/*
- * Copyright 2014 Facebook, Inc.
+ * Copyright 2016 Facebook, Inc.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* @author Andrei Alexandrescu (andrei.alexandrescu@fb.com)
*/
-#ifndef FOLLY_BASE_CONV_H_
-#define FOLLY_BASE_CONV_H_
+#pragma once
-#include "folly/FBString.h"
-#include "folly/Likely.h"
-#include "folly/Preprocessor.h"
-#include "folly/Range.h"
-
-#include <boost/implicit_cast.hpp>
-#include <type_traits>
+#include <algorithm>
+#include <cctype>
+#include <climits>
+#include <cstddef>
#include <limits>
+#include <stdexcept>
#include <string>
#include <tuple>
-#include <stdexcept>
+#include <type_traits>
#include <typeinfo>
+#include <utility>
-#include <limits.h>
-
-// V8 JavaScript implementation
-#include <double-conversion/double-conversion.h>
+#include <boost/implicit_cast.hpp>
+#include <double-conversion/double-conversion.h> // V8 JavaScript implementation
-#define FOLLY_RANGE_CHECK(condition, message) \
- ((condition) ? (void)0 : throw std::range_error( \
- (__FILE__ "(" + std::to_string((long long int) __LINE__) + "): " \
- + (message)).c_str()))
+#include <folly/Demangle.h>
+#include <folly/FBString.h>
+#include <folly/Likely.h>
+#include <folly/Range.h>
+#include <folly/portability/Math.h>
namespace folly {
-/*******************************************************************************
- * Integral to integral
- ******************************************************************************/
+class ConversionError : public std::range_error {
+ public:
+ // Keep this in sync with kErrorStrings in Conv.cpp
+ enum Code {
+ SUCCESS,
+ EMPTY_INPUT_STRING,
+ NO_DIGITS,
+ BOOL_OVERFLOW,
+ BOOL_INVALID_VALUE,
+ NON_DIGIT_CHAR,
+ INVALID_LEADING_CHAR,
+ POSITIVE_OVERFLOW,
+ NEGATIVE_OVERFLOW,
+ STRING_TO_FLOAT_ERROR,
+ NON_WHITESPACE_AFTER_END,
+ ARITH_POSITIVE_OVERFLOW,
+ ARITH_NEGATIVE_OVERFLOW,
+ ARITH_LOSS_OF_PRECISION,
+ NUM_ERROR_CODES, // has to be the last entry
+ };
+
+ ConversionError(const std::string& str, Code code)
+ : std::range_error(str), code_(code) {}
+
+ ConversionError(const char* str, Code code)
+ : std::range_error(str), code_(code) {}
+
+ Code errorCode() const { return code_; }
+
+ private:
+ Code code_;
+};
+
+namespace detail {
+
+ConversionError makeConversionError(
+ ConversionError::Code code,
+ const char* input,
+ size_t inputLen);
+
+inline ConversionError makeConversionError(
+ ConversionError::Code code,
+ const std::string& str) {
+ return makeConversionError(code, str.data(), str.size());
+}
+
+inline ConversionError makeConversionError(
+ ConversionError::Code code,
+ StringPiece sp) {
+ return makeConversionError(code, sp.data(), sp.size());
+}
/**
- * Checked conversion from integral to integral. The checks are only
- * performed when meaningful, e.g. conversion from int to long goes
- * unchecked.
+ * Enforce that the suffix following a number is made up only of whitespace.
*/
-template <class Tgt, class Src>
-typename std::enable_if<
- std::is_integral<Src>::value && std::is_integral<Tgt>::value,
- Tgt>::type
-to(const Src & value) {
- /* static */ if (std::numeric_limits<Tgt>::max()
- < std::numeric_limits<Src>::max()) {
- FOLLY_RANGE_CHECK(
- (!greater_than<Tgt, std::numeric_limits<Tgt>::max()>(value)),
- "Overflow"
- );
+inline ConversionError::Code enforceWhitespaceErr(StringPiece sp) {
+ for (auto c : sp) {
+ if (!std::isspace(c)) {
+ return ConversionError::NON_WHITESPACE_AFTER_END;
+ }
}
- /* static */ if (std::is_signed<Src>::value &&
- (!std::is_signed<Tgt>::value || sizeof(Src) > sizeof(Tgt))) {
- FOLLY_RANGE_CHECK(
- (!less_than<Tgt, std::numeric_limits<Tgt>::min()>(value)),
- "Negative overflow"
- );
+ return ConversionError::SUCCESS;
+}
+
+/**
+ * Keep this implementation around for prettyToDouble().
+ */
+inline void enforceWhitespace(StringPiece sp) {
+ auto err = enforceWhitespaceErr(sp);
+ if (err != ConversionError::SUCCESS) {
+ throw detail::makeConversionError(err, sp);
}
- return static_cast<Tgt>(value);
}
-/*******************************************************************************
- * Floating point to floating point
- ******************************************************************************/
+/**
+ * A simple std::pair-like wrapper to wrap both a value and an error
+ */
+template <typename T>
+struct ConversionResult {
+ explicit ConversionResult(T v) : value(v) {}
+ explicit ConversionResult(ConversionError::Code e) : error(e) {}
+
+ bool success() const {
+ return error == ConversionError::SUCCESS;
+ }
+
+ T value;
+ ConversionError::Code error{ConversionError::SUCCESS};
+};
+}
+/**
+ * The identity conversion function.
+ * to<T>(T) returns itself for all types T.
+ */
template <class Tgt, class Src>
-typename std::enable_if<
- std::is_floating_point<Tgt>::value && std::is_floating_point<Src>::value,
- Tgt>::type
+typename std::enable_if<std::is_same<Tgt, Src>::value, Tgt>::type
to(const Src & value) {
- /* static */ if (std::numeric_limits<Tgt>::max() <
- std::numeric_limits<Src>::max()) {
- FOLLY_RANGE_CHECK(value <= std::numeric_limits<Tgt>::max(),
- "Overflow");
- FOLLY_RANGE_CHECK(value >= -std::numeric_limits<Tgt>::max(),
- "Negative overflow");
- }
- return boost::implicit_cast<Tgt>(value);
+ return value;
+}
+
+template <class Tgt, class Src>
+typename std::enable_if<std::is_same<Tgt, Src>::value, Tgt>::type
+to(Src && value) {
+ return std::forward<Src>(value);
}
/*******************************************************************************
typename std::tuple_element<
sizeof...(Ts),
std::tuple<T, Ts...> >::type const&
- getLastElement(const T& v, const Ts&... vs) {
+ getLastElement(const T&, const Ts&... vs) {
return getLastElement(vs...);
}
template <typename IntegerType>
constexpr unsigned int
digitsEnough() {
- return ceil((double(sizeof(IntegerType) * CHAR_BIT) * M_LN2) / M_LN10);
+ return (unsigned int)(ceil(sizeof(IntegerType) * CHAR_BIT * M_LN2 / M_LN10));
}
-inline unsigned int
-unsafeTelescope128(char * buffer, unsigned int room, unsigned __int128 x) {
+inline size_t
+unsafeTelescope128(char * buffer, size_t room, unsigned __int128 x) {
typedef unsigned __int128 Usrc;
- unsigned int p = room - 1;
+ size_t p = room - 1;
while (x >= (Usrc(1) << 64)) { // Using 128-bit division while needed
const auto y = x / 10;
*/
inline uint32_t digits10(uint64_t v) {
+#ifdef __x86_64__
+
+ // For this arch we can get a little help from specialized CPU instructions
+ // which can count leading zeroes; 64 minus that is appx. log (base 2).
+ // Use that to approximate base-10 digits (log_10) and then adjust if needed.
+
+ // 10^i, defined for i 0 through 19.
+ // This is 20 * 8 == 160 bytes, which fits neatly into 5 cache lines
+ // (assuming a cache line size of 64).
+ static const uint64_t powersOf10[20] FOLLY_ALIGNED(64) = {
+ 1,
+ 10,
+ 100,
+ 1000,
+ 10000,
+ 100000,
+ 1000000,
+ 10000000,
+ 100000000,
+ 1000000000,
+ 10000000000,
+ 100000000000,
+ 1000000000000,
+ 10000000000000,
+ 100000000000000,
+ 1000000000000000,
+ 10000000000000000,
+ 100000000000000000,
+ 1000000000000000000,
+ 10000000000000000000UL,
+ };
+
+ // "count leading zeroes" operation not valid; for 0; special case this.
+ if UNLIKELY (! v) {
+ return 1;
+ }
+
+ // bits is in the ballpark of log_2(v).
+ const uint8_t leadingZeroes = __builtin_clzll(v);
+ const auto bits = 63 - leadingZeroes;
+
+ // approximate log_10(v) == log_10(2) * bits.
+ // Integer magic below: 77/256 is appx. 0.3010 (log_10(2)).
+ // The +1 is to make this the ceiling of the log_10 estimate.
+ const uint32_t minLength = 1 + ((bits * 77) >> 8);
+
+ // return that log_10 lower bound, plus adjust if input >= 10^(that bound)
+ // in case there's a small error and we misjudged length.
+ return minLength + (uint32_t) (UNLIKELY (v >= powersOf10[minLength]));
+
+#else
+
uint32_t result = 1;
for (;;) {
if (LIKELY(v < 10)) return result;
v /= 10000U;
result += 4;
}
+
+#endif
}
/**
*result += value;
}
-/**
- * Ubiquitous helper template for writing string appenders
- */
-template <class T> struct IsSomeString {
- enum { value = std::is_same<T, std::string>::value
- || std::is_same<T, fbstring>::value };
-};
+template<class T>
+constexpr typename std::enable_if<
+ std::is_same<T, char>::value,
+ size_t>::type
+estimateSpaceNeeded(T) {
+ return 1;
+}
/**
* Everything implicitly convertible to const char* gets appended.
}
}
+template<class Src>
+typename std::enable_if<
+ std::is_convertible<Src, const char*>::value,
+ size_t>::type
+estimateSpaceNeeded(Src value) {
+ const char *c = value;
+ if (c) {
+ return folly::StringPiece(value).size();
+ };
+ return 0;
+}
+
+template<class Src>
+typename std::enable_if<
+ (std::is_convertible<Src, folly::StringPiece>::value ||
+ IsSomeString<Src>::value) &&
+ !std::is_convertible<Src, const char*>::value,
+ size_t>::type
+estimateSpaceNeeded(Src value) {
+ return folly::StringPiece(value).size();
+}
+
+template <>
+inline size_t estimateSpaceNeeded(std::nullptr_t /* value */) {
+ return 0;
+}
+
+template<class Src>
+typename std::enable_if<
+ std::is_pointer<Src>::value &&
+ IsSomeString<std::remove_pointer<Src>>::value,
+ size_t>::type
+estimateSpaceNeeded(Src value) {
+ return value->size();
+}
+
/**
* Strings get appended, too.
*/
toAppend(__int128 value, Tgt * result) {
typedef unsigned __int128 Usrc;
char buffer[detail::digitsEnough<unsigned __int128>() + 1];
- unsigned int p;
+ size_t p;
if (value < 0) {
- p = detail::unsafeTelescope128(buffer, sizeof(buffer), Usrc(-value));
+ p = detail::unsafeTelescope128(buffer, sizeof(buffer), -Usrc(value));
buffer[--p] = '-';
} else {
p = detail::unsafeTelescope128(buffer, sizeof(buffer), value);
void
toAppend(unsigned __int128 value, Tgt * result) {
char buffer[detail::digitsEnough<unsigned __int128>()];
- unsigned int p;
+ size_t p;
p = detail::unsafeTelescope128(buffer, sizeof(buffer), value);
result->append(buffer + p, buffer + sizeof(buffer));
}
+template<class T>
+constexpr typename std::enable_if<
+ std::is_same<T, __int128>::value,
+ size_t>::type
+estimateSpaceNeeded(T) {
+ return detail::digitsEnough<__int128>();
+}
+
+template<class T>
+constexpr typename std::enable_if<
+ std::is_same<T, unsigned __int128>::value,
+ size_t>::type
+estimateSpaceNeeded(T) {
+ return detail::digitsEnough<unsigned __int128>();
+}
+
#endif
/**
}
}
+template <class Src>
+typename std::enable_if<
+ std::is_integral<Src>::value && std::is_signed<Src>::value
+ && sizeof(Src) >= 4 && sizeof(Src) < 16,
+ size_t>::type
+estimateSpaceNeeded(Src value) {
+ if (value < 0) {
+ // When "value" is the smallest negative, negating it would evoke
+ // undefined behavior, so, instead of writing "-value" below, we write
+ // "~static_cast<uint64_t>(value) + 1"
+ return 1 + digits10(~static_cast<uint64_t>(value) + 1);
+ }
+
+ return digits10(static_cast<uint64_t>(value));
+}
+
/**
* As above, but for uint32_t and uint64_t.
*/
&& IsSomeString<Tgt>::value && sizeof(Src) >= 4>::type
toAppend(Src value, Tgt * result) {
char buffer[20];
- result->append(buffer, buffer + uint64ToBufferUnsafe(value, buffer));
+ result->append(buffer, uint64ToBufferUnsafe(value, buffer));
+}
+
+template <class Src>
+typename std::enable_if<
+ std::is_integral<Src>::value && !std::is_signed<Src>::value
+ && sizeof(Src) >= 4 && sizeof(Src) < 16,
+ size_t>::type
+estimateSpaceNeeded(Src value) {
+ return digits10(value);
}
/**
toAppend<Tgt>(static_cast<Intermediate>(value), result);
}
-#if defined(__clang__) || __GNUC_PREREQ(4, 7)
-// std::underlying_type became available by gcc 4.7.0
+template <class Src>
+typename std::enable_if<
+ std::is_integral<Src>::value
+ && sizeof(Src) < 4
+ && !std::is_same<Src, char>::value,
+ size_t>::type
+estimateSpaceNeeded(Src value) {
+ typedef typename
+ std::conditional<std::is_signed<Src>::value, int64_t, uint64_t>::type
+ Intermediate;
+ return estimateSpaceNeeded(static_cast<Intermediate>(value));
+}
/**
* Enumerated values get appended as integers.
static_cast<typename std::underlying_type<Src>::type>(value), result);
}
-#else
-
-/**
- * Enumerated values get appended as integers.
- */
-template <class Tgt, class Src>
+template <class Src>
typename std::enable_if<
- std::is_enum<Src>::value && IsSomeString<Tgt>::value>::type
-toAppend(Src value, Tgt * result) {
- /* static */ if (Src(-1) < 0) {
- /* static */ if (sizeof(Src) <= sizeof(int)) {
- toAppend(static_cast<int>(value), result);
- } else {
- toAppend(static_cast<long>(value), result);
- }
- } else {
- /* static */ if (sizeof(Src) <= sizeof(int)) {
- toAppend(static_cast<unsigned int>(value), result);
- } else {
- toAppend(static_cast<unsigned long>(value), result);
- }
- }
+ std::is_enum<Src>::value, size_t>::type
+estimateSpaceNeeded(Src value) {
+ return estimateSpaceNeeded(
+ static_cast<typename std::underlying_type<Src>::type>(value));
}
-#endif // gcc 4.7 onwards
-
/*******************************************************************************
* Conversions from floating-point types to string types.
******************************************************************************/
+namespace detail {
+constexpr int kConvMaxDecimalInShortestLow = -6;
+constexpr int kConvMaxDecimalInShortestHigh = 21;
+} // folly::detail
+
/** Wrapper around DoubleToStringConverter **/
template <class Tgt, class Src>
typename std::enable_if<
using namespace double_conversion;
DoubleToStringConverter
conv(DoubleToStringConverter::NO_FLAGS,
- "infinity", "NaN", 'E',
- -6, // decimal in shortest low
- 21, // decimal in shortest high
+ "Infinity", "NaN", 'E',
+ detail::kConvMaxDecimalInShortestLow,
+ detail::kConvMaxDecimalInShortestHigh,
6, // max leading padding zeros
1); // max trailing padding zeros
char buffer[256];
}
/**
- * Variadic conversion to string. Appends each element in turn.
+ * Upper bound of the length of the output from
+ * DoubleToStringConverter::ToShortest(double, StringBuilder*),
+ * as used in toAppend(double, string*).
*/
+template <class Src>
+typename std::enable_if<
+ std::is_floating_point<Src>::value, size_t>::type
+estimateSpaceNeeded(Src value) {
+ // kBase10MaximalLength is 17. We add 1 for decimal point,
+ // e.g. 10.0/9 is 17 digits and 18 characters, including the decimal point.
+ constexpr int kMaxMantissaSpace =
+ double_conversion::DoubleToStringConverter::kBase10MaximalLength + 1;
+ // strlen("E-") + digits10(numeric_limits<double>::max_exponent10)
+ constexpr int kMaxExponentSpace = 2 + 3;
+ static const int kMaxPositiveSpace = std::max({
+ // E.g. 1.1111111111111111E-100.
+ kMaxMantissaSpace + kMaxExponentSpace,
+ // E.g. 0.000001.1111111111111111, if kConvMaxDecimalInShortestLow is -6.
+ kMaxMantissaSpace - detail::kConvMaxDecimalInShortestLow,
+ // If kConvMaxDecimalInShortestHigh is 21, then 1e21 is the smallest
+ // number > 1 which ToShortest outputs in exponential notation,
+ // so 21 is the longest non-exponential number > 1.
+ detail::kConvMaxDecimalInShortestHigh
+ });
+ return kMaxPositiveSpace + (value < 0); // +1 for minus sign, if negative
+}
+
+/**
+ * This can be specialized, together with adding specialization
+ * for estimateSpaceNeed for your type, so that we allocate
+ * as much as you need instead of the default
+ */
+template<class Src>
+struct HasLengthEstimator : std::false_type {};
+
+template <class Src>
+constexpr typename std::enable_if<
+ !std::is_fundamental<Src>::value
+#if FOLLY_HAVE_INT128_T
+ // On OSX 10.10, is_fundamental<__int128> is false :-O
+ && !std::is_same<__int128, Src>::value
+ && !std::is_same<unsigned __int128, Src>::value
+#endif
+ && !IsSomeString<Src>::value
+ && !std::is_convertible<Src, const char*>::value
+ && !std::is_convertible<Src, StringPiece>::value
+ && !std::is_enum<Src>::value
+ && !HasLengthEstimator<Src>::value,
+ size_t>::type
+estimateSpaceNeeded(const Src&) {
+ return sizeof(Src) + 1; // dumbest best effort ever?
+}
+
+namespace detail {
+
+template <class Tgt>
+typename std::enable_if<IsSomeString<Tgt>::value, size_t>::type
+estimateSpaceToReserve(size_t sofar, Tgt*) {
+ return sofar;
+}
+
+template <class T, class... Ts>
+size_t estimateSpaceToReserve(size_t sofar, const T& v, const Ts&... vs) {
+ return estimateSpaceToReserve(sofar + estimateSpaceNeeded(v), vs...);
+}
+
+template<class...Ts>
+void reserveInTarget(const Ts&...vs) {
+ getLastElement(vs...)->reserve(estimateSpaceToReserve(0, vs...));
+}
+
+template<class Delimiter, class...Ts>
+void reserveInTargetDelim(const Delimiter& d, const Ts&...vs) {
+ static_assert(sizeof...(vs) >= 2, "Needs at least 2 args");
+ size_t fordelim = (sizeof...(vs) - 2) *
+ estimateSpaceToReserve(0, d, static_cast<std::string*>(nullptr));
+ getLastElement(vs...)->reserve(estimateSpaceToReserve(fordelim, vs...));
+}
+
+/**
+ * Variadic base case: append one element
+ */
+template <class T, class Tgt>
+typename std::enable_if<
+ IsSomeString<typename std::remove_pointer<Tgt>::type>
+ ::value>::type
+toAppendStrImpl(const T& v, Tgt result) {
+ toAppend(v, result);
+}
+
template <class T, class... Ts>
typename std::enable_if<sizeof...(Ts) >= 2
&& IsSomeString<
typename std::remove_pointer<
typename detail::last_element<Ts...>::type
>::type>::value>::type
-toAppend(const T& v, const Ts&... vs) {
+toAppendStrImpl(const T& v, const Ts&... vs) {
+ toAppend(v, getLastElement(vs...));
+ toAppendStrImpl(vs...);
+}
+
+template <class Delimiter, class T, class Tgt>
+typename std::enable_if<
+ IsSomeString<typename std::remove_pointer<Tgt>::type>::value>::type
+toAppendDelimStrImpl(const Delimiter& /* delim */, const T& v, Tgt result) {
+ toAppend(v, result);
+}
+
+template <class Delimiter, class T, class... Ts>
+typename std::enable_if<sizeof...(Ts) >= 2
+ && IsSomeString<
+ typename std::remove_pointer<
+ typename detail::last_element<Ts...>::type
+ >::type>::value>::type
+toAppendDelimStrImpl(const Delimiter& delim, const T& v, const Ts&... vs) {
+ // we are really careful here, calling toAppend with just one element does
+ // not try to estimate space needed (as we already did that). If we call
+ // toAppend(v, delim, ....) we would do unnecesary size calculation
toAppend(v, detail::getLastElement(vs...));
+ toAppend(delim, detail::getLastElement(vs...));
+ toAppendDelimStrImpl(delim, vs...);
+}
+} // folly::detail
+
+
+/**
+ * Variadic conversion to string. Appends each element in turn.
+ * If we have two or more things to append, we it will not reserve
+ * the space for them and will depend on strings exponential growth.
+ * If you just append once consider using toAppendFit which reserves
+ * the space needed (but does not have exponential as a result).
+ *
+ * Custom implementations of toAppend() can be provided in the same namespace as
+ * the type to customize printing. estimateSpaceNeed() may also be provided to
+ * avoid reallocations in toAppendFit():
+ *
+ * namespace other_namespace {
+ *
+ * template <class String>
+ * void toAppend(const OtherType&, String* out);
+ *
+ * // optional
+ * size_t estimateSpaceNeeded(const OtherType&);
+ *
+ * }
+ */
+template <class... Ts>
+typename std::enable_if<sizeof...(Ts) >= 3
+ && IsSomeString<
+ typename std::remove_pointer<
+ typename detail::last_element<Ts...>::type
+ >::type>::value>::type
+toAppend(const Ts&... vs) {
+ ::folly::detail::toAppendStrImpl(vs...);
+}
+
+#ifdef _MSC_VER
+// Special case pid_t on MSVC, because it's a void* rather than an
+// integral type. We can't do a global special case because this is already
+// dangerous enough (as most pointers will implicitly convert to a void*)
+// just doing it for MSVC.
+template <class Tgt>
+void toAppend(const pid_t a, Tgt* res) {
+ toAppend(uint64_t(a), res);
+}
+#endif
+
+/**
+ * Special version of the call that preallocates exaclty as much memory
+ * as need for arguments to be stored in target. This means we are
+ * not doing exponential growth when we append. If you are using it
+ * in a loop you are aiming at your foot with a big perf-destroying
+ * bazooka.
+ * On the other hand if you are appending to a string once, this
+ * will probably save a few calls to malloc.
+ */
+template <class... Ts>
+typename std::enable_if<
+ IsSomeString<
+ typename std::remove_pointer<
+ typename detail::last_element<Ts...>::type
+ >::type>::value>::type
+toAppendFit(const Ts&... vs) {
+ ::folly::detail::reserveInTarget(vs...);
toAppend(vs...);
}
+template <class Ts>
+void toAppendFit(const Ts&) {}
+
/**
* Variadic base case: do nothing.
*/
template <class Tgt>
-typename std::enable_if<IsSomeString<Tgt>::value>::type
-toAppend(Tgt* result) {
-}
+typename std::enable_if<IsSomeString<Tgt>::value>::type toAppend(
+ Tgt* /* result */) {}
/**
* Variadic base case: do nothing.
*/
template <class Delimiter, class Tgt>
-typename std::enable_if<IsSomeString<Tgt>::value>::type
-toAppendDelim(const Delimiter& delim, Tgt* result) {
-}
+typename std::enable_if<IsSomeString<Tgt>::value>::type toAppendDelim(
+ const Delimiter& /* delim */, Tgt* /* result */) {}
/**
* 1 element: same as toAppend.
*/
template <class Delimiter, class T, class Tgt>
-typename std::enable_if<IsSomeString<Tgt>::value>::type
-toAppendDelim(const Delimiter& delim, const T& v, Tgt* tgt) {
+typename std::enable_if<IsSomeString<Tgt>::value>::type toAppendDelim(
+ const Delimiter& /* delim */, const T& v, Tgt* tgt) {
toAppend(v, tgt);
}
/**
- * Append to string with a delimiter in between elements.
+ * Append to string with a delimiter in between elements. Check out
+ * comments for toAppend for details about memory allocation.
*/
-template <class Delimiter, class T, class... Ts>
-typename std::enable_if<sizeof...(Ts) >= 2
+template <class Delimiter, class... Ts>
+typename std::enable_if<sizeof...(Ts) >= 3
&& IsSomeString<
typename std::remove_pointer<
typename detail::last_element<Ts...>::type
>::type>::value>::type
-toAppendDelim(const Delimiter& delim, const T& v, const Ts&... vs) {
- toAppend(v, delim, detail::getLastElement(vs...));
+toAppendDelim(const Delimiter& delim, const Ts&... vs) {
+ detail::toAppendDelimStrImpl(delim, vs...);
+}
+
+/**
+ * Detail in comment for toAppendFit
+ */
+template <class Delimiter, class... Ts>
+typename std::enable_if<
+ IsSomeString<
+ typename std::remove_pointer<
+ typename detail::last_element<Ts...>::type
+ >::type>::value>::type
+toAppendDelimFit(const Delimiter& delim, const Ts&... vs) {
+ detail::reserveInTargetDelim(delim, vs...);
toAppendDelim(delim, vs...);
}
+template <class De, class Ts>
+void toAppendDelimFit(const De&, const Ts&) {}
+
/**
* to<SomeString>(v1, v2, ...) uses toAppend() (see below) as back-end
* for all types.
*/
template <class Tgt, class... Ts>
-typename std::enable_if<IsSomeString<Tgt>::value, Tgt>::type
+typename std::enable_if<
+ IsSomeString<Tgt>::value && (
+ sizeof...(Ts) != 1 ||
+ !std::is_same<Tgt, typename detail::last_element<Ts...>::type>::value),
+ Tgt>::type
to(const Ts&... vs) {
Tgt result;
- toAppend(vs..., &result);
+ toAppendFit(vs..., &result);
return result;
}
+/**
+ * toDelim<SomeString>(SomeString str) returns itself.
+ */
+template <class Tgt, class Delim, class Src>
+typename std::enable_if<IsSomeString<Tgt>::value &&
+ std::is_same<Tgt, Src>::value,
+ Tgt>::type
+toDelim(const Delim& /* delim */, const Src& value) {
+ return value;
+}
+
/**
* toDelim<SomeString>(delim, v1, v2, ...) uses toAppendDelim() as
* back-end for all types.
*/
template <class Tgt, class Delim, class... Ts>
-typename std::enable_if<IsSomeString<Tgt>::value, Tgt>::type
+typename std::enable_if<
+ IsSomeString<Tgt>::value && (
+ sizeof...(Ts) != 1 ||
+ !std::is_same<Tgt, typename detail::last_element<Ts...>::type>::value),
+ Tgt>::type
toDelim(const Delim& delim, const Ts&... vs) {
Tgt result;
- toAppendDelim(delim, vs..., &result);
+ toAppendDelimFit(delim, vs..., &result);
return result;
}
namespace detail {
-/**
- * Finds the first non-digit in a string. The number of digits
- * searched depends on the precision of the Tgt integral. Assumes the
- * string starts with NO whitespace and NO sign.
- *
- * The semantics of the routine is:
- * for (;; ++b) {
- * if (b >= e || !isdigit(*b)) return b;
- * }
- *
- * Complete unrolling marks bottom-line (i.e. entire conversion)
- * improvements of 20%.
- */
- template <class Tgt>
- const char* findFirstNonDigit(const char* b, const char* e) {
- for (; b < e; ++b) {
- auto const c = static_cast<unsigned>(*b) - '0';
- if (c >= 10) break;
- }
- return b;
- }
+ConversionResult<bool> str_to_bool(StringPiece* src);
- // Maximum value of number when represented as a string
- template <class T> struct MaxString {
- static const char*const value;
- };
+template <typename T>
+ConversionResult<T> str_to_floating(StringPiece* src);
+extern template ConversionResult<float> str_to_floating<float>(
+ StringPiece* src);
+extern template ConversionResult<double> str_to_floating<double>(
+ StringPiece* src);
-/*
- * Lookup tables that converts from a decimal character value to an integral
- * binary value, shifted by a decimal "shift" multiplier.
- * For all character values in the range '0'..'9', the table at those
- * index locations returns the actual decimal value shifted by the multiplier.
- * For all other values, the lookup table returns an invalid OOR value.
- */
-// Out-of-range flag value, larger than the largest value that can fit in
-// four decimal bytes (9999), but four of these added up together should
-// still not overflow uint16_t.
-constexpr int32_t OOR = 10000;
-
-__attribute__((aligned(16))) constexpr uint16_t shift1[] = {
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 0-9
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 10
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 20
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 30
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, 0, // 40
- 1, 2, 3, 4, 5, 6, 7, 8, 9, OOR, OOR,
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 60
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 70
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 80
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 90
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 100
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 110
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 120
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 130
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 140
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 150
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 160
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 170
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 180
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 190
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 200
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 210
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 220
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 230
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 240
- OOR, OOR, OOR, OOR, OOR, OOR // 250
-};
+template <class Tgt>
+ConversionResult<Tgt> digits_to(const char* b, const char* e);
+
+extern template ConversionResult<char> digits_to<char>(
+ const char*,
+ const char*);
+extern template ConversionResult<signed char> digits_to<signed char>(
+ const char*,
+ const char*);
+extern template ConversionResult<unsigned char> digits_to<unsigned char>(
+ const char*,
+ const char*);
+
+extern template ConversionResult<short> digits_to<short>(
+ const char*,
+ const char*);
+extern template ConversionResult<unsigned short> digits_to<unsigned short>(
+ const char*,
+ const char*);
+
+extern template ConversionResult<int> digits_to<int>(const char*, const char*);
+extern template ConversionResult<unsigned int> digits_to<unsigned int>(
+ const char*,
+ const char*);
+
+extern template ConversionResult<long> digits_to<long>(
+ const char*,
+ const char*);
+extern template ConversionResult<unsigned long> digits_to<unsigned long>(
+ const char*,
+ const char*);
+
+extern template ConversionResult<long long> digits_to<long long>(
+ const char*,
+ const char*);
+extern template ConversionResult<unsigned long long>
+digits_to<unsigned long long>(const char*, const char*);
-__attribute__((aligned(16))) constexpr uint16_t shift10[] = {
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 0-9
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 10
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 20
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 30
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, 0, // 40
- 10, 20, 30, 40, 50, 60, 70, 80, 90, OOR, OOR,
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 60
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 70
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 80
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 90
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 100
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 110
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 120
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 130
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 140
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 150
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 160
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 170
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 180
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 190
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 200
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 210
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 220
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 230
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 240
- OOR, OOR, OOR, OOR, OOR, OOR // 250
-};
+#if FOLLY_HAVE_INT128_T
+extern template ConversionResult<__int128> digits_to<__int128>(
+ const char*,
+ const char*);
+extern template ConversionResult<unsigned __int128>
+digits_to<unsigned __int128>(const char*, const char*);
+#endif
-__attribute__((aligned(16))) constexpr uint16_t shift100[] = {
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 0-9
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 10
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 20
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 30
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, 0, // 40
- 100, 200, 300, 400, 500, 600, 700, 800, 900, OOR, OOR,
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 60
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 70
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 80
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 90
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 100
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 110
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 120
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 130
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 140
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 150
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 160
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 170
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 180
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 190
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 200
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 210
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 220
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 230
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 240
- OOR, OOR, OOR, OOR, OOR, OOR // 250
-};
+template <class T>
+ConversionResult<T> str_to_integral(StringPiece* src);
-__attribute__((aligned(16))) constexpr uint16_t shift1000[] = {
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 0-9
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 10
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 20
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 30
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, 0, // 40
- 1000, 2000, 3000, 4000, 5000, 6000, 7000, 8000, 9000, OOR, OOR,
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 60
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 70
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 80
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 90
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 100
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 110
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 120
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 130
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 140
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 150
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 160
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 170
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 180
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 190
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 200
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 210
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 220
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 230
- OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, OOR, // 240
- OOR, OOR, OOR, OOR, OOR, OOR // 250
-};
+extern template ConversionResult<char> str_to_integral<char>(StringPiece* src);
+extern template ConversionResult<signed char> str_to_integral<signed char>(
+ StringPiece* src);
+extern template ConversionResult<unsigned char> str_to_integral<unsigned char>(
+ StringPiece* src);
-/**
- * String represented as a pair of pointers to char to unsigned
- * integrals. Assumes NO whitespace before or after, and also that the
- * string is composed entirely of digits. Tgt must be unsigned, and no
- * sign is allowed in the string (even it's '+'). String may be empty,
- * in which case digits_to throws.
- */
- template <class Tgt>
- Tgt digits_to(const char * b, const char * e) {
-
- static_assert(!std::is_signed<Tgt>::value, "Unsigned type expected");
- assert(b <= e);
-
- const size_t size = e - b;
-
- /* Although the string is entirely made of digits, we still need to
- * check for overflow.
- */
- if (size >= std::numeric_limits<Tgt>::digits10 + 1) {
- // Leading zeros? If so, recurse to keep things simple
- if (b < e && *b == '0') {
- for (++b;; ++b) {
- if (b == e) return 0; // just zeros, e.g. "0000"
- if (*b != '0') return digits_to<Tgt>(b, e);
- }
- }
- FOLLY_RANGE_CHECK(size == std::numeric_limits<Tgt>::digits10 + 1 &&
- strncmp(b, detail::MaxString<Tgt>::value, size) <= 0,
- "Numeric overflow upon conversion");
- }
+extern template ConversionResult<short> str_to_integral<short>(
+ StringPiece* src);
+extern template ConversionResult<unsigned short>
+str_to_integral<unsigned short>(StringPiece* src);
- // Here we know that the number won't overflow when
- // converted. Proceed without checks.
+extern template ConversionResult<int> str_to_integral<int>(StringPiece* src);
+extern template ConversionResult<unsigned int> str_to_integral<unsigned int>(
+ StringPiece* src);
- Tgt result = 0;
+extern template ConversionResult<long> str_to_integral<long>(StringPiece* src);
+extern template ConversionResult<unsigned long> str_to_integral<unsigned long>(
+ StringPiece* src);
- for (; e - b >= 4; b += 4) {
- result *= 10000;
- const int32_t r0 = shift1000[static_cast<size_t>(b[0])];
- const int32_t r1 = shift100[static_cast<size_t>(b[1])];
- const int32_t r2 = shift10[static_cast<size_t>(b[2])];
- const int32_t r3 = shift1[static_cast<size_t>(b[3])];
- const auto sum = r0 + r1 + r2 + r3;
- assert(sum < OOR && "Assumption: string only has digits");
- result += sum;
- }
+extern template ConversionResult<long long> str_to_integral<long long>(
+ StringPiece* src);
+extern template ConversionResult<unsigned long long>
+str_to_integral<unsigned long long>(StringPiece* src);
- switch (e - b) {
- case 3: {
- const int32_t r0 = shift100[static_cast<size_t>(b[0])];
- const int32_t r1 = shift10[static_cast<size_t>(b[1])];
- const int32_t r2 = shift1[static_cast<size_t>(b[2])];
- const auto sum = r0 + r1 + r2;
- assert(sum < OOR && "Assumption: string only has digits");
- return result * 1000 + sum;
- }
- case 2: {
- const int32_t r0 = shift10[static_cast<size_t>(b[0])];
- const int32_t r1 = shift1[static_cast<size_t>(b[1])];
- const auto sum = r0 + r1;
- assert(sum < OOR && "Assumption: string only has digits");
- return result * 100 + sum;
- }
- case 1: {
- const int32_t sum = shift1[static_cast<size_t>(b[0])];
- assert(sum < OOR && "Assumption: string only has digits");
- return result * 10 + sum;
- }
- }
+#if FOLLY_HAVE_INT128_T
+extern template ConversionResult<__int128> str_to_integral<__int128>(
+ StringPiece* src);
+extern template ConversionResult<unsigned __int128>
+str_to_integral<unsigned __int128>(StringPiece* src);
+#endif
- assert(b == e);
- FOLLY_RANGE_CHECK(size > 0, "Found no digits to convert in input");
- return result;
- }
+template <typename T>
+typename std::enable_if<std::is_same<T, bool>::value, ConversionResult<T>>::type
+convertTo(StringPiece* src) {
+ return str_to_bool(src);
+}
+
+template <typename T>
+typename std::enable_if<
+ std::is_floating_point<T>::value, ConversionResult<T>>::type
+convertTo(StringPiece* src) {
+ return str_to_floating<T>(src);
+}
+template <typename T>
+typename std::enable_if<
+ std::is_integral<T>::value && !std::is_same<T, bool>::value,
+ ConversionResult<T>>::type
+convertTo(StringPiece* src) {
+ return str_to_integral<T>(src);
+}
- bool str_to_bool(StringPiece * src);
+template <typename T>
+struct WrapperInfo { using type = T; };
-} // namespace detail
+template <typename T, typename Gen>
+typename std::enable_if<
+ std::is_same<typename WrapperInfo<T>::type, T>::value, T>::type
+inline wrap(ConversionResult<typename WrapperInfo<T>::type> res, Gen&& gen) {
+ if (LIKELY(res.success())) {
+ return res.value;
+ }
+ throw detail::makeConversionError(res.error, gen());
+}
+
+} // namespace detail
/**
* String represented as a pair of pointers to char to unsigned
* integrals. Assumes NO whitespace before or after.
*/
-template <class Tgt>
+template <typename Tgt>
typename std::enable_if<
- std::is_integral<Tgt>::value && !std::is_signed<Tgt>::value
- && !std::is_same<typename std::remove_cv<Tgt>::type, bool>::value,
- Tgt>::type
-to(const char * b, const char * e) {
- return detail::digits_to<Tgt>(b, e);
+ std::is_integral<typename detail::WrapperInfo<Tgt>::type>::value &&
+ !std::is_same<typename detail::WrapperInfo<Tgt>::type, bool>::value,
+ Tgt>::type
+to(const char* b, const char* e) {
+ auto res = detail::digits_to<typename detail::WrapperInfo<Tgt>::type>(b, e);
+ return detail::wrap<Tgt>(res, [&] { return StringPiece(b, e); });
}
+/*******************************************************************************
+ * Conversions from string types to arithmetic types.
+ ******************************************************************************/
+
/**
- * String represented as a pair of pointers to char to signed
- * integrals. Assumes NO whitespace before or after. Allows an
- * optional leading sign.
+ * Parsing strings to numeric types. These routines differ from
+ * parseTo(str, numeric) routines in that they take a POINTER TO a StringPiece
+ * and alter that StringPiece to reflect progress information.
*/
-template <class Tgt>
+template <typename Tgt>
typename std::enable_if<
- std::is_integral<Tgt>::value && std::is_signed<Tgt>::value,
- Tgt>::type
-to(const char * b, const char * e) {
- FOLLY_RANGE_CHECK(b < e, "Empty input string in conversion to integral");
- if (!isdigit(*b)) {
- if (*b == '-') {
- Tgt result = -to<typename std::make_unsigned<Tgt>::type>(b + 1, e);
- FOLLY_RANGE_CHECK(result <= 0, "Negative overflow.");
- return result;
- }
- FOLLY_RANGE_CHECK(*b == '+', "Invalid lead character");
- ++b;
+ std::is_arithmetic<typename detail::WrapperInfo<Tgt>::type>::value>::type
+parseTo(StringPiece* src, Tgt& out) {
+ auto res = detail::convertTo<typename detail::WrapperInfo<Tgt>::type>(src);
+ out = detail::wrap<Tgt>(res, [&] { return *src; });
+}
+
+template <typename Tgt>
+typename std::enable_if<
+ std::is_arithmetic<typename detail::WrapperInfo<Tgt>::type>::value>::type
+parseTo(StringPiece src, Tgt& out) {
+ auto res = detail::convertTo<typename detail::WrapperInfo<Tgt>::type>(&src);
+ if (LIKELY(res.success())) {
+ res.error = detail::enforceWhitespaceErr(src);
}
- Tgt result = to<typename std::make_unsigned<Tgt>::type>(b, e);
- FOLLY_RANGE_CHECK(result >= 0, "Overflow.");
- return result;
+ out = detail::wrap<Tgt>(res, [&] { return src; });
}
-/**
- * Parsing strings to integrals. These routines differ from
- * to<integral>(string) in that they take a POINTER TO a StringPiece
- * and alter that StringPiece to reflect progress information.
- */
+/*******************************************************************************
+ * Integral / Floating Point to integral / Floating Point
+ ******************************************************************************/
/**
- * StringPiece to integrals, with progress information. Alters the
- * StringPiece parameter to munch the already-parsed characters.
+ * Unchecked conversion from arithmetic to boolean. This is different from the
+ * other arithmetic conversions because we use the C convention of treating any
+ * non-zero value as true, instead of range checking.
*/
-template <class Tgt>
+template <class Tgt, class Src>
typename std::enable_if<
- std::is_integral<Tgt>::value
- && !std::is_same<typename std::remove_cv<Tgt>::type, bool>::value,
- Tgt>::type
-to(StringPiece * src) {
+ std::is_arithmetic<Src>::value && !std::is_same<Tgt, Src>::value &&
+ std::is_same<Tgt, bool>::value,
+ Tgt>::type
+to(const Src& value) {
+ return value != Src();
+}
- auto b = src->data(), past = src->data() + src->size();
- for (;; ++b) {
- FOLLY_RANGE_CHECK(b < past, "No digits found in input string");
- if (!isspace(*b)) break;
- }
+namespace detail {
- auto m = b;
-
- // First digit is customized because we test for sign
- bool negative = false;
- /* static */ if (std::is_signed<Tgt>::value) {
- if (!isdigit(*m)) {
- if (*m == '-') {
- negative = true;
- } else {
- FOLLY_RANGE_CHECK(*m == '+', "Invalid leading character in conversion"
- " to integral");
- }
- ++b;
- ++m;
+/**
+ * Checked conversion from integral to integral. The checks are only
+ * performed when meaningful, e.g. conversion from int to long goes
+ * unchecked.
+ */
+template <class Tgt, class Src>
+typename std::enable_if<
+ std::is_integral<Src>::value && !std::is_same<Tgt, Src>::value &&
+ !std::is_same<Tgt, bool>::value &&
+ std::is_integral<Tgt>::value,
+ ConversionResult<Tgt>>::type
+convertTo(const Src& value) {
+ /* static */ if (
+ std::numeric_limits<Tgt>::max() < std::numeric_limits<Src>::max()) {
+ if (greater_than<Tgt, std::numeric_limits<Tgt>::max()>(value)) {
+ return ConversionResult<Tgt>(ConversionError::ARITH_POSITIVE_OVERFLOW);
}
}
- FOLLY_RANGE_CHECK(m < past, "No digits found in input string");
- FOLLY_RANGE_CHECK(isdigit(*m), "Non-digit character found");
- m = detail::findFirstNonDigit<Tgt>(m + 1, past);
-
- Tgt result;
- /* static */ if (!std::is_signed<Tgt>::value) {
- result = detail::digits_to<typename std::make_unsigned<Tgt>::type>(b, m);
- } else {
- auto t = detail::digits_to<typename std::make_unsigned<Tgt>::type>(b, m);
- if (negative) {
- result = -t;
- FOLLY_RANGE_CHECK(is_non_positive(result), "Negative overflow");
- } else {
- result = t;
- FOLLY_RANGE_CHECK(is_non_negative(result), "Overflow");
+ /* static */ if (
+ std::is_signed<Src>::value &&
+ (!std::is_signed<Tgt>::value || sizeof(Src) > sizeof(Tgt))) {
+ if (less_than<Tgt, std::numeric_limits<Tgt>::min()>(value)) {
+ return ConversionResult<Tgt>(ConversionError::ARITH_NEGATIVE_OVERFLOW);
}
}
- src->advance(m - src->data());
- return result;
+ return ConversionResult<Tgt>(static_cast<Tgt>(value));
}
/**
- * StringPiece to bool, with progress information. Alters the
- * StringPiece parameter to munch the already-parsed characters.
+ * Checked conversion from floating to floating. The checks are only
+ * performed when meaningful, e.g. conversion from float to double goes
+ * unchecked.
*/
-template <class Tgt>
+template <class Tgt, class Src>
typename std::enable_if<
- std::is_same<typename std::remove_cv<Tgt>::type, bool>::value,
- Tgt>::type
-to(StringPiece * src) {
- return detail::str_to_bool(src);
+ std::is_floating_point<Tgt>::value && std::is_floating_point<Src>::value &&
+ !std::is_same<Tgt, Src>::value,
+ ConversionResult<Tgt>>::type
+convertTo(const Src& value) {
+ /* static */ if (
+ std::numeric_limits<Tgt>::max() < std::numeric_limits<Src>::max()) {
+ if (value > std::numeric_limits<Tgt>::max()) {
+ return ConversionResult<Tgt>(ConversionError::ARITH_POSITIVE_OVERFLOW);
+ }
+ if (value < std::numeric_limits<Tgt>::lowest()) {
+ return ConversionResult<Tgt>(ConversionError::ARITH_NEGATIVE_OVERFLOW);
+ }
+ }
+ return ConversionResult<Tgt>(boost::implicit_cast<Tgt>(value));
}
-namespace detail {
-
/**
- * Enforce that the suffix following a number is made up only of whitespace.
+ * Check if a floating point value can safely be converted to an
+ * integer value without triggering undefined behaviour.
*/
-inline void enforceWhitespace(const char* b, const char* e) {
- for (; b != e; ++b) {
- FOLLY_RANGE_CHECK(isspace(*b), to<std::string>("Non-whitespace: ", *b));
+template <typename Tgt, typename Src>
+inline typename std::enable_if<
+ std::is_floating_point<Src>::value && std::is_integral<Tgt>::value &&
+ !std::is_same<Tgt, bool>::value,
+ bool>::type
+checkConversion(const Src& value) {
+ constexpr Src tgtMaxAsSrc = static_cast<Src>(std::numeric_limits<Tgt>::max());
+ constexpr Src tgtMinAsSrc = static_cast<Src>(std::numeric_limits<Tgt>::min());
+ if (value >= tgtMaxAsSrc) {
+ if (value > tgtMaxAsSrc) {
+ return false;
+ }
+ const Src mmax = folly::nextafter(tgtMaxAsSrc, Src());
+ if (static_cast<Tgt>(value - mmax) >
+ std::numeric_limits<Tgt>::max() - static_cast<Tgt>(mmax)) {
+ return false;
+ }
+ } else if (std::is_signed<Tgt>::value && value <= tgtMinAsSrc) {
+ if (value < tgtMinAsSrc) {
+ return false;
+ }
+ const Src mmin = folly::nextafter(tgtMinAsSrc, Src());
+ if (static_cast<Tgt>(value - mmin) <
+ std::numeric_limits<Tgt>::min() - static_cast<Tgt>(mmin)) {
+ return false;
+ }
}
+ return true;
}
-} // namespace detail
+// Integers can always safely be converted to floating point values
+template <typename Tgt, typename Src>
+constexpr typename std::enable_if<
+ std::is_integral<Src>::value && std::is_floating_point<Tgt>::value,
+ bool>::type
+checkConversion(const Src&) {
+ return true;
+}
+
+// Also, floating point values can always be safely converted to bool
+// Per the standard, any floating point value that is not zero will yield true
+template <typename Tgt, typename Src>
+constexpr typename std::enable_if<
+ std::is_floating_point<Src>::value && std::is_same<Tgt, bool>::value,
+ bool>::type
+checkConversion(const Src&) {
+ return true;
+}
/**
- * String or StringPiece to integrals. Accepts leading and trailing
- * whitespace, but no non-space trailing characters.
+ * Checked conversion from integral to floating point and back. The
+ * result must be convertible back to the source type without loss of
+ * precision. This seems Draconian but sometimes is what's needed, and
+ * complements existing routines nicely. For various rounding
+ * routines, see <math>.
*/
-template <class Tgt>
+template <typename Tgt, typename Src>
typename std::enable_if<
- std::is_integral<Tgt>::value,
- Tgt>::type
-to(StringPiece src) {
- Tgt result = to<Tgt>(&src);
- detail::enforceWhitespace(src.data(), src.data() + src.size());
- return result;
+ (std::is_integral<Src>::value && std::is_floating_point<Tgt>::value) ||
+ (std::is_floating_point<Src>::value && std::is_integral<Tgt>::value),
+ ConversionResult<Tgt>>::type
+convertTo(const Src& value) {
+ if (LIKELY(checkConversion<Tgt>(value))) {
+ Tgt result = static_cast<Tgt>(value);
+ if (LIKELY(checkConversion<Src>(result))) {
+ Src witness = static_cast<Src>(result);
+ if (LIKELY(value == witness)) {
+ return ConversionResult<Tgt>(result);
+ }
+ }
+ }
+ return ConversionResult<Tgt>(ConversionError::ARITH_LOSS_OF_PRECISION);
}
-/*******************************************************************************
- * Conversions from string types to floating-point types.
- ******************************************************************************/
+template <typename Tgt, typename Src>
+inline std::string errorValue(const Src& value) {
+#ifdef FOLLY_HAS_RTTI
+ return to<std::string>("(", demangle(typeid(Tgt)), ") ", value);
+#else
+ return to<std::string>(value);
+#endif
+}
-/**
- * StringPiece to double, with progress information. Alters the
- * StringPiece parameter to munch the already-parsed characters.
- */
-template <class Tgt>
-inline typename std::enable_if<
- std::is_floating_point<Tgt>::value,
- Tgt>::type
-to(StringPiece *const src) {
- using namespace double_conversion;
- static StringToDoubleConverter
- conv(StringToDoubleConverter::ALLOW_TRAILING_JUNK
- | StringToDoubleConverter::ALLOW_LEADING_SPACES,
- 0.0,
- // return this for junk input string
- std::numeric_limits<double>::quiet_NaN(),
- nullptr, nullptr);
-
- FOLLY_RANGE_CHECK(!src->empty(), "No digits found in input string");
-
- int length;
- auto result = conv.StringToDouble(src->data(), src->size(),
- &length); // processed char count
-
- if (!std::isnan(result)) {
- src->advance(length);
- return result;
- }
+template <typename Tgt, typename Src>
+using IsArithToArith = std::integral_constant<
+ bool,
+ !std::is_same<Tgt, Src>::value && !std::is_same<Tgt, bool>::value &&
+ std::is_arithmetic<Src>::value &&
+ std::is_arithmetic<Tgt>::value>;
- for (;; src->advance(1)) {
- if (src->empty()) {
- throw std::range_error("Unable to convert an empty string"
- " to a floating point value.");
- }
- if (!isspace(src->front())) {
- break;
- }
- }
+} // namespace detail
- // Was that "inf[inity]"?
- if (src->size() >= 3 && toupper((*src)[0]) == 'I'
- && toupper((*src)[1]) == 'N' && toupper((*src)[2]) == 'F') {
- if (src->size() >= 8 &&
- toupper((*src)[3]) == 'I' &&
- toupper((*src)[4]) == 'N' &&
- toupper((*src)[5]) == 'I' &&
- toupper((*src)[6]) == 'T' &&
- toupper((*src)[7]) == 'Y') {
- src->advance(8);
- } else {
- src->advance(3);
- }
- return std::numeric_limits<Tgt>::infinity();
- }
+template <typename Tgt, typename Src>
+typename std::enable_if<
+ detail::IsArithToArith<
+ typename detail::WrapperInfo<Tgt>::type, Src>::value, Tgt>::type
+to(const Src& value) {
+ auto res = detail::convertTo<typename detail::WrapperInfo<Tgt>::type>(value);
+ return detail::wrap<Tgt>(res, [&] {
+ return detail::errorValue<typename detail::WrapperInfo<Tgt>::type>(value);
+ });
+}
- // Was that "-inf[inity]"?
- if (src->size() >= 4 && toupper((*src)[0]) == '-'
- && toupper((*src)[1]) == 'I' && toupper((*src)[2]) == 'N'
- && toupper((*src)[3]) == 'F') {
- if (src->size() >= 9 &&
- toupper((*src)[4]) == 'I' &&
- toupper((*src)[5]) == 'N' &&
- toupper((*src)[6]) == 'I' &&
- toupper((*src)[7]) == 'T' &&
- toupper((*src)[8]) == 'Y') {
- src->advance(9);
- } else {
- src->advance(4);
- }
- return -std::numeric_limits<Tgt>::infinity();
- }
+/*******************************************************************************
+ * Custom Conversions
+ *
+ * Any type can be used with folly::to by implementing parseTo. The
+ * implementation should be provided in the namespace of the type to facilitate
+ * argument-dependent lookup:
+ *
+ * namespace other_namespace {
+ * void parseTo(::folly::StringPiece, OtherType&);
+ * }
+ ******************************************************************************/
+template <class T>
+typename std::enable_if<std::is_enum<T>::value>::type
+parseTo(StringPiece in, T& out) {
+ typename std::underlying_type<T>::type tmp;
+ parseTo(in, tmp);
+ out = static_cast<T>(tmp);
+}
- // "nan"?
- if (src->size() >= 3 && toupper((*src)[0]) == 'N'
- && toupper((*src)[1]) == 'A' && toupper((*src)[2]) == 'N') {
- src->advance(3);
- return std::numeric_limits<Tgt>::quiet_NaN();
- }
+inline void parseTo(StringPiece in, StringPiece& out) {
+ out = in;
+}
- // "-nan"?
- if (src->size() >= 4 &&
- toupper((*src)[0]) == '-' &&
- toupper((*src)[1]) == 'N' &&
- toupper((*src)[2]) == 'A' &&
- toupper((*src)[3]) == 'N') {
- src->advance(4);
- return -std::numeric_limits<Tgt>::quiet_NaN();
- }
+inline void parseTo(StringPiece in, std::string& out) {
+ out.clear();
+ out.append(in.data(), in.size());
+}
- // All bets are off
- throw std::range_error("Unable to convert \"" + src->toString()
- + "\" to a floating point value.");
+inline void parseTo(StringPiece in, fbstring& out) {
+ out.clear();
+ out.append(in.data(), in.size());
}
/**
- * Any string, const char*, or StringPiece to double.
+ * String or StringPiece to target conversion. Accepts leading and trailing
+ * whitespace, but no non-space trailing characters.
*/
+
template <class Tgt>
-typename std::enable_if<
- std::is_floating_point<Tgt>::value,
- Tgt>::type
+typename std::enable_if<!std::is_same<StringPiece, Tgt>::value, Tgt>::type
to(StringPiece src) {
- Tgt result = to<double>(&src);
- detail::enforceWhitespace(src.data(), src.data() + src.size());
+ Tgt result;
+ parseTo(src, result);
return result;
}
-/*******************************************************************************
- * Integral to floating point and back
- ******************************************************************************/
-
-/**
- * Checked conversion from integral to flating point and back. The
- * result must be convertible back to the source type without loss of
- * precision. This seems Draconian but sometimes is what's needed, and
- * complements existing routines nicely. For various rounding
- * routines, see <math>.
- */
-template <class Tgt, class Src>
-typename std::enable_if<
- (std::is_integral<Src>::value && std::is_floating_point<Tgt>::value)
- ||
- (std::is_floating_point<Src>::value && std::is_integral<Tgt>::value),
- Tgt>::type
-to(const Src & value) {
- Tgt result = value;
- auto witness = static_cast<Src>(result);
- if (value != witness) {
- throw std::range_error(
- to<std::string>("to<>: loss of precision when converting ", value,
- " to type ", typeid(Tgt).name()).c_str());
- }
+template <class Tgt>
+Tgt to(StringPiece* src) {
+ Tgt result;
+ parseTo(src, result);
return result;
}
* Enum to anything and back
******************************************************************************/
-#if defined(__clang__) || __GNUC_PREREQ(4, 7)
-// std::underlying_type became available by gcc 4.7.0
-
template <class Tgt, class Src>
-typename std::enable_if<std::is_enum<Src>::value, Tgt>::type
+typename std::enable_if<
+ std::is_enum<Src>::value && !std::is_same<Src, Tgt>::value, Tgt>::type
to(const Src & value) {
return to<Tgt>(static_cast<typename std::underlying_type<Src>::type>(value));
}
template <class Tgt, class Src>
-typename std::enable_if<std::is_enum<Tgt>::value, Tgt>::type
+typename std::enable_if<
+ std::is_enum<Tgt>::value && !std::is_same<Src, Tgt>::value, Tgt>::type
to(const Src & value) {
return static_cast<Tgt>(to<typename std::underlying_type<Tgt>::type>(value));
}
-#else
-
-template <class Tgt, class Src>
-typename std::enable_if<std::is_enum<Src>::value, Tgt>::type
-to(const Src & value) {
- /* static */ if (Src(-1) < 0) {
- /* static */ if (sizeof(Src) <= sizeof(int)) {
- return to<Tgt>(static_cast<int>(value));
- } else {
- return to<Tgt>(static_cast<long>(value));
- }
- } else {
- /* static */ if (sizeof(Src) <= sizeof(int)) {
- return to<Tgt>(static_cast<unsigned int>(value));
- } else {
- return to<Tgt>(static_cast<unsigned long>(value));
- }
- }
-}
-
-template <class Tgt, class Src>
-typename std::enable_if<std::is_enum<Tgt>::value, Tgt>::type
-to(const Src & value) {
- /* static */ if (Tgt(-1) < 0) {
- /* static */ if (sizeof(Tgt) <= sizeof(int)) {
- return static_cast<Tgt>(to<int>(value));
- } else {
- return static_cast<Tgt>(to<long>(value));
- }
- } else {
- /* static */ if (sizeof(Tgt) <= sizeof(int)) {
- return static_cast<Tgt>(to<unsigned int>(value));
- } else {
- return static_cast<Tgt>(to<unsigned long>(value));
- }
- }
-}
-
-#endif // gcc 4.7 onwards
-
} // namespace folly
-
-// FOLLY_CONV_INTERNAL is defined by Conv.cpp. Keep the FOLLY_RANGE_CHECK
-// macro for use in Conv.cpp, but #undefine it everywhere else we are included,
-// to avoid defining this global macro name in other files that include Conv.h.
-#ifndef FOLLY_CONV_INTERNAL
-#undef FOLLY_RANGE_CHECK
-#endif
-
-#endif /* FOLLY_BASE_CONV_H_ */