2018-05-13 16:18:45 -04:00
|
|
|
extern crate chrono;
|
|
|
|
|
2018-05-17 22:56:46 -04:00
|
|
|
use chrono::Datelike;
|
|
|
|
use chrono::Local;
|
|
|
|
use std::collections::HashMap;
|
2018-05-15 00:50:14 -04:00
|
|
|
use std::vec::Vec;
|
2018-05-13 16:18:45 -04:00
|
|
|
|
2018-05-15 00:50:14 -04:00
|
|
|
#[cfg(test)]
|
2018-05-17 00:31:57 -04:00
|
|
|
mod test_python_compat;
|
2018-05-18 23:50:53 -04:00
|
|
|
#[cfg(test)]
|
|
|
|
mod tests;
|
|
|
|
|
|
|
|
#[derive(Debug, PartialEq)]
|
|
|
|
pub enum ParseError {
|
|
|
|
InvalidMonth,
|
|
|
|
}
|
2018-05-15 00:50:14 -04:00
|
|
|
|
2018-05-18 23:50:53 -04:00
|
|
|
enum ParseInternalError {
|
|
|
|
// Errors that indicate internal bugs
|
|
|
|
YMDEarlyResolve,
|
|
|
|
YMDValueUnset,
|
|
|
|
|
|
|
|
// Python-style errors
|
|
|
|
ValueError(String),
|
|
|
|
}
|
|
|
|
|
|
|
|
type ParseResult<I> = Result<I, ParseError>;
|
|
|
|
type ParseIResult<I> = Result<I, ParseInternalError>;
|
|
|
|
|
|
|
|
#[derive(Debug, PartialEq)]
|
2018-05-15 00:50:14 -04:00
|
|
|
pub enum Token {
|
|
|
|
Alpha(String),
|
|
|
|
Numeric(String),
|
|
|
|
Separator(String),
|
|
|
|
}
|
|
|
|
|
|
|
|
pub struct Tokenizer {
|
|
|
|
token_stack: Vec<Token>,
|
|
|
|
parse_string: String,
|
|
|
|
}
|
|
|
|
|
|
|
|
#[derive(Debug)]
|
|
|
|
enum ParseState {
|
|
|
|
Empty,
|
|
|
|
Alpha,
|
|
|
|
AlphaDecimal,
|
|
|
|
Numeric,
|
|
|
|
NumericDecimal,
|
|
|
|
}
|
|
|
|
|
|
|
|
impl Tokenizer {
|
|
|
|
fn new(parse_string: String) -> Self {
|
|
|
|
Tokenizer {
|
|
|
|
token_stack: Vec::new(),
|
|
|
|
parse_string: parse_string.chars().rev().collect(),
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
impl Iterator for Tokenizer {
|
|
|
|
type Item = Token;
|
|
|
|
|
|
|
|
fn next(&mut self) -> Option<Self::Item> {
|
2018-05-17 00:31:57 -04:00
|
|
|
if !self.token_stack.is_empty() {
|
|
|
|
return Some(self.token_stack.pop().unwrap());
|
|
|
|
};
|
|
|
|
if self.parse_string.is_empty() {
|
|
|
|
return None;
|
|
|
|
};
|
2018-05-13 16:18:45 -04:00
|
|
|
|
2018-05-15 00:50:14 -04:00
|
|
|
let mut char_stack: Vec<char> = Vec::new();
|
|
|
|
let mut seen_letters = false;
|
|
|
|
let mut state = ParseState::Empty;
|
|
|
|
|
|
|
|
while let Some(next) = self.parse_string.pop() {
|
|
|
|
println!("{} - {:?}", next, state);
|
|
|
|
match state {
|
2018-05-17 00:31:57 -04:00
|
|
|
ParseState::Empty => {
|
|
|
|
if next.is_numeric() {
|
|
|
|
state = ParseState::Numeric;
|
|
|
|
char_stack.push(next);
|
|
|
|
} else if next.is_alphabetic() {
|
|
|
|
state = ParseState::Alpha;
|
|
|
|
seen_letters = true;
|
|
|
|
char_stack.push(next);
|
|
|
|
} else if next.is_whitespace() {
|
|
|
|
char_stack.push(' ');
|
|
|
|
break;
|
|
|
|
} else {
|
|
|
|
char_stack.push(next);
|
|
|
|
break;
|
|
|
|
}
|
2018-05-15 00:50:14 -04:00
|
|
|
}
|
2018-05-17 00:31:57 -04:00
|
|
|
ParseState::Alpha => {
|
|
|
|
if next.is_alphabetic() {
|
|
|
|
char_stack.push(next);
|
|
|
|
} else if next == '.' {
|
|
|
|
state = ParseState::AlphaDecimal;
|
|
|
|
char_stack.push(next);
|
|
|
|
} else {
|
|
|
|
// We don't recognize the character, so push it back
|
|
|
|
// to be handled later.
|
|
|
|
self.parse_string.push(next);
|
|
|
|
break;
|
|
|
|
}
|
2018-05-15 00:50:14 -04:00
|
|
|
}
|
2018-05-17 00:31:57 -04:00
|
|
|
ParseState::AlphaDecimal => {
|
|
|
|
if next == '.' || next.is_alphabetic() {
|
|
|
|
char_stack.push(next);
|
|
|
|
} else if next.is_numeric() && char_stack.last().unwrap().clone() == '.' {
|
|
|
|
char_stack.push(next);
|
|
|
|
state = ParseState::NumericDecimal;
|
|
|
|
} else {
|
|
|
|
self.parse_string.push(next);
|
|
|
|
break;
|
|
|
|
}
|
2018-05-15 00:50:14 -04:00
|
|
|
}
|
2018-05-17 00:31:57 -04:00
|
|
|
ParseState::Numeric => {
|
|
|
|
if next.is_numeric() {
|
|
|
|
char_stack.push(next);
|
|
|
|
} else if next == '.' || (next == ',' && char_stack.len() >= 2) {
|
|
|
|
char_stack.push(next);
|
|
|
|
state = ParseState::NumericDecimal;
|
|
|
|
} else {
|
|
|
|
// We don't recognize the character, so push it back
|
|
|
|
// to be handled later
|
|
|
|
self.parse_string.push(next);
|
|
|
|
break;
|
|
|
|
}
|
2018-05-15 00:50:14 -04:00
|
|
|
}
|
2018-05-17 00:31:57 -04:00
|
|
|
ParseState::NumericDecimal => {
|
|
|
|
if next == '.' || next.is_numeric() {
|
|
|
|
char_stack.push(next);
|
|
|
|
} else if next.is_alphabetic() && char_stack.last().unwrap().clone() == '.' {
|
|
|
|
char_stack.push(next);
|
|
|
|
state = ParseState::AlphaDecimal;
|
|
|
|
} else {
|
|
|
|
self.parse_string.push(next);
|
|
|
|
break;
|
|
|
|
}
|
2018-05-15 00:50:14 -04:00
|
|
|
}
|
|
|
|
}
|
2018-05-17 00:31:57 -04:00
|
|
|
}
|
2018-05-15 00:50:14 -04:00
|
|
|
|
|
|
|
// I like Python's version of this much better:
|
|
|
|
// needs_split = seen_letters or char_stack.count('.') > 1 or char_stack[-1] in '.,'
|
2018-05-17 00:31:57 -04:00
|
|
|
let dot_count = char_stack.iter().fold(0, |count, character| {
|
|
|
|
count + (if character == &'.' { 1 } else { 0 })
|
|
|
|
});
|
|
|
|
let needs_split = seen_letters || dot_count > 1 || char_stack.last().unwrap() == &'.'
|
|
|
|
|| char_stack.last().unwrap() == &',';
|
2018-05-15 00:50:14 -04:00
|
|
|
let final_string = char_stack.into_iter().collect();
|
|
|
|
|
|
|
|
let mut tokens = match state {
|
|
|
|
ParseState::Empty => vec![Token::Separator(final_string)],
|
|
|
|
ParseState::Alpha => vec![Token::Alpha(final_string)],
|
|
|
|
ParseState::Numeric => vec![Token::Numeric(final_string)],
|
|
|
|
ParseState::AlphaDecimal => {
|
|
|
|
if needs_split {
|
2018-05-17 00:31:57 -04:00
|
|
|
decimal_split(&final_string, false)
|
2018-05-15 00:50:14 -04:00
|
|
|
} else {
|
|
|
|
vec![Token::Alpha(final_string)]
|
|
|
|
}
|
|
|
|
}
|
|
|
|
ParseState::NumericDecimal => {
|
|
|
|
if needs_split {
|
|
|
|
decimal_split(&final_string, dot_count == 0)
|
|
|
|
} else {
|
|
|
|
vec![Token::Numeric(final_string)]
|
|
|
|
}
|
|
|
|
}
|
2018-05-17 00:31:57 -04:00
|
|
|
}.into_iter()
|
|
|
|
.rev()
|
|
|
|
.collect();
|
2018-05-15 00:50:14 -04:00
|
|
|
|
|
|
|
self.token_stack.append(&mut tokens);
|
|
|
|
// UNWRAP: Previous match guaranteed that at least one token was added
|
|
|
|
Some(self.token_stack.pop().unwrap())
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
fn decimal_split(characters: &str, cast_period: bool) -> Vec<Token> {
|
|
|
|
let mut token_stack: Vec<Token> = Vec::new();
|
|
|
|
let mut char_stack: Vec<char> = Vec::new();
|
|
|
|
let mut state = ParseState::Empty;
|
|
|
|
|
2018-05-17 00:31:57 -04:00
|
|
|
for c in characters.chars() {
|
|
|
|
match state {
|
|
|
|
ParseState::Empty => {
|
|
|
|
if c.is_alphabetic() {
|
|
|
|
char_stack.push(c);
|
|
|
|
state = ParseState::Alpha;
|
|
|
|
} else if c.is_numeric() {
|
|
|
|
char_stack.push(c);
|
|
|
|
state = ParseState::Numeric;
|
|
|
|
} else {
|
|
|
|
let character = if cast_period { '.' } else { c };
|
|
|
|
token_stack.push(Token::Separator(character.to_string()));
|
|
|
|
}
|
2018-05-15 00:50:14 -04:00
|
|
|
}
|
2018-05-17 00:31:57 -04:00
|
|
|
ParseState::Alpha => {
|
|
|
|
if c.is_alphabetic() {
|
|
|
|
char_stack.push(c);
|
|
|
|
} else {
|
|
|
|
token_stack.push(Token::Alpha(char_stack.iter().collect()));
|
|
|
|
char_stack.clear();
|
|
|
|
let character = if cast_period { '.' } else { c };
|
|
|
|
token_stack.push(Token::Separator(character.to_string()));
|
|
|
|
state = ParseState::Empty;
|
|
|
|
}
|
2018-05-15 00:50:14 -04:00
|
|
|
}
|
2018-05-17 00:31:57 -04:00
|
|
|
ParseState::Numeric => {
|
|
|
|
if c.is_numeric() {
|
|
|
|
char_stack.push(c);
|
|
|
|
} else {
|
|
|
|
token_stack.push(Token::Numeric(char_stack.iter().collect()));
|
|
|
|
char_stack.clear();
|
|
|
|
let character = if cast_period { '.' } else { c };
|
|
|
|
token_stack.push(Token::Separator(character.to_string()));
|
|
|
|
state = ParseState::Empty;
|
|
|
|
}
|
2018-05-15 00:50:14 -04:00
|
|
|
}
|
2018-05-17 00:31:57 -04:00
|
|
|
_ => panic!("Invalid parse state during decimal_split()"),
|
|
|
|
}
|
|
|
|
}
|
2018-05-15 00:50:14 -04:00
|
|
|
|
|
|
|
match state {
|
|
|
|
ParseState::Alpha => token_stack.push(Token::Alpha(char_stack.iter().collect())),
|
|
|
|
ParseState::Numeric => token_stack.push(Token::Numeric(char_stack.iter().collect())),
|
|
|
|
ParseState::Empty => (),
|
2018-05-17 00:31:57 -04:00
|
|
|
_ => panic!("Invalid parse state during decimal_split()"),
|
2018-05-15 00:50:14 -04:00
|
|
|
}
|
|
|
|
|
|
|
|
token_stack
|
|
|
|
}
|
|
|
|
|
|
|
|
pub fn tokenize(parse_string: &str) -> Vec<Token> {
|
|
|
|
let tokenizer = Tokenizer::new(parse_string.to_owned());
|
|
|
|
tokenizer.collect()
|
|
|
|
}
|
2018-05-17 22:56:46 -04:00
|
|
|
|
|
|
|
fn parse_info(vec: Vec<Vec<&str>>) -> HashMap<String, usize> {
|
|
|
|
let mut m = HashMap::new();
|
|
|
|
|
|
|
|
if vec.len() == 1 {
|
|
|
|
for (i, val) in vec.get(0).unwrap().into_iter().enumerate() {
|
|
|
|
m.insert(val.to_lowercase(), i);
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
for (i, val_vec) in vec.into_iter().enumerate() {
|
|
|
|
for val in val_vec.into_iter() {
|
|
|
|
m.insert(val.to_lowercase(), i);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
m
|
|
|
|
}
|
|
|
|
|
|
|
|
struct ParserInfo {
|
|
|
|
jump: HashMap<String, usize>,
|
|
|
|
weekday: HashMap<String, usize>,
|
|
|
|
months: HashMap<String, usize>,
|
|
|
|
hms: HashMap<String, usize>,
|
|
|
|
ampm: HashMap<String, usize>,
|
|
|
|
utczone: HashMap<String, usize>,
|
|
|
|
pertain: HashMap<String, usize>,
|
|
|
|
tzoffset: HashMap<String, usize>,
|
|
|
|
dayfirst: bool,
|
|
|
|
yearfirst: bool,
|
|
|
|
year: u32,
|
|
|
|
century: u32,
|
|
|
|
}
|
|
|
|
|
|
|
|
impl Default for ParserInfo {
|
|
|
|
fn default() -> Self {
|
|
|
|
let year = Local::now().year();
|
|
|
|
let century = year / 100 * 100;
|
|
|
|
|
|
|
|
ParserInfo {
|
2018-05-18 23:50:53 -04:00
|
|
|
jump: parse_info(vec![
|
|
|
|
vec![
|
|
|
|
" ", ".", ",", ";", "-", "/", "'", "at", "on", "and", "ad", "m", "t", "of",
|
|
|
|
"st", "nd", "rd", "th",
|
|
|
|
],
|
|
|
|
]),
|
2018-05-17 22:56:46 -04:00
|
|
|
weekday: parse_info(vec![
|
|
|
|
vec!["Mon", "Monday"],
|
|
|
|
vec!["Tue", "Tues", "Tuesday"],
|
|
|
|
vec!["Wed", "Wednesday"],
|
|
|
|
vec!["Thu", "Thurs", "Thursday"],
|
|
|
|
vec!["Fri", "Friday"],
|
|
|
|
vec!["Sat", "Saturday"],
|
|
|
|
vec!["Sun", "Sunday"],
|
|
|
|
]),
|
|
|
|
months: parse_info(vec![
|
|
|
|
vec!["Jan", "January"],
|
|
|
|
vec!["Feb", "February"],
|
|
|
|
vec!["Mar", "March"],
|
|
|
|
vec!["Apr", "April"],
|
|
|
|
vec!["May"],
|
|
|
|
vec!["Jun", "June"],
|
|
|
|
vec!["Jul", "July"],
|
|
|
|
vec!["Aug", "August"],
|
|
|
|
vec!["Sep", "Sept", "September"],
|
|
|
|
vec!["Oct", "October"],
|
|
|
|
vec!["Nov", "November"],
|
|
|
|
vec!["Dec", "December"],
|
|
|
|
]),
|
|
|
|
hms: parse_info(vec![
|
|
|
|
vec!["h", "hour", "hours"],
|
|
|
|
vec!["m", "minute", "minutes"],
|
|
|
|
vec!["s", "second", "seconds"],
|
|
|
|
]),
|
2018-05-18 23:50:53 -04:00
|
|
|
ampm: parse_info(vec![vec!["am", "a"], vec!["pm", "p"]]),
|
|
|
|
utczone: parse_info(vec![vec!["UTC", "GMT", "Z"]]),
|
2018-05-17 22:56:46 -04:00
|
|
|
pertain: parse_info(vec![vec!["of"]]),
|
|
|
|
tzoffset: parse_info(vec![vec![]]),
|
|
|
|
dayfirst: false,
|
|
|
|
yearfirst: false,
|
|
|
|
year: year as u32,
|
|
|
|
century: century as u32,
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
impl ParserInfo {
|
|
|
|
fn get_jump(&self, name: &str) -> bool {
|
|
|
|
self.jump.contains_key(&name.to_lowercase())
|
|
|
|
}
|
|
|
|
|
|
|
|
fn get_weekday(&self, name: &str) -> Option<usize> {
|
|
|
|
self.weekday.get(&name.to_lowercase()).cloned() // TODO: Why do I have to clone a primitive?
|
|
|
|
}
|
|
|
|
|
|
|
|
fn get_month(&self, name: &str) -> Option<usize> {
|
|
|
|
self.months.get(&name.to_lowercase()).map(|u| u + 1)
|
|
|
|
}
|
|
|
|
|
|
|
|
fn get_hms(&self, name: &str) -> Option<usize> {
|
|
|
|
self.hms.get(&name.to_lowercase()).cloned()
|
|
|
|
}
|
|
|
|
|
|
|
|
fn get_ampm(&self, name: &str) -> Option<usize> {
|
|
|
|
self.ampm.get(&name.to_lowercase()).cloned()
|
|
|
|
}
|
|
|
|
|
|
|
|
fn get_pertain(&self, name: &str) -> bool {
|
|
|
|
self.pertain.contains_key(&name.to_lowercase())
|
|
|
|
}
|
|
|
|
|
|
|
|
fn get_utczone(&self, name: &str) -> bool {
|
|
|
|
self.utczone.contains_key(&name.to_lowercase())
|
|
|
|
}
|
|
|
|
|
|
|
|
fn get_tzoffset(&self, name: &str) -> Option<usize> {
|
|
|
|
if self.utczone.contains_key(&name.to_lowercase()) {
|
|
|
|
Some(0)
|
|
|
|
} else {
|
|
|
|
self.tzoffset.get(&name.to_lowercase()).cloned()
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
fn convertyear(&self, year: u32, century_specified: bool) -> u32 {
|
|
|
|
let mut year = year;
|
|
|
|
|
|
|
|
if year < 100 && !century_specified {
|
|
|
|
year += self.century;
|
|
|
|
if year >= self.year + 50 {
|
|
|
|
year -= 100;
|
|
|
|
} else if year < self.year - 50 {
|
|
|
|
year += 100
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
year
|
|
|
|
}
|
2018-05-18 23:50:53 -04:00
|
|
|
}
|
|
|
|
|
|
|
|
fn days_in_month(year: i32, month: i32) -> Result<i32, ParseError> {
|
|
|
|
let leap_year = match year % 4 {
|
|
|
|
0 => year % 400 == 0,
|
|
|
|
_ => false,
|
|
|
|
};
|
|
|
|
|
|
|
|
match month {
|
|
|
|
2 => if leap_year {
|
|
|
|
Ok(29)
|
|
|
|
} else {
|
|
|
|
Ok(28)
|
|
|
|
},
|
|
|
|
1 | 3 | 5 | 7 | 8 | 10 | 12 => Ok(31),
|
|
|
|
4 | 6 | 9 | 11 => Ok(30),
|
|
|
|
_ => Err(ParseError::InvalidMonth),
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
#[derive(Debug, Hash, PartialEq, Eq)]
|
|
|
|
enum YMDLabel {
|
|
|
|
Year,
|
|
|
|
Month,
|
|
|
|
Day,
|
|
|
|
}
|
|
|
|
|
|
|
|
struct YMD {
|
|
|
|
_ymd: Vec<i32>, // TODO: This seems like a super weird way to store things
|
|
|
|
century_specified: bool,
|
|
|
|
dstridx: Option<usize>,
|
|
|
|
mstridx: Option<usize>,
|
|
|
|
ystridx: Option<usize>,
|
|
|
|
}
|
|
|
|
|
|
|
|
impl YMD {
|
|
|
|
fn could_be_day(&self, val: i32) -> ParseResult<bool> {
|
|
|
|
if self.dstridx.is_some() {
|
|
|
|
Ok(false)
|
|
|
|
} else if self.mstridx.is_none() {
|
|
|
|
Ok((1 <= val) && (val <= 31))
|
|
|
|
} else if self.ystridx.is_none() {
|
|
|
|
// UNWRAP: mstridx guaranteed to have a value
|
|
|
|
// TODO: Justify unwrap for self._ymd
|
|
|
|
let month = self._ymd[self.mstridx.unwrap()];
|
|
|
|
Ok(1 <= val && (val <= days_in_month(2000, month)?))
|
|
|
|
} else {
|
|
|
|
let month = self._ymd[self.mstridx.unwrap()];
|
|
|
|
let year = self._ymd[self.ystridx.unwrap()];
|
|
|
|
Ok(1 <= val && (val <= days_in_month(year, month)?))
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
fn append(&mut self, val: i32, label: Option<YMDLabel>) -> ParseIResult<()> {
|
|
|
|
let mut label = label;
|
|
|
|
|
|
|
|
if val > 100 {
|
|
|
|
self.century_specified = true;
|
|
|
|
match label {
|
|
|
|
None => label = Some(YMDLabel::Year),
|
|
|
|
Some(YMDLabel::Year) => (),
|
|
|
|
_ => {
|
|
|
|
return Err(ParseInternalError::ValueError(format!(
|
|
|
|
"Invalid label: {:?}",
|
|
|
|
label
|
|
|
|
)))
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
match label {
|
|
|
|
Some(YMDLabel::Month) => {
|
|
|
|
if self.mstridx.is_some() {
|
|
|
|
Err(ParseInternalError::ValueError(
|
|
|
|
"Month already set.".to_owned(),
|
|
|
|
))
|
|
|
|
} else {
|
|
|
|
self.mstridx = Some(self._ymd.len() - 1);
|
|
|
|
Ok(())
|
|
|
|
}
|
|
|
|
}
|
|
|
|
Some(YMDLabel::Day) => {
|
|
|
|
if self.dstridx.is_some() {
|
|
|
|
Err(ParseInternalError::ValueError(
|
|
|
|
"Day already set.".to_owned(),
|
|
|
|
))
|
|
|
|
} else {
|
|
|
|
self.dstridx = Some(self._ymd.len() - 1);
|
|
|
|
Ok(())
|
|
|
|
}
|
|
|
|
}
|
|
|
|
Some(YMDLabel::Year) => {
|
|
|
|
if self.ystridx.is_some() {
|
|
|
|
Err(ParseInternalError::ValueError(
|
|
|
|
"Year already set.".to_owned(),
|
|
|
|
))
|
|
|
|
} else {
|
|
|
|
self.ystridx = Some(self._ymd.len() - 1);
|
|
|
|
Ok(())
|
|
|
|
}
|
|
|
|
}
|
|
|
|
None => Err(ParseInternalError::ValueError("Missing label.".to_owned())),
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
fn resolve_from_stridxs(
|
|
|
|
&mut self,
|
|
|
|
strids: &mut HashMap<YMDLabel, usize>,
|
|
|
|
) -> ParseIResult<(i32, i32, i32)> {
|
|
|
|
if strids.len() == 2 {
|
|
|
|
let missing_key = if !strids.contains_key(&YMDLabel::Year) {
|
|
|
|
YMDLabel::Year
|
|
|
|
} else if !strids.contains_key(&YMDLabel::Month) {
|
|
|
|
YMDLabel::Month
|
|
|
|
} else {
|
|
|
|
YMDLabel::Day
|
|
|
|
};
|
|
|
|
|
|
|
|
let strids_vals: Vec<usize> = strids.values().map(|u| u.clone()).collect();
|
|
|
|
let missing_val = if !strids_vals.contains(&0) {
|
|
|
|
0
|
|
|
|
} else if !strids_vals.contains(&1) {
|
|
|
|
1
|
|
|
|
} else {
|
|
|
|
2
|
|
|
|
};
|
|
|
|
|
|
|
|
strids.insert(missing_key, missing_val);
|
|
|
|
}
|
|
|
|
|
|
|
|
if self._ymd.len() != 3 || strids.len() != 3 {
|
|
|
|
return Err(ParseInternalError::YMDEarlyResolve);
|
|
|
|
}
|
|
|
|
|
|
|
|
// TODO: Why do I have to clone &usize? Isn't it Copy?
|
|
|
|
Ok((
|
|
|
|
self._ymd[strids.get(&YMDLabel::Year).unwrap().clone()],
|
|
|
|
self._ymd[strids.get(&YMDLabel::Month).unwrap().clone()],
|
|
|
|
self._ymd[strids.get(&YMDLabel::Day).unwrap().clone()],
|
|
|
|
))
|
|
|
|
}
|
|
|
|
|
|
|
|
fn resolve_ymd(&mut self, yearfirst: bool, dayfirst: bool) -> ParseIResult<(i32, i32, i32)> {
|
|
|
|
let len_ymd = self._ymd.len();
|
|
|
|
let mut year: Option<i32> = None;
|
|
|
|
let mut month: Option<i32> = None;
|
|
|
|
let mut day: Option<i32> = None;
|
|
|
|
let mut other: Option<i32> = None;
|
|
|
|
|
|
|
|
let mut strids: HashMap<YMDLabel, usize> = HashMap::new();
|
|
|
|
self.ystridx
|
|
|
|
.map(|u| strids.insert(YMDLabel::Year, u.clone()));
|
|
|
|
self.mstridx
|
|
|
|
.map(|u| strids.insert(YMDLabel::Month, u.clone()));
|
|
|
|
self.dstridx
|
|
|
|
.map(|u| strids.insert(YMDLabel::Day, u.clone()));
|
|
|
|
|
|
|
|
// TODO: More Rustiomatic way of doing this?
|
|
|
|
if let Ok(ymd) = self.resolve_from_stridxs(&mut strids) { return Ok(ymd) };
|
|
|
|
|
|
|
|
// TODO: More Rustiomatic? Too many blocks for my liking
|
|
|
|
// Also having the array unpacking syntax is nice
|
|
|
|
if len_ymd > 3 {
|
|
|
|
return Err(ParseInternalError::ValueError(
|
|
|
|
"More than three YMD values".to_owned(),
|
|
|
|
));
|
|
|
|
} else if len_ymd == 1 || (self.mstridx.is_some() && len_ymd == 2) {
|
|
|
|
if self.mstridx.is_some() {
|
|
|
|
month = Some(self._ymd[self.mstridx.unwrap()]);
|
|
|
|
other = Some(self._ymd[self.mstridx.unwrap() - 1]);
|
|
|
|
} else {
|
|
|
|
other = Some(self._ymd[0]);
|
|
|
|
}
|
|
|
|
|
|
|
|
if len_ymd > 1 || self.mstridx.is_some() {
|
|
|
|
if other.unwrap_or(0) > 31 {
|
|
|
|
year = other;
|
|
|
|
} else {
|
|
|
|
day = other;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
} else if len_ymd == 2 {
|
|
|
|
if self._ymd[0] > 31 {
|
|
|
|
year = Some(self._ymd[0]);
|
|
|
|
month = Some(self._ymd[1]);
|
|
|
|
} else if self._ymd[1] > 31 {
|
|
|
|
month = Some(self._ymd[0]);
|
|
|
|
year = Some(self._ymd[1]);
|
|
|
|
} else if dayfirst && self._ymd[1] <= 12 {
|
|
|
|
day = Some(self._ymd[0]);
|
|
|
|
month = Some(self._ymd[1]);
|
|
|
|
} else {
|
|
|
|
month = Some(self._ymd[0]);
|
|
|
|
day = Some(self._ymd[1]);
|
|
|
|
}
|
|
|
|
} else if len_ymd == 3 {
|
|
|
|
// UNWRAP: 3 elements guarantees all indices are Some
|
|
|
|
if self.mstridx.unwrap() == 0 {
|
|
|
|
if self._ymd[1] > 31 {
|
|
|
|
month = Some(self._ymd[0]);
|
|
|
|
year = Some(self._ymd[1]);
|
|
|
|
day = Some(self._ymd[2]);
|
|
|
|
} else {
|
|
|
|
month = Some(self._ymd[0]);
|
|
|
|
day = Some(self._ymd[1]);
|
|
|
|
year = Some(self._ymd[2]);
|
|
|
|
}
|
|
|
|
} else if self.mstridx.unwrap() == 1 {
|
|
|
|
if self._ymd[0] > 31 || (yearfirst && self._ymd[2] <= 31) {
|
|
|
|
year = Some(self._ymd[0]);
|
|
|
|
month = Some(self._ymd[1]);
|
|
|
|
day = Some(self._ymd[2]);
|
|
|
|
} else {
|
|
|
|
day = Some(self._ymd[0]);
|
|
|
|
month = Some(self._ymd[1]);
|
|
|
|
year = Some(self._ymd[2]);
|
|
|
|
}
|
|
|
|
} else if self.mstridx.unwrap() == 2 {
|
|
|
|
// It was in the original docs, so: WTF!?
|
|
|
|
if self._ymd[1] > 31 {
|
|
|
|
day = Some(self._ymd[0]);
|
|
|
|
year = Some(self._ymd[1]);
|
|
|
|
month = Some(self._ymd[2]);
|
|
|
|
} else {
|
|
|
|
year = Some(self._ymd[0]);
|
|
|
|
day = Some(self._ymd[1]);
|
|
|
|
month = Some(self._ymd[2]);
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
if self._ymd[0] > 31 || self.ystridx.unwrap() == 0 ||
|
|
|
|
(yearfirst && self._ymd[1] <= 12 && self._ymd[2] <= 31) {
|
|
|
|
if dayfirst && self._ymd[2] <= 12 {
|
|
|
|
year = Some(self._ymd[0]);
|
|
|
|
day = Some(self._ymd[1]);
|
|
|
|
month = Some(self._ymd[2]);
|
|
|
|
} else {
|
|
|
|
year = Some(self._ymd[0]);
|
|
|
|
month = Some(self._ymd[1]);
|
|
|
|
day = Some(self._ymd[2]);
|
|
|
|
}
|
|
|
|
} else if self._ymd[0] > 12 ||
|
|
|
|
(dayfirst && self._ymd[1] <= 12) {
|
|
|
|
day = Some(self._ymd[0]);
|
|
|
|
month = Some(self._ymd[1]);
|
|
|
|
year = Some(self._ymd[2]);
|
|
|
|
} else {
|
|
|
|
month = Some(self._ymd[0]);
|
|
|
|
day = Some(self._ymd[1]);
|
|
|
|
year = Some(self._ymd[2]);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2018-05-18 23:58:28 -04:00
|
|
|
// TODO: Remove the error handling here
|
|
|
|
// We should be able to justify the UNWRAP, but I haven't
|
|
|
|
// convinced myself of that quite yet.
|
2018-05-18 23:50:53 -04:00
|
|
|
if !year.and(month).and(day).is_some() {
|
|
|
|
Err(ParseInternalError::YMDValueUnset)
|
|
|
|
} else {
|
|
|
|
Ok((year.unwrap(), month.unwrap(), day.unwrap()))
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|