Twine.h revision 37e3fe9ad7d7cb350cbbce0695c68d652d624bb4
1//===-- Twine.h - Fast Temporary String Concatenation -----------*- C++ -*-===// 2// 3// The LLVM Compiler Infrastructure 4// 5// This file is distributed under the University of Illinois Open Source 6// License. See LICENSE.TXT for details. 7// 8//===----------------------------------------------------------------------===// 9 10#ifndef LLVM_ADT_TWINE_H 11#define LLVM_ADT_TWINE_H 12 13#include "llvm/ADT/StringRef.h" 14#include "llvm/Support/DataTypes.h" 15#include <cassert> 16#include <string> 17 18namespace llvm { 19 template <typename T> 20 class SmallVectorImpl; 21 class StringRef; 22 class raw_ostream; 23 24 /// Twine - A lightweight data structure for efficiently representing the 25 /// concatenation of temporary values as strings. 26 /// 27 /// A Twine is a kind of rope, it represents a concatenated string using a 28 /// binary-tree, where the string is the preorder of the nodes. Since the 29 /// Twine can be efficiently rendered into a buffer when its result is used, 30 /// it avoids the cost of generating temporary values for intermediate string 31 /// results -- particularly in cases when the Twine result is never 32 /// required. By explicitly tracking the type of leaf nodes, we can also avoid 33 /// the creation of temporary strings for conversions operations (such as 34 /// appending an integer to a string). 35 /// 36 /// A Twine is not intended for use directly and should not be stored, its 37 /// implementation relies on the ability to store pointers to temporary stack 38 /// objects which may be deallocated at the end of a statement. Twines should 39 /// only be used accepted as const references in arguments, when an API wishes 40 /// to accept possibly-concatenated strings. 41 /// 42 /// Twines support a special 'null' value, which always concatenates to form 43 /// itself, and renders as an empty string. This can be returned from APIs to 44 /// effectively nullify any concatenations performed on the result. 45 /// 46 /// \b Implementation \n 47 /// 48 /// Given the nature of a Twine, it is not possible for the Twine's 49 /// concatenation method to construct interior nodes; the result must be 50 /// represented inside the returned value. For this reason a Twine object 51 /// actually holds two values, the left- and right-hand sides of a 52 /// concatenation. We also have nullary Twine objects, which are effectively 53 /// sentinel values that represent empty strings. 54 /// 55 /// Thus, a Twine can effectively have zero, one, or two children. The \see 56 /// isNullary(), \see isUnary(), and \see isBinary() predicates exist for 57 /// testing the number of children. 58 /// 59 /// We maintain a number of invariants on Twine objects (FIXME: Why): 60 /// - Nullary twines are always represented with their Kind on the left-hand 61 /// side, and the Empty kind on the right-hand side. 62 /// - Unary twines are always represented with the value on the left-hand 63 /// side, and the Empty kind on the right-hand side. 64 /// - If a Twine has another Twine as a child, that child should always be 65 /// binary (otherwise it could have been folded into the parent). 66 /// 67 /// These invariants are check by \see isValid(). 68 /// 69 /// \b Efficiency Considerations \n 70 /// 71 /// The Twine is designed to yield efficient and small code for common 72 /// situations. For this reason, the concat() method is inlined so that 73 /// concatenations of leaf nodes can be optimized into stores directly into a 74 /// single stack allocated object. 75 /// 76 /// In practice, not all compilers can be trusted to optimize concat() fully, 77 /// so we provide two additional methods (and accompanying operator+ 78 /// overloads) to guarantee that particularly important cases (cstring plus 79 /// StringRef) codegen as desired. 80 class Twine { 81 /// NodeKind - Represent the type of an argument. 82 enum NodeKind { 83 /// An empty string; the result of concatenating anything with it is also 84 /// empty. 85 NullKind, 86 87 /// The empty string. 88 EmptyKind, 89 90 /// A pointer to a Twine instance. 91 TwineKind, 92 93 /// A pointer to a C string instance. 94 CStringKind, 95 96 /// A pointer to an std::string instance. 97 StdStringKind, 98 99 /// A pointer to a StringRef instance. 100 StringRefKind, 101 102 /// A pointer to a uint64_t value, to render as an unsigned decimal 103 /// integer. 104 UDecKind, 105 106 /// A pointer to a uint64_t value, to render as an unsigned hexadecimal 107 /// integer. 108 UHexKind, 109 110 /// A pointer to a uint64_t value, to render as a signed decimal integer. 111 SDecKind 112 }; 113 114 private: 115 /// LHS - The prefix in the concatenation, which may be uninitialized for 116 /// Null or Empty kinds. 117 const void *LHS; 118 /// RHS - The suffix in the concatenation, which may be uninitialized for 119 /// Null or Empty kinds. 120 const void *RHS; 121 /// LHSKind - The NodeKind of the left hand side, \see getLHSKind(). 122 NodeKind LHSKind : 8; 123 /// RHSKind - The NodeKind of the left hand side, \see getLHSKind(). 124 NodeKind RHSKind : 8; 125 126 private: 127 /// Construct a nullary twine; the kind must be NullKind or EmptyKind. 128 explicit Twine(NodeKind Kind) 129 : LHSKind(Kind), RHSKind(EmptyKind) { 130 assert(isNullary() && "Invalid kind!"); 131 } 132 133 /// Construct a binary twine. 134 explicit Twine(const Twine &_LHS, const Twine &_RHS) 135 : LHS(&_LHS), RHS(&_RHS), LHSKind(TwineKind), RHSKind(TwineKind) { 136 assert(isValid() && "Invalid twine!"); 137 } 138 139 /// Construct a twine from explicit values. 140 explicit Twine(const void *_LHS, NodeKind _LHSKind, 141 const void *_RHS, NodeKind _RHSKind) 142 : LHS(_LHS), RHS(_RHS), LHSKind(_LHSKind), RHSKind(_RHSKind) { 143 assert(isValid() && "Invalid twine!"); 144 } 145 146 /// isNull - Check for the null twine. 147 bool isNull() const { 148 return getLHSKind() == NullKind; 149 } 150 151 /// isEmpty - Check for the empty twine. 152 bool isEmpty() const { 153 return getLHSKind() == EmptyKind; 154 } 155 156 /// isNullary - Check if this is a nullary twine (null or empty). 157 bool isNullary() const { 158 return isNull() || isEmpty(); 159 } 160 161 /// isUnary - Check if this is a unary twine. 162 bool isUnary() const { 163 return getRHSKind() == EmptyKind && !isNullary(); 164 } 165 166 /// isBinary - Check if this is a binary twine. 167 bool isBinary() const { 168 return getLHSKind() != NullKind && getRHSKind() != EmptyKind; 169 } 170 171 /// isValid - Check if this is a valid twine (satisfying the invariants on 172 /// order and number of arguments). 173 bool isValid() const { 174 // Nullary twines always have Empty on the RHS. 175 if (isNullary() && getRHSKind() != EmptyKind) 176 return false; 177 178 // Null should never appear on the RHS. 179 if (getRHSKind() == NullKind) 180 return false; 181 182 // The RHS cannot be non-empty if the LHS is empty. 183 if (getRHSKind() != EmptyKind && getLHSKind() == EmptyKind) 184 return false; 185 186 // A twine child should always be binary. 187 if (getLHSKind() == TwineKind && 188 !static_cast<const Twine*>(LHS)->isBinary()) 189 return false; 190 if (getRHSKind() == TwineKind && 191 !static_cast<const Twine*>(RHS)->isBinary()) 192 return false; 193 194 return true; 195 } 196 197 /// getLHSKind - Get the NodeKind of the left-hand side. 198 NodeKind getLHSKind() const { return LHSKind; } 199 200 /// getRHSKind - Get the NodeKind of the left-hand side. 201 NodeKind getRHSKind() const { return RHSKind; } 202 203 /// printOneChild - Print one child from a twine. 204 void printOneChild(raw_ostream &OS, const void *Ptr, NodeKind Kind) const; 205 206 /// printOneChildRepr - Print the representation of one child from a twine. 207 void printOneChildRepr(raw_ostream &OS, const void *Ptr, 208 NodeKind Kind) const; 209 210 public: 211 /// @name Constructors 212 /// @{ 213 214 /// Construct from an empty string. 215 /*implicit*/ Twine() : LHSKind(EmptyKind), RHSKind(EmptyKind) { 216 assert(isValid() && "Invalid twine!"); 217 } 218 219 /// Construct from a C string. 220 /// 221 /// We take care here to optimize "" into the empty twine -- this will be 222 /// optimized out for string constants. This allows Twine arguments have 223 /// default "" values, without introducing unnecessary string constants. 224 /*implicit*/ Twine(const char *Str) 225 : RHSKind(EmptyKind) { 226 if (Str[0] != '\0') { 227 LHS = Str; 228 LHSKind = CStringKind; 229 } else 230 LHSKind = EmptyKind; 231 232 assert(isValid() && "Invalid twine!"); 233 } 234 235 /// Construct from an std::string. 236 /*implicit*/ Twine(const std::string &Str) 237 : LHS(&Str), LHSKind(StdStringKind), RHSKind(EmptyKind) { 238 assert(isValid() && "Invalid twine!"); 239 } 240 241 /// Construct from a StringRef. 242 /*implicit*/ Twine(const StringRef &Str) 243 : LHS(&Str), LHSKind(StringRefKind), RHSKind(EmptyKind) { 244 assert(isValid() && "Invalid twine!"); 245 } 246 247 // FIXME: Unfortunately, to make sure this is as efficient as possible we 248 // need extra binary constructors from particular types. We can't rely on 249 // the compiler to be smart enough to fold operator+()/concat() down to the 250 // right thing. Yet. 251 252 /// Construct as the concatenation of a C string and a StringRef. 253 /*implicit*/ Twine(const char *_LHS, const StringRef &_RHS) 254 : LHS(_LHS), RHS(&_RHS), LHSKind(CStringKind), RHSKind(StringRefKind) { 255 assert(isValid() && "Invalid twine!"); 256 } 257 258 /// Construct as the concatenation of a StringRef and a C string. 259 /*implicit*/ Twine(const StringRef &_LHS, const char *_RHS) 260 : LHS(&_LHS), RHS(_RHS), LHSKind(StringRefKind), RHSKind(CStringKind) { 261 assert(isValid() && "Invalid twine!"); 262 } 263 264 /// Create a 'null' string, which is an empty string that always 265 /// concatenates to form another empty string. 266 static Twine createNull() { 267 return Twine(NullKind); 268 } 269 270 /// @} 271 /// @name Numeric Conversions 272 /// @{ 273 274 /// Construct a twine to print \arg Val as an unsigned decimal integer. 275 static Twine utostr(const uint64_t &Val) { 276 return Twine(&Val, UDecKind, 0, EmptyKind); 277 } 278 279 /// Construct a twine to print \arg Val as a signed decimal integer. 280 static Twine itostr(const int64_t &Val) { 281 return Twine(&Val, SDecKind, 0, EmptyKind); 282 } 283 284 // Construct a twine to print \arg Val as an unsigned hexadecimal integer. 285 static Twine utohexstr(const uint64_t &Val) { 286 return Twine(&Val, UHexKind, 0, EmptyKind); 287 } 288 289 // Construct a twine to print \arg Val as an unsigned hexadecimal 290 // integer. This routine is provided as a convenience to sign extend values 291 // before printing. 292 static Twine itohexstr(const int64_t &Val) { 293 return Twine(&Val, UHexKind, 0, EmptyKind); 294 } 295 296 /// @} 297 /// @name String Operations 298 /// @{ 299 300 Twine concat(const Twine &Suffix) const; 301 302 /// @} 303 /// @name Output & Conversion. 304 /// @{ 305 306 /// str - Return the twine contents as a std::string. 307 std::string str() const; 308 309 /// toVector - Write the concatenated string into the given SmallString or 310 /// SmallVector. 311 void toVector(SmallVectorImpl<char> &Out) const; 312 313 /// print - Write the concatenated string represented by this twine to the 314 /// stream \arg OS. 315 void print(raw_ostream &OS) const; 316 317 /// dump - Dump the concatenated string represented by this twine to stderr. 318 void dump() const; 319 320 /// print - Write the representation of this twine to the stream \arg OS. 321 void printRepr(raw_ostream &OS) const; 322 323 /// dumpRepr - Dump the representation of this twine to stderr. 324 void dumpRepr() const; 325 326 /// @} 327 }; 328 329 /// @name Twine Inline Implementations 330 /// @{ 331 332 inline Twine Twine::concat(const Twine &Suffix) const { 333 // Concatenation with null is null. 334 if (isNull() || Suffix.isNull()) 335 return Twine(NullKind); 336 337 // Concatenation with empty yields the other side. 338 if (isEmpty()) 339 return Suffix; 340 if (Suffix.isEmpty()) 341 return *this; 342 343 // Otherwise we need to create a new node, taking care to fold in unary 344 // twines. 345 const void *NewLHS = this, *NewRHS = &Suffix; 346 NodeKind NewLHSKind = TwineKind, NewRHSKind = TwineKind; 347 if (isUnary()) { 348 NewLHS = LHS; 349 NewLHSKind = getLHSKind(); 350 } 351 if (Suffix.isUnary()) { 352 NewRHS = Suffix.LHS; 353 NewRHSKind = Suffix.getLHSKind(); 354 } 355 356 return Twine(NewLHS, NewLHSKind, NewRHS, NewRHSKind); 357 } 358 359 inline Twine operator+(const Twine &LHS, const Twine &RHS) { 360 return LHS.concat(RHS); 361 } 362 363 /// Additional overload to guarantee simplified codegen; this is equivalent to 364 /// concat(). 365 366 inline Twine operator+(const char *LHS, const StringRef &RHS) { 367 return Twine(LHS, RHS); 368 } 369 370 /// Additional overload to guarantee simplified codegen; this is equivalent to 371 /// concat(). 372 373 inline Twine operator+(const StringRef &LHS, const char *RHS) { 374 return Twine(LHS, RHS); 375 } 376 377 inline raw_ostream &operator<<(raw_ostream &OS, const Twine &RHS) { 378 RHS.print(OS); 379 return OS; 380 } 381 382 /// @} 383} 384 385#endif 386