1 //===-- StringRef.cpp - Lightweight String References ---------------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 #include "llvm/ADT/StringRef.h"
11 #include "llvm/ADT/APInt.h"
12 #include "llvm/ADT/OwningPtr.h"
17 // MSVC emits references to this into the translation units which reference it.
19 const size_t StringRef::npos;
22 static char ascii_tolower(char x) {
23 if (x >= 'A' && x <= 'Z')
28 static char ascii_toupper(char x) {
29 if (x >= 'a' && x <= 'z')
34 static bool ascii_isdigit(char x) {
35 return x >= '0' && x <= '9';
38 /// compare_lower - Compare strings, ignoring case.
39 int StringRef::compare_lower(StringRef RHS) const {
40 for (size_t I = 0, E = min(Length, RHS.Length); I != E; ++I) {
41 unsigned char LHC = ascii_tolower(Data[I]);
42 unsigned char RHC = ascii_tolower(RHS.Data[I]);
44 return LHC < RHC ? -1 : 1;
47 if (Length == RHS.Length)
49 return Length < RHS.Length ? -1 : 1;
52 /// compare_numeric - Compare strings, handle embedded numbers.
53 int StringRef::compare_numeric(StringRef RHS) const {
54 for (size_t I = 0, E = min(Length, RHS.Length); I != E; ++I) {
55 // Check for sequences of digits.
56 if (ascii_isdigit(Data[I]) && ascii_isdigit(RHS.Data[I])) {
57 // The longer sequence of numbers is considered larger.
58 // This doesn't really handle prefixed zeros well.
60 for (J = I + 1; J != E + 1; ++J) {
61 bool ld = J < Length && ascii_isdigit(Data[J]);
62 bool rd = J < RHS.Length && ascii_isdigit(RHS.Data[J]);
68 // The two number sequences have the same length (J-I), just memcmp them.
69 if (int Res = compareMemory(Data + I, RHS.Data + I, J - I))
70 return Res < 0 ? -1 : 1;
71 // Identical number sequences, continue search after the numbers.
75 if (Data[I] != RHS.Data[I])
76 return (unsigned char)Data[I] < (unsigned char)RHS.Data[I] ? -1 : 1;
78 if (Length == RHS.Length)
80 return Length < RHS.Length ? -1 : 1;
83 // Compute the edit distance between the two given strings.
84 unsigned StringRef::edit_distance(llvm::StringRef Other,
85 bool AllowReplacements,
86 unsigned MaxEditDistance) {
87 // The algorithm implemented below is the "classic"
88 // dynamic-programming algorithm for computing the Levenshtein
89 // distance, which is described here:
91 // http://en.wikipedia.org/wiki/Levenshtein_distance
93 // Although the algorithm is typically described using an m x n
94 // array, only two rows are used at a time, so this implemenation
95 // just keeps two separate vectors for those two rows.
97 size_type n = Other.size();
99 const unsigned SmallBufferSize = 64;
100 unsigned SmallBuffer[SmallBufferSize];
101 llvm::OwningArrayPtr<unsigned> Allocated;
102 unsigned *previous = SmallBuffer;
103 if (2*(n + 1) > SmallBufferSize) {
104 previous = new unsigned [2*(n+1)];
105 Allocated.reset(previous);
107 unsigned *current = previous + (n + 1);
109 for (unsigned i = 0; i <= n; ++i)
112 for (size_type y = 1; y <= m; ++y) {
114 unsigned BestThisRow = current[0];
116 for (size_type x = 1; x <= n; ++x) {
117 if (AllowReplacements) {
118 current[x] = min(previous[x-1] + ((*this)[y-1] == Other[x-1]? 0u:1u),
119 min(current[x-1], previous[x])+1);
122 if ((*this)[y-1] == Other[x-1]) current[x] = previous[x-1];
123 else current[x] = min(current[x-1], previous[x]) + 1;
125 BestThisRow = min(BestThisRow, current[x]);
128 if (MaxEditDistance && BestThisRow > MaxEditDistance)
129 return MaxEditDistance + 1;
131 unsigned *tmp = current;
136 unsigned Result = previous[n];
140 //===----------------------------------------------------------------------===//
142 //===----------------------------------------------------------------------===//
144 std::string StringRef::lower() const {
145 std::string Result(size(), char());
146 for (size_type i = 0, e = size(); i != e; ++i) {
147 Result[i] = ascii_tolower(Data[i]);
152 std::string StringRef::upper() const {
153 std::string Result(size(), char());
154 for (size_type i = 0, e = size(); i != e; ++i) {
155 Result[i] = ascii_toupper(Data[i]);
160 //===----------------------------------------------------------------------===//
162 //===----------------------------------------------------------------------===//
165 /// find - Search for the first string \arg Str in the string.
167 /// \return - The index of the first occurrence of \arg Str, or npos if not
169 size_t StringRef::find(StringRef Str, size_t From) const {
170 size_t N = Str.size();
174 // For short haystacks or unsupported needles fall back to the naive algorithm
175 if (Length < 16 || N > 255 || N == 0) {
176 for (size_t e = Length - N + 1, i = min(From, e); i != e; ++i)
177 if (substr(i, N).equals(Str))
185 // Build the bad char heuristic table, with uint8_t to reduce cache thrashing.
186 uint8_t BadCharSkip[256];
187 std::memset(BadCharSkip, N, 256);
188 for (unsigned i = 0; i != N-1; ++i)
189 BadCharSkip[(uint8_t)Str[i]] = N-1-i;
191 unsigned Len = Length-From, Pos = From;
193 if (substr(Pos, N).equals(Str)) // See if this is the correct substring.
196 // Otherwise skip the appropriate number of bytes.
197 uint8_t Skip = BadCharSkip[(uint8_t)(*this)[Pos+N-1]];
205 /// rfind - Search for the last string \arg Str in the string.
207 /// \return - The index of the last occurrence of \arg Str, or npos if not
209 size_t StringRef::rfind(StringRef Str) const {
210 size_t N = Str.size();
213 for (size_t i = Length - N + 1, e = 0; i != e;) {
215 if (substr(i, N).equals(Str))
221 /// find_first_of - Find the first character in the string that is in \arg
222 /// Chars, or npos if not found.
224 /// Note: O(size() + Chars.size())
225 StringRef::size_type StringRef::find_first_of(StringRef Chars,
227 std::bitset<1 << CHAR_BIT> CharBits;
228 for (size_type i = 0; i != Chars.size(); ++i)
229 CharBits.set((unsigned char)Chars[i]);
231 for (size_type i = min(From, Length), e = Length; i != e; ++i)
232 if (CharBits.test((unsigned char)Data[i]))
237 /// find_first_not_of - Find the first character in the string that is not
238 /// \arg C or npos if not found.
239 StringRef::size_type StringRef::find_first_not_of(char C, size_t From) const {
240 for (size_type i = min(From, Length), e = Length; i != e; ++i)
246 /// find_first_not_of - Find the first character in the string that is not
247 /// in the string \arg Chars, or npos if not found.
249 /// Note: O(size() + Chars.size())
250 StringRef::size_type StringRef::find_first_not_of(StringRef Chars,
252 std::bitset<1 << CHAR_BIT> CharBits;
253 for (size_type i = 0; i != Chars.size(); ++i)
254 CharBits.set((unsigned char)Chars[i]);
256 for (size_type i = min(From, Length), e = Length; i != e; ++i)
257 if (!CharBits.test((unsigned char)Data[i]))
262 /// find_last_of - Find the last character in the string that is in \arg C,
263 /// or npos if not found.
265 /// Note: O(size() + Chars.size())
266 StringRef::size_type StringRef::find_last_of(StringRef Chars,
268 std::bitset<1 << CHAR_BIT> CharBits;
269 for (size_type i = 0; i != Chars.size(); ++i)
270 CharBits.set((unsigned char)Chars[i]);
272 for (size_type i = min(From, Length) - 1, e = -1; i != e; --i)
273 if (CharBits.test((unsigned char)Data[i]))
278 //===----------------------------------------------------------------------===//
279 // Helpful Algorithms
280 //===----------------------------------------------------------------------===//
282 /// count - Return the number of non-overlapped occurrences of \arg Str in
284 size_t StringRef::count(StringRef Str) const {
286 size_t N = Str.size();
289 for (size_t i = 0, e = Length - N + 1; i != e; ++i)
290 if (substr(i, N).equals(Str))
295 static unsigned GetAutoSenseRadix(StringRef &Str) {
296 if (Str.startswith("0x")) {
299 } else if (Str.startswith("0b")) {
302 } else if (Str.startswith("0")) {
310 /// GetAsUnsignedInteger - Workhorse method that converts a integer character
311 /// sequence of radix up to 36 to an unsigned long long value.
312 static bool GetAsUnsignedInteger(StringRef Str, unsigned Radix,
313 unsigned long long &Result) {
314 // Autosense radix if not specified.
316 Radix = GetAutoSenseRadix(Str);
318 // Empty strings (after the radix autosense) are invalid.
319 if (Str.empty()) return true;
321 // Parse all the bytes of the string given this radix. Watch for overflow.
323 while (!Str.empty()) {
325 if (Str[0] >= '0' && Str[0] <= '9')
326 CharVal = Str[0]-'0';
327 else if (Str[0] >= 'a' && Str[0] <= 'z')
328 CharVal = Str[0]-'a'+10;
329 else if (Str[0] >= 'A' && Str[0] <= 'Z')
330 CharVal = Str[0]-'A'+10;
334 // If the parsed value is larger than the integer radix, the string is
336 if (CharVal >= Radix)
339 // Add in this character.
340 unsigned long long PrevResult = Result;
341 Result = Result*Radix+CharVal;
343 // Check for overflow.
344 if (Result < PrevResult)
353 bool StringRef::getAsInteger(unsigned Radix, unsigned long long &Result) const {
354 return GetAsUnsignedInteger(*this, Radix, Result);
358 bool StringRef::getAsInteger(unsigned Radix, long long &Result) const {
359 unsigned long long ULLVal;
361 // Handle positive strings first.
362 if (empty() || front() != '-') {
363 if (GetAsUnsignedInteger(*this, Radix, ULLVal) ||
364 // Check for value so large it overflows a signed value.
365 (long long)ULLVal < 0)
371 // Get the positive part of the value.
372 if (GetAsUnsignedInteger(substr(1), Radix, ULLVal) ||
373 // Reject values so large they'd overflow as negative signed, but allow
374 // "-0". This negates the unsigned so that the negative isn't undefined
375 // on signed overflow.
376 (long long)-ULLVal > 0)
383 bool StringRef::getAsInteger(unsigned Radix, int &Result) const {
385 if (getAsInteger(Radix, Val) ||
392 bool StringRef::getAsInteger(unsigned Radix, unsigned &Result) const {
393 unsigned long long Val;
394 if (getAsInteger(Radix, Val) ||
395 (unsigned)Val != Val)
401 bool StringRef::getAsInteger(unsigned Radix, APInt &Result) const {
402 StringRef Str = *this;
404 // Autosense radix if not specified.
406 Radix = GetAutoSenseRadix(Str);
408 assert(Radix > 1 && Radix <= 36);
410 // Empty strings (after the radix autosense) are invalid.
411 if (Str.empty()) return true;
413 // Skip leading zeroes. This can be a significant improvement if
414 // it means we don't need > 64 bits.
415 while (!Str.empty() && Str.front() == '0')
418 // If it was nothing but zeroes....
420 Result = APInt(64, 0);
424 // (Over-)estimate the required number of bits.
425 unsigned Log2Radix = 0;
426 while ((1U << Log2Radix) < Radix) Log2Radix++;
427 bool IsPowerOf2Radix = ((1U << Log2Radix) == Radix);
429 unsigned BitWidth = Log2Radix * Str.size();
430 if (BitWidth < Result.getBitWidth())
431 BitWidth = Result.getBitWidth(); // don't shrink the result
433 Result = Result.zext(BitWidth);
435 APInt RadixAP, CharAP; // unused unless !IsPowerOf2Radix
436 if (!IsPowerOf2Radix) {
437 // These must have the same bit-width as Result.
438 RadixAP = APInt(BitWidth, Radix);
439 CharAP = APInt(BitWidth, 0);
442 // Parse all the bytes of the string given this radix.
444 while (!Str.empty()) {
446 if (Str[0] >= '0' && Str[0] <= '9')
447 CharVal = Str[0]-'0';
448 else if (Str[0] >= 'a' && Str[0] <= 'z')
449 CharVal = Str[0]-'a'+10;
450 else if (Str[0] >= 'A' && Str[0] <= 'Z')
451 CharVal = Str[0]-'A'+10;
455 // If the parsed value is larger than the integer radix, the string is
457 if (CharVal >= Radix)
460 // Add in this character.
461 if (IsPowerOf2Radix) {
462 Result <<= Log2Radix;