qsv_dateparser/

datetime.rs

#![allow(deprecated)]
use crate::timezone;
use anyhow::{Result, anyhow};
use chrono::format::{Item, ParseResult, Parsed, parse as parse_items};
use chrono::prelude::*;
use regex::Regex;

macro_rules! regex {
    ($re:literal $(,)?) => {{
        static RE: std::sync::OnceLock<regex::Regex> = std::sync::OnceLock::new();
        RE.get_or_init(|| {
            regex::RegexBuilder::new($re)
                .unicode(false)
                .build()
                .expect("invalid regex literal")
        })
    }};
}

/// Compiles a strftime format literal into `chrono` format items exactly once,
/// mirroring [`regex!`].
///
/// `chrono`'s `parse_from_str` / `datetime_from_str` convenience methods walk
/// the format string through `StrftimeItems` on *every* call. Since every
/// format in this file is a literal, that walk is pure repeated work, and it is
/// paid several times per input because the parsers try formats in an
/// `or_else` chain. Hoisting it into a `OnceLock` leaves the parse itself
/// untouched: each method below reduces to exactly the same
/// `parse(&mut Parsed, input, items)` plus `Parsed::to_*` that the chrono
/// convenience method performs internally.
///
/// The format is a literal, so the resulting items borrow `'static` and the
/// `expect` is checked once at first use rather than per call.
macro_rules! fmt_items {
    ($fmt:literal $(,)?) => {{
        static ITEMS: std::sync::OnceLock<Vec<chrono::format::Item<'static>>> =
            std::sync::OnceLock::new();
        ITEMS
            .get_or_init(|| {
                chrono::format::StrftimeItems::new($fmt)
                    .parse()
                    .expect("invalid strftime literal")
            })
            .as_slice()
    }};
}
/// Lookup table of bytes that may legally appear in an accepted date format:
/// ASCII alphanumerics, ASCII whitespace (`\s` under `unicode(false)` =
/// space, `\t`, `\n`, `\x0B`, `\x0C`, `\r`), and the separators `- + / : . ,`.
const fn build_date_byte_table() -> [bool; 256] {
    let mut table = [false; 256];
    let mut i = 0usize;
    while i < 256 {
        let b = i as u8;
        table[i] = b.is_ascii_alphanumeric()
            || matches!(b, b' ' | 0x09..=0x0D)
            || matches!(b, b'-' | b'+' | b'/' | b':' | b'.' | b',');
        i += 1;
    }
    table
}

static DATE_BYTE: [bool; 256] = build_date_byte_table();

/// Bytes matched by `\s` in these patterns, which are all built with
/// `unicode(false)`: space plus `\t \n \v \f \r`.
///
/// `u8::is_ascii_whitespace` is not a substitute — it omits `\v` (0x0B), so
/// using it here would silently disagree with the regexes.
const fn build_ws_table() -> [bool; 256] {
    let mut table = [false; 256];
    let mut i = 0usize;
    while i < 256 {
        table[i] = matches!(i as u8, b' ' | 0x09..=0x0D);
        i += 1;
    }
    table
}

static DATE_WS: [bool; 256] = build_ws_table();

/// Cheap structural pre-filter run before the regex dispatch chain.
///
/// Any byte outside [`DATE_BYTE`] (e.g. `_`, `#`, `(`, or any non-ASCII byte)
/// means the input cannot be a date, so we can bail before running 5-6 failing
/// regex probes. This is the common, hot case for non-date string columns. It is
/// intentionally conservative: it rejects nothing that currently parses.
/// The table collapses the per-byte test to a single load + branch.
#[inline]
fn cannot_be_date(input: &str) -> bool {
    input.bytes().any(|b| !DATE_BYTE[b as usize])
}

/// Extracts the trailing timezone token without asking the regex engine for
/// capture positions.
///
/// The `_z` parsers used a `(?P<tz>…)` group purely to locate this token, but
/// capture tracking forces the regex crate onto a much slower engine: on
/// `month_mdy_hms_z`'s pattern, `captures()` costs 188 ns against 18 ns for
/// `is_match()` or `find()`, and a reused `capture_locations` buffer only gets
/// it to 179 ns — so it is the engine, not the allocation.
///
/// Instead the caller runs a prefix regex (the same pattern minus the tz
/// group) with `find()`, and passes the match end here. `prefix_end` is where
/// the greedy prefix stopped, so the rest of the input is the timezone.
///
/// Returns `None` whenever the remainder is not unambiguously a timezone
/// token, which is the caller's signal to fall back to the original
/// capture-based path. Because the fallback still runs the full pattern, this
/// cannot change any result: it either produces exactly what the capture group
/// would have, or declines and lets the slow path decide.
///
/// `require_ws` mirrors the difference between the two tz groups: `\s*` in
/// `ymd_hms_z` versus `\s+` in `month_mdy_hms_z`. It matters because the
/// prefix's own trailing `\s*` may already have eaten the separator, and the
/// caller trims the token anyway, so the whitespace has to be accounted for
/// here rather than inferred from the remainder alone.
#[inline]
fn tz_suffix(input: &str, prefix_end: usize, require_ws: bool) -> Option<&str> {
    let bytes = input.as_bytes();
    let rest = input.get(prefix_end..)?;

    if require_ws {
        let leading_ws = rest
            .as_bytes()
            .first()
            .is_some_and(|&b| DATE_WS[b as usize]);
        let trailing_ws = prefix_end
            .checked_sub(1)
            .is_some_and(|i| DATE_WS[bytes[i] as usize]);
        if !leading_ws && !trailing_ws {
            return None;
        }
    }

    // `\s*` then 3-6 of `[+-:a-zA-Z0-9]`, and nothing else.
    let token = rest.trim_start_matches(|c: char| (c as u32) < 128 && DATE_WS[c as usize]);
    if !(3..=6).contains(&token.len()) {
        return None;
    }
    token
        .bytes()
        .all(|b| b.is_ascii_alphanumeric() || matches!(b, b'+' | b'-' | b':'))
        .then_some(token)
}

/// Scratch space for the month-name parsers, which must strip separators from
/// the input before handing it to chrono.
///
/// Comfortably covers every shape those regexes accept in practice — the
/// longest realistic input, `September 17, 2012 at 10:09am PST`, is 33 bytes.
/// The patterns allow unbounded runs of whitespace, so a pathological input can
/// still exceed this; that falls back to allocating, which makes the size a
/// performance choice rather than a correctness one.
const NORMALIZE_SCRATCH: usize = 64;

/// Copies `input` into `buf`, dropping every byte in `strip`, and then removes
/// the first `"at"` when `drop_at` is set.
///
/// This replaces `String::replace`, whose allocation cost 56 ns of
/// `month_mdy_hms_z` — comparable to the entire format parse. Returns `None`
/// when the input does not fit, leaving the caller to allocate as before.
///
/// The `"at"` removal deliberately mirrors `String::find`, which matches the
/// first occurrence anywhere rather than only a standalone word. No month name
/// or abbreviation contains a lowercase `at`, so the two agree on every input
/// the callers' regexes admit, but keeping the behaviour identical avoids
/// having to prove that separately.
#[inline]
fn normalize_into<'b>(
    input: &str,
    buf: &'b mut [u8; NORMALIZE_SCRATCH],
    strip: &[u8],
    drop_at: bool,
) -> Option<&'b str> {
    if input.len() > buf.len() {
        return None;
    }

    let mut n = 0;
    for &b in input.as_bytes() {
        if !strip.contains(&b) {
            buf[n] = b;
            n += 1;
        }
    }

    if drop_at && let Some(pos) = buf[..n].windows(2).position(|w| w == b"at") {
        buf.copy_within(pos + 2..n, pos);
        n -= 2;
    }

    // Only whole ASCII bytes are ever dropped, so this cannot split a
    // multi-byte character; the check is a cheap way to say so without unsafe.
    std::str::from_utf8(&buf[..n]).ok()
}

/// Which format a date-time input's time-of-day portion dispatches to.
///
/// These are parse-dispatch buckets, not an exhaustive account of what the
/// family regexes admit. `\d{1,2}:\d{2}(?::\d{2})?(?:\.\d{1,9})?\s*(?:am|pm|AM|PM)?`
/// also admits a fraction with no seconds field — `hh:mm.fff`, with or without
/// an AM/PM marker — which buckets as [`Self::Hm`] or [`Self::ImP`] and then
/// fails against that format. It is malformed rather than a real shape, and
/// `unsupported_shapes_still_fail` pins it as unsupported.
///
/// Each bucket is matched by at most one format string per family, so
/// classifying up front replaces a chain of up to five trial parses — all but
/// the last guaranteed to fail — with a single attempt.
#[derive(Clone, Copy, PartialEq, Eq)]
enum TimeShape {
    /// `hh:mm`
    Hm,
    /// `hh:mm:ss`
    Hms,
    /// `hh:mm:ss.fff`
    HmsF,
    /// `hh:mm AM/PM`
    ImP,
    /// `hh:mm:ss AM/PM`
    ImsP,
    /// `hh:mm:ss.fff AM/PM`.
    ///
    /// Families whose regex admits both a fraction and an AM/PM marker parse
    /// this with the same `%I:%M:%S%.f %P` they use for [`Self::ImsP`], since
    /// `%.f` consumes nothing when there is no period. The two month-name
    /// families reject it at their regex gate and map it to `None`.
    HmsFP,
}

/// Classifies the time-of-day portion of an input that has already passed a
/// family regex, from a single byte scan.
///
/// Callers pass the exact string they are about to hand to the parser, not the
/// raw input. `month_mdy_hms` strips both `,` and `.` beforehand, so
/// classifying the raw input would mistake the period in `Sept. 17, 2012` for
/// fractional seconds. `month_dmy_hms` strips only `,` — it must keep a
/// fractional-seconds period for [`TimeShape::HmsF`] to be reachable.
///
/// A colon count of 2 distinguishes `%H:%M:%S` from `%H:%M`; the callers'
/// regexes cap the time at two colons and admit none anywhere else. The AM/PM
/// marker can only ever be the final two bytes, and chrono's `%P` is
/// case-insensitive, so a single `| 32` comparison covers `am`/`AM`/`Am`.
#[inline]
fn time_shape(input: &str) -> TimeShape {
    let bytes = input.as_bytes();

    let ampm = bytes.len() >= 2 && {
        let [.., ap, m] = bytes else { unreachable!() };
        m | 32 == b'm' && matches!(ap | 32, b'a' | b'p')
    };

    let mut colons = 0_u8;
    let mut fraction = false;
    for &b in bytes {
        match b {
            b':' => colons += 1,
            b'.' => fraction = true,
            _ => {}
        }
    }

    match (colons >= 2, fraction, ampm) {
        (true, true, true) => TimeShape::HmsFP,
        (true, true, false) => TimeShape::HmsF,
        (true, false, true) => TimeShape::ImsP,
        (true, false, false) => TimeShape::Hms,
        (false, _, true) => TimeShape::ImP,
        (false, _, false) => TimeShape::Hm,
    }
}

/// Returns true when the year field of a slash-separated date (the digits
/// after the second `/`) is exactly 2 digits wide. Callers' regexes guarantee
/// the `d{1,2}/d{1,2}/d{2,4}` shape, but the scan is panic-free regardless.
///
/// Used to dispatch between the `%y` and `%Y` chrono format families: `%y`
/// consumes at most 2 digits, so it always fails on 3-4 digit years, and `%Y`
/// is never reached on a 2-digit year that `%y` accepts (both families apply
/// identical date-validity rules), so picking one family by year width is
/// result-preserving and halves the trial-parse chain.
#[inline]
fn slash_year_is_two_digits(bytes: &[u8]) -> bool {
    let mut slashes = 0u8;
    let mut year_len = 0usize;
    for &b in bytes {
        if b == b'/' {
            slashes += 1;
        } else if slashes == 2 {
            if b.is_ascii_digit() {
                year_len += 1;
            } else {
                break;
            }
        }
    }
    year_len == 2
}

/// Parse struct has methods implemented parsers for accepted formats.
pub struct Parse<'z, Tz2> {
    tz: &'z Tz2,
    default_time: NaiveTime,
    prefer_dmy: bool,
}

impl<'z, Tz2> Parse<'z, Tz2>
where
    Tz2: TimeZone,
{
    /// Create a new instance of [`Parse`] with a custom parsing timezone that handles the
    /// datetime string without time offset.
    pub const fn new(tz: &'z Tz2, default_time: NaiveTime) -> Self {
        Self {
            tz,
            default_time,
            prefer_dmy: false,
        }
    }

    pub const fn prefer_dmy(&mut self, yes: bool) -> &Self {
        self.prefer_dmy = yes;
        self
    }

    /// Create a new instance of [`Parse`] with a custom parsing timezone that handles the
    /// datetime string without time offset, and the date parsing preference.
    pub const fn new_with_preference(
        tz: &'z Tz2,
        default_time: NaiveTime,
        prefer_dmy: bool,
    ) -> Self {
        Self {
            tz,
            default_time,
            prefer_dmy,
        }
    }

    /// Drop-in replacement for `Tz::datetime_from_str` taking pre-compiled items.
    ///
    /// Note that `Parsed` **must** be constructed fresh for every attempt.
    /// `Parsed::set_*` returns `Err` when a field is set twice to conflicting
    /// values, so reusing one `Parsed` across the `or_else` chains below would
    /// produce silently wrong results rather than a compile error.
    #[inline]
    fn dt_from_items(&self, input: &str, items: &[Item<'static>]) -> ParseResult<DateTime<Tz2>> {
        let mut parsed = Parsed::new();
        parse_items(&mut parsed, input, items.iter())?;
        parsed.to_datetime_with_timezone(self.tz)
    }

    /// Drop-in replacement for `NaiveDateTime::parse_from_str` taking
    /// pre-compiled items. See [`Self::dt_from_items`] on `Parsed` reuse.
    #[inline]
    fn naive_dt_from_items(input: &str, items: &[Item<'static>]) -> ParseResult<NaiveDateTime> {
        let mut parsed = Parsed::new();
        parse_items(&mut parsed, input, items.iter())?;
        parsed.to_naive_datetime_with_offset(0)
    }

    /// Drop-in replacement for `NaiveDate::parse_from_str` taking pre-compiled
    /// items. See [`Self::dt_from_items`] on `Parsed` reuse.
    #[inline]
    fn naive_date_from_items(input: &str, items: &[Item<'static>]) -> ParseResult<NaiveDate> {
        let mut parsed = Parsed::new();
        parse_items(&mut parsed, input, items.iter())?;
        parsed.to_naive_date()
    }

    /// This method tries to parse the input datetime string with a list of accepted formats. See
    /// more examples from [`Parse`], [`crate::parse()`] and [`crate::parse_with_timezone()`].
    ///
    /// Order rationale: the regex-gated families are tried first because their
    /// `is_match` gate rejects non-matching inputs cheaply. The two parsers
    /// without a family regex gate — `unix_timestamp` (runs `fast_float2`) and
    /// `rfc2822` (runs `parse_from_rfc2822`) — are tried last to avoid paying
    /// their cost on the common ISO/slash dates. Each still applies its own cheap
    /// byte pre-filter before the heavy parse (`unix_timestamp` checks the first
    /// byte against the leads `fast_float2` accepts; `rfc2822` requires a `:`).
    ///
    /// This reorder is result-preserving:
    /// - A `fast_float2`-parseable input (a pure finite number; `inf`/`nan` are
    ///   rejected as non-finite) matches no family gate (they all require `/`,
    ///   an interior `-`, or a letters+space shape a bare number lacks), so it
    ///   still reaches `unix_timestamp`.
    /// - An `rfc2822` input always carries a timezone, which makes the
    ///   `$`-anchored `month_dmy_*` regexes fail; conversely `month_dmy_*` only
    ///   succeeds without a timezone, which makes `rfc2822` fail. The two are
    ///   mutually exclusive, so deferring `rfc2822` cannot change any result.
    ///
    /// Within that order, the first byte selects which families can match at
    /// all. Every family gate is `^`-anchored on either a digit or a letter, so
    /// a letter-leading input cannot match any of the numeric families and vice
    /// versa. Running them anyway costs roughly 5-10 ns each in regex startup
    /// even when the first byte rejects immediately, which is most of the cost
    /// of a non-date word — the dominant input in a `qsv stats --infer-dates`
    /// run over a text column.
    ///
    /// `rfc2822` stays in the digit branch as well as the letter branch: the
    /// day-of-week is optional in RFC 2822, so `02 Jun 2021 06:31:39 GMT`
    /// parses and leads with a digit.
    #[inline]
    pub fn parse(&self, input: &str) -> Result<DateTime<Utc>> {
        if cannot_be_date(input) {
            return Err(anyhow!("{} did not match any formats.", input));
        }
        let Some(&first) = input.as_bytes().first() else {
            return Err(anyhow!("{} did not match any formats.", input));
        };

        let parsed = if first.is_ascii_digit() {
            self.slash_mdy_family(input)
                .or_else(|| self.slash_ymd_family(input))
                .or_else(|| self.ymd_family(input))
                .or_else(|| self.month_ymd(input))
                .or_else(|| self.month_dmy_family(input))
                .or_else(|| self.unix_timestamp(input))
                .or_else(|| self.rfc2822(input))
        } else if first.is_ascii_alphabetic() {
            // `month_mdy_family` is the only letter-anchored gate.
            // `unix_timestamp` is excluded by its own lead-byte pre-filter.
            self.month_mdy_family(input).or_else(|| self.rfc2822(input))
        } else {
            // `+`, `-`, `.` and the separators that survive `cannot_be_date`.
            // No family gate can match, but a signed or bare-decimal timestamp
            // can; `rfc2822` is kept for its own leading-whitespace handling.
            self.unix_timestamp(input).or_else(|| self.rfc2822(input))
        };

        parsed.unwrap_or_else(|| Err(anyhow!("{} did not match any formats.", input)))
    }

    #[inline]
    fn ymd_family(&self, input: &str) -> Option<Result<DateTime<Utc>>> {
        let re: &Regex = regex! {
            r"^\d{4}-\d{2}"

        };

        if !re.is_match(input) {
            return None;
        }
        self.rfc3339(input)
            .or_else(|| self.ymd_hms(input))
            .or_else(|| self.ymd_hms_z(input))
            .or_else(|| self.ymd(input))
            .or_else(|| self.ymd_z(input))
    }

    #[inline]
    fn month_mdy_family(&self, input: &str) -> Option<Result<DateTime<Utc>>> {
        let re: &Regex = regex! {
            r"^[a-zA-Z]{3,9}\.?\s+\d{1,2}"
        };

        if !re.is_match(input) {
            return None;
        }
        self.month_mdy_hms(input)
            .or_else(|| self.month_mdy_hms_z(input))
            .or_else(|| self.month_mdy(input))
    }

    #[inline]
    fn month_dmy_family(&self, input: &str) -> Option<Result<DateTime<Utc>>> {
        let re: &Regex = regex! {r"^\d{1,2}\s+[a-zA-Z]{3,9}"
        };

        if !re.is_match(input) {
            return None;
        }
        self.month_dmy_hms(input).or_else(|| self.month_dmy(input))
    }

    #[inline]
    fn slash_mdy_family(&self, input: &str) -> Option<Result<DateTime<Utc>>> {
        let re: &Regex = regex! {r"^\d{1,2}/\d{1,2}"
        };
        if !re.is_match(input) {
            return None;
        }
        if self.prefer_dmy {
            self.slash_dmy_hms(input)
                .or_else(|| self.slash_dmy(input))
                .or_else(|| self.slash_mdy_hms(input))
                .or_else(|| self.slash_mdy(input))
        } else {
            self.slash_mdy_hms(input)
                .or_else(|| self.slash_mdy(input))
                .or_else(|| self.slash_dmy_hms(input))
                .or_else(|| self.slash_dmy(input))
        }
    }

    #[inline]
    fn slash_ymd_family(&self, input: &str) -> Option<Result<DateTime<Utc>>> {
        let re: &Regex = regex! {r"^[0-9]{4}/[0-9]{1,2}"};
        if !re.is_match(input) {
            return None;
        }
        self.slash_ymd_hms(input).or_else(|| self.slash_ymd(input))
    }

    // unix timestamp
    // - 0
    // - -770172300
    // - 1671673426.123456789
    #[inline]
    fn unix_timestamp(&self, input: &str) -> Option<Result<DateTime<Utc>>> {
        // Cheap pre-filter before the heavier float parse: only finite numbers
        // are accepted, so the first byte must be a digit, sign, or dot
        // (`fast_float2` rejects leading whitespace and empty input; bare
        // `inf`/`nan` are excluded here, matching the non-finite check below).
        // This is the last-resort numeric parser, so most inputs reaching it
        // are non-numeric.
        let &b0 = input.as_bytes().first()?;
        if !(b0.is_ascii_digit() || matches!(b0, b'+' | b'-' | b'.')) {
            return None;
        }

        let ts_sec_val: f64 = if let Ok(val) = fast_float2::parse(input) {
            val
        } else {
            return None;
        };

        // Reject non-finite values (`+inf`, `-nan`, … pass the lead-byte filter
        // above): the `as i64` cast below would otherwise turn `nan` into 0
        // (1970-01-01) and `inf` into i64::MAX nanos (2262-04-11) — never the
        // intended reading of the input.
        if !ts_sec_val.is_finite() {
            return None;
        }

        // convert the timestamp seconds value to nanoseconds
        let ts_ns_val = ts_sec_val * 1_000_000_000_f64;

        let result = Utc.timestamp_nanos(ts_ns_val as i64).with_timezone(&Utc);
        Some(Ok(result))
    }

    // rfc3339
    // - 2021-05-01T01:17:02.604456Z
    // - 2017-11-25T22:34:50Z
    #[inline]
    fn rfc3339(&self, input: &str) -> Option<Result<DateTime<Utc>>> {
        DateTime::parse_from_rfc3339(input)
            .ok()
            .map(|parsed| parsed.with_timezone(&Utc))
            .map(Ok)
    }

    // rfc2822
    // - Wed, 02 Jun 2021 06:31:39 GMT
    #[inline]
    fn rfc2822(&self, input: &str) -> Option<Result<DateTime<Utc>>> {
        // Fast pre-filter: every RFC2822 datetime carries a time-of-day
        // (`hour ":" minute`), so it always contains ':'. Skip the
        // `parse_from_rfc2822` attempt for colon-free inputs. This is the
        // last-resort parser, so most inputs reaching it are non-rfc2822.
        if !input.as_bytes().contains(&b':') {
            return None;
        }
        DateTime::parse_from_rfc2822(input)
            .ok()
            .map(|parsed| parsed.with_timezone(&Utc))
            .map(Ok)
    }

    // yyyy-mm-dd hh:mm:ss  (separator is space OR ISO 8601 'T')
    // - 2014-04-26 05:24:37 PM
    // - 2021-04-30 21:14
    // - 2021-04-30 21:14:10
    // - 2021-04-30 21:14:10.052282
    // - 2014-04-26 17:24:37.123
    // - 2014-04-26 17:24:37.3186369
    // - 2012-08-03 18:31:59.257000000
    // - 2020-01-15T08:00
    // - 2020-01-15T08:00:00
    // - 2020-01-15T08:00:00.123456
    // - 2012-03-19 10:11:59.318 PM
    #[inline]
    fn ymd_hms(&self, input: &str) -> Option<Result<DateTime<Utc>>> {
        let re: &Regex = regex! {
                r"^\d{4}-\d{2}-\d{2}[T\s]+\d{2}:\d{2}(?::\d{2})?(?:\.\d{1,9})?\s*(?:am|pm|AM|PM)?$"

        };
        if !re.is_match(input) {
            return None;
        }

        // Byte 10 is the date/time separator. The regex guarantees the input
        // has at least 16 bytes and that byte 10 is either 'T' or ASCII
        // whitespace, so the single byte picks the format family.
        let items = match (input.as_bytes()[10] == b'T', time_shape(input)) {
            (true, TimeShape::Hms) => fmt_items!("%Y-%m-%dT%H:%M:%S"),
            (true, TimeShape::Hm) => fmt_items!("%Y-%m-%dT%H:%M"),
            (true, TimeShape::HmsF) => fmt_items!("%Y-%m-%dT%H:%M:%S%.f"),
            (true, TimeShape::ImsP | TimeShape::HmsFP) => {
                fmt_items!("%Y-%m-%dT%I:%M:%S%.f %P")
            }
            (true, TimeShape::ImP) => fmt_items!("%Y-%m-%dT%I:%M %P"),
            (false, TimeShape::Hms) => fmt_items!("%Y-%m-%d %H:%M:%S"),
            (false, TimeShape::Hm) => fmt_items!("%Y-%m-%d %H:%M"),
            (false, TimeShape::HmsF) => fmt_items!("%Y-%m-%d %H:%M:%S%.f"),
            (false, TimeShape::ImsP | TimeShape::HmsFP) => {
                fmt_items!("%Y-%m-%d %I:%M:%S%.f %P")
            }
            (false, TimeShape::ImP) => fmt_items!("%Y-%m-%d %I:%M %P"),
        };

        self.dt_from_items(input, items)
            .ok()
            .map(|parsed| parsed.with_timezone(&Utc))
            .map(Ok)
    }

    // yyyy-mm-dd hh:mm:ss z
    // - 2017-11-25 13:31:15 PST
    // - 2017-11-25 13:31 PST
    // - 2014-12-16 06:20:00 UTC
    // - 2014-12-16 06:20:00 GMT
    // - 2014-04-26 13:13:43 +0800
    // - 2014-04-26 13:13:44 +09:00
    // - 2012-08-03 18:31:59.257000000 +0000
    // - 2015-09-30 18:48:56.35272715 UTC
    #[inline]
    fn ymd_hms_z(&self, input: &str) -> Option<Result<DateTime<Utc>>> {
        // Fast pre-filter: bare dates "YYYY-MM-DD" are 10 chars; valid inputs need space + time
        if input.len() < 17 || !input.as_bytes()[10].is_ascii_whitespace() {
            return None;
        }
        // Fast path: locate the timezone with `find()` on the pattern minus the
        // tz group, which keeps the regex engine off the capture-tracking path.
        // See `tz_suffix` — on failure this falls through to the original
        // capture-based match, so no input can change meaning.
        let prefix: &Regex = regex! {r"^\d{4}-\d{2}-\d{2}\s+\d{2}:\d{2}(?::\d{2})?(?:\.\d{1,9})?"};
        let fast_tz = prefix
            .find(input)
            .and_then(|m| tz_suffix(input, m.end(), false));

        let re: &Regex = regex! {
                r"^\d{4}-\d{2}-\d{2}\s+\d{2}:\d{2}(?::\d{2})?(?:\.\d{1,9})?(?P<tz>\s*[+-:a-zA-Z0-9]{3,6})$"
        };

        if let Some(tz) = fast_tz.or_else(|| {
            re.captures(input)
                .and_then(|caps| caps.name("tz").map(|m| m.as_str().trim()))
        }) {
            let parse_from_str = Self::naive_dt_from_items;
            return match timezone::parse(tz) {
                Ok(offset) => parse_from_str(input, fmt_items!("%Y-%m-%d %H:%M:%S %Z"))
                    .or_else(|_| parse_from_str(input, fmt_items!("%Y-%m-%d %H:%M %Z")))
                    .or_else(|_| parse_from_str(input, fmt_items!("%Y-%m-%d %H:%M:%S%.f %Z")))
                    .ok()
                    .and_then(|parsed| offset.from_local_datetime(&parsed).single())
                    .map(|datetime| datetime.with_timezone(&Utc))
                    .map(Ok),
                Err(err) => Some(Err(err)),
            };
        }
        None
    }

    // yyyy-mm-dd
    // - 2021-02-21
    #[inline]
    fn ymd(&self, input: &str) -> Option<Result<DateTime<Utc>>> {
        let re: &Regex = regex! {r"^\d{4}-\d{2}-\d{2}$"
        };

        if !re.is_match(input) {
            return None;
        }
        let now = Utc::now()
            .date()
            .and_time(self.default_time)?
            .with_timezone(self.tz);
        Self::naive_date_from_items(input, fmt_items!("%Y-%m-%d"))
            .ok()
            .map(|parsed| parsed.and_time(now.time()))
            .and_then(|datetime| self.tz.from_local_datetime(&datetime).single())
            .map(|at_tz| at_tz.with_timezone(&Utc))
            .map(Ok)
    }

    // yyyy-mm-dd z
    // - 2021-02-21 PST
    // - 2021-02-21 UTC
    // - 2020-07-20+08:00 (yyyy-mm-dd-07:00)
    #[inline]
    fn ymd_z(&self, input: &str) -> Option<Result<DateTime<Utc>>> {
        // Fast pre-filter: bare date "YYYY-MM-DD" is exactly 10 chars; timezone appended = longer
        if input.len() <= 10 {
            return None;
        }
        let re: &Regex = regex! {r"^\d{4}-\d{2}-\d{2}(?P<tz>\s*[+-:a-zA-Z0-9]{3,6})$"
        };
        if let Some(caps) = re.captures(input)
            && let Some(matched_tz) = caps.name("tz")
        {
            return match timezone::parse(matched_tz.as_str().trim()) {
                Ok(offset) => {
                    let now = Utc::now()
                        .date()
                        .and_time(self.default_time)?
                        .with_timezone(&offset);
                    Self::naive_date_from_items(input, fmt_items!("%Y-%m-%d %Z"))
                        .ok()
                        .map(|parsed| parsed.and_time(now.time()))
                        .and_then(|datetime| offset.from_local_datetime(&datetime).single())
                        .map(|at_tz| at_tz.with_timezone(&Utc))
                        .map(Ok)
                }
                Err(err) => Some(Err(err)),
            };
        }
        None
    }

    // yyyy-mon-dd
    // - 2021-Feb-21
    #[inline]
    fn month_ymd(&self, input: &str) -> Option<Result<DateTime<Utc>>> {
        let re: &Regex = regex! {r"^\d{4}-\w{3,9}-\d{2}$"
        };
        if !re.is_match(input) {
            return None;
        }

        let now = Utc::now()
            .date()
            .and_time(self.default_time)?
            .with_timezone(self.tz);
        Self::naive_date_from_items(input, fmt_items!("%Y-%m-%d"))
            .or_else(|_| Self::naive_date_from_items(input, fmt_items!("%Y-%b-%d")))
            .ok()
            .map(|parsed| parsed.and_time(now.time()))
            .and_then(|datetime| self.tz.from_local_datetime(&datetime).single())
            .map(|at_tz| at_tz.with_timezone(&Utc))
            .map(Ok)
    }

    // Mon dd, yyyy, hh:mm:ss
    // - May 8, 2009 5:57:51 PM
    // - September 17, 2012 10:09am
    // - September 17, 2012, 10:10:09
    #[inline]
    fn month_mdy_hms(&self, input: &str) -> Option<Result<DateTime<Utc>>> {
        let re: &Regex = regex! {
                r"^[a-zA-Z]{3,9}\.?\s+\d{1,2},\s+\d{2,4},?\s+\d{1,2}:\d{2}(?::\d{2})?\s*(?:am|pm|AM|PM)?$"
        };
        if !re.is_match(input) {
            return None;
        }

        // The regex above enforces \s+ after any comma or period, so removing bare ',' or '.'
        // is equivalent to the previous `replace(", ", " ").replace(". ", " ")` for all
        // inputs that reach this point — marginally-malformed inputs (e.g. "May 27,2012 …")
        // still fail to parse after stripping because the digits run together.
        let mut buf = [0_u8; NORMALIZE_SCRATCH];
        let fallback;
        let dt = match normalize_into(input, &mut buf, b",.", false) {
            Some(s) => s,
            None => {
                fallback = input.replace([',', '.'], "");
                fallback.as_str()
            }
        };
        // Classify `dt`, not `input`: the regex admits a period after an
        // abbreviated month ("Sept. 17, 2012"), which would otherwise read as
        // fractional seconds. This family's regex has no fractional-seconds
        // group at all, and the strip removes any period regardless, so the
        // two fraction-bearing shapes cannot occur — and never had a format.
        let items = match time_shape(dt) {
            TimeShape::Hms => fmt_items!("%B %d %Y %H:%M:%S"),
            TimeShape::Hm => fmt_items!("%B %d %Y %H:%M"),
            TimeShape::ImsP => fmt_items!("%B %d %Y %I:%M:%S %P"),
            TimeShape::ImP => fmt_items!("%B %d %Y %I:%M %P"),
            TimeShape::HmsF | TimeShape::HmsFP => return None,
        };
        self.dt_from_items(dt, items)
            .ok()
            .map(|at_tz| at_tz.with_timezone(&Utc))
            .map(Ok)
    }

    // Mon dd, yyyy hh:mm:ss z
    // - May 02, 2021 15:51:31 UTC
    // - May 02, 2021 15:51 UTC
    // - May 26, 2021, 12:49 AM PDT
    // - September 17, 2012 at 10:09am PST
    #[inline]
    fn month_mdy_hms_z(&self, input: &str) -> Option<Result<DateTime<Utc>>> {
        // Fast pre-filter: must contain an isolated 4-digit year — eliminates "May 27 02:45:27".
        // Skip the O(n) scan entirely for inputs too short to hold a valid month+day+year+time+tz.
        if input.len() < 20 {
            return None;
        }
        let bytes = input.as_bytes();
        let has_year = (0..bytes.len().saturating_sub(3)).any(|i| {
            bytes[i..i + 4].iter().all(|b| b.is_ascii_digit())
                && (i == 0 || !bytes[i - 1].is_ascii_digit())
                && bytes.get(i + 4).is_none_or(|b| !b.is_ascii_digit())
        });
        if !has_year {
            return None;
        }
        // Fast path — see the twin comment in ymd_hms_z. This pattern's tz
        // group requires whitespace (`\s+`), but the prefix's own trailing
        // `\s*` may already have consumed it, so `tz_suffix` is told to accept
        // the separator on either side of the boundary.
        let prefix: &Regex = regex! {
                r"^[a-zA-Z]{3,9}\s+\d{1,2},?\s+\d{4}\s*,?(?:at)?\s+\d{2}:\d{2}(?::\d{2})?\s*(?:am|pm|AM|PM)?"
        };
        let fast_tz = prefix
            .find(input)
            .and_then(|m| tz_suffix(input, m.end(), true));

        let re: &Regex = regex! {
                r"^[a-zA-Z]{3,9}\s+\d{1,2},?\s+\d{4}\s*,?(?:at)?\s+\d{2}:\d{2}(?::\d{2})?\s*(?:am|pm|AM|PM)?(?P<tz>\s+[+-:a-zA-Z0-9]{3,6})$",
        };
        if let Some(tz) = fast_tz.or_else(|| {
            re.captures(input)
                .and_then(|caps| caps.name("tz").map(|m| m.as_str().trim()))
        }) {
            let parse_from_str = Self::naive_dt_from_items;
            return match timezone::parse(tz) {
                Ok(offset) => {
                    let mut buf = [0_u8; NORMALIZE_SCRATCH];
                    let fallback;
                    let dt: &str = match normalize_into(input, &mut buf, b",", true) {
                        Some(s) => s,
                        None => {
                            let mut owned = input.replace(',', "");
                            if let Some(pos) = owned.find("at") {
                                owned.replace_range(pos..pos + 2, "");
                            }
                            fallback = owned;
                            fallback.as_str()
                        }
                    };
                    parse_from_str(dt, fmt_items!("%B %d %Y %H:%M:%S %Z"))
                        .or_else(|_| parse_from_str(dt, fmt_items!("%B %d %Y %H:%M %Z")))
                        .or_else(|_| parse_from_str(dt, fmt_items!("%B %d %Y %I:%M:%S %P %Z")))
                        .or_else(|_| parse_from_str(dt, fmt_items!("%B %d %Y %I:%M %P %Z")))
                        .ok()
                        .and_then(|parsed| offset.from_local_datetime(&parsed).single())
                        .map(|datetime| datetime.with_timezone(&Utc))
                        .map(Ok)
                }
                Err(err) => Some(Err(err)),
            };
        }
        None
    }

    // Mon dd, yyyy
    // - May 25, 2021
    // - oct 7, 1970
    // - oct 7, 70
    // - oct. 7, 1970
    // - oct. 7, 70
    // - October 7, 1970
    #[inline]
    fn month_mdy(&self, input: &str) -> Option<Result<DateTime<Utc>>> {
        let re: &Regex = regex! {r"^[a-zA-Z]{3,9}\.?\s+\d{1,2},\s+\d{2,4}$"
        };
        if !re.is_match(input) {
            return None;
        }

        let now = Utc::now()
            .date()
            .and_time(self.default_time)?
            .with_timezone(self.tz);
        // The regex above enforces \s+ after any comma or period, so removing bare ',' or '.'
        // is equivalent to the previous `replace(", ", " ").replace(". ", " ")` for all
        // inputs that reach this point.
        let mut buf = [0_u8; NORMALIZE_SCRATCH];
        let fallback;
        let dt = match normalize_into(input, &mut buf, b",.", false) {
            Some(s) => s,
            None => {
                fallback = input.replace([',', '.'], "");
                fallback.as_str()
            }
        };
        Self::naive_date_from_items(dt, fmt_items!("%B %d %y"))
            .or_else(|_| Self::naive_date_from_items(dt, fmt_items!("%B %d %Y")))
            .ok()
            .map(|parsed| parsed.and_time(now.time()))
            .and_then(|datetime| self.tz.from_local_datetime(&datetime).single())
            .map(|at_tz| at_tz.with_timezone(&Utc))
            .map(Ok)
    }

    // dd Mon yyyy hh:mm:ss
    // - 12 Feb 2006, 19:17
    // - 12 Feb 2006 19:17
    // - 14 May 2019 19:11:40.164
    #[inline]
    fn month_dmy_hms(&self, input: &str) -> Option<Result<DateTime<Utc>>> {
        // Fast pre-filter: time component always contains ':', skip regex for date-only inputs.
        if !input.as_bytes().contains(&b':') {
            return None;
        }
        let re: &Regex = regex! {
                r"^\d{1,2}\s+[a-zA-Z]{3,9}\s+\d{2,4},?\s+\d{1,2}:[0-9]{2}(?::[0-9]{2})?(?:\.[0-9]{1,9})?$"
        };
        if !re.is_match(input) {
            return None;
        }

        let mut buf = [0_u8; NORMALIZE_SCRATCH];
        let fallback;
        let dt = match normalize_into(input, &mut buf, b",", false) {
            Some(s) => s,
            None => {
                fallback = input.replace(',', "");
                fallback.as_str()
            }
        };
        // This family's regex has no am/pm alternative, so the AM/PM shapes
        // cannot occur here. The chain previously ended in `%I:%M:%S %P` and
        // `%I:%M %P`, which were therefore unreachable; dropping them changes
        // no result.
        let items = match time_shape(dt) {
            TimeShape::Hms => fmt_items!("%d %B %Y %H:%M:%S"),
            TimeShape::Hm => fmt_items!("%d %B %Y %H:%M"),
            TimeShape::HmsF => fmt_items!("%d %B %Y %H:%M:%S%.f"),
            TimeShape::ImP | TimeShape::ImsP | TimeShape::HmsFP => return None,
        };
        self.dt_from_items(dt, items)
            .ok()
            .map(|at_tz| at_tz.with_timezone(&Utc))
            .map(Ok)
    }

    // dd Mon yyyy
    // - 7 oct 70
    // - 7 oct 1970
    // - 03 February 2013
    // - 1 July 2013
    #[inline]
    fn month_dmy(&self, input: &str) -> Option<Result<DateTime<Utc>>> {
        let re: &Regex = regex! {r"^\d{1,2}\s+[a-zA-Z]{3,9}\s+\d{2,4}$"
        };
        if !re.is_match(input) {
            return None;
        }

        let now = Utc::now()
            .date()
            .and_time(self.default_time)?
            .with_timezone(self.tz);
        // Fast path: if the last 4 bytes are all digits and preceded by a space, it's a
        // 4-digit year — skip the always-failing %d %B %y (2-digit year) attempt.
        let bytes = input.as_bytes();
        let len = bytes.len();
        let four_digit_year = len >= 5
            && bytes[len - 4..].iter().all(|b| b.is_ascii_digit())
            && bytes[len - 5].is_ascii_whitespace();
        let parsed = if four_digit_year {
            Self::naive_date_from_items(input, fmt_items!("%d %B %Y"))
        } else {
            Self::naive_date_from_items(input, fmt_items!("%d %B %y"))
                .or_else(|_| Self::naive_date_from_items(input, fmt_items!("%d %B %Y")))
        };
        parsed
            .ok()
            .map(|parsed| parsed.and_time(now.time()))
            .and_then(|datetime| self.tz.from_local_datetime(&datetime).single())
            .map(|at_tz| at_tz.with_timezone(&Utc))
            .map(Ok)
    }

    // mm/dd/yyyy hh:mm:ss
    // - 4/8/2014 22:05
    // - 04/08/2014 22:05
    // - 4/8/14 22:05
    // - 04/2/2014 03:00:51
    // - 8/8/1965 12:00:00 AM
    // - 8/8/1965 01:00:01 PM
    // - 8/8/1965 01:00 PM
    // - 8/8/1965 1:00 PM
    // - 8/8/1965 12:00 AM
    // - 4/02/2014 03:00:51
    // - 03/19/2012 10:11:59
    // - 03/19/2012 10:11:59.3186369
    // - 03/19/2012 10:11:59.318 PM
    #[inline]
    fn slash_mdy_hms(&self, input: &str) -> Option<Result<DateTime<Utc>>> {
        let re: &Regex = regex! {
                r"^\d{1,2}/\d{1,2}/\d{2,4}\s+\d{1,2}:\d{2}(?::\d{2})?(?:\.\d{1,9})?\s*(?:am|pm|AM|PM)?$"
        };
        if !re.is_match(input) {
            return None;
        }

        // Dispatch on year width (see slash_year_is_two_digits) and on the
        // time shape (see time_shape) instead of trying all 10 formats. This
        // picks the one format that can match, so `MM/DD/YYYY hh:mm:ss AM/PM`
        // no longer burns three guaranteed-failing attempts before the fourth.
        let items = match (
            slash_year_is_two_digits(input.as_bytes()),
            time_shape(input),
        ) {
            (true, TimeShape::Hms) => fmt_items!("%m/%d/%y %H:%M:%S"),
            (true, TimeShape::Hm) => fmt_items!("%m/%d/%y %H:%M"),
            (true, TimeShape::HmsF) => fmt_items!("%m/%d/%y %H:%M:%S%.f"),
            (true, TimeShape::ImsP | TimeShape::HmsFP) => {
                fmt_items!("%m/%d/%y %I:%M:%S%.f %P")
            }
            (true, TimeShape::ImP) => fmt_items!("%m/%d/%y %I:%M %P"),
            (false, TimeShape::Hms) => fmt_items!("%m/%d/%Y %H:%M:%S"),
            (false, TimeShape::Hm) => fmt_items!("%m/%d/%Y %H:%M"),
            (false, TimeShape::HmsF) => fmt_items!("%m/%d/%Y %H:%M:%S%.f"),
            (false, TimeShape::ImsP | TimeShape::HmsFP) => {
                fmt_items!("%m/%d/%Y %I:%M:%S%.f %P")
            }
            (false, TimeShape::ImP) => fmt_items!("%m/%d/%Y %I:%M %P"),
        };
        self.dt_from_items(input, items)
            .ok()
            .map(|at_tz| at_tz.with_timezone(&Utc))
            .map(Ok)
    }

    // dd/mm/yyyy hh:mm:ss
    // - 8/4/2014 22:05
    // - 08/04/2014 22:05
    // - 8/4/14 22:05
    // - 2/04/2014 03:00:51
    // - 8/8/1965 12:00:00 AM
    // - 8/8/1965 01:00:01 PM
    // - 8/8/1965 01:00 PM
    // - 8/8/1965 1:00 PM
    // - 8/8/1965 12:00 AM
    // - 02/4/2014 03:00:51
    // - 19/03/2012 10:11:59
    // - 19/03/2012 10:11:59.3186369
    // - 19/03/2012 10:11:59.318 PM
    #[inline]
    fn slash_dmy_hms(&self, input: &str) -> Option<Result<DateTime<Utc>>> {
        let re: &Regex = regex! {
                r"^\d{1,2}/\d{1,2}/\d{2,4}\s+\d{1,2}:\d{2}(?::\d{2})?(?:\.\d{1,9})?\s*(?:am|pm|AM|PM)?$"
        };
        if !re.is_match(input) {
            return None;
        }

        // Dispatch on year width and time shape — see the twin comment in
        // slash_mdy_hms.
        let items = match (
            slash_year_is_two_digits(input.as_bytes()),
            time_shape(input),
        ) {
            (true, TimeShape::Hms) => fmt_items!("%d/%m/%y %H:%M:%S"),
            (true, TimeShape::Hm) => fmt_items!("%d/%m/%y %H:%M"),
            (true, TimeShape::HmsF) => fmt_items!("%d/%m/%y %H:%M:%S%.f"),
            (true, TimeShape::ImsP | TimeShape::HmsFP) => {
                fmt_items!("%d/%m/%y %I:%M:%S%.f %P")
            }
            (true, TimeShape::ImP) => fmt_items!("%d/%m/%y %I:%M %P"),
            (false, TimeShape::Hms) => fmt_items!("%d/%m/%Y %H:%M:%S"),
            (false, TimeShape::Hm) => fmt_items!("%d/%m/%Y %H:%M"),
            (false, TimeShape::HmsF) => fmt_items!("%d/%m/%Y %H:%M:%S%.f"),
            (false, TimeShape::ImsP | TimeShape::HmsFP) => {
                fmt_items!("%d/%m/%Y %I:%M:%S%.f %P")
            }
            (false, TimeShape::ImP) => fmt_items!("%d/%m/%Y %I:%M %P"),
        };
        self.dt_from_items(input, items)
            .ok()
            .map(|at_tz| at_tz.with_timezone(&Utc))
            .map(Ok)
    }

    // mm/dd/yyyy
    // - 3/31/2014
    // - 03/31/2014
    // - 08/21/71
    // - 8/1/71
    #[inline]
    fn slash_mdy(&self, input: &str) -> Option<Result<DateTime<Utc>>> {
        let re: &Regex = regex! {r"^\d{1,2}/\d{1,2}/\d{2,4}$"
        };
        if !re.is_match(input) {
            return None;
        }

        let now = Utc::now()
            .date()
            .and_time(self.default_time)?
            .with_timezone(self.tz);
        let fmt = if slash_year_is_two_digits(input.as_bytes()) {
            fmt_items!("%m/%d/%y")
        } else {
            fmt_items!("%m/%d/%Y")
        };
        Self::naive_date_from_items(input, fmt)
            .ok()
            .map(|parsed| parsed.and_time(now.time()))
            .and_then(|datetime| self.tz.from_local_datetime(&datetime).single())
            .map(|at_tz| at_tz.with_timezone(&Utc))
            .map(Ok)
    }

    // dd/mm/yyyy
    // - 31/3/2014
    // - 31/03/2014
    // - 21/08/71
    // - 1/8/71
    #[inline]
    fn slash_dmy(&self, input: &str) -> Option<Result<DateTime<Utc>>> {
        let re: &Regex = regex! {r"^[0-9]{1,2}/[0-9]{1,2}/[0-9]{2,4}$"
        };
        if !re.is_match(input) {
            return None;
        }

        let now = Utc::now()
            .date()
            .and_time(self.default_time)?
            .with_timezone(self.tz);
        let fmt = if slash_year_is_two_digits(input.as_bytes()) {
            fmt_items!("%d/%m/%y")
        } else {
            fmt_items!("%d/%m/%Y")
        };
        Self::naive_date_from_items(input, fmt)
            .ok()
            .map(|parsed| parsed.and_time(now.time()))
            .and_then(|datetime| self.tz.from_local_datetime(&datetime).single())
            .map(|at_tz| at_tz.with_timezone(&Utc))
            .map(Ok)
    }

    // yyyy/mm/dd hh:mm:ss
    // - 2014/4/8 22:05
    // - 2014/04/08 22:05
    // - 2014/04/2 03:00:51
    // - 2014/4/02 03:00:51
    // - 2012/03/19 10:11:59
    // - 2012/03/19 10:11:59.3186369
    // - 2012/03/19 10:11:59.318 PM
    #[inline]
    fn slash_ymd_hms(&self, input: &str) -> Option<Result<DateTime<Utc>>> {
        let re: &Regex = regex! {
                r"^[0-9]{4}/[0-9]{1,2}/[0-9]{1,2}\s+[0-9]{1,2}:[0-9]{2}(?::[0-9]{2})?(?:\.[0-9]{1,9})?\s*(?:am|pm|AM|PM)?$"
        };
        if !re.is_match(input) {
            return None;
        }

        let items = match time_shape(input) {
            TimeShape::Hms => fmt_items!("%Y/%m/%d %H:%M:%S"),
            TimeShape::Hm => fmt_items!("%Y/%m/%d %H:%M"),
            TimeShape::HmsF => fmt_items!("%Y/%m/%d %H:%M:%S%.f"),
            TimeShape::ImsP | TimeShape::HmsFP => fmt_items!("%Y/%m/%d %I:%M:%S%.f %P"),
            TimeShape::ImP => fmt_items!("%Y/%m/%d %I:%M %P"),
        };
        self.dt_from_items(input, items)
            .ok()
            .map(|at_tz| at_tz.with_timezone(&Utc))
            .map(Ok)
    }

    // yyyy/mm/dd
    // - 2014/3/31
    // - 2014/03/31
    #[inline]
    fn slash_ymd(&self, input: &str) -> Option<Result<DateTime<Utc>>> {
        let re: &Regex = regex! {r"^[0-9]{4}/[0-9]{1,2}/[0-9]{1,2}$"
        };
        if !re.is_match(input) {
            return None;
        }

        let now = Utc::now()
            .date()
            .and_time(self.default_time)?
            .with_timezone(self.tz);
        Self::naive_date_from_items(input, fmt_items!("%Y/%m/%d"))
            .ok()
            .map(|parsed| parsed.and_time(now.time()))
            .and_then(|datetime| self.tz.from_local_datetime(&datetime).single())
            .map(|at_tz| at_tz.with_timezone(&Utc))
            .map(Ok)
    }
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn unix_timestamp() {
        let parse = Parse::new(&Utc, Utc::now().time());

        let test_cases = vec![
            ("0", Utc.ymd(1970, 1, 1).and_hms(0, 0, 0)),
            ("0000000000", Utc.ymd(1970, 1, 1).and_hms(0, 0, 0)),
            ("0000000000000", Utc.ymd(1970, 1, 1).and_hms(0, 0, 0)),
            ("0000000000000000000", Utc.ymd(1970, 1, 1).and_hms(0, 0, 0)),
            ("-770172300", Utc.ymd(1945, 8, 5).and_hms(23, 15, 0)),
            (
                "1671673426.123456789",
                Utc.ymd(2022, 12, 22).and_hms_nano(1, 43, 46, 123456768),
            ),
            ("1511648546", Utc.ymd(2017, 11, 25).and_hms(22, 22, 26)),
            (
                "1620036248.420",
                Utc.ymd(2021, 5, 3).and_hms_milli(10, 4, 8, 420),
            ),
            (
                "1620036248.717915136",
                Utc.ymd(2021, 5, 3).and_hms_nano(10, 4, 8, 717915136),
            ),
        ];

        for &(input, want) in test_cases.iter() {
            assert_eq!(
                parse.unix_timestamp(input).unwrap().unwrap(),
                want,
                "unix_timestamp/{}",
                input
            )
        }
        assert!(parse.unix_timestamp("15116").is_some());
        assert!(
            parse
                .unix_timestamp("16200248727179150001620024872717915000") //DevSkim: ignore DS173237
                .is_some()
        );
        assert!(parse.unix_timestamp("not-a-ts").is_none());
        // Non-finite floats must be rejected, whether caught by the lead-byte
        // pre-filter (bare `inf`/`nan`) or the is_finite check (signed forms).
        for input in [
            "inf", "nan", "INF", "NaN", "infinity", "+inf", "-inf", "-nan",
        ] {
            assert!(
                parse.unix_timestamp(input).is_none(),
                "unix_timestamp must reject non-finite {input}"
            );
        }
    }

    #[test]
    fn rfc3339() {
        let parse = Parse::new(&Utc, Utc::now().time());

        let test_cases = [
            (
                "2021-05-01T01:17:02.604456Z",
                Utc.ymd(2021, 5, 1).and_hms_nano(1, 17, 2, 604456000),
            ),
            (
                "2017-11-25T22:34:50Z",
                Utc.ymd(2017, 11, 25).and_hms(22, 34, 50),
            ),
        ];

        for &(input, want) in test_cases.iter() {
            assert_eq!(
                parse.rfc3339(input).unwrap().unwrap(),
                want,
                "rfc3339/{}",
                input
            )
        }
        assert!(parse.rfc3339("2017-11-25 22:34:50").is_none());
        assert!(parse.rfc3339("not-date-time").is_none());
    }

    #[test]
    fn rfc2822() {
        let parse = Parse::new(&Utc, Utc::now().time());

        let test_cases = [
            (
                "Wed, 02 Jun 2021 06:31:39 GMT",
                Utc.ymd(2021, 6, 2).and_hms(6, 31, 39),
            ),
            (
                "Wed, 02 Jun 2021 06:31:39 PDT",
                Utc.ymd(2021, 6, 2).and_hms(13, 31, 39),
            ),
        ];

        for &(input, want) in test_cases.iter() {
            assert_eq!(
                parse.rfc2822(input).unwrap().unwrap(),
                want,
                "rfc2822/{}",
                input
            )
        }
        assert!(parse.rfc2822("02 Jun 2021 06:31:39").is_none());
        assert!(parse.rfc2822("not-date-time").is_none());
    }

    #[test]
    fn ymd_hms() {
        let parse = Parse::new(&Utc, Utc::now().time());

        let test_cases = [
            ("2021-04-30 21:14", Utc.ymd(2021, 4, 30).and_hms(21, 14, 0)),
            (
                "2021-04-30 21:14:10",
                Utc.ymd(2021, 4, 30).and_hms(21, 14, 10),
            ),
            (
                "2021-04-30 21:14:10.052282",
                Utc.ymd(2021, 4, 30).and_hms_micro(21, 14, 10, 52282),
            ),
            (
                "2014-04-26 05:24:37 PM",
                Utc.ymd(2014, 4, 26).and_hms(17, 24, 37),
            ),
            (
                "2014-04-26 17:24:37.123",
                Utc.ymd(2014, 4, 26).and_hms_milli(17, 24, 37, 123),
            ),
            (
                "2014-04-26 17:24:37.3186369",
                Utc.ymd(2014, 4, 26).and_hms_nano(17, 24, 37, 318636900),
            ),
            (
                "2012-08-03 18:31:59.257000000",
                Utc.ymd(2012, 8, 3).and_hms_nano(18, 31, 59, 257000000),
            ),
            // ISO 8601 with 'T' separator and no timezone (naive wall-clock).
            // Must agree with the space-separated form on the same wall-clock instant.
            ("2020-01-15T08:00", Utc.ymd(2020, 1, 15).and_hms(8, 0, 0)),
            ("2020-01-15T08:00:00", Utc.ymd(2020, 1, 15).and_hms(8, 0, 0)),
            (
                "2020-01-15T08:00:00.123",
                Utc.ymd(2020, 1, 15).and_hms_milli(8, 0, 0, 123),
            ),
            (
                "2020-01-15T08:00:00.123456",
                Utc.ymd(2020, 1, 15).and_hms_micro(8, 0, 0, 123456),
            ),
            (
                "2020-01-15T08:00:00.123456789",
                Utc.ymd(2020, 1, 15).and_hms_nano(8, 0, 0, 123456789),
            ),
        ];

        for &(input, want) in test_cases.iter() {
            assert_eq!(
                parse.ymd_hms(input).unwrap().unwrap(),
                want,
                "ymd_hms/{}",
                input
            )
        }
        assert!(parse.ymd_hms("not-date-time").is_none());

        // T and space separators must produce the same instant.
        let t_form = parse.ymd_hms("2020-01-15T08:00:00").unwrap().unwrap();
        let space_form = parse.ymd_hms("2020-01-15 08:00:00").unwrap().unwrap();
        assert_eq!(t_form, space_form, "T-separator vs space disagree");
    }

    #[test]
    fn ymd_hms_z() {
        let parse = Parse::new(&Utc, Utc::now().time());

        let test_cases = [
            (
                "2017-11-25 13:31:15 PST",
                Utc.ymd(2017, 11, 25).and_hms(21, 31, 15),
            ),
            (
                "2017-11-25 13:31 PST",
                Utc.ymd(2017, 11, 25).and_hms(21, 31, 0),
            ),
            (
                "2014-12-16 06:20:00 UTC",
                Utc.ymd(2014, 12, 16).and_hms(6, 20, 0),
            ),
            (
                "2014-12-16 06:20:00 GMT",
                Utc.ymd(2014, 12, 16).and_hms(6, 20, 0),
            ),
            (
                "2014-04-26 13:13:43 +0800",
                Utc.ymd(2014, 4, 26).and_hms(5, 13, 43),
            ),
            (
                "2014-04-26 13:13:44 +09:00",
                Utc.ymd(2014, 4, 26).and_hms(4, 13, 44),
            ),
            (
                "2012-08-03 18:31:59.257000000 +0000",
                Utc.ymd(2012, 8, 3).and_hms_nano(18, 31, 59, 257000000),
            ),
            (
                "2015-09-30 18:48:56.35272715 UTC",
                Utc.ymd(2015, 9, 30).and_hms_nano(18, 48, 56, 352727150),
            ),
        ];

        for &(input, want) in test_cases.iter() {
            assert_eq!(
                parse.ymd_hms_z(input).unwrap().unwrap(),
                want,
                "ymd_hms_z/{}",
                input
            )
        }
        assert!(parse.ymd_hms_z("not-date-time").is_none());
        // Pre-filter boundary: exactly 16 chars is rejected by length guard (< 17)
        assert!(parse.ymd_hms_z("2021-04-30 21:14").is_none()); // 16 chars, rejected by length guard
        // 17 chars but byte[10] is not whitespace — rejected by whitespace check
        assert!(parse.ymd_hms_z("2021-04-30X21:14Z").is_none()); // 17 chars, byte[10]='X' not space
        // 17 chars with whitespace at byte[10] proceeds to regex but regex rejects malformed input
        assert!(parse.ymd_hms_z("2021-04-30 21:1XZ").is_none()); // 17 chars, byte[10]=' ', regex rejects
    }

    #[test]
    fn ymd() {
        let parse = Parse::new(&Utc, Utc::now().time());

        let test_cases = [(
            "2021-02-21",
            Utc.ymd(2021, 2, 21).and_time(Utc::now().time()),
        )];

        for &(input, want) in test_cases.iter() {
            assert_eq!(
                parse
                    .ymd(input)
                    .unwrap()
                    .unwrap()
                    .trunc_subsecs(0)
                    .with_second(0)
                    .unwrap(),
                want.unwrap().trunc_subsecs(0).with_second(0).unwrap(),
                "ymd/{}",
                input
            )
        }
        assert!(parse.ymd("not-date-time").is_none());
    }

    #[test]
    fn ymd_z() {
        let parse = Parse::new(&Utc, Utc::now().time());
        let now_at_pst = Utc::now().with_timezone(&FixedOffset::west(8 * 3600));
        let now_at_cst = Utc::now().with_timezone(&FixedOffset::east(8 * 3600));

        let test_cases = [
            (
                "2021-02-21 PST",
                FixedOffset::west(8 * 3600)
                    .ymd(2021, 2, 21)
                    .and_time(now_at_pst.time())
                    .map(|dt| dt.with_timezone(&Utc)),
            ),
            (
                "2021-02-21 UTC",
                FixedOffset::west(0)
                    .ymd(2021, 2, 21)
                    .and_time(Utc::now().time())
                    .map(|dt| dt.with_timezone(&Utc)),
            ),
            (
                "2020-07-20+08:00",
                FixedOffset::east(8 * 3600)
                    .ymd(2020, 7, 20)
                    .and_time(now_at_cst.time())
                    .map(|dt| dt.with_timezone(&Utc)),
            ),
        ];

        for &(input, want) in test_cases.iter() {
            assert_eq!(
                parse
                    .ymd_z(input)
                    .unwrap()
                    .unwrap()
                    .trunc_subsecs(0)
                    .with_second(0)
                    .unwrap(),
                want.unwrap().trunc_subsecs(0).with_second(0).unwrap(),
                "ymd_z/{}",
                input
            )
        }
        assert!(parse.ymd_z("not-date-time").is_none());
        // Pre-filter boundary: exactly 10 chars (bare date) is rejected (<= 10 guard), 11+ proceeds
        assert!(parse.ymd_z("2021-02-21").is_none()); // exactly 10 chars, rejected
        assert!(parse.ymd_z("2021-02-21X").is_none()); // 11 chars, proceeds to regex but regex rejects
    }

    #[test]
    fn month_ymd() {
        let parse = Parse::new(&Utc, Utc::now().time());

        let test_cases = [(
            "2021-Feb-21",
            Utc.ymd(2021, 2, 21).and_time(Utc::now().time()),
        )];

        for &(input, want) in test_cases.iter() {
            assert_eq!(
                parse
                    .month_ymd(input)
                    .unwrap()
                    .unwrap()
                    .trunc_subsecs(0)
                    .with_second(0)
                    .unwrap(),
                want.unwrap().trunc_subsecs(0).with_second(0).unwrap(),
                "month_ymd/{}",
                input
            )
        }
        assert!(parse.month_ymd("not-date-time").is_none());
    }

    #[test]
    fn month_mdy_hms() {
        let parse = Parse::new(&Utc, Utc::now().time());

        let test_cases = [
            (
                "May 8, 2009 5:57:51 PM",
                Utc.ymd(2009, 5, 8).and_hms(17, 57, 51),
            ),
            (
                "September 17, 2012 10:09am",
                Utc.ymd(2012, 9, 17).and_hms(10, 9, 0),
            ),
            (
                "September 17, 2012, 10:10:09",
                Utc.ymd(2012, 9, 17).and_hms(10, 10, 9),
            ),
        ];

        for &(input, want) in test_cases.iter() {
            assert_eq!(
                parse.month_mdy_hms(input).unwrap().unwrap(),
                want,
                "month_mdy_hms/{}",
                input
            )
        }
        assert!(parse.month_mdy_hms("not-date-time").is_none());
    }

    #[test]
    fn month_mdy_hms_z() {
        let parse = Parse::new(&Utc, Utc::now().time());

        let test_cases = [
            (
                "May 02, 2021 15:51:31 UTC",
                Utc.ymd(2021, 5, 2).and_hms(15, 51, 31),
            ),
            (
                "May 02, 2021 15:51 UTC",
                Utc.ymd(2021, 5, 2).and_hms(15, 51, 0),
            ),
            (
                "May 26, 2021, 12:49 AM PDT",
                Utc.ymd(2021, 5, 26).and_hms(7, 49, 0),
            ),
            (
                "September 17, 2012 at 10:09am PST",
                Utc.ymd(2012, 9, 17).and_hms(18, 9, 0),
            ),
        ];

        for &(input, want) in test_cases.iter() {
            assert_eq!(
                parse.month_mdy_hms_z(input).unwrap().unwrap(),
                want,
                "month_mdy_hms_z/{}",
                input
            )
        }
        assert!(parse.month_mdy_hms_z("not-date-time").is_none());
        // Pre-filter: 20+ chars required; no isolated 4-digit year → has_year=false, rejected
        assert!(parse.month_mdy_hms_z("May 27, 02:45:27 XX PST").is_none()); // 23 chars, no 4-digit year
        // Pre-filter: 20+ chars with isolated 4-digit sequence → has_year=true, regex rejects format
        assert!(parse.month_mdy_hms_z("May 27 1234 something PST").is_none()); // 25 chars, has_year=true but regex rejects
    }

    #[test]
    fn month_mdy() {
        let parse = Parse::new(&Utc, Utc::now().time());

        let test_cases = [
            (
                "May 25, 2021",
                Utc.ymd(2021, 5, 25).and_time(Utc::now().time()),
            ),
            (
                "oct 7, 1970",
                Utc.ymd(1970, 10, 7).and_time(Utc::now().time()),
            ),
            (
                "oct 7, 70",
                Utc.ymd(1970, 10, 7).and_time(Utc::now().time()),
            ),
            (
                "oct. 7, 1970",
                Utc.ymd(1970, 10, 7).and_time(Utc::now().time()),
            ),
            (
                "oct. 7, 70",
                Utc.ymd(1970, 10, 7).and_time(Utc::now().time()),
            ),
            (
                "October 7, 1970",
                Utc.ymd(1970, 10, 7).and_time(Utc::now().time()),
            ),
        ];

        for &(input, want) in test_cases.iter() {
            assert_eq!(
                parse
                    .month_mdy(input)
                    .unwrap()
                    .unwrap()
                    .trunc_subsecs(0)
                    .with_second(0)
                    .unwrap(),
                want.unwrap().trunc_subsecs(0).with_second(0).unwrap(),
                "month_mdy/{}",
                input
            )
        }
        assert!(parse.month_mdy("not-date-time").is_none());
    }

    #[test]
    fn month_dmy_hms() {
        let parse = Parse::new(&Utc, Utc::now().time());

        let test_cases = [
            (
                "12 Feb 2006, 19:17",
                Utc.ymd(2006, 2, 12).and_hms(19, 17, 0),
            ),
            ("12 Feb 2006 19:17", Utc.ymd(2006, 2, 12).and_hms(19, 17, 0)),
            (
                "14 May 2019 19:11:40.164",
                Utc.ymd(2019, 5, 14).and_hms_milli(19, 11, 40, 164),
            ),
        ];

        for &(input, want) in test_cases.iter() {
            assert_eq!(
                parse.month_dmy_hms(input).unwrap().unwrap(),
                want,
                "month_dmy_hms/{}",
                input
            )
        }
        assert!(parse.month_dmy_hms("not-date-time").is_none());
    }

    #[test]
    fn month_dmy() {
        let parse = Parse::new(&Utc, Utc::now().time());

        let test_cases = [
            ("7 oct 70", Utc.ymd(1970, 10, 7).and_time(Utc::now().time())),
            (
                "7 oct 1970",
                Utc.ymd(1970, 10, 7).and_time(Utc::now().time()),
            ),
            (
                "03 February 2013",
                Utc.ymd(2013, 2, 3).and_time(Utc::now().time()),
            ),
            (
                "1 July 2013",
                Utc.ymd(2013, 7, 1).and_time(Utc::now().time()),
            ),
        ];

        for &(input, want) in test_cases.iter() {
            assert_eq!(
                parse
                    .month_dmy(input)
                    .unwrap()
                    .unwrap()
                    .trunc_subsecs(0)
                    .with_second(0)
                    .unwrap(),
                want.unwrap().trunc_subsecs(0).with_second(0).unwrap(),
                "month_dmy/{}",
                input
            )
        }
        assert!(parse.month_dmy("not-date-time").is_none());
    }

    // Explicitly tests the `four_digit_year` fast path in `month_dmy` (skips `%d %B %y`) and
    // the else-branch fallback that tries `%d %B %y` first then `%d %B %Y`.
    #[test]
    fn month_dmy_year_fast_path() {
        let parse = Parse::new(&Utc, Utc::now().time());

        // Fast path: 4-digit year — `four_digit_year` is true, goes directly to `%d %B %Y`
        let four_digit = parse.month_dmy("14 May 2019").unwrap().unwrap();
        assert_eq!(four_digit.year(), 2019);
        assert_eq!(four_digit.month(), 5);
        assert_eq!(four_digit.day(), 14);

        // Else-branch: 2-digit year — `four_digit_year` is false, tries `%d %B %y` first
        // chrono %y: 00–68 → 2000–2068, so "19" → 2019 (not 1919)
        let two_digit = parse.month_dmy("14 May 19").unwrap().unwrap();
        assert_eq!(two_digit.year(), 2019);
        assert_eq!(two_digit.month(), 5);
        assert_eq!(two_digit.day(), 14);
    }

    #[test]
    fn slash_mdy_hms() {
        let parse = Parse::new(&Utc, Utc::now().time());

        let test_cases = vec![
            ("4/8/2014 22:05", Utc.ymd(2014, 4, 8).and_hms(22, 5, 0)),
            ("04/08/2014 22:05", Utc.ymd(2014, 4, 8).and_hms(22, 5, 0)),
            ("4/8/14 22:05", Utc.ymd(2014, 4, 8).and_hms(22, 5, 0)),
            ("04/2/2014 03:00:51", Utc.ymd(2014, 4, 2).and_hms(3, 0, 51)),
            ("8/8/1965 12:00:00 AM", Utc.ymd(1965, 8, 8).and_hms(0, 0, 0)),
            (
                "8/8/1965 01:00:01 PM",
                Utc.ymd(1965, 8, 8).and_hms(13, 0, 1),
            ),
            ("8/8/1965 01:00 PM", Utc.ymd(1965, 8, 8).and_hms(13, 0, 0)),
            ("8/8/1965 1:00 PM", Utc.ymd(1965, 8, 8).and_hms(13, 0, 0)),
            ("8/8/1965 12:00 AM", Utc.ymd(1965, 8, 8).and_hms(0, 0, 0)),
            ("4/02/2014 03:00:51", Utc.ymd(2014, 4, 2).and_hms(3, 0, 51)),
            (
                "03/19/2012 10:11:59",
                Utc.ymd(2012, 3, 19).and_hms(10, 11, 59),
            ),
            (
                "03/19/2012 10:11:59.3186369",
                Utc.ymd(2012, 3, 19).and_hms_nano(10, 11, 59, 318636900),
            ),
        ];

        for &(input, want) in test_cases.iter() {
            assert_eq!(
                parse.slash_mdy_hms(input).unwrap().unwrap(),
                want,
                "slash_mdy_hms/{}",
                input
            )
        }
        assert!(parse.slash_mdy_hms("not-date-time").is_none());
    }

    #[test]
    fn slash_mdy() {
        let parse = Parse::new(&Utc, Utc::now().time());

        let test_cases = [
            (
                "3/31/2014",
                Utc.ymd(2014, 3, 31).and_time(Utc::now().time()),
            ),
            (
                "03/31/2014",
                Utc.ymd(2014, 3, 31).and_time(Utc::now().time()),
            ),
            ("08/21/71", Utc.ymd(1971, 8, 21).and_time(Utc::now().time())),
            ("8/1/71", Utc.ymd(1971, 8, 1).and_time(Utc::now().time())),
        ];

        for &(input, want) in test_cases.iter() {
            assert_eq!(
                parse
                    .slash_mdy(input)
                    .unwrap()
                    .unwrap()
                    .trunc_subsecs(0)
                    .with_second(0)
                    .unwrap(),
                want.unwrap().trunc_subsecs(0).with_second(0).unwrap(),
                "slash_mdy/{}",
                input
            )
        }
        assert!(parse.slash_mdy("not-date-time").is_none());
    }

    #[test]
    fn slash_dmy() {
        let mut parse = Parse::new(&Utc, Utc::now().time());

        let test_cases = [
            (
                "31/3/2014",
                Utc.ymd(2014, 3, 31).and_time(Utc::now().time()),
            ),
            (
                "13/11/2014",
                Utc.ymd(2014, 11, 13).and_time(Utc::now().time()),
            ),
            ("21/08/71", Utc.ymd(1971, 8, 21).and_time(Utc::now().time())),
            ("1/8/71", Utc.ymd(1971, 8, 1).and_time(Utc::now().time())),
        ];

        for &(input, want) in test_cases.iter() {
            assert_eq!(
                parse
                    .prefer_dmy(true)
                    .slash_dmy(input)
                    .unwrap()
                    .unwrap()
                    .trunc_subsecs(0)
                    .with_second(0)
                    .unwrap(),
                want.unwrap().trunc_subsecs(0).with_second(0).unwrap(),
                "slash_dmy/{}",
                input
            )
        }
        assert!(parse.slash_dmy("not-date-time").is_none());
    }

    #[test]
    fn slash_ymd_hms() {
        let parse = Parse::new(&Utc, Utc::now().time());

        let test_cases = [
            ("2014/4/8 22:05", Utc.ymd(2014, 4, 8).and_hms(22, 5, 0)),
            ("2014/04/08 22:05", Utc.ymd(2014, 4, 8).and_hms(22, 5, 0)),
            ("2014/04/2 03:00:51", Utc.ymd(2014, 4, 2).and_hms(3, 0, 51)),
            ("2014/4/02 03:00:51", Utc.ymd(2014, 4, 2).and_hms(3, 0, 51)),
            (
                "2012/03/19 10:11:59",
                Utc.ymd(2012, 3, 19).and_hms(10, 11, 59),
            ),
            (
                "2012/03/19 10:11:59.3186369",
                Utc.ymd(2012, 3, 19).and_hms_nano(10, 11, 59, 318636900),
            ),
        ];

        for &(input, want) in test_cases.iter() {
            assert_eq!(
                parse.slash_ymd_hms(input).unwrap().unwrap(),
                want,
                "slash_ymd_hms/{}",
                input
            )
        }
        assert!(parse.slash_ymd_hms("not-date-time").is_none());
    }

    #[test]
    fn slash_ymd() {
        let parse = Parse::new(&Utc, Utc::now().time());

        let test_cases = [
            (
                "2014/3/31",
                Utc.ymd(2014, 3, 31).and_time(Utc::now().time()),
            ),
            (
                "2014/03/31",
                Utc.ymd(2014, 3, 31).and_time(Utc::now().time()),
            ),
        ];

        for &(input, want) in test_cases.iter() {
            assert_eq!(
                parse
                    .slash_ymd(input)
                    .unwrap()
                    .unwrap()
                    .trunc_subsecs(0)
                    .with_second(0)
                    .unwrap(),
                want.unwrap().trunc_subsecs(0).with_second(0).unwrap(),
                "slash_ymd/{}",
                input
            )
        }
        assert!(parse.slash_ymd("not-date-time").is_none());
    }

    #[test]
    fn time_shape_classification() {
        use TimeShape::{Hm, Hms, HmsF, HmsFP, ImP, ImsP};

        let cases = [
            ("2021-04-30 21:14", Hm),
            ("2021-04-30T21:14", Hm),
            ("2021-04-30 21:14:10", Hms),
            ("2021-04-30 21:14:10.052282", HmsF),
            ("8/8/1965 12:00 AM", ImP),
            ("8/8/1965 12:00am", ImP),
            ("8/8/1965 01:00:01 PM", ImsP),
            ("September 17 2012 10:09am", ImP),
            ("03/19/2012 10:11:59.318 PM", HmsFP),
            // A one-colon time with a fraction has no format either; it lands
            // in a shape whose format still rejects it, so it keeps failing.
            ("03/19/2012 10:11.123", Hm),
        ];
        for (input, want) in cases {
            assert!(
                time_shape(input) == want,
                "time_shape misclassified {input}"
            );
        }
    }

    /// Every `fmt_items!` literal in this file must be a valid strftime format.
    ///
    /// The macro compiles its literal lazily on first use and `expect`s the
    /// result, so an invalid format would panic the first time some input
    /// happened to reach that particular link of an `or_else` chain — possibly
    /// only in production. The literals are extracted from this file's own
    /// source rather than re-listed here, so the check cannot drift out of
    /// sync as parsers are added or reworked.
    #[test]
    fn every_fmt_items_literal_is_valid() {
        let src = include_str!("datetime.rs");
        // Matches uses, not the macro definition (which has no `!`).
        let call = Regex::new(r#"fmt_items!\("([^"]*)"\)"#).unwrap();

        let mut checked = 0_usize;
        for caps in call.captures_iter(src) {
            let fmt = &caps[1];
            assert!(
                chrono::format::StrftimeItems::new(fmt).parse().is_ok(),
                "invalid strftime literal: {fmt}"
            );
            checked += 1;
        }
        // Guard against the extraction silently matching nothing.
        assert!(
            checked >= 50,
            "expected to check every fmt_items! literal, only found {checked}"
        );
    }

    /// Fractional seconds combined with an AM/PM marker (issue #12).
    ///
    /// The four families whose regex admits both a fraction and an AM/PM
    /// marker parse this with `%I:%M:%S%.f %P`. That one format also covers
    /// the fraction-less `ImsP` shape, because `%.f` consumes nothing when
    /// there is no period, so it replaced `%I:%M:%S %P` rather than joining
    /// it — which is why the plain AM/PM cases are re-asserted here too.
    #[test]
    fn fractional_seconds_with_ampm() {
        let parse = Parse::new(&Utc, Utc::now().time());

        let cases = [
            (
                "03/19/2012 10:11:59.318 PM",
                Utc.ymd(2012, 3, 19).and_hms_milli(22, 11, 59, 318),
            ),
            (
                "3/19/2012 1:11:59.318 am",
                Utc.ymd(2012, 3, 19).and_hms_milli(1, 11, 59, 318),
            ),
            (
                "03/19/12 10:11:59.318 PM",
                Utc.ymd(2012, 3, 19).and_hms_milli(22, 11, 59, 318),
            ),
            (
                "2012/03/19 10:11:59.318 PM",
                Utc.ymd(2012, 3, 19).and_hms_milli(22, 11, 59, 318),
            ),
            (
                "2012-03-19 10:11:59.318 PM",
                Utc.ymd(2012, 3, 19).and_hms_milli(22, 11, 59, 318),
            ),
            (
                "2012-03-19T10:11:59.318 PM",
                Utc.ymd(2012, 3, 19).and_hms_milli(22, 11, 59, 318),
            ),
            (
                "2012-03-19 10:11:59.3186369 PM",
                Utc.ymd(2012, 3, 19).and_hms_nano(22, 11, 59, 318636900),
            ),
            // Unchanged by the format swap: no period, so `%.f` matches empty.
            (
                "03/19/2012 10:11:59 PM",
                Utc.ymd(2012, 3, 19).and_hms(22, 11, 59),
            ),
            (
                "2012-03-19 10:11:59 PM",
                Utc.ymd(2012, 3, 19).and_hms(22, 11, 59),
            ),
        ];

        for (input, want) in cases {
            assert_eq!(parse.parse(input).unwrap(), want, "parse/{input}");
        }

        // slash_dmy_hms's arms are only reached day-first, so the cases above
        // never exercise them. Both year widths, since the parser dispatches
        // on year width before it dispatches on time shape.
        let dmy = Parse::new_with_preference(&Utc, Utc::now().time(), true);
        let dmy_cases = [
            (
                "19/03/2012 10:11:59.318 PM",
                Utc.ymd(2012, 3, 19).and_hms_milli(22, 11, 59, 318),
            ),
            (
                "19/03/12 10:11:59.318 PM",
                Utc.ymd(2012, 3, 19).and_hms_milli(22, 11, 59, 318),
            ),
            (
                "9/3/2012 1:11:59.318 am",
                Utc.ymd(2012, 3, 9).and_hms_milli(1, 11, 59, 318),
            ),
            // Fraction-less, to pin that the format swap left it alone.
            (
                "19/03/2012 10:11:59 PM",
                Utc.ymd(2012, 3, 19).and_hms(22, 11, 59),
            ),
        ];

        for (input, want) in dmy_cases {
            assert_eq!(dmy.parse(input).unwrap(), want, "prefer_dmy/{input}");
        }
    }

    /// Inputs that resemble an accepted shape but do not parse. Pinned so that
    /// format-chain refactors stay result-preserving: a classifier that newly
    /// accepts any of these has widened the accepted input set, which is a
    /// behavior change rather than a performance optimization.
    ///
    /// These are independent gaps, not one gap. Some are turned away by their
    /// family regex before any format is tried; others clear the regex and
    /// then match no format. Each case carries its own reason below.
    #[test]
    fn unsupported_shapes_still_fail() {
        let parse = Parse::new(&Utc, Utc::now().time());

        for input in [
            // Rejected at the regex gate: month_mdy_hms has no
            // fractional-seconds group at all, and strips `.` before parsing.
            "May 8, 2009 5:57:51.123 PM",
            // Rejected at the regex gate: month_dmy_hms has no am/pm
            // alternative, which is also why that chain's two `%I ... %P`
            // formats were unreachable and have been removed. Note the first
            // of these carries no fractional seconds — it is the missing
            // am/pm alternative alone that rejects it.
            "14 May 2019 07:11:40 PM",
            "14 May 2019 07:11:40.164 PM",
            // Clears the regex, matches no format: a fraction with no seconds
            // field. The optional groups admit it, but it buckets as Hm/ImP
            // and fails there. Malformed rather than a real shape. Note the
            // first two carry no AM/PM marker.
            "03/19/2012 10:11.123",
            "2021-04-30 21:14.052282",
            "03/19/2012 10:11.123 PM",
            // Clears the regex, matches no format: `%I` only accepts a 1-12
            // hour, so a 24-hour reading cannot carry an AM/PM marker.
            "03/19/2012 22:11:59.318 PM",
        ] {
            assert!(
                parse.parse(input).is_err(),
                "{input} is expected to remain unsupported"
            );
        }
    }
}