added `format::parse` and accompanying tests.

This commit is contained in:
Kang Seonghoon 2015-02-04 02:58:40 +09:00
parent dfa92ef56d
commit aebcedf37b
2 changed files with 553 additions and 47 deletions

View File

@ -7,18 +7,22 @@
*/
use std::fmt;
use std::iter;
use std::usize;
use {Datelike, Timelike};
use Weekday;
use div::{div_floor, mod_floor};
use duration::Duration;
use offset::Offset;
use naive::date::NaiveDate;
use naive::time::NaiveTime;
use self::parsed::Parsed;
pub use self::strftime::StrftimeItems;
/// Padding characters for numeric items.
#[derive(Copy, PartialEq, Eq, Debug)]
#[derive(Copy, Clone, PartialEq, Eq, Debug)]
pub enum Pad {
/// No padding.
None,
@ -29,79 +33,115 @@ pub enum Pad {
}
/// Numeric item types.
#[derive(Copy, PartialEq, Eq, Debug)]
/// They have associated formatting width (FW) and parsing width (PW).
///
/// The **formatting width** is the minimal width to be formatted.
/// If the number is too short, and the padding is not `Pad::None`, then it is left-padded.
/// If the number is too long or (in some cases) negative, it is printed as is.
///
/// The **parsing width** is the maximal width to be scanned.
/// The parser only tries to consume from one to given number of digits (greedily).
/// It also trims the preceding whitespaces if any.
/// It cannot parse the negative number, so some date and time cannot be formatted then
/// parsed with the same formatting items.
#[derive(Copy, Clone, PartialEq, Eq, Debug)]
pub enum Numeric {
/// Full Gregorian year.
/// Full Gregorian year (FW=PW=4).
Year,
/// Gregorian year divided by 100 (century number). Always rounds towards minus infinity.
/// Gregorian year divided by 100 (century number; FW=PW=2).
/// Always rounds towards minus infinity.
YearDiv100,
/// Gregorian year modulo 100. Cannot be negative.
/// Gregorian year modulo 100 (FW=PW=2). Cannot be negative.
YearMod100,
/// Year in the ISO week date.
/// Year in the ISO week date (FW=PW=4).
IsoYear,
/// Year in the ISO week date, divided by 100. Always rounds towards minus infinity.
/// Year in the ISO week date, divided by 100 (FW=PW=2). Always rounds towards minus infinity.
IsoYearDiv100,
/// Year in the ISO week date, modulo 100. Cannot be negative.
/// Year in the ISO week date, modulo 100 (FW=PW=2). Cannot be negative.
IsoYearMod100,
/// Month.
/// Month (FW=PW=2).
Month,
/// Day of the month.
/// Day of the month (FW=PW=2).
Day,
/// Week number, where the week 1 starts at the first Sunday of January.
/// Week number, where the week 1 starts at the first Sunday of January (FW=PW=2).
WeekFromSun,
/// Week number, where the week 1 starts at the first Monday of January.
/// Week number, where the week 1 starts at the first Monday of January (FW=PW=2).
WeekFromMon,
/// Week number in the ISO week date.
/// Week number in the ISO week date (FW=PW=2).
IsoWeek,
/// Day of the week, where Sunday = 0 and Saturday = 6.
/// Day of the week, where Sunday = 0 and Saturday = 6 (FW=PW=1).
NumDaysFromSun,
/// Day of the week, where Monday = 1 and Sunday = 7.
/// Day of the week, where Monday = 1 and Sunday = 7 (FW=PW=1).
WeekdayFromMon,
/// Day of the year.
/// Day of the year (FW=PW=3).
Ordinal,
/// Hour number in the 24-hour clocks.
/// Hour number in the 24-hour clocks (FW=PW=2).
Hour,
/// Hour number in the 12-hour clocks.
/// Hour number in the 12-hour clocks (FW=PW=2).
Hour12,
/// The number of minutes since the last whole hour.
/// The number of minutes since the last whole hour (FW=PW=2).
Minute,
/// The number of seconds since the last whole minute.
/// The number of seconds since the last whole minute (FW=PW=2).
Second,
/// The number of nanoseconds since the last whole second.
/// The number of nanoseconds since the last whole second (FW=PW=9).
Nanosecond,
/// The number of non-leap seconds since January 1, 1970 0:00:00 UTC.
/// The number of non-leap seconds since January 1, 1970 0:00:00 UTC (FW=1, PW=infinity).
Timestamp,
}
/// Fixed-format item types.
#[derive(Copy, PartialEq, Eq, Debug)]
///
/// They have their own rules of formatting and parsing.
/// Otherwise noted, they print in the specified cases but parse case-insensitively.
#[derive(Copy, Clone, PartialEq, Eq, Debug)]
pub enum Fixed {
/// Abbreviated month names.
///
/// Prints a three-letter-long name in the title case, reads the same name in any case.
ShortMonthName,
/// Full month names.
///
/// Prints a full name in the title case, reads either a short or full name in any case.
LongMonthName,
/// Abbreviated day of the week names.
///
/// Prints a three-letter-long name in the title case, reads the same name in any case.
ShortWeekdayName,
/// FUll day of the week names.
/// Full day of the week names.
///
/// Prints a full name in the title case, reads either a short or full name in any case.
LongWeekdayName,
/// AM/PM in upper cases.
/// AM/PM.
///
/// Prints in lower case, reads in any case.
LowerAmPm,
/// AM/PM in lower cases.
/// AM/PM.
///
/// Prints in upper case, reads in any case.
UpperAmPm,
/// Timezone name.
///
/// It does not support parsing, its use in the parser is an immediate failure.
TimezoneName,
/// Offset from the local time to UTC (`+09:00` or `-04:00` or `+00:00`).
///
/// In the parser, the colon can be omitted and/or surrounded with any amount of whitespaces.
/// The offset is limited from `-24:00` to `+24:00`, which is same to `FixedOffset`'s range.
TimezoneOffset,
/// Offset from the local time to UTC (`+09:00` or `-04:00` or `Z`).
///
/// In the parser, the colon can be omitted and/or surrounded with any amount of whitespaces,
/// and `Z` can be either in upper case or in lower case.
/// The offset is limited from `-24:00` to `+24:00`, which is same to `FixedOffset`'s range.
TimezoneOffsetZ,
}
/// A single formatting item. This is used for both formatting and parsing.
#[derive(Copy, PartialEq, Eq, Debug)]
#[derive(Copy, Clone, PartialEq, Eq, Debug)]
pub enum Item<'a> {
/// A literally printed and parsed text.
Literal(&'a str),
/// Whitespace. Prints literally but parses zero or more whitespace.
/// Whitespace. Prints literally but reads zero or more whitespace.
Space(&'a str),
/// Numeric item. Can be optionally padded to the maximal length (if any) when formatting;
/// the parser simply ignores any padded whitespace and zeroes.
@ -119,38 +159,34 @@ macro_rules! num0 { ($x:ident) => (Item::Numeric(Numeric::$x, Pad::Zero)) }
macro_rules! nums { ($x:ident) => (Item::Numeric(Numeric::$x, Pad::Space)) }
macro_rules! fix { ($x:ident) => (Item::Fixed(Fixed::$x)) }
/// Abbreviated month names.
static SHORT_MONTHS: [&'static str; 12] =
["Jan", "Feb", "Mar", "Apr", "May", "Jun", "Jul", "Aug", "Sep", "Oct", "Nov", "Dec"];
/// Full month names.
static LONG_MONTHS: [&'static str; 12] =
["January", "February", "March", "April", "May", "June",
"July", "August", "September", "October", "November", "December"];
/// Abbreviated weekday names.
static SHORT_WEEKDAYS: [&'static str; 7] =
["Mon", "Tue", "Wed", "Thu", "Fri", "Sat", "Sun"];
/// FUll weekday names.
static LONG_WEEKDAYS: [&'static str; 7] =
["Monday", "Tuesday", "Wednesday", "Thursday", "Friday", "Saturday", "Sunday"];
/// Tries to format given arguments with given formatting items.
/// Internally used by `DelayedFormat`.
pub fn format<'a, I>(w: &mut fmt::Formatter, date: Option<&NaiveDate>, time: Option<&NaiveTime>,
off: Option<&(String, Duration)>, items: I) -> fmt::Result
where I: Iterator<Item=Item<'a>> {
// full and abbreviated month and weekday names
static SHORT_MONTHS: [&'static str; 12] =
["Jan", "Feb", "Mar", "Apr", "May", "Jun", "Jul", "Aug", "Sep", "Oct", "Nov", "Dec"];
static LONG_MONTHS: [&'static str; 12] =
["January", "February", "March", "April", "May", "June",
"July", "August", "September", "October", "November", "December"];
static SHORT_WEEKDAYS: [&'static str; 7] =
["Mon", "Tue", "Wed", "Thu", "Fri", "Sat", "Sun"];
static LONG_WEEKDAYS: [&'static str; 7] =
["Monday", "Tuesday", "Wednesday", "Thursday", "Friday", "Saturday", "Sunday"];
for item in items {
match item {
Item::Literal(s) | Item::Space(s) => try!(write!(w, "{}", s)),
Item::Numeric(spec, pad) => {
use self::Numeric::*;
let week_from_sun =
|&: d: &NaiveDate| (d.ordinal() - d.weekday().num_days_from_sunday() + 7) / 7;
let week_from_mon =
|&: d: &NaiveDate| (d.ordinal() - d.weekday().num_days_from_monday() + 7) / 7;
let (width, v) = match spec {
Year => (4, date.map(|d| d.year() as i64)),
YearDiv100 => (2, date.map(|d| div_floor(d.year() as i64, 100))),
@ -177,6 +213,7 @@ pub fn format<'a, I>(w: &mut fmt::Formatter, date: Option<&NaiveDate>, time: Opt
(_, _) => None
}),
};
if let Some(v) = v {
match pad {
Pad::None => try!(write!(w, "{}", v)),
@ -190,6 +227,7 @@ pub fn format<'a, I>(w: &mut fmt::Formatter, date: Option<&NaiveDate>, time: Opt
Item::Fixed(spec) => {
use self::Fixed::*;
let ret = match spec {
ShortMonthName =>
date.map(|d| write!(w, "{}", SHORT_MONTHS[d.month0() as usize])),
@ -197,7 +235,7 @@ pub fn format<'a, I>(w: &mut fmt::Formatter, date: Option<&NaiveDate>, time: Opt
date.map(|d| write!(w, "{}", LONG_MONTHS[d.month0() as usize])),
ShortWeekdayName =>
date.map(|d| write!(w, "{}",
SHORT_WEEKDAYS[d.weekday() .num_days_from_monday() as usize])),
SHORT_WEEKDAYS[d.weekday().num_days_from_monday() as usize])),
LongWeekdayName =>
date.map(|d| write!(w, "{}",
LONG_WEEKDAYS[d.weekday().num_days_from_monday() as usize])),
@ -224,6 +262,7 @@ pub fn format<'a, I>(w: &mut fmt::Formatter, date: Option<&NaiveDate>, time: Opt
}
}),
};
match ret {
Some(ret) => try!(ret),
None => return Err(fmt::Error), // insufficient arguments for given format
@ -237,6 +276,276 @@ pub fn format<'a, I>(w: &mut fmt::Formatter, date: Option<&NaiveDate>, time: Opt
Ok(())
}
/// Tries to parse given string into `parsed` with given formatting items.
/// Returns true when the entire string has been parsed (otherwise `parsed` should not be used).
/// There should be no trailing string after parsing; use a stray `Item::Space` to trim whitespaces.
///
/// This particular date and time parser is:
///
/// - Greedy. It will consume the longest possible prefix.
/// For example, `April` is always consumed entirely when the long month name is requested;
/// it equally accepts `Apr`, but prefers the longer prefix in this case.
/// - Padding-agnostic (for numeric items). The `Pad` field is completely ignored,
/// so one can prepend any number of whitespace then any number of zeroes before numbers.
/// - (Still) obeying the intrinsic parsing width. This allows, for example, parsing `HHMMSS`.
pub fn parse<'a, I>(parsed: &mut Parsed, mut s: &str, items: I) -> bool
where I: Iterator<Item=Item<'a>> {
// lowercased month and weekday names
static LONG_MONTHS: [&'static str; 12] =
["january", "february", "march", "april", "may", "june",
"july", "august", "september", "october", "november", "december"];
static LONG_WEEKDAYS: [&'static str; 7] =
["monday", "tuesday", "wednesday", "thursday", "friday", "saturday", "sunday"];
// tries to parse the month index (0 through 11) with the first three ASCII letters.
fn parse_short_month(s: &str) -> Option<u8> {
if s.len() < 3 { return None; }
let s = s.as_bytes();
match [s[0] | 32, s[1] | 32, s[2] | 32] {
[b'j',b'a',b'n'] => Some(0),
[b'f',b'e',b'b'] => Some(1),
[b'm',b'a',b'r'] => Some(2),
[b'a',b'p',b'r'] => Some(3),
[b'm',b'a',b'y'] => Some(4),
[b'j',b'u',b'n'] => Some(5),
[b'j',b'u',b'l'] => Some(6),
[b'a',b'u',b'g'] => Some(7),
[b's',b'e',b'p'] => Some(8),
[b'o',b'c',b't'] => Some(9),
[b'n',b'o',b'v'] => Some(10),
[b'd',b'e',b'c'] => Some(11),
_ => None
}
}
// tries to parse the weekday with the first three ASCII letters.
fn parse_short_weekday(s: &str) -> Option<Weekday> {
if s.len() < 3 { return None; }
let s = s.as_bytes();
match [s[0] | 32, s[1] | 32, s[2] | 32] {
[b'm',b'o',b'n'] => Some(Weekday::Mon),
[b't',b'u',b'e'] => Some(Weekday::Tue),
[b'w',b'e',b'd'] => Some(Weekday::Wed),
[b't',b'h',b'u'] => Some(Weekday::Thu),
[b'f',b'r',b'i'] => Some(Weekday::Fri),
[b's',b'a',b't'] => Some(Weekday::Sat),
[b's',b'u',b'n'] => Some(Weekday::Sun),
_ => None
}
}
// tries to consume `\s*[-+]\d\d[\s:]*\d\d` and return an offset in seconds
fn parse_timezone_offset(mut s: &str, allow_zulu: bool) -> Option<(&str, i32)> {
s = s.trim_left();
// + or -, or Z/z if `allow_zulu` is true
let negative = if s.starts_with("+") {
false
} else if s.starts_with("-") {
true
} else if allow_zulu && (s.starts_with("z") || s.starts_with("Z")) {
return Some((&s[1..], 0));
} else {
return None;
};
s = &s[1..];
// hours (00--24, where 24 is allowed only with 24:00)
// the range check happens later for this reason.
let hours = match s.as_bytes() {
[h1 @ b'0'...b'2', h2 @ b'0'...b'9', ..] =>
(h1 - b'0') as i32 * 10 + (h2 - b'0') as i32,
_ => return None
};
s = &s[2..];
// optional colons and whitespaces
s = s.trim_left_matches(|&: c: char| c == ':' || c.is_whitespace());
// minutes (00--59)
let minutes = match s.as_bytes() {
[m1 @ b'0'...b'5', m2 @ b'0'...b'9', ..] =>
(m1 - b'0') as i32 * 10 + (m2 - b'0') as i32,
_ => return None
};
s = &s[2..];
let seconds = hours * 3600 + minutes * 60;
if seconds > 86400 { return None; } // range check for hours
Some((s, if negative {-seconds} else {seconds}))
}
// compares two slice case-insensitively (in ASCII).
// assumes the `pattern` is already converted to lower case.
fn equals_ascii_nocase(s: &str, pattern: &str) -> bool {
iter::order::equals(s.as_bytes().iter().map(|&: &c| match c { b'A'...b'Z' => c + 32,
_ => c }),
pattern.as_bytes().iter().cloned())
}
for item in items {
match item {
Item::Literal(prefix) => {
if !s.starts_with(prefix) { return false; }
s = &s[prefix.len()..];
}
Item::Space(_) => {
s = s.trim_left();
}
Item::Numeric(spec, _pad) => {
use self::Numeric::*;
fn set_weekday_with_num_days_from_sunday(p: &mut Parsed, v: i64) -> bool {
p.set_weekday(match v {
0 => Weekday::Sun, 1 => Weekday::Mon, 2 => Weekday::Tue,
3 => Weekday::Wed, 4 => Weekday::Thu, 5 => Weekday::Fri,
6 => Weekday::Sat, _ => return false
})
}
fn set_weekday_with_number_from_monday(p: &mut Parsed, v: i64) -> bool {
p.set_weekday(match v {
1 => Weekday::Mon, 2 => Weekday::Tue, 3 => Weekday::Wed,
4 => Weekday::Thu, 5 => Weekday::Fri, 6 => Weekday::Sat,
7 => Weekday::Sun, _ => return false
})
}
let (width, set): (usize, fn(&mut Parsed, i64) -> bool) = match spec {
Year => (4, Parsed::set_year),
YearDiv100 => (2, Parsed::set_year_div_100),
YearMod100 => (2, Parsed::set_year_mod_100),
IsoYear => (4, Parsed::set_isoyear),
IsoYearDiv100 => (2, Parsed::set_isoyear_div_100),
IsoYearMod100 => (2, Parsed::set_isoyear_mod_100),
Month => (2, Parsed::set_month),
Day => (2, Parsed::set_day),
WeekFromSun => (2, Parsed::set_week_from_sun),
WeekFromMon => (2, Parsed::set_week_from_mon),
IsoWeek => (2, Parsed::set_isoweek),
NumDaysFromSun => (1, set_weekday_with_num_days_from_sunday),
WeekdayFromMon => (1, set_weekday_with_number_from_monday),
Ordinal => (3, Parsed::set_ordinal),
Hour => (2, Parsed::set_hour),
Hour12 => (2, Parsed::set_hour12),
Minute => (2, Parsed::set_minute),
Second => (2, Parsed::set_second),
Nanosecond => (9, Parsed::set_nanosecond),
Timestamp => (usize::MAX, Parsed::set_timestamp),
};
// strip zero or more whitespaces
s = s.trim_left();
// scan digits
let mut win = s.as_bytes();
if win.len() > width { win = &win[..width]; }
let upto = win.iter().position(|&c| c < b'0' || b'9' < c).unwrap_or(win.len());
if upto == 0 { return false; } // no digits detected
if let Ok(v) = s[..upto].parse() {
if !set(parsed, v) { return false; }
s = &s[upto..];
} else {
// overflow. this is possible with `Timestamp` for example
return false;
}
}
Item::Fixed(spec) => {
use self::Fixed::*;
match spec {
ShortMonthName => {
if let Some(month0) = parse_short_month(s) {
if !parsed.set_month(month0 as i64 + 1) { return false; }
s = &s[3..];
} else {
return false;
}
}
LongMonthName => {
if let Some(month0) = parse_short_month(s) {
// three-letter abbreviation is a prefix of the corresponding long name
let long = LONG_MONTHS[month0 as usize];
if s.len() >= long.len() && equals_ascii_nocase(&s[3..long.len()],
&long[3..]) {
// *optionally* consume the long form if possible
s = &s[long.len()..];
} else {
s = &s[3..];
}
if !parsed.set_month(month0 as i64 + 1) { return false; }
} else {
return false;
}
}
ShortWeekdayName => {
if let Some(weekday) = parse_short_weekday(s) {
if !parsed.set_weekday(weekday) { return false; }
s = &s[3..];
} else {
return false;
}
}
LongWeekdayName => {
if let Some(weekday) = parse_short_weekday(s) {
// three-letter abbreviation is a prefix of the corresponding long name
let long = LONG_WEEKDAYS[weekday.num_days_from_monday() as usize];
if s.len() >= long.len() && equals_ascii_nocase(&s[3..long.len()],
&long[3..]) {
// *optionally* consume the long form if possible
s = &s[long.len()..];
} else {
s = &s[3..];
}
if !parsed.set_weekday(weekday) { return false; }
} else {
return false;
}
}
LowerAmPm | UpperAmPm => {
if s.len() < 2 { return false; }
let ampm = match [s.as_bytes()[0] | 32, s.as_bytes()[1] | 32] {
[b'a',b'm'] => false,
[b'p',b'm'] => true,
_ => return false
};
if !parsed.set_ampm(ampm) { return false; }
s = &s[2..];
}
// not supported in the parser
TimezoneName => return false,
TimezoneOffset | TimezoneOffsetZ => {
let allow_zulu = spec == TimezoneOffsetZ;
if let Some((s_, offset)) = parse_timezone_offset(s, allow_zulu) {
s = s_;
if !parsed.set_offset(offset as i64) { return false; }
} else {
return false;
}
}
}
}
Item::Error => {
return false;
}
}
}
// if there are trailling chars, it is an error
s.is_empty()
}
/// A *temporary* object which can be used as an argument to `format!` or others.
/// This is normally constructed via `format` methods of each date and time type.
#[derive(Debug)]
@ -276,3 +585,200 @@ pub mod parsed;
pub mod strftime;
#[cfg(test)]
#[test]
fn test_parse() {
macro_rules! check {
($fmt:expr, $items:expr; _) => ({
assert!(!parse(&mut Parsed::new(), $fmt, $items.iter().cloned()));
});
($fmt:expr, $items:expr; $($k:ident: $v:expr),*) => ({
let mut parsed = Parsed::new();
assert!(parse(&mut parsed, $fmt, $items.iter().cloned()));
assert_eq!(parsed, Parsed { $($k: Some($v),)* ..Parsed::new() });
});
}
// empty string
check!("", []; );
check!(" ", []; _);
check!("a", []; _);
// whitespaces
check!("", [sp!("")]; );
check!(" ", [sp!("")]; );
check!("\t", [sp!("")]; );
check!(" \n\r \n", [sp!("")]; );
check!("a", [sp!("")]; _);
// literal
check!("", [lit!("a")]; _);
check!(" ", [lit!("a")]; _);
check!("a", [lit!("a")]; );
check!("aa", [lit!("a")]; _);
check!("A", [lit!("a")]; _);
check!("xy", [lit!("xy")]; );
check!("xy", [lit!("x"), lit!("y")]; );
check!("x y", [lit!("x"), lit!("y")]; _);
check!("xy", [lit!("x"), sp!(""), lit!("y")]; );
check!("x y", [lit!("x"), sp!(""), lit!("y")]; );
// numeric
check!("1987", [num!(Year)]; year_div_100: 19, year_mod_100: 87);
check!("1987 ", [num!(Year)]; _);
check!("0x12", [num!(Year)]; _);
check!("2015", [num!(Year)]; year_div_100: 20, year_mod_100: 15);
check!("0000", [num!(Year)]; year_div_100: 0, year_mod_100: 0);
check!("9999", [num!(Year)]; year_div_100: 99, year_mod_100: 99);
check!(" \t987", [num!(Year)]; year_div_100: 9, year_mod_100: 87);
check!("5", [num!(Year)]; year_div_100: 0, year_mod_100: 5);
check!("-42", [num!(Year)]; _);
check!("+42", [num!(Year)]; _);
check!("5\0", [num!(Year)]; _);
check!("\05", [num!(Year)]; _);
check!("", [num!(Year)]; _);
check!("12345", [num!(Year), lit!("5")]; year_div_100: 12, year_mod_100: 34);
check!("12345", [nums!(Year), lit!("5")]; year_div_100: 12, year_mod_100: 34);
check!("12345", [num0!(Year), lit!("5")]; year_div_100: 12, year_mod_100: 34);
check!("12341234", [num!(Year), num!(Year)]; year_div_100: 12, year_mod_100: 34);
check!("1234 1234", [num!(Year), num!(Year)]; year_div_100: 12, year_mod_100: 34);
check!("1234 1235", [num!(Year), num!(Year)]; _);
check!("1234 1234", [num!(Year), lit!("x"), num!(Year)]; _);
check!("1234x1234", [num!(Year), lit!("x"), num!(Year)];
year_div_100: 12, year_mod_100: 34);
check!("1234xx1234", [num!(Year), lit!("x"), num!(Year)]; _);
check!("1234 x 1234", [num!(Year), lit!("x"), num!(Year)]; _);
// various numeric fields
check!("1234 5678",
[num!(Year), num!(IsoYear)];
year_div_100: 12, year_mod_100: 34, isoyear_div_100: 56, isoyear_mod_100: 78);
check!("12 34 56 78",
[num!(YearDiv100), num!(YearMod100), num!(IsoYearDiv100), num!(IsoYearMod100)];
year_div_100: 12, year_mod_100: 34, isoyear_div_100: 56, isoyear_mod_100: 78);
check!("1 2 3 4 5 6",
[num!(Month), num!(Day), num!(WeekFromSun), num!(WeekFromMon), num!(IsoWeek),
num!(NumDaysFromSun)];
month: 1, day: 2, week_from_sun: 3, week_from_mon: 4, isoweek: 5, weekday: Weekday::Sat);
check!("7 89 01",
[num!(WeekdayFromMon), num!(Ordinal), num!(Hour12)];
weekday: Weekday::Sun, ordinal: 89, hour_mod_12: 1);
check!("23 45 6 78901234 567890123",
[num!(Hour), num!(Minute), num!(Second), num!(Nanosecond), num!(Timestamp)];
hour_div_12: 1, hour_mod_12: 11, minute: 45, second: 6, nanosecond: 78_901_234,
timestamp: 567_890_123);
// fixed: month and weekday names
check!("apr", [fix!(ShortMonthName)]; month: 4);
check!("Apr", [fix!(ShortMonthName)]; month: 4);
check!("APR", [fix!(ShortMonthName)]; month: 4);
check!("ApR", [fix!(ShortMonthName)]; month: 4);
check!("April", [fix!(ShortMonthName)]; _);
check!("A", [fix!(ShortMonthName)]; _);
check!("Sol", [fix!(ShortMonthName)]; _);
check!("Apr", [fix!(LongMonthName)]; month: 4);
check!("Apri", [fix!(LongMonthName)]; _);
check!("April", [fix!(LongMonthName)]; month: 4);
check!("Aprill", [fix!(LongMonthName)]; _);
check!("Aprill", [fix!(LongMonthName), lit!("l")]; month: 4);
check!("Aprl", [fix!(LongMonthName), lit!("l")]; month: 4);
check!("April", [fix!(LongMonthName), lit!("il")]; _); // do not backtrack
check!("thu", [fix!(ShortWeekdayName)]; weekday: Weekday::Thu);
check!("Thu", [fix!(ShortWeekdayName)]; weekday: Weekday::Thu);
check!("THU", [fix!(ShortWeekdayName)]; weekday: Weekday::Thu);
check!("tHu", [fix!(ShortWeekdayName)]; weekday: Weekday::Thu);
check!("Thursday", [fix!(ShortWeekdayName)]; _);
check!("T", [fix!(ShortWeekdayName)]; _);
check!("The", [fix!(ShortWeekdayName)]; _);
check!("Nop", [fix!(ShortWeekdayName)]; _);
check!("Thu", [fix!(LongWeekdayName)]; weekday: Weekday::Thu);
check!("Thur", [fix!(LongWeekdayName)]; _);
check!("Thurs", [fix!(LongWeekdayName)]; _);
check!("Thursday", [fix!(LongWeekdayName)]; weekday: Weekday::Thu);
check!("Thursdays", [fix!(LongWeekdayName)]; _);
check!("Thursdays", [fix!(LongWeekdayName), lit!("s")]; weekday: Weekday::Thu);
check!("Thus", [fix!(LongWeekdayName), lit!("s")]; weekday: Weekday::Thu);
check!("Thursday", [fix!(LongWeekdayName), lit!("rsday")]; _); // do not backtrack
// fixed: am/pm
check!("am", [fix!(LowerAmPm)]; hour_div_12: 0);
check!("pm", [fix!(LowerAmPm)]; hour_div_12: 1);
check!("AM", [fix!(LowerAmPm)]; hour_div_12: 0);
check!("PM", [fix!(LowerAmPm)]; hour_div_12: 1);
check!("am", [fix!(UpperAmPm)]; hour_div_12: 0);
check!("pm", [fix!(UpperAmPm)]; hour_div_12: 1);
check!("AM", [fix!(UpperAmPm)]; hour_div_12: 0);
check!("PM", [fix!(UpperAmPm)]; hour_div_12: 1);
check!("Am", [fix!(LowerAmPm)]; hour_div_12: 0);
check!(" Am", [fix!(LowerAmPm)]; _);
check!("ame", [fix!(LowerAmPm)]; _);
check!("a", [fix!(LowerAmPm)]; _);
check!("p", [fix!(LowerAmPm)]; _);
check!("x", [fix!(LowerAmPm)]; _);
check!("xx", [fix!(LowerAmPm)]; _);
check!("", [fix!(LowerAmPm)]; _);
// fixed: timezone offsets
check!("+00:00", [fix!(TimezoneOffset)]; offset: 0);
check!("-00:00", [fix!(TimezoneOffset)]; offset: 0);
check!("+00:01", [fix!(TimezoneOffset)]; offset: 60);
check!("-00:01", [fix!(TimezoneOffset)]; offset: -60);
check!("+00:30", [fix!(TimezoneOffset)]; offset: 30 * 60);
check!("-00:30", [fix!(TimezoneOffset)]; offset: -30 * 60);
check!("+04:56", [fix!(TimezoneOffset)]; offset: 296 * 60);
check!("-04:56", [fix!(TimezoneOffset)]; offset: -296 * 60);
check!("+24:00", [fix!(TimezoneOffset)]; offset: 24 * 60 * 60);
check!("-24:00", [fix!(TimezoneOffset)]; offset: -24 * 60 * 60);
check!("+24:01", [fix!(TimezoneOffset)]; _);
check!("-24:01", [fix!(TimezoneOffset)]; _);
check!("+00:59", [fix!(TimezoneOffset)]; offset: 59 * 60);
check!("+00:60", [fix!(TimezoneOffset)]; _);
check!("+00:99", [fix!(TimezoneOffset)]; _);
check!("+99:00", [fix!(TimezoneOffset)]; _);
check!("#12:34", [fix!(TimezoneOffset)]; _);
check!("12:34", [fix!(TimezoneOffset)]; _);
check!("+12:34 ", [fix!(TimezoneOffset)]; _);
check!(" +12:34", [fix!(TimezoneOffset)]; offset: 754 * 60);
check!("\t -12:34", [fix!(TimezoneOffset)]; offset: -754 * 60);
check!("+", [fix!(TimezoneOffset)]; _);
check!("+1", [fix!(TimezoneOffset)]; _);
check!("+12", [fix!(TimezoneOffset)]; _);
check!("+123", [fix!(TimezoneOffset)]; _);
check!("+1234", [fix!(TimezoneOffset)]; offset: 754 * 60);
check!("+12345", [fix!(TimezoneOffset)]; _);
check!("+12345", [fix!(TimezoneOffset), num!(Day)]; offset: 754 * 60, day: 5);
check!("Z", [fix!(TimezoneOffset)]; _);
check!("Z", [fix!(TimezoneOffsetZ)]; offset: 0);
check!("z", [fix!(TimezoneOffsetZ)]; offset: 0);
check!("Y", [fix!(TimezoneOffsetZ)]; _);
check!("Zulu", [fix!(TimezoneOffsetZ), lit!("ulu")]; offset: 0);
check!("zulu", [fix!(TimezoneOffsetZ), lit!("ulu")]; offset: 0);
check!("+1234ulu", [fix!(TimezoneOffsetZ), lit!("ulu")]; offset: 754 * 60);
check!("+12:34ulu", [fix!(TimezoneOffsetZ), lit!("ulu")]; offset: 754 * 60);
check!("???", [fix!(TimezoneName)]; _); // not allowed
// some practical examples
check!("2015-02-04T14:37:05+09:00",
[num!(Year), lit!("-"), num!(Month), lit!("-"), num!(Day), lit!("T"),
num!(Hour), lit!(":"), num!(Minute), lit!(":"), num!(Second), fix!(TimezoneOffset)];
year_div_100: 20, year_mod_100: 15, month: 2, day: 4,
hour_div_12: 1, hour_mod_12: 2, minute: 37, second: 5, offset: 32400);
check!("Mon, 10 Jun 2013 09:32:37 GMT",
[fix!(ShortWeekdayName), lit!(","), sp!(" "), num!(Day), sp!(" "),
fix!(ShortMonthName), sp!(" "), num!(Year), sp!(" "), num!(Hour), lit!(":"),
num!(Minute), lit!(":"), num!(Second), sp!(" "), lit!("GMT")];
year_div_100: 20, year_mod_100: 13, month: 6, day: 10, weekday: Weekday::Mon,
hour_div_12: 0, hour_mod_12: 9, minute: 32, second: 37);
check!("20060102150405",
[num!(Year), num!(Month), num!(Day), num!(Hour), num!(Minute), num!(Second)];
year_div_100: 20, year_mod_100: 6, month: 1, day: 2,
hour_div_12: 1, hour_mod_12: 3, minute: 4, second: 5);
check!("3:14PM",
[num!(Hour12), lit!(":"), num!(Minute), fix!(LowerAmPm)];
hour_div_12: 1, hour_mod_12: 3, minute: 14);
check!("12345678901234.56789",
[num!(Timestamp), lit!("."), num!(Nanosecond)];
nanosecond: 56_789, timestamp: 12_345_678_901_234);
}

View File

@ -21,7 +21,7 @@ use datetime::DateTime;
/// Parsed parts of date and time.
#[allow(missing_copy_implementations)]
#[derive(Clone, Debug)]
#[derive(Clone, PartialEq, Debug)]
pub struct Parsed {
/// Year divided by 100. Implies that the year is >= 1 BCE.
///