X-Git-Url: http://plrg.eecs.uci.edu/git/?a=blobdiff_plain;f=folly%2FFBString.h;h=f7b58ccc24ab9aaf9c39700e48375fe1577cb90d;hb=17d04308e64ee7a11ad68f4b4b4c03498c3c8844;hp=b9d45a98dc21aa0d6d6b23a1accd3b7444f31146;hpb=52fceaf485e63d459d2df5858ffdbeaafe851dbf;p=folly.git diff --git a/folly/FBString.h b/folly/FBString.h index b9d45a98..f7b58ccc 100644 --- a/folly/FBString.h +++ b/folly/FBString.h @@ -1,5 +1,5 @@ /* - * Copyright 2014 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. @@ -20,94 +20,71 @@ #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 -// libc++ doesn't provide this header -#ifndef _LIBCPP_VERSION // 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 -#endif - +// 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 +// When used as std::string replacement always disable assertions. +#ifndef NDEBUG +#define NDEBUG +#define FOLLY_DEFINED_NDEBUG_FOR_FBSTRING +#endif // NDEBUG + // Handle the cases where the fbcode version (folly/Malloc.h) is included // 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 "folly/Traits.h" -#include "folly/Malloc.h" -#include "folly/Hash.h" +#include +#include +#include +#include +#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 // Ignore shadowing warnings within this file, so includers can use -Wshadow. #pragma GCC diagnostic push @@ -115,6 +92,7 @@ // 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 @@ -129,7 +107,10 @@ namespace folly { // 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__) +# if __has_attribute(__no_sanitize__) +# define FBSTRING_DISABLE_ADDRESS_SANITIZER \ + __attribute__((__no_sanitize__("address"), __noinline__)) +# elif __has_attribute(__no_address_safety_analysis__) # define FBSTRING_DISABLE_ADDRESS_SANITIZER \ __attribute__((__no_address_safety_analysis__, __noinline__)) # elif __has_attribute(__no_sanitize_address__) @@ -283,19 +264,10 @@ private: }; */ -/** - * gcc-4.7 throws what appears to be some false positive uninitialized - * warnings for the members of the MediumLarge struct. So, mute them here. - */ -#if defined(__GNUC__) && !defined(__clang__) -# pragma GCC diagnostic push -# pragma GCC diagnostic ignored "-Wuninitialized" -#endif - /** * This is the core of the string. The code should work on 32- and - * 64-bit architectures and with any Char size. Porting to big endian - * architectures would require some changes. + * 64-bit and both big- and little-endianan architectures with any + * Char size. * * The storage is selected as follows (assuming we store one-byte * characters on a 64-bit machine): (a) "small" strings between 0 and @@ -307,49 +279,50 @@ private: * reference-counted and copied lazily. the reference count is * allocated right before the character array. * - * The discriminator between these three strategies sits in the two - * most significant bits of the rightmost char of the storage. If - * neither is set, then the string is small (and its length sits in - * the lower-order bits of that rightmost character). If the MSb is - * set, the string is medium width. If the second MSb is set, then the - * string is large. + * The discriminator between these three strategies sits in two + * bits of the rightmost char of the storage. If neither is set, then the + * string is small (and its length sits in the lower-order bits on + * little-endian or the high-order bits on big-endian of that + * rightmost character). If the MSb is set, the string is medium width. + * If the second MSb is set, then the string is large. On little-endian, + * these 2 bits are the 2 MSbs of MediumLarge::capacity_, while on + * big-endian, these 2 bits are the 2 LSbs. This keeps both little-endian + * and big-endian fbstring_core equivalent with merely different ops used + * to extract capacity/category. */ template class fbstring_core { +protected: + static constexpr bool kIsLittleEndian = + __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__; + static constexpr bool kIsBigEndian = + __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__; + static_assert( + kIsLittleEndian || kIsBigEndian, "unable to identify endianness"); public: - 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); - } + fbstring_core() noexcept { reset(); } 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_)); - const size_t size = rhs.smallSize(); - if (size == 0) { - ml_.capacity_ = rhs.ml_.capacity_; - writeTerminator(); - } else { - // Just write the whole thing, don't look at details. In - // particular we need to copy capacity anyway because we want - // to set the size (don't forget that the last character, - // which stores a short string's length, is shared with the - // ml_.capacity field). - ml_ = rhs.ml_; - } - assert(category() == isSmall && this->size() == rhs.size()); - } else if (rhs.category() == isLarge) { + 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"); + // Just write the whole thing, don't look at details. In + // particular we need to copy capacity anyway because we want + // to set the size (don't forget that the last character, + // which stores a short string's length, is shared with the + // ml_.capacity field). + ml_ = rhs.ml_; + 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. @@ -363,22 +336,19 @@ 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_.setCapacity(allocSize / sizeof(Char) - 1, Category::isMedium); + assert(category() == Category::isMedium); } assert(size() == rhs.size()); assert(memcmp(data(), rhs.data(), size() * sizeof(Char)) == 0); } fbstring_core(fbstring_core&& goner) noexcept { - if (goner.category() == isSmall) { - // Just copy, leave the goner in peace - new(this) fbstring_core(goner.small_, goner.smallSize()); - } else { - // Take goner's guts - ml_ = goner.ml_; + // Take goner's guts + ml_ = goner.ml_; + if (goner.category() != Category::isSmall) { // Clean goner's carcass - goner.setSmallSize(0); + goner.reset(); } } @@ -387,13 +357,25 @@ public: // 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. @@ -418,6 +400,7 @@ 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. @@ -425,26 +408,24 @@ 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_.setCapacity(allocSize / sizeof(Char) - 1, 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_.setCapacity(effectiveCapacity, Category::isLarge); } writeTerminator(); - assert(this->size() == size); - assert(memcmp(this->data(), data, size * sizeof(Char)) == 0); } ~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; } @@ -469,11 +450,11 @@ public: ml_.data_ = data; ml_.size_ = size; // Don't forget about null terminator - ml_.capacity_ = (allocatedSize - 1) | isMedium; + ml_.setCapacity(allocatedSize - 1, Category::isMedium); } else { // No need for the memory free(data); - setSmallSize(0); + reset(); } } @@ -494,11 +475,11 @@ 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); @@ -516,44 +497,27 @@ 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 @@ -563,13 +527,11 @@ 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 @@ -585,7 +547,7 @@ public: // we have + 1 above. RefCounted::decrementRefs(ml_.data_); ml_.data_ = newRC->data_; - ml_.capacity_ = minCapacity | isLarge; + ml_.setCapacity(minCapacity, Category::isLarge); // size remains unchanged } else { // String is not shared, so let's try to realloc (if needed) @@ -595,12 +557,12 @@ public: RefCounted::reallocate(ml_.data_, ml_.size_, ml_.capacity(), minCapacity); ml_.data_ = newRC->data_; - ml_.capacity_ = minCapacity | isLarge; + ml_.setCapacity(minCapacity, 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 @@ -616,7 +578,7 @@ public: (ml_.capacity() + 1) * sizeof(Char), capacityBytes)); writeTerminator(); - ml_.capacity_ = (capacityBytes / sizeof(Char) - 1) | isMedium; + ml_.setCapacity(capacityBytes / sizeof(Char) - 1, Category::isMedium); } else { // Conversion from medium to large string fbstring_core nascent; @@ -630,7 +592,7 @@ public: assert(capacity() >= minCapacity); } } else { - assert(category() == isSmall); + assert(category() == Category::isSmall); if (minCapacity > maxMediumSize) { // large auto const newRC = RefCounted::create(& minCapacity); @@ -639,7 +601,7 @@ public: // No need for writeTerminator(), we wrote it above with + 1. ml_.data_ = newRC->data_; ml_.size_ = size; - ml_.capacity_ = minCapacity | isLarge; + ml_.setCapacity(minCapacity, Category::isLarge); assert(capacity() >= minCapacity); } else if (minCapacity > maxSmallSize) { // medium @@ -652,7 +614,7 @@ 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_.setCapacity(allocSizeBytes / sizeof(Char) - 1, Category::isMedium); } else { // small // Nothing to do, everything stays put @@ -665,12 +627,11 @@ 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); @@ -683,7 +644,7 @@ 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); @@ -693,12 +654,11 @@ 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); @@ -711,21 +671,21 @@ public: 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. @@ -736,41 +696,36 @@ 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: // Disabled fbstring_core & operator=(const fbstring_core & rhs); - struct MediumLarge { - Char * data_; - size_t size_; - size_t capacity_; - - size_t capacity() const { - return capacity_ & capacityExtractMask; - } - }; + // Equivalent to setSmallSize(0), but with specialized + // writeTerminator which doesn't re-check the category after + // capacity_ is overwritten. + void reset() { + // Only initialize the tag, will set the MSBs (i.e. the small + // string size) to zero too. + ml_.capacity_ = kIsLittleEndian + ? maxSmallSize << (8 * (sizeof(size_t) - sizeof(Char))) + : maxSmallSize << 2; + small_[0] = '\0'; + assert(category() == Category::isSmall && size() == 0); + } struct RefCounted { std::atomic refCount_; @@ -839,39 +794,68 @@ private: } }; + typedef std::conditional::type + category_type; + + enum class Category : category_type { + isSmall = 0, + isMedium = kIsLittleEndian + ? sizeof(size_t) == 4 ? 0x80000000 : 0x8000000000000000 + : 0x2, + isLarge = kIsLittleEndian + ? sizeof(size_t) == 4 ? 0x40000000 : 0x4000000000000000 + : 0x1, + }; + + Category category() const { + // works for both big-endian and little-endian + return static_cast(ml_.capacity_ & categoryExtractMask); + } + + struct MediumLarge { + Char * data_; + size_t size_; + size_t capacity_; + + size_t capacity() const { + return kIsLittleEndian + ? capacity_ & capacityExtractMask + : capacity_ >> 2; + } + + void setCapacity(size_t cap, Category cat) { + capacity_ = kIsLittleEndian + ? cap | static_cast(cat) + : (cap << 2) | static_cast(cat); + } + }; + union { - mutable Char small_[sizeof(MediumLarge) / sizeof(Char)]; - mutable MediumLarge ml_; + Char small_[sizeof(MediumLarge) / sizeof(Char)]; + MediumLarge ml_; }; - enum { + enum : size_t { lastChar = sizeof(MediumLarge) - 1, maxSmallSize = lastChar / sizeof(Char), maxMediumSize = 254 / sizeof(Char), // coincides with the small // bin size in dlmalloc - categoryExtractMask = sizeof(size_t) == 4 ? 0xC0000000 : 0xC000000000000000, - capacityExtractMask = ~categoryExtractMask, + categoryExtractMask = kIsLittleEndian + ? sizeof(size_t) == 4 ? 0xC0000000 : 0xC000000000000000 + : 0x3, + capacityExtractMask = kIsLittleEndian + ? ~categoryExtractMask + : 0x0 /*unused*/, }; static_assert(!(sizeof(MediumLarge) % sizeof(Char)), "Corrupt memory layout for fbstring."); - enum Category { - isSmall = 0, - isMedium = sizeof(size_t) == 4 ? 0x80000000 : 0x8000000000000000, - isLarge = sizeof(size_t) == 4 ? 0x40000000 : 0x4000000000000000, - }; - - Category category() const { - // Assumes little endian - return static_cast(ml_.capacity_ & categoryExtractMask); - } - size_t smallSize() const { - assert(category() == isSmall && - static_cast(small_[maxSmallSize]) - <= static_cast(maxSmallSize)); - return static_cast(maxSmallSize) - - static_cast(small_[maxSmallSize]); + assert(category() == Category::isSmall); + auto shift = kIsLittleEndian ? 0 : 2; + auto smallShifted = static_cast(small_[maxSmallSize]) >> shift; + assert(static_cast(maxSmallSize) >= smallShifted); + return static_cast(maxSmallSize) - smallShifted; } void setSmallSize(size_t s) { @@ -879,14 +863,13 @@ private: // so don't assume anything about the previous value of // small_[maxSmallSize]. assert(s <= maxSmallSize); - small_[maxSmallSize] = maxSmallSize - s; + small_[maxSmallSize] = kIsLittleEndian + ? maxSmallSize - s + : (maxSmallSize - s) << 2; + writeTerminator(); } }; -#if defined(__GNUC__) && !defined(__clang__) -# pragma GCC diagnostic pop -#endif - #ifndef _LIBSTDCXX_FBSTRING /** * Dummy fbstring core that uses an actual std::string. This doesn't @@ -973,7 +956,7 @@ class basic_fbstring { size() <= max_size() && capacity() <= max_size() && size() <= capacity() && - (begin()[size()] == Storage::TERMINATOR || begin()[size()] == '\0'); + begin()[size()] == '\0'; } struct Invariant; @@ -1029,7 +1012,23 @@ private: public: // C++11 21.4.2 construct/copy/destroy - explicit basic_fbstring(const A& a = A()) noexcept { + + // Note: while the following two constructors can be (and previously were) + // collapsed into one constructor written this way: + // + // explicit basic_fbstring(const A& a = A()) noexcept { } + // + // This can cause Clang (at least version 3.7) to fail with the error: + // "chosen constructor is explicit in copy-initialization ... + // in implicit initialization of field '(x)' with omitted initializer" + // + // if used in a struct which is default-initialized. Hence the split into + // these two separate constructors. + + basic_fbstring() noexcept : basic_fbstring(A()) { + } + + explicit basic_fbstring(const A&) noexcept { } basic_fbstring(const basic_fbstring& str) @@ -1053,20 +1052,19 @@ public: assign(str, pos, n); } - /* 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; - })) { + /* implicit */ basic_fbstring(const value_type* s, const A& /*a*/ = A()) + : 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()) + basic_fbstring(const value_type* s, size_type n, const A& /*a*/ = A()) : store_(s, n) { } - basic_fbstring(size_type n, value_type c, const A& a = A()) { + basic_fbstring(size_type n, value_type c, const A& /*a*/ = A()) { auto const data = store_.expand_noinit(n); fbstring_detail::pod_fill(data, data + n, c); store_.writeTerminator(); @@ -1076,7 +1074,7 @@ public: basic_fbstring(InIt begin, InIt end, typename std::enable_if< !std::is_same::type, - value_type*>::value, const A>::type & a = A()) { + value_type*>::value, const A>::type & /*a*/ = A()) { assign(begin, end); } @@ -1288,14 +1286,10 @@ public: // 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); } @@ -2004,11 +1998,18 @@ 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); @@ -2056,7 +2057,7 @@ 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, @@ -2098,24 +2099,44 @@ 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; @@ -2125,11 +2146,28 @@ 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; @@ -2141,11 +2179,22 @@ 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); @@ -2154,6 +2203,16 @@ 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, @@ -2288,20 +2347,20 @@ 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; @@ -2340,6 +2399,9 @@ operator<<( os.setstate(std::ios_base::badbit | std::ios_base::failbit); } } +#elif defined(_MSC_VER) + // MSVC doesn't define __ostream_insert + os.write(str.data(), str.size()); #else std::__ostream_insert(os, str.data(), str.size()); #endif @@ -2453,43 +2515,41 @@ _GLIBCXX_END_NAMESPACE_VERSION // // Handle interaction with different C++ standard libraries, which // expect these types to be in different namespaces. -namespace std { -template -struct hash > : private hash { - size_t operator()(const folly::basic_fbstring & s) const { - return hash::operator()(s.c_str()); - } -}; +#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()); \ + } \ + }; -template <> -struct hash< ::folly::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 { + +FOLLY_FBSTRING_HASH +} // namespace std + +#if FOLLY_HAVE_DEPRECATED_ASSOC #if defined(_GLIBCXX_SYMVER) && !defined(__BIONIC__) namespace __gnu_cxx { -template -struct hash > : private hash { - size_t operator()(const folly::basic_fbstring & s) const { - return hash::operator()(s.c_str()); - } -}; - -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 __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 @@ -2499,4 +2559,9 @@ struct hash< ::folly::fbstring> { #undef FBSTRING_LIKELY #undef FBSTRING_UNLIKELY +#ifdef FOLLY_DEFINED_NDEBUG_FOR_FBSTRING +#undef NDEBUG +#undef FOLLY_DEFINED_NDEBUG_FOR_FBSTRING +#endif // FOLLY_DEFINED_NDEBUG_FOR_FBSTRING + #endif // FOLLY_BASE_FBSTRING_H_