocalStorage()) {
     S.Diag(Attr.getLoc(), diag::err_thread_non_global) << "__declspec(thread)";
     return;
   }
   VD->addAttr(::new (S.Context) ThreadAttr(
       Attr.getRange(), S.Context, Attr.getAttributeSpellingListIndex()));
 }
 
+static void handleAbiTagAttr(Sema &S, Decl *D, const AttributeList &Attr) {
+  SmallVector<StringRef, 4> Tags;
+  for (unsigned I = 0, E = Attr.getNumArgs(); I != E; ++I) {
+    StringRef Tag;
+    if (!S.checkStringLiteralArgumentAttr(Attr, I, Tag))
+      return;
+    Tags.push_back(Tag);
+  }
+
+  if (const auto *NS = dyn_cast<NamespaceDecl>(D)) {
+    if (!NS->isInline()) {
+      S.Diag(Attr.getLoc(), diag::warn_attr_abi_tag_namespace) << 0;
+      return;
+    }
+    if (NS->isAnonymousNamespace()) {
+      S.Diag(Attr.getLoc(), diag::warn_attr_abi_tag_namespace) << 1;
+      return;
+    }
+    if (Attr.getNumArgs() == 0)
+      Tags.push_back(NS->getName());
+  } else if (!checkAttributeAtLeastNumArgs(S, Attr, 1))
+    return;
+
+  // Store tags sorted and without duplicates.
+  std::sort(Tags.begin(), Tags.end());
+  Tags.erase(std::unique(Tags.begin(), Tags.end()), Tags.end());
+
+  D->addAttr(::new (S.Context)
+             AbiTagAttr(Attr.getRange(), S.Context, Tags.data(), Tags.size(),
+                        Attr.getAttributeSpellingListIndex()));
+
+  // FIXME: remove this warning as soon as mangled part is ready.
+  S.Diag(Attr.getRange().getBegin(), diag::warn_attribute_ignored)
+        << Attr.getName();
+}
+
 static void handleARMInterruptAttr(Sema &S, Decl *D,
                                    const AttributeList &Attr) {
   // Check the attribute arguments.
   if (Attr.getNumArgs() > 1) {
     S.Diag(Attr.getLoc(), diag::err_attribute_too_many_arguments)
       << Attr.getName() << 1;
     return;
   }
case AttributeList::AT_MSInheritance:
     handleMSInheritanceAttr(S, D, Attr);
     break;
   case AttributeList::AT_SelectAny:
     handleSimpleAttribute<SelectAnyAttr>(S, D, Attr);
     break;
   case AttributeList::AT_Thread:
     handleDeclspecThreadAttr(S, D, Attr);
     break;
+  case AttributeList::AT_AbiTag:
+    handleAbiTagAttr(S, D, Attr);
+    break;
 
   // Thread safety attributes:
   case AttributeList::AT_AssertExclusiveLock:
     handleAssertExclusiveLockAttr(S, D, Attr);
     break;
   case AttributeList::AT_AssertSharedLock:
     handleAssertSharedLockAttr(S, D, Attr);
     break;

cfe/trunk/test/SemaCXX/attr-abi-tag-syntax.cpp

+// RUN: %clang_cc1 -std=c++11 -fsyntax-only -verify %s
+
+namespace N1 {
+
+namespace __attribute__((__abi_tag__)) {}
+// expected-warning@-1 {{'abi_tag' attribute on non-inline namespace ignored}}
+
+namespace N __attribute__((__abi_tag__)) {}
+// expected-warning@-1 {{'abi_tag' attribute on non-inline namespace ignored}}
+
+} // namespace N1
+
+namespace N2 {
+
+inline namespace __attribute__((__abi_tag__)) {}
+// expected-warning@-1 {{'abi_tag' attribute on anonymous namespace ignored}}
+
+inline namespace N __attribute__((__abi_tag__)) {}
+// FIXME: remove this warning as soon as attribute fully supported.
+// expected-warning@-2 {{'__abi_tag__' attribute ignored}}
+
+} // namespcace N2
+
+__attribute__((abi_tag("B", "A"))) extern int a1;
+// FIXME: remove this warning as soon as attribute fully supported.
+// expected-warning@-2 {{'abi_tag' attribute ignored}}
+
+__attribute__((abi_tag("A", "B"))) extern int a1;
+// expected-note@-1 {{previous declaration is here}}
+// FIXME: remove this warning as soon as attribute fully supported.
+// expected-warning@-3 {{'abi_tag' attribute ignored}}
+
+__attribute__((abi_tag("A", "C"))) extern int a1;
+// expected-error@-1 {{'abi_tag' C missing in original declaration}}
+// FIXME: remove this warning as soon as attribute fully supported.
+// expected-warning@-3 {{'abi_tag' attribute ignored}}
+
+extern int a2;
+// expected-note@-1 {{previous declaration is here}}
+__attribute__((abi_tag("A")))extern int a2;
+// expected-error@-1 {{cannot add 'abi_tag' attribute in a redeclaration}}
+// FIXME: remove this warning as soon as attribute fully supported.
+// expected-warning@-3 {{'abi_tag' attribute ignored}}