[Demangle] use std::string_view::data rather than &*std::string_view::begin

[nickdesaulniers]

Summary

To fix expensive check builds that were failing when using MSVC's
std::string_view::iterator::operator*, I added a few expressions like
&*std::string_view::begin. @nico pointed out that this is literally the
same thing and more clearly expressed as std::string_view::data.

Link: https://github.com/llvm/llvm-project/issues/63740

[ldionne]

LGTM -- this would fail with libc++ bounded iterators as well, since you might end up dereferencing a past-the-end iterator in case the string is empty.

[philnik]

Thanks for the cleanup! LGTM with green CI.

[nickdesaulniers]

• rebase to rekick ci

[thakis]

llvm/lib/Demangle/MicrosoftDemangle.cpp

2408

I think you no longer need the .empty() check once you use data(). If Name _is_ empty, then *Name.begin() is I think UB, but Name.data() isn't.

[nickdesaulniers]

• simplify NMangled calculation using size() rather than data()

libcxxabi/src/demangle/ItaniumDemangle.h

public:
 
   void printLeft(OutputBuffer &OB) const override {
     const size_t N = FloatData<Float>::mangled_size;
     if (Contents.size() >= N) {
       union {
         Float value;
         char buf[sizeof(Float)];
       };
-      const char *t = &*Contents.begin();
+      const char *t = Contents.data();
       const char *last = t + N;
       char *e = buf;
       for (; t != last; ++t, ++e) {
         unsigned d1 = isdigit(*t) ? static_cast<unsigned>(*t - '0')
                                   : static_cast<unsigned>(*t - 'a' + 10);
         ++t;
         unsigned d0 = isdigit(*t) ? static_cast<unsigned>(*t - '0')
                                   : static_cast<unsigned>(*t - 'a' + 10);
if (Qual.empty())
       return nullptr;
 
     // extension            ::= U <objc-name> <objc-type>  # objc-type<identifier>
     if (starts_with(Qual, "objcproto")) {
       constexpr size_t Len = sizeof("objcproto") - 1;
       std::string_view ProtoSourceName(Qual.data() + Len, Qual.size() - Len);
       std::string_view Proto;
       {
-        ScopedOverride<const char *> SaveFirst(First, &*ProtoSourceName.begin()),
+        ScopedOverride<const char *> SaveFirst(First, ProtoSourceName.data()),
             SaveLast(Last, &*ProtoSourceName.rbegin() + 1);
         Proto = parseBareSourceName();
       }
       if (Proto.empty())
         return nullptr;
       Node *Child = getDerived().parseQualifiedType();
       if (Child == nullptr)
         return nullptr;

llvm/include/llvm/Demangle/ItaniumDemangle.h

public:
 
   void printLeft(OutputBuffer &OB) const override {
     const size_t N = FloatData<Float>::mangled_size;
     if (Contents.size() >= N) {
       union {
         Float value;
         char buf[sizeof(Float)];
       };
-      const char *t = &*Contents.begin();
+      const char *t = Contents.data();
       const char *last = t + N;
       char *e = buf;
       for (; t != last; ++t, ++e) {
         unsigned d1 = isdigit(*t) ? static_cast<unsigned>(*t - '0')
                                   : static_cast<unsigned>(*t - 'a' + 10);
         ++t;
         unsigned d0 = isdigit(*t) ? static_cast<unsigned>(*t - '0')
                                   : static_cast<unsigned>(*t - 'a' + 10);
if (Qual.empty())
       return nullptr;
 
     // extension            ::= U <objc-name> <objc-type>  # objc-type<identifier>
     if (llvm::itanium_demangle::starts_with(Qual, "objcproto")) {
       constexpr size_t Len = sizeof("objcproto") - 1;
       std::string_view ProtoSourceName(Qual.data() + Len, Qual.size() - Len);
       std::string_view Proto;
       {
-        ScopedOverride<const char *> SaveFirst(First,
-                                               &*ProtoSourceName.begin()),
+        ScopedOverride<const char *> SaveFirst(First, ProtoSourceName.data()),
             SaveLast(Last, &*ProtoSourceName.rbegin() + 1);
         Proto = parseBareSourceName();
       }
       if (Proto.empty())
         return nullptr;
       Node *Child = getDerived().parseQualifiedType();
       if (Child == nullptr)
         return nullptr;

llvm/lib/Demangle/ItaniumDemangle.cpp

template<typename ...Ts> static bool anyWantNewline(Ts ...Vs) {
     for (bool B : {wantsNewline(Vs)...})
       if (B)
         return true;
     return false;
   }
 
   void printStr(const char *S) { fprintf(stderr, "%s", S); }
   void print(std::string_view SV) {
-    fprintf(stderr, "\"%.*s\"", (int)SV.size(), &*SV.begin());
+    fprintf(stderr, "\"%.*s\"", (int)SV.size(), SV.data());
   }
   void print(const Node *N) {
     if (N)
       N->visit(std::ref(*this));
     else
       printStr("<null>");
   }
   void print(NodeArray A) {

llvm/lib/Demangle/MicrosoftDemangle.cpp

demanglePointerCVQualifiers(std::string_view &MangledName) {
   DEMANGLE_UNREACHABLE;
 }
 
 std::string_view Demangler::copyString(std::string_view Borrowed) {
   char *Stable = Arena.allocUnalignedBuffer(Borrowed.size());
   // This is not a micro-optimization, it avoids UB, should Borrowed be an null
   // buffer.
   if (Borrowed.size())
-    std::memcpy(Stable, &*Borrowed.begin(), Borrowed.size());
+    std::memcpy(Stable, Borrowed.data(), Borrowed.size());
 
   return {Stable, Borrowed.size()};
 }
 
 SpecialTableSymbolNode *
 Demangler::demangleSpecialTableSymbolNode(std::string_view &MangledName,
                                           SpecialIntrinsicKind K) {
   NamedIdentifierNode *NI = Arena.alloc<NamedIdentifierNode>();
SymbolNode *Demangler::demangleMD5Name(std::string_view &MangledName) {
   // This is an MD5 mangled name.  We can't demangle it, just return the
   // mangled name.
   // An MD5 mangled name is ??@ followed by 32 characters and a terminating @.
   size_t MD5Last = MangledName.find('@', strlen("??@"));
   if (MD5Last == std::string_view::npos) {
     Error = true;
     return nullptr;
   }
-  const char *Start = &*MangledName.begin();
+  const char *Start = MangledName.data();
   const size_t StartSize = MangledName.size();
   MangledName.remove_prefix(MD5Last + 1);
 
   // There are two additional special cases for MD5 names:
   // 1. For complete object locators where the object name is long enough
   //    for the object to have an MD5 name, the complete object locator is
   //    called ??@...@??_R4@ (with a trailing "??_R4@" instead of the usual
   //    leading "??_R4". This is handled here.
void Demangler::dumpBackReferences() {
   OutputBuffer OB;
   for (size_t I = 0; I < Backrefs.FunctionParamCount; ++I) {
     OB.setCurrentPosition(0);
 
     TypeNode *T = Backrefs.FunctionParams[I];
     T->output(OB, OF_Default);
 
     std::string_view B = OB;
-    std::printf("  [%d] - %.*s\n", (int)I, (int)B.size(), &*B.begin());
+    std::printf("  [%d] - %.*s\n", (int)I, (int)B.size(), B.data());
   }
   std::free(OB.getBuffer());
 
   if (Backrefs.FunctionParamCount > 0)
     std::printf("\n");
   std::printf("%d name backreferences\n", (int)Backrefs.NamesCount);
   for (size_t I = 0; I < Backrefs.NamesCount; ++I) {
     std::printf("  [%d] - %.*s\n", (int)I, (int)Backrefs.Names[I]->Name.size(),
-                &*Backrefs.Names[I]->Name.begin());
+                Backrefs.Names[I]->Name.data());
   }
   if (Backrefs.NamesCount > 0)
     std::printf("\n");
 }
 
 char *llvm::microsoftDemangle(std::string_view MangledName, size_t *NMangled,
                               int *Status, MSDemangleFlags Flags) {
   Demangler D;
 
   std::string_view Name{MangledName};
   SymbolNode *AST = D.parse(Name);
   if (!D.Error && NMangled)
-    *NMangled = Name.empty() ? MangledName.size()
-                             : &*Name.begin() - &*MangledName.begin();
+    *NMangled = MangledName.size() - Name.size();
 
   if (Flags & MSDF_DumpBackrefs)
     D.dumpBackReferences();
 
   OutputFlags OF = OF_Default;
   if (Flags & MSDF_NoCallingConvention)
     OF = OutputFlags(OF | OF_NoCallingConvention);
   if (Flags & MSDF_NoAccessSpecifier)
     OF = OutputFlags(OF | OF_NoAccessSpecifier);