2 * Copyright 2016 Facebook, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
17 // @author Mark Rabkin (mrabkin@fb.com)
18 // @author Andrei Alexandrescu (andrei.alexandrescu@fb.com)
20 #ifndef FOLLY_RANGE_H_
21 #define FOLLY_RANGE_H_
23 #include <folly/Portability.h>
24 #include <folly/FBString.h>
25 #include <folly/SpookyHashV2.h>
28 #include <boost/operators.hpp>
32 #include <glog/logging.h>
36 #include <type_traits>
38 // libc++ doesn't provide this header, nor does msvc
39 #ifdef FOLLY_HAVE_BITS_CXXCONFIG_H
40 // This file appears in two locations: inside fbcode and in the
41 // libstdc++ source code (when embedding fbstring as std::string).
42 // To aid in this schizophrenic use, two macros are defined in
44 // _LIBSTDCXX_FBSTRING - Set inside libstdc++. This is useful to
45 // gate use inside fbcode v. libstdc++
46 #include <bits/c++config.h>
49 #include <folly/CpuId.h>
50 #include <folly/Traits.h>
51 #include <folly/Likely.h>
52 #include <folly/detail/RangeCommon.h>
53 #include <folly/detail/RangeSse42.h>
55 // Ignore shadowing warnings within this file, so includers can use -Wshadow.
56 #pragma GCC diagnostic push
57 #pragma GCC diagnostic ignored "-Wshadow"
61 template <class T> class Range;
64 * Finds the first occurrence of needle in haystack. The algorithm is on
65 * average faster than O(haystack.size() * needle.size()) but not as fast
66 * as Boyer-Moore. On the upside, it does not do any upfront
67 * preprocessing and does not allocate memory.
69 template <class T, class Comp = std::equal_to<typename Range<T>::value_type>>
70 inline size_t qfind(const Range<T> & haystack,
71 const Range<T> & needle,
75 * Finds the first occurrence of needle in haystack. The result is the
76 * offset reported to the beginning of haystack, or string::npos if
77 * needle wasn't found.
80 size_t qfind(const Range<T> & haystack,
81 const typename Range<T>::value_type& needle);
84 * Finds the last occurrence of needle in haystack. The result is the
85 * offset reported to the beginning of haystack, or string::npos if
86 * needle wasn't found.
89 size_t rfind(const Range<T> & haystack,
90 const typename Range<T>::value_type& needle);
94 * Finds the first occurrence of any element of needle in
95 * haystack. The algorithm is O(haystack.size() * needle.size()).
98 inline size_t qfind_first_of(const Range<T> & haystack,
99 const Range<T> & needle);
102 * Small internal helper - returns the value just before an iterator.
107 * For random-access iterators, the value before is simply i[-1].
109 template <class Iter>
110 typename std::enable_if<
111 std::is_same<typename std::iterator_traits<Iter>::iterator_category,
112 std::random_access_iterator_tag>::value,
113 typename std::iterator_traits<Iter>::reference>::type
114 value_before(Iter i) {
119 * For all other iterators, we need to use the decrement operator.
121 template <class Iter>
122 typename std::enable_if<
123 !std::is_same<typename std::iterator_traits<Iter>::iterator_category,
124 std::random_access_iterator_tag>::value,
125 typename std::iterator_traits<Iter>::reference>::type
126 value_before(Iter i) {
131 * Use IsCharPointer<T>::type to enable const char* or char*.
132 * Use IsCharPointer<T>::const_type to enable only const char*.
134 template <class T> struct IsCharPointer {};
137 struct IsCharPointer<char*> {
142 struct IsCharPointer<const char*> {
143 typedef int const_type;
147 } // namespace detail
150 * Range abstraction keeping a pair of iterators. We couldn't use
151 * boost's similar range abstraction because we need an API identical
152 * with the former StringPiece class, which is used by a lot of other
153 * code. This abstraction does fulfill the needs of boost's
154 * range-oriented algorithms though.
156 * (Keep memory lifetime in mind when using this class, since it
157 * doesn't manage the data it refers to - just like an iterator
160 template <class Iter>
161 class Range : private boost::totally_ordered<Range<Iter> > {
163 typedef std::size_t size_type;
164 typedef Iter iterator;
165 typedef Iter const_iterator;
166 typedef typename std::remove_reference<
167 typename std::iterator_traits<Iter>::reference>::type
169 typedef typename std::iterator_traits<Iter>::reference reference;
172 * For MutableStringPiece and MutableByteRange we define StringPiece
173 * and ByteRange as const_range_type (for everything else its just
174 * identity). We do that to enable operations such as find with
175 * args which are const.
177 typedef typename std::conditional<
178 std::is_same<Iter, char*>::value
179 || std::is_same<Iter, unsigned char*>::value,
180 Range<const value_type*>,
181 Range<Iter>>::type const_range_type;
183 typedef std::char_traits<typename std::remove_const<value_type>::type>
186 static const size_type npos;
188 // Works for all iterators
189 constexpr Range() : b_(), e_() {
192 constexpr Range(const Range&) = default;
193 constexpr Range(Range&&) = default;
196 // Works for all iterators
197 constexpr Range(Iter start, Iter end) : b_(start), e_(end) {
200 // Works only for random-access iterators
201 constexpr Range(Iter start, size_t size)
202 : b_(start), e_(start + size) { }
204 # if !__clang__ || __CLANG_PREREQ(3, 7) // Clang 3.6 crashes on this line
205 /* implicit */ Range(std::nullptr_t) = delete;
208 template <class T = Iter, typename detail::IsCharPointer<T>::type = 0>
209 constexpr /* implicit */ Range(Iter str)
210 : b_(str), e_(str + constexpr_strlen(str)) {}
212 template <class T = Iter, typename detail::IsCharPointer<T>::const_type = 0>
213 /* implicit */ Range(const std::string& str)
214 : b_(str.data()), e_(b_ + str.size()) {}
216 template <class T = Iter, typename detail::IsCharPointer<T>::const_type = 0>
217 Range(const std::string& str, std::string::size_type startFrom) {
218 if (UNLIKELY(startFrom > str.size())) {
219 throw std::out_of_range("index out of range");
221 b_ = str.data() + startFrom;
222 e_ = str.data() + str.size();
225 template <class T = Iter, typename detail::IsCharPointer<T>::const_type = 0>
226 Range(const std::string& str,
227 std::string::size_type startFrom,
228 std::string::size_type size) {
229 if (UNLIKELY(startFrom > str.size())) {
230 throw std::out_of_range("index out of range");
232 b_ = str.data() + startFrom;
233 if (str.size() - startFrom < size) {
234 e_ = str.data() + str.size();
240 Range(const Range& other,
242 size_type length = npos)
243 : Range(other.subpiece(first, length))
246 template <class T = Iter, typename detail::IsCharPointer<T>::const_type = 0>
247 /* implicit */ Range(const fbstring& str)
248 : b_(str.data()), e_(b_ + str.size()) { }
250 template <class T = Iter, typename detail::IsCharPointer<T>::const_type = 0>
251 Range(const fbstring& str, fbstring::size_type startFrom) {
252 if (UNLIKELY(startFrom > str.size())) {
253 throw std::out_of_range("index out of range");
255 b_ = str.data() + startFrom;
256 e_ = str.data() + str.size();
259 template <class T = Iter, typename detail::IsCharPointer<T>::const_type = 0>
260 Range(const fbstring& str, fbstring::size_type startFrom,
261 fbstring::size_type size) {
262 if (UNLIKELY(startFrom > str.size())) {
263 throw std::out_of_range("index out of range");
265 b_ = str.data() + startFrom;
266 if (str.size() - startFrom < size) {
267 e_ = str.data() + str.size();
273 // Allow implicit conversion from Range<const char*> (aka StringPiece) to
274 // Range<const unsigned char*> (aka ByteRange), as they're both frequently
275 // used to represent ranges of bytes. Allow explicit conversion in the other
277 template <class OtherIter, typename std::enable_if<
278 (std::is_same<Iter, const unsigned char*>::value &&
279 (std::is_same<OtherIter, const char*>::value ||
280 std::is_same<OtherIter, char*>::value)), int>::type = 0>
281 /* implicit */ Range(const Range<OtherIter>& other)
282 : b_(reinterpret_cast<const unsigned char*>(other.begin())),
283 e_(reinterpret_cast<const unsigned char*>(other.end())) {
286 template <class OtherIter, typename std::enable_if<
287 (std::is_same<Iter, unsigned char*>::value &&
288 std::is_same<OtherIter, char*>::value), int>::type = 0>
289 /* implicit */ Range(const Range<OtherIter>& other)
290 : b_(reinterpret_cast<unsigned char*>(other.begin())),
291 e_(reinterpret_cast<unsigned char*>(other.end())) {
294 template <class OtherIter, typename std::enable_if<
295 (std::is_same<Iter, const char*>::value &&
296 (std::is_same<OtherIter, const unsigned char*>::value ||
297 std::is_same<OtherIter, unsigned char*>::value)), int>::type = 0>
298 explicit Range(const Range<OtherIter>& other)
299 : b_(reinterpret_cast<const char*>(other.begin())),
300 e_(reinterpret_cast<const char*>(other.end())) {
303 template <class OtherIter, typename std::enable_if<
304 (std::is_same<Iter, char*>::value &&
305 std::is_same<OtherIter, unsigned char*>::value), int>::type = 0>
306 explicit Range(const Range<OtherIter>& other)
307 : b_(reinterpret_cast<char*>(other.begin())),
308 e_(reinterpret_cast<char*>(other.end())) {
311 // Allow implicit conversion from Range<From> to Range<To> if From is
312 // implicitly convertible to To.
313 template <class OtherIter, typename std::enable_if<
314 (!std::is_same<Iter, OtherIter>::value &&
315 std::is_convertible<OtherIter, Iter>::value), int>::type = 0>
316 constexpr /* implicit */ Range(const Range<OtherIter>& other)
321 // Allow explicit conversion from Range<From> to Range<To> if From is
322 // explicitly convertible to To.
323 template <class OtherIter, typename std::enable_if<
324 (!std::is_same<Iter, OtherIter>::value &&
325 !std::is_convertible<OtherIter, Iter>::value &&
326 std::is_constructible<Iter, const OtherIter&>::value), int>::type = 0>
327 constexpr explicit Range(const Range<OtherIter>& other)
332 Range& operator=(const Range& rhs) & = default;
333 Range& operator=(Range&& rhs) & = default;
340 void assign(Iter start, Iter end) {
345 void reset(Iter start, size_type size) {
350 // Works only for Range<const char*>
351 void reset(const std::string& str) {
352 reset(str.data(), str.size());
355 size_type size() const {
359 size_type walk_size() const {
360 return std::distance(b_, e_);
362 bool empty() const { return b_ == e_; }
363 Iter data() const { return b_; }
364 Iter start() const { return b_; }
365 Iter begin() const { return b_; }
366 Iter end() const { return e_; }
367 Iter cbegin() const { return b_; }
368 Iter cend() const { return e_; }
369 value_type& front() {
375 return detail::value_before(e_);
377 const value_type& front() const {
381 const value_type& back() const {
383 return detail::value_before(e_);
385 // Works only for Range<const char*> and Range<char*>
386 std::string str() const { return std::string(b_, size()); }
387 std::string toString() const { return str(); }
388 // Works only for Range<const char*> and Range<char*>
389 fbstring fbstr() const { return fbstring(b_, size()); }
390 fbstring toFbstring() const { return fbstr(); }
392 const_range_type castToConst() const {
393 return const_range_type(*this);
396 // Works only for Range<const char*> and Range<char*>
397 int compare(const const_range_type& o) const {
398 const size_type tsize = this->size();
399 const size_type osize = o.size();
400 const size_type msize = std::min(tsize, osize);
401 int r = traits_type::compare(data(), o.data(), msize);
402 if (r == 0 && tsize != osize) {
403 // We check the signed bit of the subtraction and bit shift it
404 // to produce either 0 or 2. The subtraction yields the
405 // comparison values of either -1 or 1.
406 r = (static_cast<int>(
407 (osize - tsize) >> (CHAR_BIT * sizeof(size_t) - 1)) << 1) - 1;
412 value_type& operator[](size_t i) {
413 DCHECK_GT(size(), i);
417 const value_type& operator[](size_t i) const {
418 DCHECK_GT(size(), i);
422 value_type& at(size_t i) {
423 if (i >= size()) throw std::out_of_range("index out of range");
427 const value_type& at(size_t i) const {
428 if (i >= size()) throw std::out_of_range("index out of range");
432 // Do NOT use this function, which was left behind for backwards
433 // compatibility. Use SpookyHashV2 instead -- it is faster, and produces
434 // a 64-bit hash, which means dramatically fewer collisions in large maps.
435 // (The above advice does not apply if you are targeting a 32-bit system.)
437 // Works only for Range<const char*> and Range<char*>
438 uint32_t hash() const {
439 // Taken from fbi/nstring.h:
440 // Quick and dirty bernstein hash...fine for short ascii strings
441 uint32_t hash = 5381;
442 for (size_t ix = 0; ix < size(); ix++) {
443 hash = ((hash << 5) + hash) + b_[ix];
448 void advance(size_type n) {
449 if (UNLIKELY(n > size())) {
450 throw std::out_of_range("index out of range");
455 void subtract(size_type n) {
456 if (UNLIKELY(n > size())) {
457 throw std::out_of_range("index out of range");
472 Range subpiece(size_type first, size_type length = npos) const {
473 if (UNLIKELY(first > size())) {
474 throw std::out_of_range("index out of range");
477 return Range(b_ + first, std::min(length, size() - first));
480 // string work-alike functions
481 size_type find(const_range_type str) const {
482 return qfind(castToConst(), str);
485 size_type find(const_range_type str, size_t pos) const {
486 if (pos > size()) return std::string::npos;
487 size_t ret = qfind(castToConst().subpiece(pos), str);
488 return ret == npos ? ret : ret + pos;
491 size_type find(Iter s, size_t pos, size_t n) const {
492 if (pos > size()) return std::string::npos;
493 auto forFinding = castToConst();
495 pos ? forFinding.subpiece(pos) : forFinding, const_range_type(s, n));
496 return ret == npos ? ret : ret + pos;
499 // Works only for Range<(const) (unsigned) char*> which have Range(Iter) ctor
500 size_type find(const Iter s) const {
501 return qfind(castToConst(), const_range_type(s));
504 // Works only for Range<(const) (unsigned) char*> which have Range(Iter) ctor
505 size_type find(const Iter s, size_t pos) const {
506 if (pos > size()) return std::string::npos;
507 size_type ret = qfind(castToConst().subpiece(pos), const_range_type(s));
508 return ret == npos ? ret : ret + pos;
511 size_type find(value_type c) const {
512 return qfind(castToConst(), c);
515 size_type rfind(value_type c) const {
516 return folly::rfind(castToConst(), c);
519 size_type find(value_type c, size_t pos) const {
520 if (pos > size()) return std::string::npos;
521 size_type ret = qfind(castToConst().subpiece(pos), c);
522 return ret == npos ? ret : ret + pos;
525 size_type find_first_of(const_range_type needles) const {
526 return qfind_first_of(castToConst(), needles);
529 size_type find_first_of(const_range_type needles, size_t pos) const {
530 if (pos > size()) return std::string::npos;
531 size_type ret = qfind_first_of(castToConst().subpiece(pos), needles);
532 return ret == npos ? ret : ret + pos;
535 // Works only for Range<(const) (unsigned) char*> which have Range(Iter) ctor
536 size_type find_first_of(Iter needles) const {
537 return find_first_of(const_range_type(needles));
540 // Works only for Range<(const) (unsigned) char*> which have Range(Iter) ctor
541 size_type find_first_of(Iter needles, size_t pos) const {
542 return find_first_of(const_range_type(needles), pos);
545 size_type find_first_of(Iter needles, size_t pos, size_t n) const {
546 return find_first_of(const_range_type(needles, n), pos);
549 size_type find_first_of(value_type c) const {
553 size_type find_first_of(value_type c, size_t pos) const {
558 * Determine whether the range contains the given subrange or item.
560 * Note: Call find() directly if the index is needed.
562 bool contains(const const_range_type& other) const {
563 return find(other) != std::string::npos;
566 bool contains(const value_type& other) const {
567 return find(other) != std::string::npos;
570 void swap(Range& rhs) {
571 std::swap(b_, rhs.b_);
572 std::swap(e_, rhs.e_);
576 * Does this Range start with another range?
578 bool startsWith(const const_range_type& other) const {
579 return size() >= other.size()
580 && castToConst().subpiece(0, other.size()) == other;
582 bool startsWith(value_type c) const {
583 return !empty() && front() == c;
587 * Does this Range end with another range?
589 bool endsWith(const const_range_type& other) const {
590 return size() >= other.size()
591 && castToConst().subpiece(size() - other.size()) == other;
593 bool endsWith(value_type c) const {
594 return !empty() && back() == c;
598 * Remove the items in [b, e), as long as this subrange is at the beginning
599 * or end of the Range.
601 * Required for boost::algorithm::trim()
603 void erase(Iter b, Iter e) {
606 } else if (e == e_) {
609 throw std::out_of_range("index out of range");
614 * Remove the given prefix and return true if the range starts with the given
615 * prefix; return false otherwise.
617 bool removePrefix(const const_range_type& prefix) {
618 return startsWith(prefix) && (b_ += prefix.size(), true);
620 bool removePrefix(value_type prefix) {
621 return startsWith(prefix) && (++b_, true);
625 * Remove the given suffix and return true if the range ends with the given
626 * suffix; return false otherwise.
628 bool removeSuffix(const const_range_type& suffix) {
629 return endsWith(suffix) && (e_ -= suffix.size(), true);
631 bool removeSuffix(value_type suffix) {
632 return endsWith(suffix) && (--e_, true);
636 * Replaces the content of the range, starting at position 'pos', with
637 * contents of 'replacement'. Entire 'replacement' must fit into the
638 * range. Returns false if 'replacements' does not fit. Example use:
640 * char in[] = "buffer";
641 * auto msp = MutablesStringPiece(input);
642 * EXPECT_TRUE(msp.replaceAt(2, "tt"));
643 * EXPECT_EQ(msp, "butter");
645 * // not enough space
646 * EXPECT_FALSE(msp.replace(msp.size() - 1, "rr"));
647 * EXPECT_EQ(msp, "butter"); // unchanged
649 bool replaceAt(size_t pos, const_range_type replacement) {
650 if (size() < pos + replacement.size()) {
654 std::copy(replacement.begin(), replacement.end(), begin() + pos);
660 * Replaces all occurences of 'source' with 'dest'. Returns number
661 * of replacements made. Source and dest have to have the same
662 * length. Throws if the lengths are different. If 'source' is a
663 * pattern that is overlapping with itself, we perform sequential
664 * replacement: "aaaaaaa".replaceAll("aa", "ba") --> "bababaa"
668 * char in[] = "buffer";
669 * auto msp = MutablesStringPiece(input);
670 * EXPECT_EQ(msp.replaceAll("ff","tt"), 1);
671 * EXPECT_EQ(msp, "butter");
673 size_t replaceAll(const_range_type source, const_range_type dest) {
674 if (source.size() != dest.size()) {
675 throw std::invalid_argument(
676 "replacement must have the same size as source");
684 size_t num_replaced = 0;
685 size_type found = std::string::npos;
686 while ((found = find(source, pos)) != std::string::npos) {
687 replaceAt(found, dest);
688 pos += source.size();
696 * Splits this `Range` `[b, e)` in the position `i` dictated by the next
697 * occurence of `delimiter`.
699 * Returns a new `Range` `[b, i)` and adjusts this range to start right after
700 * the delimiter's position. This range will be empty if the delimiter is not
701 * found. If called on an empty `Range`, both this and the returned `Range`
706 * folly::StringPiece s("sample string for split_next");
707 * auto p = s.split_step(' ');
709 * // prints "string for split_next"
717 * void tokenize(StringPiece s, char delimiter) {
718 * while (!s.empty()) {
719 * cout << s.split_step(delimiter);
723 * @author: Marcelo Juchem <marcelo@fb.com>
725 Range split_step(value_type delimiter) {
726 auto i = std::find(b_, e_, delimiter);
729 b_ = i == e_ ? e_ : std::next(i);
734 Range split_step(Range delimiter) {
735 auto i = find(delimiter);
736 Range result(b_, i == std::string::npos ? size() : i);
738 b_ = result.end() == e_ ? e_ : std::next(result.end(), delimiter.size());
744 * Convenience method that calls `split_step()` and passes the result to a
745 * functor, returning whatever the functor does. Any additional arguments
746 * `args` passed to this function are perfectly forwarded to the functor.
748 * Say you have a functor with this signature:
750 * Foo fn(Range r) { }
752 * `split_step()`'s return type will be `Foo`. It works just like:
754 * auto result = fn(myRange.split_step(' '));
756 * A functor returning `void` is also supported.
760 * void do_some_parsing(folly::StringPiece s) {
761 * auto version = s.split_step(' ', [&](folly::StringPiece x) {
763 * throw std::invalid_argument("empty string");
765 * return std::strtoull(x.begin(), x.end(), 16);
772 * void parse(folly::StringPiece s) {
773 * s.split_step(' ', parse_field, bar, 10);
774 * s.split_step('\t', parse_field, baz, 20);
776 * auto const kludge = [](folly::StringPiece x, int &out, int def) {
780 * parse_field(x, out, def);
784 * s.split_step('\t', kludge, gaz);
785 * s.split_step(' ', kludge, foo);
794 * static parse_field(folly::StringPiece s, int &out, int def) {
796 * out = folly::to<int>(s);
797 * } catch (std::exception const &) {
803 * @author: Marcelo Juchem <marcelo@fb.com>
805 template <typename TProcess, typename... Args>
806 auto split_step(value_type delimiter, TProcess &&process, Args &&...args)
807 -> decltype(process(std::declval<Range>(), std::forward<Args>(args)...))
808 { return process(split_step(delimiter), std::forward<Args>(args)...); }
810 template <typename TProcess, typename... Args>
811 auto split_step(Range delimiter, TProcess &&process, Args &&...args)
812 -> decltype(process(std::declval<Range>(), std::forward<Args>(args)...))
813 { return process(split_step(delimiter), std::forward<Args>(args)...); }
819 template <class Iter>
820 const typename Range<Iter>::size_type Range<Iter>::npos = std::string::npos;
823 void swap(Range<T>& lhs, Range<T>& rhs) {
828 * Create a range from two iterators, with type deduction.
830 template <class Iter>
831 Range<Iter> range(Iter first, Iter last) {
832 return Range<Iter>(first, last);
836 * Creates a range to reference the contents of a contiguous-storage container.
838 // Use pointers for types with '.data()' member
839 template <class Collection,
840 class T = typename std::remove_pointer<
841 decltype(std::declval<Collection>().data())>::type>
842 Range<T*> range(Collection&& v) {
843 return Range<T*>(v.data(), v.data() + v.size());
846 template <class T, size_t n>
847 Range<T*> range(T (&array)[n]) {
848 return Range<T*>(array, array + n);
851 typedef Range<const char*> StringPiece;
852 typedef Range<char*> MutableStringPiece;
853 typedef Range<const unsigned char*> ByteRange;
854 typedef Range<unsigned char*> MutableByteRange;
856 inline std::ostream& operator<<(std::ostream& os,
857 const StringPiece piece) {
858 os.write(piece.start(), piece.size());
862 inline std::ostream& operator<<(std::ostream& os,
863 const MutableStringPiece piece) {
864 os.write(piece.start(), piece.size());
869 * Templated comparison operators
873 inline bool operator==(const Range<T>& lhs, const Range<T>& rhs) {
874 return lhs.size() == rhs.size() && lhs.compare(rhs) == 0;
878 inline bool operator<(const Range<T>& lhs, const Range<T>& rhs) {
879 return lhs.compare(rhs) < 0;
883 * Specializations of comparison operators for StringPiece
888 template <class A, class B>
889 struct ComparableAsStringPiece {
892 (std::is_convertible<A, StringPiece>::value
893 && std::is_same<B, StringPiece>::value)
895 (std::is_convertible<B, StringPiece>::value
896 && std::is_same<A, StringPiece>::value)
900 } // namespace detail
903 * operator== through conversion for Range<const char*>
905 template <class T, class U>
907 std::enable_if<detail::ComparableAsStringPiece<T, U>::value, bool>::type
908 operator==(const T& lhs, const U& rhs) {
909 return StringPiece(lhs) == StringPiece(rhs);
913 * operator< through conversion for Range<const char*>
915 template <class T, class U>
917 std::enable_if<detail::ComparableAsStringPiece<T, U>::value, bool>::type
918 operator<(const T& lhs, const U& rhs) {
919 return StringPiece(lhs) < StringPiece(rhs);
923 * operator> through conversion for Range<const char*>
925 template <class T, class U>
927 std::enable_if<detail::ComparableAsStringPiece<T, U>::value, bool>::type
928 operator>(const T& lhs, const U& rhs) {
929 return StringPiece(lhs) > StringPiece(rhs);
933 * operator< through conversion for Range<const char*>
935 template <class T, class U>
937 std::enable_if<detail::ComparableAsStringPiece<T, U>::value, bool>::type
938 operator<=(const T& lhs, const U& rhs) {
939 return StringPiece(lhs) <= StringPiece(rhs);
943 * operator> through conversion for Range<const char*>
945 template <class T, class U>
947 std::enable_if<detail::ComparableAsStringPiece<T, U>::value, bool>::type
948 operator>=(const T& lhs, const U& rhs) {
949 return StringPiece(lhs) >= StringPiece(rhs);
952 // Do NOT use this, use SpookyHashV2 instead, see commment on hash() above.
953 struct StringPieceHash {
954 std::size_t operator()(const StringPiece str) const {
955 return static_cast<std::size_t>(str.hash());
960 * Finds substrings faster than brute force by borrowing from Boyer-Moore
962 template <class T, class Comp>
963 size_t qfind(const Range<T>& haystack,
964 const Range<T>& needle,
966 // Don't use std::search, use a Boyer-Moore-like trick by comparing
967 // the last characters first
968 auto const nsize = needle.size();
969 if (haystack.size() < nsize) {
970 return std::string::npos;
972 if (!nsize) return 0;
973 auto const nsize_1 = nsize - 1;
974 auto const lastNeedle = needle[nsize_1];
976 // Boyer-Moore skip value for the last char in the needle. Zero is
977 // not a valid value; skip will be computed the first time it's
979 std::string::size_type skip = 0;
981 auto i = haystack.begin();
982 auto iEnd = haystack.end() - nsize_1;
985 // Boyer-Moore: match the last element in the needle
986 while (!eq(i[nsize_1], lastNeedle)) {
989 return std::string::npos;
992 // Here we know that the last char matches
993 // Continue in pedestrian mode
994 for (size_t j = 0; ; ) {
996 if (!eq(i[j], needle[j])) {
997 // Not found, we can skip
998 // Compute the skip value lazily
1001 while (skip <= nsize_1 && !eq(needle[nsize_1 - skip], lastNeedle)) {
1008 // Check if done searching
1011 return i - haystack.begin();
1015 return std::string::npos;
1020 inline size_t qfind_first_byte_of(const StringPiece haystack,
1021 const StringPiece needles) {
1022 static auto const qfind_first_byte_of_fn =
1023 folly::CpuId().sse42() ? qfind_first_byte_of_sse42
1024 : qfind_first_byte_of_nosse;
1025 return qfind_first_byte_of_fn(haystack, needles);
1028 } // namespace detail
1030 template <class T, class Comp>
1031 size_t qfind_first_of(const Range<T> & haystack,
1032 const Range<T> & needles,
1034 auto ret = std::find_first_of(haystack.begin(), haystack.end(),
1035 needles.begin(), needles.end(),
1037 return ret == haystack.end() ? std::string::npos : ret - haystack.begin();
1040 struct AsciiCaseSensitive {
1041 bool operator()(char lhs, char rhs) const {
1047 * Check if two ascii characters are case insensitive equal.
1048 * The difference between the lower/upper case characters are the 6-th bit.
1049 * We also check they are alpha chars, in case of xor = 32.
1051 struct AsciiCaseInsensitive {
1052 bool operator()(char lhs, char rhs) const {
1054 if (k == 0) return true;
1055 if (k != 32) return false;
1057 return (k >= 'a' && k <= 'z');
1062 size_t qfind(const Range<T>& haystack,
1063 const typename Range<T>::value_type& needle) {
1064 auto pos = std::find(haystack.begin(), haystack.end(), needle);
1065 return pos == haystack.end() ? std::string::npos : pos - haystack.data();
1069 size_t rfind(const Range<T>& haystack,
1070 const typename Range<T>::value_type& needle) {
1071 for (auto i = haystack.size(); i-- > 0; ) {
1072 if (haystack[i] == needle) {
1076 return std::string::npos;
1079 // specialization for StringPiece
1081 inline size_t qfind(const Range<const char*>& haystack, const char& needle) {
1082 auto pos = static_cast<const char*>(
1083 ::memchr(haystack.data(), needle, haystack.size()));
1084 return pos == nullptr ? std::string::npos : pos - haystack.data();
1087 #if FOLLY_HAVE_MEMRCHR
1089 inline size_t rfind(const Range<const char*>& haystack, const char& needle) {
1090 auto pos = static_cast<const char*>(
1091 ::memrchr(haystack.data(), needle, haystack.size()));
1092 return pos == nullptr ? std::string::npos : pos - haystack.data();
1096 // specialization for ByteRange
1098 inline size_t qfind(const Range<const unsigned char*>& haystack,
1099 const unsigned char& needle) {
1100 auto pos = static_cast<const unsigned char*>(
1101 ::memchr(haystack.data(), needle, haystack.size()));
1102 return pos == nullptr ? std::string::npos : pos - haystack.data();
1105 #if FOLLY_HAVE_MEMRCHR
1107 inline size_t rfind(const Range<const unsigned char*>& haystack,
1108 const unsigned char& needle) {
1109 auto pos = static_cast<const unsigned char*>(
1110 ::memrchr(haystack.data(), needle, haystack.size()));
1111 return pos == nullptr ? std::string::npos : pos - haystack.data();
1116 size_t qfind_first_of(const Range<T>& haystack,
1117 const Range<T>& needles) {
1118 return qfind_first_of(haystack, needles, AsciiCaseSensitive());
1121 // specialization for StringPiece
1123 inline size_t qfind_first_of(const Range<const char*>& haystack,
1124 const Range<const char*>& needles) {
1125 return detail::qfind_first_byte_of(haystack, needles);
1128 // specialization for ByteRange
1130 inline size_t qfind_first_of(const Range<const unsigned char*>& haystack,
1131 const Range<const unsigned char*>& needles) {
1132 return detail::qfind_first_byte_of(StringPiece(haystack),
1133 StringPiece(needles));
1136 template<class Key, class Enable>
1140 struct hasher<folly::Range<T*>,
1141 typename std::enable_if<std::is_pod<T>::value, void>::type> {
1142 size_t operator()(folly::Range<T*> r) const {
1143 return hash::SpookyHashV2::Hash64(r.begin(), r.size() * sizeof(T), 0);
1147 } // !namespace folly
1149 #pragma GCC diagnostic pop
1151 FOLLY_ASSUME_FBVECTOR_COMPATIBLE_1(folly::Range);
1153 #endif // FOLLY_RANGE_H_