[clang] Implement objc_non_runtime_protocol to remove protocol metadata

Summary:
Motivated by the new objc_direct attribute, this change adds a new
attribute that remotes metadata from Protocols that the programmer knows
isn't going to be used at runtime. We simply have the frontend skip
generating any protocol metadata entries (e.g. OBJC_CLASS_NAME,
_OBJC_$_PROTOCOL_INSTANCE_METHDOS, _OBJC_PROTOCOL, etc) for a protocol
marked with `__attribute__((objc_non_runtime_protocol))`.

There are a few APIs used to retrieve a protocol at runtime.
`@protocol(SomeProtocol)` will now error out of the requested protocol
is marked with attribute. `objc_getProtocol` will return `NULL` which
is consistent with the behavior of a non-existing protocol.

Subscribers: cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D75574
diff --git a/clang/lib/CodeGen/CGObjC.cpp b/clang/lib/CodeGen/CGObjC.cpp
index 99b896a..f905e17e 100644
--- a/clang/lib/CodeGen/CGObjC.cpp
+++ b/clang/lib/CodeGen/CGObjC.cpp
@@ -445,6 +445,75 @@
                              Method);
 }
 
+static void AppendFirstImpliedRuntimeProtocols(
+    const ObjCProtocolDecl *PD,
+    llvm::UniqueVector<const ObjCProtocolDecl *> &PDs) {
+  if (!PD->isNonRuntimeProtocol()) {
+    const auto *Can = PD->getCanonicalDecl();
+    PDs.insert(Can);
+    return;
+  }
+
+  for (const auto *ParentPD : PD->protocols())
+    AppendFirstImpliedRuntimeProtocols(ParentPD, PDs);
+}
+
+std::vector<const ObjCProtocolDecl *>
+CGObjCRuntime::GetRuntimeProtocolList(ObjCProtocolDecl::protocol_iterator begin,
+                                      ObjCProtocolDecl::protocol_iterator end) {
+  std::vector<const ObjCProtocolDecl *> RuntimePds;
+  llvm::DenseSet<const ObjCProtocolDecl *> NonRuntimePDs;
+
+  for (; begin != end; ++begin) {
+    const auto *It = *begin;
+    const auto *Can = It->getCanonicalDecl();
+    if (Can->isNonRuntimeProtocol())
+      NonRuntimePDs.insert(Can);
+    else
+      RuntimePds.push_back(Can);
+  }
+
+  // If there are no non-runtime protocols then we can just stop now.
+  if (NonRuntimePDs.empty())
+    return RuntimePds;
+
+  // Else we have to search through the non-runtime protocol's inheritancy
+  // hierarchy DAG stopping whenever a branch either finds a runtime protocol or
+  // a non-runtime protocol without any parents. These are the "first-implied"
+  // protocols from a non-runtime protocol.
+  llvm::UniqueVector<const ObjCProtocolDecl *> FirstImpliedProtos;
+  for (const auto *PD : NonRuntimePDs)
+    AppendFirstImpliedRuntimeProtocols(PD, FirstImpliedProtos);
+
+  // Walk the Runtime list to get all protocols implied via the inclusion of
+  // this protocol, e.g. all protocols it inherits from including itself.
+  llvm::DenseSet<const ObjCProtocolDecl *> AllImpliedProtocols;
+  for (const auto *PD : RuntimePds) {
+    const auto *Can = PD->getCanonicalDecl();
+    AllImpliedProtocols.insert(Can);
+    Can->getImpliedProtocols(AllImpliedProtocols);
+  }
+
+  // Similar to above, walk the list of first-implied protocols to find the set
+  // all the protocols implied excluding the listed protocols themselves since
+  // they are not yet a part of the `RuntimePds` list.
+  for (const auto *PD : FirstImpliedProtos) {
+    PD->getImpliedProtocols(AllImpliedProtocols);
+  }
+
+  // From the first-implied list we have to finish building the final protocol
+  // list. If a protocol in the first-implied list was already implied via some
+  // inheritance path through some other protocols then it would be redundant to
+  // add it here and so we skip over it.
+  for (const auto *PD : FirstImpliedProtos) {
+    if (!AllImpliedProtocols.contains(PD)) {
+      RuntimePds.push_back(PD);
+    }
+  }
+
+  return RuntimePds;
+}
+
 /// Instead of '[[MyClass alloc] init]', try to generate
 /// 'objc_alloc_init(MyClass)'. This provides a code size improvement on the
 /// caller side, as well as the optimized objc_alloc.