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 # if !__clang__ || __CLANG_PREREQ(3, 7) // Clang 3.6 crashes on this line
209 /* implicit */ Range(std::nullptr_t) = delete;
212 template <class T = Iter, typename detail::IsCharPointer<T>::type = 0>
213 constexpr /* implicit */ Range(Iter str)
214 : b_(str), e_(str + constexpr_strlen(str)) {}
216 template <class T = Iter, typename detail::IsCharPointer<T>::const_type = 0>
217 /* implicit */ Range(const std::string& str)
218 : b_(str.data()), e_(b_ + str.size()) {}
220 template <class T = Iter, typename detail::IsCharPointer<T>::const_type = 0>
221 Range(const std::string& str, std::string::size_type startFrom) {
222 if (UNLIKELY(startFrom > str.size())) {
223 std::__throw_out_of_range("index out of range");
225 b_ = str.data() + startFrom;
226 e_ = str.data() + str.size();
229 template <class T = Iter, typename detail::IsCharPointer<T>::const_type = 0>
230 Range(const std::string& str,
231 std::string::size_type startFrom,
232 std::string::size_type size) {
233 if (UNLIKELY(startFrom > str.size())) {
234 std::__throw_out_of_range("index out of range");
236 b_ = str.data() + startFrom;
237 if (str.size() - startFrom < size) {
238 e_ = str.data() + str.size();
244 Range(const Range& other,
246 size_type length = npos)
247 : Range(other.subpiece(first, length))
250 template <class T = Iter, typename detail::IsCharPointer<T>::const_type = 0>
251 /* implicit */ Range(const fbstring& str)
252 : b_(str.data()), e_(b_ + str.size()) { }
254 template <class T = Iter, typename detail::IsCharPointer<T>::const_type = 0>
255 Range(const fbstring& str, fbstring::size_type startFrom) {
256 if (UNLIKELY(startFrom > str.size())) {
257 std::__throw_out_of_range("index out of range");
259 b_ = str.data() + startFrom;
260 e_ = str.data() + str.size();
263 template <class T = Iter, typename detail::IsCharPointer<T>::const_type = 0>
264 Range(const fbstring& str, fbstring::size_type startFrom,
265 fbstring::size_type size) {
266 if (UNLIKELY(startFrom > str.size())) {
267 std::__throw_out_of_range("index out of range");
269 b_ = str.data() + startFrom;
270 if (str.size() - startFrom < size) {
271 e_ = str.data() + str.size();
277 // Allow implicit conversion from Range<const char*> (aka StringPiece) to
278 // Range<const unsigned char*> (aka ByteRange), as they're both frequently
279 // used to represent ranges of bytes. Allow explicit conversion in the other
281 template <class OtherIter, typename std::enable_if<
282 (std::is_same<Iter, const unsigned char*>::value &&
283 (std::is_same<OtherIter, const char*>::value ||
284 std::is_same<OtherIter, char*>::value)), int>::type = 0>
285 /* implicit */ Range(const Range<OtherIter>& other)
286 : b_(reinterpret_cast<const unsigned char*>(other.begin())),
287 e_(reinterpret_cast<const unsigned char*>(other.end())) {
290 template <class OtherIter, typename std::enable_if<
291 (std::is_same<Iter, unsigned char*>::value &&
292 std::is_same<OtherIter, char*>::value), int>::type = 0>
293 /* implicit */ Range(const Range<OtherIter>& other)
294 : b_(reinterpret_cast<unsigned char*>(other.begin())),
295 e_(reinterpret_cast<unsigned char*>(other.end())) {
298 template <class OtherIter, typename std::enable_if<
299 (std::is_same<Iter, const char*>::value &&
300 (std::is_same<OtherIter, const unsigned char*>::value ||
301 std::is_same<OtherIter, unsigned char*>::value)), int>::type = 0>
302 explicit Range(const Range<OtherIter>& other)
303 : b_(reinterpret_cast<const char*>(other.begin())),
304 e_(reinterpret_cast<const char*>(other.end())) {
307 template <class OtherIter, typename std::enable_if<
308 (std::is_same<Iter, char*>::value &&
309 std::is_same<OtherIter, unsigned char*>::value), int>::type = 0>
310 explicit Range(const Range<OtherIter>& other)
311 : b_(reinterpret_cast<char*>(other.begin())),
312 e_(reinterpret_cast<char*>(other.end())) {
315 // Allow implicit conversion from Range<From> to Range<To> if From is
316 // implicitly convertible to To.
317 template <class OtherIter, typename std::enable_if<
318 (!std::is_same<Iter, OtherIter>::value &&
319 std::is_convertible<OtherIter, Iter>::value), int>::type = 0>
320 constexpr /* implicit */ Range(const Range<OtherIter>& other)
325 // Allow explicit conversion from Range<From> to Range<To> if From is
326 // explicitly convertible to To.
327 template <class OtherIter, typename std::enable_if<
328 (!std::is_same<Iter, OtherIter>::value &&
329 !std::is_convertible<OtherIter, Iter>::value &&
330 std::is_constructible<Iter, const OtherIter&>::value), int>::type = 0>
331 constexpr explicit Range(const Range<OtherIter>& other)
337 * Allow explicit construction of Range() from a std::array of a
340 * For instance, this allows constructing StringPiece from a
341 * std::array<char, N> or a std::array<const char, N>
346 typename = typename std::enable_if<
347 std::is_convertible<const T*, Iter>::value>::type>
348 constexpr explicit Range(const std::array<T, N>& array)
349 : b_{array.empty() ? nullptr : &array.at(0)},
350 e_{array.empty() ? nullptr : &array.at(0) + N} {}
355 typename std::enable_if<std::is_convertible<T*, Iter>::value>::type>
356 constexpr explicit Range(std::array<T, N>& array)
357 : b_{array.empty() ? nullptr : &array.at(0)},
358 e_{array.empty() ? nullptr : &array.at(0) + N} {}
360 Range& operator=(const Range& rhs) & = default;
361 Range& operator=(Range&& rhs) & = default;
368 void assign(Iter start, Iter end) {
373 void reset(Iter start, size_type size) {
378 // Works only for Range<const char*>
379 void reset(const std::string& str) {
380 reset(str.data(), str.size());
383 constexpr size_type size() const {
384 // It would be nice to assert(b_ <= e_) here. This can be achieved even
385 // in a C++11 compatible constexpr function:
386 // http://ericniebler.com/2014/09/27/assert-and-constexpr-in-cxx11/
387 // Unfortunately current gcc versions have a bug causing it to reject
388 // this check in a constexpr function:
389 // https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71448
390 return size_type(e_ - b_);
392 constexpr size_type walk_size() const {
393 return size_type(std::distance(b_, e_));
395 constexpr bool empty() const {
398 constexpr Iter data() const {
401 constexpr Iter start() const {
404 constexpr Iter begin() const {
407 constexpr Iter end() const {
410 constexpr Iter cbegin() const {
413 constexpr Iter cend() const {
416 value_type& front() {
422 return detail::value_before(e_);
424 const value_type& front() const {
428 const value_type& back() const {
430 return detail::value_before(e_);
432 // Works only for Range<const char*> and Range<char*>
433 std::string str() const { return std::string(b_, size()); }
434 std::string toString() const { return str(); }
435 // Works only for Range<const char*> and Range<char*>
436 fbstring fbstr() const { return fbstring(b_, size()); }
437 fbstring toFbstring() const { return fbstr(); }
439 const_range_type castToConst() const {
440 return const_range_type(*this);
443 // Works only for Range<const char*> and Range<char*>
444 int compare(const const_range_type& o) const {
445 const size_type tsize = this->size();
446 const size_type osize = o.size();
447 const size_type msize = std::min(tsize, osize);
448 int r = traits_type::compare(data(), o.data(), msize);
449 if (r == 0 && tsize != osize) {
450 // We check the signed bit of the subtraction and bit shift it
451 // to produce either 0 or 2. The subtraction yields the
452 // comparison values of either -1 or 1.
453 r = (static_cast<int>(
454 (osize - tsize) >> (CHAR_BIT * sizeof(size_t) - 1)) << 1) - 1;
459 value_type& operator[](size_t i) {
460 DCHECK_GT(size(), i);
464 const value_type& operator[](size_t i) const {
465 DCHECK_GT(size(), i);
469 value_type& at(size_t i) {
470 if (i >= size()) std::__throw_out_of_range("index out of range");
474 const value_type& at(size_t i) const {
475 if (i >= size()) std::__throw_out_of_range("index out of range");
479 // Do NOT use this function, which was left behind for backwards
480 // compatibility. Use SpookyHashV2 instead -- it is faster, and produces
481 // a 64-bit hash, which means dramatically fewer collisions in large maps.
482 // (The above advice does not apply if you are targeting a 32-bit system.)
484 // Works only for Range<const char*> and Range<char*>
487 // ** WANT TO GET RID OF THIS LINT? **
489 // A) Use a better hash function (*cough*folly::Hash*cough*), but
490 // only if you don't serialize data in a format that depends on
491 // this formula (ie the writer and reader assume this exact hash
492 // function is used).
494 // B) If you have to use this exact function then make your own hasher
495 // object and copy the body over (see thrift example: D3972362).
496 // https://github.com/facebook/fbthrift/commit/f8ed502e24ab4a32a9d5f266580
497 FOLLY_DEPRECATED("Replace with folly::Hash if the hash is not serialized")
498 uint32_t hash() const {
499 // Taken from fbi/nstring.h:
500 // Quick and dirty bernstein hash...fine for short ascii strings
501 uint32_t hash = 5381;
502 for (size_t ix = 0; ix < size(); ix++) {
503 hash = ((hash << 5) + hash) + b_[ix];
508 void advance(size_type n) {
509 if (UNLIKELY(n > size())) {
510 std::__throw_out_of_range("index out of range");
515 void subtract(size_type n) {
516 if (UNLIKELY(n > size())) {
517 std::__throw_out_of_range("index out of range");
522 Range subpiece(size_type first, size_type length = npos) const {
523 if (UNLIKELY(first > size())) {
524 std::__throw_out_of_range("index out of range");
527 return Range(b_ + first, std::min(length, size() - first));
530 // unchecked versions
531 void uncheckedAdvance(size_type n) {
532 DCHECK_LE(n, size());
536 void uncheckedSubtract(size_type n) {
537 DCHECK_LE(n, size());
541 Range uncheckedSubpiece(size_type first, size_type length = npos) const {
542 DCHECK_LE(first, size());
543 return Range(b_ + first, std::min(length, size() - first));
556 // string work-alike functions
557 size_type find(const_range_type str) const {
558 return qfind(castToConst(), str);
561 size_type find(const_range_type str, size_t pos) const {
562 if (pos > size()) return std::string::npos;
563 size_t ret = qfind(castToConst().subpiece(pos), str);
564 return ret == npos ? ret : ret + pos;
567 size_type find(Iter s, size_t pos, size_t n) const {
568 if (pos > size()) return std::string::npos;
569 auto forFinding = castToConst();
571 pos ? forFinding.subpiece(pos) : forFinding, const_range_type(s, n));
572 return ret == npos ? ret : ret + pos;
575 // Works only for Range<(const) (unsigned) char*> which have Range(Iter) ctor
576 size_type find(const Iter s) const {
577 return qfind(castToConst(), const_range_type(s));
580 // Works only for Range<(const) (unsigned) char*> which have Range(Iter) ctor
581 size_type find(const Iter s, size_t pos) const {
582 if (pos > size()) return std::string::npos;
583 size_type ret = qfind(castToConst().subpiece(pos), const_range_type(s));
584 return ret == npos ? ret : ret + pos;
587 size_type find(value_type c) const {
588 return qfind(castToConst(), c);
591 size_type rfind(value_type c) const {
592 return folly::rfind(castToConst(), c);
595 size_type find(value_type c, size_t pos) const {
596 if (pos > size()) return std::string::npos;
597 size_type ret = qfind(castToConst().subpiece(pos), c);
598 return ret == npos ? ret : ret + pos;
601 size_type find_first_of(const_range_type needles) const {
602 return qfind_first_of(castToConst(), needles);
605 size_type find_first_of(const_range_type needles, size_t pos) const {
606 if (pos > size()) return std::string::npos;
607 size_type ret = qfind_first_of(castToConst().subpiece(pos), needles);
608 return ret == npos ? ret : ret + pos;
611 // Works only for Range<(const) (unsigned) char*> which have Range(Iter) ctor
612 size_type find_first_of(Iter needles) const {
613 return find_first_of(const_range_type(needles));
616 // Works only for Range<(const) (unsigned) char*> which have Range(Iter) ctor
617 size_type find_first_of(Iter needles, size_t pos) const {
618 return find_first_of(const_range_type(needles), pos);
621 size_type find_first_of(Iter needles, size_t pos, size_t n) const {
622 return find_first_of(const_range_type(needles, n), pos);
625 size_type find_first_of(value_type c) const {
629 size_type find_first_of(value_type c, size_t pos) const {
634 * Determine whether the range contains the given subrange or item.
636 * Note: Call find() directly if the index is needed.
638 bool contains(const const_range_type& other) const {
639 return find(other) != std::string::npos;
642 bool contains(const value_type& other) const {
643 return find(other) != std::string::npos;
646 void swap(Range& rhs) {
647 std::swap(b_, rhs.b_);
648 std::swap(e_, rhs.e_);
652 * Does this Range start with another range?
654 bool startsWith(const const_range_type& other) const {
655 return size() >= other.size()
656 && castToConst().subpiece(0, other.size()) == other;
658 bool startsWith(value_type c) const {
659 return !empty() && front() == c;
663 * Does this Range end with another range?
665 bool endsWith(const const_range_type& other) const {
666 return size() >= other.size()
667 && castToConst().subpiece(size() - other.size()) == other;
669 bool endsWith(value_type c) const {
670 return !empty() && back() == c;
674 * Remove the items in [b, e), as long as this subrange is at the beginning
675 * or end of the Range.
677 * Required for boost::algorithm::trim()
679 void erase(Iter b, Iter e) {
682 } else if (e == e_) {
685 std::__throw_out_of_range("index out of range");
690 * Remove the given prefix and return true if the range starts with the given
691 * prefix; return false otherwise.
693 bool removePrefix(const const_range_type& prefix) {
694 return startsWith(prefix) && (b_ += prefix.size(), true);
696 bool removePrefix(value_type prefix) {
697 return startsWith(prefix) && (++b_, true);
701 * Remove the given suffix and return true if the range ends with the given
702 * suffix; return false otherwise.
704 bool removeSuffix(const const_range_type& suffix) {
705 return endsWith(suffix) && (e_ -= suffix.size(), true);
707 bool removeSuffix(value_type suffix) {
708 return endsWith(suffix) && (--e_, true);
712 * Replaces the content of the range, starting at position 'pos', with
713 * contents of 'replacement'. Entire 'replacement' must fit into the
714 * range. Returns false if 'replacements' does not fit. Example use:
716 * char in[] = "buffer";
717 * auto msp = MutablesStringPiece(input);
718 * EXPECT_TRUE(msp.replaceAt(2, "tt"));
719 * EXPECT_EQ(msp, "butter");
721 * // not enough space
722 * EXPECT_FALSE(msp.replace(msp.size() - 1, "rr"));
723 * EXPECT_EQ(msp, "butter"); // unchanged
725 bool replaceAt(size_t pos, const_range_type replacement) {
726 if (size() < pos + replacement.size()) {
730 std::copy(replacement.begin(), replacement.end(), begin() + pos);
736 * Replaces all occurences of 'source' with 'dest'. Returns number
737 * of replacements made. Source and dest have to have the same
738 * length. Throws if the lengths are different. If 'source' is a
739 * pattern that is overlapping with itself, we perform sequential
740 * replacement: "aaaaaaa".replaceAll("aa", "ba") --> "bababaa"
744 * char in[] = "buffer";
745 * auto msp = MutablesStringPiece(input);
746 * EXPECT_EQ(msp.replaceAll("ff","tt"), 1);
747 * EXPECT_EQ(msp, "butter");
749 size_t replaceAll(const_range_type source, const_range_type dest) {
750 if (source.size() != dest.size()) {
751 throw std::invalid_argument(
752 "replacement must have the same size as source");
760 size_t num_replaced = 0;
761 size_type found = std::string::npos;
762 while ((found = find(source, pos)) != std::string::npos) {
763 replaceAt(found, dest);
764 pos += source.size();
772 * Splits this `Range` `[b, e)` in the position `i` dictated by the next
773 * occurence of `delimiter`.
775 * Returns a new `Range` `[b, i)` and adjusts this range to start right after
776 * the delimiter's position. This range will be empty if the delimiter is not
777 * found. If called on an empty `Range`, both this and the returned `Range`
782 * folly::StringPiece s("sample string for split_next");
783 * auto p = s.split_step(' ');
785 * // prints "string for split_next"
793 * void tokenize(StringPiece s, char delimiter) {
794 * while (!s.empty()) {
795 * cout << s.split_step(delimiter);
799 * @author: Marcelo Juchem <marcelo@fb.com>
801 Range split_step(value_type delimiter) {
802 auto i = std::find(b_, e_, delimiter);
805 b_ = i == e_ ? e_ : std::next(i);
810 Range split_step(Range delimiter) {
811 auto i = find(delimiter);
812 Range result(b_, i == std::string::npos ? size() : i);
814 b_ = result.end() == e_
818 typename std::iterator_traits<Iter>::difference_type(
825 * Convenience method that calls `split_step()` and passes the result to a
826 * functor, returning whatever the functor does. Any additional arguments
827 * `args` passed to this function are perfectly forwarded to the functor.
829 * Say you have a functor with this signature:
831 * Foo fn(Range r) { }
833 * `split_step()`'s return type will be `Foo`. It works just like:
835 * auto result = fn(myRange.split_step(' '));
837 * A functor returning `void` is also supported.
841 * void do_some_parsing(folly::StringPiece s) {
842 * auto version = s.split_step(' ', [&](folly::StringPiece x) {
844 * throw std::invalid_argument("empty string");
846 * return std::strtoull(x.begin(), x.end(), 16);
853 * void parse(folly::StringPiece s) {
854 * s.split_step(' ', parse_field, bar, 10);
855 * s.split_step('\t', parse_field, baz, 20);
857 * auto const kludge = [](folly::StringPiece x, int &out, int def) {
861 * parse_field(x, out, def);
865 * s.split_step('\t', kludge, gaz);
866 * s.split_step(' ', kludge, foo);
875 * static parse_field(folly::StringPiece s, int &out, int def) {
877 * out = folly::to<int>(s);
878 * } catch (std::exception const &) {
884 * @author: Marcelo Juchem <marcelo@fb.com>
886 template <typename TProcess, typename... Args>
887 auto split_step(value_type delimiter, TProcess &&process, Args &&...args)
888 -> decltype(process(std::declval<Range>(), std::forward<Args>(args)...))
889 { return process(split_step(delimiter), std::forward<Args>(args)...); }
891 template <typename TProcess, typename... Args>
892 auto split_step(Range delimiter, TProcess &&process, Args &&...args)
893 -> decltype(process(std::declval<Range>(), std::forward<Args>(args)...))
894 { return process(split_step(delimiter), std::forward<Args>(args)...); }
900 template <class Iter>
901 const typename Range<Iter>::size_type Range<Iter>::npos = std::string::npos;
904 void swap(Range<T>& lhs, Range<T>& rhs) {
909 * Create a range from two iterators, with type deduction.
911 template <class Iter>
912 constexpr Range<Iter> range(Iter first, Iter last) {
913 return Range<Iter>(first, last);
917 * Creates a range to reference the contents of a contiguous-storage container.
919 // Use pointers for types with '.data()' member
922 class T = typename std::remove_pointer<
923 decltype(std::declval<Collection>().data())>::type>
924 constexpr Range<T*> range(Collection&& v) {
925 return Range<T*>(v.data(), v.data() + v.size());
928 template <class T, size_t n>
929 constexpr Range<T*> range(T (&array)[n]) {
930 return Range<T*>(array, array + n);
933 template <class T, size_t n>
934 constexpr Range<const T*> range(const std::array<T, n>& array) {
935 return Range<const T*>{array};
938 typedef Range<const char*> StringPiece;
939 typedef Range<char*> MutableStringPiece;
940 typedef Range<const unsigned char*> ByteRange;
941 typedef Range<unsigned char*> MutableByteRange;
943 inline std::ostream& operator<<(std::ostream& os,
944 const StringPiece piece) {
945 os.write(piece.start(), std::streamsize(piece.size()));
949 inline std::ostream& operator<<(std::ostream& os,
950 const MutableStringPiece piece) {
951 os.write(piece.start(), std::streamsize(piece.size()));
956 * Templated comparison operators
960 inline bool operator==(const Range<T>& lhs, const Range<T>& rhs) {
961 return lhs.size() == rhs.size() && lhs.compare(rhs) == 0;
965 inline bool operator<(const Range<T>& lhs, const Range<T>& rhs) {
966 return lhs.compare(rhs) < 0;
970 * Specializations of comparison operators for StringPiece
975 template <class A, class B>
976 struct ComparableAsStringPiece {
979 (std::is_convertible<A, StringPiece>::value
980 && std::is_same<B, StringPiece>::value)
982 (std::is_convertible<B, StringPiece>::value
983 && std::is_same<A, StringPiece>::value)
987 } // namespace detail
990 * operator== through conversion for Range<const char*>
992 template <class T, class U>
994 std::enable_if<detail::ComparableAsStringPiece<T, U>::value, bool>::type
995 operator==(const T& lhs, const U& rhs) {
996 return StringPiece(lhs) == StringPiece(rhs);
1000 * operator< through conversion for Range<const char*>
1002 template <class T, class U>
1004 std::enable_if<detail::ComparableAsStringPiece<T, U>::value, bool>::type
1005 operator<(const T& lhs, const U& rhs) {
1006 return StringPiece(lhs) < StringPiece(rhs);
1010 * operator> through conversion for Range<const char*>
1012 template <class T, class U>
1014 std::enable_if<detail::ComparableAsStringPiece<T, U>::value, bool>::type
1015 operator>(const T& lhs, const U& rhs) {
1016 return StringPiece(lhs) > StringPiece(rhs);
1020 * operator< through conversion for Range<const char*>
1022 template <class T, class U>
1024 std::enable_if<detail::ComparableAsStringPiece<T, U>::value, bool>::type
1025 operator<=(const T& lhs, const U& rhs) {
1026 return StringPiece(lhs) <= StringPiece(rhs);
1030 * operator> through conversion for Range<const char*>
1032 template <class T, class U>
1034 std::enable_if<detail::ComparableAsStringPiece<T, U>::value, bool>::type
1035 operator>=(const T& lhs, const U& rhs) {
1036 return StringPiece(lhs) >= StringPiece(rhs);
1040 * Finds substrings faster than brute force by borrowing from Boyer-Moore
1042 template <class T, class Comp>
1043 size_t qfind(const Range<T>& haystack,
1044 const Range<T>& needle,
1046 // Don't use std::search, use a Boyer-Moore-like trick by comparing
1047 // the last characters first
1048 auto const nsize = needle.size();
1049 if (haystack.size() < nsize) {
1050 return std::string::npos;
1052 if (!nsize) return 0;
1053 auto const nsize_1 = nsize - 1;
1054 auto const lastNeedle = needle[nsize_1];
1056 // Boyer-Moore skip value for the last char in the needle. Zero is
1057 // not a valid value; skip will be computed the first time it's
1059 std::string::size_type skip = 0;
1061 auto i = haystack.begin();
1062 auto iEnd = haystack.end() - nsize_1;
1065 // Boyer-Moore: match the last element in the needle
1066 while (!eq(i[nsize_1], lastNeedle)) {
1069 return std::string::npos;
1072 // Here we know that the last char matches
1073 // Continue in pedestrian mode
1074 for (size_t j = 0; ; ) {
1076 if (!eq(i[j], needle[j])) {
1077 // Not found, we can skip
1078 // Compute the skip value lazily
1081 while (skip <= nsize_1 && !eq(needle[nsize_1 - skip], lastNeedle)) {
1088 // Check if done searching
1091 return size_t(i - haystack.begin());
1095 return std::string::npos;
1100 inline size_t qfind_first_byte_of(const StringPiece haystack,
1101 const StringPiece needles) {
1102 static auto const qfind_first_byte_of_fn =
1103 folly::CpuId().sse42() ? qfind_first_byte_of_sse42
1104 : qfind_first_byte_of_nosse;
1105 return qfind_first_byte_of_fn(haystack, needles);
1108 } // namespace detail
1110 template <class T, class Comp>
1111 size_t qfind_first_of(const Range<T> & haystack,
1112 const Range<T> & needles,
1114 auto ret = std::find_first_of(haystack.begin(), haystack.end(),
1115 needles.begin(), needles.end(),
1117 return ret == haystack.end() ? std::string::npos : ret - haystack.begin();
1120 struct AsciiCaseSensitive {
1121 bool operator()(char lhs, char rhs) const {
1127 * Check if two ascii characters are case insensitive equal.
1128 * The difference between the lower/upper case characters are the 6-th bit.
1129 * We also check they are alpha chars, in case of xor = 32.
1131 struct AsciiCaseInsensitive {
1132 bool operator()(char lhs, char rhs) const {
1134 if (k == 0) return true;
1135 if (k != 32) return false;
1137 return (k >= 'a' && k <= 'z');
1142 size_t qfind(const Range<T>& haystack,
1143 const typename Range<T>::value_type& needle) {
1144 auto pos = std::find(haystack.begin(), haystack.end(), needle);
1145 return pos == haystack.end() ? std::string::npos : pos - haystack.data();
1149 size_t rfind(const Range<T>& haystack,
1150 const typename Range<T>::value_type& needle) {
1151 for (auto i = haystack.size(); i-- > 0; ) {
1152 if (haystack[i] == needle) {
1156 return std::string::npos;
1159 // specialization for StringPiece
1161 inline size_t qfind(const Range<const char*>& haystack, const char& needle) {
1162 // memchr expects a not-null pointer, early return if the range is empty.
1163 if (haystack.empty()) {
1164 return std::string::npos;
1166 auto pos = static_cast<const char*>(
1167 ::memchr(haystack.data(), needle, haystack.size()));
1168 return pos == nullptr ? std::string::npos : pos - haystack.data();
1172 inline size_t rfind(const Range<const char*>& haystack, const char& needle) {
1173 // memchr expects a not-null pointer, early return if the range is empty.
1174 if (haystack.empty()) {
1175 return std::string::npos;
1177 auto pos = static_cast<const char*>(
1178 ::memrchr(haystack.data(), needle, haystack.size()));
1179 return pos == nullptr ? std::string::npos : pos - haystack.data();
1182 // specialization for ByteRange
1184 inline size_t qfind(const Range<const unsigned char*>& haystack,
1185 const unsigned char& needle) {
1186 // memchr expects a not-null pointer, early return if the range is empty.
1187 if (haystack.empty()) {
1188 return std::string::npos;
1190 auto pos = static_cast<const unsigned char*>(
1191 ::memchr(haystack.data(), needle, haystack.size()));
1192 return pos == nullptr ? std::string::npos : pos - haystack.data();
1196 inline size_t rfind(const Range<const unsigned char*>& haystack,
1197 const unsigned char& needle) {
1198 // memchr expects a not-null pointer, early return if the range is empty.
1199 if (haystack.empty()) {
1200 return std::string::npos;
1202 auto pos = static_cast<const unsigned char*>(
1203 ::memrchr(haystack.data(), needle, haystack.size()));
1204 return pos == nullptr ? std::string::npos : pos - haystack.data();
1208 size_t qfind_first_of(const Range<T>& haystack,
1209 const Range<T>& needles) {
1210 return qfind_first_of(haystack, needles, AsciiCaseSensitive());
1213 // specialization for StringPiece
1215 inline size_t qfind_first_of(const Range<const char*>& haystack,
1216 const Range<const char*>& needles) {
1217 return detail::qfind_first_byte_of(haystack, needles);
1220 // specialization for ByteRange
1222 inline size_t qfind_first_of(const Range<const unsigned char*>& haystack,
1223 const Range<const unsigned char*>& needles) {
1224 return detail::qfind_first_byte_of(StringPiece(haystack),
1225 StringPiece(needles));
1228 template<class Key, class Enable>
1232 struct hasher<folly::Range<T*>,
1233 typename std::enable_if<std::is_pod<T>::value, void>::type> {
1234 size_t operator()(folly::Range<T*> r) const {
1235 return hash::SpookyHashV2::Hash64(r.begin(), r.size() * sizeof(T), 0);
1240 * Ubiquitous helper template for knowing what's a string
1242 template <class T> struct IsSomeString {
1243 enum { value = std::is_same<T, std::string>::value
1244 || std::is_same<T, fbstring>::value };
1247 } // !namespace folly
1249 #pragma GCC diagnostic pop
1251 FOLLY_ASSUME_FBVECTOR_COMPATIBLE_1(folly::Range);