dd3f1ecf71
Even though the change is ES6 spec compliant, we decided to revert to be consistent with other browsers and work on fixing the spec. Original issue's description: > Make dates default to the local timezone if none specified > > In ES5, dates were supposed to default to UTC if no timezone was specified. However, this changed in ES6, which specified that dates should be in the local timezone if no timezone was specified. This CL updates our behavior to match that part of the ES6 spec. > BUG=chromium:391730, v8:4242 > LOG=Y > Committed: https://crrev.com/f06754a8e1d305a43560705f6c167d85d40e602d > Cr-Commit-Position: refs/heads/master@{#29854} BUG=chromium:543320,chromium:539813 LOG=NO Review URL: https://codereview.chromium.org/1403153003 Cr-Commit-Position: refs/heads/master@{#31295}
371 lines
11 KiB
C++
371 lines
11 KiB
C++
// Copyright 2011 the V8 project authors. All rights reserved.
|
|
// Use of this source code is governed by a BSD-style license that can be
|
|
// found in the LICENSE file.
|
|
|
|
#ifndef V8_DATEPARSER_H_
|
|
#define V8_DATEPARSER_H_
|
|
|
|
#include "src/allocation.h"
|
|
#include "src/char-predicates.h"
|
|
#include "src/scanner.h"
|
|
|
|
namespace v8 {
|
|
namespace internal {
|
|
|
|
class DateParser : public AllStatic {
|
|
public:
|
|
// Parse the string as a date. If parsing succeeds, return true after
|
|
// filling out the output array as follows (all integers are Smis):
|
|
// [0]: year
|
|
// [1]: month (0 = Jan, 1 = Feb, ...)
|
|
// [2]: day
|
|
// [3]: hour
|
|
// [4]: minute
|
|
// [5]: second
|
|
// [6]: millisecond
|
|
// [7]: UTC offset in seconds, or null value if no timezone specified
|
|
// If parsing fails, return false (content of output array is not defined).
|
|
template <typename Char>
|
|
static bool Parse(Vector<Char> str, FixedArray* output, UnicodeCache* cache);
|
|
|
|
enum {
|
|
YEAR, MONTH, DAY, HOUR, MINUTE, SECOND, MILLISECOND, UTC_OFFSET, OUTPUT_SIZE
|
|
};
|
|
|
|
private:
|
|
// Range testing
|
|
static inline bool Between(int x, int lo, int hi) {
|
|
return static_cast<unsigned>(x - lo) <= static_cast<unsigned>(hi - lo);
|
|
}
|
|
|
|
// Indicates a missing value.
|
|
static const int kNone = kMaxInt;
|
|
|
|
// Maximal number of digits used to build the value of a numeral.
|
|
// Remaining digits are ignored.
|
|
static const int kMaxSignificantDigits = 9;
|
|
|
|
// InputReader provides basic string parsing and character classification.
|
|
template <typename Char>
|
|
class InputReader BASE_EMBEDDED {
|
|
public:
|
|
InputReader(UnicodeCache* unicode_cache, Vector<Char> s)
|
|
: index_(0),
|
|
buffer_(s),
|
|
unicode_cache_(unicode_cache) {
|
|
Next();
|
|
}
|
|
|
|
int position() { return index_; }
|
|
|
|
// Advance to the next character of the string.
|
|
void Next() {
|
|
ch_ = (index_ < buffer_.length()) ? buffer_[index_] : 0;
|
|
index_++;
|
|
}
|
|
|
|
// Read a string of digits as an unsigned number. Cap value at
|
|
// kMaxSignificantDigits, but skip remaining digits if the numeral
|
|
// is longer.
|
|
int ReadUnsignedNumeral() {
|
|
int n = 0;
|
|
int i = 0;
|
|
while (IsAsciiDigit()) {
|
|
if (i < kMaxSignificantDigits) n = n * 10 + ch_ - '0';
|
|
i++;
|
|
Next();
|
|
}
|
|
return n;
|
|
}
|
|
|
|
// Read a word (sequence of chars. >= 'A'), fill the given buffer with a
|
|
// lower-case prefix, and pad any remainder of the buffer with zeroes.
|
|
// Return word length.
|
|
int ReadWord(uint32_t* prefix, int prefix_size) {
|
|
int len;
|
|
for (len = 0; IsAsciiAlphaOrAbove(); Next(), len++) {
|
|
if (len < prefix_size) prefix[len] = AsciiAlphaToLower(ch_);
|
|
}
|
|
for (int i = len; i < prefix_size; i++) prefix[i] = 0;
|
|
return len;
|
|
}
|
|
|
|
// The skip methods return whether they actually skipped something.
|
|
bool Skip(uint32_t c) {
|
|
if (ch_ == c) {
|
|
Next();
|
|
return true;
|
|
}
|
|
return false;
|
|
}
|
|
|
|
inline bool SkipWhiteSpace();
|
|
inline bool SkipParentheses();
|
|
|
|
// Character testing/classification. Non-ASCII digits are not supported.
|
|
bool Is(uint32_t c) const { return ch_ == c; }
|
|
bool IsEnd() const { return ch_ == 0; }
|
|
bool IsAsciiDigit() const { return IsDecimalDigit(ch_); }
|
|
bool IsAsciiAlphaOrAbove() const { return ch_ >= 'A'; }
|
|
bool IsAsciiSign() const { return ch_ == '+' || ch_ == '-'; }
|
|
|
|
// Return 1 for '+' and -1 for '-'.
|
|
int GetAsciiSignValue() const { return 44 - static_cast<int>(ch_); }
|
|
|
|
private:
|
|
int index_;
|
|
Vector<Char> buffer_;
|
|
uint32_t ch_;
|
|
UnicodeCache* unicode_cache_;
|
|
};
|
|
|
|
enum KeywordType {
|
|
INVALID, MONTH_NAME, TIME_ZONE_NAME, TIME_SEPARATOR, AM_PM
|
|
};
|
|
|
|
struct DateToken {
|
|
public:
|
|
bool IsInvalid() { return tag_ == kInvalidTokenTag; }
|
|
bool IsUnknown() { return tag_ == kUnknownTokenTag; }
|
|
bool IsNumber() { return tag_ == kNumberTag; }
|
|
bool IsSymbol() { return tag_ == kSymbolTag; }
|
|
bool IsWhiteSpace() { return tag_ == kWhiteSpaceTag; }
|
|
bool IsEndOfInput() { return tag_ == kEndOfInputTag; }
|
|
bool IsKeyword() { return tag_ >= kKeywordTagStart; }
|
|
|
|
int length() { return length_; }
|
|
|
|
int number() {
|
|
DCHECK(IsNumber());
|
|
return value_;
|
|
}
|
|
KeywordType keyword_type() {
|
|
DCHECK(IsKeyword());
|
|
return static_cast<KeywordType>(tag_);
|
|
}
|
|
int keyword_value() {
|
|
DCHECK(IsKeyword());
|
|
return value_;
|
|
}
|
|
char symbol() {
|
|
DCHECK(IsSymbol());
|
|
return static_cast<char>(value_);
|
|
}
|
|
bool IsSymbol(char symbol) {
|
|
return IsSymbol() && this->symbol() == symbol;
|
|
}
|
|
bool IsKeywordType(KeywordType tag) {
|
|
return tag_ == tag;
|
|
}
|
|
bool IsFixedLengthNumber(int length) {
|
|
return IsNumber() && length_ == length;
|
|
}
|
|
bool IsAsciiSign() {
|
|
return tag_ == kSymbolTag && (value_ == '-' || value_ == '+');
|
|
}
|
|
int ascii_sign() {
|
|
DCHECK(IsAsciiSign());
|
|
return 44 - value_;
|
|
}
|
|
bool IsKeywordZ() {
|
|
return IsKeywordType(TIME_ZONE_NAME) && length_ == 1 && value_ == 0;
|
|
}
|
|
bool IsUnknown(int character) {
|
|
return IsUnknown() && value_ == character;
|
|
}
|
|
// Factory functions.
|
|
static DateToken Keyword(KeywordType tag, int value, int length) {
|
|
return DateToken(tag, length, value);
|
|
}
|
|
static DateToken Number(int value, int length) {
|
|
return DateToken(kNumberTag, length, value);
|
|
}
|
|
static DateToken Symbol(char symbol) {
|
|
return DateToken(kSymbolTag, 1, symbol);
|
|
}
|
|
static DateToken EndOfInput() {
|
|
return DateToken(kEndOfInputTag, 0, -1);
|
|
}
|
|
static DateToken WhiteSpace(int length) {
|
|
return DateToken(kWhiteSpaceTag, length, -1);
|
|
}
|
|
static DateToken Unknown() {
|
|
return DateToken(kUnknownTokenTag, 1, -1);
|
|
}
|
|
static DateToken Invalid() {
|
|
return DateToken(kInvalidTokenTag, 0, -1);
|
|
}
|
|
|
|
private:
|
|
enum TagType {
|
|
kInvalidTokenTag = -6,
|
|
kUnknownTokenTag = -5,
|
|
kWhiteSpaceTag = -4,
|
|
kNumberTag = -3,
|
|
kSymbolTag = -2,
|
|
kEndOfInputTag = -1,
|
|
kKeywordTagStart = 0
|
|
};
|
|
DateToken(int tag, int length, int value)
|
|
: tag_(tag),
|
|
length_(length),
|
|
value_(value) { }
|
|
|
|
int tag_;
|
|
int length_; // Number of characters.
|
|
int value_;
|
|
};
|
|
|
|
template <typename Char>
|
|
class DateStringTokenizer {
|
|
public:
|
|
explicit DateStringTokenizer(InputReader<Char>* in)
|
|
: in_(in), next_(Scan()) { }
|
|
DateToken Next() {
|
|
DateToken result = next_;
|
|
next_ = Scan();
|
|
return result;
|
|
}
|
|
|
|
DateToken Peek() {
|
|
return next_;
|
|
}
|
|
bool SkipSymbol(char symbol) {
|
|
if (next_.IsSymbol(symbol)) {
|
|
next_ = Scan();
|
|
return true;
|
|
}
|
|
return false;
|
|
}
|
|
|
|
private:
|
|
DateToken Scan();
|
|
|
|
InputReader<Char>* in_;
|
|
DateToken next_;
|
|
};
|
|
|
|
static int ReadMilliseconds(DateToken number);
|
|
|
|
// KeywordTable maps names of months, time zones, am/pm to numbers.
|
|
class KeywordTable : public AllStatic {
|
|
public:
|
|
// Look up a word in the keyword table and return an index.
|
|
// 'pre' contains a prefix of the word, zero-padded to size kPrefixLength
|
|
// and 'len' is the word length.
|
|
static int Lookup(const uint32_t* pre, int len);
|
|
// Get the type of the keyword at index i.
|
|
static KeywordType GetType(int i) {
|
|
return static_cast<KeywordType>(array[i][kTypeOffset]);
|
|
}
|
|
// Get the value of the keyword at index i.
|
|
static int GetValue(int i) { return array[i][kValueOffset]; }
|
|
|
|
static const int kPrefixLength = 3;
|
|
static const int kTypeOffset = kPrefixLength;
|
|
static const int kValueOffset = kTypeOffset + 1;
|
|
static const int kEntrySize = kValueOffset + 1;
|
|
static const int8_t array[][kEntrySize];
|
|
};
|
|
|
|
class TimeZoneComposer BASE_EMBEDDED {
|
|
public:
|
|
TimeZoneComposer() : sign_(kNone), hour_(kNone), minute_(kNone) {}
|
|
void Set(int offset_in_hours) {
|
|
sign_ = offset_in_hours < 0 ? -1 : 1;
|
|
hour_ = offset_in_hours * sign_;
|
|
minute_ = 0;
|
|
}
|
|
void SetSign(int sign) { sign_ = sign < 0 ? -1 : 1; }
|
|
void SetAbsoluteHour(int hour) { hour_ = hour; }
|
|
void SetAbsoluteMinute(int minute) { minute_ = minute; }
|
|
bool IsExpecting(int n) const {
|
|
return hour_ != kNone && minute_ == kNone && TimeComposer::IsMinute(n);
|
|
}
|
|
bool IsUTC() const { return hour_ == 0 && minute_ == 0; }
|
|
bool Write(FixedArray* output);
|
|
bool IsEmpty() { return hour_ == kNone; }
|
|
private:
|
|
int sign_;
|
|
int hour_;
|
|
int minute_;
|
|
};
|
|
|
|
class TimeComposer BASE_EMBEDDED {
|
|
public:
|
|
TimeComposer() : index_(0), hour_offset_(kNone) {}
|
|
bool IsEmpty() const { return index_ == 0; }
|
|
bool IsExpecting(int n) const {
|
|
return (index_ == 1 && IsMinute(n)) ||
|
|
(index_ == 2 && IsSecond(n)) ||
|
|
(index_ == 3 && IsMillisecond(n));
|
|
}
|
|
bool Add(int n) {
|
|
return index_ < kSize ? (comp_[index_++] = n, true) : false;
|
|
}
|
|
bool AddFinal(int n) {
|
|
if (!Add(n)) return false;
|
|
while (index_ < kSize) comp_[index_++] = 0;
|
|
return true;
|
|
}
|
|
void SetHourOffset(int n) { hour_offset_ = n; }
|
|
bool Write(FixedArray* output);
|
|
|
|
static bool IsMinute(int x) { return Between(x, 0, 59); }
|
|
static bool IsHour(int x) { return Between(x, 0, 23); }
|
|
static bool IsSecond(int x) { return Between(x, 0, 59); }
|
|
|
|
private:
|
|
static bool IsHour12(int x) { return Between(x, 0, 12); }
|
|
static bool IsMillisecond(int x) { return Between(x, 0, 999); }
|
|
|
|
static const int kSize = 4;
|
|
int comp_[kSize];
|
|
int index_;
|
|
int hour_offset_;
|
|
};
|
|
|
|
class DayComposer BASE_EMBEDDED {
|
|
public:
|
|
DayComposer() : index_(0), named_month_(kNone), is_iso_date_(false) {}
|
|
bool IsEmpty() const { return index_ == 0; }
|
|
bool Add(int n) {
|
|
if (index_ < kSize) {
|
|
comp_[index_] = n;
|
|
index_++;
|
|
return true;
|
|
}
|
|
return false;
|
|
}
|
|
void SetNamedMonth(int n) { named_month_ = n; }
|
|
bool Write(FixedArray* output);
|
|
void set_iso_date() { is_iso_date_ = true; }
|
|
static bool IsMonth(int x) { return Between(x, 1, 12); }
|
|
static bool IsDay(int x) { return Between(x, 1, 31); }
|
|
|
|
private:
|
|
static const int kSize = 3;
|
|
int comp_[kSize];
|
|
int index_;
|
|
int named_month_;
|
|
// If set, ensures that data is always parsed in year-month-date order.
|
|
bool is_iso_date_;
|
|
};
|
|
|
|
// Tries to parse an ES5 Date Time String. Returns the next token
|
|
// to continue with in the legacy date string parser. If parsing is
|
|
// complete, returns DateToken::EndOfInput(). If terminally unsuccessful,
|
|
// returns DateToken::Invalid(). Otherwise parsing continues in the
|
|
// legacy parser.
|
|
template <typename Char>
|
|
static DateParser::DateToken ParseES5DateTime(
|
|
DateStringTokenizer<Char>* scanner, DayComposer* day, TimeComposer* time,
|
|
TimeZoneComposer* tz);
|
|
};
|
|
|
|
|
|
} // namespace internal
|
|
} // namespace v8
|
|
|
|
#endif // V8_DATEPARSER_H_
|