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 "llvm/Support/ErrorHandling.h" 16 #include <cassert> 17 #include <string> 18 19 namespace llvm { 20 template <typename T> 21 class SmallVectorImpl; 22 class StringRef; 23 class raw_ostream; 24 25 /// Twine - A lightweight data structure for efficiently representing the 26 /// concatenation of temporary values as strings. 27 /// 28 /// A Twine is a kind of rope, it represents a concatenated string using a 29 /// binary-tree, where the string is the preorder of the nodes. Since the 30 /// Twine can be efficiently rendered into a buffer when its result is used, 31 /// it avoids the cost of generating temporary values for intermediate string 32 /// results -- particularly in cases when the Twine result is never 33 /// required. By explicitly tracking the type of leaf nodes, we can also avoid 34 /// the creation of temporary strings for conversions operations (such as 35 /// appending an integer to a string). 36 /// 37 /// A Twine is not intended for use directly and should not be stored, its 38 /// implementation relies on the ability to store pointers to temporary stack 39 /// objects which may be deallocated at the end of a statement. Twines should 40 /// only be used accepted as const references in arguments, when an API wishes 41 /// to accept possibly-concatenated strings. 42 /// 43 /// Twines support a special 'null' value, which always concatenates to form 44 /// itself, and renders as an empty string. This can be returned from APIs to 45 /// effectively nullify any concatenations performed on the result. 46 /// 47 /// \b Implementation \n 48 /// 49 /// Given the nature of a Twine, it is not possible for the Twine's 50 /// concatenation method to construct interior nodes; the result must be 51 /// represented inside the returned value. For this reason a Twine object 52 /// actually holds two values, the left- and right-hand sides of a 53 /// concatenation. We also have nullary Twine objects, which are effectively 54 /// sentinel values that represent empty strings. 55 /// 56 /// Thus, a Twine can effectively have zero, one, or two children. The \see 57 /// isNullary(), \see isUnary(), and \see isBinary() predicates exist for 58 /// testing the number of children. 59 /// 60 /// We maintain a number of invariants on Twine objects (FIXME: Why): 61 /// - Nullary twines are always represented with their Kind on the left-hand 62 /// side, and the Empty kind on the right-hand side. 63 /// - Unary twines are always represented with the value on the left-hand 64 /// side, and the Empty kind on the right-hand side. 65 /// - If a Twine has another Twine as a child, that child should always be 66 /// binary (otherwise it could have been folded into the parent). 67 /// 68 /// These invariants are check by \see isValid(). 69 /// 70 /// \b Efficiency Considerations \n 71 /// 72 /// The Twine is designed to yield efficient and small code for common 73 /// situations. For this reason, the concat() method is inlined so that 74 /// concatenations of leaf nodes can be optimized into stores directly into a 75 /// single stack allocated object. 76 /// 77 /// In practice, not all compilers can be trusted to optimize concat() fully, 78 /// so we provide two additional methods (and accompanying operator+ 79 /// overloads) to guarantee that particularly important cases (cstring plus 80 /// StringRef) codegen as desired. 81 class Twine { 82 /// NodeKind - Represent the type of an argument. 83 enum NodeKind { 84 /// An empty string; the result of concatenating anything with it is also 85 /// empty. 86 NullKind, 87 88 /// The empty string. 89 EmptyKind, 90 91 /// A pointer to a Twine instance. 92 TwineKind, 93 94 /// A pointer to a C string instance. 95 CStringKind, 96 97 /// A pointer to an std::string instance. 98 StdStringKind, 99 100 /// A pointer to a StringRef instance. 101 StringRefKind, 102 103 /// A char value reinterpreted as a pointer, to render as a character. 104 CharKind, 105 106 /// An unsigned int value reinterpreted as a pointer, to render as an 107 /// unsigned decimal integer. 108 DecUIKind, 109 110 /// An int value reinterpreted as a pointer, to render as a signed 111 /// decimal integer. 112 DecIKind, 113 114 /// A pointer to an unsigned long value, to render as an unsigned decimal 115 /// integer. 116 DecULKind, 117 118 /// A pointer to a long value, to render as a signed decimal integer. 119 DecLKind, 120 121 /// A pointer to an unsigned long long value, to render as an unsigned 122 /// decimal integer. 123 DecULLKind, 124 125 /// A pointer to a long long value, to render as a signed decimal integer. 126 DecLLKind, 127 128 /// A pointer to a uint64_t value, to render as an unsigned hexadecimal 129 /// integer. 130 UHexKind 131 }; 132 133 union Child 134 { 135 const Twine *twine; 136 const char *cString; 137 const std::string *stdString; 138 const StringRef *stringRef; 139 char character; 140 unsigned int decUI; 141 int decI; 142 const unsigned long *decUL; 143 const long *decL; 144 const unsigned long long *decULL; 145 const long long *decLL; 146 const uint64_t *uHex; 147 }; 148 149 private: 150 /// LHS - The prefix in the concatenation, which may be uninitialized for 151 /// Null or Empty kinds. 152 Child LHS; 153 /// RHS - The suffix in the concatenation, which may be uninitialized for 154 /// Null or Empty kinds. 155 Child RHS; 156 // enums stored as unsigned chars to save on space while some compilers 157 // don't support specifying the backing type for an enum 158 /// LHSKind - The NodeKind of the left hand side, \see getLHSKind(). 159 unsigned char LHSKind; 160 /// RHSKind - The NodeKind of the left hand side, \see getLHSKind(). 161 unsigned char RHSKind; 162 163 private: 164 /// Construct a nullary twine; the kind must be NullKind or EmptyKind. 165 explicit Twine(NodeKind Kind) 166 : LHSKind(Kind), RHSKind(EmptyKind) { 167 assert(isNullary() && "Invalid kind!"); 168 } 169 170 /// Construct a binary twine. 171 explicit Twine(const Twine &_LHS, const Twine &_RHS) 172 : LHSKind(TwineKind), RHSKind(TwineKind) { 173 LHS.twine = &_LHS; 174 RHS.twine = &_RHS; 175 assert(isValid() && "Invalid twine!"); 176 } 177 178 /// Construct a twine from explicit values. 179 explicit Twine(Child _LHS, NodeKind _LHSKind, 180 Child _RHS, NodeKind _RHSKind) 181 : LHS(_LHS), RHS(_RHS), LHSKind(_LHSKind), RHSKind(_RHSKind) { 182 assert(isValid() && "Invalid twine!"); 183 } 184 185 /// isNull - Check for the null twine. 186 bool isNull() const { 187 return getLHSKind() == NullKind; 188 } 189 190 /// isEmpty - Check for the empty twine. 191 bool isEmpty() const { 192 return getLHSKind() == EmptyKind; 193 } 194 195 /// isNullary - Check if this is a nullary twine (null or empty). 196 bool isNullary() const { 197 return isNull() || isEmpty(); 198 } 199 200 /// isUnary - Check if this is a unary twine. 201 bool isUnary() const { 202 return getRHSKind() == EmptyKind && !isNullary(); 203 } 204 205 /// isBinary - Check if this is a binary twine. 206 bool isBinary() const { 207 return getLHSKind() != NullKind && getRHSKind() != EmptyKind; 208 } 209 210 /// isValid - Check if this is a valid twine (satisfying the invariants on 211 /// order and number of arguments). 212 bool isValid() const { 213 // Nullary twines always have Empty on the RHS. 214 if (isNullary() && getRHSKind() != EmptyKind) 215 return false; 216 217 // Null should never appear on the RHS. 218 if (getRHSKind() == NullKind) 219 return false; 220 221 // The RHS cannot be non-empty if the LHS is empty. 222 if (getRHSKind() != EmptyKind && getLHSKind() == EmptyKind) 223 return false; 224 225 // A twine child should always be binary. 226 if (getLHSKind() == TwineKind && 227 !LHS.twine->isBinary()) 228 return false; 229 if (getRHSKind() == TwineKind && 230 !RHS.twine->isBinary()) 231 return false; 232 233 return true; 234 } 235 236 /// getLHSKind - Get the NodeKind of the left-hand side. 237 NodeKind getLHSKind() const { return (NodeKind) LHSKind; } 238 239 /// getRHSKind - Get the NodeKind of the left-hand side. 240 NodeKind getRHSKind() const { return (NodeKind) RHSKind; } 241 242 /// printOneChild - Print one child from a twine. 243 void printOneChild(raw_ostream &OS, Child Ptr, NodeKind Kind) const; 244 245 /// printOneChildRepr - Print the representation of one child from a twine. 246 void printOneChildRepr(raw_ostream &OS, Child Ptr, 247 NodeKind Kind) const; 248 249 public: 250 /// @name Constructors 251 /// @{ 252 253 /// Construct from an empty string. 254 /*implicit*/ Twine() : LHSKind(EmptyKind), RHSKind(EmptyKind) { 255 assert(isValid() && "Invalid twine!"); 256 } 257 258 /// Construct from a C string. 259 /// 260 /// We take care here to optimize "" into the empty twine -- this will be 261 /// optimized out for string constants. This allows Twine arguments have 262 /// default "" values, without introducing unnecessary string constants. 263 /*implicit*/ Twine(const char *Str) 264 : RHSKind(EmptyKind) { 265 if (Str[0] != '\0') { 266 LHS.cString = Str; 267 LHSKind = CStringKind; 268 } else 269 LHSKind = EmptyKind; 270 271 assert(isValid() && "Invalid twine!"); 272 } 273 274 /// Construct from an std::string. 275 /*implicit*/ Twine(const std::string &Str) 276 : LHSKind(StdStringKind), RHSKind(EmptyKind) { 277 LHS.stdString = &Str; 278 assert(isValid() && "Invalid twine!"); 279 } 280 281 /// Construct from a StringRef. 282 /*implicit*/ Twine(const StringRef &Str) 283 : LHSKind(StringRefKind), RHSKind(EmptyKind) { 284 LHS.stringRef = &Str; 285 assert(isValid() && "Invalid twine!"); 286 } 287 288 /// Construct from a char. 289 explicit Twine(char Val) 290 : LHSKind(CharKind), RHSKind(EmptyKind) { 291 LHS.character = Val; 292 } 293 294 /// Construct from a signed char. 295 explicit Twine(signed char Val) 296 : LHSKind(CharKind), RHSKind(EmptyKind) { 297 LHS.character = static_cast<char>(Val); 298 } 299 300 /// Construct from an unsigned char. 301 explicit Twine(unsigned char Val) 302 : LHSKind(CharKind), RHSKind(EmptyKind) { 303 LHS.character = static_cast<char>(Val); 304 } 305 306 /// Construct a twine to print \arg Val as an unsigned decimal integer. 307 explicit Twine(unsigned Val) 308 : LHSKind(DecUIKind), RHSKind(EmptyKind) { 309 LHS.decUI = Val; 310 } 311 312 /// Construct a twine to print \arg Val as a signed decimal integer. 313 explicit Twine(int Val) 314 : LHSKind(DecIKind), RHSKind(EmptyKind) { 315 LHS.decI = Val; 316 } 317 318 /// Construct a twine to print \arg Val as an unsigned decimal integer. 319 explicit Twine(const unsigned long &Val) 320 : LHSKind(DecULKind), RHSKind(EmptyKind) { 321 LHS.decUL = &Val; 322 } 323 324 /// Construct a twine to print \arg Val as a signed decimal integer. 325 explicit Twine(const long &Val) 326 : LHSKind(DecLKind), RHSKind(EmptyKind) { 327 LHS.decL = &Val; 328 } 329 330 /// Construct a twine to print \arg Val as an unsigned decimal integer. 331 explicit Twine(const unsigned long long &Val) 332 : LHSKind(DecULLKind), RHSKind(EmptyKind) { 333 LHS.decULL = &Val; 334 } 335 336 /// Construct a twine to print \arg Val as a signed decimal integer. 337 explicit Twine(const long long &Val) 338 : LHSKind(DecLLKind), RHSKind(EmptyKind) { 339 LHS.decLL = &Val; 340 } 341 342 // FIXME: Unfortunately, to make sure this is as efficient as possible we 343 // need extra binary constructors from particular types. We can't rely on 344 // the compiler to be smart enough to fold operator+()/concat() down to the 345 // right thing. Yet. 346 347 /// Construct as the concatenation of a C string and a StringRef. 348 /*implicit*/ Twine(const char *_LHS, const StringRef &_RHS) 349 : LHSKind(CStringKind), RHSKind(StringRefKind) { 350 LHS.cString = _LHS; 351 RHS.stringRef = &_RHS; 352 assert(isValid() && "Invalid twine!"); 353 } 354 355 /// Construct as the concatenation of a StringRef and a C string. 356 /*implicit*/ Twine(const StringRef &_LHS, const char *_RHS) 357 : LHSKind(StringRefKind), RHSKind(CStringKind) { 358 LHS.stringRef = &_LHS; 359 RHS.cString = _RHS; 360 assert(isValid() && "Invalid twine!"); 361 } 362 363 /// Create a 'null' string, which is an empty string that always 364 /// concatenates to form another empty string. 365 static Twine createNull() { 366 return Twine(NullKind); 367 } 368 369 /// @} 370 /// @name Numeric Conversions 371 /// @{ 372 373 // Construct a twine to print \arg Val as an unsigned hexadecimal integer. 374 static Twine utohexstr(const uint64_t &Val) { 375 Child LHS, RHS; 376 LHS.uHex = &Val; 377 RHS.twine = 0; 378 return Twine(LHS, UHexKind, RHS, EmptyKind); 379 } 380 381 /// @} 382 /// @name Predicate Operations 383 /// @{ 384 385 /// isTriviallyEmpty - Check if this twine is trivially empty; a false 386 /// return value does not necessarily mean the twine is empty. 387 bool isTriviallyEmpty() const { 388 return isNullary(); 389 } 390 391 /// isSingleStringRef - Return true if this twine can be dynamically 392 /// accessed as a single StringRef value with getSingleStringRef(). 393 bool isSingleStringRef() const { 394 if (getRHSKind() != EmptyKind) return false; 395 396 switch (getLHSKind()) { 397 case EmptyKind: 398 case CStringKind: 399 case StdStringKind: 400 case StringRefKind: 401 return true; 402 default: 403 return false; 404 } 405 } 406 407 /// @} 408 /// @name String Operations 409 /// @{ 410 411 Twine concat(const Twine &Suffix) const; 412 413 /// @} 414 /// @name Output & Conversion. 415 /// @{ 416 417 /// str - Return the twine contents as a std::string. 418 std::string str() const; 419 420 /// toVector - Write the concatenated string into the given SmallString or 421 /// SmallVector. 422 void toVector(SmallVectorImpl<char> &Out) const; 423 424 /// getSingleStringRef - This returns the twine as a single StringRef. This 425 /// method is only valid if isSingleStringRef() is true. 426 StringRef getSingleStringRef() const { 427 assert(isSingleStringRef() &&"This cannot be had as a single stringref!"); 428 switch (getLHSKind()) { 429 default: llvm_unreachable("Out of sync with isSingleStringRef"); 430 case EmptyKind: return StringRef(); 431 case CStringKind: return StringRef(LHS.cString); 432 case StdStringKind: return StringRef(*LHS.stdString); 433 case StringRefKind: return *LHS.stringRef; 434 } 435 } 436 437 /// toStringRef - This returns the twine as a single StringRef if it can be 438 /// represented as such. Otherwise the twine is written into the given 439 /// SmallVector and a StringRef to the SmallVector's data is returned. 440 StringRef toStringRef(SmallVectorImpl<char> &Out) const; 441 442 /// toNullTerminatedStringRef - This returns the twine as a single null 443 /// terminated StringRef if it can be represented as such. Otherwise the 444 /// twine is written into the given SmallVector and a StringRef to the 445 /// SmallVector's data is returned. 446 /// 447 /// The returned StringRef's size does not include the null terminator. 448 StringRef toNullTerminatedStringRef(SmallVectorImpl<char> &Out) const; 449 450 /// print - Write the concatenated string represented by this twine to the 451 /// stream \arg OS. 452 void print(raw_ostream &OS) const; 453 454 /// dump - Dump the concatenated string represented by this twine to stderr. 455 void dump() const; 456 457 /// print - Write the representation of this twine to the stream \arg OS. 458 void printRepr(raw_ostream &OS) const; 459 460 /// dumpRepr - Dump the representation of this twine to stderr. 461 void dumpRepr() const; 462 463 /// @} 464 }; 465 466 /// @name Twine Inline Implementations 467 /// @{ 468 469 inline Twine Twine::concat(const Twine &Suffix) const { 470 // Concatenation with null is null. 471 if (isNull() || Suffix.isNull()) 472 return Twine(NullKind); 473 474 // Concatenation with empty yields the other side. 475 if (isEmpty()) 476 return Suffix; 477 if (Suffix.isEmpty()) 478 return *this; 479 480 // Otherwise we need to create a new node, taking care to fold in unary 481 // twines. 482 Child NewLHS, NewRHS; 483 NewLHS.twine = this; 484 NewRHS.twine = &Suffix; 485 NodeKind NewLHSKind = TwineKind, NewRHSKind = TwineKind; 486 if (isUnary()) { 487 NewLHS = LHS; 488 NewLHSKind = getLHSKind(); 489 } 490 if (Suffix.isUnary()) { 491 NewRHS = Suffix.LHS; 492 NewRHSKind = Suffix.getLHSKind(); 493 } 494 495 return Twine(NewLHS, NewLHSKind, NewRHS, NewRHSKind); 496 } 497 498 inline Twine operator+(const Twine &LHS, const Twine &RHS) { 499 return LHS.concat(RHS); 500 } 501 502 /// Additional overload to guarantee simplified codegen; this is equivalent to 503 /// concat(). 504 505 inline Twine operator+(const char *LHS, const StringRef &RHS) { 506 return Twine(LHS, RHS); 507 } 508 509 /// Additional overload to guarantee simplified codegen; this is equivalent to 510 /// concat(). 511 512 inline Twine operator+(const StringRef &LHS, const char *RHS) { 513 return Twine(LHS, RHS); 514 } 515 516 inline raw_ostream &operator<<(raw_ostream &OS, const Twine &RHS) { 517 RHS.print(OS); 518 return OS; 519 } 520 521 /// @} 522 } 523 524 #endif 525