2 * Copyright 2017 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)
22 #include <folly/FBString.h>
23 #include <folly/Portability.h>
24 #include <folly/SpookyHashV2.h>
25 #include <folly/portability/BitsFunctexcept.h>
26 #include <folly/portability/Constexpr.h>
27 #include <folly/portability/String.h>
29 #include <boost/operators.hpp>
30 #include <glog/logging.h>
39 #include <type_traits>
41 // libc++ doesn't provide this header, nor does msvc
42 #ifdef FOLLY_HAVE_BITS_CXXCONFIG_H
43 // This file appears in two locations: inside fbcode and in the
44 // libstdc++ source code (when embedding fbstring as std::string).
45 // To aid in this schizophrenic use, two macros are defined in
47 // _LIBSTDCXX_FBSTRING - Set inside libstdc++. This is useful to
48 // gate use inside fbcode v. libstdc++
49 #include <bits/c++config.h>
52 #include <folly/CpuId.h>
53 #include <folly/Traits.h>
54 #include <folly/Likely.h>
55 #include <folly/detail/RangeCommon.h>
56 #include <folly/detail/RangeSse42.h>
58 // Ignore shadowing warnings within this file, so includers can use -Wshadow.
59 #pragma GCC diagnostic push
60 #pragma GCC diagnostic ignored "-Wshadow"
64 template <class T> class Range;
67 * Finds the first occurrence of needle in haystack. The algorithm is on
68 * average faster than O(haystack.size() * needle.size()) but not as fast
69 * as Boyer-Moore. On the upside, it does not do any upfront
70 * preprocessing and does not allocate memory.
72 template <class T, class Comp = std::equal_to<typename Range<T>::value_type>>
73 inline size_t qfind(const Range<T> & haystack,
74 const Range<T> & needle,
78 * Finds the first occurrence of needle in haystack. The result is the
79 * offset reported to the beginning of haystack, or string::npos if
80 * needle wasn't found.
83 size_t qfind(const Range<T> & haystack,
84 const typename Range<T>::value_type& needle);
87 * Finds the last occurrence of needle in haystack. The result is the
88 * offset reported to the beginning of haystack, or string::npos if
89 * needle wasn't found.
92 size_t rfind(const Range<T> & haystack,
93 const typename Range<T>::value_type& needle);
97 * Finds the first occurrence of any element of needle in
98 * haystack. The algorithm is O(haystack.size() * needle.size()).
101 inline size_t qfind_first_of(const Range<T> & haystack,
102 const Range<T> & needle);
105 * Small internal helper - returns the value just before an iterator.
110 * For random-access iterators, the value before is simply i[-1].
112 template <class Iter>
113 typename std::enable_if<
114 std::is_same<typename std::iterator_traits<Iter>::iterator_category,
115 std::random_access_iterator_tag>::value,
116 typename std::iterator_traits<Iter>::reference>::type
117 value_before(Iter i) {
122 * For all other iterators, we need to use the decrement operator.
124 template <class Iter>
125 typename std::enable_if<
126 !std::is_same<typename std::iterator_traits<Iter>::iterator_category,
127 std::random_access_iterator_tag>::value,
128 typename std::iterator_traits<Iter>::reference>::type
129 value_before(Iter i) {
134 * Use IsCharPointer<T>::type to enable const char* or char*.
135 * Use IsCharPointer<T>::const_type to enable only const char*.
137 template <class T> struct IsCharPointer {};
140 struct IsCharPointer<char*> {
145 struct IsCharPointer<const char*> {
146 typedef int const_type;
150 } // namespace detail
153 * Range abstraction keeping a pair of iterators. We couldn't use
154 * boost's similar range abstraction because we need an API identical
155 * with the former StringPiece class, which is used by a lot of other
156 * code. This abstraction does fulfill the needs of boost's
157 * range-oriented algorithms though.
159 * (Keep memory lifetime in mind when using this class, since it
160 * doesn't manage the data it refers to - just like an iterator
163 template <class Iter>
164 class Range : private boost::totally_ordered<Range<Iter> > {
166 typedef std::size_t size_type;
167 typedef Iter iterator;
168 typedef Iter const_iterator;
169 typedef typename std::remove_reference<
170 typename std::iterator_traits<Iter>::reference>::type
172 using difference_type = typename std::iterator_traits<Iter>::difference_type;
173 typedef typename std::iterator_traits<Iter>::reference reference;
176 * For MutableStringPiece and MutableByteRange we define StringPiece
177 * and ByteRange as const_range_type (for everything else its just
178 * identity). We do that to enable operations such as find with
179 * args which are const.
181 typedef typename std::conditional<
182 std::is_same<Iter, char*>::value
183 || std::is_same<Iter, unsigned char*>::value,
184 Range<const value_type*>,
185 Range<Iter>>::type const_range_type;
187 typedef std::char_traits<typename std::remove_const<value_type>::type>
190 static const size_type npos;
192 // Works for all iterators
193 constexpr Range() : b_(), e_() {
196 constexpr Range(const Range&) = default;
197 constexpr Range(Range&&) = default;
200 // Works for all iterators
201 constexpr Range(Iter start, Iter end) : b_(start), e_(end) {
204 // Works only for random-access iterators
205 constexpr Range(Iter start, size_t size)
206 : b_(start), e_(start + size) { }
208 /* implicit */ Range(std::nullptr_t) = delete;
210 template <class T = Iter, typename detail::IsCharPointer<T>::type = 0>
211 constexpr /* implicit */ Range(Iter str)
212 : b_(str), e_(str + constexpr_strlen(str)) {}
214 template <class T = Iter, typename detail::IsCharPointer<T>::const_type = 0>
215 /* implicit */ Range(const std::string& str)
216 : b_(str.data()), e_(b_ + str.size()) {}
218 template <class T = Iter, typename detail::IsCharPointer<T>::const_type = 0>
219 Range(const std::string& str, std::string::size_type startFrom) {
220 if (UNLIKELY(startFrom > str.size())) {
221 std::__throw_out_of_range("index out of range");
223 b_ = str.data() + startFrom;
224 e_ = str.data() + str.size();
227 template <class T = Iter, typename detail::IsCharPointer<T>::const_type = 0>
228 Range(const std::string& str,
229 std::string::size_type startFrom,
230 std::string::size_type size) {
231 if (UNLIKELY(startFrom > str.size())) {
232 std::__throw_out_of_range("index out of range");
234 b_ = str.data() + startFrom;
235 if (str.size() - startFrom < size) {
236 e_ = str.data() + str.size();
242 Range(const Range& other,
244 size_type length = npos)
245 : Range(other.subpiece(first, length))
248 template <class T = Iter, typename detail::IsCharPointer<T>::const_type = 0>
249 /* implicit */ Range(const fbstring& str)
250 : b_(str.data()), e_(b_ + str.size()) { }
252 template <class T = Iter, typename detail::IsCharPointer<T>::const_type = 0>
253 Range(const fbstring& str, fbstring::size_type startFrom) {
254 if (UNLIKELY(startFrom > str.size())) {
255 std::__throw_out_of_range("index out of range");
257 b_ = str.data() + startFrom;
258 e_ = str.data() + str.size();
261 template <class T = Iter, typename detail::IsCharPointer<T>::const_type = 0>
262 Range(const fbstring& str, fbstring::size_type startFrom,
263 fbstring::size_type size) {
264 if (UNLIKELY(startFrom > str.size())) {
265 std::__throw_out_of_range("index out of range");
267 b_ = str.data() + startFrom;
268 if (str.size() - startFrom < size) {
269 e_ = str.data() + str.size();
275 // Allow implicit conversion from Range<const char*> (aka StringPiece) to
276 // Range<const unsigned char*> (aka ByteRange), as they're both frequently
277 // used to represent ranges of bytes. Allow explicit conversion in the other
279 template <class OtherIter, typename std::enable_if<
280 (std::is_same<Iter, const unsigned char*>::value &&
281 (std::is_same<OtherIter, const char*>::value ||
282 std::is_same<OtherIter, char*>::value)), int>::type = 0>
283 /* implicit */ Range(const Range<OtherIter>& other)
284 : b_(reinterpret_cast<const unsigned char*>(other.begin())),
285 e_(reinterpret_cast<const unsigned char*>(other.end())) {
288 template <class OtherIter, typename std::enable_if<
289 (std::is_same<Iter, unsigned char*>::value &&
290 std::is_same<OtherIter, char*>::value), int>::type = 0>
291 /* implicit */ Range(const Range<OtherIter>& other)
292 : b_(reinterpret_cast<unsigned char*>(other.begin())),
293 e_(reinterpret_cast<unsigned char*>(other.end())) {
296 template <class OtherIter, typename std::enable_if<
297 (std::is_same<Iter, const char*>::value &&
298 (std::is_same<OtherIter, const unsigned char*>::value ||
299 std::is_same<OtherIter, unsigned char*>::value)), int>::type = 0>
300 explicit Range(const Range<OtherIter>& other)
301 : b_(reinterpret_cast<const char*>(other.begin())),
302 e_(reinterpret_cast<const char*>(other.end())) {
305 template <class OtherIter, typename std::enable_if<
306 (std::is_same<Iter, char*>::value &&
307 std::is_same<OtherIter, unsigned char*>::value), int>::type = 0>
308 explicit Range(const Range<OtherIter>& other)
309 : b_(reinterpret_cast<char*>(other.begin())),
310 e_(reinterpret_cast<char*>(other.end())) {
313 // Allow implicit conversion from Range<From> to Range<To> if From is
314 // implicitly convertible to To.
315 template <class OtherIter, typename std::enable_if<
316 (!std::is_same<Iter, OtherIter>::value &&
317 std::is_convertible<OtherIter, Iter>::value), int>::type = 0>
318 constexpr /* implicit */ Range(const Range<OtherIter>& other)
323 // Allow explicit conversion from Range<From> to Range<To> if From is
324 // explicitly convertible to To.
325 template <class OtherIter, typename std::enable_if<
326 (!std::is_same<Iter, OtherIter>::value &&
327 !std::is_convertible<OtherIter, Iter>::value &&
328 std::is_constructible<Iter, const OtherIter&>::value), int>::type = 0>
329 constexpr explicit Range(const Range<OtherIter>& other)
335 * Allow explicit construction of Range() from a std::array of a
338 * For instance, this allows constructing StringPiece from a
339 * std::array<char, N> or a std::array<const char, N>
344 typename = typename std::enable_if<
345 std::is_convertible<const T*, Iter>::value>::type>
346 constexpr explicit Range(const std::array<T, N>& array)
347 : b_{array.empty() ? nullptr : &array.at(0)},
348 e_{array.empty() ? nullptr : &array.at(0) + N} {}
353 typename std::enable_if<std::is_convertible<T*, Iter>::value>::type>
354 constexpr explicit Range(std::array<T, N>& array)
355 : b_{array.empty() ? nullptr : &array.at(0)},
356 e_{array.empty() ? nullptr : &array.at(0) + N} {}
358 Range& operator=(const Range& rhs) & = default;
359 Range& operator=(Range&& rhs) & = default;
366 void assign(Iter start, Iter end) {
371 void reset(Iter start, size_type size) {
376 // Works only for Range<const char*>
377 void reset(const std::string& str) {
378 reset(str.data(), str.size());
381 constexpr size_type size() const {
382 // It would be nice to assert(b_ <= e_) here. This can be achieved even
383 // in a C++11 compatible constexpr function:
384 // http://ericniebler.com/2014/09/27/assert-and-constexpr-in-cxx11/
385 // Unfortunately current gcc versions have a bug causing it to reject
386 // this check in a constexpr function:
387 // https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71448
388 return size_type(e_ - b_);
390 constexpr size_type walk_size() const {
391 return size_type(std::distance(b_, e_));
393 constexpr bool empty() const {
396 constexpr Iter data() const {
399 constexpr Iter start() const {
402 constexpr Iter begin() const {
405 constexpr Iter end() const {
408 constexpr Iter cbegin() const {
411 constexpr Iter cend() const {
414 value_type& front() {
420 return detail::value_before(e_);
422 const value_type& front() const {
426 const value_type& back() const {
428 return detail::value_before(e_);
430 // Works only for Range<const char*> and Range<char*>
431 std::string str() const { return std::string(b_, size()); }
432 std::string toString() const { return str(); }
433 // Works only for Range<const char*> and Range<char*>
434 fbstring fbstr() const { return fbstring(b_, size()); }
435 fbstring toFbstring() const { return fbstr(); }
437 const_range_type castToConst() const {
438 return const_range_type(*this);
441 // Works only for Range<const char*> and Range<char*>
442 int compare(const const_range_type& o) const {
443 const size_type tsize = this->size();
444 const size_type osize = o.size();
445 const size_type msize = std::min(tsize, osize);
446 int r = traits_type::compare(data(), o.data(), msize);
447 if (r == 0 && tsize != osize) {
448 // We check the signed bit of the subtraction and bit shift it
449 // to produce either 0 or 2. The subtraction yields the
450 // comparison values of either -1 or 1.
451 r = (static_cast<int>(
452 (osize - tsize) >> (CHAR_BIT * sizeof(size_t) - 1)) << 1) - 1;
457 value_type& operator[](size_t i) {
458 DCHECK_GT(size(), i);
462 const value_type& operator[](size_t i) const {
463 DCHECK_GT(size(), i);
467 value_type& at(size_t i) {
468 if (i >= size()) std::__throw_out_of_range("index out of range");
472 const value_type& at(size_t i) const {
473 if (i >= size()) std::__throw_out_of_range("index out of range");
477 // Do NOT use this function, which was left behind for backwards
478 // compatibility. Use SpookyHashV2 instead -- it is faster, and produces
479 // a 64-bit hash, which means dramatically fewer collisions in large maps.
480 // (The above advice does not apply if you are targeting a 32-bit system.)
482 // Works only for Range<const char*> and Range<char*>
485 // ** WANT TO GET RID OF THIS LINT? **
487 // A) Use a better hash function (*cough*folly::Hash*cough*), but
488 // only if you don't serialize data in a format that depends on
489 // this formula (ie the writer and reader assume this exact hash
490 // function is used).
492 // B) If you have to use this exact function then make your own hasher
493 // object and copy the body over (see thrift example: D3972362).
494 // https://github.com/facebook/fbthrift/commit/f8ed502e24ab4a32a9d5f266580
495 FOLLY_DEPRECATED("Replace with folly::Hash if the hash is not serialized")
496 uint32_t hash() const {
497 // Taken from fbi/nstring.h:
498 // Quick and dirty bernstein hash...fine for short ascii strings
499 uint32_t hash = 5381;
500 for (size_t ix = 0; ix < size(); ix++) {
501 hash = ((hash << 5) + hash) + b_[ix];
506 void advance(size_type n) {
507 if (UNLIKELY(n > size())) {
508 std::__throw_out_of_range("index out of range");
513 void subtract(size_type n) {
514 if (UNLIKELY(n > size())) {
515 std::__throw_out_of_range("index out of range");
520 Range subpiece(size_type first, size_type length = npos) const {
521 if (UNLIKELY(first > size())) {
522 std::__throw_out_of_range("index out of range");
525 return Range(b_ + first, std::min(length, size() - first));
528 // unchecked versions
529 void uncheckedAdvance(size_type n) {
530 DCHECK_LE(n, size());
534 void uncheckedSubtract(size_type n) {
535 DCHECK_LE(n, size());
539 Range uncheckedSubpiece(size_type first, size_type length = npos) const {
540 DCHECK_LE(first, size());
541 return Range(b_ + first, std::min(length, size() - first));
554 // string work-alike functions
555 size_type find(const_range_type str) const {
556 return qfind(castToConst(), str);
559 size_type find(const_range_type str, size_t pos) const {
560 if (pos > size()) return std::string::npos;
561 size_t ret = qfind(castToConst().subpiece(pos), str);
562 return ret == npos ? ret : ret + pos;
565 size_type find(Iter s, size_t pos, size_t n) const {
566 if (pos > size()) return std::string::npos;
567 auto forFinding = castToConst();
569 pos ? forFinding.subpiece(pos) : forFinding, const_range_type(s, n));
570 return ret == npos ? ret : ret + pos;
573 // Works only for Range<(const) (unsigned) char*> which have Range(Iter) ctor
574 size_type find(const Iter s) const {
575 return qfind(castToConst(), const_range_type(s));
578 // Works only for Range<(const) (unsigned) char*> which have Range(Iter) ctor
579 size_type find(const Iter s, size_t pos) const {
580 if (pos > size()) return std::string::npos;
581 size_type ret = qfind(castToConst().subpiece(pos), const_range_type(s));
582 return ret == npos ? ret : ret + pos;
585 size_type find(value_type c) const {
586 return qfind(castToConst(), c);
589 size_type rfind(value_type c) const {
590 return folly::rfind(castToConst(), c);
593 size_type find(value_type c, size_t pos) const {
594 if (pos > size()) return std::string::npos;
595 size_type ret = qfind(castToConst().subpiece(pos), c);
596 return ret == npos ? ret : ret + pos;
599 size_type find_first_of(const_range_type needles) const {
600 return qfind_first_of(castToConst(), needles);
603 size_type find_first_of(const_range_type needles, size_t pos) const {
604 if (pos > size()) return std::string::npos;
605 size_type ret = qfind_first_of(castToConst().subpiece(pos), needles);
606 return ret == npos ? ret : ret + pos;
609 // Works only for Range<(const) (unsigned) char*> which have Range(Iter) ctor
610 size_type find_first_of(Iter needles) const {
611 return find_first_of(const_range_type(needles));
614 // Works only for Range<(const) (unsigned) char*> which have Range(Iter) ctor
615 size_type find_first_of(Iter needles, size_t pos) const {
616 return find_first_of(const_range_type(needles), pos);
619 size_type find_first_of(Iter needles, size_t pos, size_t n) const {
620 return find_first_of(const_range_type(needles, n), pos);
623 size_type find_first_of(value_type c) const {
627 size_type find_first_of(value_type c, size_t pos) const {
632 * Determine whether the range contains the given subrange or item.
634 * Note: Call find() directly if the index is needed.
636 bool contains(const const_range_type& other) const {
637 return find(other) != std::string::npos;
640 bool contains(const value_type& other) const {
641 return find(other) != std::string::npos;
644 void swap(Range& rhs) {
645 std::swap(b_, rhs.b_);
646 std::swap(e_, rhs.e_);
650 * Does this Range start with another range?
652 bool startsWith(const const_range_type& other) const {
653 return size() >= other.size()
654 && castToConst().subpiece(0, other.size()) == other;
656 bool startsWith(value_type c) const {
657 return !empty() && front() == c;
660 template <class Comp>
661 bool startsWith(const const_range_type& other, Comp&& eq) const {
662 if (size() < other.size()) {
665 auto const trunc = subpiece(0, other.size());
667 trunc.begin(), trunc.end(), other.begin(), std::forward<Comp>(eq));
671 * Does this Range end with another range?
673 bool endsWith(const const_range_type& other) const {
674 return size() >= other.size()
675 && castToConst().subpiece(size() - other.size()) == other;
677 bool endsWith(value_type c) const {
678 return !empty() && back() == c;
681 template <class Comp>
682 bool endsWith(const const_range_type& other, Comp&& eq) const {
683 if (size() < other.size()) {
686 auto const trunc = subpiece(size() - other.size());
688 trunc.begin(), trunc.end(), other.begin(), std::forward<Comp>(eq));
692 * Remove the items in [b, e), as long as this subrange is at the beginning
693 * or end of the Range.
695 * Required for boost::algorithm::trim()
697 void erase(Iter b, Iter e) {
700 } else if (e == e_) {
703 std::__throw_out_of_range("index out of range");
708 * Remove the given prefix and return true if the range starts with the given
709 * prefix; return false otherwise.
711 bool removePrefix(const const_range_type& prefix) {
712 return startsWith(prefix) && (b_ += prefix.size(), true);
714 bool removePrefix(value_type prefix) {
715 return startsWith(prefix) && (++b_, true);
719 * Remove the given suffix and return true if the range ends with the given
720 * suffix; return false otherwise.
722 bool removeSuffix(const const_range_type& suffix) {
723 return endsWith(suffix) && (e_ -= suffix.size(), true);
725 bool removeSuffix(value_type suffix) {
726 return endsWith(suffix) && (--e_, true);
730 * Replaces the content of the range, starting at position 'pos', with
731 * contents of 'replacement'. Entire 'replacement' must fit into the
732 * range. Returns false if 'replacements' does not fit. Example use:
734 * char in[] = "buffer";
735 * auto msp = MutablesStringPiece(input);
736 * EXPECT_TRUE(msp.replaceAt(2, "tt"));
737 * EXPECT_EQ(msp, "butter");
739 * // not enough space
740 * EXPECT_FALSE(msp.replace(msp.size() - 1, "rr"));
741 * EXPECT_EQ(msp, "butter"); // unchanged
743 bool replaceAt(size_t pos, const_range_type replacement) {
744 if (size() < pos + replacement.size()) {
748 std::copy(replacement.begin(), replacement.end(), begin() + pos);
754 * Replaces all occurences of 'source' with 'dest'. Returns number
755 * of replacements made. Source and dest have to have the same
756 * length. Throws if the lengths are different. If 'source' is a
757 * pattern that is overlapping with itself, we perform sequential
758 * replacement: "aaaaaaa".replaceAll("aa", "ba") --> "bababaa"
762 * char in[] = "buffer";
763 * auto msp = MutablesStringPiece(input);
764 * EXPECT_EQ(msp.replaceAll("ff","tt"), 1);
765 * EXPECT_EQ(msp, "butter");
767 size_t replaceAll(const_range_type source, const_range_type dest) {
768 if (source.size() != dest.size()) {
769 throw std::invalid_argument(
770 "replacement must have the same size as source");
778 size_t num_replaced = 0;
779 size_type found = std::string::npos;
780 while ((found = find(source, pos)) != std::string::npos) {
781 replaceAt(found, dest);
782 pos += source.size();
790 * Splits this `Range` `[b, e)` in the position `i` dictated by the next
791 * occurence of `delimiter`.
793 * Returns a new `Range` `[b, i)` and adjusts this range to start right after
794 * the delimiter's position. This range will be empty if the delimiter is not
795 * found. If called on an empty `Range`, both this and the returned `Range`
800 * folly::StringPiece s("sample string for split_next");
801 * auto p = s.split_step(' ');
803 * // prints "string for split_next"
811 * void tokenize(StringPiece s, char delimiter) {
812 * while (!s.empty()) {
813 * cout << s.split_step(delimiter);
817 * @author: Marcelo Juchem <marcelo@fb.com>
819 Range split_step(value_type delimiter) {
820 auto i = std::find(b_, e_, delimiter);
823 b_ = i == e_ ? e_ : std::next(i);
828 Range split_step(Range delimiter) {
829 auto i = find(delimiter);
830 Range result(b_, i == std::string::npos ? size() : i);
832 b_ = result.end() == e_
836 typename std::iterator_traits<Iter>::difference_type(
843 * Convenience method that calls `split_step()` and passes the result to a
844 * functor, returning whatever the functor does. Any additional arguments
845 * `args` passed to this function are perfectly forwarded to the functor.
847 * Say you have a functor with this signature:
849 * Foo fn(Range r) { }
851 * `split_step()`'s return type will be `Foo`. It works just like:
853 * auto result = fn(myRange.split_step(' '));
855 * A functor returning `void` is also supported.
859 * void do_some_parsing(folly::StringPiece s) {
860 * auto version = s.split_step(' ', [&](folly::StringPiece x) {
862 * throw std::invalid_argument("empty string");
864 * return std::strtoull(x.begin(), x.end(), 16);
871 * void parse(folly::StringPiece s) {
872 * s.split_step(' ', parse_field, bar, 10);
873 * s.split_step('\t', parse_field, baz, 20);
875 * auto const kludge = [](folly::StringPiece x, int &out, int def) {
879 * parse_field(x, out, def);
883 * s.split_step('\t', kludge, gaz);
884 * s.split_step(' ', kludge, foo);
893 * static parse_field(folly::StringPiece s, int &out, int def) {
895 * out = folly::to<int>(s);
896 * } catch (std::exception const &) {
902 * @author: Marcelo Juchem <marcelo@fb.com>
904 template <typename TProcess, typename... Args>
905 auto split_step(value_type delimiter, TProcess &&process, Args &&...args)
906 -> decltype(process(std::declval<Range>(), std::forward<Args>(args)...))
907 { return process(split_step(delimiter), std::forward<Args>(args)...); }
909 template <typename TProcess, typename... Args>
910 auto split_step(Range delimiter, TProcess &&process, Args &&...args)
911 -> decltype(process(std::declval<Range>(), std::forward<Args>(args)...))
912 { return process(split_step(delimiter), std::forward<Args>(args)...); }
918 template <class Iter>
919 const typename Range<Iter>::size_type Range<Iter>::npos = std::string::npos;
922 void swap(Range<T>& lhs, Range<T>& rhs) {
927 * Create a range from two iterators, with type deduction.
929 template <class Iter>
930 constexpr Range<Iter> range(Iter first, Iter last) {
931 return Range<Iter>(first, last);
935 * Creates a range to reference the contents of a contiguous-storage container.
937 // Use pointers for types with '.data()' member
940 class T = typename std::remove_pointer<
941 decltype(std::declval<Collection>().data())>::type>
942 constexpr Range<T*> range(Collection&& v) {
943 return Range<T*>(v.data(), v.data() + v.size());
946 template <class T, size_t n>
947 constexpr Range<T*> range(T (&array)[n]) {
948 return Range<T*>(array, array + n);
951 template <class T, size_t n>
952 constexpr Range<const T*> range(const std::array<T, n>& array) {
953 return Range<const T*>{array};
956 typedef Range<const char*> StringPiece;
957 typedef Range<char*> MutableStringPiece;
958 typedef Range<const unsigned char*> ByteRange;
959 typedef Range<unsigned char*> MutableByteRange;
961 inline std::ostream& operator<<(std::ostream& os,
962 const StringPiece piece) {
963 os.write(piece.start(), std::streamsize(piece.size()));
967 inline std::ostream& operator<<(std::ostream& os,
968 const MutableStringPiece piece) {
969 os.write(piece.start(), std::streamsize(piece.size()));
974 * Templated comparison operators
978 inline bool operator==(const Range<T>& lhs, const Range<T>& rhs) {
979 return lhs.size() == rhs.size() && lhs.compare(rhs) == 0;
983 inline bool operator<(const Range<T>& lhs, const Range<T>& rhs) {
984 return lhs.compare(rhs) < 0;
988 * Specializations of comparison operators for StringPiece
993 template <class A, class B>
994 struct ComparableAsStringPiece {
997 (std::is_convertible<A, StringPiece>::value
998 && std::is_same<B, StringPiece>::value)
1000 (std::is_convertible<B, StringPiece>::value
1001 && std::is_same<A, StringPiece>::value)
1005 } // namespace detail
1008 * operator== through conversion for Range<const char*>
1010 template <class T, class U>
1012 std::enable_if<detail::ComparableAsStringPiece<T, U>::value, bool>::type
1013 operator==(const T& lhs, const U& rhs) {
1014 return StringPiece(lhs) == StringPiece(rhs);
1018 * operator< through conversion for Range<const char*>
1020 template <class T, class U>
1022 std::enable_if<detail::ComparableAsStringPiece<T, U>::value, bool>::type
1023 operator<(const T& lhs, const U& rhs) {
1024 return StringPiece(lhs) < StringPiece(rhs);
1028 * operator> through conversion for Range<const char*>
1030 template <class T, class U>
1032 std::enable_if<detail::ComparableAsStringPiece<T, U>::value, bool>::type
1033 operator>(const T& lhs, const U& rhs) {
1034 return StringPiece(lhs) > StringPiece(rhs);
1038 * operator< through conversion for Range<const char*>
1040 template <class T, class U>
1042 std::enable_if<detail::ComparableAsStringPiece<T, U>::value, bool>::type
1043 operator<=(const T& lhs, const U& rhs) {
1044 return StringPiece(lhs) <= StringPiece(rhs);
1048 * operator> through conversion for Range<const char*>
1050 template <class T, class U>
1052 std::enable_if<detail::ComparableAsStringPiece<T, U>::value, bool>::type
1053 operator>=(const T& lhs, const U& rhs) {
1054 return StringPiece(lhs) >= StringPiece(rhs);
1058 * Finds substrings faster than brute force by borrowing from Boyer-Moore
1060 template <class T, class Comp>
1061 size_t qfind(const Range<T>& haystack,
1062 const Range<T>& needle,
1064 // Don't use std::search, use a Boyer-Moore-like trick by comparing
1065 // the last characters first
1066 auto const nsize = needle.size();
1067 if (haystack.size() < nsize) {
1068 return std::string::npos;
1070 if (!nsize) return 0;
1071 auto const nsize_1 = nsize - 1;
1072 auto const lastNeedle = needle[nsize_1];
1074 // Boyer-Moore skip value for the last char in the needle. Zero is
1075 // not a valid value; skip will be computed the first time it's
1077 std::string::size_type skip = 0;
1079 auto i = haystack.begin();
1080 auto iEnd = haystack.end() - nsize_1;
1083 // Boyer-Moore: match the last element in the needle
1084 while (!eq(i[nsize_1], lastNeedle)) {
1087 return std::string::npos;
1090 // Here we know that the last char matches
1091 // Continue in pedestrian mode
1092 for (size_t j = 0; ; ) {
1094 if (!eq(i[j], needle[j])) {
1095 // Not found, we can skip
1096 // Compute the skip value lazily
1099 while (skip <= nsize_1 && !eq(needle[nsize_1 - skip], lastNeedle)) {
1106 // Check if done searching
1109 return size_t(i - haystack.begin());
1113 return std::string::npos;
1118 inline size_t qfind_first_byte_of(const StringPiece haystack,
1119 const StringPiece needles) {
1120 static auto const qfind_first_byte_of_fn =
1121 folly::CpuId().sse42() ? qfind_first_byte_of_sse42
1122 : qfind_first_byte_of_nosse;
1123 return qfind_first_byte_of_fn(haystack, needles);
1126 } // namespace detail
1128 template <class T, class Comp>
1129 size_t qfind_first_of(const Range<T> & haystack,
1130 const Range<T> & needles,
1132 auto ret = std::find_first_of(haystack.begin(), haystack.end(),
1133 needles.begin(), needles.end(),
1135 return ret == haystack.end() ? std::string::npos : ret - haystack.begin();
1138 struct AsciiCaseSensitive {
1139 bool operator()(char lhs, char rhs) const {
1145 * Check if two ascii characters are case insensitive equal.
1146 * The difference between the lower/upper case characters are the 6-th bit.
1147 * We also check they are alpha chars, in case of xor = 32.
1149 struct AsciiCaseInsensitive {
1150 bool operator()(char lhs, char rhs) const {
1152 if (k == 0) return true;
1153 if (k != 32) return false;
1155 return (k >= 'a' && k <= 'z');
1160 size_t qfind(const Range<T>& haystack,
1161 const typename Range<T>::value_type& needle) {
1162 auto pos = std::find(haystack.begin(), haystack.end(), needle);
1163 return pos == haystack.end() ? std::string::npos : pos - haystack.data();
1167 size_t rfind(const Range<T>& haystack,
1168 const typename Range<T>::value_type& needle) {
1169 for (auto i = haystack.size(); i-- > 0; ) {
1170 if (haystack[i] == needle) {
1174 return std::string::npos;
1177 // specialization for StringPiece
1179 inline size_t qfind(const Range<const char*>& haystack, const char& needle) {
1180 // memchr expects a not-null pointer, early return if the range is empty.
1181 if (haystack.empty()) {
1182 return std::string::npos;
1184 auto pos = static_cast<const char*>(
1185 ::memchr(haystack.data(), needle, haystack.size()));
1186 return pos == nullptr ? std::string::npos : pos - haystack.data();
1190 inline size_t rfind(const Range<const char*>& haystack, const char& needle) {
1191 // memchr expects a not-null pointer, early return if the range is empty.
1192 if (haystack.empty()) {
1193 return std::string::npos;
1195 auto pos = static_cast<const char*>(
1196 ::memrchr(haystack.data(), needle, haystack.size()));
1197 return pos == nullptr ? std::string::npos : pos - haystack.data();
1200 // specialization for ByteRange
1202 inline size_t qfind(const Range<const unsigned char*>& haystack,
1203 const unsigned char& needle) {
1204 // memchr expects a not-null pointer, early return if the range is empty.
1205 if (haystack.empty()) {
1206 return std::string::npos;
1208 auto pos = static_cast<const unsigned char*>(
1209 ::memchr(haystack.data(), needle, haystack.size()));
1210 return pos == nullptr ? std::string::npos : pos - haystack.data();
1214 inline size_t rfind(const Range<const unsigned char*>& haystack,
1215 const unsigned char& needle) {
1216 // memchr expects a not-null pointer, early return if the range is empty.
1217 if (haystack.empty()) {
1218 return std::string::npos;
1220 auto pos = static_cast<const unsigned char*>(
1221 ::memrchr(haystack.data(), needle, haystack.size()));
1222 return pos == nullptr ? std::string::npos : pos - haystack.data();
1226 size_t qfind_first_of(const Range<T>& haystack,
1227 const Range<T>& needles) {
1228 return qfind_first_of(haystack, needles, AsciiCaseSensitive());
1231 // specialization for StringPiece
1233 inline size_t qfind_first_of(const Range<const char*>& haystack,
1234 const Range<const char*>& needles) {
1235 return detail::qfind_first_byte_of(haystack, needles);
1238 // specialization for ByteRange
1240 inline size_t qfind_first_of(const Range<const unsigned char*>& haystack,
1241 const Range<const unsigned char*>& needles) {
1242 return detail::qfind_first_byte_of(StringPiece(haystack),
1243 StringPiece(needles));
1246 template<class Key, class Enable>
1250 struct hasher<folly::Range<T*>,
1251 typename std::enable_if<std::is_pod<T>::value, void>::type> {
1252 size_t operator()(folly::Range<T*> r) const {
1253 return hash::SpookyHashV2::Hash64(r.begin(), r.size() * sizeof(T), 0);
1258 * Ubiquitous helper template for knowing what's a string
1260 template <class T> struct IsSomeString {
1261 enum { value = std::is_same<T, std::string>::value
1262 || std::is_same<T, fbstring>::value };
1265 } // !namespace folly
1267 #pragma GCC diagnostic pop
1269 FOLLY_ASSUME_FBVECTOR_COMPATIBLE_1(folly::Range);