#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_
#define FBSTRING_UNLIKELY(x) (x)
#endif
-// Ignore shadowing warnings within this file, so includers can use -Wshadow.
#pragma GCC diagnostic push
+// Ignore shadowing warnings within this file, so includers can use -Wshadow.
#pragma GCC diagnostic ignored "-Wshadow"
+// GCC 4.9 has a false positive in setSmallSize (probably
+// https://gcc.gnu.org/bugzilla/show_bug.cgi?id=59124), disable
+// compile-time array bound checking.
+#pragma GCC diagnostic ignored "-Warray-bounds"
// FBString cannot use throw when replacing std::string, though it may still
// use std::__throw_*
# endif
# endif
#elif defined (__GNUC__) && \
- (__GNUC__ == 4) && \
- (__GNUC_MINOR__ >= 8) && \
__SANITIZE_ADDRESS__
# define FBSTRING_DISABLE_ADDRESS_SANITIZER \
__attribute__((__no_address_safety_analysis__, __noinline__))
void shrink(size_t delta);
// Expands the string by delta characters (i.e. after this call
// size() will report the old size() plus delta) but without
- // initializing the expanded region. Returns a pointer to the memory
- // to be initialized (the beginning of the expanded portion). The
- // caller is expected to fill the expanded area appropriately.
+ // initializing the expanded region. The expanded region is
+ // zero-terminated. Returns a pointer to the memory to be
+ // initialized (the beginning of the expanded portion). The caller
+ // is expected to fill the expanded area appropriately.
Char* expand_noinit(size_t delta);
// Expands the string by one character and sets the last character
// to c.
* to extract capacity/category.
*/
template <class Char> 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 { reset(); }
auto const allocSize =
goodMallocSize((1 + rhs.ml_.size_) * sizeof(Char));
ml_.data_ = static_cast<Char*>(checkedMalloc(allocSize));
+ // Also copies terminator.
fbstring_detail::pod_copy(rhs.ml_.data_,
- // 1 for terminator
rhs.ml_.data_ + rhs.ml_.size_ + 1,
ml_.data_);
- // No need for writeTerminator() here, we copied one extra
- // element just above.
ml_.size_ = rhs.ml_.size_;
ml_.setCapacity(allocSize / sizeof(Char) - 1, Category::isMedium);
assert(category() == Category::isMedium);
if (reinterpret_cast<size_t>(data) & (sizeof(size_t) - 1)) {
fbstring_detail::pod_copy(data, data + size, small_);
} else {
- // Copy one word (64 bits) at a time
+ // Copy one word at a time
const size_t byteSize = size * sizeof(Char);
- if (byteSize > 2 * sizeof(size_t)) {
- // Copy three words
+ constexpr size_t wordWidth = sizeof(size_t);
+ switch ((byteSize + wordWidth - 1) / wordWidth) { // Number of words.
+ case 3:
ml_.capacity_ = reinterpret_cast<const size_t*>(data)[2];
- copyTwo:
+ case 2:
ml_.size_ = reinterpret_cast<const size_t*>(data)[1];
- copyOne:
+ case 1:
ml_.data_ = *reinterpret_cast<Char**>(const_cast<Char*>(data));
- } else if (byteSize > sizeof(size_t)) {
- // Copy two words
- goto copyTwo;
- } else if (size > 0) {
- // Copy one word
- goto copyOne;
+ case 0:
+ break;
}
}
setSmallSize(size);
- return;
- } else if (size <= maxMediumSize) {
- // Medium strings are allocated normally. Don't forget to
- // allocate one extra Char for the terminating null.
- auto const allocSize = goodMallocSize((1 + size) * sizeof(Char));
- ml_.data_ = static_cast<Char*>(checkedMalloc(allocSize));
- fbstring_detail::pod_copy(data, data + size, ml_.data_);
- ml_.size_ = size;
- 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_.setCapacity(effectiveCapacity, Category::isLarge);
+ if (size <= maxMediumSize) {
+ // Medium strings are allocated normally. Don't forget to
+ // allocate one extra Char for the terminating null.
+ auto const allocSize = goodMallocSize((1 + size) * sizeof(Char));
+ ml_.data_ = static_cast<Char*>(checkedMalloc(allocSize));
+ fbstring_detail::pod_copy(data, data + size, ml_.data_);
+ ml_.size_ = size;
+ 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_.setCapacity(effectiveCapacity, Category::isLarge);
+ }
+ ml_.data_[size] = '\0';
}
- writeTerminator();
}
~fbstring_core() noexcept {
// If this fails, someone placed the wrong capacity in an
// fbstring.
assert(effectiveCapacity >= ml_.capacity());
+ // Also copies terminator.
fbstring_detail::pod_copy(ml_.data_, ml_.data_ + ml_.size_ + 1,
newRC->data_);
RefCounted::decrementRefs(ml_.data_);
ml_.data_ = newRC->data_;
- // No need to call writeTerminator(), we have + 1 above.
}
return ml_.data_;
}
// handling.
assert(ml_.size_ >= delta);
ml_.size_ -= delta;
- writeTerminator();
+ ml_.data_[ml_.size_] = '\0';
} else {
assert(ml_.size_ >= delta);
// Shared large string, must make unique. This is because of the
// call to reserve.
minCapacity = std::max(minCapacity, ml_.capacity());
auto const newRC = RefCounted::create(& minCapacity);
+ // Also copies terminator.
fbstring_detail::pod_copy(ml_.data_, ml_.data_ + ml_.size_ + 1,
newRC->data_);
- // Done with the old data. No need to call writeTerminator(),
- // we have + 1 above.
RefCounted::decrementRefs(ml_.data_);
ml_.data_ = newRC->data_;
ml_.setCapacity(minCapacity, Category::isLarge);
ml_.capacity(), minCapacity);
ml_.data_ = newRC->data_;
ml_.setCapacity(minCapacity, Category::isLarge);
- writeTerminator();
}
assert(capacity() >= minCapacity);
}
// Keep the string at medium size. Don't forget to allocate
// one extra Char for the terminating null.
size_t capacityBytes = goodMallocSize((1 + minCapacity) * sizeof(Char));
+ // Also copies terminator.
ml_.data_ = static_cast<Char *>(
smartRealloc(
ml_.data_,
- ml_.size_ * sizeof(Char),
+ (ml_.size_ + 1) * sizeof(Char),
(ml_.capacity() + 1) * sizeof(Char),
capacityBytes));
- writeTerminator();
ml_.setCapacity(capacityBytes / sizeof(Char) - 1, Category::isMedium);
} else {
// Conversion from medium to large string
// Will recurse to another branch of this function
nascent.reserve(minCapacity);
nascent.ml_.size_ = ml_.size_;
- fbstring_detail::pod_copy(ml_.data_, ml_.data_ + ml_.size_,
+ // Also copies terminator.
+ fbstring_detail::pod_copy(ml_.data_, ml_.data_ + ml_.size_ + 1,
nascent.ml_.data_);
nascent.swap(*this);
- writeTerminator();
assert(capacity() >= minCapacity);
}
} else {
// large
auto const newRC = RefCounted::create(& minCapacity);
auto const size = smallSize();
+ // Also copies terminator.
fbstring_detail::pod_copy(small_, small_ + size + 1, newRC->data_);
- // No need for writeTerminator(), we wrote it above with + 1.
ml_.data_ = newRC->data_;
ml_.size_ = size;
ml_.setCapacity(minCapacity, Category::isLarge);
// Don't forget to allocate one extra Char for the terminating null
auto const allocSizeBytes =
goodMallocSize((1 + minCapacity) * sizeof(Char));
- auto const data = static_cast<Char*>(checkedMalloc(allocSizeBytes));
+ auto const pData = static_cast<Char*>(checkedMalloc(allocSizeBytes));
auto const size = smallSize();
- fbstring_detail::pod_copy(small_, small_ + size + 1, data);
- // No need for writeTerminator(), we wrote it above with + 1.
- ml_.data_ = data;
+ // Also copies terminator.
+ fbstring_detail::pod_copy(small_, small_ + size + 1, pData);
+ ml_.data_ = pData;
ml_.size_ = size;
ml_.setCapacity(allocSizeBytes / sizeof(Char) - 1, Category::isMedium);
} else {
// Category can't be small - we took care of that above
assert(category() == Category::isMedium || category() == Category::isLarge);
ml_.size_ = newSz;
- writeTerminator();
+ ml_.data_[newSz] = '\0';
assert(size() == newSz);
return ml_.data_ + sz;
}
size_t sz;
if (category() == Category::isSmall) {
sz = smallSize();
- if (sz < maxSmallSize) {
+ if (FBSTRING_LIKELY(sz < maxSmallSize)) {
small_[sz] = c;
setSmallSize(sz + 1);
return;
assert(category() == Category::isMedium || category() == Category::isLarge);
ml_.size_ = sz + 1;
ml_.data_[sz] = c;
- writeTerminator();
+ ml_.data_[sz + 1] = '\0';
}
size_t size() const {
return category() == Category::isLarge && RefCounted::refs(ml_.data_) > 1;
}
- void writeTerminator() {
- if (category() == Category::isSmall) {
- const auto s = smallSize();
- if (s != maxSmallSize) {
- small_[s] = '\0';
- }
- } else {
- ml_.data_[ml_.size_] = '\0';
- }
- }
-
private:
// Disabled
fbstring_core & operator=(const fbstring_core & rhs);
- // Equivalent to setSmallSize(0), but with specialized
- // writeTerminator which doesn't re-check the category after
- // capacity_ is overwritten.
+ // Equivalent to setSmallSize(0) but a few ns faster in
+ // microbenchmarks.
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;
size_t smallSize() const {
assert(category() == Category::isSmall);
- auto shift = kIsLittleEndian ? 0 : 2;
+ constexpr auto shift = kIsLittleEndian ? 0 : 2;
auto smallShifted = static_cast<size_t>(small_[maxSmallSize]) >> shift;
assert(static_cast<size_t>(maxSmallSize) >= smallShifted);
return static_cast<size_t>(maxSmallSize) - smallShifted;
// so don't assume anything about the previous value of
// small_[maxSmallSize].
assert(s <= maxSmallSize);
- small_[maxSmallSize] = kIsLittleEndian
- ? maxSmallSize - s
- : (maxSmallSize - s) << 2;
- writeTerminator();
+ constexpr auto shift = kIsLittleEndian ? 0 : 2;
+ small_[maxSmallSize] = (maxSmallSize - s) << shift;
+ small_[s] = '\0';
+ assert(category() == Category::isSmall && size() == s);
}
};
}
#endif
- basic_fbstring(const basic_fbstring& str, size_type pos,
- size_type n = npos, const A& a = A()) {
+ basic_fbstring(const basic_fbstring& str,
+ size_type pos,
+ size_type n = npos,
+ const A& /* a */ = A()) {
assign(str, pos, n);
}
}
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();
+ auto const pData = store_.expand_noinit(n);
+ fbstring_detail::pod_fill(pData, pData + n, c);
}
template <class InIt>
}
basic_fbstring& operator=(const basic_fbstring& lhs) {
+ Invariant checker(*this);
+
if (FBSTRING_UNLIKELY(&lhs == this)) {
return *this;
}
auto const srcSize = lhs.size();
if (capacity() >= srcSize && !store_.isShared()) {
// great, just copy the contents
- if (oldSize < srcSize)
+ if (oldSize < srcSize) {
store_.expand_noinit(srcSize - oldSize);
- else
+ } else {
store_.shrink(oldSize - srcSize);
+ }
assert(size() == srcSize);
- fbstring_detail::pod_copy(lhs.begin(), lhs.end(), begin());
- store_.writeTerminator();
+ auto srcData = lhs.data();
+ fbstring_detail::pod_copy(
+ srcData, srcData + srcSize, store_.mutable_data());
} else {
// need to reallocate, so we may as well create a brand new string
basic_fbstring(lhs).swap(*this);
}
basic_fbstring& operator=(value_type c) {
+ Invariant checker(*this);
+
if (empty()) {
store_.expand_noinit(1);
} else if (store_.isShared()) {
} else {
store_.shrink(size() - 1);
}
- *store_.mutable_data() = c;
- store_.writeTerminator();
+ front() = c;
return *this;
}
}
void resize(const size_type n, const value_type c = value_type()) {
+ Invariant checker(*this);
+
auto size = this->size();
if (n <= size) {
store_.shrink(size - n);
} else {
- // Do this in two steps to minimize slack memory copied (see
- // smartRealloc).
- auto const capacity = this->capacity();
- assert(capacity >= size);
- if (size < capacity) {
- auto delta = std::min(n, capacity) - size;
- store_.expand_noinit(delta);
- fbstring_detail::pod_fill(begin() + size, end(), c);
- size += delta;
- if (size == n) {
- store_.writeTerminator();
- return;
- }
- assert(size < n);
- }
auto const delta = n - size;
- store_.expand_noinit(delta);
- fbstring_detail::pod_fill(end() - delta, end(), c);
- store_.writeTerminator();
+ auto pData = store_.expand_noinit(delta);
+ fbstring_detail::pod_fill(pData, pData + delta, c);
}
assert(this->size() == n);
}
}
basic_fbstring& append(const value_type* s, size_type n) {
-#ifndef NDEBUG
Invariant checker(*this);
- (void) checker;
-#endif
+
if (FBSTRING_UNLIKELY(!n)) {
// Unlikely but must be done
return *this;
basic_fbstring& assign(const value_type* s, const size_type n) {
Invariant checker(*this);
- (void) checker;
+
+ // s can alias this, we need to use pod_move.
if (size() >= n) {
- std::copy(s, s + n, begin());
+ fbstring_detail::pod_move(s, s + n, store_.mutable_data());
resize(n);
assert(size() == n);
} else {
const value_type *const s2 = s + size();
- std::copy(s, s2, begin());
+ fbstring_detail::pod_move(s, s2, store_.mutable_data());
append(s2, n - size());
assert(size() == n);
}
- store_.writeTerminator();
assert(size() == n);
return *this;
}
return begin() + pos;
}
+#ifndef _LIBSTDCXX_FBSTRING
+ private:
+ typedef std::basic_istream<value_type, traits_type> istream_type;
+
+ public:
+ friend inline istream_type& getline(istream_type& is,
+ basic_fbstring& str,
+ value_type delim) {
+ Invariant checker(str);
+
+ str.clear();
+ size_t size = 0;
+ while (true) {
+ size_t avail = str.capacity() - size;
+ // fbstring has 1 byte extra capacity for the null terminator,
+ // and getline null-terminates the read string.
+ is.getline(str.store_.expand_noinit(avail), avail + 1, delim);
+ size += is.gcount();
+
+ if (is.bad() || is.eof() || !is.fail()) {
+ // Done by either failure, end of file, or normal read.
+ if (!is.bad() && !is.eof()) {
+ --size; // gcount() also accounts for the delimiter.
+ }
+ str.resize(size);
+ break;
+ }
+
+ assert(size == str.size());
+ assert(size == str.capacity());
+ // Start at minimum allocation 63 + terminator = 64.
+ str.reserve(std::max<size_t>(63, 3 * size / 2));
+ // Clear the error so we can continue reading.
+ is.clear();
+ }
+ return is;
+ }
+
+ friend inline istream_type& getline(istream_type& is, basic_fbstring& str) {
+ return getline(is, str, '\n');
+ }
+#endif
+
private:
template <int i> class Selector {};
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 iterator oldEnd = end();
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,
- begin() + pos + n);
+ // Also copies terminator.
+ fbstring_detail::pod_move(&*p, &*oldEnd - n + 1, begin() + pos + n);
std::fill(begin() + pos, begin() + pos + n, c);
} else {
append(n - (end() - p), c);
append(iterator(p), oldEnd);
std::fill(iterator(p), oldEnd, c);
}
- store_.writeTerminator();
return begin() + pos;
}
iterator insertImpl(const_iterator i,
FwdIterator s1, FwdIterator s2, std::forward_iterator_tag) {
Invariant checker(*this);
- (void) checker;
+
const size_type pos = i - begin();
const typename std::iterator_traits<FwdIterator>::difference_type n2 =
std::distance(s1, s2);
begin() + old_size + newElems);
std::copy(s1, t, begin() + pos);
}
- store_.writeTerminator();
return begin() + pos;
}
basic_fbstring& erase(size_type pos = 0, size_type n = npos) {
Invariant checker(*this);
- (void) checker;
+
enforce(pos <= length(), std::__throw_out_of_range, "");
procrustes(n, length() - pos);
std::copy(begin() + pos + n, end(), begin() + pos);
basic_fbstring& replace(size_type pos, size_type n1,
StrOrLength s_or_n2, NumOrChar n_or_c) {
Invariant checker(*this);
- (void) checker;
+
enforce(pos <= size(), std::__throw_out_of_range, "");
procrustes(n1, length() - pos);
const iterator b = begin() + pos;
}
private:
- template <class FwdIterator>
- bool replaceAliased(iterator i1, iterator i2,
- FwdIterator s1, FwdIterator s2, std::false_type) {
+ template <class FwdIterator>
+ bool replaceAliased(iterator /* i1 */,
+ iterator /* i2 */,
+ FwdIterator /* s1 */,
+ FwdIterator /* s2 */,
+ std::false_type) {
return false;
}
void replaceImpl(iterator i1, iterator i2,
FwdIterator s1, FwdIterator s2, std::forward_iterator_tag) {
Invariant checker(*this);
- (void) checker;
// Handle aliased replace
if (replaceAliased(i1, i2, s1, s2,
const auto len = basic_fbstring<E, T, A, S>::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));
+ rhs.insert(rhs.begin(), lhs, lhs + len);
+ return rhs;
}
// Meh, no go. Do it by hand since we have len already.
basic_fbstring<E, T, A, S> result;
//
if (rhs.capacity() > rhs.size()) {
// Good, at least we don't need to reallocate
- return std::move(rhs.insert(rhs.begin(), lhs));
+ rhs.insert(rhs.begin(), lhs);
+ return rhs;
}
// Meh, no go. Forward to operator+(E, const&).
auto const& rhsC = rhs;
return os;
}
-#ifndef _LIBSTDCXX_FBSTRING
-
-template <typename E, class T, class A, class S>
-inline
-std::basic_istream<typename basic_fbstring<E, T, A, S>::value_type,
- typename basic_fbstring<E, T, A, S>::traits_type>&
-getline(
- std::basic_istream<typename basic_fbstring<E, T, A, S>::value_type,
- typename basic_fbstring<E, T, A, S>::traits_type>& is,
- basic_fbstring<E, T, A, S>& str,
- typename basic_fbstring<E, T, A, S>::value_type delim) {
- // Use the nonstandard getdelim()
- char * buf = nullptr;
- size_t size = 0;
- for (;;) {
- // This looks quadratic but it really depends on realloc
- auto const newSize = size + 128;
- buf = static_cast<char*>(checkedRealloc(buf, newSize));
- is.getline(buf + size, newSize - size, delim);
- if (is.bad() || is.eof() || !is.fail()) {
- // done by either failure, end of file, or normal read
- size += std::strlen(buf + size);
- break;
- }
- // Here we have failed due to too short a buffer
- // Minus one to discount the terminating '\0'
- size = newSize - 1;
- assert(buf[size] == 0);
- // Clear the error so we can continue reading
- is.clear();
- }
- basic_fbstring<E, T, A, S> result(buf, size, size + 1,
- AcquireMallocatedString());
- result.swap(str);
- return is;
-}
-
-template <typename E, class T, class A, class S>
-inline
-std::basic_istream<typename basic_fbstring<E, T, A, S>::value_type,
- typename basic_fbstring<E, T, A, S>::traits_type>&
-getline(
- std::basic_istream<typename basic_fbstring<E, T, A, S>::value_type,
- typename basic_fbstring<E, T, A, S>::traits_type>& is,
- basic_fbstring<E, T, A, S>& str) {
- // Just forward to the version with a delimiter
- return getline(is, str, '\n');
-}
-
-#endif
-
template <typename E1, class T, class A, class S>
const typename basic_fbstring<E1, T, A, S>::size_type
basic_fbstring<E1, T, A, S>::npos =
#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_
projects / folly.git / blobdiff