Export of internal Abseil changes

--
9beb68204986a015c9cb065b9fae4f9a8879a788 by Abseil Team <absl-team@google.com>:

Move Base64EscapeInternal and CalculateBase64EscapedLenInternal to an internal header.

PiperOrigin-RevId: 288917378

--
90acfbe03b3f9f6de3ffa49c39343dfaa2c5d38c by Greg Falcon <gfalcon@google.com>:

Update macos CI script to support the ALTERNATE_OPTIONS environment variable.

PiperOrigin-RevId: 288913564

--
f1572e870678cdcda6b48cb39780d1ad984e4c1b by Derek Mauro <dmauro@google.com>:

Makes absl::NullSafeStringView constexpr
Fixes https://github.com/abseil/abseil-cpp/issues/583

PiperOrigin-RevId: 288906940

--
d28a8471e32c10caa64bfffe6d6d4d0a8d144013 by Abseil Team <absl-team@google.com>:

absl::GetFlag is lock free for small trivially copyable types.

PiperOrigin-RevId: 288768172

--
2643b8ed1a1dc836b38ab9e46538a1af129ffd67 by Gennadiy Rozental <rogeeff@google.com>:

Eliminate call to callback from flag initialization.

We do not need to have this invocation inside FlagImpl::Init since SetCallback performs invocation anyways. Calling InitCallback from inside of Init complicates separation of value initialization from data guard initialization, which is about to happen.

PiperOrigin-RevId: 288732526

--
22caa880b7a4cb6da34e16a2e064a473c99e880b by Abseil Team <absl-team@google.com>:

Fix the documentation on how to create a null string_view.

PiperOrigin-RevId: 288727968

--
10727c5cadc561837141176f4c9b9717cec9233a by Greg Falcon <gfalcon@google.com>:

Change CI scripts for gcc to use the ALTERNATE_OPTIONS file as well.

PiperOrigin-RevId: 288718855

--
5d1e2dd6c7fb12af8aa4337a0f61872f5f0c5992 by Greg Falcon <gfalcon@google.com>:

Add an option for using inline namespaces in Abseil.

PiperOrigin-RevId: 288614491
GitOrigin-RevId: 9beb68204986a015c9cb065b9fae4f9a8879a788
Change-Id: If9acd46301e3df8cb231b4c16f7ed651bf4fb3c3
This commit is contained in:
Abseil Team 2020-01-09 09:58:48 -08:00 committed by Derek Mauro
parent 63ee2f8877
commit b3aaac8a37
19 changed files with 527 additions and 255 deletions

View file

@ -90,8 +90,42 @@
// not support forward declarations of its own types, nor does it support
// user-provided specialization of Abseil templates. Code that violates these
// rules may be broken without warning.)
#if !defined(ABSL_OPTION_USE_INLINE_NAMESPACE) || \
!defined(ABSL_OPTION_INLINE_NAMESPACE_NAME)
#error options.h is misconfigured.
#endif
// Check that ABSL_OPTION_INLINE_NAMESPACE_NAME is neither "head" nor ""
#if defined(__cplusplus) && ABSL_OPTION_USE_INLINE_NAMESPACE == 1
#define ABSL_INTERNAL_DO_TOKEN_STR(x) #x
#define ABSL_INTERNAL_TOKEN_STR(x) ABSL_INTERNAL_DO_TOKEN_STR(x)
#define ABSL_INTERNAL_INLINE_NAMESPACE_STR \
ABSL_INTERNAL_TOKEN_STR(ABSL_OPTION_INLINE_NAMESPACE_NAME)
static_assert(ABSL_INTERNAL_INLINE_NAMESPACE_STR[0] != '\0',
"options.h misconfigured: ABSL_OPTION_INLINE_NAMESPACE_NAME must "
"not be empty.");
static_assert(ABSL_INTERNAL_INLINE_NAMESPACE_STR[0] != 'h' ||
ABSL_INTERNAL_INLINE_NAMESPACE_STR[1] != 'e' ||
ABSL_INTERNAL_INLINE_NAMESPACE_STR[2] != 'a' ||
ABSL_INTERNAL_INLINE_NAMESPACE_STR[3] != 'd' ||
ABSL_INTERNAL_INLINE_NAMESPACE_STR[4] != '\0',
"options.h misconfigured: ABSL_OPTION_INLINE_NAMESPACE_NAME must "
"be changed to a new, unique identifier name.");
#endif
#if ABSL_OPTION_USE_INLINE_NAMESPACE == 0
#define ABSL_NAMESPACE_BEGIN
#define ABSL_NAMESPACE_END
#elif ABSL_OPTION_USE_INLINE_NAMESPACE == 1
#define ABSL_NAMESPACE_BEGIN \
inline namespace ABSL_OPTION_INLINE_NAMESPACE_NAME {
#define ABSL_NAMESPACE_END }
#else
#error options.h is misconfigured.
#endif
// -----------------------------------------------------------------------------
// Compiler Feature Checks

View file

@ -185,4 +185,30 @@
#define ABSL_OPTION_USE_STD_VARIANT 2
// ABSL_OPTION_USE_INLINE_NAMESPACE
// ABSL_OPTION_INLINE_NAMESPACE_NAME
//
// These options controls whether all entities in the absl namespace are
// contained within an inner inline namespace. This does not affect the
// user-visible API of Abseil, but it changes the mangled names of all symbols.
//
// This can be useful as a version tag if you are distributing Abseil in
// precompiled form. This will prevent a binary library build of Abseil with
// one inline namespace being used with headers configured with a different
// inline namespace name. Binary packagers are reminded that Abseil does not
// guarantee any ABI stability in Abseil, so any update of Abseil or
// configuration change in such a binary package should be combined with a
// new, unique value for the inline namespace name.
//
// A value of 0 means not to use inline namespaces.
//
// A value of 1 means to use an inline namespace with the given name inside
// namespace absl. If this is set, ABSL_OPTION_INLINE_NAMESPACE_NAME must also
// be changed to a new, unique identifier name. In particular "head" is not
// allowed.
#define ABSL_OPTION_USE_INLINE_NAMESPACE 0
#define ABSL_OPTION_INLINE_NAMESPACE_NAME head
#endif // ABSL_BASE_OPTIONS_H_

View file

@ -45,4 +45,15 @@
#define ABSL_FLAGS_STRIP_HELP ABSL_FLAGS_STRIP_NAMES
#endif
// ABSL_FLAGS_INTERNAL_ATOMIC_DOUBLE_WORD macro is used for using atomics with
// double words, e.g. absl::Duration.
// For reasons in bug https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80878, modern
// versions of GCC do not support cmpxchg16b instruction in standard atomics.
#ifdef ABSL_FLAGS_INTERNAL_ATOMIC_DOUBLE_WORD
#error "ABSL_FLAGS_INTERNAL_ATOMIC_DOUBLE_WORD should not be defined."
#elif defined(__clang__) && defined(__x86_64__) && \
defined(__GCC_HAVE_SYNC_COMPARE_AND_SWAP_16)
#define ABSL_FLAGS_INTERNAL_ATOMIC_DOUBLE_WORD 1
#endif
#endif // ABSL_FLAGS_CONFIG_H_

View file

@ -20,28 +20,6 @@
namespace absl {
ABSL_NAMESPACE_BEGIN
// We want to validate the type mismatch between type definition and
// declaration. The lock-free implementation does not allow us to do it,
// so in debug builds we always use the slower implementation, which always
// validates the type.
#ifndef NDEBUG
#define ABSL_FLAGS_ATOMIC_GET(T) \
T GetFlag(const absl::Flag<T>& flag) { return flag.Get(); }
#else
#define ABSL_FLAGS_ATOMIC_GET(T) \
T GetFlag(const absl::Flag<T>& flag) { \
T result; \
if (flag.AtomicGet(&result)) { \
return result; \
} \
return flag.Get(); \
}
#endif
ABSL_FLAGS_INTERNAL_FOR_EACH_LOCK_FREE(ABSL_FLAGS_ATOMIC_GET)
#undef ABSL_FLAGS_ATOMIC_GET
// This global nutex protects on-demand construction of flag objects in MSVC
// builds.
#if defined(_MSC_VER) && !defined(__clang__)

View file

@ -29,6 +29,8 @@
#ifndef ABSL_FLAGS_FLAG_H_
#define ABSL_FLAGS_FLAG_H_
#include <type_traits>
#include "absl/base/attributes.h"
#include "absl/base/casts.h"
#include "absl/flags/config.h"
@ -181,23 +183,42 @@ class Flag {
//
// // FLAGS_firstname is a Flag of type `std::string`
// std::string first_name = absl::GetFlag(FLAGS_firstname);
template <typename T>
template <typename T,
typename std::enable_if<
!flags_internal::IsAtomicFlagTypeTrait<T>::value, int>::type = 0>
ABSL_MUST_USE_RESULT T GetFlag(const absl::Flag<T>& flag) {
#define ABSL_FLAGS_INTERNAL_LOCK_FREE_VALIDATE(BIT) \
static_assert( \
!std::is_same<T, BIT>::value, \
"Do not specify explicit template parameters to absl::GetFlag");
ABSL_FLAGS_INTERNAL_FOR_EACH_LOCK_FREE(ABSL_FLAGS_INTERNAL_LOCK_FREE_VALIDATE)
#undef ABSL_FLAGS_INTERNAL_LOCK_FREE_VALIDATE
return flag.Get();
}
// We want to validate the type mismatch between type definition and
// declaration. The lock-free implementation does not allow us to do it,
// so in debug builds we always use the slower implementation, which always
// validates the type.
#ifndef NDEBUG
template <typename T,
typename std::enable_if<
flags_internal::IsAtomicFlagTypeTrait<T>::value, int>::type = 0>
ABSL_MUST_USE_RESULT T GetFlag(const absl::Flag<T>& flag) {
return flag.Get();
}
#else
// Overload for `GetFlag()` for types that support lock-free reads.
#define ABSL_FLAGS_INTERNAL_LOCK_FREE_EXPORT(T) \
ABSL_MUST_USE_RESULT T GetFlag(const absl::Flag<T>& flag);
ABSL_FLAGS_INTERNAL_FOR_EACH_LOCK_FREE(ABSL_FLAGS_INTERNAL_LOCK_FREE_EXPORT)
#undef ABSL_FLAGS_INTERNAL_LOCK_FREE_EXPORT
template <typename T,
typename std::enable_if<
flags_internal::IsAtomicFlagTypeTrait<T>::value, int>::type = 0>
ABSL_MUST_USE_RESULT T GetFlag(const absl::Flag<T>& flag) {
// T might not be default constructible.
union U {
T value;
U() {}
};
U result;
if (flag.AtomicGet(&result.value)) {
return result.value;
}
return flag.Get();
}
#endif
// SetFlag()
//

View file

@ -259,22 +259,22 @@ class CommandLineFlag {
virtual void Read(void* dst) const = 0;
};
// This macro is the "source of truth" for the list of supported flag types we
// expect to perform lock free operations on. Specifically it generates code,
// a one argument macro operating on a type, supplied as a macro argument, for
// each type in the list.
#define ABSL_FLAGS_INTERNAL_FOR_EACH_LOCK_FREE(A) \
A(bool) \
A(short) \
A(unsigned short) \
A(int) \
A(unsigned int) \
A(long) \
A(unsigned long) \
A(long long) \
A(unsigned long long) \
A(double) \
A(float)
// This macro is the "source of truth" for the list of supported flag built-in
// types.
#define ABSL_FLAGS_INTERNAL_BUILTIN_TYPES(A) \
A(bool) \
A(short) \
A(unsigned short) \
A(int) \
A(unsigned int) \
A(long) \
A(unsigned long) \
A(long long) \
A(unsigned long long) \
A(double) \
A(float) \
A(std::string) \
A(std::vector<std::string>)
} // namespace flags_internal
ABSL_NAMESPACE_END

View file

@ -16,6 +16,7 @@
#include "absl/flags/internal/flag.h"
#include "absl/base/optimization.h"
#include "absl/flags/config.h"
#include "absl/flags/usage_config.h"
#include "absl/synchronization/mutex.h"
@ -35,9 +36,7 @@ namespace {
bool ShouldValidateFlagValue(FlagOpFn flag_type_id) {
#define DONT_VALIDATE(T) \
if (flag_type_id == &flags_internal::FlagOps<T>) return false;
ABSL_FLAGS_INTERNAL_FOR_EACH_LOCK_FREE(DONT_VALIDATE)
DONT_VALIDATE(std::string)
DONT_VALIDATE(std::vector<std::string>)
ABSL_FLAGS_INTERNAL_BUILTIN_TYPES(DONT_VALIDATE)
#undef DONT_VALIDATE
return true;
@ -85,7 +84,6 @@ void FlagImpl::Init() {
cur_ = MakeInitValue().release();
StoreAtomic();
inited_.store(true, std::memory_order_release);
InvokeCallback();
}
}
@ -264,8 +262,15 @@ void FlagImpl::StoreAtomic() {
if (data_size <= sizeof(int64_t)) {
int64_t t = 0;
std::memcpy(&t, cur_, data_size);
atomic_.store(t, std::memory_order_release);
atomics_.small_atomic.store(t, std::memory_order_release);
}
#if defined(ABSL_FLAGS_INTERNAL_ATOMIC_DOUBLE_WORD)
else if (data_size <= sizeof(FlagsInternalTwoWordsType)) {
FlagsInternalTwoWordsType t{0, 0};
std::memcpy(&t, cur_, data_size);
atomics_.big_atomic.store(t, std::memory_order_release);
}
#endif
}
void FlagImpl::Write(const void* src, const flags_internal::FlagOpFn src_op) {

View file

@ -20,6 +20,7 @@
#include <cstring>
#include "absl/base/thread_annotations.h"
#include "absl/flags/config.h"
#include "absl/flags/internal/commandlineflag.h"
#include "absl/flags/internal/registry.h"
#include "absl/memory/memory.h"
@ -30,7 +31,61 @@ namespace absl {
ABSL_NAMESPACE_BEGIN
namespace flags_internal {
constexpr int64_t AtomicInit() { return 0xababababababababll; }
// The minimum atomic size we believe to generate lock free code, i.e. all
// trivially copyable types not bigger this size generate lock free code.
static constexpr int kMinLockFreeAtomicSize = 8;
// The same as kMinLockFreeAtomicSize but maximum atomic size. As double words
// might use two registers, we want to dispatch the logic for them.
#if defined(ABSL_FLAGS_INTERNAL_ATOMIC_DOUBLE_WORD)
static constexpr int kMaxLockFreeAtomicSize = 16;
#else
static constexpr int kMaxLockFreeAtomicSize = 8;
#endif
// We can use atomic in cases when it fits in the register, trivially copyable
// in order to make memcpy operations.
template <typename T>
struct IsAtomicFlagTypeTrait {
static constexpr bool value =
(sizeof(T) <= kMaxLockFreeAtomicSize &&
type_traits_internal::is_trivially_copyable<T>::value);
};
// Clang does not always produce cmpxchg16b instruction when alignment of a 16
// bytes type is not 16.
struct alignas(16) FlagsInternalTwoWordsType {
int64_t first;
int64_t second;
};
constexpr bool operator==(const FlagsInternalTwoWordsType& that,
const FlagsInternalTwoWordsType& other) {
return that.first == other.first && that.second == other.second;
}
constexpr bool operator!=(const FlagsInternalTwoWordsType& that,
const FlagsInternalTwoWordsType& other) {
return !(that == other);
}
constexpr int64_t SmallAtomicInit() { return 0xababababababababll; }
template <typename T, typename S = void>
struct BestAtomicType {
using type = int64_t;
static constexpr int64_t AtomicInit() { return SmallAtomicInit(); }
};
template <typename T>
struct BestAtomicType<
T, typename std::enable_if<(kMinLockFreeAtomicSize < sizeof(T) &&
sizeof(T) <= kMaxLockFreeAtomicSize),
void>::type> {
using type = FlagsInternalTwoWordsType;
static constexpr FlagsInternalTwoWordsType AtomicInit() {
return {SmallAtomicInit(), SmallAtomicInit()};
}
};
template <typename T>
class Flag;
@ -182,14 +237,15 @@ class FlagImpl {
// it replaces `dst` with the new value.
bool TryParse(void** dst, absl::string_view value, std::string* err) const
ABSL_EXCLUSIVE_LOCKS_REQUIRED(*DataGuard());
template <typename T>
bool AtomicGet(T* v) const {
const int64_t r = atomic_.load(std::memory_order_acquire);
if (r != flags_internal::AtomicInit()) {
std::memcpy(v, &r, sizeof(T));
using U = flags_internal::BestAtomicType<T>;
const typename U::type r = atomics_.template load<T>();
if (r != U::AtomicInit()) {
std::memcpy(static_cast<void*>(v), &r, sizeof(T));
return true;
}
return false;
}
@ -271,7 +327,34 @@ class FlagImpl {
int64_t counter_ ABSL_GUARDED_BY(*DataGuard()) = 0;
// For some types, a copy of the current value is kept in an atomically
// accessible field.
std::atomic<int64_t> atomic_{flags_internal::AtomicInit()};
union Atomics {
// Using small atomic for small types.
std::atomic<int64_t> small_atomic;
template <typename T,
typename K = typename std::enable_if<
(sizeof(T) <= kMinLockFreeAtomicSize), void>::type>
int64_t load() const {
return small_atomic.load(std::memory_order_acquire);
}
#if defined(ABSL_FLAGS_INTERNAL_ATOMIC_DOUBLE_WORD)
// Using big atomics for big types.
std::atomic<FlagsInternalTwoWordsType> big_atomic;
template <typename T, typename K = typename std::enable_if<
(kMinLockFreeAtomicSize < sizeof(T) &&
sizeof(T) <= kMaxLockFreeAtomicSize),
void>::type>
FlagsInternalTwoWordsType load() const {
return big_atomic.load(std::memory_order_acquire);
}
constexpr Atomics()
: big_atomic{FlagsInternalTwoWordsType{SmallAtomicInit(),
SmallAtomicInit()}} {}
#else
constexpr Atomics() : small_atomic{SmallAtomicInit()} {}
#endif
};
Atomics atomics_{};
struct CallbackData {
FlagCallback func;

View file

@ -280,9 +280,7 @@ void CheckDefaultValuesParsingRoundtrip() {
#define IGNORE_TYPE(T) \
if (flag->IsOfType<T>()) return;
ABSL_FLAGS_INTERNAL_FOR_EACH_LOCK_FREE(IGNORE_TYPE)
IGNORE_TYPE(std::string)
IGNORE_TYPE(std::vector<std::string>)
ABSL_FLAGS_INTERNAL_BUILTIN_TYPES(IGNORE_TYPE)
#undef IGNORE_TYPE
flag->CheckDefaultValueParsingRoundtrip();

View file

@ -37,6 +37,7 @@ cc_library(
"internal/charconv_bigint.h",
"internal/charconv_parse.cc",
"internal/charconv_parse.h",
"internal/escaping.cc",
"internal/memutil.cc",
"internal/memutil.h",
"internal/stl_type_traits.h",
@ -54,6 +55,7 @@ cc_library(
"ascii.h",
"charconv.h",
"escaping.h",
"internal/escaping.h",
"match.h",
"numbers.h",
"str_cat.h",

View file

@ -38,6 +38,8 @@ absl_cc_library(
"internal/charconv_bigint.h"
"internal/charconv_parse.cc"
"internal/charconv_parse.h"
"internal/escaping.cc"
"internal/escaping.h"
"internal/memutil.cc"
"internal/memutil.h"
"internal/stl_type_traits.h"

View file

@ -26,6 +26,7 @@
#include "absl/base/internal/raw_logging.h"
#include "absl/base/internal/unaligned_access.h"
#include "absl/strings/internal/char_map.h"
#include "absl/strings/internal/escaping.h"
#include "absl/strings/internal/resize_uninitialized.h"
#include "absl/strings/internal/utf8.h"
#include "absl/strings/str_cat.h"
@ -764,176 +765,9 @@ constexpr signed char kUnWebSafeBase64[] = {
};
/* clang-format on */
size_t CalculateBase64EscapedLenInternal(size_t input_len, bool do_padding) {
// Base64 encodes three bytes of input at a time. If the input is not
// divisible by three, we pad as appropriate.
//
// (from https://tools.ietf.org/html/rfc3548)
// Special processing is performed if fewer than 24 bits are available
// at the end of the data being encoded. A full encoding quantum is
// always completed at the end of a quantity. When fewer than 24 input
// bits are available in an input group, zero bits are added (on the
// right) to form an integral number of 6-bit groups. Padding at the
// end of the data is performed using the '=' character. Since all base
// 64 input is an integral number of octets, only the following cases
// can arise:
// Base64 encodes each three bytes of input into four bytes of output.
size_t len = (input_len / 3) * 4;
if (input_len % 3 == 0) {
// (from https://tools.ietf.org/html/rfc3548)
// (1) the final quantum of encoding input is an integral multiple of 24
// bits; here, the final unit of encoded output will be an integral
// multiple of 4 characters with no "=" padding,
} else if (input_len % 3 == 1) {
// (from https://tools.ietf.org/html/rfc3548)
// (2) the final quantum of encoding input is exactly 8 bits; here, the
// final unit of encoded output will be two characters followed by two
// "=" padding characters, or
len += 2;
if (do_padding) {
len += 2;
}
} else { // (input_len % 3 == 2)
// (from https://tools.ietf.org/html/rfc3548)
// (3) the final quantum of encoding input is exactly 16 bits; here, the
// final unit of encoded output will be three characters followed by one
// "=" padding character.
len += 3;
if (do_padding) {
len += 1;
}
}
assert(len >= input_len); // make sure we didn't overflow
return len;
}
size_t Base64EscapeInternal(const unsigned char* src, size_t szsrc, char* dest,
size_t szdest, const char* base64,
bool do_padding) {
static const char kPad64 = '=';
if (szsrc * 4 > szdest * 3) return 0;
char* cur_dest = dest;
const unsigned char* cur_src = src;
char* const limit_dest = dest + szdest;
const unsigned char* const limit_src = src + szsrc;
// Three bytes of data encodes to four characters of cyphertext.
// So we can pump through three-byte chunks atomically.
if (szsrc >= 3) { // "limit_src - 3" is UB if szsrc < 3.
while (cur_src < limit_src - 3) { // While we have >= 32 bits.
uint32_t in = absl::big_endian::Load32(cur_src) >> 8;
cur_dest[0] = base64[in >> 18];
in &= 0x3FFFF;
cur_dest[1] = base64[in >> 12];
in &= 0xFFF;
cur_dest[2] = base64[in >> 6];
in &= 0x3F;
cur_dest[3] = base64[in];
cur_dest += 4;
cur_src += 3;
}
}
// To save time, we didn't update szdest or szsrc in the loop. So do it now.
szdest = limit_dest - cur_dest;
szsrc = limit_src - cur_src;
/* now deal with the tail (<=3 bytes) */
switch (szsrc) {
case 0:
// Nothing left; nothing more to do.
break;
case 1: {
// One byte left: this encodes to two characters, and (optionally)
// two pad characters to round out the four-character cypherblock.
if (szdest < 2) return 0;
uint32_t in = cur_src[0];
cur_dest[0] = base64[in >> 2];
in &= 0x3;
cur_dest[1] = base64[in << 4];
cur_dest += 2;
szdest -= 2;
if (do_padding) {
if (szdest < 2) return 0;
cur_dest[0] = kPad64;
cur_dest[1] = kPad64;
cur_dest += 2;
szdest -= 2;
}
break;
}
case 2: {
// Two bytes left: this encodes to three characters, and (optionally)
// one pad character to round out the four-character cypherblock.
if (szdest < 3) return 0;
uint32_t in = absl::big_endian::Load16(cur_src);
cur_dest[0] = base64[in >> 10];
in &= 0x3FF;
cur_dest[1] = base64[in >> 4];
in &= 0x00F;
cur_dest[2] = base64[in << 2];
cur_dest += 3;
szdest -= 3;
if (do_padding) {
if (szdest < 1) return 0;
cur_dest[0] = kPad64;
cur_dest += 1;
szdest -= 1;
}
break;
}
case 3: {
// Three bytes left: same as in the big loop above. We can't do this in
// the loop because the loop above always reads 4 bytes, and the fourth
// byte is past the end of the input.
if (szdest < 4) return 0;
uint32_t in = (cur_src[0] << 16) + absl::big_endian::Load16(cur_src + 1);
cur_dest[0] = base64[in >> 18];
in &= 0x3FFFF;
cur_dest[1] = base64[in >> 12];
in &= 0xFFF;
cur_dest[2] = base64[in >> 6];
in &= 0x3F;
cur_dest[3] = base64[in];
cur_dest += 4;
szdest -= 4;
break;
}
default:
// Should not be reached: blocks of 4 bytes are handled
// in the while loop before this switch statement.
ABSL_RAW_LOG(FATAL, "Logic problem? szsrc = %zu", szsrc);
break;
}
return (cur_dest - dest);
}
constexpr char kBase64Chars[] =
"ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/";
constexpr char kWebSafeBase64Chars[] =
"ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789-_";
template <typename String>
void Base64EscapeInternal(const unsigned char* src, size_t szsrc, String* dest,
bool do_padding, const char* base64_chars) {
const size_t calc_escaped_size =
CalculateBase64EscapedLenInternal(szsrc, do_padding);
strings_internal::STLStringResizeUninitialized(dest, calc_escaped_size);
const size_t escaped_len = Base64EscapeInternal(
src, szsrc, &(*dest)[0], dest->size(), base64_chars, do_padding);
assert(calc_escaped_size == escaped_len);
dest->erase(escaped_len);
}
template <typename String>
bool Base64UnescapeInternal(const char* src, size_t slen, String* dest,
const signed char* unbase64) {
@ -1068,26 +902,30 @@ bool WebSafeBase64Unescape(absl::string_view src, std::string* dest) {
}
void Base64Escape(absl::string_view src, std::string* dest) {
Base64EscapeInternal(reinterpret_cast<const unsigned char*>(src.data()),
src.size(), dest, true, kBase64Chars);
strings_internal::Base64EscapeInternal(
reinterpret_cast<const unsigned char*>(src.data()), src.size(), dest,
true, strings_internal::kBase64Chars);
}
void WebSafeBase64Escape(absl::string_view src, std::string* dest) {
Base64EscapeInternal(reinterpret_cast<const unsigned char*>(src.data()),
src.size(), dest, false, kWebSafeBase64Chars);
strings_internal::Base64EscapeInternal(
reinterpret_cast<const unsigned char*>(src.data()), src.size(), dest,
false, kWebSafeBase64Chars);
}
std::string Base64Escape(absl::string_view src) {
std::string dest;
Base64EscapeInternal(reinterpret_cast<const unsigned char*>(src.data()),
src.size(), &dest, true, kBase64Chars);
strings_internal::Base64EscapeInternal(
reinterpret_cast<const unsigned char*>(src.data()), src.size(), &dest,
true, strings_internal::kBase64Chars);
return dest;
}
std::string WebSafeBase64Escape(absl::string_view src) {
std::string dest;
Base64EscapeInternal(reinterpret_cast<const unsigned char*>(src.data()),
src.size(), &dest, false, kWebSafeBase64Chars);
strings_internal::Base64EscapeInternal(
reinterpret_cast<const unsigned char*>(src.data()), src.size(), &dest,
false, kWebSafeBase64Chars);
return dest;
}

View file

@ -0,0 +1,180 @@
// Copyright 2020 The Abseil Authors.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// https://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#include "absl/strings/internal/escaping.h"
#include "absl/base/internal/endian.h"
#include "absl/base/internal/raw_logging.h"
namespace absl {
ABSL_NAMESPACE_BEGIN
namespace strings_internal {
const char kBase64Chars[] =
"ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/";
size_t CalculateBase64EscapedLenInternal(size_t input_len, bool do_padding) {
// Base64 encodes three bytes of input at a time. If the input is not
// divisible by three, we pad as appropriate.
//
// (from https://tools.ietf.org/html/rfc3548)
// Special processing is performed if fewer than 24 bits are available
// at the end of the data being encoded. A full encoding quantum is
// always completed at the end of a quantity. When fewer than 24 input
// bits are available in an input group, zero bits are added (on the
// right) to form an integral number of 6-bit groups. Padding at the
// end of the data is performed using the '=' character. Since all base
// 64 input is an integral number of octets, only the following cases
// can arise:
// Base64 encodes each three bytes of input into four bytes of output.
size_t len = (input_len / 3) * 4;
if (input_len % 3 == 0) {
// (from https://tools.ietf.org/html/rfc3548)
// (1) the final quantum of encoding input is an integral multiple of 24
// bits; here, the final unit of encoded output will be an integral
// multiple of 4 characters with no "=" padding,
} else if (input_len % 3 == 1) {
// (from https://tools.ietf.org/html/rfc3548)
// (2) the final quantum of encoding input is exactly 8 bits; here, the
// final unit of encoded output will be two characters followed by two
// "=" padding characters, or
len += 2;
if (do_padding) {
len += 2;
}
} else { // (input_len % 3 == 2)
// (from https://tools.ietf.org/html/rfc3548)
// (3) the final quantum of encoding input is exactly 16 bits; here, the
// final unit of encoded output will be three characters followed by one
// "=" padding character.
len += 3;
if (do_padding) {
len += 1;
}
}
assert(len >= input_len); // make sure we didn't overflow
return len;
}
size_t Base64EscapeInternal(const unsigned char* src, size_t szsrc, char* dest,
size_t szdest, const char* base64,
bool do_padding) {
static const char kPad64 = '=';
if (szsrc * 4 > szdest * 3) return 0;
char* cur_dest = dest;
const unsigned char* cur_src = src;
char* const limit_dest = dest + szdest;
const unsigned char* const limit_src = src + szsrc;
// Three bytes of data encodes to four characters of cyphertext.
// So we can pump through three-byte chunks atomically.
if (szsrc >= 3) { // "limit_src - 3" is UB if szsrc < 3.
while (cur_src < limit_src - 3) { // While we have >= 32 bits.
uint32_t in = absl::big_endian::Load32(cur_src) >> 8;
cur_dest[0] = base64[in >> 18];
in &= 0x3FFFF;
cur_dest[1] = base64[in >> 12];
in &= 0xFFF;
cur_dest[2] = base64[in >> 6];
in &= 0x3F;
cur_dest[3] = base64[in];
cur_dest += 4;
cur_src += 3;
}
}
// To save time, we didn't update szdest or szsrc in the loop. So do it now.
szdest = limit_dest - cur_dest;
szsrc = limit_src - cur_src;
/* now deal with the tail (<=3 bytes) */
switch (szsrc) {
case 0:
// Nothing left; nothing more to do.
break;
case 1: {
// One byte left: this encodes to two characters, and (optionally)
// two pad characters to round out the four-character cypherblock.
if (szdest < 2) return 0;
uint32_t in = cur_src[0];
cur_dest[0] = base64[in >> 2];
in &= 0x3;
cur_dest[1] = base64[in << 4];
cur_dest += 2;
szdest -= 2;
if (do_padding) {
if (szdest < 2) return 0;
cur_dest[0] = kPad64;
cur_dest[1] = kPad64;
cur_dest += 2;
szdest -= 2;
}
break;
}
case 2: {
// Two bytes left: this encodes to three characters, and (optionally)
// one pad character to round out the four-character cypherblock.
if (szdest < 3) return 0;
uint32_t in = absl::big_endian::Load16(cur_src);
cur_dest[0] = base64[in >> 10];
in &= 0x3FF;
cur_dest[1] = base64[in >> 4];
in &= 0x00F;
cur_dest[2] = base64[in << 2];
cur_dest += 3;
szdest -= 3;
if (do_padding) {
if (szdest < 1) return 0;
cur_dest[0] = kPad64;
cur_dest += 1;
szdest -= 1;
}
break;
}
case 3: {
// Three bytes left: same as in the big loop above. We can't do this in
// the loop because the loop above always reads 4 bytes, and the fourth
// byte is past the end of the input.
if (szdest < 4) return 0;
uint32_t in = (cur_src[0] << 16) + absl::big_endian::Load16(cur_src + 1);
cur_dest[0] = base64[in >> 18];
in &= 0x3FFFF;
cur_dest[1] = base64[in >> 12];
in &= 0xFFF;
cur_dest[2] = base64[in >> 6];
in &= 0x3F;
cur_dest[3] = base64[in];
cur_dest += 4;
szdest -= 4;
break;
}
default:
// Should not be reached: blocks of 4 bytes are handled
// in the while loop before this switch statement.
ABSL_RAW_LOG(FATAL, "Logic problem? szsrc = %zu", szsrc);
break;
}
return (cur_dest - dest);
}
} // namespace strings_internal
ABSL_NAMESPACE_END
} // namespace absl

View file

@ -0,0 +1,58 @@
// Copyright 2020 The Abseil Authors.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// https://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#ifndef ABSL_STRINGS_INTERNAL_ESCAPING_H_
#define ABSL_STRINGS_INTERNAL_ESCAPING_H_
#include <cassert>
#include "absl/strings/internal/resize_uninitialized.h"
namespace absl {
ABSL_NAMESPACE_BEGIN
namespace strings_internal {
ABSL_CONST_INIT extern const char kBase64Chars[];
// Calculates how long a string will be when it is base64 encoded given its
// length and whether or not the result should be padded.
size_t CalculateBase64EscapedLenInternal(size_t input_len, bool do_padding);
// Base64-encodes `src` using the alphabet provided in `base64` and writes the
// result to `dest`. If `do_padding` is true, `dest` is padded with '=' chars
// until its length is a multiple of 3. Returns the length of `dest`.
size_t Base64EscapeInternal(const unsigned char* src, size_t szsrc, char* dest,
size_t szdest, const char* base64, bool do_padding);
// Base64-encodes `src` using the alphabet provided in `base64` and writes the
// result to `dest`. If `do_padding` is true, `dest` is padded with '=' chars
// until its length is a multiple of 3.
template <typename String>
void Base64EscapeInternal(const unsigned char* src, size_t szsrc, String* dest,
bool do_padding, const char* base64_chars) {
const size_t calc_escaped_size =
CalculateBase64EscapedLenInternal(szsrc, do_padding);
STLStringResizeUninitialized(dest, calc_escaped_size);
const size_t escaped_len = Base64EscapeInternal(
src, szsrc, &(*dest)[0], dest->size(), base64_chars, do_padding);
assert(calc_escaped_size == escaped_len);
dest->erase(escaped_len);
}
} // namespace strings_internal
ABSL_NAMESPACE_END
} // namespace absl
#endif // ABSL_STRINGS_INTERNAL_ESCAPING_H_

View file

@ -122,7 +122,7 @@ ABSL_NAMESPACE_BEGIN
//
// You may create a null `string_view` in two ways:
//
// absl::string_view sv();
// absl::string_view sv;
// absl::string_view sv(nullptr, 0);
//
// For the above, `sv.data() == nullptr`, `sv.length() == 0`, and
@ -605,7 +605,7 @@ inline string_view ClippedSubstr(string_view s, size_t pos,
// Creates an `absl::string_view` from a pointer `p` even if it's null-valued.
// This function should be used where an `absl::string_view` can be created from
// a possibly-null pointer.
inline string_view NullSafeStringView(const char* p) {
constexpr string_view NullSafeStringView(const char* p) {
return p ? string_view(p) : string_view();
}

View file

@ -931,6 +931,31 @@ TEST(StringViewTest, NullSafeStringView) {
}
}
TEST(StringViewTest, ConstexprNullSafeStringView) {
{
constexpr absl::string_view s = absl::NullSafeStringView(nullptr);
EXPECT_EQ(nullptr, s.data());
EXPECT_EQ(0, s.size());
EXPECT_EQ(absl::string_view(), s);
}
#if !defined(_MSC_VER) || _MSC_VER >= 1910
// MSVC 2017+ is required for good constexpr string_view support.
// See the implementation of `absl::string_view::StrlenInternal()`.
{
static constexpr char kHi[] = "hi";
absl::string_view s = absl::NullSafeStringView(kHi);
EXPECT_EQ(kHi, s.data());
EXPECT_EQ(strlen(kHi), s.size());
EXPECT_EQ(absl::string_view("hi"), s);
}
{
constexpr absl::string_view s = absl::NullSafeStringView("hello");
EXPECT_EQ(s.size(), 5);
EXPECT_EQ("hello", s);
}
#endif
}
TEST(StringViewTest, ConstexprCompiles) {
constexpr absl::string_view sp;
#ifdef ABSL_HAVE_STRING_VIEW_FROM_NULLPTR

View file

@ -22,6 +22,7 @@
#define ABSL_OPTION_USE_STD_OPTIONAL 0
#define ABSL_OPTION_USE_STD_STRING_VIEW 0
#define ABSL_OPTION_USE_STD_VARIANT 0
#define ABSL_OPTION_USE_INLINE_NAMESPACE 1
#define ABSL_OPTION_INLINE_NAMESPACE_NAME ns
#endif // ABSL_BASE_OPTIONS_H_

View file

@ -55,7 +55,8 @@ for std in ${STD}; do
for exceptions_mode in ${EXCEPTIONS_MODE}; do
echo "--------------------------------------------------------------------"
time docker run \
--volume="${ABSEIL_ROOT}:/abseil-cpp:ro" \
--volume="${ABSEIL_ROOT}:/abseil-cpp-ro:ro" \
--tmpfs=/abseil-cpp \
--workdir=/abseil-cpp \
--cap-add=SYS_PTRACE \
--rm \
@ -63,18 +64,23 @@ for std in ${STD}; do
-e BAZEL_CXXOPTS="-std=${std}" \
${DOCKER_EXTRA_ARGS:-} \
${DOCKER_CONTAINER} \
/usr/local/bin/bazel test ... \
--compilation_mode="${compilation_mode}" \
--copt="${exceptions_mode}" \
--copt=-Werror \
--define="absl=1" \
--keep_going \
--show_timestamps \
--test_env="GTEST_INSTALL_FAILURE_SIGNAL_HANDLER=1" \
--test_env="TZDIR=/abseil-cpp/absl/time/internal/cctz/testdata/zoneinfo" \
--test_output=errors \
--test_tag_filters=-benchmark \
${BAZEL_EXTRA_ARGS:-}
/bin/sh -c "
cp -r /abseil-cpp-ro/* /abseil-cpp/
if [ -n \"${ALTERNATE_OPTIONS:-}\" ]; then
cp ${ALTERNATE_OPTIONS:-} absl/base/options.h || exit 1
fi
/usr/local/bin/bazel test ... \
--compilation_mode=\"${compilation_mode}\" \
--copt=\"${exceptions_mode}\" \
--copt=-Werror \
--define=\"absl=1\" \
--keep_going \
--show_timestamps \
--test_env=\"GTEST_INSTALL_FAILURE_SIGNAL_HANDLER=1\" \
--test_env=\"TZDIR=/abseil-cpp/absl/time/internal/cctz/testdata/zoneinfo\" \
--test_output=errors \
--test_tag_filters=-benchmark \
${BAZEL_EXTRA_ARGS:-}"
done
done
done

View file

@ -41,6 +41,10 @@ echo "---------------"
cd ${ABSEIL_ROOT}
if [ -n "${ALTERNATE_OPTIONS:-}" ]; then
cp ${ALTERNATE_OPTIONS:-} absl/base/options.h || exit 1
fi
${BAZEL_BIN} test ... \
--copt=-Werror \
--keep_going \