llvmAsmParser.y

//===-- llvmAsmParser.y - Parser for llvm assembly files --------*- C++ -*-===//
// 
//                     The LLVM Compiler Infrastructure
//
// This file was developed by the LLVM research group and is distributed under
// the University of Illinois Open Source License. See LICENSE.TXT for details.
// 
//===----------------------------------------------------------------------===//
//
//  This file implements the bison parser for LLVM assembly languages files.
//
//===----------------------------------------------------------------------===//

%{
#include "ParserInternals.h"
#include "llvm/Instructions.h"
#include "llvm/Module.h"
#include "llvm/SymbolTable.h"
#include "llvm/Support/GetElementPtrTypeIterator.h"
#include "llvm/ADT/STLExtras.h"
#include <algorithm>
#include <iostream>
#include <list>
#include <utility>

int yyerror(const char *ErrorMsg); // Forward declarations to prevent "implicit
int yylex();                       // declaration" of xxx warnings.
int yyparse();

namespace llvm {
  std::string CurFilename;
}
using namespace llvm;

static Module *ParserResult;

// DEBUG_UPREFS - Define this symbol if you want to enable debugging output
// relating to upreferences in the input stream.
//
//#define DEBUG_UPREFS 1
#ifdef DEBUG_UPREFS
#define UR_OUT(X) std::cerr << X
#else
#define UR_OUT(X)
#endif

#define YYERROR_VERBOSE 1

// HACK ALERT: This variable is used to implement the automatic conversion of
// variable argument instructions from their old to new forms.  When this
// compatiblity "Feature" is removed, this should be too.
//
static BasicBlock *CurBB;
static bool ObsoleteVarArgs;


// This contains info used when building the body of a function.  It is
// destroyed when the function is completed.
//
typedef std::vector<Value *> ValueList;           // Numbered defs
static void ResolveDefinitions(std::map<const Type *,ValueList> &LateResolvers,
                               std::map<const Type *,ValueList> *FutureLateResolvers = 0);

static struct PerModuleInfo {
  Module *CurrentModule;
  std::map<const Type *, ValueList> Values; // Module level numbered definitions
  std::map<const Type *,ValueList> LateResolveValues;
  std::vector<PATypeHolder>    Types;
  std::map<ValID, PATypeHolder> LateResolveTypes;

  /// PlaceHolderInfo - When temporary placeholder objects are created, remember
  /// how they were referenced and one which line of the input they came from so
  /// that we can resolve them later and print error messages as appropriate.
  std::map<Value*, std::pair<ValID, int> > PlaceHolderInfo;

  // GlobalRefs - This maintains a mapping between <Type, ValID>'s and forward
  // references to global values.  Global values may be referenced before they
  // are defined, and if so, the temporary object that they represent is held
  // here.  This is used for forward references of GlobalValues.
  //
  typedef std::map<std::pair<const PointerType *,
                             ValID>, GlobalValue*> GlobalRefsType;
  GlobalRefsType GlobalRefs;

  void ModuleDone() {
    // If we could not resolve some functions at function compilation time
    // (calls to functions before they are defined), resolve them now...  Types
    // are resolved when the constant pool has been completely parsed.
    //
    ResolveDefinitions(LateResolveValues);

    // Check to make sure that all global value forward references have been
    // resolved!
    //
    if (!GlobalRefs.empty()) {
      std::string UndefinedReferences = "Unresolved global references exist:\n";
      
      for (GlobalRefsType::iterator I = GlobalRefs.begin(), E =GlobalRefs.end();
           I != E; ++I) {
        UndefinedReferences += "  " + I->first.first->getDescription() + " " +
                               I->first.second.getName() + "\n";
      }
      ThrowException(UndefinedReferences);
    }

    Values.clear();         // Clear out function local definitions
    Types.clear();
    CurrentModule = 0;
  }


  // GetForwardRefForGlobal - Check to see if there is a forward reference
  // for this global.  If so, remove it from the GlobalRefs map and return it.
  // If not, just return null.
  GlobalValue *GetForwardRefForGlobal(const PointerType *PTy, ValID ID) {
    // Check to see if there is a forward reference to this global variable...
    // if there is, eliminate it and patch the reference to use the new def'n.
    GlobalRefsType::iterator I = GlobalRefs.find(std::make_pair(PTy, ID));
    GlobalValue *Ret = 0;
    if (I != GlobalRefs.end()) {
      Ret = I->second;
      GlobalRefs.erase(I);
    }
    return Ret;
  }
} CurModule;

static struct PerFunctionInfo {
  Function *CurrentFunction;     // Pointer to current function being created

  std::map<const Type*, ValueList> Values;   // Keep track of #'d definitions
  std::map<const Type*, ValueList> LateResolveValues;
  std::vector<PATypeHolder> Types;
  std::map<ValID, PATypeHolder> LateResolveTypes;
  bool isDeclare;                // Is this function a forward declararation?

  /// BBForwardRefs - When we see forward references to basic blocks, keep
  /// track of them here.
  std::map<BasicBlock*, std::pair<ValID, int> > BBForwardRefs;
  std::vector<BasicBlock*> NumberedBlocks;
  unsigned NextBBNum;

  inline PerFunctionInfo() {
    CurrentFunction = 0;
    isDeclare = false;
  }

  inline void FunctionStart(Function *M) {
    CurrentFunction = M;
    NextBBNum = 0;
  }

  void FunctionDone() {
    NumberedBlocks.clear();

    // Any forward referenced blocks left?
    if (!BBForwardRefs.empty())
      ThrowException("Undefined reference to label " +
                     BBForwardRefs.begin()->second.first.getName());

    // Resolve all forward references now.
    ResolveDefinitions(LateResolveValues, &CurModule.LateResolveValues);

    Values.clear();         // Clear out function local definitions
    Types.clear();          // Clear out function local types
    CurrentFunction = 0;
    isDeclare = false;
  }
} CurFun;  // Info for the current function...

static bool inFunctionScope() { return CurFun.CurrentFunction != 0; }


//===----------------------------------------------------------------------===//
//               Code to handle definitions of all the types
//===----------------------------------------------------------------------===//

static int InsertValue(Value *V,
                  std::map<const Type*,ValueList> &ValueTab = CurFun.Values) {
  if (V->hasName()) return -1;           // Is this a numbered definition?

  // Yes, insert the value into the value table...
  ValueList &List = ValueTab[V->getType()];
  List.push_back(V);
  return List.size()-1;
}

static const Type *getTypeVal(const ValID &D, bool DoNotImprovise = false) {
  switch (D.Type) {
  case ValID::NumberVal:               // Is it a numbered definition?
    // Module constants occupy the lowest numbered slots...
    if ((unsigned)D.Num < CurModule.Types.size()) 
      return CurModule.Types[(unsigned)D.Num];
    break;
  case ValID::NameVal:                 // Is it a named definition?
    if (const Type *N = CurModule.CurrentModule->getTypeByName(D.Name)) {
      D.destroy();  // Free old strdup'd memory...
      return N;
    }
    break;
  default:
    ThrowException("Internal parser error: Invalid symbol type reference!");
  }

  // If we reached here, we referenced either a symbol that we don't know about
  // or an id number that hasn't been read yet.  We may be referencing something
  // forward, so just create an entry to be resolved later and get to it...
  //
  if (DoNotImprovise) return 0;  // Do we just want a null to be returned?

  std::map<ValID, PATypeHolder> &LateResolver = inFunctionScope() ? 
    CurFun.LateResolveTypes : CurModule.LateResolveTypes;
  
  std::map<ValID, PATypeHolder>::iterator I = LateResolver.find(D);
  if (I != LateResolver.end()) {
    return I->second;
  }

  Type *Typ = OpaqueType::get();
  LateResolver.insert(std::make_pair(D, Typ));
  return Typ;
}

static Value *lookupInSymbolTable(const Type *Ty, const std::string &Name) {
  SymbolTable &SymTab = 
    inFunctionScope() ? CurFun.CurrentFunction->getSymbolTable() :
                        CurModule.CurrentModule->getSymbolTable();
  return SymTab.lookup(Ty, Name);
}

// getValNonImprovising - Look up the value specified by the provided type and
// the provided ValID.  If the value exists and has already been defined, return
// it.  Otherwise return null.
//
static Value *getValNonImprovising(const Type *Ty, const ValID &D) {
  if (isa<FunctionType>(Ty))
    ThrowException("Functions are not values and "
                   "must be referenced as pointers");

  switch (D.Type) {
  case ValID::NumberVal: {                 // Is it a numbered definition?
    unsigned Num = (unsigned)D.Num;

    // Module constants occupy the lowest numbered slots...
    std::map<const Type*,ValueList>::iterator VI = CurModule.Values.find(Ty);
    if (VI != CurModule.Values.end()) {
      if (Num < VI->second.size()) 
        return VI->second[Num];
      Num -= VI->second.size();
    }

    // Make sure that our type is within bounds
    VI = CurFun.Values.find(Ty);
    if (VI == CurFun.Values.end()) return 0;

    // Check that the number is within bounds...
    if (VI->second.size() <= Num) return 0;
  
    return VI->second[Num];
  }

  case ValID::NameVal: {                // Is it a named definition?
    Value *N = lookupInSymbolTable(Ty, std::string(D.Name));
    if (N == 0) return 0;

    D.destroy();  // Free old strdup'd memory...
    return N;
  }

  // Check to make sure that "Ty" is an integral type, and that our 
  // value will fit into the specified type...
  case ValID::ConstSIntVal:    // Is it a constant pool reference??
    if (!ConstantSInt::isValueValidForType(Ty, D.ConstPool64))
      ThrowException("Signed integral constant '" +
                     itostr(D.ConstPool64) + "' is invalid for type '" + 
                     Ty->getDescription() + "'!");
    return ConstantSInt::get(Ty, D.ConstPool64);

  case ValID::ConstUIntVal:     // Is it an unsigned const pool reference?
    if (!ConstantUInt::isValueValidForType(Ty, D.UConstPool64)) {
      if (!ConstantSInt::isValueValidForType(Ty, D.ConstPool64)) {
        ThrowException("Integral constant '" + utostr(D.UConstPool64) +
                       "' is invalid or out of range!");
      } else {     // This is really a signed reference.  Transmogrify.
        return ConstantSInt::get(Ty, D.ConstPool64);
      }
    } else {
      return ConstantUInt::get(Ty, D.UConstPool64);
    }

  case ValID::ConstFPVal:        // Is it a floating point const pool reference?
    if (!ConstantFP::isValueValidForType(Ty, D.ConstPoolFP))
      ThrowException("FP constant invalid for type!!");
    return ConstantFP::get(Ty, D.ConstPoolFP);
    
  case ValID::ConstNullVal:      // Is it a null value?
    if (!isa<PointerType>(Ty))
      ThrowException("Cannot create a a non pointer null!");
    return ConstantPointerNull::get(cast<PointerType>(Ty));
    
  case ValID::ConstUndefVal:      // Is it an undef value?
    return UndefValue::get(Ty);
    
  case ValID::ConstantVal:       // Fully resolved constant?
    if (D.ConstantValue->getType() != Ty)
      ThrowException("Constant expression type different from required type!");
    return D.ConstantValue;

  default:
    assert(0 && "Unhandled case!");
    return 0;
  }   // End of switch

  assert(0 && "Unhandled case!");
  return 0;
}

// getVal - This function is identical to getValNonImprovising, except that if a
// value is not already defined, it "improvises" by creating a placeholder var
// that looks and acts just like the requested variable.  When the value is
// defined later, all uses of the placeholder variable are replaced with the
// real thing.
//
static Value *getVal(const Type *Ty, const ValID &ID) {
  if (Ty == Type::LabelTy)
    ThrowException("Cannot use a basic block here");

  // See if the value has already been defined.
  Value *V = getValNonImprovising(Ty, ID);
  if (V) return V;

  // If we reached here, we referenced either a symbol that we don't know about
  // or an id number that hasn't been read yet.  We may be referencing something
  // forward, so just create an entry to be resolved later and get to it...
  //
  V = new Argument(Ty);

  // Remember where this forward reference came from.  FIXME, shouldn't we try
  // to recycle these things??
  CurModule.PlaceHolderInfo.insert(std::make_pair(V, std::make_pair(ID,
                                                               llvmAsmlineno)));

  if (inFunctionScope())
    InsertValue(V, CurFun.LateResolveValues);
  else 
    InsertValue(V, CurModule.LateResolveValues);
  return V;
}

/// getBBVal - This is used for two purposes:
///  * If isDefinition is true, a new basic block with the specified ID is being
///    defined.
///  * If isDefinition is true, this is a reference to a basic block, which may
///    or may not be a forward reference.
///
static BasicBlock *getBBVal(const ValID &ID, bool isDefinition = false) {
  assert(inFunctionScope() && "Can't get basic block at global scope!");

  std::string Name;
  BasicBlock *BB = 0;
  switch (ID.Type) {
  default: ThrowException("Illegal label reference " + ID.getName());
  case ValID::NumberVal:                // Is it a numbered definition?
    if (unsigned(ID.Num) >= CurFun.NumberedBlocks.size())
      CurFun.NumberedBlocks.resize(ID.Num+1);
    BB = CurFun.NumberedBlocks[ID.Num];
    break;
  case ValID::NameVal:                  // Is it a named definition?
    Name = ID.Name;
    if (Value *N = CurFun.CurrentFunction->
                   getSymbolTable().lookup(Type::LabelTy, Name))
      BB = cast<BasicBlock>(N);
    break;
  }

  // See if the block has already been defined.
  if (BB) {
    // If this is the definition of the block, make sure the existing value was
    // just a forward reference.  If it was a forward reference, there will be
    // an entry for it in the PlaceHolderInfo map.
    if (isDefinition && !CurFun.BBForwardRefs.erase(BB))
      // The existing value was a definition, not a forward reference.
      ThrowException("Redefinition of label " + ID.getName());

    ID.destroy();                       // Free strdup'd memory.
    return BB;
  }

  // Otherwise this block has not been seen before.
  BB = new BasicBlock("", CurFun.CurrentFunction);
  if (ID.Type == ValID::NameVal) {
    BB->setName(ID.Name);
  } else {
    CurFun.NumberedBlocks[ID.Num] = BB;
  }

  // If this is not a definition, keep track of it so we can use it as a forward
  // reference.
  if (!isDefinition) {
    // Remember where this forward reference came from.
    CurFun.BBForwardRefs[BB] = std::make_pair(ID, llvmAsmlineno);
  } else {
    // The forward declaration could have been inserted anywhere in the
    // function: insert it into the correct place now.
    CurFun.CurrentFunction->getBasicBlockList().remove(BB);
    CurFun.CurrentFunction->getBasicBlockList().push_back(BB);
  }
  ID.destroy();
  return BB;
}


//===----------------------------------------------------------------------===//
//              Code to handle forward references in instructions
//===----------------------------------------------------------------------===//
//
// This code handles the late binding needed with statements that reference
// values not defined yet... for example, a forward branch, or the PHI node for
// a loop body.
//
// This keeps a table (CurFun.LateResolveValues) of all such forward references
// and back patchs after we are done.
//

// ResolveDefinitions - If we could not resolve some defs at parsing 
// time (forward branches, phi functions for loops, etc...) resolve the 
// defs now...
//
static void ResolveDefinitions(std::map<const Type*,ValueList> &LateResolvers,
                               std::map<const Type*,ValueList> *FutureLateResolvers) {
  // Loop over LateResolveDefs fixing up stuff that couldn't be resolved
  for (std::map<const Type*,ValueList>::iterator LRI = LateResolvers.begin(),
         E = LateResolvers.end(); LRI != E; ++LRI) {
    ValueList &List = LRI->second;
    while (!List.empty()) {
      Value *V = List.back();
      List.pop_back();

      std::map<Value*, std::pair<ValID, int> >::iterator PHI =
        CurModule.PlaceHolderInfo.find(V);
      assert(PHI != CurModule.PlaceHolderInfo.end() && "Placeholder error!");

      ValID &DID = PHI->second.first;

      Value *TheRealValue = getValNonImprovising(LRI->first, DID);
      if (TheRealValue) {
        V->replaceAllUsesWith(TheRealValue);
        delete V;
        CurModule.PlaceHolderInfo.erase(PHI);
      } else if (FutureLateResolvers) {
        // Functions have their unresolved items forwarded to the module late
        // resolver table
        InsertValue(V, *FutureLateResolvers);
      } else {
        if (DID.Type == ValID::NameVal)
          ThrowException("Reference to an invalid definition: '" +DID.getName()+
                         "' of type '" + V->getType()->getDescription() + "'",
                         PHI->second.second);
        else
          ThrowException("Reference to an invalid definition: #" +
                         itostr(DID.Num) + " of type '" + 
                         V->getType()->getDescription() + "'",
                         PHI->second.second);
      }
    }
  }

  LateResolvers.clear();
}

// ResolveTypeTo - A brand new type was just declared.  This means that (if
// name is not null) things referencing Name can be resolved.  Otherwise, things
// refering to the number can be resolved.  Do this now.
//
static void ResolveTypeTo(char *Name, const Type *ToTy) {
  std::vector<PATypeHolder> &Types = inFunctionScope() ? 
     CurFun.Types : CurModule.Types;

   ValID D;
   if (Name) D = ValID::create(Name);
   else      D = ValID::create((int)Types.size());

   std::map<ValID, PATypeHolder> &LateResolver = inFunctionScope() ? 
     CurFun.LateResolveTypes : CurModule.LateResolveTypes;
  
   std::map<ValID, PATypeHolder>::iterator I = LateResolver.find(D);
   if (I != LateResolver.end()) {
     ((DerivedType*)I->second.get())->refineAbstractTypeTo(ToTy);
     LateResolver.erase(I);
   }
}

// ResolveTypes - At this point, all types should be resolved.  Any that aren't
// are errors.
//
static void ResolveTypes(std::map<ValID, PATypeHolder> &LateResolveTypes) {
  if (!LateResolveTypes.empty()) {
    const ValID &DID = LateResolveTypes.begin()->first;

    if (DID.Type == ValID::NameVal)
      ThrowException("Reference to an invalid type: '" +DID.getName() + "'");
    else
      ThrowException("Reference to an invalid type: #" + itostr(DID.Num));
  }
}

// setValueName - Set the specified value to the name given.  The name may be
// null potentially, in which case this is a noop.  The string passed in is
// assumed to be a malloc'd string buffer, and is free'd by this function.
//
static void setValueName(Value *V, char *NameStr) {
  if (NameStr) {
    std::string Name(NameStr);      // Copy string
    free(NameStr);                  // Free old string

    if (V->getType() == Type::VoidTy) 
      ThrowException("Can't assign name '" + Name+"' to value with void type!");
    
    assert(inFunctionScope() && "Must be in function scope!");
    SymbolTable &ST = CurFun.CurrentFunction->getSymbolTable();
    if (ST.lookup(V->getType(), Name))
      ThrowException("Redefinition of value named '" + Name + "' in the '" +
                     V->getType()->getDescription() + "' type plane!");
    
    // Set the name.
    V->setName(Name, &ST);
  }
}

/// ParseGlobalVariable - Handle parsing of a global.  If Initializer is null,
/// this is a declaration, otherwise it is a definition.
static void ParseGlobalVariable(char *NameStr,GlobalValue::LinkageTypes Linkage,
                                bool isConstantGlobal, const Type *Ty,
                                Constant *Initializer) {
  if (isa<FunctionType>(Ty))
    ThrowException("Cannot declare global vars of function type!");

  const PointerType *PTy = PointerType::get(Ty); 

  std::string Name;
  if (NameStr) {
    Name = NameStr;      // Copy string
    free(NameStr);       // Free old string
  }

  // See if this global value was forward referenced.  If so, recycle the
  // object.
  ValID ID; 
  if (!Name.empty()) {
    ID = ValID::create((char*)Name.c_str());
  } else {
    ID = ValID::create((int)CurModule.Values[PTy].size());
  }

  if (GlobalValue *FWGV = CurModule.GetForwardRefForGlobal(PTy, ID)) {
    // Move the global to the end of the list, from whereever it was 
    // previously inserted.
    GlobalVariable *GV = cast<GlobalVariable>(FWGV);
    CurModule.CurrentModule->getGlobalList().remove(GV);
    CurModule.CurrentModule->getGlobalList().push_back(GV);
    GV->setInitializer(Initializer);
    GV->setLinkage(Linkage);
    GV->setConstant(isConstantGlobal);
    InsertValue(GV, CurModule.Values);
    return;
  }

  // If this global has a name, check to see if there is already a definition
  // of this global in the module.  If so, merge as appropriate.  Note that
  // this is really just a hack around problems in the CFE.  :(
  if (!Name.empty()) {
    // We are a simple redefinition of a value, check to see if it is defined
    // the same as the old one.
    if (GlobalVariable *EGV = 
                CurModule.CurrentModule->getGlobalVariable(Name, Ty)) {
      // We are allowed to redefine a global variable in two circumstances:
      // 1. If at least one of the globals is uninitialized or 
      // 2. If both initializers have the same value.
      //
      if (!EGV->hasInitializer() || !Initializer ||
          EGV->getInitializer() == Initializer) {

        // Make sure the existing global version gets the initializer!  Make
        // sure that it also gets marked const if the new version is.
        if (Initializer && !EGV->hasInitializer())
          EGV->setInitializer(Initializer);
        if (isConstantGlobal)
          EGV->setConstant(true);
        EGV->setLinkage(Linkage);
        return;
      }

      ThrowException("Redefinition of global variable named '" + Name + 
                     "' in the '" + Ty->getDescription() + "' type plane!");
    }
  }

  // Otherwise there is no existing GV to use, create one now.
  GlobalVariable *GV =
    new GlobalVariable(Ty, isConstantGlobal, Linkage, Initializer, Name, 
                       CurModule.CurrentModule);
  InsertValue(GV, CurModule.Values);
}

// setTypeName - Set the specified type to the name given.  The name may be
// null potentially, in which case this is a noop.  The string passed in is
// assumed to be a malloc'd string buffer, and is freed by this function.
//
// This function returns true if the type has already been defined, but is
// allowed to be redefined in the specified context.  If the name is a new name
// for the type plane, it is inserted and false is returned.
static bool setTypeName(const Type *T, char *NameStr) {
  assert(!inFunctionScope() && "Can't give types function-local names!");
  if (NameStr == 0) return false;
  
  std::string Name(NameStr);      // Copy string
  free(NameStr);                  // Free old string

  // We don't allow assigning names to void type
  if (T == Type::VoidTy) 
    ThrowException("Can't assign name '" + Name + "' to the void type!");

  // Set the type name, checking for conflicts as we do so.
  bool AlreadyExists = CurModule.CurrentModule->addTypeName(Name, T);

  if (AlreadyExists) {   // Inserting a name that is already defined???
    const Type *Existing = CurModule.CurrentModule->getTypeByName(Name);
    assert(Existing && "Conflict but no matching type?");

    // There is only one case where this is allowed: when we are refining an
    // opaque type.  In this case, Existing will be an opaque type.
    if (const OpaqueType *OpTy = dyn_cast<OpaqueType>(Existing)) {
      // We ARE replacing an opaque type!
      const_cast<OpaqueType*>(OpTy)->refineAbstractTypeTo(T);
      return true;
    }

    // Otherwise, this is an attempt to redefine a type. That's okay if
    // the redefinition is identical to the original. This will be so if
    // Existing and T point to the same Type object. In this one case we
    // allow the equivalent redefinition.
    if (Existing == T) return true;  // Yes, it's equal.

    // Any other kind of (non-equivalent) redefinition is an error.
    ThrowException("Redefinition of type named '" + Name + "' in the '" +
                   T->getDescription() + "' type plane!");
  }

  return false;
}

//===----------------------------------------------------------------------===//
// Code for handling upreferences in type names...
//

// TypeContains - Returns true if Ty directly contains E in it.
//
static bool TypeContains(const Type *Ty, const Type *E) {
  return find(Ty->subtype_begin(), Ty->subtype_end(), E) != Ty->subtype_end();
}

namespace {
  struct UpRefRecord {
    // NestingLevel - The number of nesting levels that need to be popped before
    // this type is resolved.
    unsigned NestingLevel;
    
    // LastContainedTy - This is the type at the current binding level for the
    // type.  Every time we reduce the nesting level, this gets updated.
    const Type *LastContainedTy;

    // UpRefTy - This is the actual opaque type that the upreference is
    // represented with.
    OpaqueType *UpRefTy;

    UpRefRecord(unsigned NL, OpaqueType *URTy)
      : NestingLevel(NL), LastContainedTy(URTy), UpRefTy(URTy) {}
  };
}

// UpRefs - A list of the outstanding upreferences that need to be resolved.
static std::vector<UpRefRecord> UpRefs;

/// HandleUpRefs - Every time we finish a new layer of types, this function is
/// called.  It loops through the UpRefs vector, which is a list of the
/// currently active types.  For each type, if the up reference is contained in
/// the newly completed type, we decrement the level count.  When the level
/// count reaches zero, the upreferenced type is the type that is passed in:
/// thus we can complete the cycle.
///
static PATypeHolder HandleUpRefs(const Type *ty) {
  if (!ty->isAbstract()) return ty;
  PATypeHolder Ty(ty);
  UR_OUT("Type '" << Ty->getDescription() << 
         "' newly formed.  Resolving upreferences.\n" <<
         UpRefs.size() << " upreferences active!\n");

  // If we find any resolvable upreferences (i.e., those whose NestingLevel goes
  // to zero), we resolve them all together before we resolve them to Ty.  At
  // the end of the loop, if there is anything to resolve to Ty, it will be in
  // this variable.
  OpaqueType *TypeToResolve = 0;

  for (unsigned i = 0; i != UpRefs.size(); ++i) {
    UR_OUT("  UR#" << i << " - TypeContains(" << Ty->getDescription() << ", " 
           << UpRefs[i].second->getDescription() << ") = " 
           << (TypeContains(Ty, UpRefs[i].second) ? "true" : "false") << "\n");
    if (TypeContains(Ty, UpRefs[i].LastContainedTy)) {
      // Decrement level of upreference
      unsigned Level = --UpRefs[i].NestingLevel;
      UpRefs[i].LastContainedTy = Ty;
      UR_OUT("  Uplevel Ref Level = " << Level << "\n");
      if (Level == 0) {                     // Upreference should be resolved! 
        if (!TypeToResolve) {
          TypeToResolve = UpRefs[i].UpRefTy;
        } else {
          UR_OUT("  * Resolving upreference for "
                 << UpRefs[i].second->getDescription() << "\n";
                 std::string OldName = UpRefs[i].UpRefTy->getDescription());
          UpRefs[i].UpRefTy->refineAbstractTypeTo(TypeToResolve);
          UR_OUT("  * Type '" << OldName << "' refined upreference to: "
                 << (const void*)Ty << ", " << Ty->getDescription() << "\n");
        }
        UpRefs.erase(UpRefs.begin()+i);     // Remove from upreference list...
        --i;                                // Do not skip the next element...
      }
    }
  }

  if (TypeToResolve) {
    UR_OUT("  * Resolving upreference for "
           << UpRefs[i].second->getDescription() << "\n";
           std::string OldName = TypeToResolve->getDescription());
    TypeToResolve->refineAbstractTypeTo(Ty);
  }

  return Ty;
}


//===----------------------------------------------------------------------===//
//            RunVMAsmParser - Define an interface to this parser
//===----------------------------------------------------------------------===//
//
Module *llvm::RunVMAsmParser(const std::string &Filename, FILE *F) {
  llvmAsmin = F;
  CurFilename = Filename;
  llvmAsmlineno = 1;      // Reset the current line number...
  ObsoleteVarArgs = false;

  // Allocate a new module to read
  CurModule.CurrentModule = new Module(Filename);

  yyparse();       // Parse the file, potentially throwing exception

  Module *Result = ParserResult;

  // Check to see if they called va_start but not va_arg..
  if (!ObsoleteVarArgs)
    if (Function *F = Result->getNamedFunction("llvm.va_start"))
      if (F->asize() == 1) {
        std::cerr << "WARNING: this file uses obsolete features.  "
                  << "Assemble and disassemble to update it.\n";
        ObsoleteVarArgs = true;
      }

  if (ObsoleteVarArgs) {
    // If the user is making use of obsolete varargs intrinsics, adjust them for
    // the user.
    if (Function *F = Result->getNamedFunction("llvm.va_start")) {
      assert(F->asize() == 1 && "Obsolete va_start takes 1 argument!");

      const Type *RetTy = F->getFunctionType()->getParamType(0);
      RetTy = cast<PointerType>(RetTy)->getElementType();
      Function *NF = Result->getOrInsertFunction("llvm.va_start", RetTy, 0);
      
      while (!F->use_empty()) {
        CallInst *CI = cast<CallInst>(F->use_back());
        Value *V = new CallInst(NF, "", CI);
        new StoreInst(V, CI->getOperand(1), CI);
        CI->getParent()->getInstList().erase(CI);
      }
      Result->getFunctionList().erase(F);
    }
    
    if (Function *F = Result->getNamedFunction("llvm.va_end")) {
      assert(F->asize() == 1 && "Obsolete va_end takes 1 argument!");
      const Type *ArgTy = F->getFunctionType()->getParamType(0);
      ArgTy = cast<PointerType>(ArgTy)->getElementType();
      Function *NF = Result->getOrInsertFunction("llvm.va_end", Type::VoidTy,
                                                 ArgTy, 0);

      while (!F->use_empty()) {
        CallInst *CI = cast<CallInst>(F->use_back());
        Value *V = new LoadInst(CI->getOperand(1), "", CI);
        new CallInst(NF, V, "", CI);
        CI->getParent()->getInstList().erase(CI);
      }
      Result->getFunctionList().erase(F);
    }

    if (Function *F = Result->getNamedFunction("llvm.va_copy")) {
      assert(F->asize() == 2 && "Obsolete va_copy takes 2 argument!");
      const Type *ArgTy = F->getFunctionType()->getParamType(0);
      ArgTy = cast<PointerType>(ArgTy)->getElementType();
      Function *NF = Result->getOrInsertFunction("llvm.va_copy", ArgTy,
                                                 ArgTy, 0);

      while (!F->use_empty()) {
        CallInst *CI = cast<CallInst>(F->use_back());
        Value *V = new CallInst(NF, CI->getOperand(2), "", CI);
        new StoreInst(V, CI->getOperand(1), CI);
        CI->getParent()->getInstList().erase(CI);
      }
      Result->getFunctionList().erase(F);
    }
  }

  llvmAsmin = stdin;    // F is about to go away, don't use it anymore...
  ParserResult = 0;

  return Result;
}

%}

%union {
  llvm::Module                           *ModuleVal;
  llvm::Function                         *FunctionVal;
  std::pair<llvm::PATypeHolder*, char*>  *ArgVal;
  llvm::BasicBlock                       *BasicBlockVal;
  llvm::TerminatorInst                   *TermInstVal;
  llvm::Instruction                      *InstVal;
  llvm::Constant                         *ConstVal;

  const llvm::Type                       *PrimType;
  llvm::PATypeHolder                     *TypeVal;
  llvm::Value                            *ValueVal;

  std::vector<std::pair<llvm::PATypeHolder*,char*> > *ArgList;
  std::vector<llvm::Value*>              *ValueList;
  std::list<llvm::PATypeHolder>          *TypeList;
  std::list<std::pair<llvm::Value*,
                      llvm::BasicBlock*> > *PHIList; // Represent the RHS of PHI node
  std::vector<std::pair<llvm::Constant*, llvm::BasicBlock*> > *JumpTable;
  std::vector<llvm::Constant*>           *ConstVector;

  llvm::GlobalValue::LinkageTypes         Linkage;
  int64_t                           SInt64Val;
  uint64_t                          UInt64Val;
  int                               SIntVal;
  unsigned                          UIntVal;
  double                            FPVal;
  bool                              BoolVal;

  char                             *StrVal;   // This memory is strdup'd!
  llvm::ValID                             ValIDVal; // strdup'd memory maybe!

  llvm::Instruction::BinaryOps            BinaryOpVal;
  llvm::Instruction::TermOps              TermOpVal;
  llvm::Instruction::MemoryOps            MemOpVal;
  llvm::Instruction::OtherOps             OtherOpVal;
  llvm::Module::Endianness                Endianness;
}

%type <ModuleVal>     Module FunctionList
%type <FunctionVal>   Function FunctionProto FunctionHeader BasicBlockList
%type <BasicBlockVal> BasicBlock InstructionList
%type <TermInstVal>   BBTerminatorInst
%type <InstVal>       Inst InstVal MemoryInst
%type <ConstVal>      ConstVal ConstExpr
%type <ConstVector>   ConstVector
%type <ArgList>       ArgList ArgListH
%type <ArgVal>        ArgVal
%type <PHIList>       PHIList
%type <ValueList>     ValueRefList ValueRefListE  // For call param lists
%type <ValueList>     IndexList                   // For GEP derived indices
%type <TypeList>      TypeListI ArgTypeListI
%type <JumpTable>     JumpTable
%type <BoolVal>       GlobalType                  // GLOBAL or CONSTANT?
%type <BoolVal>       OptVolatile                 // 'volatile' or not
%type <Linkage>       OptLinkage
%type <Endianness>    BigOrLittle

// ValueRef - Unresolved reference to a definition or BB
%type <ValIDVal>      ValueRef ConstValueRef SymbolicValueRef
%type <ValueVal>      ResolvedVal            // <type> <valref> pair
// Tokens and types for handling constant integer values
//
// ESINT64VAL - A negative number within long long range
%token <SInt64Val> ESINT64VAL

// EUINT64VAL - A positive number within uns. long long range
%token <UInt64Val> EUINT64VAL
%type  <SInt64Val> EINT64VAL

%token  <SIntVal>   SINTVAL   // Signed 32 bit ints...
%token  <UIntVal>   UINTVAL   // Unsigned 32 bit ints...
%type   <SIntVal>   INTVAL
%token  <FPVal>     FPVAL     // Float or Double constant

// Built in types...
%type  <TypeVal> Types TypesV UpRTypes UpRTypesV
%type  <PrimType> SIntType UIntType IntType FPType PrimType   // Classifications
%token <PrimType> VOID BOOL SBYTE UBYTE SHORT USHORT INT UINT LONG ULONG
%token <PrimType> FLOAT DOUBLE TYPE LABEL

%token <StrVal> VAR_ID LABELSTR STRINGCONSTANT
%type  <StrVal> Name OptName OptAssign


%token IMPLEMENTATION ZEROINITIALIZER TRUETOK FALSETOK BEGINTOK ENDTOK
%token DECLARE GLOBAL CONSTANT VOLATILE
%token TO DOTDOTDOT NULL_TOK UNDEF CONST INTERNAL LINKONCE WEAK  APPENDING
%token OPAQUE NOT EXTERNAL TARGET TRIPLE ENDIAN POINTERSIZE LITTLE BIG
%token DEPLIBS 

// Basic Block Terminating Operators 
%token <TermOpVal> RET BR SWITCH INVOKE UNWIND UNREACHABLE

// Binary Operators 
%type  <BinaryOpVal> ArithmeticOps LogicalOps SetCondOps // Binops Subcatagories
%token <BinaryOpVal> ADD SUB MUL DIV REM AND OR XOR
%token <BinaryOpVal> SETLE SETGE SETLT SETGT SETEQ SETNE  // Binary Comarators

// Memory Instructions
%token <MemOpVal> MALLOC ALLOCA FREE LOAD STORE GETELEMENTPTR

// Other Operators
%type  <OtherOpVal> ShiftOps
%token <OtherOpVal> PHI_TOK CALL CAST SELECT SHL SHR VAARG VANEXT
%token VA_ARG // FIXME: OBSOLETE

%start Module
%%

// Handle constant integer size restriction and conversion...
//
INTVAL : SINTVAL;
INTVAL : UINTVAL {
  if ($1 > (uint32_t)INT32_MAX)     // Outside of my range!
    ThrowException("Value too large for type!");
  $$ = (int32_t)$1;
};


EINT64VAL : ESINT64VAL;      // These have same type and can't cause problems...
EINT64VAL : EUINT64VAL {
  if ($1 > (uint64_t)INT64_MAX)     // Outside of my range!
    ThrowException("Value too large for type!");
  $$ = (int64_t)$1;
};

// Operations that are notably excluded from this list include: 
// RET, BR, & SWITCH because they end basic blocks and are treated specially.
//
ArithmeticOps: ADD | SUB | MUL | DIV | REM;
LogicalOps   : AND | OR | XOR;
SetCondOps   : SETLE | SETGE | SETLT | SETGT | SETEQ | SETNE;

ShiftOps  : SHL | SHR;

// These are some types that allow classification if we only want a particular 
// thing... for example, only a signed, unsigned, or integral type.
SIntType :  LONG |  INT |  SHORT | SBYTE;
UIntType : ULONG | UINT | USHORT | UBYTE;
IntType  : SIntType | UIntType;
FPType   : FLOAT | DOUBLE;

// OptAssign - Value producing statements have an optional assignment component
OptAssign : Name '=' {
    $$ = $1;
  }
  | /*empty*/ { 
    $$ = 0; 
  };

OptLinkage : INTERNAL  { $$ = GlobalValue::InternalLinkage; } |
             LINKONCE  { $$ = GlobalValue::LinkOnceLinkage; } |
             WEAK      { $$ = GlobalValue::WeakLinkage; } |
             APPENDING { $$ = GlobalValue::AppendingLinkage; } |
             /*empty*/ { $$ = GlobalValue::ExternalLinkage; };

//===----------------------------------------------------------------------===//
// Types includes all predefined types... except void, because it can only be
// used in specific contexts (function returning void for example).  To have
// access to it, a user must explicitly use TypesV.
//

// TypesV includes all of 'Types', but it also includes the void type.
TypesV    : Types    | VOID { $$ = new PATypeHolder($1); };
UpRTypesV : UpRTypes | VOID { $$ = new PATypeHolder($1); };

Types     : UpRTypes {
    if (!UpRefs.empty())
      ThrowException("Invalid upreference in type: " + (*$1)->getDescription());
    $$ = $1;
  };