1 //===-- StringRef.cpp - Lightweight String References ---------------------===// 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 #include "llvm/ADT/StringRef.h" 11 #include "llvm/ADT/APInt.h" 12 #include "llvm/ADT/OwningPtr.h" 13 #include <bitset> 14 15 using namespace llvm; 16 17 // MSVC emits references to this into the translation units which reference it. 18 #ifndef _MSC_VER 19 const size_t StringRef::npos; 20 #endif 21 22 static char ascii_tolower(char x) { 23 if (x >= 'A' && x <= 'Z') 24 return x - 'A' + 'a'; 25 return x; 26 } 27 28 static bool ascii_isdigit(char x) { 29 return x >= '0' && x <= '9'; 30 } 31 32 /// compare_lower - Compare strings, ignoring case. 33 int StringRef::compare_lower(StringRef RHS) const { 34 for (size_t I = 0, E = min(Length, RHS.Length); I != E; ++I) { 35 unsigned char LHC = ascii_tolower(Data[I]); 36 unsigned char RHC = ascii_tolower(RHS.Data[I]); 37 if (LHC != RHC) 38 return LHC < RHC ? -1 : 1; 39 } 40 41 if (Length == RHS.Length) 42 return 0; 43 return Length < RHS.Length ? -1 : 1; 44 } 45 46 /// compare_numeric - Compare strings, handle embedded numbers. 47 int StringRef::compare_numeric(StringRef RHS) const { 48 for (size_t I = 0, E = min(Length, RHS.Length); I != E; ++I) { 49 // Check for sequences of digits. 50 if (ascii_isdigit(Data[I]) && ascii_isdigit(RHS.Data[I])) { 51 // The longer sequence of numbers is considered larger. 52 // This doesn't really handle prefixed zeros well. 53 size_t J; 54 for (J = I + 1; J != E + 1; ++J) { 55 bool ld = J < Length && ascii_isdigit(Data[J]); 56 bool rd = J < RHS.Length && ascii_isdigit(RHS.Data[J]); 57 if (ld != rd) 58 return rd ? -1 : 1; 59 if (!rd) 60 break; 61 } 62 // The two number sequences have the same length (J-I), just memcmp them. 63 if (int Res = compareMemory(Data + I, RHS.Data + I, J - I)) 64 return Res < 0 ? -1 : 1; 65 // Identical number sequences, continue search after the numbers. 66 I = J - 1; 67 continue; 68 } 69 if (Data[I] != RHS.Data[I]) 70 return (unsigned char)Data[I] < (unsigned char)RHS.Data[I] ? -1 : 1; 71 } 72 if (Length == RHS.Length) 73 return 0; 74 return Length < RHS.Length ? -1 : 1; 75 } 76 77 // Compute the edit distance between the two given strings. 78 unsigned StringRef::edit_distance(llvm::StringRef Other, 79 bool AllowReplacements, 80 unsigned MaxEditDistance) { 81 // The algorithm implemented below is the "classic" 82 // dynamic-programming algorithm for computing the Levenshtein 83 // distance, which is described here: 84 // 85 // http://en.wikipedia.org/wiki/Levenshtein_distance 86 // 87 // Although the algorithm is typically described using an m x n 88 // array, only two rows are used at a time, so this implemenation 89 // just keeps two separate vectors for those two rows. 90 size_type m = size(); 91 size_type n = Other.size(); 92 93 const unsigned SmallBufferSize = 64; 94 unsigned SmallBuffer[SmallBufferSize]; 95 llvm::OwningArrayPtr<unsigned> Allocated; 96 unsigned *previous = SmallBuffer; 97 if (2*(n + 1) > SmallBufferSize) { 98 previous = new unsigned [2*(n+1)]; 99 Allocated.reset(previous); 100 } 101 unsigned *current = previous + (n + 1); 102 103 for (unsigned i = 0; i <= n; ++i) 104 previous[i] = i; 105 106 for (size_type y = 1; y <= m; ++y) { 107 current[0] = y; 108 unsigned BestThisRow = current[0]; 109 110 for (size_type x = 1; x <= n; ++x) { 111 if (AllowReplacements) { 112 current[x] = min(previous[x-1] + ((*this)[y-1] == Other[x-1]? 0u:1u), 113 min(current[x-1], previous[x])+1); 114 } 115 else { 116 if ((*this)[y-1] == Other[x-1]) current[x] = previous[x-1]; 117 else current[x] = min(current[x-1], previous[x]) + 1; 118 } 119 BestThisRow = min(BestThisRow, current[x]); 120 } 121 122 if (MaxEditDistance && BestThisRow > MaxEditDistance) 123 return MaxEditDistance + 1; 124 125 unsigned *tmp = current; 126 current = previous; 127 previous = tmp; 128 } 129 130 unsigned Result = previous[n]; 131 return Result; 132 } 133 134 //===----------------------------------------------------------------------===// 135 // String Searching 136 //===----------------------------------------------------------------------===// 137 138 139 /// find - Search for the first string \arg Str in the string. 140 /// 141 /// \return - The index of the first occurrence of \arg Str, or npos if not 142 /// found. 143 size_t StringRef::find(StringRef Str, size_t From) const { 144 size_t N = Str.size(); 145 if (N > Length) 146 return npos; 147 148 // For short haystacks or unsupported needles fall back to the naive algorithm 149 if (Length < 16 || N > 255 || N == 0) { 150 for (size_t e = Length - N + 1, i = min(From, e); i != e; ++i) 151 if (substr(i, N).equals(Str)) 152 return i; 153 return npos; 154 } 155 156 if (From >= Length) 157 return npos; 158 159 // Build the bad char heuristic table, with uint8_t to reduce cache thrashing. 160 uint8_t BadCharSkip[256]; 161 std::memset(BadCharSkip, N, 256); 162 for (unsigned i = 0; i != N-1; ++i) 163 BadCharSkip[(uint8_t)Str[i]] = N-1-i; 164 165 unsigned Len = Length-From, Pos = From; 166 while (Len >= N) { 167 if (substr(Pos, N).equals(Str)) // See if this is the correct substring. 168 return Pos; 169 170 // Otherwise skip the appropriate number of bytes. 171 uint8_t Skip = BadCharSkip[(uint8_t)(*this)[Pos+N-1]]; 172 Len -= Skip; 173 Pos += Skip; 174 } 175 176 return npos; 177 } 178 179 /// rfind - Search for the last string \arg Str in the string. 180 /// 181 /// \return - The index of the last occurrence of \arg Str, or npos if not 182 /// found. 183 size_t StringRef::rfind(StringRef Str) const { 184 size_t N = Str.size(); 185 if (N > Length) 186 return npos; 187 for (size_t i = Length - N + 1, e = 0; i != e;) { 188 --i; 189 if (substr(i, N).equals(Str)) 190 return i; 191 } 192 return npos; 193 } 194 195 /// find_first_of - Find the first character in the string that is in \arg 196 /// Chars, or npos if not found. 197 /// 198 /// Note: O(size() + Chars.size()) 199 StringRef::size_type StringRef::find_first_of(StringRef Chars, 200 size_t From) const { 201 std::bitset<1 << CHAR_BIT> CharBits; 202 for (size_type i = 0; i != Chars.size(); ++i) 203 CharBits.set((unsigned char)Chars[i]); 204 205 for (size_type i = min(From, Length), e = Length; i != e; ++i) 206 if (CharBits.test((unsigned char)Data[i])) 207 return i; 208 return npos; 209 } 210 211 /// find_first_not_of - Find the first character in the string that is not 212 /// \arg C or npos if not found. 213 StringRef::size_type StringRef::find_first_not_of(char C, size_t From) const { 214 for (size_type i = min(From, Length), e = Length; i != e; ++i) 215 if (Data[i] != C) 216 return i; 217 return npos; 218 } 219 220 /// find_first_not_of - Find the first character in the string that is not 221 /// in the string \arg Chars, or npos if not found. 222 /// 223 /// Note: O(size() + Chars.size()) 224 StringRef::size_type StringRef::find_first_not_of(StringRef Chars, 225 size_t From) const { 226 std::bitset<1 << CHAR_BIT> CharBits; 227 for (size_type i = 0; i != Chars.size(); ++i) 228 CharBits.set((unsigned char)Chars[i]); 229 230 for (size_type i = min(From, Length), e = Length; i != e; ++i) 231 if (!CharBits.test((unsigned char)Data[i])) 232 return i; 233 return npos; 234 } 235 236 /// find_last_of - Find the last character in the string that is in \arg C, 237 /// or npos if not found. 238 /// 239 /// Note: O(size() + Chars.size()) 240 StringRef::size_type StringRef::find_last_of(StringRef Chars, 241 size_t From) const { 242 std::bitset<1 << CHAR_BIT> CharBits; 243 for (size_type i = 0; i != Chars.size(); ++i) 244 CharBits.set((unsigned char)Chars[i]); 245 246 for (size_type i = min(From, Length) - 1, e = -1; i != e; --i) 247 if (CharBits.test((unsigned char)Data[i])) 248 return i; 249 return npos; 250 } 251 252 //===----------------------------------------------------------------------===// 253 // Helpful Algorithms 254 //===----------------------------------------------------------------------===// 255 256 /// count - Return the number of non-overlapped occurrences of \arg Str in 257 /// the string. 258 size_t StringRef::count(StringRef Str) const { 259 size_t Count = 0; 260 size_t N = Str.size(); 261 if (N > Length) 262 return 0; 263 for (size_t i = 0, e = Length - N + 1; i != e; ++i) 264 if (substr(i, N).equals(Str)) 265 ++Count; 266 return Count; 267 } 268 269 static unsigned GetAutoSenseRadix(StringRef &Str) { 270 if (Str.startswith("0x")) { 271 Str = Str.substr(2); 272 return 16; 273 } else if (Str.startswith("0b")) { 274 Str = Str.substr(2); 275 return 2; 276 } else if (Str.startswith("0")) { 277 return 8; 278 } else { 279 return 10; 280 } 281 } 282 283 284 /// GetAsUnsignedInteger - Workhorse method that converts a integer character 285 /// sequence of radix up to 36 to an unsigned long long value. 286 static bool GetAsUnsignedInteger(StringRef Str, unsigned Radix, 287 unsigned long long &Result) { 288 // Autosense radix if not specified. 289 if (Radix == 0) 290 Radix = GetAutoSenseRadix(Str); 291 292 // Empty strings (after the radix autosense) are invalid. 293 if (Str.empty()) return true; 294 295 // Parse all the bytes of the string given this radix. Watch for overflow. 296 Result = 0; 297 while (!Str.empty()) { 298 unsigned CharVal; 299 if (Str[0] >= '0' && Str[0] <= '9') 300 CharVal = Str[0]-'0'; 301 else if (Str[0] >= 'a' && Str[0] <= 'z') 302 CharVal = Str[0]-'a'+10; 303 else if (Str[0] >= 'A' && Str[0] <= 'Z') 304 CharVal = Str[0]-'A'+10; 305 else 306 return true; 307 308 // If the parsed value is larger than the integer radix, the string is 309 // invalid. 310 if (CharVal >= Radix) 311 return true; 312 313 // Add in this character. 314 unsigned long long PrevResult = Result; 315 Result = Result*Radix+CharVal; 316 317 // Check for overflow. 318 if (Result < PrevResult) 319 return true; 320 321 Str = Str.substr(1); 322 } 323 324 return false; 325 } 326 327 bool StringRef::getAsInteger(unsigned Radix, unsigned long long &Result) const { 328 return GetAsUnsignedInteger(*this, Radix, Result); 329 } 330 331 332 bool StringRef::getAsInteger(unsigned Radix, long long &Result) const { 333 unsigned long long ULLVal; 334 335 // Handle positive strings first. 336 if (empty() || front() != '-') { 337 if (GetAsUnsignedInteger(*this, Radix, ULLVal) || 338 // Check for value so large it overflows a signed value. 339 (long long)ULLVal < 0) 340 return true; 341 Result = ULLVal; 342 return false; 343 } 344 345 // Get the positive part of the value. 346 if (GetAsUnsignedInteger(substr(1), Radix, ULLVal) || 347 // Reject values so large they'd overflow as negative signed, but allow 348 // "-0". This negates the unsigned so that the negative isn't undefined 349 // on signed overflow. 350 (long long)-ULLVal > 0) 351 return true; 352 353 Result = -ULLVal; 354 return false; 355 } 356 357 bool StringRef::getAsInteger(unsigned Radix, int &Result) const { 358 long long Val; 359 if (getAsInteger(Radix, Val) || 360 (int)Val != Val) 361 return true; 362 Result = Val; 363 return false; 364 } 365 366 bool StringRef::getAsInteger(unsigned Radix, unsigned &Result) const { 367 unsigned long long Val; 368 if (getAsInteger(Radix, Val) || 369 (unsigned)Val != Val) 370 return true; 371 Result = Val; 372 return false; 373 } 374 375 bool StringRef::getAsInteger(unsigned Radix, APInt &Result) const { 376 StringRef Str = *this; 377 378 // Autosense radix if not specified. 379 if (Radix == 0) 380 Radix = GetAutoSenseRadix(Str); 381 382 assert(Radix > 1 && Radix <= 36); 383 384 // Empty strings (after the radix autosense) are invalid. 385 if (Str.empty()) return true; 386 387 // Skip leading zeroes. This can be a significant improvement if 388 // it means we don't need > 64 bits. 389 while (!Str.empty() && Str.front() == '0') 390 Str = Str.substr(1); 391 392 // If it was nothing but zeroes.... 393 if (Str.empty()) { 394 Result = APInt(64, 0); 395 return false; 396 } 397 398 // (Over-)estimate the required number of bits. 399 unsigned Log2Radix = 0; 400 while ((1U << Log2Radix) < Radix) Log2Radix++; 401 bool IsPowerOf2Radix = ((1U << Log2Radix) == Radix); 402 403 unsigned BitWidth = Log2Radix * Str.size(); 404 if (BitWidth < Result.getBitWidth()) 405 BitWidth = Result.getBitWidth(); // don't shrink the result 406 else 407 Result = Result.zext(BitWidth); 408 409 APInt RadixAP, CharAP; // unused unless !IsPowerOf2Radix 410 if (!IsPowerOf2Radix) { 411 // These must have the same bit-width as Result. 412 RadixAP = APInt(BitWidth, Radix); 413 CharAP = APInt(BitWidth, 0); 414 } 415 416 // Parse all the bytes of the string given this radix. 417 Result = 0; 418 while (!Str.empty()) { 419 unsigned CharVal; 420 if (Str[0] >= '0' && Str[0] <= '9') 421 CharVal = Str[0]-'0'; 422 else if (Str[0] >= 'a' && Str[0] <= 'z') 423 CharVal = Str[0]-'a'+10; 424 else if (Str[0] >= 'A' && Str[0] <= 'Z') 425 CharVal = Str[0]-'A'+10; 426 else 427 return true; 428 429 // If the parsed value is larger than the integer radix, the string is 430 // invalid. 431 if (CharVal >= Radix) 432 return true; 433 434 // Add in this character. 435 if (IsPowerOf2Radix) { 436 Result <<= Log2Radix; 437 Result |= CharVal; 438 } else { 439 Result *= RadixAP; 440 CharAP = CharVal; 441 Result += CharAP; 442 } 443 444 Str = Str.substr(1); 445 } 446 447 return false; 448 } 449