tvl-depot/absl/strings/internal/str_format/parser.h
Abseil Team d902eb869b Export of internal Abseil changes.
--
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
2019-04-16 15:40:40 -04:00

322 lines
11 KiB
C++

#ifndef ABSL_STRINGS_INTERNAL_STR_FORMAT_PARSER_H_
#define ABSL_STRINGS_INTERNAL_STR_FORMAT_PARSER_H_
#include <limits.h>
#include <stddef.h>
#include <stdlib.h>
#include <cassert>
#include <initializer_list>
#include <iosfwd>
#include <iterator>
#include <memory>
#include <vector>
#include "absl/strings/internal/str_format/checker.h"
#include "absl/strings/internal/str_format/extension.h"
namespace absl {
namespace str_format_internal {
// The analyzed properties of a single specified conversion.
struct UnboundConversion {
UnboundConversion()
: flags() /* This is required to zero all the fields of flags. */ {
flags.basic = true;
}
class InputValue {
public:
void set_value(int value) {
assert(value >= 0);
value_ = value;
}
int value() const { return value_; }
// Marks the value as "from arg". aka the '*' format.
// Requires `value >= 1`.
// When set, is_from_arg() return true and get_from_arg() returns the
// original value.
// `value()`'s return value is unspecfied in this state.
void set_from_arg(int value) {
assert(value > 0);
value_ = -value - 1;
}
bool is_from_arg() const { return value_ < -1; }
int get_from_arg() const {
assert(is_from_arg());
return -value_ - 1;
}
private:
int value_ = -1;
};
// No need to initialize. It will always be set in the parser.
int arg_position;
InputValue width;
InputValue precision;
Flags flags;
LengthMod length_mod;
ConversionChar conv;
};
// Consume conversion spec prefix (not including '%') of [p, end) if valid.
// Examples of valid specs would be e.g.: "s", "d", "-12.6f".
// If valid, it returns the first character following the conversion spec,
// and the spec part is broken down and returned in 'conv'.
// If invalid, returns nullptr.
const char* ConsumeUnboundConversion(const char* p, const char* end,
UnboundConversion* conv, int* next_arg);
// Helper tag class for the table below.
// It allows fast `char -> ConversionChar/LengthMod` checking and conversions.
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
// 'consumer'.
// Text runs will be passed by calling
// Consumer::Append(string_view);
// ConversionItems will be passed by calling
// Consumer::ConvertOne(UnboundConversion, string_view);
// In the case of ConvertOne, the string_view that is passed is the
// portion of the format string corresponding to the conversion, not including
// the leading %. On success, it returns true. On failure, it stops and returns
// false.
template <typename Consumer>
bool ParseFormatString(string_view src, Consumer consumer) {
int next_arg = 0;
const char* p = src.data();
const char* const end = p + src.size();
while (p != end) {
const char* percent = static_cast<const char*>(memchr(p, '%', end - p));
if (!percent) {
// We found the last substring.
return consumer.Append(string_view(p, end - p));
}
// We found a percent, so push the text run then process the percent.
if (ABSL_PREDICT_FALSE(!consumer.Append(string_view(p, percent - p)))) {
return false;
}
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.
// The only way to get `next_arg < 0` here is to have a positional
// argument first which sets next_arg to -1 and then a non-positional
// argument.
return false;
}
p = percent + 2;
// 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;
}
} 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;
}
// Always returns true, or fails to compile in a constexpr context if s does not
// point to a constexpr char array.
constexpr bool EnsureConstexpr(string_view s) {
return s.empty() || s[0] == s[0];
}
class ParsedFormatBase {
public:
explicit ParsedFormatBase(string_view format, bool allow_ignored,
std::initializer_list<Conv> convs);
ParsedFormatBase(const ParsedFormatBase& other) { *this = other; }
ParsedFormatBase(ParsedFormatBase&& other) { *this = std::move(other); }
ParsedFormatBase& operator=(const ParsedFormatBase& other) {
if (this == &other) return *this;
has_error_ = other.has_error_;
items_ = other.items_;
size_t text_size = items_.empty() ? 0 : items_.back().text_end;
data_.reset(new char[text_size]);
memcpy(data_.get(), other.data_.get(), text_size);
return *this;
}
ParsedFormatBase& operator=(ParsedFormatBase&& other) {
if (this == &other) return *this;
has_error_ = other.has_error_;
data_ = std::move(other.data_);
items_ = std::move(other.items_);
// Reset the vector to make sure the invariants hold.
other.items_.clear();
return *this;
}
template <typename Consumer>
bool ProcessFormat(Consumer consumer) const {
const char* const base = data_.get();
string_view text(base, 0);
for (const auto& item : items_) {
const char* const end = text.data() + text.size();
text = string_view(end, (base + item.text_end) - end);
if (item.is_conversion) {
if (!consumer.ConvertOne(item.conv, text)) return false;
} else {
if (!consumer.Append(text)) return false;
}
}
return !has_error_;
}
bool has_error() const { return has_error_; }
private:
// Returns whether the conversions match and if !allow_ignored it verifies
// that all conversions are used by the format.
bool MatchesConversions(bool allow_ignored,
std::initializer_list<Conv> convs) const;
struct ParsedFormatConsumer;
struct ConversionItem {
bool is_conversion;
// Points to the past-the-end location of this element in the data_ array.
size_t text_end;
UnboundConversion conv;
};
bool has_error_;
std::unique_ptr<char[]> data_;
std::vector<ConversionItem> items_;
};
// A value type representing a preparsed format. These can be created, copied
// around, and reused to speed up formatting loops.
// The user must specify through the template arguments the conversion
// characters used in the format. This will be checked at compile time.
//
// This class uses Conv enum values to specify each argument.
// This allows for more flexibility as you can specify multiple possible
// conversion characters for each argument.
// ParsedFormat<char...> is a simplified alias for when the user only
// needs to specify a single conversion character for each argument.
//
// Example:
// // Extended format supports multiple characters per argument:
// using MyFormat = ExtendedParsedFormat<Conv::d | Conv::x>;
// MyFormat GetFormat(bool use_hex) {
// if (use_hex) return MyFormat("foo %x bar");
// return MyFormat("foo %d bar");
// }
// // 'format' can be used with any value that supports 'd' and 'x',
// // like `int`.
// auto format = GetFormat(use_hex);
// value = StringF(format, i);
//
// This class also supports runtime format checking with the ::New() and
// ::NewAllowIgnored() factory functions.
// This is the only API that allows the user to pass a runtime specified format
// string. These factory functions will return NULL if the format does not match
// the conversions requested by the user.
template <str_format_internal::Conv... C>
class ExtendedParsedFormat : public str_format_internal::ParsedFormatBase {
public:
explicit ExtendedParsedFormat(string_view format)
#if ABSL_INTERNAL_ENABLE_FORMAT_CHECKER
__attribute__((
enable_if(str_format_internal::EnsureConstexpr(format),
"Format std::string is not constexpr."),
enable_if(str_format_internal::ValidFormatImpl<C...>(format),
"Format specified does not match the template arguments.")))
#endif // ABSL_INTERNAL_ENABLE_FORMAT_CHECKER
: ExtendedParsedFormat(format, false) {
}
// ExtendedParsedFormat factory function.
// The user still has to specify the conversion characters, but they will not
// be checked at compile time. Instead, it will be checked at runtime.
// This delays the checking to runtime, but allows the user to pass
// dynamically sourced formats.
// It returns NULL if the format does not match the conversion characters.
// The user is responsible for checking the return value before using it.
//
// The 'New' variant will check that all the specified arguments are being
// consumed by the format and return NULL if any argument is being ignored.
// The 'NewAllowIgnored' variant will not verify this and will allow formats
// that ignore arguments.
static std::unique_ptr<ExtendedParsedFormat> New(string_view format) {
return New(format, false);
}
static std::unique_ptr<ExtendedParsedFormat> NewAllowIgnored(
string_view format) {
return New(format, true);
}
private:
static std::unique_ptr<ExtendedParsedFormat> New(string_view format,
bool allow_ignored) {
std::unique_ptr<ExtendedParsedFormat> conv(
new ExtendedParsedFormat(format, allow_ignored));
if (conv->has_error()) return nullptr;
return conv;
}
ExtendedParsedFormat(string_view s, bool allow_ignored)
: ParsedFormatBase(s, allow_ignored, {C...}) {}
};
} // namespace str_format_internal
} // namespace absl
#endif // ABSL_STRINGS_INTERNAL_STR_FORMAT_PARSER_H_