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)
22 #include <folly/FBString.h>
23 #include <folly/Portability.h>
24 #include <folly/SpookyHashV2.h>
25 #include <folly/portability/Constexpr.h>
26 #include <folly/portability/String.h>
29 #include <boost/operators.hpp>
33 #include <glog/logging.h>
37 #include <type_traits>
39 // libc++ doesn't provide this header, nor does msvc
40 #ifdef FOLLY_HAVE_BITS_CXXCONFIG_H
41 // This file appears in two locations: inside fbcode and in the
42 // libstdc++ source code (when embedding fbstring as std::string).
43 // To aid in this schizophrenic use, two macros are defined in
45 // _LIBSTDCXX_FBSTRING - Set inside libstdc++. This is useful to
46 // gate use inside fbcode v. libstdc++
47 #include <bits/c++config.h>
50 #include <folly/CpuId.h>
51 #include <folly/Traits.h>
52 #include <folly/Likely.h>
53 #include <folly/detail/RangeCommon.h>
54 #include <folly/detail/RangeSse42.h>
56 // Ignore shadowing warnings within this file, so includers can use -Wshadow.
57 #pragma GCC diagnostic push
58 #pragma GCC diagnostic ignored "-Wshadow"
62 template <class T> class Range;
65 * Finds the first occurrence of needle in haystack. The algorithm is on
66 * average faster than O(haystack.size() * needle.size()) but not as fast
67 * as Boyer-Moore. On the upside, it does not do any upfront
68 * preprocessing and does not allocate memory.
70 template <class T, class Comp = std::equal_to<typename Range<T>::value_type>>
71 inline size_t qfind(const Range<T> & haystack,
72 const Range<T> & needle,
76 * Finds the first occurrence of needle in haystack. The result is the
77 * offset reported to the beginning of haystack, or string::npos if
78 * needle wasn't found.
81 size_t qfind(const Range<T> & haystack,
82 const typename Range<T>::value_type& needle);
85 * Finds the last occurrence of needle in haystack. The result is the
86 * offset reported to the beginning of haystack, or string::npos if
87 * needle wasn't found.
90 size_t rfind(const Range<T> & haystack,
91 const typename Range<T>::value_type& needle);
95 * Finds the first occurrence of any element of needle in
96 * haystack. The algorithm is O(haystack.size() * needle.size()).
99 inline size_t qfind_first_of(const Range<T> & haystack,
100 const Range<T> & needle);
103 * Small internal helper - returns the value just before an iterator.
108 * For random-access iterators, the value before is simply i[-1].
110 template <class Iter>
111 typename std::enable_if<
112 std::is_same<typename std::iterator_traits<Iter>::iterator_category,
113 std::random_access_iterator_tag>::value,
114 typename std::iterator_traits<Iter>::reference>::type
115 value_before(Iter i) {
120 * For all other iterators, we need to use the decrement operator.
122 template <class Iter>
123 typename std::enable_if<
124 !std::is_same<typename std::iterator_traits<Iter>::iterator_category,
125 std::random_access_iterator_tag>::value,
126 typename std::iterator_traits<Iter>::reference>::type
127 value_before(Iter i) {
132 * Use IsCharPointer<T>::type to enable const char* or char*.
133 * Use IsCharPointer<T>::const_type to enable only const char*.
135 template <class T> struct IsCharPointer {};
138 struct IsCharPointer<char*> {
143 struct IsCharPointer<const char*> {
144 typedef int const_type;
148 } // namespace detail
151 * Range abstraction keeping a pair of iterators. We couldn't use
152 * boost's similar range abstraction because we need an API identical
153 * with the former StringPiece class, which is used by a lot of other
154 * code. This abstraction does fulfill the needs of boost's
155 * range-oriented algorithms though.
157 * (Keep memory lifetime in mind when using this class, since it
158 * doesn't manage the data it refers to - just like an iterator
161 template <class Iter>
162 class Range : private boost::totally_ordered<Range<Iter> > {
164 typedef std::size_t size_type;
165 typedef Iter iterator;
166 typedef Iter const_iterator;
167 typedef typename std::remove_reference<
168 typename std::iterator_traits<Iter>::reference>::type
170 typedef typename std::iterator_traits<Iter>::reference reference;
173 * For MutableStringPiece and MutableByteRange we define StringPiece
174 * and ByteRange as const_range_type (for everything else its just
175 * identity). We do that to enable operations such as find with
176 * args which are const.
178 typedef typename std::conditional<
179 std::is_same<Iter, char*>::value
180 || std::is_same<Iter, unsigned char*>::value,
181 Range<const value_type*>,
182 Range<Iter>>::type const_range_type;
184 typedef std::char_traits<typename std::remove_const<value_type>::type>
187 static const size_type npos;
189 // Works for all iterators
190 constexpr Range() : b_(), e_() {
193 constexpr Range(const Range&) = default;
194 constexpr Range(Range&&) = default;
197 // Works for all iterators
198 constexpr Range(Iter start, Iter end) : b_(start), e_(end) {
201 // Works only for random-access iterators
202 constexpr Range(Iter start, size_t size)
203 : b_(start), e_(start + size) { }
205 # if !__clang__ || __CLANG_PREREQ(3, 7) // Clang 3.6 crashes on this line
206 /* implicit */ Range(std::nullptr_t) = delete;
209 template <class T = Iter, typename detail::IsCharPointer<T>::type = 0>
210 constexpr /* implicit */ Range(Iter str)
211 : b_(str), e_(str + constexpr_strlen(str)) {}
213 template <class T = Iter, typename detail::IsCharPointer<T>::const_type = 0>
214 /* implicit */ Range(const std::string& str)
215 : b_(str.data()), e_(b_ + str.size()) {}
217 template <class T = Iter, typename detail::IsCharPointer<T>::const_type = 0>
218 Range(const std::string& str, std::string::size_type startFrom) {
219 if (UNLIKELY(startFrom > str.size())) {
220 throw std::out_of_range("index out of range");
222 b_ = str.data() + startFrom;
223 e_ = str.data() + str.size();
226 template <class T = Iter, typename detail::IsCharPointer<T>::const_type = 0>
227 Range(const std::string& str,
228 std::string::size_type startFrom,
229 std::string::size_type size) {
230 if (UNLIKELY(startFrom > str.size())) {
231 throw std::out_of_range("index out of range");
233 b_ = str.data() + startFrom;
234 if (str.size() - startFrom < size) {
235 e_ = str.data() + str.size();
241 Range(const Range& other,
243 size_type length = npos)
244 : Range(other.subpiece(first, length))
247 template <class T = Iter, typename detail::IsCharPointer<T>::const_type = 0>
248 /* implicit */ Range(const fbstring& str)
249 : b_(str.data()), e_(b_ + str.size()) { }
251 template <class T = Iter, typename detail::IsCharPointer<T>::const_type = 0>
252 Range(const fbstring& str, fbstring::size_type startFrom) {
253 if (UNLIKELY(startFrom > str.size())) {
254 throw std::out_of_range("index out of range");
256 b_ = str.data() + startFrom;
257 e_ = str.data() + str.size();
260 template <class T = Iter, typename detail::IsCharPointer<T>::const_type = 0>
261 Range(const fbstring& str, fbstring::size_type startFrom,
262 fbstring::size_type size) {
263 if (UNLIKELY(startFrom > str.size())) {
264 throw std::out_of_range("index out of range");
266 b_ = str.data() + startFrom;
267 if (str.size() - startFrom < size) {
268 e_ = str.data() + str.size();
274 // Allow implicit conversion from Range<const char*> (aka StringPiece) to
275 // Range<const unsigned char*> (aka ByteRange), as they're both frequently
276 // used to represent ranges of bytes. Allow explicit conversion in the other
278 template <class OtherIter, typename std::enable_if<
279 (std::is_same<Iter, const unsigned char*>::value &&
280 (std::is_same<OtherIter, const char*>::value ||
281 std::is_same<OtherIter, char*>::value)), int>::type = 0>
282 /* implicit */ Range(const Range<OtherIter>& other)
283 : b_(reinterpret_cast<const unsigned char*>(other.begin())),
284 e_(reinterpret_cast<const unsigned char*>(other.end())) {
287 template <class OtherIter, typename std::enable_if<
288 (std::is_same<Iter, unsigned char*>::value &&
289 std::is_same<OtherIter, char*>::value), int>::type = 0>
290 /* implicit */ Range(const Range<OtherIter>& other)
291 : b_(reinterpret_cast<unsigned char*>(other.begin())),
292 e_(reinterpret_cast<unsigned char*>(other.end())) {
295 template <class OtherIter, typename std::enable_if<
296 (std::is_same<Iter, const char*>::value &&
297 (std::is_same<OtherIter, const unsigned char*>::value ||
298 std::is_same<OtherIter, unsigned char*>::value)), int>::type = 0>
299 explicit Range(const Range<OtherIter>& other)
300 : b_(reinterpret_cast<const char*>(other.begin())),
301 e_(reinterpret_cast<const char*>(other.end())) {
304 template <class OtherIter, typename std::enable_if<
305 (std::is_same<Iter, char*>::value &&
306 std::is_same<OtherIter, unsigned char*>::value), int>::type = 0>
307 explicit Range(const Range<OtherIter>& other)
308 : b_(reinterpret_cast<char*>(other.begin())),
309 e_(reinterpret_cast<char*>(other.end())) {
312 // Allow implicit conversion from Range<From> to Range<To> if From is
313 // implicitly convertible to To.
314 template <class OtherIter, typename std::enable_if<
315 (!std::is_same<Iter, OtherIter>::value &&
316 std::is_convertible<OtherIter, Iter>::value), int>::type = 0>
317 constexpr /* implicit */ Range(const Range<OtherIter>& other)
322 // Allow explicit conversion from Range<From> to Range<To> if From is
323 // explicitly convertible to To.
324 template <class OtherIter, typename std::enable_if<
325 (!std::is_same<Iter, OtherIter>::value &&
326 !std::is_convertible<OtherIter, Iter>::value &&
327 std::is_constructible<Iter, const OtherIter&>::value), int>::type = 0>
328 constexpr explicit Range(const Range<OtherIter>& other)
333 Range& operator=(const Range& rhs) & = default;
334 Range& operator=(Range&& rhs) & = default;
341 void assign(Iter start, Iter end) {
346 void reset(Iter start, size_type size) {
351 // Works only for Range<const char*>
352 void reset(const std::string& str) {
353 reset(str.data(), str.size());
356 constexpr size_type size() const {
357 // It would be nice to assert(b_ <= e_) here. This can be achieved even
358 // in a C++11 compatible constexpr function:
359 // http://ericniebler.com/2014/09/27/assert-and-constexpr-in-cxx11/
360 // Unfortunately current gcc versions have a bug causing it to reject
361 // this check in a constexpr function:
362 // https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71448
365 size_type walk_size() const {
366 return std::distance(b_, e_);
368 bool empty() const { return b_ == e_; }
369 Iter data() const { return b_; }
370 Iter start() const { return b_; }
371 Iter begin() const { return b_; }
372 Iter end() const { return e_; }
373 Iter cbegin() const { return b_; }
374 Iter cend() const { return e_; }
375 value_type& front() {
381 return detail::value_before(e_);
383 const value_type& front() const {
387 const value_type& back() const {
389 return detail::value_before(e_);
391 // Works only for Range<const char*> and Range<char*>
392 std::string str() const { return std::string(b_, size()); }
393 std::string toString() const { return str(); }
394 // Works only for Range<const char*> and Range<char*>
395 fbstring fbstr() const { return fbstring(b_, size()); }
396 fbstring toFbstring() const { return fbstr(); }
398 const_range_type castToConst() const {
399 return const_range_type(*this);
402 // Works only for Range<const char*> and Range<char*>
403 int compare(const const_range_type& o) const {
404 const size_type tsize = this->size();
405 const size_type osize = o.size();
406 const size_type msize = std::min(tsize, osize);
407 int r = traits_type::compare(data(), o.data(), msize);
408 if (r == 0 && tsize != osize) {
409 // We check the signed bit of the subtraction and bit shift it
410 // to produce either 0 or 2. The subtraction yields the
411 // comparison values of either -1 or 1.
412 r = (static_cast<int>(
413 (osize - tsize) >> (CHAR_BIT * sizeof(size_t) - 1)) << 1) - 1;
418 value_type& operator[](size_t i) {
419 DCHECK_GT(size(), i);
423 const value_type& operator[](size_t i) const {
424 DCHECK_GT(size(), i);
428 value_type& at(size_t i) {
429 if (i >= size()) throw std::out_of_range("index out of range");
433 const value_type& at(size_t i) const {
434 if (i >= size()) throw std::out_of_range("index out of range");
438 // Do NOT use this function, which was left behind for backwards
439 // compatibility. Use SpookyHashV2 instead -- it is faster, and produces
440 // a 64-bit hash, which means dramatically fewer collisions in large maps.
441 // (The above advice does not apply if you are targeting a 32-bit system.)
443 // Works only for Range<const char*> and Range<char*>
444 uint32_t hash() const {
445 // Taken from fbi/nstring.h:
446 // Quick and dirty bernstein hash...fine for short ascii strings
447 uint32_t hash = 5381;
448 for (size_t ix = 0; ix < size(); ix++) {
449 hash = ((hash << 5) + hash) + b_[ix];
454 void advance(size_type n) {
455 if (UNLIKELY(n > size())) {
456 throw std::out_of_range("index out of range");
461 void subtract(size_type n) {
462 if (UNLIKELY(n > size())) {
463 throw std::out_of_range("index out of range");
468 Range subpiece(size_type first, size_type length = npos) const {
469 if (UNLIKELY(first > size())) {
470 throw std::out_of_range("index out of range");
473 return Range(b_ + first, std::min(length, size() - first));
476 // unchecked versions
477 void uncheckedAdvance(size_type n) {
478 DCHECK_LE(n, size());
482 void uncheckedSubtract(size_type n) {
483 DCHECK_LE(n, size());
487 Range uncheckedSubpiece(size_type first, size_type length = npos) const {
488 DCHECK_LE(first, size());
489 return Range(b_ + first, std::min(length, size() - first));
502 // string work-alike functions
503 size_type find(const_range_type str) const {
504 return qfind(castToConst(), str);
507 size_type find(const_range_type str, size_t pos) const {
508 if (pos > size()) return std::string::npos;
509 size_t ret = qfind(castToConst().subpiece(pos), str);
510 return ret == npos ? ret : ret + pos;
513 size_type find(Iter s, size_t pos, size_t n) const {
514 if (pos > size()) return std::string::npos;
515 auto forFinding = castToConst();
517 pos ? forFinding.subpiece(pos) : forFinding, const_range_type(s, n));
518 return ret == npos ? ret : ret + pos;
521 // Works only for Range<(const) (unsigned) char*> which have Range(Iter) ctor
522 size_type find(const Iter s) const {
523 return qfind(castToConst(), const_range_type(s));
526 // Works only for Range<(const) (unsigned) char*> which have Range(Iter) ctor
527 size_type find(const Iter s, size_t pos) const {
528 if (pos > size()) return std::string::npos;
529 size_type ret = qfind(castToConst().subpiece(pos), const_range_type(s));
530 return ret == npos ? ret : ret + pos;
533 size_type find(value_type c) const {
534 return qfind(castToConst(), c);
537 size_type rfind(value_type c) const {
538 return folly::rfind(castToConst(), c);
541 size_type find(value_type c, size_t pos) const {
542 if (pos > size()) return std::string::npos;
543 size_type ret = qfind(castToConst().subpiece(pos), c);
544 return ret == npos ? ret : ret + pos;
547 size_type find_first_of(const_range_type needles) const {
548 return qfind_first_of(castToConst(), needles);
551 size_type find_first_of(const_range_type needles, size_t pos) const {
552 if (pos > size()) return std::string::npos;
553 size_type ret = qfind_first_of(castToConst().subpiece(pos), needles);
554 return ret == npos ? ret : ret + pos;
557 // Works only for Range<(const) (unsigned) char*> which have Range(Iter) ctor
558 size_type find_first_of(Iter needles) const {
559 return find_first_of(const_range_type(needles));
562 // Works only for Range<(const) (unsigned) char*> which have Range(Iter) ctor
563 size_type find_first_of(Iter needles, size_t pos) const {
564 return find_first_of(const_range_type(needles), pos);
567 size_type find_first_of(Iter needles, size_t pos, size_t n) const {
568 return find_first_of(const_range_type(needles, n), pos);
571 size_type find_first_of(value_type c) const {
575 size_type find_first_of(value_type c, size_t pos) const {
580 * Determine whether the range contains the given subrange or item.
582 * Note: Call find() directly if the index is needed.
584 bool contains(const const_range_type& other) const {
585 return find(other) != std::string::npos;
588 bool contains(const value_type& other) const {
589 return find(other) != std::string::npos;
592 void swap(Range& rhs) {
593 std::swap(b_, rhs.b_);
594 std::swap(e_, rhs.e_);
598 * Does this Range start with another range?
600 bool startsWith(const const_range_type& other) const {
601 return size() >= other.size()
602 && castToConst().subpiece(0, other.size()) == other;
604 bool startsWith(value_type c) const {
605 return !empty() && front() == c;
609 * Does this Range end with another range?
611 bool endsWith(const const_range_type& other) const {
612 return size() >= other.size()
613 && castToConst().subpiece(size() - other.size()) == other;
615 bool endsWith(value_type c) const {
616 return !empty() && back() == c;
620 * Remove the items in [b, e), as long as this subrange is at the beginning
621 * or end of the Range.
623 * Required for boost::algorithm::trim()
625 void erase(Iter b, Iter e) {
628 } else if (e == e_) {
631 throw std::out_of_range("index out of range");
636 * Remove the given prefix and return true if the range starts with the given
637 * prefix; return false otherwise.
639 bool removePrefix(const const_range_type& prefix) {
640 return startsWith(prefix) && (b_ += prefix.size(), true);
642 bool removePrefix(value_type prefix) {
643 return startsWith(prefix) && (++b_, true);
647 * Remove the given suffix and return true if the range ends with the given
648 * suffix; return false otherwise.
650 bool removeSuffix(const const_range_type& suffix) {
651 return endsWith(suffix) && (e_ -= suffix.size(), true);
653 bool removeSuffix(value_type suffix) {
654 return endsWith(suffix) && (--e_, true);
658 * Replaces the content of the range, starting at position 'pos', with
659 * contents of 'replacement'. Entire 'replacement' must fit into the
660 * range. Returns false if 'replacements' does not fit. Example use:
662 * char in[] = "buffer";
663 * auto msp = MutablesStringPiece(input);
664 * EXPECT_TRUE(msp.replaceAt(2, "tt"));
665 * EXPECT_EQ(msp, "butter");
667 * // not enough space
668 * EXPECT_FALSE(msp.replace(msp.size() - 1, "rr"));
669 * EXPECT_EQ(msp, "butter"); // unchanged
671 bool replaceAt(size_t pos, const_range_type replacement) {
672 if (size() < pos + replacement.size()) {
676 std::copy(replacement.begin(), replacement.end(), begin() + pos);
682 * Replaces all occurences of 'source' with 'dest'. Returns number
683 * of replacements made. Source and dest have to have the same
684 * length. Throws if the lengths are different. If 'source' is a
685 * pattern that is overlapping with itself, we perform sequential
686 * replacement: "aaaaaaa".replaceAll("aa", "ba") --> "bababaa"
690 * char in[] = "buffer";
691 * auto msp = MutablesStringPiece(input);
692 * EXPECT_EQ(msp.replaceAll("ff","tt"), 1);
693 * EXPECT_EQ(msp, "butter");
695 size_t replaceAll(const_range_type source, const_range_type dest) {
696 if (source.size() != dest.size()) {
697 throw std::invalid_argument(
698 "replacement must have the same size as source");
706 size_t num_replaced = 0;
707 size_type found = std::string::npos;
708 while ((found = find(source, pos)) != std::string::npos) {
709 replaceAt(found, dest);
710 pos += source.size();
718 * Splits this `Range` `[b, e)` in the position `i` dictated by the next
719 * occurence of `delimiter`.
721 * Returns a new `Range` `[b, i)` and adjusts this range to start right after
722 * the delimiter's position. This range will be empty if the delimiter is not
723 * found. If called on an empty `Range`, both this and the returned `Range`
728 * folly::StringPiece s("sample string for split_next");
729 * auto p = s.split_step(' ');
731 * // prints "string for split_next"
739 * void tokenize(StringPiece s, char delimiter) {
740 * while (!s.empty()) {
741 * cout << s.split_step(delimiter);
745 * @author: Marcelo Juchem <marcelo@fb.com>
747 Range split_step(value_type delimiter) {
748 auto i = std::find(b_, e_, delimiter);
751 b_ = i == e_ ? e_ : std::next(i);
756 Range split_step(Range delimiter) {
757 auto i = find(delimiter);
758 Range result(b_, i == std::string::npos ? size() : i);
760 b_ = result.end() == e_ ? e_ : std::next(result.end(), delimiter.size());
766 * Convenience method that calls `split_step()` and passes the result to a
767 * functor, returning whatever the functor does. Any additional arguments
768 * `args` passed to this function are perfectly forwarded to the functor.
770 * Say you have a functor with this signature:
772 * Foo fn(Range r) { }
774 * `split_step()`'s return type will be `Foo`. It works just like:
776 * auto result = fn(myRange.split_step(' '));
778 * A functor returning `void` is also supported.
782 * void do_some_parsing(folly::StringPiece s) {
783 * auto version = s.split_step(' ', [&](folly::StringPiece x) {
785 * throw std::invalid_argument("empty string");
787 * return std::strtoull(x.begin(), x.end(), 16);
794 * void parse(folly::StringPiece s) {
795 * s.split_step(' ', parse_field, bar, 10);
796 * s.split_step('\t', parse_field, baz, 20);
798 * auto const kludge = [](folly::StringPiece x, int &out, int def) {
802 * parse_field(x, out, def);
806 * s.split_step('\t', kludge, gaz);
807 * s.split_step(' ', kludge, foo);
816 * static parse_field(folly::StringPiece s, int &out, int def) {
818 * out = folly::to<int>(s);
819 * } catch (std::exception const &) {
825 * @author: Marcelo Juchem <marcelo@fb.com>
827 template <typename TProcess, typename... Args>
828 auto split_step(value_type delimiter, TProcess &&process, Args &&...args)
829 -> decltype(process(std::declval<Range>(), std::forward<Args>(args)...))
830 { return process(split_step(delimiter), std::forward<Args>(args)...); }
832 template <typename TProcess, typename... Args>
833 auto split_step(Range delimiter, TProcess &&process, Args &&...args)
834 -> decltype(process(std::declval<Range>(), std::forward<Args>(args)...))
835 { return process(split_step(delimiter), std::forward<Args>(args)...); }
841 template <class Iter>
842 const typename Range<Iter>::size_type Range<Iter>::npos = std::string::npos;
845 void swap(Range<T>& lhs, Range<T>& rhs) {
850 * Create a range from two iterators, with type deduction.
852 template <class Iter>
853 Range<Iter> range(Iter first, Iter last) {
854 return Range<Iter>(first, last);
858 * Creates a range to reference the contents of a contiguous-storage container.
860 // Use pointers for types with '.data()' member
861 template <class Collection,
862 class T = typename std::remove_pointer<
863 decltype(std::declval<Collection>().data())>::type>
864 Range<T*> range(Collection&& v) {
865 return Range<T*>(v.data(), v.data() + v.size());
868 template <class T, size_t n>
869 Range<T*> range(T (&array)[n]) {
870 return Range<T*>(array, array + n);
873 typedef Range<const char*> StringPiece;
874 typedef Range<char*> MutableStringPiece;
875 typedef Range<const unsigned char*> ByteRange;
876 typedef Range<unsigned char*> MutableByteRange;
878 inline std::ostream& operator<<(std::ostream& os,
879 const StringPiece piece) {
880 os.write(piece.start(), piece.size());
884 inline std::ostream& operator<<(std::ostream& os,
885 const MutableStringPiece piece) {
886 os.write(piece.start(), piece.size());
891 * Templated comparison operators
895 inline bool operator==(const Range<T>& lhs, const Range<T>& rhs) {
896 return lhs.size() == rhs.size() && lhs.compare(rhs) == 0;
900 inline bool operator<(const Range<T>& lhs, const Range<T>& rhs) {
901 return lhs.compare(rhs) < 0;
905 * Specializations of comparison operators for StringPiece
910 template <class A, class B>
911 struct ComparableAsStringPiece {
914 (std::is_convertible<A, StringPiece>::value
915 && std::is_same<B, StringPiece>::value)
917 (std::is_convertible<B, StringPiece>::value
918 && std::is_same<A, StringPiece>::value)
922 } // namespace detail
925 * operator== through conversion for Range<const char*>
927 template <class T, class U>
929 std::enable_if<detail::ComparableAsStringPiece<T, U>::value, bool>::type
930 operator==(const T& lhs, const U& rhs) {
931 return StringPiece(lhs) == StringPiece(rhs);
935 * operator< through conversion for Range<const char*>
937 template <class T, class U>
939 std::enable_if<detail::ComparableAsStringPiece<T, U>::value, bool>::type
940 operator<(const T& lhs, const U& rhs) {
941 return StringPiece(lhs) < StringPiece(rhs);
945 * operator> through conversion for Range<const char*>
947 template <class T, class U>
949 std::enable_if<detail::ComparableAsStringPiece<T, U>::value, bool>::type
950 operator>(const T& lhs, const U& rhs) {
951 return StringPiece(lhs) > StringPiece(rhs);
955 * operator< through conversion for Range<const char*>
957 template <class T, class U>
959 std::enable_if<detail::ComparableAsStringPiece<T, U>::value, bool>::type
960 operator<=(const T& lhs, const U& rhs) {
961 return StringPiece(lhs) <= StringPiece(rhs);
965 * operator> through conversion for Range<const char*>
967 template <class T, class U>
969 std::enable_if<detail::ComparableAsStringPiece<T, U>::value, bool>::type
970 operator>=(const T& lhs, const U& rhs) {
971 return StringPiece(lhs) >= StringPiece(rhs);
974 // Do NOT use this, use SpookyHashV2 instead, see commment on hash() above.
975 struct StringPieceHash {
976 std::size_t operator()(const StringPiece str) const {
977 return static_cast<std::size_t>(str.hash());
982 * Finds substrings faster than brute force by borrowing from Boyer-Moore
984 template <class T, class Comp>
985 size_t qfind(const Range<T>& haystack,
986 const Range<T>& needle,
988 // Don't use std::search, use a Boyer-Moore-like trick by comparing
989 // the last characters first
990 auto const nsize = needle.size();
991 if (haystack.size() < nsize) {
992 return std::string::npos;
994 if (!nsize) return 0;
995 auto const nsize_1 = nsize - 1;
996 auto const lastNeedle = needle[nsize_1];
998 // Boyer-Moore skip value for the last char in the needle. Zero is
999 // not a valid value; skip will be computed the first time it's
1001 std::string::size_type skip = 0;
1003 auto i = haystack.begin();
1004 auto iEnd = haystack.end() - nsize_1;
1007 // Boyer-Moore: match the last element in the needle
1008 while (!eq(i[nsize_1], lastNeedle)) {
1011 return std::string::npos;
1014 // Here we know that the last char matches
1015 // Continue in pedestrian mode
1016 for (size_t j = 0; ; ) {
1018 if (!eq(i[j], needle[j])) {
1019 // Not found, we can skip
1020 // Compute the skip value lazily
1023 while (skip <= nsize_1 && !eq(needle[nsize_1 - skip], lastNeedle)) {
1030 // Check if done searching
1033 return i - haystack.begin();
1037 return std::string::npos;
1042 inline size_t qfind_first_byte_of(const StringPiece haystack,
1043 const StringPiece needles) {
1044 static auto const qfind_first_byte_of_fn =
1045 folly::CpuId().sse42() ? qfind_first_byte_of_sse42
1046 : qfind_first_byte_of_nosse;
1047 return qfind_first_byte_of_fn(haystack, needles);
1050 } // namespace detail
1052 template <class T, class Comp>
1053 size_t qfind_first_of(const Range<T> & haystack,
1054 const Range<T> & needles,
1056 auto ret = std::find_first_of(haystack.begin(), haystack.end(),
1057 needles.begin(), needles.end(),
1059 return ret == haystack.end() ? std::string::npos : ret - haystack.begin();
1062 struct AsciiCaseSensitive {
1063 bool operator()(char lhs, char rhs) const {
1069 * Check if two ascii characters are case insensitive equal.
1070 * The difference between the lower/upper case characters are the 6-th bit.
1071 * We also check they are alpha chars, in case of xor = 32.
1073 struct AsciiCaseInsensitive {
1074 bool operator()(char lhs, char rhs) const {
1076 if (k == 0) return true;
1077 if (k != 32) return false;
1079 return (k >= 'a' && k <= 'z');
1084 size_t qfind(const Range<T>& haystack,
1085 const typename Range<T>::value_type& needle) {
1086 auto pos = std::find(haystack.begin(), haystack.end(), needle);
1087 return pos == haystack.end() ? std::string::npos : pos - haystack.data();
1091 size_t rfind(const Range<T>& haystack,
1092 const typename Range<T>::value_type& needle) {
1093 for (auto i = haystack.size(); i-- > 0; ) {
1094 if (haystack[i] == needle) {
1098 return std::string::npos;
1101 // specialization for StringPiece
1103 inline size_t qfind(const Range<const char*>& haystack, const char& needle) {
1104 // memchr expects a not-null pointer, early return if the range is empty.
1105 if (haystack.empty()) {
1106 return std::string::npos;
1108 auto pos = static_cast<const char*>(
1109 ::memchr(haystack.data(), needle, haystack.size()));
1110 return pos == nullptr ? std::string::npos : pos - haystack.data();
1114 inline size_t rfind(const Range<const char*>& haystack, const char& needle) {
1115 // memchr expects a not-null pointer, early return if the range is empty.
1116 if (haystack.empty()) {
1117 return std::string::npos;
1119 auto pos = static_cast<const char*>(
1120 ::memrchr(haystack.data(), needle, haystack.size()));
1121 return pos == nullptr ? std::string::npos : pos - haystack.data();
1124 // specialization for ByteRange
1126 inline size_t qfind(const Range<const unsigned char*>& haystack,
1127 const unsigned char& needle) {
1128 // memchr expects a not-null pointer, early return if the range is empty.
1129 if (haystack.empty()) {
1130 return std::string::npos;
1132 auto pos = static_cast<const unsigned char*>(
1133 ::memchr(haystack.data(), needle, haystack.size()));
1134 return pos == nullptr ? std::string::npos : pos - haystack.data();
1138 inline size_t rfind(const Range<const unsigned char*>& haystack,
1139 const unsigned char& needle) {
1140 // memchr expects a not-null pointer, early return if the range is empty.
1141 if (haystack.empty()) {
1142 return std::string::npos;
1144 auto pos = static_cast<const unsigned char*>(
1145 ::memrchr(haystack.data(), needle, haystack.size()));
1146 return pos == nullptr ? std::string::npos : pos - haystack.data();
1150 size_t qfind_first_of(const Range<T>& haystack,
1151 const Range<T>& needles) {
1152 return qfind_first_of(haystack, needles, AsciiCaseSensitive());
1155 // specialization for StringPiece
1157 inline size_t qfind_first_of(const Range<const char*>& haystack,
1158 const Range<const char*>& needles) {
1159 return detail::qfind_first_byte_of(haystack, needles);
1162 // specialization for ByteRange
1164 inline size_t qfind_first_of(const Range<const unsigned char*>& haystack,
1165 const Range<const unsigned char*>& needles) {
1166 return detail::qfind_first_byte_of(StringPiece(haystack),
1167 StringPiece(needles));
1170 template<class Key, class Enable>
1174 struct hasher<folly::Range<T*>,
1175 typename std::enable_if<std::is_pod<T>::value, void>::type> {
1176 size_t operator()(folly::Range<T*> r) const {
1177 return hash::SpookyHashV2::Hash64(r.begin(), r.size() * sizeof(T), 0);
1182 * Ubiquitous helper template for knowing what's a string
1184 template <class T> struct IsSomeString {
1185 enum { value = std::is_same<T, std::string>::value
1186 || std::is_same<T, fbstring>::value };
1189 } // !namespace folly
1191 #pragma GCC diagnostic pop
1193 FOLLY_ASSUME_FBVECTOR_COMPATIBLE_1(folly::Range);