X-Git-Url: http://plrg.eecs.uci.edu/git/?p=folly.git;a=blobdiff_plain;f=folly%2FFBString.h;h=e492e8f560f04eb317c340f6965528f298d13c50;hp=7c465870e9fad0a03c5731e448b2019a36b37b9f;hb=275ca94d04e44f28cfa411668eb1c1dd8db90b80;hpb=d006f324c88d45bb7f01d5fefea2ee2e52c09e2c diff --git a/folly/FBString.h b/folly/FBString.h index 7c465870..e492e8f5 100644 --- a/folly/FBString.h +++ b/folly/FBString.h @@ -1,5 +1,5 @@ /* - * Copyright 2013 Facebook, Inc. + * Copyright 2015 Facebook, Inc. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. @@ -20,52 +20,14 @@ #ifndef FOLLY_BASE_FBSTRING_H_ #define FOLLY_BASE_FBSTRING_H_ -/** - fbstring's behavior can be configured via two macro definitions, as - follows. Normally, fbstring does not write a '\0' at the end of - each string whenever it changes the underlying characters. Instead, - it lazily writes the '\0' whenever either c_str() or data() - called. - - This is standard-compliant behavior and may save costs in some - circumstances. However, it may be surprising to some client code - because c_str() and data() are const member functions (fbstring - uses the "mutable" storage class for its own state). - - In order to appease client code that expects fbstring to be - zero-terminated at all times, if the preprocessor symbol - FBSTRING_CONSERVATIVE is defined, fbstring does exactly that, - i.e. it goes the extra mile to guarantee a '\0' is always planted - at the end of its data. - - On the contrary, if the desire is to debug faulty client code that - unduly assumes the '\0' is present, fbstring plants a '^' (i.e., - emphatically NOT a zero) at the end of each string if - FBSTRING_PERVERSE is defined. (Calling c_str() or data() still - writes the '\0', of course.) - - The preprocessor symbols FBSTRING_PERVERSE and - FBSTRING_CONSERVATIVE cannot be defined simultaneously. This is - enforced during preprocessing. -*/ - -//#define FBSTRING_PERVERSE -//#define FBSTRING_CONSERVATIVE - -#ifdef FBSTRING_PERVERSE -#ifdef FBSTRING_CONSERVATIVE -#error Cannot define both FBSTRING_PERVERSE and FBSTRING_CONSERVATIVE. -#endif -#endif +#include +#include +#include // This file appears in two locations: inside fbcode and in the // libstdc++ source code (when embedding fbstring as std::string). -// To aid in this schizophrenic use, two macros are defined in -// c++config.h: -// _LIBSTDCXX_FBSTRING - Set inside libstdc++. This is useful to -// gate use inside fbcode v. libstdc++ -#include - +// To aid in this schizophrenic use, _LIBSTDCXX_FBSTRING is defined in +// libstdc++'s c++config.h, to gate use inside fbcode v. libstdc++. #ifdef _LIBSTDCXX_FBSTRING #pragma GCC system_header @@ -74,32 +36,58 @@ // either before or after this inclusion. #ifdef FOLLY_MALLOC_H_ #undef FOLLY_MALLOC_H_ -#include "basic_fbstring_malloc.h" +#include "basic_fbstring_malloc.h" // nolint #else -#include "basic_fbstring_malloc.h" +#include "basic_fbstring_malloc.h" // nolint #undef FOLLY_MALLOC_H_ #endif #else // !_LIBSTDCXX_FBSTRING +#include + +// libc++ doesn't provide this header, nor does msvc +#ifdef FOLLY_HAVE_BITS_CXXCONFIG_H +#include +#endif + #include #include #include +#include + +#include +#include +#include +#include -#include "folly/Traits.h" -#include "folly/Malloc.h" -#include "folly/Hash.h" +#if FOLLY_HAVE_DEPRECATED_ASSOC +#ifdef _GLIBCXX_SYMVER +#include +#include +#endif +#endif #endif // We defined these here rather than including Likely.h to avoid // redefinition errors when fbstring is imported into libstdc++. +#if defined(__GNUC__) && __GNUC__ >= 4 #define FBSTRING_LIKELY(x) (__builtin_expect((x), 1)) #define FBSTRING_UNLIKELY(x) (__builtin_expect((x), 0)) +#else +#define FBSTRING_LIKELY(x) (x) +#define FBSTRING_UNLIKELY(x) (x) +#endif -#include -#include -#include +// Ignore shadowing warnings within this file, so includers can use -Wshadow. +#pragma GCC diagnostic push +#pragma GCC diagnostic ignored "-Wshadow" + +// FBString cannot use throw when replacing std::string, though it may still +// use std::__throw_* +// nolint +#define throw FOLLY_FBSTRING_MAY_NOT_USE_THROW #ifdef _LIBSTDCXX_FBSTRING namespace std _GLIBCXX_VISIBILITY(default) { @@ -108,6 +96,30 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION namespace folly { #endif +// Different versions of gcc/clang support different versions of +// the address sanitizer attribute. Unfortunately, this attribute +// has issues when inlining is used, so disable that as well. +#if defined(__clang__) +# if __has_feature(address_sanitizer) +# if __has_attribute(__no_address_safety_analysis__) +# define FBSTRING_DISABLE_ADDRESS_SANITIZER \ + __attribute__((__no_address_safety_analysis__, __noinline__)) +# elif __has_attribute(__no_sanitize_address__) +# define FBSTRING_DISABLE_ADDRESS_SANITIZER \ + __attribute__((__no_sanitize_address__, __noinline__)) +# endif +# endif +#elif defined (__GNUC__) && \ + (__GNUC__ == 4) && \ + (__GNUC_MINOR__ >= 8) && \ + __SANITIZE_ADDRESS__ +# define FBSTRING_DISABLE_ADDRESS_SANITIZER \ + __attribute__((__no_address_safety_analysis__, __noinline__)) +#endif +#ifndef FBSTRING_DISABLE_ADDRESS_SANITIZER +# define FBSTRING_DISABLE_ADDRESS_SANITIZER +#endif + namespace fbstring_detail { template @@ -116,7 +128,6 @@ OutIt copy_n(InIt b, typename std::iterator_traits::difference_type n, OutIt d) { for (; n != 0; --n, ++b, ++d) { - assert((const void*)&*d != &*b); *d = *b; } return d; @@ -277,22 +288,25 @@ private: */ template class fbstring_core { public: - fbstring_core() { + fbstring_core() noexcept { // Only initialize the tag, will set the MSBs (i.e. the small // string size) to zero too ml_.capacity_ = maxSmallSize << (8 * (sizeof(size_t) - sizeof(Char))); // or: setSmallSize(0); writeTerminator(); - assert(category() == isSmall && size() == 0); + assert(category() == Category::isSmall && size() == 0); } fbstring_core(const fbstring_core & rhs) { assert(&rhs != this); // Simplest case first: small strings are bitblitted - if (rhs.category() == isSmall) { - assert(offsetof(MediumLarge, data_) == 0); - assert(offsetof(MediumLarge, size_) == sizeof(ml_.data_)); - assert(offsetof(MediumLarge, capacity_) == 2 * sizeof(ml_.data_)); + if (rhs.category() == Category::isSmall) { + static_assert(offsetof(MediumLarge, data_) == 0, + "fbstring layout failure"); + static_assert(offsetof(MediumLarge, size_) == sizeof(ml_.data_), + "fbstring layout failure"); + static_assert(offsetof(MediumLarge, capacity_) == 2 * sizeof(ml_.data_), + "fbstring layout failure"); const size_t size = rhs.smallSize(); if (size == 0) { ml_.capacity_ = rhs.ml_.capacity_; @@ -305,12 +319,12 @@ public: // ml_.capacity field). ml_ = rhs.ml_; } - assert(category() == isSmall && this->size() == rhs.size()); - } else if (rhs.category() == isLarge) { + assert(category() == Category::isSmall && this->size() == rhs.size()); + } else if (rhs.category() == Category::isLarge) { // Large strings are just refcounted ml_ = rhs.ml_; RefCounted::incrementRefs(ml_.data_); - assert(category() == isLarge && size() == rhs.size()); + assert(category() == Category::isLarge && size() == rhs.size()); } else { // Medium strings are copied eagerly. Don't forget to allocate // one extra Char for the null terminator. @@ -324,15 +338,16 @@ public: // No need for writeTerminator() here, we copied one extra // element just above. ml_.size_ = rhs.ml_.size_; - ml_.capacity_ = (allocSize / sizeof(Char) - 1) | isMedium; - assert(category() == isMedium); + ml_.capacity_ = (allocSize / sizeof(Char) - 1) + | static_cast(Category::isMedium); + assert(category() == Category::isMedium); } assert(size() == rhs.size()); assert(memcmp(data(), rhs.data(), size() * sizeof(Char)) == 0); } - fbstring_core(fbstring_core&& goner) { - if (goner.category() == isSmall) { + fbstring_core(fbstring_core&& goner) noexcept { + if (goner.category() == Category::isSmall) { // Just copy, leave the goner in peace new(this) fbstring_core(goner.small_, goner.smallSize()); } else { @@ -343,14 +358,30 @@ public: } } - fbstring_core(const Char *const data, const size_t size) { + // NOTE(agallagher): The word-aligned copy path copies bytes which are + // outside the range of the string, and makes address sanitizer unhappy, + // so just disable it on this function. + fbstring_core(const Char *const data, const size_t size) + FBSTRING_DISABLE_ADDRESS_SANITIZER { +#ifndef NDEBUG +#ifndef _LIBSTDCXX_FBSTRING + SCOPE_EXIT { + assert(this->size() == size); + assert(memcmp(this->data(), data, size * sizeof(Char)) == 0); + }; +#endif +#endif + // Simplest case first: small strings are bitblitted if (size <= maxSmallSize) { // Layout is: Char* data_, size_t size_, size_t capacity_ - /*static_*/assert(sizeof(*this) == sizeof(Char*) + 2 * sizeof(size_t)); - /*static_*/assert(sizeof(Char*) == sizeof(size_t)); + static_assert(sizeof(*this) == sizeof(Char*) + 2 * sizeof(size_t), + "fbstring has unexpected size"); + static_assert(sizeof(Char*) == sizeof(size_t), + "fbstring size assumption violation"); // sizeof(size_t) must be a power of 2 - /*static_*/assert((sizeof(size_t) & (sizeof(size_t) - 1)) == 0); + static_assert((sizeof(size_t) & (sizeof(size_t) - 1)) == 0, + "fbstring size assumption violation"); // If data is aligned, use fast word-wise copying. Otherwise, // use conservative memcpy. @@ -375,6 +406,7 @@ public: } } setSmallSize(size); + return; } else if (size <= maxMediumSize) { // Medium strings are allocated normally. Don't forget to // allocate one extra Char for the terminating null. @@ -382,26 +414,26 @@ public: ml_.data_ = static_cast(checkedMalloc(allocSize)); fbstring_detail::pod_copy(data, data + size, ml_.data_); ml_.size_ = size; - ml_.capacity_ = (allocSize / sizeof(Char) - 1) | isMedium; + ml_.capacity_ = (allocSize / sizeof(Char) - 1) + | static_cast(Category::isMedium); } else { // Large strings are allocated differently size_t effectiveCapacity = size; auto const newRC = RefCounted::create(data, & effectiveCapacity); ml_.data_ = newRC->data_; ml_.size_ = size; - ml_.capacity_ = effectiveCapacity | isLarge; + ml_.capacity_ = effectiveCapacity + | static_cast(Category::isLarge); } writeTerminator(); - assert(this->size() == size); - assert(memcmp(this->data(), data, size * sizeof(Char)) == 0); } - ~fbstring_core() { + ~fbstring_core() noexcept { auto const c = category(); - if (c == isSmall) { + if (c == Category::isSmall) { return; } - if (c == isMedium) { + if (c == Category::isMedium) { free(ml_.data_); return; } @@ -409,22 +441,25 @@ public: } // Snatches a previously mallocated string. The parameter "size" - // is the size of the string, and the parameter "capacity" is the size - // of the mallocated block. The string must be \0-terminated, so - // data[size] == '\0' and capacity >= size + 1. + // is the size of the string, and the parameter "allocatedSize" + // is the size of the mallocated block. The string must be + // \0-terminated, so allocatedSize >= size + 1 and data[size] == '\0'. // - // So if you want a 2-character string, pass malloc(3) as "data", pass 2 as - // "size", and pass 3 as "capacity". - fbstring_core(Char *const data, const size_t size, - const size_t capacity, + // So if you want a 2-character string, pass malloc(3) as "data", + // pass 2 as "size", and pass 3 as "allocatedSize". + fbstring_core(Char * const data, + const size_t size, + const size_t allocatedSize, AcquireMallocatedString) { if (size > 0) { - assert(capacity > size); + assert(allocatedSize >= size + 1); assert(data[size] == '\0'); // Use the medium string storage ml_.data_ = data; ml_.size_ = size; - ml_.capacity_ = capacity | isMedium; + // Don't forget about null terminator + ml_.capacity_ = (allocatedSize - 1) + | static_cast(Category::isMedium); } else { // No need for the memory free(data); @@ -449,11 +484,11 @@ public: Char * mutable_data() { auto const c = category(); - if (c == isSmall) { + if (c == Category::isSmall) { return small_; } - assert(c == isMedium || c == isLarge); - if (c == isLarge && RefCounted::refs(ml_.data_) > 1) { + assert(c == Category::isMedium || c == Category::isLarge); + if (c == Category::isLarge && RefCounted::refs(ml_.data_) > 1) { // Ensure unique. size_t effectiveCapacity = ml_.capacity(); auto const newRC = RefCounted::create(& effectiveCapacity); @@ -471,44 +506,27 @@ public: const Char * c_str() const { auto const c = category(); -#ifdef FBSTRING_PERVERSE - if (c == isSmall) { - assert(small_[smallSize()] == TERMINATOR || smallSize() == maxSmallSize - || small_[smallSize()] == '\0'); - small_[smallSize()] = '\0'; - return small_; - } - assert(c == isMedium || c == isLarge); - assert(ml_.data_[ml_.size_] == TERMINATOR || ml_.data_[ml_.size_] == '\0'); - ml_.data_[ml_.size_] = '\0'; -#elif defined(FBSTRING_CONSERVATIVE) - if (c == isSmall) { + if (c == Category::isSmall) { assert(small_[smallSize()] == '\0'); return small_; } - assert(c == isMedium || c == isLarge); + assert(c == Category::isMedium || c == Category::isLarge); assert(ml_.data_[ml_.size_] == '\0'); -#else - if (c == isSmall) { - small_[smallSize()] = '\0'; - return small_; - } - assert(c == isMedium || c == isLarge); - ml_.data_[ml_.size_] = '\0'; -#endif return ml_.data_; } void shrink(const size_t delta) { - if (category() == isSmall) { + if (category() == Category::isSmall) { // Check for underflow assert(delta <= smallSize()); setSmallSize(smallSize() - delta); - } else if (category() == isMedium || RefCounted::refs(ml_.data_) == 1) { + } else if (category() == Category::isMedium || + RefCounted::refs(ml_.data_) == 1) { // Medium strings and unique large strings need no special // handling. assert(ml_.size_ >= delta); ml_.size_ -= delta; + writeTerminator(); } else { assert(ml_.size_ >= delta); // Shared large string, must make unique. This is because of the @@ -518,13 +536,11 @@ public: fbstring_core(ml_.data_, ml_.size_ - delta).swap(*this); } // No need to write the terminator. - return; } - writeTerminator(); } void reserve(size_t minCapacity) { - if (category() == isLarge) { + if (category() == Category::isLarge) { // Ensure unique if (RefCounted::refs(ml_.data_) > 1) { // We must make it unique regardless; in-place reallocation is @@ -540,7 +556,8 @@ public: // we have + 1 above. RefCounted::decrementRefs(ml_.data_); ml_.data_ = newRC->data_; - ml_.capacity_ = minCapacity | isLarge; + ml_.capacity_ = minCapacity + | static_cast(Category::isLarge); // size remains unchanged } else { // String is not shared, so let's try to realloc (if needed) @@ -550,12 +567,13 @@ public: RefCounted::reallocate(ml_.data_, ml_.size_, ml_.capacity(), minCapacity); ml_.data_ = newRC->data_; - ml_.capacity_ = minCapacity | isLarge; + ml_.capacity_ = minCapacity + | static_cast(Category::isLarge); writeTerminator(); } assert(capacity() >= minCapacity); } - } else if (category() == isMedium) { + } else if (category() == Category::isMedium) { // String is not shared if (minCapacity <= ml_.capacity()) { return; // nothing to do, there's enough room @@ -568,10 +586,11 @@ public: smartRealloc( ml_.data_, ml_.size_ * sizeof(Char), - ml_.capacity() * sizeof(Char), + (ml_.capacity() + 1) * sizeof(Char), capacityBytes)); writeTerminator(); - ml_.capacity_ = (capacityBytes / sizeof(Char) - 1) | isMedium; + ml_.capacity_ = (capacityBytes / sizeof(Char) - 1) + | static_cast(Category::isMedium); } else { // Conversion from medium to large string fbstring_core nascent; @@ -585,7 +604,7 @@ public: assert(capacity() >= minCapacity); } } else { - assert(category() == isSmall); + assert(category() == Category::isSmall); if (minCapacity > maxMediumSize) { // large auto const newRC = RefCounted::create(& minCapacity); @@ -594,7 +613,8 @@ public: // No need for writeTerminator(), we wrote it above with + 1. ml_.data_ = newRC->data_; ml_.size_ = size; - ml_.capacity_ = minCapacity | isLarge; + ml_.capacity_ = minCapacity + | static_cast(Category::isLarge); assert(capacity() >= minCapacity); } else if (minCapacity > maxSmallSize) { // medium @@ -607,7 +627,8 @@ public: // No need for writeTerminator(), we wrote it above with + 1. ml_.data_ = data; ml_.size_ = size; - ml_.capacity_ = (allocSizeBytes / sizeof(Char) - 1) | isMedium; + ml_.capacity_ = (allocSizeBytes / sizeof(Char) - 1) + | static_cast(Category::isMedium); } else { // small // Nothing to do, everything stays put @@ -620,12 +641,11 @@ public: // Strategy is simple: make room, then change size assert(capacity() >= size()); size_t sz, newSz; - if (category() == isSmall) { + if (category() == Category::isSmall) { sz = smallSize(); newSz = sz + delta; if (newSz <= maxSmallSize) { setSmallSize(newSz); - writeTerminator(); return small_ + sz; } reserve(newSz); @@ -638,7 +658,7 @@ public: } assert(capacity() >= newSz); // Category can't be small - we took care of that above - assert(category() == isMedium || category() == isLarge); + assert(category() == Category::isMedium || category() == Category::isLarge); ml_.size_ = newSz; writeTerminator(); assert(size() == newSz); @@ -648,39 +668,38 @@ public: void push_back(Char c) { assert(capacity() >= size()); size_t sz; - if (category() == isSmall) { + if (category() == Category::isSmall) { sz = smallSize(); if (sz < maxSmallSize) { - setSmallSize(sz + 1); small_[sz] = c; - writeTerminator(); + setSmallSize(sz + 1); return; } reserve(maxSmallSize * 2); } else { sz = ml_.size_; if (sz == capacity()) { // always true for isShared() - reserve(sz * 3 / 2); // ensures not shared + reserve(1 + sz * 3 / 2); // ensures not shared } } assert(!isShared()); assert(capacity() >= sz + 1); // Category can't be small - we took care of that above - assert(category() == isMedium || category() == isLarge); + assert(category() == Category::isMedium || category() == Category::isLarge); ml_.size_ = sz + 1; ml_.data_[sz] = c; writeTerminator(); } size_t size() const { - return category() == isSmall ? smallSize() : ml_.size_; + return category() == Category::isSmall ? smallSize() : ml_.size_; } size_t capacity() const { switch (category()) { - case isSmall: + case Category::isSmall: return maxSmallSize; - case isLarge: + case Category::isLarge: // For large-sized strings, a multi-referenced chunk has no // available capacity. This is because any attempt to append // data would trigger a new allocation. @@ -691,26 +710,18 @@ public: } bool isShared() const { - return category() == isLarge && RefCounted::refs(ml_.data_) > 1; + return category() == Category::isLarge && RefCounted::refs(ml_.data_) > 1; } -#ifdef FBSTRING_PERVERSE - enum { TERMINATOR = '^' }; -#else - enum { TERMINATOR = '\0' }; -#endif - void writeTerminator() { -#if defined(FBSTRING_PERVERSE) || defined(FBSTRING_CONSERVATIVE) - if (category() == isSmall) { + if (category() == Category::isSmall) { const auto s = smallSize(); if (s != maxSmallSize) { - small_[s] = TERMINATOR; + small_[s] = '\0'; } } else { - ml_.data_[ml_.size_] = TERMINATOR; + ml_.data_[ml_.size_] = '\0'; } -#endif } private: @@ -795,8 +806,8 @@ private: }; union { - mutable Char small_[sizeof(MediumLarge) / sizeof(Char)]; - mutable MediumLarge ml_; + Char small_[sizeof(MediumLarge) / sizeof(Char)]; + MediumLarge ml_; }; enum { @@ -810,7 +821,10 @@ private: static_assert(!(sizeof(MediumLarge) % sizeof(Char)), "Corrupt memory layout for fbstring."); - enum Category { + typedef std::conditional::type + category_type; + + enum class Category : category_type { isSmall = 0, isMedium = sizeof(size_t) == 4 ? 0x80000000 : 0x8000000000000000, isLarge = sizeof(size_t) == 4 ? 0x40000000 : 0x4000000000000000, @@ -822,7 +836,9 @@ private: } size_t smallSize() const { - assert(category() == isSmall && small_[maxSmallSize] <= maxSmallSize); + assert(category() == Category::isSmall && + static_cast(small_[maxSmallSize]) + <= static_cast(maxSmallSize)); return static_cast(maxSmallSize) - static_cast(small_[maxSmallSize]); } @@ -833,6 +849,7 @@ private: // small_[maxSmallSize]. assert(s <= maxSmallSize); small_[maxSmallSize] = maxSmallSize - s; + writeTerminator(); } }; @@ -926,7 +943,7 @@ class basic_fbstring { size() <= max_size() && capacity() <= max_size() && size() <= capacity() && - (begin()[size()] == Storage::TERMINATOR || begin()[size()] == '\0'); + begin()[size()] == '\0'; } struct Invariant; @@ -981,8 +998,8 @@ private: } public: - // 21.3.1 construct/copy/destroy - explicit basic_fbstring(const A& a = A()) { + // C++11 21.4.2 construct/copy/destroy + explicit basic_fbstring(const A& a = A()) noexcept { } basic_fbstring(const basic_fbstring& str) @@ -990,7 +1007,8 @@ public: } // Move constructor - basic_fbstring(basic_fbstring&& goner) : store_(std::move(goner.store_)) { + basic_fbstring(basic_fbstring&& goner) noexcept + : store_(std::move(goner.store_)) { } #ifndef _LIBSTDCXX_FBSTRING @@ -1006,12 +1024,11 @@ public: } /* implicit */ basic_fbstring(const value_type* s, const A& a = A()) - : store_(s, s ? traits_type::length(s) : ({ - basic_fbstring err = __PRETTY_FUNCTION__; - err += ": null pointer initializer not valid"; - std::__throw_logic_error(err.c_str()); - 0; - })) { + : store_(s, s + ? traits_type::length(s) + : (std::__throw_logic_error( + "basic_fbstring: null pointer initializer not valid"), + 0)) { } basic_fbstring(const value_type* s, size_type n, const A& a = A()) @@ -1043,7 +1060,12 @@ public: : store_(s, n, c, a) { } - ~basic_fbstring() { + // Construction from initialization list + basic_fbstring(std::initializer_list il) { + assign(il.begin(), il.end()); + } + + ~basic_fbstring() noexcept { } basic_fbstring& operator=(const basic_fbstring& lhs) { @@ -1069,9 +1091,12 @@ public: } // Move assignment - basic_fbstring& operator=(basic_fbstring&& goner) { - // Self move assignment is illegal, see 17.6.4.9 for the explanation - assert(&goner != this); + basic_fbstring& operator=(basic_fbstring&& goner) noexcept { + if (FBSTRING_UNLIKELY(&goner == this)) { + // Compatibility with std::basic_string<>, + // C++11 21.4.2 [string.cons] / 23 requires self-move-assignment support. + return *this; + } // No need of this anymore this->~basic_fbstring(); // Move the goner into this @@ -1114,11 +1139,17 @@ public: return *this; } - // 21.3.2 iterators: + basic_fbstring& operator=(std::initializer_list il) { + return assign(il.begin(), il.end()); + } + + // C++11 21.4.3 iterators: iterator begin() { return store_.mutable_data(); } const_iterator begin() const { return store_.data(); } + const_iterator cbegin() const { return begin(); } + iterator end() { return store_.mutable_data() + store_.size(); } @@ -1127,6 +1158,8 @@ public: return store_.data() + store_.size(); } + const_iterator cend() const { return end(); } + reverse_iterator rbegin() { return reverse_iterator(end()); } @@ -1135,6 +1168,8 @@ public: return const_reverse_iterator(end()); } + const_reverse_iterator crbegin() const { return rbegin(); } + reverse_iterator rend() { return reverse_iterator(begin()); } @@ -1143,6 +1178,8 @@ public: return const_reverse_iterator(begin()); } + const_reverse_iterator crend() const { return rend(); } + // Added by C++11 // C++11 21.4.5, element access: const value_type& front() const { return *begin(); } @@ -1162,7 +1199,7 @@ public: store_.shrink(1); } - // 21.3.3 capacity: + // C++11 21.4.4 capacity: size_type size() const { return store_.size(); } size_type length() const { return size(); } @@ -1206,20 +1243,24 @@ public: store_.reserve(res_arg); } + void shrink_to_fit() { + // Shrink only if slack memory is sufficiently large + if (capacity() < size() * 3 / 2) { + return; + } + basic_fbstring(cbegin(), cend()).swap(*this); + } + void clear() { resize(0); } bool empty() const { return size() == 0; } - // 21.3.4 element access: + // C++11 21.4.5 element access: const_reference operator[](size_type pos) const { - return *(c_str() + pos); + return *(begin() + pos); } reference operator[](size_type pos) { - if (pos == size()) { - // Just call c_str() to make sure '\0' is present - c_str(); - } return *(begin() + pos); } @@ -1233,7 +1274,7 @@ public: return (*this)[n]; } - // 21.3.5 modifiers: + // C++11 21.4.6 modifiers: basic_fbstring& operator+=(const basic_fbstring& str) { return append(str); } @@ -1247,6 +1288,11 @@ public: return *this; } + basic_fbstring& operator+=(std::initializer_list il) { + append(il); + return *this; + } + basic_fbstring& append(const basic_fbstring& str) { #ifndef NDEBUG auto desiredSize = size() + str.size(); @@ -1316,6 +1362,10 @@ public: return *this; } + basic_fbstring& append(std::initializer_list il) { + return append(il.begin(), il.end()); + } + void push_back(const value_type c) { // primitive store_.push_back(c); } @@ -1325,6 +1375,10 @@ public: return assign(str.data(), str.size()); } + basic_fbstring& assign(basic_fbstring&& str) { + return *this = std::move(str); + } + basic_fbstring& assign(const basic_fbstring& str, const size_type pos, size_type n) { const size_type sz = str.size(); @@ -1355,6 +1409,10 @@ public: return assign(s, traits_type::length(s)); } + basic_fbstring& assign(std::initializer_list il) { + return assign(il.begin(), il.end()); + } + template basic_fbstring& assign(ItOrLength first_or_n, ItOrChar last_or_c) { return replace(begin(), end(), first_or_n, last_or_c); @@ -1387,7 +1445,7 @@ public: return *this; } - iterator insert(const iterator p, const value_type c) { + iterator insert(const_iterator p, const value_type c) { const size_type pos = p - begin(); insert(p, 1, c); return begin() + pos; @@ -1396,10 +1454,11 @@ public: private: template class Selector {}; - basic_fbstring& insertImplDiscr(iterator p, - size_type n, value_type c, Selector<1>) { + iterator insertImplDiscr(const_iterator p, + size_type n, value_type c, Selector<1>) { Invariant checker(*this); (void) checker; + auto const pos = p - begin(); assert(p >= begin() && p <= end()); if (capacity() - size() < n) { const size_type sz = p - begin(); @@ -1407,33 +1466,33 @@ private: p = begin() + sz; } const iterator oldEnd = end(); - if( n < size_type(oldEnd - p)) { + if (n < size_type(oldEnd - p)) { append(oldEnd - n, oldEnd); //std::copy( // reverse_iterator(oldEnd - n), // reverse_iterator(p), // reverse_iterator(oldEnd)); - fbstring_detail::pod_move(&*p, &*oldEnd - n, &*p + n); - std::fill(p, p + n, c); + fbstring_detail::pod_move(&*p, &*oldEnd - n, + begin() + pos + n); + std::fill(begin() + pos, begin() + pos + n, c); } else { append(n - (end() - p), c); - append(p, oldEnd); - std::fill(p, oldEnd, c); + append(iterator(p), oldEnd); + std::fill(iterator(p), oldEnd, c); } store_.writeTerminator(); - return *this; + return begin() + pos; } template - basic_fbstring& insertImplDiscr(iterator i, - InputIter b, InputIter e, Selector<0>) { - insertImpl(i, b, e, + iterator insertImplDiscr(const_iterator i, + InputIter b, InputIter e, Selector<0>) { + return insertImpl(i, b, e, typename std::iterator_traits::iterator_category()); - return *this; } template - void insertImpl(iterator i, + iterator insertImpl(const_iterator i, FwdIterator s1, FwdIterator s2, std::forward_iterator_tag) { Invariant checker(*this); (void) checker; @@ -1455,9 +1514,9 @@ private: const iterator tailBegin = end() - n2; store_.expand_noinit(n2); fbstring_detail::pod_copy(tailBegin, tailBegin + n2, end() - n2); - std::copy(reverse_iterator(tailBegin), reverse_iterator(i), + std::copy(const_reverse_iterator(tailBegin), const_reverse_iterator(i), reverse_iterator(tailBegin + n2)); - std::copy(s1, s2, i); + std::copy(s1, s2, begin() + pos); } else { FwdIterator t = s1; const size_type old_size = size(); @@ -1467,27 +1526,35 @@ private: std::copy(t, s2, begin() + old_size); fbstring_detail::pod_copy(data() + pos, data() + old_size, begin() + old_size + newElems); - std::copy(s1, t, i); + std::copy(s1, t, begin() + pos); } store_.writeTerminator(); + return begin() + pos; } template - void insertImpl(iterator i, - InputIterator b, InputIterator e, std::input_iterator_tag) { + iterator insertImpl(const_iterator i, + InputIterator b, InputIterator e, + std::input_iterator_tag) { + const auto pos = i - begin(); basic_fbstring temp(begin(), i); for (; b != e; ++b) { temp.push_back(*b); } - temp.append(i, end()); + temp.append(i, cend()); swap(temp); + return begin() + pos; } public: template - void insert(iterator p, ItOrLength first_or_n, ItOrChar last_or_c) { + iterator insert(const_iterator p, ItOrLength first_or_n, ItOrChar last_or_c) { Selector::is_specialized> sel; - insertImplDiscr(p, first_or_n, last_or_c, sel); + return insertImplDiscr(p, first_or_n, last_or_c, sel); + } + + iterator insert(const_iterator p, std::initializer_list il) { + return insert(p, il.begin(), il.end()); } basic_fbstring& erase(size_type pos = 0, size_type n = npos) { @@ -1595,15 +1662,15 @@ private: } private: - template + template bool replaceAliased(iterator i1, iterator i2, - FwdIterator s1, FwdIterator s2, P*) { + FwdIterator s1, FwdIterator s2, std::false_type) { return false; } template bool replaceAliased(iterator i1, iterator i2, - FwdIterator s1, FwdIterator s2, value_type*) { + FwdIterator s1, FwdIterator s2, std::true_type) { static const std::less_equal le = std::less_equal(); const bool aliased = le(&*begin(), &*s1) && le(&*s1, &*end()); @@ -1618,7 +1685,6 @@ private: return true; } -public: template void replaceImpl(iterator i1, iterator i2, FwdIterator s1, FwdIterator s2, std::forward_iterator_tag) { @@ -1626,7 +1692,10 @@ public: (void) checker; // Handle aliased replace - if (replaceAliased(i1, i2, s1, s2, &*s1)) { + if (replaceAliased(i1, i2, s1, s2, + std::integral_constant::value || + std::is_same::value>())) { return; } @@ -1683,7 +1752,6 @@ public: store_.swap(rhs.store_); } - // 21.3.6 string operations: const value_type* c_str() const { return store_.c_str(); } @@ -1702,7 +1770,9 @@ public: const size_type nsize) const { if (!nsize) return pos; auto const size = this->size(); - if (nsize + pos > size) return npos; + // nsize + pos can overflow (eg pos == npos), guard against that by checking + // that nsize + pos does not wrap around. + if (nsize + pos > size || nsize + pos < pos) return npos; // Don't use std::search, use a Boyer-Moore-like trick by comparing // the last characters first auto const haystack = data(); @@ -1899,11 +1969,18 @@ public: return find_last_not_of(&c, pos, 1); } - basic_fbstring substr(size_type pos = 0, size_type n = npos) const { + basic_fbstring substr(size_type pos = 0, size_type n = npos) const& { enforce(pos <= size(), std::__throw_out_of_range, ""); return basic_fbstring(data() + pos, std::min(n, size() - pos)); } + basic_fbstring substr(size_type pos = 0, size_type n = npos) && { + enforce(pos <= size(), std::__throw_out_of_range, ""); + erase(0, pos); + if (n < size()) resize(n); + return std::move(*this); + } + int compare(const basic_fbstring& str) const { // FIX due to Goncalo N M de Carvalho July 18, 2005 return compare(0, size(), str); @@ -1951,7 +2028,7 @@ private: }; // non-member functions -// C++11 21.4.8.1/2 +// C++11 21.4.8.1/1 template inline basic_fbstring operator+(const basic_fbstring& lhs, @@ -1993,24 +2070,44 @@ basic_fbstring operator+(basic_fbstring&& lhs, return std::move(lhs.append(rhs)); } +// C++11 21.4.8.1/5 template inline basic_fbstring operator+( - const typename basic_fbstring::value_type* lhs, + const E* lhs, const basic_fbstring& rhs) { // basic_fbstring result; - const typename basic_fbstring::size_type len = - basic_fbstring::traits_type::length(lhs); + const auto len = basic_fbstring::traits_type::length(lhs); + result.reserve(len + rhs.size()); + result.append(lhs, len).append(rhs); + return result; +} + +// C++11 21.4.8.1/6 +template +inline +basic_fbstring operator+( + const E* lhs, + basic_fbstring&& rhs) { + // + const auto len = basic_fbstring::traits_type::length(lhs); + if (rhs.capacity() >= len + rhs.size()) { + // Good, at least we don't need to reallocate + return std::move(rhs.insert(rhs.begin(), lhs, lhs + len)); + } + // Meh, no go. Do it by hand since we have len already. + basic_fbstring result; result.reserve(len + rhs.size()); result.append(lhs, len).append(rhs); return result; } +// C++11 21.4.8.1/7 template inline basic_fbstring operator+( - typename basic_fbstring::value_type lhs, + E lhs, const basic_fbstring& rhs) { basic_fbstring result; @@ -2020,11 +2117,28 @@ basic_fbstring operator+( return result; } +// C++11 21.4.8.1/8 +template +inline +basic_fbstring operator+( + E lhs, + basic_fbstring&& rhs) { + // + if (rhs.capacity() > rhs.size()) { + // Good, at least we don't need to reallocate + return std::move(rhs.insert(rhs.begin(), lhs)); + } + // Meh, no go. Forward to operator+(E, const&). + auto const& rhsC = rhs; + return lhs + rhsC; +} + +// C++11 21.4.8.1/9 template inline basic_fbstring operator+( const basic_fbstring& lhs, - const typename basic_fbstring::value_type* rhs) { + const E* rhs) { typedef typename basic_fbstring::size_type size_type; typedef typename basic_fbstring::traits_type traits_type; @@ -2036,11 +2150,22 @@ basic_fbstring operator+( return result; } +// C++11 21.4.8.1/10 +template +inline +basic_fbstring operator+( + basic_fbstring&& lhs, + const E* rhs) { + // + return std::move(lhs += rhs); +} + +// C++11 21.4.8.1/11 template inline basic_fbstring operator+( const basic_fbstring& lhs, - typename basic_fbstring::value_type rhs) { + E rhs) { basic_fbstring result; result.reserve(lhs.size() + 1); @@ -2049,11 +2174,21 @@ basic_fbstring operator+( return result; } +// C++11 21.4.8.1/12 +template +inline +basic_fbstring operator+( + basic_fbstring&& lhs, + E rhs) { + // + return std::move(lhs += rhs); +} + template inline bool operator==(const basic_fbstring& lhs, const basic_fbstring& rhs) { - return lhs.compare(rhs) == 0; } + return lhs.size() == rhs.size() && lhs.compare(rhs) == 0; } template inline @@ -2158,7 +2293,7 @@ bool operator>=(const typename basic_fbstring::value_type* lhs, return !(lhs < rhs); } -// subclause 21.3.7.8: +// C++11 21.4.8.8 template void swap(basic_fbstring& lhs, basic_fbstring& rhs) { lhs.swap(rhs); @@ -2183,20 +2318,20 @@ std::basic_istream< auto err = __ios_base::goodbit; if (sentry) { auto n = is.width(); - if (n == 0) { + if (n <= 0) { n = str.max_size(); } str.erase(); - auto got = is.rdbuf()->sgetc(); - for (; extracted != n && got != T::eof() && !isspace(got); ++extracted) { - // Whew. We get to store this guy + for (auto got = is.rdbuf()->sgetc(); extracted != size_t(n); ++extracted) { + if (got == T::eof()) { + err |= __ios_base::eofbit; + is.width(0); + break; + } + if (isspace(got)) break; str.push_back(got); got = is.rdbuf()->snextc(); } - if (got == T::eof()) { - err |= __ios_base::eofbit; - is.width(0); - } } if (!extracted) { err |= __ios_base::failbit; @@ -2215,7 +2350,29 @@ operator<<( std::basic_ostream::value_type, typename basic_fbstring::traits_type>& os, const basic_fbstring& str) { - os.write(str.data(), str.size()); +#if _LIBCPP_VERSION + typename std::basic_ostream< + typename basic_fbstring::value_type, + typename basic_fbstring::traits_type>::sentry __s(os); + if (__s) { + typedef std::ostreambuf_iterator< + typename basic_fbstring::value_type, + typename basic_fbstring::traits_type> _Ip; + size_t __len = str.size(); + bool __left = + (os.flags() & std::ios_base::adjustfield) == std::ios_base::left; + if (__pad_and_output(_Ip(os), + str.data(), + __left ? str.data() + __len : str.data(), + str.data() + __len, + os, + os.fill()).failed()) { + os.setstate(std::ios_base::badbit | std::ios_base::failbit); + } + } +#else + std::__ostream_insert(os, str.data(), str.size()); +#endif return os; } @@ -2231,7 +2388,7 @@ getline( basic_fbstring& str, typename basic_fbstring::value_type delim) { // Use the nonstandard getdelim() - char * buf = NULL; + char * buf = nullptr; size_t size = 0; for (;;) { // This looks quadratic but it really depends on realloc @@ -2322,17 +2479,51 @@ _GLIBCXX_END_NAMESPACE_VERSION #ifndef _LIBSTDCXX_FBSTRING +// Hash functions to make fbstring usable with e.g. hash_map +// +// Handle interaction with different C++ standard libraries, which +// expect these types to be in different namespaces. + +#define FOLLY_FBSTRING_HASH1(T) \ + template <> \ + struct hash< ::folly::basic_fbstring > { \ + size_t operator()(const ::folly::fbstring& s) const { \ + return ::folly::hash::fnv32_buf(s.data(), s.size()); \ + } \ + }; + +// The C++11 standard says that these four are defined +#define FOLLY_FBSTRING_HASH \ + FOLLY_FBSTRING_HASH1(char) \ + FOLLY_FBSTRING_HASH1(char16_t) \ + FOLLY_FBSTRING_HASH1(char32_t) \ + FOLLY_FBSTRING_HASH1(wchar_t) + namespace std { -template <> -struct hash< ::folly::fbstring> { - size_t operator()(const ::folly::fbstring& s) const { - return ::folly::hash::fnv32_buf(s.data(), s.size()); - } -}; -} + +FOLLY_FBSTRING_HASH + +} // namespace std + +#if FOLLY_HAVE_DEPRECATED_ASSOC +#if defined(_GLIBCXX_SYMVER) && !defined(__BIONIC__) +namespace __gnu_cxx { + +FOLLY_FBSTRING_HASH + +} // namespace __gnu_cxx +#endif // _GLIBCXX_SYMVER && !__BIONIC__ +#endif // FOLLY_HAVE_DEPRECATED_ASSOC + +#undef FOLLY_FBSTRING_HASH +#undef FOLLY_FBSTRING_HASH1 #endif // _LIBSTDCXX_FBSTRING +#pragma GCC diagnostic pop + +#undef FBSTRING_DISABLE_ADDRESS_SANITIZER +#undef throw #undef FBSTRING_LIKELY #undef FBSTRING_UNLIKELY