maurice/snix
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Export of internal Abseil changes.

Browse source 
--
babbb6421068af3831870fd5995444437ace6769 by Derek Mauro <dmauro@google.com>:

Rollback of:

Make raw_hash_set_test less flaky.

Split the timing loop into chunks so that we are less suceptible to
antogantistic processes.

PiperOrigin-RevId: 243854490

--
a2711f17a712f6d09799bf32363d67526737b486 by CJ Johnson <johnsoncj@google.com>:

Relocates IsAtLeastForwardIterator to internal/inlined_vector.h

PiperOrigin-RevId: 243846090

--
6c14cdbeb9a61022c27f8957654f930d8abf2fc1 by Matt Kulukundis <kfm@google.com>:

Make raw_hash_set_test less flaky.

Split the timing loop into chunks so that we are less suceptible to
antogantistic processes.

PiperOrigin-RevId: 243824289

--
ee6072a6b6e0ac653622524ceb09db3b9e870f96 by Samuel Benzaquen <sbenza@google.com>:

Improve format parser performance.
Replace the main switch with a lookup in the existing tag table.
Improve the ABI of ConsumeUnboundConversion a little.

PiperOrigin-RevId: 243824112

--
24b9e6476dfa4be8d644359eab8ac6816492f187 by Abseil Team <absl-team@google.com>:

Fix DR numbers: 3800 ? 3080, 3801 ? 3081.

PiperOrigin-RevId: 243804213

--
0660404074707e197684f07cc0bffe4a9c35cd2f by Abseil Team <absl-team@google.com>:

Internal change.

PiperOrigin-RevId: 243757359

--
ba0f5bb9b8584d75c4ffc44ff3cb8c691796ffc6 by Xiaoyi Zhang <zhangxy@google.com>:

Consolidate ABSL_INTERNAL_UNALIGNED_* implementation into memcpy.
The compiler should be good enough to optimize these operations.
See https://github.com/abseil/abseil-cpp/issues/269 for background.

PiperOrigin-RevId: 243323941

--
00853a8756548df7217513c562d604b4ee5c6ab9 by Eric Fiselier <ericwf@google.com>:

Reexport memory.h from optional.h for compatibility between libc++ and
libstdc++.

PiperOrigin-RevId: 243313425
GitOrigin-RevId: babbb6421068af3831870fd5995444437ace6769
Change-Id: Ic53c127ad857a431ad60c98b27cc585fed50a3e3
This commit is contained in:
Abseil Team 2019-04-16 12:11:35 -07:00 • committed by Alex Strelnikov
parent a02f62f456
commit d902eb869b
10 changed files with 186 additions and 328 deletions
Download patch file Download diff file Expand all files Collapse all files

165
absl/base/internal/unaligned_access.h
View file

@ -25,15 +25,6 @@
// unaligned APIs // unaligned APIs
// Portable handling of unaligned loads, stores, and copies. // Portable handling of unaligned loads, stores, and copies.
// On some platforms, like ARM, the copy functions can be more efficient
// then a load and a store.
//
// It is possible to implement all of these these using constant-length memcpy
// calls, which is portable and will usually be inlined into simple loads and
// stores if the architecture supports it. However, such inlining usually
// happens in a pass that's quite late in compilation, which means the resulting
// loads and stores cannot participate in many other optimizations, leading to
// overall worse code.
// The unaligned API is C++ only. The declarations use C++ features // The unaligned API is C++ only. The declarations use C++ features
// (namespaces, inline) which are absent or incompatible in C. // (namespaces, inline) which are absent or incompatible in C.
@ -108,164 +99,8 @@ inline void UnalignedStore64(void *p, uint64_t v) {
#define ABSL_INTERNAL_UNALIGNED_STORE64(_p, _val) \ #define ABSL_INTERNAL_UNALIGNED_STORE64(_p, _val) \
(absl::base_internal::UnalignedStore64(_p, _val)) (absl::base_internal::UnalignedStore64(_p, _val))
#elif defined(UNDEFINED_BEHAVIOR_SANITIZER)
namespace absl {
namespace base_internal {
inline uint16_t UnalignedLoad16(const void *p) {
uint16_t t;
memcpy(&t, p, sizeof t);
return t;
}
inline uint32_t UnalignedLoad32(const void *p) {
uint32_t t;
memcpy(&t, p, sizeof t);
return t;
}
inline uint64_t UnalignedLoad64(const void *p) {
uint64_t t;
memcpy(&t, p, sizeof t);
return t;
}
inline void UnalignedStore16(void *p, uint16_t v) { memcpy(p, &v, sizeof v); }
inline void UnalignedStore32(void *p, uint32_t v) { memcpy(p, &v, sizeof v); }
inline void UnalignedStore64(void *p, uint64_t v) { memcpy(p, &v, sizeof v); }
} // namespace base_internal
} // namespace absl
#define ABSL_INTERNAL_UNALIGNED_LOAD16(_p) \
(absl::base_internal::UnalignedLoad16(_p))
#define ABSL_INTERNAL_UNALIGNED_LOAD32(_p) \
(absl::base_internal::UnalignedLoad32(_p))
#define ABSL_INTERNAL_UNALIGNED_LOAD64(_p) \
(absl::base_internal::UnalignedLoad64(_p))
#define ABSL_INTERNAL_UNALIGNED_STORE16(_p, _val) \
(absl::base_internal::UnalignedStore16(_p, _val))
#define ABSL_INTERNAL_UNALIGNED_STORE32(_p, _val) \
(absl::base_internal::UnalignedStore32(_p, _val))
#define ABSL_INTERNAL_UNALIGNED_STORE64(_p, _val) \
(absl::base_internal::UnalignedStore64(_p, _val))
#elif defined(__x86_64__) || defined(_M_X64) || defined(__i386) || \
defined(_M_IX86) || defined(__ppc__) || defined(__PPC__) || \
defined(__ppc64__) || defined(__PPC64__)
// x86 and x86-64 can perform unaligned loads/stores directly;
// modern PowerPC hardware can also do unaligned integer loads and stores;
// but note: the FPU still sends unaligned loads and stores to a trap handler!
#define ABSL_INTERNAL_UNALIGNED_LOAD16(_p) \
(*reinterpret_cast<const uint16_t *>(_p))
#define ABSL_INTERNAL_UNALIGNED_LOAD32(_p) \
(*reinterpret_cast<const uint32_t *>(_p))
#define ABSL_INTERNAL_UNALIGNED_LOAD64(_p) \
(*reinterpret_cast<const uint64_t *>(_p))
#define ABSL_INTERNAL_UNALIGNED_STORE16(_p, _val) \
(*reinterpret_cast<uint16_t *>(_p) = (_val))
#define ABSL_INTERNAL_UNALIGNED_STORE32(_p, _val) \
(*reinterpret_cast<uint32_t *>(_p) = (_val))
#define ABSL_INTERNAL_UNALIGNED_STORE64(_p, _val) \
(*reinterpret_cast<uint64_t *>(_p) = (_val))
#elif defined(__arm__) && \
!defined(__ARM_ARCH_5__) && \
!defined(__ARM_ARCH_5T__) && \
!defined(__ARM_ARCH_5TE__) && \
!defined(__ARM_ARCH_5TEJ__) && \
!defined(__ARM_ARCH_6__) && \
!defined(__ARM_ARCH_6J__) && \
!defined(__ARM_ARCH_6K__) && \
!defined(__ARM_ARCH_6Z__) && \
!defined(__ARM_ARCH_6ZK__) && \
!defined(__ARM_ARCH_6T2__)
// ARMv7 and newer support native unaligned accesses, but only of 16-bit
// and 32-bit values (not 64-bit); older versions either raise a fatal signal,
// do an unaligned read and rotate the words around a bit, or do the reads very
// slowly (trip through kernel mode). There's no simple #define that says just
// "ARMv7 or higher", so we have to filter away all ARMv5 and ARMv6
// sub-architectures. Newer gcc (>= 4.6) set an __ARM_FEATURE_ALIGNED #define,
// so in time, maybe we can move on to that.
//
// This is a mess, but there's not much we can do about it.
//
// To further complicate matters, only LDR instructions (single reads) are
// allowed to be unaligned, not LDRD (two reads) or LDM (many reads). Unless we
// explicitly tell the compiler that these accesses can be unaligned, it can and
// will combine accesses. On armcc, the way to signal this is done by accessing
// through the type (uint32_t __packed *), but GCC has no such attribute
// (it ignores __attribute__((packed)) on individual variables). However,
// we can tell it that a _struct_ is unaligned, which has the same effect,
// so we do that.
namespace absl {
namespace base_internal {
struct Unaligned16Struct {
uint16_t value;
uint8_t dummy; // To make the size non-power-of-two.
} ABSL_ATTRIBUTE_PACKED;
struct Unaligned32Struct {
uint32_t value;
uint8_t dummy; // To make the size non-power-of-two.
} ABSL_ATTRIBUTE_PACKED;
} // namespace base_internal
} // namespace absl
#define ABSL_INTERNAL_UNALIGNED_LOAD16(_p) \
((reinterpret_cast<const ::absl::base_internal::Unaligned16Struct *>(_p)) \
->value)
#define ABSL_INTERNAL_UNALIGNED_LOAD32(_p) \
((reinterpret_cast<const ::absl::base_internal::Unaligned32Struct *>(_p)) \
->value)
#define ABSL_INTERNAL_UNALIGNED_STORE16(_p, _val) \
((reinterpret_cast< ::absl::base_internal::Unaligned16Struct *>(_p)) \
->value = (_val))
#define ABSL_INTERNAL_UNALIGNED_STORE32(_p, _val) \
((reinterpret_cast< ::absl::base_internal::Unaligned32Struct *>(_p)) \
->value = (_val))
namespace absl {
namespace base_internal {
inline uint64_t UnalignedLoad64(const void *p) {
uint64_t t;
memcpy(&t, p, sizeof t);
return t;
}
inline void UnalignedStore64(void *p, uint64_t v) { memcpy(p, &v, sizeof v); }
} // namespace base_internal
} // namespace absl
#define ABSL_INTERNAL_UNALIGNED_LOAD64(_p) \
(absl::base_internal::UnalignedLoad64(_p))
#define ABSL_INTERNAL_UNALIGNED_STORE64(_p, _val) \
(absl::base_internal::UnalignedStore64(_p, _val))
#else #else
// ABSL_INTERNAL_NEED_ALIGNED_LOADS is defined when the underlying platform
// doesn't support unaligned access.
#define ABSL_INTERNAL_NEED_ALIGNED_LOADS
// These functions are provided for architectures that don't support
// unaligned loads and stores.
namespace absl { namespace absl {
namespace base_internal { namespace base_internal {

25
absl/container/inlined_vector.h
View file

@ -73,17 +73,12 @@ class InlinedVector {
using AllocatorTraits = typename Storage::AllocatorTraits; using AllocatorTraits = typename Storage::AllocatorTraits;
template <typename Iterator> template <typename Iterator>
using IsAtLeastForwardIterator = std::is_convertible< using EnableIfAtLeastForwardIterator = absl::enable_if_t<
typename std::iterator_traits<Iterator>::iterator_category, inlined_vector_internal::IsAtLeastForwardIterator<Iterator>::value>;
std::forward_iterator_tag>;
template <typename Iterator> template <typename Iterator>
using EnableIfAtLeastForwardIterator = using DisableIfAtLeastForwardIterator = absl::enable_if_t<
absl::enable_if_t<IsAtLeastForwardIterator<Iterator>::value>; !inlined_vector_internal::IsAtLeastForwardIterator<Iterator>::value>;
template <typename Iterator>
using DisableIfAtLeastForwardIterator =
absl::enable_if_t<!IsAtLeastForwardIterator<Iterator>::value>;
using rvalue_reference = typename Storage::rvalue_reference; using rvalue_reference = typename Storage::rvalue_reference;
@ -1060,7 +1055,9 @@ class InlinedVector {
template <typename ForwardIt> template <typename ForwardIt>
void AssignForwardRange(ForwardIt first, ForwardIt last) { void AssignForwardRange(ForwardIt first, ForwardIt last) {
static_assert(IsAtLeastForwardIterator<ForwardIt>::value, ""); static_assert(absl::inlined_vector_internal::IsAtLeastForwardIterator<
ForwardIt>::value,
"");
auto length = std::distance(first, last); auto length = std::distance(first, last);
@ -1084,7 +1081,9 @@ class InlinedVector {
template <typename ForwardIt> template <typename ForwardIt>
void AppendForwardRange(ForwardIt first, ForwardIt last) { void AppendForwardRange(ForwardIt first, ForwardIt last) {
static_assert(IsAtLeastForwardIterator<ForwardIt>::value, ""); static_assert(absl::inlined_vector_internal::IsAtLeastForwardIterator<
ForwardIt>::value,
"");
auto length = std::distance(first, last); auto length = std::distance(first, last);
reserve(size() + length); reserve(size() + length);
@ -1113,7 +1112,9 @@ class InlinedVector {
template <typename ForwardIt> template <typename ForwardIt>
iterator InsertWithForwardRange(const_iterator position, ForwardIt first, iterator InsertWithForwardRange(const_iterator position, ForwardIt first,
ForwardIt last) { ForwardIt last) {
static_assert(IsAtLeastForwardIterator<ForwardIt>::value, ""); static_assert(absl::inlined_vector_internal::IsAtLeastForwardIterator<
ForwardIt>::value,
"");
assert(position >= begin() && position <= end()); assert(position >= begin() && position <= end());
if (ABSL_PREDICT_FALSE(first == last)) if (ABSL_PREDICT_FALSE(first == last))

9
absl/container/internal/inlined_vector.h
View file

@ -26,6 +26,11 @@
namespace absl { namespace absl {
namespace inlined_vector_internal { namespace inlined_vector_internal {
template <typename Iterator>
using IsAtLeastForwardIterator = std::is_convertible<
typename std::iterator_traits<Iterator>::iterator_category,
std::forward_iterator_tag>;
template <typename InlinedVector> template <typename InlinedVector>
class Storage; class Storage;
@ -89,9 +94,7 @@ class Storage<InlinedVector<T, N, A>> {
void AddSize(size_type count) { GetSizeAndIsAllocated() += count << 1; } void AddSize(size_type count) { GetSizeAndIsAllocated() += count << 1; }
void SetAllocatedData(pointer data) { void SetAllocatedData(pointer data) { data_.allocated.allocated_data = data; }
data_.allocated.allocated_data = data;
}
void SetAllocatedCapacity(size_type capacity) { void SetAllocatedCapacity(size_type capacity) {
data_.allocated.allocated_capacity = capacity; data_.allocated.allocated_capacity = capacity;

2
absl/hash/BUILD.bazel
View file

@ -35,10 +35,10 @@ cc_library(
copts = ABSL_DEFAULT_COPTS, copts = ABSL_DEFAULT_COPTS,
linkopts = ABSL_DEFAULT_LINKOPTS, linkopts = ABSL_DEFAULT_LINKOPTS,
deps = [ deps = [
":city",
"//absl/base:core_headers", "//absl/base:core_headers",
"//absl/base:endian", "//absl/base:endian",
"//absl/container:fixed_array", "//absl/container:fixed_array",
"//absl/hash:city",
"//absl/meta:type_traits", "//absl/meta:type_traits",
"//absl/numeric:int128", "//absl/numeric:int128",
"//absl/strings", "//absl/strings",

4
absl/strings/charconv.h
View file

@ -49,9 +49,9 @@ struct from_chars_result {
// this only supports the `double` and `float` types. // this only supports the `double` and `float` types.
// //
// This interface incorporates the proposed resolutions for library issues // This interface incorporates the proposed resolutions for library issues
// DR 3800 and DR 3801. If these are adopted with different wording, // DR 3080 and DR 3081. If these are adopted with different wording,
// Abseil's behavior will change to match the standard. (The behavior most // Abseil's behavior will change to match the standard. (The behavior most
// likely to change is for DR 3801, which says what `value` will be set to in // likely to change is for DR 3081, which says what `value` will be set to in
// the case of overflow and underflow. Code that wants to avoid possible // the case of overflow and underflow. Code that wants to avoid possible
// breaking changes in this area should not depend on `value` when the returned // breaking changes in this area should not depend on `value` when the returned
// from_chars_result indicates a range error.) // from_chars_result indicates a range error.)

9
absl/strings/internal/str_format/bind_test.cc
View file

@ -9,16 +9,11 @@ namespace absl {
namespace str_format_internal { namespace str_format_internal {
namespace { namespace {
template <typename T, size_t N>
size_t ArraySize(T (&)[N]) {
return N;
}
class FormatBindTest : public ::testing::Test { class FormatBindTest : public ::testing::Test {
public: public:
bool Extract(const char *s, UnboundConversion *props, int *next) const { bool Extract(const char *s, UnboundConversion *props, int *next) const {
absl::string_view src = s; return ConsumeUnboundConversion(s, s + strlen(s), props, next) ==
return ConsumeUnboundConversion(&src, props, next) && src.empty(); s + strlen(s);
} }
}; };

137
absl/strings/internal/str_format/parser.cc
View file

@ -15,6 +15,44 @@
namespace absl { namespace absl {
namespace str_format_internal { namespace str_format_internal {
using CC = ConversionChar::Id;
using LM = LengthMod::Id;
ABSL_CONST_INIT const ConvTag kTags[256] = {
{}, {}, {}, {}, {}, {}, {}, {}, // 00-07
{}, {}, {}, {}, {}, {}, {}, {}, // 08-0f
{}, {}, {}, {}, {}, {}, {}, {}, // 10-17
{}, {}, {}, {}, {}, {}, {}, {}, // 18-1f
{}, {}, {}, {}, {}, {}, {}, {}, // 20-27
{}, {}, {}, {}, {}, {}, {}, {}, // 28-2f
{}, {}, {}, {}, {}, {}, {}, {}, // 30-37
{}, {}, {}, {}, {}, {}, {}, {}, // 38-3f
{}, CC::A, {}, CC::C, {}, CC::E, CC::F, CC::G, // @ABCDEFG
{}, {}, {}, {}, LM::L, {}, {}, {}, // HIJKLMNO
{}, {}, {}, CC::S, {}, {}, {}, {}, // PQRSTUVW
CC::X, {}, {}, {}, {}, {}, {}, {}, // XYZ[\]^_
{}, CC::a, {}, CC::c, CC::d, CC::e, CC::f, CC::g, // `abcdefg
LM::h, CC::i, LM::j, {}, LM::l, {}, CC::n, CC::o, // hijklmno
CC::p, LM::q, {}, CC::s, LM::t, CC::u, {}, {}, // pqrstuvw
CC::x, {}, LM::z, {}, {}, {}, {}, {}, // xyz{|}!
{}, {}, {}, {}, {}, {}, {}, {}, // 80-87
{}, {}, {}, {}, {}, {}, {}, {}, // 88-8f
{}, {}, {}, {}, {}, {}, {}, {}, // 90-97
{}, {}, {}, {}, {}, {}, {}, {}, // 98-9f
{}, {}, {}, {}, {}, {}, {}, {}, // a0-a7
{}, {}, {}, {}, {}, {}, {}, {}, // a8-af
{}, {}, {}, {}, {}, {}, {}, {}, // b0-b7
{}, {}, {}, {}, {}, {}, {}, {}, // b8-bf
{}, {}, {}, {}, {}, {}, {}, {}, // c0-c7
{}, {}, {}, {}, {}, {}, {}, {}, // c8-cf
{}, {}, {}, {}, {}, {}, {}, {}, // d0-d7
{}, {}, {}, {}, {}, {}, {}, {}, // d8-df
{}, {}, {}, {}, {}, {}, {}, {}, // e0-e7
{}, {}, {}, {}, {}, {}, {}, {}, // e8-ef
{}, {}, {}, {}, {}, {}, {}, {}, // f0-f7
{}, {}, {}, {}, {}, {}, {}, {}, // f8-ff
};
namespace { namespace {
bool CheckFastPathSetting(const UnboundConversion& conv) { bool CheckFastPathSetting(const UnboundConversion& conv) {
@ -36,59 +74,16 @@ bool CheckFastPathSetting(const UnboundConversion& conv) {
return should_be_basic == conv.flags.basic; return should_be_basic == conv.flags.basic;
} }
// Keep a single table for all the conversion chars and length modifiers.
// We invert the length modifiers to make them negative so that we can easily
// test for them.
// Everything else is `none`, which is a negative constant.
using CC = ConversionChar::Id;
using LM = LengthMod::Id;
static constexpr std::int8_t none = -128;
static constexpr std::int8_t kIds[] = {
none, none, none, none, none, none, none, none, // 00-07
none, none, none, none, none, none, none, none, // 08-0f
none, none, none, none, none, none, none, none, // 10-17
none, none, none, none, none, none, none, none, // 18-1f
none, none, none, none, none, none, none, none, // 20-27
none, none, none, none, none, none, none, none, // 28-2f
none, none, none, none, none, none, none, none, // 30-37
none, none, none, none, none, none, none, none, // 38-3f
none, CC::A, none, CC::C, none, CC::E, CC::F, CC::G, // @ABCDEFG
none, none, none, none, ~LM::L, none, none, none, // HIJKLMNO
none, none, none, CC::S, none, none, none, none, // PQRSTUVW
CC::X, none, none, none, none, none, none, none, // XYZ[\]^_
none, CC::a, none, CC::c, CC::d, CC::e, CC::f, CC::g, // `abcdefg
~LM::h, CC::i, ~LM::j, none, ~LM::l, none, CC::n, CC::o, // hijklmno
CC::p, ~LM::q, none, CC::s, ~LM::t, CC::u, none, none, // pqrstuvw
CC::x, none, ~LM::z, none, none, none, none, none, // xyz{|}~!
none, none, none, none, none, none, none, none, // 80-87
none, none, none, none, none, none, none, none, // 88-8f
none, none, none, none, none, none, none, none, // 90-97
none, none, none, none, none, none, none, none, // 98-9f
none, none, none, none, none, none, none, none, // a0-a7
none, none, none, none, none, none, none, none, // a8-af
none, none, none, none, none, none, none, none, // b0-b7
none, none, none, none, none, none, none, none, // b8-bf
none, none, none, none, none, none, none, none, // c0-c7
none, none, none, none, none, none, none, none, // c8-cf
none, none, none, none, none, none, none, none, // d0-d7
none, none, none, none, none, none, none, none, // d8-df
none, none, none, none, none, none, none, none, // e0-e7
none, none, none, none, none, none, none, none, // e8-ef
none, none, none, none, none, none, none, none, // f0-f7
none, none, none, none, none, none, none, none, // f8-ff
};
template <bool is_positional> template <bool is_positional>
bool ConsumeConversion(string_view *src, UnboundConversion *conv, const char *ConsumeConversion(const char *pos, const char *const end,
int *next_arg) { UnboundConversion *conv, int *next_arg) {
const char *pos = src->data(); const char* const original_pos = pos;
const char *const end = pos + src->size();
char c; char c;
// Read the next char into `c` and update `pos`. Returns false if there are // Read the next char into `c` and update `pos`. Returns false if there are
// no more chars to read. // no more chars to read.
#define ABSL_FORMAT_PARSER_INTERNAL_GET_CHAR() \ #define ABSL_FORMAT_PARSER_INTERNAL_GET_CHAR() \
do { \ do { \
if (ABSL_PREDICT_FALSE(pos == end)) return false; \ if (ABSL_PREDICT_FALSE(pos == end)) return nullptr; \
c = *pos++; \ c = *pos++; \
} while (0) } while (0)
@ -111,10 +106,10 @@ bool ConsumeConversion(string_view *src, UnboundConversion *conv,
if (is_positional) { if (is_positional) {
ABSL_FORMAT_PARSER_INTERNAL_GET_CHAR(); ABSL_FORMAT_PARSER_INTERNAL_GET_CHAR();
if (ABSL_PREDICT_FALSE(c < '1' || c > '9')) return false; if (ABSL_PREDICT_FALSE(c < '1' || c > '9')) return nullptr;
conv->arg_position = parse_digits(); conv->arg_position = parse_digits();
assert(conv->arg_position > 0); assert(conv->arg_position > 0);
if (ABSL_PREDICT_FALSE(c != '$')) return false; if (ABSL_PREDICT_FALSE(c != '$')) return nullptr;
} }
ABSL_FORMAT_PARSER_INTERNAL_GET_CHAR(); ABSL_FORMAT_PARSER_INTERNAL_GET_CHAR();
@ -129,10 +124,9 @@ bool ConsumeConversion(string_view *src, UnboundConversion *conv,
conv->flags.basic = false; conv->flags.basic = false;
for (; c <= '0';) { for (; c <= '0';) {
// FIXME: We might be able to speed this up reusing the kIds lookup table // FIXME: We might be able to speed this up reusing the lookup table from
// from above. // above. It might require changing Flags to be a plain integer where we
// It might require changing Flags to be a plain integer where we can |= a // can |= a value.
// value.
switch (c) { switch (c) {
case '-': case '-':
conv->flags.left = true; conv->flags.left = true;
@ -160,20 +154,20 @@ flags_done:
if (c >= '0') { if (c >= '0') {
int maybe_width = parse_digits(); int maybe_width = parse_digits();
if (!is_positional && c == '$') { if (!is_positional && c == '$') {
if (ABSL_PREDICT_FALSE(*next_arg != 0)) return false; if (ABSL_PREDICT_FALSE(*next_arg != 0)) return nullptr;
// Positional conversion. // Positional conversion.
*next_arg = -1; *next_arg = -1;
conv->flags = Flags(); conv->flags = Flags();
conv->flags.basic = true; conv->flags.basic = true;
return ConsumeConversion<true>(src, conv, next_arg); return ConsumeConversion<true>(original_pos, end, conv, next_arg);
} }
conv->width.set_value(maybe_width); conv->width.set_value(maybe_width);
} else if (c == '*') { } else if (c == '*') {
ABSL_FORMAT_PARSER_INTERNAL_GET_CHAR(); ABSL_FORMAT_PARSER_INTERNAL_GET_CHAR();
if (is_positional) { if (is_positional) {
if (ABSL_PREDICT_FALSE(c < '1' || c > '9')) return false; if (ABSL_PREDICT_FALSE(c < '1' || c > '9')) return nullptr;
conv->width.set_from_arg(parse_digits()); conv->width.set_from_arg(parse_digits());
if (ABSL_PREDICT_FALSE(c != '$')) return false; if (ABSL_PREDICT_FALSE(c != '$')) return nullptr;
ABSL_FORMAT_PARSER_INTERNAL_GET_CHAR(); ABSL_FORMAT_PARSER_INTERNAL_GET_CHAR();
} else { } else {
conv->width.set_from_arg(++*next_arg); conv->width.set_from_arg(++*next_arg);
@ -188,9 +182,9 @@ flags_done:
} else if (c == '*') { } else if (c == '*') {
ABSL_FORMAT_PARSER_INTERNAL_GET_CHAR(); ABSL_FORMAT_PARSER_INTERNAL_GET_CHAR();
if (is_positional) { if (is_positional) {
if (ABSL_PREDICT_FALSE(c < '1' || c > '9')) return false; if (ABSL_PREDICT_FALSE(c < '1' || c > '9')) return nullptr;
conv->precision.set_from_arg(parse_digits()); conv->precision.set_from_arg(parse_digits());
if (c != '$') return false; if (c != '$') return nullptr;
ABSL_FORMAT_PARSER_INTERNAL_GET_CHAR(); ABSL_FORMAT_PARSER_INTERNAL_GET_CHAR();
} else { } else {
conv->precision.set_from_arg(++*next_arg); conv->precision.set_from_arg(++*next_arg);
@ -201,14 +195,14 @@ flags_done:
} }
} }
std::int8_t id = kIds[static_cast<unsigned char>(c)]; auto tag = GetTagForChar(c);
if (id < 0) { if (ABSL_PREDICT_FALSE(!tag.is_conv())) {
if (ABSL_PREDICT_FALSE(id == none)) return false; if (ABSL_PREDICT_FALSE(!tag.is_length())) return nullptr;
// It is a length modifier. // It is a length modifier.
using str_format_internal::LengthMod; using str_format_internal::LengthMod;
LengthMod length_mod = LengthMod::FromId(static_cast<LM>(~id)); LengthMod length_mod = tag.as_length();
ABSL_FORMAT_PARSER_INTERNAL_GET_CHAR(); ABSL_FORMAT_PARSER_INTERNAL_GET_CHAR();
if (c == 'h' && length_mod.id() == LengthMod::h) { if (c == 'h' && length_mod.id() == LengthMod::h) {
conv->length_mod = LengthMod::FromId(LengthMod::hh); conv->length_mod = LengthMod::FromId(LengthMod::hh);
@ -219,25 +213,24 @@ flags_done:
} else { } else {
conv->length_mod = length_mod; conv->length_mod = length_mod;
} }
id = kIds[static_cast<unsigned char>(c)]; tag = GetTagForChar(c);
if (ABSL_PREDICT_FALSE(id < 0)) return false; if (ABSL_PREDICT_FALSE(!tag.is_conv())) return nullptr;
} }
assert(CheckFastPathSetting(*conv)); assert(CheckFastPathSetting(*conv));
(void)(&CheckFastPathSetting); (void)(&CheckFastPathSetting);
conv->conv = ConversionChar::FromId(static_cast<CC>(id)); conv->conv = tag.as_conv();
if (!is_positional) conv->arg_position = ++*next_arg; if (!is_positional) conv->arg_position = ++*next_arg;
*src = string_view(pos, end - pos); return pos;
return true;
} }
} // namespace } // namespace
bool ConsumeUnboundConversion(string_view *src, UnboundConversion *conv, const char *ConsumeUnboundConversion(const char *p, const char *end,
int *next_arg) { UnboundConversion *conv, int *next_arg) {
if (*next_arg < 0) return ConsumeConversion<true>(src, conv, next_arg); if (*next_arg < 0) return ConsumeConversion<true>(p, end, conv, next_arg);
return ConsumeConversion<false>(src, conv, next_arg); return ConsumeConversion<false>(p, end, conv, next_arg);
} }
struct ParsedFormatBase::ParsedFormatConsumer { struct ParsedFormatBase::ParsedFormatConsumer {

122
absl/strings/internal/str_format/parser.h
View file

@ -63,17 +63,45 @@ struct UnboundConversion {
ConversionChar conv; ConversionChar conv;
}; };
// Consume conversion spec prefix (not including '%') of '*src' if valid. // Consume conversion spec prefix (not including '%') of [p, end) if valid.
// Examples of valid specs would be e.g.: "s", "d", "-12.6f". // Examples of valid specs would be e.g.: "s", "d", "-12.6f".
// If valid, the front of src is advanced such that src becomes the // If valid, it returns the first character following the conversion spec,
// part following the conversion spec, and the spec part is broken down and // and the spec part is broken down and returned in 'conv'.
// returned in 'conv'. // If invalid, returns nullptr.
// If invalid, returns false and leaves 'src' unmodified. const char* ConsumeUnboundConversion(const char* p, const char* end,
// For example: UnboundConversion* conv, int* next_arg);
// Given "d9", returns "d", and leaves src="9",
// Given "!", returns "" and leaves src="!". // Helper tag class for the table below.
bool ConsumeUnboundConversion(string_view* src, UnboundConversion* conv, // It allows fast `char -> ConversionChar/LengthMod` checking and conversions.
int* next_arg); class ConvTag {
public:
constexpr ConvTag(ConversionChar::Id id) : tag_(id) {} // NOLINT
// We invert the length modifiers to make them negative so that we can easily
// test for them.
constexpr ConvTag(LengthMod::Id id) : tag_(~id) {} // NOLINT
// Everything else is -128, which is negative to make is_conv() simpler.
constexpr ConvTag() : tag_(-128) {}
bool is_conv() const { return tag_ >= 0; }
bool is_length() const { return tag_ < 0 && tag_ != -128; }
ConversionChar as_conv() const {
assert(is_conv());
return ConversionChar::FromId(static_cast<ConversionChar::Id>(tag_));
}
LengthMod as_length() const {
assert(is_length());
return LengthMod::FromId(static_cast<LengthMod::Id>(~tag_));
}
private:
std::int8_t tag_;
};
extern const ConvTag kTags[256];
// Keep a single table for all the conversion chars and length modifiers.
inline ConvTag GetTagForChar(char c) {
return kTags[static_cast<unsigned char>(c)];
}
// Parse the format string provided in 'src' and pass the identified items into // Parse the format string provided in 'src' and pass the identified items into
// 'consumer'. // 'consumer'.
@ -88,52 +116,54 @@ bool ConsumeUnboundConversion(string_view* src, UnboundConversion* conv,
template <typename Consumer> template <typename Consumer>
bool ParseFormatString(string_view src, Consumer consumer) { bool ParseFormatString(string_view src, Consumer consumer) {
int next_arg = 0; int next_arg = 0;
while (!src.empty()) { const char* p = src.data();
const char* percent = const char* const end = p + src.size();
static_cast<const char*>(memchr(src.data(), '%', src.size())); while (p != end) {
const char* percent = static_cast<const char*>(memchr(p, '%', end - p));
if (!percent) { if (!percent) {
// We found the last substring. // We found the last substring.
return consumer.Append(src); return consumer.Append(string_view(p, end - p));
} }
// We found a percent, so push the text run then process the percent. // We found a percent, so push the text run then process the percent.
size_t percent_loc = percent - src.data(); if (ABSL_PREDICT_FALSE(!consumer.Append(string_view(p, percent - p)))) {
if (!consumer.Append(string_view(src.data(), percent_loc))) return false; return false;
if (percent + 1 >= src.data() + src.size()) return false;
UnboundConversion conv;
switch (percent[1]) {
case '%':
if (!consumer.Append("%")) return false;
src.remove_prefix(percent_loc + 2);
continue;
#define PARSER_CASE(ch) \
case #ch[0]: \
src.remove_prefix(percent_loc + 2); \
conv.conv = ConversionChar::FromId(ConversionChar::ch); \
conv.arg_position = ++next_arg; \
break;
ABSL_CONVERSION_CHARS_EXPAND_(PARSER_CASE, );
#undef PARSER_CASE
default:
src.remove_prefix(percent_loc + 1);
if (!ConsumeUnboundConversion(&src, &conv, &next_arg)) return false;
break;
} }
if (next_arg == 0) { if (ABSL_PREDICT_FALSE(percent + 1 >= end)) return false;
auto tag = GetTagForChar(percent[1]);
if (tag.is_conv()) {
if (ABSL_PREDICT_FALSE(next_arg < 0)) {
// This indicates an error in the format std::string. // This indicates an error in the format std::string.
// The only way to get next_arg == 0 is to have a positional argument // The only way to get `next_arg < 0` here is to have a positional
// first which sets next_arg to -1 and then a non-positional argument // argument first which sets next_arg to -1 and then a non-positional
// which does ++next_arg. // argument.
// Checking here seems to be the cheapeast place to do it.
return false; return false;
} }
if (!consumer.ConvertOne( p = percent + 2;
conv, string_view(percent + 1, src.data() - (percent + 1)))) {
// Keep this case separate from the one below.
// ConvertOne is more efficient when the compiler can see that the `basic`
// flag is set.
UnboundConversion conv;
conv.conv = tag.as_conv();
conv.arg_position = ++next_arg;
if (ABSL_PREDICT_FALSE(
!consumer.ConvertOne(conv, string_view(percent + 1, 1)))) {
return false; return false;
} }
} else if (percent[1] != '%') {
UnboundConversion conv;
p = ConsumeUnboundConversion(percent + 1, end, &conv, &next_arg);
if (ABSL_PREDICT_FALSE(p == nullptr)) return false;
if (ABSL_PREDICT_FALSE(!consumer.ConvertOne(
conv, string_view(percent + 1, p - (percent + 1))))) {
return false;
}
} else {
if (ABSL_PREDICT_FALSE(!consumer.Append("%"))) return false;
p = percent + 2;
continue;
}
} }
return true; return true;
} }

25
absl/strings/internal/str_format/parser_test.cc
View file

@ -1,6 +1,8 @@
#include "absl/strings/internal/str_format/parser.h" #include "absl/strings/internal/str_format/parser.h"
#include <string.h> #include <string.h>
#include "gmock/gmock.h"
#include "gtest/gtest.h" #include "gtest/gtest.h"
#include "absl/base/macros.h" #include "absl/base/macros.h"
@ -9,6 +11,8 @@ namespace str_format_internal {
namespace { namespace {
using testing::Pair;
TEST(LengthModTest, Names) { TEST(LengthModTest, Names) {
struct Expectation { struct Expectation {
int line; int line;
@ -63,20 +67,21 @@ TEST(ConversionCharTest, Names) {
class ConsumeUnboundConversionTest : public ::testing::Test { class ConsumeUnboundConversionTest : public ::testing::Test {
public: public:
typedef UnboundConversion Props; std::pair<string_view, string_view> Consume(string_view src) {
string_view Consume(string_view* src) {
int next = 0; int next = 0;
const char* prev_begin = src->data();
o = UnboundConversion(); // refresh o = UnboundConversion(); // refresh
ConsumeUnboundConversion(src, &o, &next); const char* p = ConsumeUnboundConversion(
return {prev_begin, static_cast<size_t>(src->data() - prev_begin)}; src.data(), src.data() + src.size(), &o, &next);
if (!p) return {{}, src};
return {string_view(src.data(), p - src.data()),
string_view(p, src.data() + src.size() - p)};
} }
bool Run(const char *fmt, bool force_positional = false) { bool Run(const char *fmt, bool force_positional = false) {
string_view src = fmt;
int next = force_positional ? -1 : 0; int next = force_positional ? -1 : 0;
o = UnboundConversion(); // refresh o = UnboundConversion(); // refresh
return ConsumeUnboundConversion(&src, &o, &next) && src.empty(); return ConsumeUnboundConversion(fmt, fmt + strlen(fmt), &o, &next) ==
fmt + strlen(fmt);
} }
UnboundConversion o; UnboundConversion o;
}; };
@ -104,11 +109,7 @@ TEST_F(ConsumeUnboundConversionTest, ConsumeSpecification) {
}; };
for (const auto& e : kExpect) { for (const auto& e : kExpect) {
SCOPED_TRACE(e.line); SCOPED_TRACE(e.line);
string_view src = e.src; EXPECT_THAT(Consume(e.src), Pair(e.out, e.src_post));
EXPECT_EQ(e.src, src);
string_view out = Consume(&src);
EXPECT_EQ(e.out, out);
EXPECT_EQ(e.src_post, src);
} }
} }

4
absl/strings/numbers.cc
View file

@ -54,7 +54,7 @@ bool SimpleAtof(absl::string_view str, float* out) {
// not all non-whitespace characters consumed // not all non-whitespace characters consumed
return false; return false;
} }
// from_chars() with DR 3801's current wording will return max() on // from_chars() with DR 3081's current wording will return max() on
// overflow. SimpleAtof returns infinity instead. // overflow. SimpleAtof returns infinity instead.
if (result.ec == std::errc::result_out_of_range) { if (result.ec == std::errc::result_out_of_range) {
if (*out > 1.0) { if (*out > 1.0) {
@ -80,7 +80,7 @@ bool SimpleAtod(absl::string_view str, double* out) {
// not all non-whitespace characters consumed // not all non-whitespace characters consumed
return false; return false;
} }
// from_chars() with DR 3801's current wording will return max() on // from_chars() with DR 3081's current wording will return max() on
// overflow. SimpleAtod returns infinity instead. // overflow. SimpleAtod returns infinity instead.
if (result.ec == std::errc::result_out_of_range) { if (result.ec == std::errc::result_out_of_range) {
if (*out > 1.0) { if (*out > 1.0) {