sley_core/

lib.rs

#![cfg_attr(not(test), deny(clippy::unwrap_used, clippy::expect_used))]

use std::borrow::Borrow;
use std::error::Error;
use std::fmt;
use std::ops::Deref;
use std::path::{Path, PathBuf};
use std::str::FromStr;
use std::sync::OnceLock;

pub const UPSTREAM_GIT_COMPAT_VERSION: &str = "2.55.0";

static ORIGINAL_CWD: OnceLock<Option<PathBuf>> = OnceLock::new();

pub fn set_original_cwd(path: Option<PathBuf>) {
    let _ = ORIGINAL_CWD.set(path);
}

pub fn original_cwd() -> Option<PathBuf> {
    ORIGINAL_CWD.get()?.clone()
}

#[derive(Debug, Default, Clone, PartialEq, Eq)]
pub enum DateMode {
    #[default]
    Default,
    Local,
    Raw,
    RawLocal,
    Unix,
    Short,
    ShortLocal,
    Iso,
    IsoLocal,
    IsoStrict,
    IsoStrictLocal,
    Rfc2822,
    Rfc2822Local,
    Relative,
    Human,
    HumanLocal,
    Strftime {
        template: String,
        local: bool,
    },
}

impl DateMode {
    pub fn parse(value: &str) -> Option<Self> {
        if let Some(template) = value.strip_prefix("format:") {
            return Some(Self::Strftime {
                template: template.to_string(),
                local: false,
            });
        }
        if let Some(template) = value.strip_prefix("format-local:") {
            return Some(Self::Strftime {
                template: template.to_string(),
                local: true,
            });
        }
        if value == "tformat:" || value.starts_with("tformat:") {
            return Some(Self::Strftime {
                template: value["tformat:".len()..].to_string(),
                local: false,
            });
        }
        if value == "auto:" || value.starts_with("auto:") {
            return Some(Self::Default);
        }
        Some(match value {
            "default" => Self::Default,
            "default-local" | "local" => Self::Local,
            "raw" => Self::Raw,
            "raw-local" => Self::RawLocal,
            "unix" => Self::Unix,
            "short" => Self::Short,
            "short-local" => Self::ShortLocal,
            "iso" | "iso8601" => Self::Iso,
            "iso-local" | "iso8601-local" => Self::IsoLocal,
            "iso-strict" | "iso8601-strict" => Self::IsoStrict,
            "iso-strict-local" | "iso8601-strict-local" => Self::IsoStrictLocal,
            "rfc" | "rfc2822" => Self::Rfc2822,
            "rfc-local" | "rfc2822-local" => Self::Rfc2822Local,
            "relative" | "relative-local" => Self::Relative,
            "human" => Self::Human,
            "human-local" => Self::HumanLocal,
            _ => return None,
        })
    }

    pub fn parse_atom_modifier(modifier: Option<&str>) -> Option<Self> {
        modifier.map_or(Some(Self::Default), Self::parse)
    }

    pub fn render(&self, timestamp: i64, timezone: &str) -> Option<String> {
        let tz = if self.is_local() { "+0000" } else { timezone };
        let parts = DateParts::from_timestamp(timestamp, tz)?;
        Some(match self {
            Self::Default | Self::Local => {
                let base = format!(
                    "{} {} {} {:02}:{:02}:{:02} {}",
                    parts.weekday,
                    MONTHS_ABBR[(parts.month - 1) as usize],
                    parts.day,
                    parts.hour,
                    parts.minute,
                    parts.second,
                    parts.year,
                );
                if self.is_local() {
                    base
                } else {
                    format!("{base} {}", parts.timezone)
                }
            }
            Self::Raw | Self::RawLocal => format!("{} {}", parts.timestamp, parts.timezone),
            Self::Unix => parts.timestamp.to_string(),
            Self::Short | Self::ShortLocal => {
                format!("{:04}-{:02}-{:02}", parts.year, parts.month, parts.day)
            }
            Self::Iso | Self::IsoLocal => format!(
                "{:04}-{:02}-{:02} {:02}:{:02}:{:02} {}",
                parts.year,
                parts.month,
                parts.day,
                parts.hour,
                parts.minute,
                parts.second,
                parts.timezone,
            ),
            Self::IsoStrict | Self::IsoStrictLocal => format!(
                "{:04}-{:02}-{:02}T{:02}:{:02}:{:02}{}",
                parts.year,
                parts.month,
                parts.day,
                parts.hour,
                parts.minute,
                parts.second,
                strict_timezone(parts.timezone),
            ),
            Self::Rfc2822 | Self::Rfc2822Local => format!(
                "{}, {} {} {:04} {:02}:{:02}:{:02} {}",
                parts.weekday,
                parts.day,
                MONTHS_ABBR[(parts.month - 1) as usize],
                parts.year,
                parts.hour,
                parts.minute,
                parts.second,
                parts.timezone,
            ),
            Self::Relative => relative_date(parts.timestamp),
            Self::Human | Self::HumanLocal => format!(
                "{} {} {} {:02}:{:02}:{:02} {} {}",
                parts.weekday,
                MONTHS_ABBR[(parts.month - 1) as usize],
                parts.day,
                parts.hour,
                parts.minute,
                parts.second,
                parts.year,
                parts.timezone,
            ),
            Self::Strftime { template, .. } => strftime(template, &parts),
        })
    }

    pub fn is_local(&self) -> bool {
        matches!(
            self,
            Self::Local
                | Self::RawLocal
                | Self::ShortLocal
                | Self::IsoLocal
                | Self::IsoStrictLocal
                | Self::Rfc2822Local
                | Self::HumanLocal
                | Self::Strftime { local: true, .. }
        )
    }
}

const MONTHS_ABBR: [&str; 12] = [
    "Jan", "Feb", "Mar", "Apr", "May", "Jun", "Jul", "Aug", "Sep", "Oct", "Nov", "Dec",
];

const MONTHS_FULL: [&str; 12] = [
    "January",
    "February",
    "March",
    "April",
    "May",
    "June",
    "July",
    "August",
    "September",
    "October",
    "November",
    "December",
];

const WEEKDAYS_FULL: [&str; 7] = [
    "Sunday",
    "Monday",
    "Tuesday",
    "Wednesday",
    "Thursday",
    "Friday",
    "Saturday",
];

struct DateParts<'a> {
    timestamp: i64,
    timezone: &'a str,
    weekday: &'static str,
    year: i64,
    month: u32,
    day: u32,
    hour: i64,
    minute: i64,
    second: i64,
}

impl<'a> DateParts<'a> {
    fn from_timestamp(timestamp: i64, timezone: &'a str) -> Option<Self> {
        const WEEKDAYS: [&str; 7] = ["Sun", "Mon", "Tue", "Wed", "Thu", "Fri", "Sat"];
        let offset_seconds = timezone_offset_seconds(timezone)?;
        let local = timestamp + offset_seconds;
        let days = local.div_euclid(86_400);
        let seconds = local.rem_euclid(86_400);
        let (year, month, day) = civil_from_days(days);
        Some(Self {
            timestamp,
            timezone,
            weekday: WEEKDAYS[(days + 4).rem_euclid(7) as usize],
            year,
            month,
            day,
            hour: seconds / 3_600,
            minute: (seconds % 3_600) / 60,
            second: seconds % 60,
        })
    }
}

fn timezone_offset_seconds(timezone: &str) -> Option<i64> {
    if timezone.len() != 5 {
        return None;
    }
    let sign = match timezone.as_bytes()[0] {
        b'+' => 1,
        b'-' => -1,
        _ => return None,
    };
    let hours = timezone[1..3].parse::<i64>().ok()?;
    let minutes = timezone[3..5].parse::<i64>().ok()?;
    Some(sign * (hours * 3_600 + minutes * 60))
}

fn strict_timezone(timezone: &str) -> String {
    let digits = timezone.strip_prefix(['+', '-']).unwrap_or(timezone);
    if digits == "0000" {
        "Z".to_string()
    } else if timezone.len() == 5 {
        format!("{}{}:{}", &timezone[..1], &timezone[1..3], &timezone[3..5])
    } else {
        timezone.to_string()
    }
}

fn strftime(template: &str, parts: &DateParts<'_>) -> String {
    let weekday_index = ["Sun", "Mon", "Tue", "Wed", "Thu", "Fri", "Sat"]
        .iter()
        .position(|day| *day == parts.weekday)
        .unwrap_or(0);
    let mut out = String::with_capacity(template.len());
    let mut chars = template.chars().peekable();
    while let Some(ch) = chars.next() {
        if ch != '%' {
            out.push(ch);
            continue;
        }
        match chars.next() {
            Some('Y') => out.push_str(&format!("{:04}", parts.year)),
            Some('y') => out.push_str(&format!("{:02}", parts.year.rem_euclid(100))),
            Some('m') => out.push_str(&format!("{:02}", parts.month)),
            Some('d') => out.push_str(&format!("{:02}", parts.day)),
            Some('e') => out.push_str(&format!("{:2}", parts.day)),
            Some('H') => out.push_str(&format!("{:02}", parts.hour)),
            Some('M') => out.push_str(&format!("{:02}", parts.minute)),
            Some('S') => out.push_str(&format!("{:02}", parts.second)),
            Some('b') | Some('h') => out.push_str(MONTHS_ABBR[(parts.month - 1) as usize]),
            Some('B') => out.push_str(MONTHS_FULL[(parts.month - 1) as usize]),
            Some('a') => out.push_str(parts.weekday),
            Some('A') => out.push_str(WEEKDAYS_FULL[weekday_index]),
            Some('%') => out.push('%'),
            Some('n') => out.push('\n'),
            Some('t') => out.push('\t'),
            Some(other) => {
                out.push('%');
                out.push(other);
            }
            None => out.push('%'),
        }
    }
    out
}

fn relative_date(timestamp: i64) -> String {
    let now = std::time::SystemTime::now()
        .duration_since(std::time::UNIX_EPOCH)
        .map(|duration| duration.as_secs() as i64)
        .unwrap_or(timestamp);
    if timestamp > now {
        return "in the future".to_string();
    }
    let diff = (now - timestamp) as u64;
    if diff < 90 {
        return format!("{diff} seconds ago");
    }
    let minutes = (diff + 30) / 60;
    if minutes < 90 {
        return format!("{minutes} minutes ago");
    }
    let hours = (diff + 1800) / 3600;
    if hours < 36 {
        return format!("{hours} hours ago");
    }
    let days = (diff + 43200) / 86400;
    if days < 14 {
        return format!("{days} days ago");
    }
    if days < 70 {
        return format!("{} weeks ago", (days + 3) / 7);
    }
    if days < 365 {
        return format!("{} months ago", (days + 15) / 30);
    }
    let years_scaled = (days * 10 + 183) / 365;
    if days < 365 * 2 {
        let months = ((days - 365) + 15) / 30;
        if months > 0 {
            return format!("1 year, {months} months ago");
        }
        return "1 year ago".to_string();
    }
    if years_scaled.is_multiple_of(10) {
        format!("{} years ago", years_scaled / 10)
    } else {
        format!("{}.{} years ago", years_scaled / 10, years_scaled % 10)
    }
}

fn civil_from_days(days: i64) -> (i64, u32, u32) {
    let days = days + 719_468;
    let era = if days >= 0 { days } else { days - 146_096 } / 146_097;
    let day_of_era = days - era * 146_097;
    let year_of_era =
        (day_of_era - day_of_era / 1460 + day_of_era / 36_524 - day_of_era / 146_096) / 365;
    let year = year_of_era + era * 400;
    let day_of_year = day_of_era - (365 * year_of_era + year_of_era / 4 - year_of_era / 100);
    let month_prime = (5 * day_of_year + 2) / 153;
    let day = day_of_year - (153 * month_prime + 2) / 5 + 1;
    let month = month_prime + if month_prime < 10 { 3 } else { -9 };
    let year = year + i64::from(month <= 2);
    (year, month as u32, day as u32)
}

fn is_scheme_char(ch: char) -> bool {
    ch.is_ascii_alphanumeric() || matches!(ch, '+' | '-' | '.')
}

/// Strip embedded credentials from `url` before showing it in user-facing output.
///
/// HTTP(S) userinfo (`user:password@host`) is replaced with `<redacted>@host`,
/// matching trace2's `GIT_TRACE2_REDACT` behavior. Non-URL strings (remote
/// names, file paths) are returned unchanged.
pub fn redact_url_for_display(url: &str) -> String {
    let mut out = String::with_capacity(url.len());
    let mut rest = url;
    while let Some(scheme_end) = rest.find("://") {
        let scheme_start = rest[..scheme_end]
            .char_indices()
            .rev()
            .find_map(|(idx, ch)| (!is_scheme_char(ch)).then_some(idx + ch.len_utf8()))
            .unwrap_or(0);
        out.push_str(&rest[..scheme_start]);

        let authority_start = scheme_end + 3;
        let authority_end = rest[authority_start..]
            .find(|ch: char| ['/', '?', '#', ' ', '\t', '\r', '\n'].contains(&ch))
            .map(|idx| authority_start + idx)
            .unwrap_or(rest.len());
        let authority = &rest[authority_start..authority_end];
        if let Some(at) = authority.rfind('@') {
            out.push_str(&rest[scheme_start..authority_start]);
            out.push_str("<redacted>@");
            out.push_str(&authority[at + 1..]);
        } else {
            out.push_str(&rest[scheme_start..authority_end]);
        }
        rest = &rest[authority_end..];
    }
    out.push_str(rest);
    out
}

/// Minimal trace2 event-target support (`GIT_TRACE2_EVENT`).
///
/// Upstream's trace2 event target writes one JSON object per line to the file
/// named by `GIT_TRACE2_EVENT`. sley emits only the `data` events the test
/// suite asserts on (`test_trace2_data` greps for the contiguous
/// `"category":"...","key":"...","value":"..."` triple), with the same field
/// order trace2's `fn_data_fl` produces. Unset/unwritable targets are
/// silently ignored, like upstream's best-effort tracing.
pub mod trace2 {
    use std::fmt::Display;
    use std::fmt::Write as _;
    use std::io::Write;
    use std::path::PathBuf;

    fn escape_json(raw: &str) -> String {
        let mut out = String::with_capacity(raw.len());
        for ch in raw.chars() {
            match ch {
                '"' => out.push_str("\\\""),
                '\\' => out.push_str("\\\\"),
                '\n' => out.push_str("\\n"),
                '\t' => out.push_str("\\t"),
                ch if (ch as u32) < 0x20 => {
                    let _ = write!(out, "\\u{:04x}", ch as u32);
                }
                ch => out.push(ch),
            }
        }
        out
    }

    enum TraceTarget {
        Stderr,
        Path(String),
    }

    fn trace_target(var: &str) -> Option<TraceTarget> {
        let target = std::env::var_os(var)?.to_string_lossy().into_owned();
        match target.as_str() {
            "1" | "true" => Some(TraceTarget::Stderr),
            _ if target.starts_with('/') => Some(TraceTarget::Path(target)),
            _ => None,
        }
    }

    fn write_target(target: &TraceTarget, bytes: &[u8]) {
        match target {
            TraceTarget::Stderr => {
                let _ = std::io::stderr().write_all(bytes);
            }
            TraceTarget::Path(path) => {
                if let Ok(mut file) = std::fs::OpenOptions::new()
                    .create(true)
                    .append(true)
                    .open(path)
                {
                    let _ = file.write_all(bytes);
                }
            }
        }
    }

    fn append_to_target(var: &str, line: &str) {
        let Some(target) = trace_target(var) else {
            return;
        };
        write_target(&target, format!("{line}\n").as_bytes());
    }

    fn redact_enabled() -> bool {
        std::env::var("GIT_TRACE2_REDACT").map_or(true, |value| value != "0")
    }

    fn maybe_redact(raw: &str) -> String {
        if redact_enabled() {
            super::redact_url_for_display(raw)
        } else {
            raw.to_string()
        }
    }

    fn quote_arg(arg: &str) -> String {
        if !arg.is_empty()
            && !arg
                .chars()
                .any(|ch| ch.is_whitespace() || matches!(ch, '\'' | '"' | '\\'))
        {
            return arg.to_string();
        }
        let mut out = String::with_capacity(arg.len() + 2);
        out.push('\'');
        for ch in arg.chars() {
            if ch == '\'' {
                out.push_str("'\\''");
            } else {
                out.push(ch);
            }
        }
        out.push('\'');
        out
    }

    fn argv0() -> String {
        let Some(arg0) = std::env::args_os().next() else {
            return "sley".to_string();
        };
        let path = PathBuf::from(arg0);
        path.file_name()
            .map(|name| name.to_string_lossy().into_owned())
            .filter(|name| !name.is_empty())
            .unwrap_or_else(|| "sley".to_string())
    }

    fn render_argv(args: &[String]) -> String {
        let mut rendered = Vec::with_capacity(args.len() + 1);
        rendered.push(quote_arg(&argv0()));
        rendered.extend(args.iter().map(|arg| quote_arg(arg)));
        rendered.join(" ")
    }

    pub fn depth() -> usize {
        std::env::var("SLEY_TRACE2_DEPTH")
            .ok()
            .and_then(|value| value.parse().ok())
            .unwrap_or(0)
    }

    fn perf_line(depth: usize, event: &str, rest: &str) {
        append_to_target(
            "GIT_TRACE2_PERF",
            &format!("d{depth} | main | {event} |  |  |  |  | {rest}"),
        );
    }

    /// Create the trace2 targets when tracing is enabled, even if this command
    /// emits no data/region/perf events — git opens the `GIT_TRACE2_EVENT` and
    /// `GIT_TRACE2_PERF` files at startup, so consumers (and test cleanups that
    /// `rm` the file) can rely on their existence.
    pub fn touch() {
        for var in ["GIT_TRACE2", "GIT_TRACE2_EVENT", "GIT_TRACE2_PERF"] {
            let Some(target) = trace_target(var) else {
                continue;
            };
            if let TraceTarget::Path(path) = target {
                let _ = std::fs::OpenOptions::new()
                    .create(true)
                    .append(true)
                    .open(path);
            }
        }
    }

    /// Emit the small normal/perf `start` records that downstream tools commonly
    /// use for argv auditing. Full trace2 lifecycle modelling remains out of
    /// scope; these records intentionally cover the stable clone/status tests.
    pub fn start(args: &[String]) {
        let argv = maybe_redact(&render_argv(args));
        append_to_target("GIT_TRACE2", &format!("start {argv}"));
        perf_line(depth(), "start", &argv);
    }

    pub fn cmd_ancestry_at_depth(depth: usize, ancestry: &[String]) {
        if ancestry.is_empty() {
            return;
        }
        append_to_target(
            "GIT_TRACE2",
            &format!("cmd_ancestry {}", ancestry.join(" <- ")),
        );
        perf_line(
            depth,
            "cmd_ancestry",
            &format!("ancestry:[{}]", ancestry.join(" ")),
        );
        let event_ancestry = ancestry
            .iter()
            .map(|name| format!("\"{}\"", escape_json(name)))
            .collect::<Vec<_>>()
            .join(",");
        append_to_target(
            "GIT_TRACE2_EVENT",
            &format!(
                "{{\"event\":\"cmd_ancestry\",\"sid\":\"sley\",\"thread\":\"main\",\"ancestry\":[{event_ancestry}]}}"
            ),
        );
    }

    pub fn cmd_name(name: &str, hierarchy: Option<&str>) {
        let rest = match hierarchy {
            Some(hierarchy) => format!("{name} ({hierarchy})"),
            None => name.to_string(),
        };
        perf_line(depth(), "cmd_name", &rest);
    }

    pub fn cmd_name_at_depth(depth: usize, name: &str, hierarchy: Option<&str>) {
        let rest = match hierarchy {
            Some(hierarchy) => format!("{name} ({hierarchy})"),
            None => name.to_string(),
        };
        perf_line(depth, "cmd_name", &rest);
    }

    pub fn child_start(class: &str, argv: &[String]) {
        child_start_with_id(class, 0, argv);
    }

    /// Record the start of a particular child/worker queue consumer.
    ///
    /// Checkout uses stable ids for each real materialization worker.  The
    /// normal target intentionally includes Git's `child_start[N]` spelling;
    /// upstream's parallel-checkout probes use that record to count workers.
    pub fn child_start_with_id(class: &str, child_id: usize, argv: &[String]) {
        let redacted: Vec<String> = argv.iter().map(|arg| maybe_redact(arg)).collect();
        let joined = redacted.join(" ");
        perf_line(
            depth(),
            "child_start",
            &format!("child_id:{child_id} class:{class} argv:[{joined}]"),
        );
        append_to_target("GIT_TRACE2", &format!("child_start[{child_id}] {joined}"));
        if let Some(target) = trace_target("GIT_TRACE2_EVENT") {
            let json_argv = redacted
                .iter()
                .map(|arg| format!("\"{}\"", escape_json(arg)))
                .collect::<Vec<_>>()
                .join(",");
            let line = format!(
                "{{\"event\":\"child_start\",\"sid\":\"sley\",\"thread\":\"main\",\"child_id\":{child_id},\"child_class\":\"{}\",\"use_shell\":false,\"argv\":[{json_argv}]}}\n",
                escape_json(class)
            );
            write_target(&target, line.as_bytes());
        }
    }

    pub fn alias(name: &str, argv: &[String]) {
        let argv = argv
            .iter()
            .map(|arg| maybe_redact(arg))
            .collect::<Vec<_>>()
            .join(" ");
        perf_line(depth(), "alias", &format!("alias:{name} argv:[{argv}]"));
    }

    /// Emit a trace2 config-parameter record to the normal and perf targets.
    pub fn def_param(key: &str, value: impl Display) {
        def_param_at_depth(depth(), key, value);
    }

    pub fn def_param_at_depth(depth: usize, key: &str, value: impl Display) {
        let value = value.to_string();
        let normal = maybe_redact(&format!("{key}={value}"));
        append_to_target("GIT_TRACE2", &format!("def_param {normal}"));
        let perf = maybe_redact(&format!("{key}:{value}"));
        perf_line(depth, "def_param", &perf);
    }

    /// Emit a trace2 `data` event (upstream `trace2_data_string` /
    /// `trace2_data_intmax`): a JSON line appended to the `GIT_TRACE2_EVENT`
    /// file when that target is enabled.
    pub fn data(category: &str, key: &str, value: impl Display) {
        let Some(target) = trace_target("GIT_TRACE2_EVENT") else {
            return;
        };
        let line = format!(
            "{{\"event\":\"data\",\"sid\":\"sley\",\"thread\":\"main\",\"nesting\":1,\"category\":\"{}\",\"key\":\"{}\",\"value\":\"{}\"}}\n",
            escape_json(category),
            escape_json(key),
            escape_json(&value.to_string()),
        );
        write_target(&target, line.as_bytes());
    }

    /// Emit a trace2 `counter` event. Git writes these for accumulated counters
    /// such as fsync hardware flushes when the event target is enabled.
    pub fn counter(category: &str, name: &str, count: impl Display) {
        let Some(target) = trace_target("GIT_TRACE2_EVENT") else {
            return;
        };
        let line = format!(
            "{{\"event\":\"counter\",\"sid\":\"sley\",\"thread\":\"main\",\"category\":\"{}\",\"name\":\"{}\",\"count\":{}}}\n",
            escape_json(category),
            escape_json(name),
            count,
        );
        write_target(&target, line.as_bytes());
    }

    /// Emit a trace2 region enter/leave pair. This is the minimal event shape
    /// Git's `test_region` helper greps for when asserting sparse-index
    /// expansion and conversion behaviour.
    pub fn region(category: &str, label: &str) {
        region_event("region_enter", category, label);
        region_event("region_leave", category, label);
    }

    fn region_event(event: &str, category: &str, label: &str) {
        let Some(target) = trace_target("GIT_TRACE2_EVENT") else {
            return;
        };
        let line = format!(
            "{{\"event\":\"{}\",\"sid\":\"sley\",\"thread\":\"main\",\"nesting\":1,\"category\":\"{}\",\"label\":\"{}\"}}\n",
            escape_json(event),
            escape_json(category),
            escape_json(label),
        );
        write_target(&target, line.as_bytes());
    }

    /// Emit the trace2 perf payload used by Git's changed-path Bloom filter
    /// tests. This intentionally writes only the grep-stable statistics string.
    pub fn bloom_statistics(
        filter_not_present: usize,
        maybe: usize,
        definitely_not: usize,
        false_positive: usize,
    ) {
        let Some(target) = trace_target("GIT_TRACE2_PERF") else {
            return;
        };
        let line = format!(
            "statistics:{{\"filter_not_present\":{filter_not_present},\"maybe\":{maybe},\"definitely_not\":{definitely_not},\"false_positive\":{false_positive}}}\n"
        );
        write_target(&target, line.as_bytes());
    }

    /// Emit a compact trace2 perf `data` row for tests that extract the
    /// read-directory statistics with pipe-field parsing.
    pub fn perf_read_directory_data(key: &str, value: impl Display) {
        let Some(target) = trace_target("GIT_TRACE2_PERF") else {
            return;
        };
        let line = format!(
            "19:00:00.000000 file.c:1 | d0 | main | data | r1 | ? | ? | read_directory | ....{key}:{value}\n"
        );
        write_target(&target, line.as_bytes());
    }

    /// Emit a trace2 perf `data` row tagged to the `setup` category (git's
    /// `trace2_data_string("setup", ...)`), used for the
    /// `implicit-bare-repository:<dir>` marker the safe.bareRepository tests
    /// grep for. Only the grep-stable `<key>:<value>` tail is significant.
    pub fn perf_setup_data(key: &str, value: impl Display) {
        let Some(target) = trace_target("GIT_TRACE2_PERF") else {
            return;
        };
        let line = format!(
            "19:00:00.000000 setup.c:1 | d0 | main | data | r0 | ? | ? | setup | ....{key}:{value}\n"
        );
        write_target(&target, line.as_bytes());
    }
}

#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash, PartialOrd, Ord)]
pub enum ObjectFormat {
    Sha1,
    Sha256,
}

impl ObjectFormat {
    pub const fn raw_len(self) -> usize {
        match self {
            Self::Sha1 => 20,
            Self::Sha256 => 32,
        }
    }

    pub const fn hex_len(self) -> usize {
        self.raw_len() * 2
    }

    pub const fn name(self) -> &'static str {
        match self {
            Self::Sha1 => "sha1",
            Self::Sha256 => "sha256",
        }
    }
}

impl FromStr for ObjectFormat {
    type Err = GitError;

    fn from_str(value: &str) -> Result<Self> {
        match value {
            "sha1" => Ok(Self::Sha1),
            "sha256" => Ok(Self::Sha256),
            other => Err(GitError::Unsupported(format!("object format {other}"))),
        }
    }
}

#[derive(Clone, Copy, PartialEq, Eq, Hash, PartialOrd, Ord)]
pub struct ObjectId {
    format: ObjectFormat,
    bytes: [u8; 32],
}

impl ObjectId {
    pub fn from_raw(format: ObjectFormat, raw: &[u8]) -> Result<Self> {
        if raw.len() != format.raw_len() {
            return Err(GitError::InvalidObjectId(format!(
                "expected {} bytes for {}, got {}",
                format.raw_len(),
                format.name(),
                raw.len()
            )));
        }
        let mut bytes = [0; 32];
        bytes[..raw.len()].copy_from_slice(raw);
        Ok(Self { format, bytes })
    }

    pub fn from_hex(format: ObjectFormat, hex: &str) -> Result<Self> {
        if hex.len() != format.hex_len() {
            return Err(GitError::InvalidObjectId(format!(
                "expected {} hex digits for {}, got {}",
                format.hex_len(),
                format.name(),
                hex.len()
            )));
        }
        let mut raw = [0; 32];
        for (i, pair) in hex.as_bytes().chunks_exact(2).enumerate() {
            raw[i] = (hex_nibble(pair[0])? << 4) | hex_nibble(pair[1])?;
        }
        Ok(Self { format, bytes: raw })
    }

    pub const fn format(&self) -> ObjectFormat {
        self.format
    }

    pub fn as_bytes(&self) -> &[u8] {
        &self.bytes[..self.format.raw_len()]
    }

    pub fn to_hex(&self) -> String {
        let mut out = String::with_capacity(self.format.hex_len());
        let _ = self.write_hex(&mut out);
        out
    }

    pub fn write_hex(&self, out: &mut impl fmt::Write) -> fmt::Result {
        write_hex_bytes(self.as_bytes(), out)
    }

    pub fn hex_prefix_matches(&self, prefix: &[u8]) -> bool {
        if prefix.len() > self.format.hex_len() {
            return false;
        }

        prefix.iter().enumerate().all(|(index, expected)| {
            let Some(expected) = hex_nibble_value(*expected) else {
                return false;
            };
            let byte = self.as_bytes()[index / 2];
            let actual = if index % 2 == 0 {
                byte >> 4
            } else {
                byte & 0x0f
            };
            actual == expected
        })
    }

    pub const fn abbrev_hex_len(&self, width: usize) -> usize {
        let hex_len = self.format.hex_len();
        if width < hex_len { width } else { hex_len }
    }

    /// The all-zero ("null") object id for `format`.
    pub fn null(format: ObjectFormat) -> Self {
        Self {
            format,
            bytes: [0; 32],
        }
    }

    /// True when every byte is zero (the null oid).
    pub fn is_null(&self) -> bool {
        self.as_bytes().iter().all(|byte| *byte == 0)
    }

    /// The id of the canonical empty tree for `format` (`4b825dc6…` for SHA-1).
    pub fn empty_tree(format: ObjectFormat) -> Self {
        Self::digest_object(format, "tree", b"")
    }

    /// The id of the canonical empty blob for `format` (`e69de29b…` for SHA-1).
    pub fn empty_blob(format: ObjectFormat) -> Self {
        Self::digest_object(format, "blob", b"")
    }

    /// Hash `"<type> <len>\0<body>"` straight into an id, bypassing the
    /// fallible length check in [`ObjectId::from_raw`] (our own digests are
    /// always the right length) so the well-known constants stay infallible.
    fn digest_object(format: ObjectFormat, object_type: &str, body: &[u8]) -> Self {
        let mut framed = Vec::with_capacity(object_type.len() + body.len() + 32);
        framed.extend_from_slice(object_type.as_bytes());
        framed.push(b' ');
        framed.extend_from_slice(body.len().to_string().as_bytes());
        framed.push(0);
        framed.extend_from_slice(body);
        let mut bytes = [0u8; 32];
        match format {
            ObjectFormat::Sha1 => bytes[..20].copy_from_slice(&sha1(&framed)),
            ObjectFormat::Sha256 => bytes[..32].copy_from_slice(&sha256(&framed)),
        }
        Self { format, bytes }
    }
}

impl fmt::Debug for ObjectId {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        f.debug_tuple("ObjectId").field(&self.to_hex()).finish()
    }
}

impl fmt::Display for ObjectId {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        self.write_hex(f)
    }
}

impl FromStr for ObjectId {
    type Err = GitError;

    /// Parse a full hex id, inferring the hash from its length (40 hex digits =
    /// SHA-1, 64 = SHA-256).
    fn from_str(text: &str) -> Result<Self> {
        let format = match text.len() {
            40 => ObjectFormat::Sha1,
            64 => ObjectFormat::Sha256,
            other => {
                return Err(GitError::InvalidObjectId(format!(
                    "expected 40 or 64 hex digits, got {other}"
                )));
            }
        };
        Self::from_hex(format, text)
    }
}

#[derive(Debug, Clone, PartialEq, Eq)]
pub struct ByteString(Vec<u8>);

impl ByteString {
    pub fn new(bytes: impl Into<Vec<u8>>) -> Self {
        Self(bytes.into())
    }

    pub fn as_bytes(&self) -> &[u8] {
        &self.0
    }
}

impl From<&str> for ByteString {
    fn from(value: &str) -> Self {
        Self(value.as_bytes().to_vec())
    }
}

/// A validated git ref name (e.g. `refs/heads/main`, `HEAD`).
#[derive(Clone, PartialEq, Eq, Hash, PartialOrd, Ord)]
pub struct FullName(String);

impl FullName {
    /// Construct a ref name, rejecting empty names, ASCII control characters,
    /// leading/trailing whitespace, and consecutive slashes.
    pub fn new(name: impl AsRef<str>) -> Result<Self> {
        let name = name.as_ref();
        validate_full_name(name)?;
        Ok(Self(name.to_string()))
    }

    pub fn as_str(&self) -> &str {
        &self.0
    }
}

impl fmt::Debug for FullName {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        f.debug_tuple("FullName").field(&self.0).finish()
    }
}

impl fmt::Display for FullName {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        f.write_str(&self.0)
    }
}

impl From<FullName> for String {
    fn from(value: FullName) -> Self {
        value.0
    }
}

impl Borrow<str> for FullName {
    fn borrow(&self) -> &str {
        &self.0
    }
}

impl AsRef<str> for FullName {
    fn as_ref(&self) -> &str {
        &self.0
    }
}

impl TryFrom<&str> for FullName {
    type Error = GitError;

    fn try_from(value: &str) -> Result<Self> {
        Self::new(value)
    }
}

impl TryFrom<String> for FullName {
    type Error = GitError;

    fn try_from(value: String) -> Result<Self> {
        validate_full_name(&value)?;
        Ok(Self(value))
    }
}

impl PartialEq<&str> for FullName {
    fn eq(&self, other: &&str) -> bool {
        self.0 == *other
    }
}

impl PartialEq<FullName> for &str {
    fn eq(&self, other: &FullName) -> bool {
        *self == other.0
    }
}

fn validate_full_name(name: &str) -> Result<()> {
    if name.is_empty() {
        return Err(GitError::InvalidFormat("ref name must not be empty".into()));
    }
    if name.chars().next().is_some_and(|ch| ch.is_whitespace())
        || name.chars().last().is_some_and(|ch| ch.is_whitespace())
    {
        return Err(GitError::InvalidFormat(
            "ref name must not have leading or trailing whitespace".into(),
        ));
    }
    if name.contains("//") {
        return Err(GitError::InvalidFormat(
            "ref name must not contain consecutive slashes".into(),
        ));
    }
    if name.bytes().any(|byte| byte.is_ascii_control()) {
        return Err(GitError::InvalidFormat(
            "ref name must not contain control characters".into(),
        ));
    }
    Ok(())
}

/// A byte string for git paths and similar on-disk identifiers.
#[derive(Debug, Clone, Default, PartialEq, Eq, Hash, PartialOrd, Ord)]
pub struct BString(Vec<u8>);

impl BString {
    pub fn new(bytes: impl Into<Vec<u8>>) -> Self {
        Self(bytes.into())
    }
    pub fn from_bytes(bytes: &[u8]) -> Self {
        Self(bytes.to_vec())
    }
    pub fn as_bytes(&self) -> &[u8] {
        &self.0
    }
    pub fn len(&self) -> usize {
        self.0.len()
    }
    pub fn is_empty(&self) -> bool {
        self.0.is_empty()
    }
    pub fn into_bytes(self) -> Vec<u8> {
        self.0
    }
}

impl From<&str> for BString {
    fn from(v: &str) -> Self {
        Self::from_bytes(v.as_bytes())
    }
}
impl From<&[u8]> for BString {
    fn from(v: &[u8]) -> Self {
        Self::from_bytes(v)
    }
}
impl<const N: usize> From<&[u8; N]> for BString {
    fn from(v: &[u8; N]) -> Self {
        Self::from_bytes(v.as_slice())
    }
}
impl From<Vec<u8>> for BString {
    fn from(v: Vec<u8>) -> Self {
        Self(v)
    }
}
impl PartialEq<&[u8]> for BString {
    fn eq(&self, o: &&[u8]) -> bool {
        self.0.as_slice() == *o
    }
}
impl<const N: usize> PartialEq<&[u8; N]> for BString {
    fn eq(&self, o: &&[u8; N]) -> bool {
        self.as_bytes() == o.as_slice()
    }
}
impl PartialEq<BString> for &[u8] {
    fn eq(&self, o: &BString) -> bool {
        *self == o.as_bytes()
    }
}
impl<const N: usize> PartialEq<BString> for &[u8; N] {
    fn eq(&self, o: &BString) -> bool {
        self.as_slice() == o.as_bytes()
    }
}
impl PartialEq<Vec<u8>> for BString {
    fn eq(&self, o: &Vec<u8>) -> bool {
        self.0 == *o
    }
}
impl PartialEq<BString> for Vec<u8> {
    fn eq(&self, o: &BString) -> bool {
        *self == o.0
    }
}

impl fmt::Display for BString {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        write!(f, "{}", String::from_utf8_lossy(&self.0))
    }
}

impl Borrow<[u8]> for BString {
    fn borrow(&self) -> &[u8] {
        self.as_bytes()
    }
}

impl Deref for BString {
    type Target = [u8];

    fn deref(&self) -> &[u8] {
        self.as_bytes()
    }
}

impl AsRef<[u8]> for BString {
    fn as_ref(&self) -> &[u8] {
        self.as_bytes()
    }
}

#[derive(Debug, Clone, PartialEq, Eq, Hash)]
pub struct RepoPath(PathBuf);

impl RepoPath {
    pub fn new(path: impl Into<PathBuf>) -> Result<Self> {
        let path = path.into();
        if path.is_absolute() {
            return Err(GitError::InvalidPath(
                "repository paths must be relative".into(),
            ));
        }
        if path.components().any(|component| {
            matches!(
                component,
                std::path::Component::ParentDir | std::path::Component::Prefix(_)
            )
        }) {
            return Err(GitError::InvalidPath(
                "repository paths must not escape".into(),
            ));
        }
        Ok(Self(path))
    }

    pub fn as_path(&self) -> &Path {
        &self.0
    }
}

/// A typed *parse-view* of a git identity line (`Name <email> <secs> <tz>`) as
/// found on a commit's `author`/`committer` or a tag's `tagger` header.
///
/// This is a read-only lens over bytes that are stored and re-serialized
/// verbatim elsewhere (see [`Signature::raw`]). It exists so callers can read
/// the typed `name`/`email`/`time` of an identity without re-implementing git's
/// ident-splitting rules, *not* as a storage format: the object model keeps the
/// original raw bytes as its source of truth, and round-tripping through this
/// view is byte-exact precisely because the raw line is retained alongside the
/// parsed fields (see [`Signature::to_ident_bytes`]).
///
/// Parse one with [`Signature::from_ident_line`]. The `time`'s timezone
/// preserves git's distinction between `+0000` (UTC) and `-0000` (a sentinel git
/// writes to mean "timezone unknown"); see [`GitTime`].
#[derive(Debug, Clone, PartialEq, Eq)]
pub struct Signature {
    /// The identity's name: the bytes before the ` <` that opens the email,
    /// with one trailing space (the separator) removed. May be empty.
    pub name: ByteString,
    /// The identity's email: the bytes between the `<` and `>` delimiters. May
    /// be empty.
    pub email: ByteString,
    /// The commit/authorship time and its timezone offset.
    pub time: GitTime,
    /// The exact original ident-line bytes this view was parsed from, retained
    /// so [`Signature::to_ident_bytes`] can reproduce the input byte-for-byte
    /// regardless of any non-canonical whitespace or formatting it contained.
    pub raw: Vec<u8>,
}

impl Signature {
    /// Parse a raw git identity line (`Name <email> <unix-secs> <tz>`) into a
    /// typed view, returning `None` when the bytes do not form a well-formed
    /// identity.
    ///
    /// The splitting mirrors git's own `split_ident_line`: the email is the run
    /// of bytes between the last `<` and the first following `>`; the name is
    /// everything before that `<` (one separating space is dropped); after the
    /// `>` come a space, the decimal Unix timestamp, a space, and the timezone
    /// token. The name and email may legitimately be empty, but a missing
    /// `<`/`>` pair, a non-numeric timestamp, or a malformed timezone token all
    /// yield `None` rather than a lossy guess — this is a *best-effort* parse
    /// that never panics. The original bytes are retained in
    /// [`Signature::raw`] so the parsed view re-serializes byte-identically.
    pub fn from_ident_line(line: &[u8]) -> Option<Self> {
        // Email is delimited by the last '<' whose matching '>' follows it, the
        // way git scans an ident from the right. Find the last '>' first, then
        // the last '<' before it.
        let mail_end = line.iter().rposition(|byte| *byte == b'>')?;
        let mail_begin = line[..mail_end].iter().rposition(|byte| *byte == b'<')? + 1;
        let email = &line[mail_begin..mail_end];

        // The name is everything before the '<', with a single trailing space
        // (the separator git inserts) trimmed if present.
        let mut name_end = mail_begin.saturating_sub(1);
        if name_end > 0 && line[name_end - 1] == b' ' {
            name_end -= 1;
        }
        let name = &line[..name_end];

        // After '>' git expects "<space><secs><space><tz>". Trim the single
        // separating space, then split the timestamp from the timezone token.
        let rest = line.get(mail_end + 1..)?;
        let rest = rest.strip_prefix(b" ")?;
        let time = GitTime::from_time_fields(rest)?;

        Some(Self {
            name: ByteString::new(name.to_vec()),
            email: ByteString::new(email.to_vec()),
            time,
            raw: line.to_vec(),
        })
    }

    /// Reproduce the original identity-line bytes.
    ///
    /// This returns [`Signature::raw`] verbatim, so for any line that
    /// [`Signature::from_ident_line`] accepted, `from_ident_line(line)?
    /// .to_ident_bytes() == line` holds byte-for-byte — including the `-0000`
    /// timezone and any non-canonical spacing the source contained.
    pub fn to_ident_bytes(&self) -> Vec<u8> {
        self.raw.clone()
    }

    /// Re-derive the canonical ident line from the parsed fields alone
    /// (`name <email> secs tz`), ignoring [`Signature::raw`].
    ///
    /// For an identity in git's canonical form this equals
    /// [`Signature::to_ident_bytes`]; it differs only when the source line
    /// carried non-canonical whitespace. Callers wanting byte-exact
    /// reproduction should use [`Signature::to_ident_bytes`]; this is provided
    /// for constructing a normalized line from typed parts.
    pub fn to_canonical_ident_bytes(&self) -> Vec<u8> {
        let mut out = Vec::with_capacity(self.raw.len());
        out.extend_from_slice(self.name.as_bytes());
        out.extend_from_slice(b" <");
        out.extend_from_slice(self.email.as_bytes());
        out.extend_from_slice(b"> ");
        out.extend_from_slice(self.time.to_ident_suffix().as_bytes());
        out
    }
}

/// A tolerant parse-view of a git identity line split git's way (ident.c's
/// `split_ident_line`). Unlike [`Signature::from_ident_line`] — which is a
/// strict, byte-exact round-trip parser — this mirrors how git's pretty-printer
/// recovers fields from *broken* idents: the email is the run between the
/// **first** `<` and the **first** following `>`, while the timestamp is located
/// by scanning **backwards** from the end of the line for the **last** `>`. That
/// split lets a corrupt ident like `Name <a@b>-<> 123 +0000` still surrender the
/// correct name (`Name`), email (`a@b`), and date (`123 +0000`).
pub struct IdentFields<'a> {
    /// Everything before the first `<`, with one trailing separator space removed.
    pub name: &'a [u8],
    /// The bytes between the first `<` and the first following `>`.
    pub email: &'a [u8],
    /// The decimal timestamp digit-run, or `None` when the line has no parseable
    /// `<digits> <±digits>` date tail (git's "person only" case).
    pub date: Option<&'a [u8]>,
    /// The timezone token (`±` plus digits), present iff `date` is.
    pub tz: Option<&'a [u8]>,
}

/// True for the whitespace bytes git's `isspace` recognizes (space, tab,
/// newline, carriage return). This deliberately excludes vertical tab (`0x0b`)
/// and form feed (`0x0c`), matching git's `sane_ctype` table — the distinction
/// that makes a vertical-tab-only date a sentinel rather than valid whitespace.
fn ident_isspace(byte: u8) -> bool {
    matches!(byte, b' ' | b'\t' | b'\n' | b'\r')
}

/// Split a git identity line the way ident.c's `split_ident_line` does,
/// returning `None` only when the line has no `<` or no following `>` (git's
/// `status < 0`). The date/timezone fields are `None` for the "person only"
/// case where no valid timestamp follows the final `>`.
pub fn split_ident_line(line: &[u8]) -> Option<IdentFields<'_>> {
    let len = line.len();
    // mail_begin: just past the first '<'.
    let lt = line.iter().position(|&byte| byte == b'<')?;
    let mail_begin = lt + 1;

    // name_end: the last non-space byte before '<' (git scans down from
    // mail_begin-2); default to the '<' position when only spaces precede it.
    let mut name_end = mail_begin - 1;
    if mail_begin >= 2 {
        let mut i = mail_begin - 2;
        loop {
            if !ident_isspace(line[i]) {
                name_end = i + 1;
                break;
            }
            if i == 0 {
                break;
            }
            i -= 1;
        }
    }
    let name = &line[..name_end];

    // mail_end: first '>' at or after mail_begin.
    let gt = line[mail_begin..].iter().position(|&byte| byte == b'>')? + mail_begin;
    let email = &line[mail_begin..gt];

    let person_only = IdentFields {
        name,
        email,
        date: None,
        tz: None,
    };

    // Date: scan from the end of the line for the LAST '>', then parse a
    // "<digits> <±digits>" tail after it (git assumes the timestamp has no '>').
    let mut cp = len - 1;
    while line[cp] != b'>' {
        if cp == 0 {
            return Some(person_only);
        }
        cp -= 1;
    }
    let mut i = cp + 1;
    while i < len && ident_isspace(line[i]) {
        i += 1;
    }
    let date_begin = i;
    while i < len && line[i].is_ascii_digit() {
        i += 1;
    }
    if i == date_begin {
        return Some(person_only);
    }
    let date = &line[date_begin..i];

    while i < len && ident_isspace(line[i]) {
        i += 1;
    }
    if i >= len || (line[i] != b'+' && line[i] != b'-') {
        return Some(person_only);
    }
    let tz_begin = i;
    i += 1;
    let tz_digits = i;
    while i < len && line[i].is_ascii_digit() {
        i += 1;
    }
    if i == tz_digits {
        return Some(person_only);
    }
    Some(IdentFields {
        name,
        email,
        date: Some(date),
        tz: Some(&line[tz_begin..i]),
    })
}

/// True when a timestamp is too large to be a valid `time_t`, mirroring git's
/// `date_overflows` for a 64-bit signed `time_t`.
fn ident_date_overflows(seconds: u64) -> bool {
    seconds >= i64::MAX as u64
}

/// Render an ident's date the way pretty.c's `show_ident_date` does: parse the
/// timestamp (git's `parse_timestamp` is unsigned/base-10 and clamps on
/// overflow), substitute the epoch sentinel (`time = 0`, timezone `+0000`) when
/// the value overflows what a `time_t` can hold, then format per `mode`. `date`
/// is the timestamp digit-run and `tz` its timezone token (as returned by
/// [`split_ident_line`]).
pub fn ident_render_date(date: &[u8], tz: &[u8], mode: &DateMode) -> String {
    let parsed = std::str::from_utf8(date)
        .ok()
        .and_then(|text| text.parse::<u64>().ok());
    let (seconds, tz_text) = match parsed {
        Some(value) if !ident_date_overflows(value) => {
            (value as i64, std::str::from_utf8(tz).unwrap_or("+0000"))
        }
        // Overflow, or a digit-run too long for u64: the epoch sentinel with a
        // forced `+0000` timezone, exactly like git's show_ident_date.
        _ => (0, "+0000"),
    };
    mode.render(seconds, tz_text).unwrap_or_default()
}

impl fmt::Display for Signature {
    /// Renders the original ident line (lossy only for bytes that are not valid
    /// UTF-8, which are replaced with `U+FFFD`). Use
    /// [`Signature::to_ident_bytes`] for the exact bytes.
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        write!(f, "{}", String::from_utf8_lossy(&self.raw))
    }
}

/// A git timestamp: a Unix time plus the committer's timezone offset.
///
/// The offset is stored as signed minutes east of UTC ([`timezone_offset_minutes`])
/// *and* a separate [`negative_utc`] flag. The flag exists because git
/// distinguishes the timezone token `-0000` from `+0000`: both are zero minutes
/// from UTC, but git writes `-0000` as a sentinel meaning "timezone unknown"
/// (e.g. for dates parsed without zone information), and that distinction is
/// part of a commit's byte-exact identity. `timezone_offset_minutes` alone
/// cannot represent it, so `negative_utc` carries the sign of a zero offset.
///
/// [`timezone_offset_minutes`]: GitTime::timezone_offset_minutes
/// [`negative_utc`]: GitTime::negative_utc
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub struct GitTime {
    /// Seconds since the Unix epoch.
    pub seconds: i64,
    /// Timezone offset east of UTC, in minutes (e.g. `+0530` -> `330`,
    /// `-0500` -> `-300`). Zero for both `+0000` and `-0000`; consult
    /// [`GitTime::negative_utc`] to tell those apart.
    pub timezone_offset_minutes: i16,
    /// `true` only when the timezone token had a negative sign with a zero
    /// magnitude (`-0000`), git's "timezone unknown" sentinel. Always `false`
    /// for any non-zero offset.
    pub negative_utc: bool,
}

impl GitTime {
    /// A `GitTime` with the given seconds and minute offset, treating a zero
    /// offset as the ordinary `+0000` (not the `-0000` sentinel). Use
    /// [`GitTime::with_negative_utc`] to construct the `-0000` case.
    pub const fn new(seconds: i64, timezone_offset_minutes: i16) -> Self {
        Self {
            seconds,
            timezone_offset_minutes,
            negative_utc: false,
        }
    }

    /// A `GitTime` whose timezone is the `-0000` sentinel ("timezone unknown").
    /// The minute offset is zero; `negative_utc` is `true`.
    pub const fn with_negative_utc(seconds: i64) -> Self {
        Self {
            seconds,
            timezone_offset_minutes: 0,
            negative_utc: true,
        }
    }

    /// Parse the `<secs> <tz>` tail of an ident line (the bytes after the
    /// `"> "` separating the email from the time), returning `None` if either
    /// field is malformed.
    fn from_time_fields(bytes: &[u8]) -> Option<Self> {
        let text = std::str::from_utf8(bytes).ok()?;
        let (seconds_text, tz_text) = text.split_once(' ')?;
        let seconds = seconds_text.parse::<i64>().ok()?;
        let (timezone_offset_minutes, negative_utc) = parse_timezone_token(tz_text)?;
        Some(Self {
            seconds,
            timezone_offset_minutes,
            negative_utc,
        })
    }

    /// The canonical `<secs> <±HHMM>` rendering of this time, as git writes it.
    /// Preserves the `-0000` sentinel.
    fn to_ident_suffix(self) -> String {
        format!("{} {}", self.seconds, self.offset_token())
    }

    /// The canonical 5-character timezone token for this offset (sign plus four
    /// digits), e.g. `+0000`, `-0500`, `+0530`. Returns `-0000` when
    /// [`GitTime::negative_utc`] is set.
    pub fn offset_token(self) -> String {
        let sign = if self.negative_utc || self.timezone_offset_minutes < 0 {
            '-'
        } else {
            '+'
        };
        let magnitude = self.timezone_offset_minutes.unsigned_abs();
        format!("{sign}{:02}{:02}", magnitude / 60, magnitude % 60)
    }
}

/// Parse a git timezone token (`±HHMM`) into `(minutes east of UTC, negative_utc)`.
///
/// Git accepts a leading `+`/`-` followed by four digits where the last two are
/// minutes. A negative sign with a zero magnitude (`-0000`) sets `negative_utc`.
/// Returns `None` for anything that is not a well-formed token.
fn parse_timezone_token(token: &str) -> Option<(i16, bool)> {
    let bytes = token.as_bytes();
    if bytes.len() != 5 {
        return None;
    }
    let negative = match bytes[0] {
        b'+' => false,
        b'-' => true,
        _ => return None,
    };
    if !bytes[1..].iter().all(u8::is_ascii_digit) {
        return None;
    }
    let hours = i16::from(bytes[1] - b'0') * 10 + i16::from(bytes[2] - b'0');
    let minutes = i16::from(bytes[3] - b'0') * 10 + i16::from(bytes[4] - b'0');
    let total = hours * 60 + minutes;
    let negative_utc = negative && total == 0;
    let signed = if negative { -total } else { total };
    Some((signed, negative_utc))
}

#[derive(Debug, Clone, PartialEq, Eq)]
pub struct Capability {
    pub name: String,
    pub value: Option<String>,
}

#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
pub enum MissingObjectKind {
    Object,
    Blob,
    Tree,
    Commit,
    Tag,
}

impl MissingObjectKind {
    pub const fn as_str(self) -> &'static str {
        match self {
            Self::Object => "object",
            Self::Blob => "blob",
            Self::Tree => "tree",
            Self::Commit => "commit",
            Self::Tag => "tag",
        }
    }
}

impl fmt::Display for MissingObjectKind {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        f.write_str(self.as_str())
    }
}

#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
pub enum MissingObjectContext {
    Read,
    Traversal,
    PackInstall,
    RevisionWalk,
    WorktreeMaterialize,
    RemoteBoundary,
}

impl MissingObjectContext {
    pub const fn as_str(self) -> &'static str {
        match self {
            Self::Read => "read",
            Self::Traversal => "traversal",
            Self::PackInstall => "pack-install",
            Self::RevisionWalk => "revision-walk",
            Self::WorktreeMaterialize => "worktree-materialize",
            Self::RemoteBoundary => "remote-boundary",
        }
    }
}

impl fmt::Display for MissingObjectContext {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        f.write_str(self.as_str())
    }
}

#[derive(Debug, Clone, PartialEq, Eq)]
pub enum NotFoundKind {
    Message(String),
    Remote {
        name: String,
    },
    Object {
        oid: ObjectId,
        kind: MissingObjectKind,
        context: Option<MissingObjectContext>,
    },
    Reference {
        name: String,
    },
    BrokenReference {
        name: String,
        target: String,
    },
    Repository {
        path: String,
    },
}

impl fmt::Display for NotFoundKind {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        match self {
            Self::Message(msg) => write!(f, "{msg}"),
            Self::Remote { name } => write!(f, "remote {name}"),
            Self::Object {
                oid,
                kind: MissingObjectKind::Object,
                ..
            } => write!(f, "object {oid}"),
            Self::Object { oid, kind, .. } => write!(f, "{kind} object {oid}"),
            Self::Reference { name } => write!(f, "{name}"),
            Self::BrokenReference { name, target } => {
                write!(f, "broken reference {name} -> {target}")
            }
            Self::Repository { path } => write!(f, "{path}"),
        }
    }
}

impl NotFoundKind {
    pub fn object_id(&self) -> Option<ObjectId> {
        match self {
            Self::Object { oid, .. } => Some(*oid),
            _ => None,
        }
    }

    pub fn missing_object_kind(&self) -> Option<MissingObjectKind> {
        match self {
            Self::Object { kind, .. } => Some(*kind),
            _ => None,
        }
    }

    pub fn missing_object_context(&self) -> Option<MissingObjectContext> {
        match self {
            Self::Object { context, .. } => *context,
            _ => None,
        }
    }
}

/// Git-compatible CLI exit status. See `git help exit-code` for the upstream taxonomy.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum CliExit {
    /// Success (exit 0).
    Ok,
    /// User-facing fatal error (exit 128).
    UserError,
    /// Invalid usage / bad arguments (exit 129).
    Usage,
    /// Command-specific exit code (e.g. grep returning 1 when no matches).
    Custom(i32),
}

impl CliExit {
    pub const fn code(self) -> i32 {
        match self {
            Self::Ok => 0,
            Self::UserError => 128,
            Self::Usage => 129,
            Self::Custom(code) => code,
        }
    }
}

#[derive(Debug, Clone, PartialEq, Eq)]
pub enum GitError {
    Io(String),
    InvalidObjectId(String),
    InvalidObject(String),
    InvalidFormat(String),
    InvalidPath(String),
    Unsupported(String),
    NotFound(NotFoundKind),
    Transaction(String),
    Command(String),
    /// Typed CLI exit with a user-facing message printed by the binary entrypoint.
    Cli(CliExit, String),
    /// Legacy explicit exit code; the message (if any) was already printed by the command.
    Exit(i32),
}

pub type Result<T> = std::result::Result<T, GitError>;

impl fmt::Display for GitError {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        match self {
            Self::Io(msg) => write!(f, "io error: {msg}"),
            Self::InvalidObjectId(msg) => write!(f, "invalid object id: {msg}"),
            Self::InvalidObject(msg) => write!(f, "invalid object: {msg}"),
            Self::InvalidFormat(msg) => write!(f, "invalid format: {msg}"),
            Self::InvalidPath(msg) => write!(f, "invalid path: {msg}"),
            Self::Unsupported(msg) => write!(f, "unsupported: {msg}"),
            Self::NotFound(kind) => write!(f, "not found: {kind}"),
            Self::Transaction(msg) => write!(f, "transaction failed: {msg}"),
            Self::Command(msg) => write!(f, "command failed: {msg}"),
            Self::Cli(_, msg) => f.write_str(msg),
            Self::Exit(code) => write!(f, "exit {code}"),
        }
    }
}

impl Error for GitError {}

impl GitError {
    pub fn usage(msg: impl Into<String>) -> Self {
        Self::Cli(CliExit::Usage, msg.into())
    }

    pub fn user_error(msg: impl Into<String>) -> Self {
        Self::Cli(CliExit::UserError, msg.into())
    }

    pub fn cli_exit(kind: CliExit, msg: impl Into<String>) -> Self {
        Self::Cli(kind, msg.into())
    }

    pub fn cli_exit_code(&self) -> i32 {
        cli_exit_code(self)
    }

    pub fn not_found(msg: impl Into<String>) -> Self {
        Self::NotFound(NotFoundKind::Message(msg.into()))
    }

    pub fn remote_not_found(name: impl Into<String>) -> Self {
        Self::NotFound(NotFoundKind::Remote { name: name.into() })
    }

    pub fn object_not_found(oid: ObjectId) -> Self {
        Self::object_kind_not_found(oid, MissingObjectKind::Object)
    }

    pub fn object_kind_not_found(oid: ObjectId, kind: MissingObjectKind) -> Self {
        Self::NotFound(NotFoundKind::Object {
            oid,
            kind,
            context: None,
        })
    }

    pub fn object_not_found_in(oid: ObjectId, context: MissingObjectContext) -> Self {
        Self::object_kind_not_found_in(oid, MissingObjectKind::Object, context)
    }

    pub fn object_kind_not_found_in(
        oid: ObjectId,
        kind: MissingObjectKind,
        context: MissingObjectContext,
    ) -> Self {
        Self::NotFound(NotFoundKind::Object {
            oid,
            kind,
            context: Some(context),
        })
    }

    pub fn reference_not_found(name: impl Into<String>) -> Self {
        Self::NotFound(NotFoundKind::Reference { name: name.into() })
    }

    pub fn broken_reference(name: impl Into<String>, target: impl Into<String>) -> Self {
        Self::NotFound(NotFoundKind::BrokenReference {
            name: name.into(),
            target: target.into(),
        })
    }

    pub fn repository_not_found(path: impl Into<String>) -> Self {
        Self::NotFound(NotFoundKind::Repository { path: path.into() })
    }

    pub fn not_found_kind(&self) -> Option<&NotFoundKind> {
        match self {
            Self::NotFound(kind) => Some(kind),
            _ => None,
        }
    }
}

impl From<std::io::Error> for GitError {
    fn from(value: std::io::Error) -> Self {
        Self::Io(value.to_string())
    }
}

/// Map a [`GitError`] to the process exit code the CLI should use.
pub fn cli_exit_code(err: &GitError) -> i32 {
    match err {
        GitError::Exit(code) => *code,
        GitError::Cli(kind, _) => kind.code(),
        // During migration, usage-style validation still returns `Command`; treat as
        // general failure until those call sites adopt `GitError::usage`.
        GitError::Command(_) => 1,
        _ => 1,
    }
}

pub fn object_id_for_bytes(
    format: ObjectFormat,
    object_type: &str,
    body: &[u8],
) -> Result<ObjectId> {
    match format {
        // Hash the `"<type> <len>\0"` header and the body as separate updates so
        // the (potentially large) body is never copied into a combined buffer just
        // to feed the digest.
        ObjectFormat::Sha1 => ObjectId::from_raw(format, &sha1_object_digest(object_type, body)),
        ObjectFormat::Sha256 => {
            let mut framed = Vec::with_capacity(object_type.len() + body.len() + 32);
            framed.extend_from_slice(object_type.as_bytes());
            framed.push(b' ');
            framed.extend_from_slice(body.len().to_string().as_bytes());
            framed.push(0);
            framed.extend_from_slice(body);
            ObjectId::from_raw(format, &sha256(&framed))
        }
    }
}

pub fn digest_bytes(format: ObjectFormat, bytes: &[u8]) -> Result<ObjectId> {
    match format {
        ObjectFormat::Sha1 => ObjectId::from_raw(format, &sha1(bytes)),
        ObjectFormat::Sha256 => ObjectId::from_raw(format, &sha256(bytes)),
    }
}

pub struct StreamingDigest {
    format: ObjectFormat,
    inner: StreamingDigestInner,
}

enum StreamingDigestInner {
    #[cfg(not(feature = "fast-sha1"))]
    Sha1(Sha1Hasher),
    #[cfg(feature = "fast-sha1")]
    Sha1(sha1::Sha1),
    Sha256(Sha256Hasher),
}

impl StreamingDigest {
    pub fn new(format: ObjectFormat) -> Self {
        let inner = match format {
            #[cfg(not(feature = "fast-sha1"))]
            ObjectFormat::Sha1 => StreamingDigestInner::Sha1(Sha1Hasher::new()),
            #[cfg(feature = "fast-sha1")]
            ObjectFormat::Sha1 => {
                use sha1::Digest;
                StreamingDigestInner::Sha1(sha1::Sha1::new())
            }
            ObjectFormat::Sha256 => StreamingDigestInner::Sha256(Sha256Hasher::new()),
        };
        Self { format, inner }
    }

    pub fn update(&mut self, data: &[u8]) {
        match &mut self.inner {
            #[cfg(not(feature = "fast-sha1"))]
            StreamingDigestInner::Sha1(hasher) => hasher.update(data),
            #[cfg(feature = "fast-sha1")]
            StreamingDigestInner::Sha1(hasher) => {
                use sha1::Digest;
                hasher.update(data);
            }
            StreamingDigestInner::Sha256(hasher) => hasher.update(data),
        }
    }

    pub fn finalize(self) -> Result<ObjectId> {
        match self.inner {
            #[cfg(not(feature = "fast-sha1"))]
            StreamingDigestInner::Sha1(hasher) => {
                ObjectId::from_raw(self.format, &hasher.finalize())
            }
            #[cfg(feature = "fast-sha1")]
            StreamingDigestInner::Sha1(hasher) => {
                use sha1::Digest;
                let bytes: [u8; 20] = hasher.finalize().into();
                ObjectId::from_raw(self.format, &bytes)
            }
            StreamingDigestInner::Sha256(hasher) => {
                ObjectId::from_raw(self.format, &hasher.finalize())
            }
        }
    }
}

pub fn to_hex(bytes: &[u8]) -> String {
    let mut out = String::with_capacity(bytes.len() * 2);
    let _ = write_hex_bytes(bytes, &mut out);
    out
}

fn write_hex_bytes(bytes: &[u8], out: &mut impl fmt::Write) -> fmt::Result {
    const HEX: &[u8; 16] = b"0123456789abcdef";
    for byte in bytes {
        out.write_char(HEX[(byte >> 4) as usize] as char)?;
        out.write_char(HEX[(byte & 0x0f) as usize] as char)?;
    }
    Ok(())
}

fn hex_nibble_value(byte: u8) -> Option<u8> {
    match byte {
        b'0'..=b'9' => Some(byte - b'0'),
        b'a'..=b'f' => Some(byte - b'a' + 10),
        b'A'..=b'F' => Some(byte - b'A' + 10),
        _ => None,
    }
}

fn hex_nibble(byte: u8) -> Result<u8> {
    hex_nibble_value(byte)
        .ok_or_else(|| GitError::InvalidObjectId(format!("non-hex byte {:?}", byte as char)))
}

// ---------------------------------------------------------------------------
// SHA-1
//
// The default is a pure-Rust streaming implementation that hashes 64-byte blocks
// straight from the caller's slices, so neither the body nor the framed object is
// copied just to be digested. Enabling the `fast-sha1` feature swaps in the
// RustCrypto `sha1` crate, which dispatches to ARMv8-SHA1 / x86 SHA-NI at runtime;
// the digests are byte-identical, so OIDs are unchanged either way.
// ---------------------------------------------------------------------------

/// SHA-1 of a raw byte slice (already-framed object, bundle prerequisite, etc.).
#[cfg(not(feature = "fast-sha1"))]
fn sha1(input: &[u8]) -> [u8; 20] {
    let mut hasher = Sha1Hasher::new();
    hasher.update(input);
    hasher.finalize()
}

/// SHA-1 of a raw byte slice using the hardware-accelerated backend.
#[cfg(feature = "fast-sha1")]
fn sha1(input: &[u8]) -> [u8; 20] {
    use sha1::{Digest, Sha1};
    let mut hasher = Sha1::new();
    hasher.update(input);
    hasher.finalize().into()
}

/// SHA-1 of a git object framed as `"<type> <len>\0<body>"`, fed as separate
/// updates so the body is never copied into a combined buffer.
#[cfg(not(feature = "fast-sha1"))]
fn sha1_object_digest(object_type: &str, body: &[u8]) -> [u8; 20] {
    let mut hasher = Sha1Hasher::new();
    hasher.update(object_type.as_bytes());
    hasher.update(b" ");
    hasher.update(body.len().to_string().as_bytes());
    hasher.update(&[0u8]);
    hasher.update(body);
    hasher.finalize()
}

#[cfg(feature = "fast-sha1")]
fn sha1_object_digest(object_type: &str, body: &[u8]) -> [u8; 20] {
    use sha1::{Digest, Sha1};
    let mut hasher = Sha1::new();
    hasher.update(object_type.as_bytes());
    hasher.update(b" ");
    hasher.update(body.len().to_string().as_bytes());
    hasher.update([0u8]);
    hasher.update(body);
    hasher.finalize().into()
}

/// Streaming pure-Rust SHA-1: feeds full 64-byte blocks directly from each
/// `update` slice and buffers only the sub-block remainder, so large inputs are
/// hashed without an intermediate copy.
#[cfg(not(feature = "fast-sha1"))]
struct Sha1Hasher {
    state: [u32; 5],
    block: [u8; 64],
    block_len: usize,
    total_len: u64,
}

#[cfg(not(feature = "fast-sha1"))]
impl Sha1Hasher {
    fn new() -> Self {
        Self {
            state: [0x67452301, 0xefcdab89, 0x98badcfe, 0x10325476, 0xc3d2e1f0],
            block: [0u8; 64],
            block_len: 0,
            total_len: 0,
        }
    }

    fn update(&mut self, mut data: &[u8]) {
        self.total_len = self.total_len.wrapping_add(data.len() as u64);
        if self.block_len > 0 {
            let take = (64 - self.block_len).min(data.len());
            self.block[self.block_len..self.block_len + take].copy_from_slice(&data[..take]);
            self.block_len += take;
            data = &data[take..];
            if self.block_len == 64 {
                let block = self.block;
                sha1_compress(&mut self.state, &block);
                self.block_len = 0;
            }
        }
        while data.len() >= 64 {
            sha1_compress(&mut self.state, &data[..64]);
            data = &data[64..];
        }
        if !data.is_empty() {
            self.block[..data.len()].copy_from_slice(data);
            self.block_len = data.len();
        }
    }

    fn finalize(mut self) -> [u8; 20] {
        let bit_len = self.total_len.wrapping_mul(8);
        // 0x80, zero pad to a 56 mod 64 boundary, then the 64-bit big-endian length.
        // From a sub-block remainder this is at most two more blocks (128 bytes).
        let mut tail = [0u8; 128];
        tail[..self.block_len].copy_from_slice(&self.block[..self.block_len]);
        tail[self.block_len] = 0x80;
        let total = if self.block_len < 56 { 64 } else { 128 };
        tail[total - 8..total].copy_from_slice(&bit_len.to_be_bytes());
        sha1_compress(&mut self.state, &tail[..64]);
        if total == 128 {
            sha1_compress(&mut self.state, &tail[64..128]);
        }
        let mut out = [0u8; 20];
        out[0..4].copy_from_slice(&self.state[0].to_be_bytes());
        out[4..8].copy_from_slice(&self.state[1].to_be_bytes());
        out[8..12].copy_from_slice(&self.state[2].to_be_bytes());
        out[12..16].copy_from_slice(&self.state[3].to_be_bytes());
        out[16..20].copy_from_slice(&self.state[4].to_be_bytes());
        out
    }
}

/// Mix one 64-byte block into the SHA-1 state. `block` must be at least 64 bytes.
#[cfg(not(feature = "fast-sha1"))]
fn sha1_compress(state: &mut [u32; 5], block: &[u8]) {
    let mut w = [0u32; 80];
    for (i, word) in w.iter_mut().take(16).enumerate() {
        let offset = i * 4;
        *word = u32::from_be_bytes([
            block[offset],
            block[offset + 1],
            block[offset + 2],
            block[offset + 3],
        ]);
    }
    for i in 16..80 {
        w[i] = (w[i - 3] ^ w[i - 8] ^ w[i - 14] ^ w[i - 16]).rotate_left(1);
    }

    let mut a = state[0];
    let mut b = state[1];
    let mut c = state[2];
    let mut d = state[3];
    let mut e = state[4];

    for (i, word) in w.iter().enumerate() {
        let (f, k) = match i {
            0..=19 => ((b & c) | ((!b) & d), 0x5a827999u32),
            20..=39 => (b ^ c ^ d, 0x6ed9eba1),
            40..=59 => ((b & c) | (b & d) | (c & d), 0x8f1bbcdc),
            _ => (b ^ c ^ d, 0xca62c1d6),
        };
        let temp = a
            .rotate_left(5)
            .wrapping_add(f)
            .wrapping_add(e)
            .wrapping_add(k)
            .wrapping_add(*word);
        e = d;
        d = c;
        c = b.rotate_left(30);
        b = a;
        a = temp;
    }

    state[0] = state[0].wrapping_add(a);
    state[1] = state[1].wrapping_add(b);
    state[2] = state[2].wrapping_add(c);
    state[3] = state[3].wrapping_add(d);
    state[4] = state[4].wrapping_add(e);
}

fn sha256(input: &[u8]) -> [u8; 32] {
    let mut hasher = Sha256Hasher::new();
    hasher.update(input);
    hasher.finalize()
}

struct Sha256Hasher {
    state: [u32; 8],
    block: [u8; 64],
    block_len: usize,
    total_len: u64,
}

impl Sha256Hasher {
    const K: [u32; 64] = [
        0x428a2f98, 0x71374491, 0xb5c0fbcf, 0xe9b5dba5, 0x3956c25b, 0x59f111f1, 0x923f82a4,
        0xab1c5ed5, 0xd807aa98, 0x12835b01, 0x243185be, 0x550c7dc3, 0x72be5d74, 0x80deb1fe,
        0x9bdc06a7, 0xc19bf174, 0xe49b69c1, 0xefbe4786, 0x0fc19dc6, 0x240ca1cc, 0x2de92c6f,
        0x4a7484aa, 0x5cb0a9dc, 0x76f988da, 0x983e5152, 0xa831c66d, 0xb00327c8, 0xbf597fc7,
        0xc6e00bf3, 0xd5a79147, 0x06ca6351, 0x14292967, 0x27b70a85, 0x2e1b2138, 0x4d2c6dfc,
        0x53380d13, 0x650a7354, 0x766a0abb, 0x81c2c92e, 0x92722c85, 0xa2bfe8a1, 0xa81a664b,
        0xc24b8b70, 0xc76c51a3, 0xd192e819, 0xd6990624, 0xf40e3585, 0x106aa070, 0x19a4c116,
        0x1e376c08, 0x2748774c, 0x34b0bcb5, 0x391c0cb3, 0x4ed8aa4a, 0x5b9cca4f, 0x682e6ff3,
        0x748f82ee, 0x78a5636f, 0x84c87814, 0x8cc70208, 0x90befffa, 0xa4506ceb, 0xbef9a3f7,
        0xc67178f2,
    ];

    fn new() -> Self {
        Self {
            state: [
                0x6a09e667u32,
                0xbb67ae85,
                0x3c6ef372,
                0xa54ff53a,
                0x510e527f,
                0x9b05688c,
                0x1f83d9ab,
                0x5be0cd19,
            ],
            block: [0u8; 64],
            block_len: 0,
            total_len: 0,
        }
    }

    fn update(&mut self, mut data: &[u8]) {
        self.total_len = self.total_len.wrapping_add(data.len() as u64);
        if self.block_len > 0 {
            let take = (64 - self.block_len).min(data.len());
            self.block[self.block_len..self.block_len + take].copy_from_slice(&data[..take]);
            self.block_len += take;
            data = &data[take..];
            if self.block_len == 64 {
                let block = self.block;
                self.compress(&block);
                self.block_len = 0;
            }
        }
        while data.len() >= 64 {
            self.compress(&data[..64]);
            data = &data[64..];
        }
        if !data.is_empty() {
            self.block[..data.len()].copy_from_slice(data);
            self.block_len = data.len();
        }
    }

    fn finalize(mut self) -> [u8; 32] {
        let bit_len = self.total_len.wrapping_mul(8);
        let mut tail = [0u8; 128];
        tail[..self.block_len].copy_from_slice(&self.block[..self.block_len]);
        tail[self.block_len] = 0x80;
        let total = if self.block_len < 56 { 64 } else { 128 };
        tail[total - 8..total].copy_from_slice(&bit_len.to_be_bytes());
        self.compress(&tail[..64]);
        if total == 128 {
            self.compress(&tail[64..128]);
        }

        let mut out = [0; 32];
        for (idx, word) in self.state.iter().enumerate() {
            out[idx * 4..idx * 4 + 4].copy_from_slice(&word.to_be_bytes());
        }
        out
    }

    fn compress(&mut self, chunk: &[u8]) {
        let mut w = [0u32; 64];
        for (i, word) in w.iter_mut().take(16).enumerate() {
            let offset = i * 4;
            *word = u32::from_be_bytes([
                chunk[offset],
                chunk[offset + 1],
                chunk[offset + 2],
                chunk[offset + 3],
            ]);
        }
        for i in 16..64 {
            let s0 = w[i - 15].rotate_right(7) ^ w[i - 15].rotate_right(18) ^ (w[i - 15] >> 3);
            let s1 = w[i - 2].rotate_right(17) ^ w[i - 2].rotate_right(19) ^ (w[i - 2] >> 10);
            w[i] = w[i - 16]
                .wrapping_add(s0)
                .wrapping_add(w[i - 7])
                .wrapping_add(s1);
        }

        let mut a = self.state[0];
        let mut b = self.state[1];
        let mut c = self.state[2];
        let mut d = self.state[3];
        let mut e = self.state[4];
        let mut f = self.state[5];
        let mut g = self.state[6];
        let mut hh = self.state[7];

        for (&word, &constant) in w.iter().zip(Self::K.iter()) {
            let s1 = e.rotate_right(6) ^ e.rotate_right(11) ^ e.rotate_right(25);
            let ch = (e & f) ^ ((!e) & g);
            let temp1 = hh
                .wrapping_add(s1)
                .wrapping_add(ch)
                .wrapping_add(constant)
                .wrapping_add(word);
            let s0 = a.rotate_right(2) ^ a.rotate_right(13) ^ a.rotate_right(22);
            let maj = (a & b) ^ (a & c) ^ (b & c);
            let temp2 = s0.wrapping_add(maj);

            hh = g;
            g = f;
            f = e;
            e = d.wrapping_add(temp1);
            d = c;
            c = b;
            b = a;
            a = temp1.wrapping_add(temp2);
        }

        self.state[0] = self.state[0].wrapping_add(a);
        self.state[1] = self.state[1].wrapping_add(b);
        self.state[2] = self.state[2].wrapping_add(c);
        self.state[3] = self.state[3].wrapping_add(d);
        self.state[4] = self.state[4].wrapping_add(e);
        self.state[5] = self.state[5].wrapping_add(f);
        self.state[6] = self.state[6].wrapping_add(g);
        self.state[7] = self.state[7].wrapping_add(hh);
    }
}

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

    #[test]
    fn sha1_blob_matches_git_known_value() {
        let oid = object_id_for_bytes(ObjectFormat::Sha1, "blob", b"hello\n")
            .expect("known blob should hash as sha1");
        assert_eq!(oid.to_hex(), "ce013625030ba8dba906f756967f9e9ca394464a");
    }

    #[test]
    fn sha256_blob_matches_git_known_value() {
        let oid = object_id_for_bytes(ObjectFormat::Sha256, "blob", b"hello\n")
            .expect("known blob should hash as sha256");
        assert_eq!(
            oid.to_hex(),
            "2cf8d83d9ee29543b34a87727421fdecb7e3f3a183d337639025de576db9ebb4"
        );
    }

    #[test]
    fn object_id_round_trips_hex() {
        let oid = ObjectId::from_hex(
            ObjectFormat::Sha1,
            "ce013625030ba8dba906f756967f9e9ca394464a",
        )
        .expect("valid sha1 hex");
        assert_eq!(oid.to_hex(), "ce013625030ba8dba906f756967f9e9ca394464a");
    }

    #[test]
    fn object_id_writes_hex_without_allocating_in_the_writer() {
        let oid = ObjectId::from_hex(
            ObjectFormat::Sha1,
            "CE013625030BA8DBA906F756967F9E9CA394464A",
        )
        .expect("valid uppercase sha1 hex");

        let mut out = String::new();
        oid.write_hex(&mut out)
            .expect("writing object id hex to a String should not fail");

        assert_eq!(out, "ce013625030ba8dba906f756967f9e9ca394464a");
        assert_eq!(oid.to_hex(), out);
        assert_eq!(format!("{oid}"), out);
    }

    #[test]
    fn object_id_matches_hex_prefixes_by_nibble() {
        let oid = ObjectId::from_hex(
            ObjectFormat::Sha1,
            "ce013625030ba8dba906f756967f9e9ca394464a",
        )
        .expect("valid sha1 hex");

        assert!(oid.hex_prefix_matches(b""));
        assert!(oid.hex_prefix_matches(b"c"));
        assert!(oid.hex_prefix_matches(b"ce013"));
        assert!(oid.hex_prefix_matches(b"CE013625"));
        assert!(oid.hex_prefix_matches(b"ce013625030ba8dba906f756967f9e9ca394464a"));

        assert!(!oid.hex_prefix_matches(b"d"));
        assert!(!oid.hex_prefix_matches(b"ce014"));
        assert!(!oid.hex_prefix_matches(b"ce01x"));

        let mut too_long = oid.to_hex();
        too_long.push('0');
        assert!(!oid.hex_prefix_matches(too_long.as_bytes()));
    }

    #[test]
    fn object_id_abbrev_hex_len_clamps_to_format_width() {
        let sha1 = ObjectId::null(ObjectFormat::Sha1);
        let sha256 = ObjectId::null(ObjectFormat::Sha256);

        assert_eq!(sha1.abbrev_hex_len(0), 0);
        assert_eq!(sha1.abbrev_hex_len(12), 12);
        assert_eq!(sha1.abbrev_hex_len(80), ObjectFormat::Sha1.hex_len());
        assert_eq!(sha256.abbrev_hex_len(80), ObjectFormat::Sha256.hex_len());
    }

    #[test]
    fn signature_parses_a_normal_ident_and_round_trips() {
        let line = b"A U Thor <author@example.com> 1700000000 +0000";
        let sig = Signature::from_ident_line(line).expect("well-formed ident parses");
        assert_eq!(sig.name.as_bytes(), b"A U Thor");
        assert_eq!(sig.email.as_bytes(), b"author@example.com");
        assert_eq!(sig.time.seconds, 1_700_000_000);
        assert_eq!(sig.time.timezone_offset_minutes, 0);
        assert!(!sig.time.negative_utc);
        // Byte-exact round-trip, and the canonical form matches here too.
        assert_eq!(sig.to_ident_bytes(), line);
        assert_eq!(sig.to_canonical_ident_bytes(), line);
    }

    #[test]
    fn signature_parses_positive_half_hour_offset() {
        let line = b"Half Hour <hh@example.com> 1500000000 +0530";
        let sig = Signature::from_ident_line(line).expect("offset ident parses");
        assert_eq!(sig.time.timezone_offset_minutes, 330);
        assert!(!sig.time.negative_utc);
        assert_eq!(sig.time.offset_token(), "+0530");
        assert_eq!(sig.to_ident_bytes(), line);
        assert_eq!(sig.to_canonical_ident_bytes(), line);
    }

    #[test]
    fn signature_parses_negative_offset() {
        let line = b"Western <w@example.com> 1500000000 -0500";
        let sig = Signature::from_ident_line(line).expect("negative offset parses");
        assert_eq!(sig.time.timezone_offset_minutes, -300);
        assert!(!sig.time.negative_utc);
        assert_eq!(sig.time.offset_token(), "-0500");
        assert_eq!(sig.to_ident_bytes(), line);
    }

    #[test]
    fn signature_preserves_negative_zero_timezone_distinct_from_positive_zero() {
        let negative = b"Unknown Zone <uz@example.com> 1500000000 -0000";
        let positive = b"Known Zone <kz@example.com> 1500000000 +0000";

        let neg = Signature::from_ident_line(negative).expect("-0000 parses");
        let pos = Signature::from_ident_line(positive).expect("+0000 parses");

        // Both are zero minutes from UTC...
        assert_eq!(neg.time.timezone_offset_minutes, 0);
        assert_eq!(pos.time.timezone_offset_minutes, 0);
        // ...but the sentinel flag distinguishes them, so the times differ.
        assert!(neg.time.negative_utc);
        assert!(!pos.time.negative_utc);
        assert_ne!(neg.time, pos.time);

        // And the distinction survives re-serialization, byte-for-byte.
        assert_eq!(neg.time.offset_token(), "-0000");
        assert_eq!(pos.time.offset_token(), "+0000");
        assert_eq!(neg.to_ident_bytes(), negative);
        assert_eq!(pos.to_ident_bytes(), positive);
        assert_eq!(neg.to_canonical_ident_bytes(), negative);
        assert_eq!(pos.to_canonical_ident_bytes(), positive);
        assert_ne!(neg.to_ident_bytes(), pos.to_ident_bytes());
    }

    #[test]
    fn signature_handles_empty_name_and_email() {
        // git permits an empty name and/or empty email; the delimiters still
        // anchor the parse.
        let line = b" <> 0 +0000";
        let sig = Signature::from_ident_line(line).expect("empty name/email parses");
        assert_eq!(sig.name.as_bytes(), b"");
        assert_eq!(sig.email.as_bytes(), b"");
        assert_eq!(sig.time.seconds, 0);
        assert_eq!(sig.to_ident_bytes(), line);
    }

    #[test]
    fn signature_keeps_angle_brackets_inside_the_name() {
        // The email is delimited by the *last* '<'/'>' pair, so a name that
        // itself contains angle brackets parses with the trailing pair as the
        // email and round-trips exactly.
        let line = b"Weird <Name> <weird@example.com> 1 +0000";
        let sig = Signature::from_ident_line(line).expect("bracketed name parses");
        assert_eq!(sig.name.as_bytes(), b"Weird <Name>");
        assert_eq!(sig.email.as_bytes(), b"weird@example.com");
        assert_eq!(sig.to_ident_bytes(), line);
    }

    #[test]
    fn signature_round_trips_non_canonical_whitespace_via_raw() {
        // An ident with two spaces before the email is not git's canonical form,
        // but the parse-view must still reproduce it byte-for-byte from `raw`.
        // (Only the canonical renderer normalizes the spacing.)
        let line = b"Spaced  <spaced@example.com> 5 +0000";
        let sig = Signature::from_ident_line(line).expect("non-canonical ident parses");
        // The name keeps the extra space (only one separator space is trimmed).
        assert_eq!(sig.name.as_bytes(), b"Spaced ");
        assert_eq!(sig.to_ident_bytes(), line);
    }

    #[test]
    fn signature_rejects_malformed_idents() {
        // No email delimiters.
        assert!(Signature::from_ident_line(b"No Email Here 0 +0000").is_none());
        // Missing the time tail entirely.
        assert!(Signature::from_ident_line(b"A U Thor <a@example.com>").is_none());
        // Non-numeric timestamp.
        assert!(Signature::from_ident_line(b"A U Thor <a@example.com> later +0000").is_none());
        // Malformed timezone token (wrong width).
        assert!(Signature::from_ident_line(b"A U Thor <a@example.com> 0 +00").is_none());
        // Timezone token missing a sign.
        assert!(Signature::from_ident_line(b"A U Thor <a@example.com> 0 0000").is_none());
    }

    #[test]
    fn git_time_constructors_set_the_sentinel() {
        assert!(!GitTime::new(0, 0).negative_utc);
        assert_eq!(GitTime::new(0, 330).offset_token(), "+0530");
        let unknown = GitTime::with_negative_utc(42);
        assert!(unknown.negative_utc);
        assert_eq!(unknown.seconds, 42);
        assert_eq!(unknown.offset_token(), "-0000");
    }

    #[test]
    fn full_name_accepts_valid_ref_names() {
        let name = FullName::new("refs/heads/main").expect("valid ref name");
        assert_eq!(name.as_str(), "refs/heads/main");
        assert_eq!(name, "refs/heads/main");
        assert_eq!(format!("{name}"), "refs/heads/main");
        assert_eq!(String::from(name.clone()), "refs/heads/main");
        let borrowed: &str = name.borrow();
        assert_eq!(borrowed, "refs/heads/main");
    }

    #[test]
    fn full_name_rejects_invalid_ref_names() {
        assert!(FullName::new("").is_err());
        assert!(FullName::new(" refs/heads/main").is_err());
        assert!(FullName::new("refs/heads/main ").is_err());
        assert!(FullName::new("refs//heads/main").is_err());
        assert!(FullName::new("refs/heads/\nmain").is_err());
    }

    #[test]
    fn cli_exit_codes_match_git_taxonomy() {
        assert_eq!(CliExit::Ok.code(), 0);
        assert_eq!(CliExit::UserError.code(), 128);
        assert_eq!(CliExit::Usage.code(), 129);
        assert_eq!(CliExit::Custom(1).code(), 1);
        assert_eq!(CliExit::Custom(5).code(), 5);
    }

    #[test]
    fn git_error_cli_exit_code_mapping() {
        assert_eq!(GitError::Exit(129).cli_exit_code(), 129);
        assert_eq!(GitError::Exit(128).cli_exit_code(), 128);
        assert_eq!(GitError::usage("unknown option").cli_exit_code(), 129);
        assert_eq!(
            GitError::user_error("not a git repository").cli_exit_code(),
            128
        );
        assert_eq!(
            GitError::cli_exit(CliExit::Custom(2), "diff found changes").cli_exit_code(),
            2
        );
        assert_eq!(GitError::Command("bad value".into()).cli_exit_code(), 1);
        assert_eq!(GitError::not_found("missing ref").cli_exit_code(), 1);
    }

    #[test]
    fn git_error_cli_displays_message_only() {
        let err = GitError::usage("unknown option `--foo'");
        assert_eq!(err.to_string(), "unknown option `--foo'");
    }

    #[test]
    fn bstring_round_trips_bytes_and_displays_lossily() {
        let path = BString::from_bytes(b"src/\xFF.txt");
        assert_eq!(path.as_bytes(), b"src/\xFF.txt");
        let borrowed: &[u8] = path.borrow();
        assert_eq!(borrowed, b"src/\xFF.txt".as_slice());
        assert_eq!(format!("{path}"), "src/\u{FFFD}.txt");
        assert_eq!(path, b"src/\xFF.txt");
        assert_eq!(path.clone().into_bytes(), b"src/\xFF.txt".to_vec());
    }

    #[test]
    fn split_ident_line_parses_well_formed_ident() {
        let f = split_ident_line(b"A U Thor <author@example.com> 1112911993 -0700")
            .expect("well formed ident should parse");
        assert_eq!(f.name, b"A U Thor");
        assert_eq!(f.email, b"author@example.com");
        assert_eq!(f.date, Some(&b"1112911993"[..]));
        assert_eq!(f.tz, Some(&b"-0700"[..]));
    }

    #[test]
    fn split_ident_line_recovers_broken_email() {
        // git inserts junk after the '>': email stops at the first '>', but the
        // timestamp is found by scanning back from the end for the last '>'.
        let f = split_ident_line(b"A U Thor <author@example.com>-<> 1112911993 -0700")
            .expect("broken-email ident should parse");
        assert_eq!(f.name, b"A U Thor");
        assert_eq!(f.email, b"author@example.com");
        assert_eq!(f.date, Some(&b"1112911993"[..]));
        assert_eq!(f.tz, Some(&b"-0700"[..]));
    }

    #[test]
    fn split_ident_line_non_numeric_date_is_person_only() {
        let f = split_ident_line(b"A U Thor <author@example.com> totally_bogus -0700")
            .expect("ident without numeric date should still parse person");
        assert_eq!(f.email, b"author@example.com");
        assert_eq!(f.date, None);
        assert_eq!(f.tz, None);
    }

    #[test]
    fn split_ident_line_whitespace_date_is_person_only() {
        // Trailing spaces after '>' with no timestamp -> no date.
        let f = split_ident_line(b"A U Thor <author@example.com>    ")
            .expect("ident with trailing whitespace should parse person");
        assert_eq!(f.date, None);
        // A vertical tab is NOT git-isspace, so it stops the space-skip and the
        // (non-digit) VT yields no date either.
        let f = split_ident_line(b"A U Thor <author@example.com>   \x0b")
            .expect("ident with non-git-whitespace suffix should parse person");
        assert_eq!(f.date, None);
    }

    #[test]
    fn split_ident_line_requires_angle_brackets() {
        assert!(split_ident_line(b"no brackets here 123 +0000").is_none());
    }

    #[test]
    fn ident_render_date_overflow_is_epoch_sentinel() {
        // 2^64 + 1 (clamps in u64 parse) and 2^64 - 2 (fits u64 but past time_t)
        // both render the epoch sentinel with a forced +0000 timezone.
        assert_eq!(
            ident_render_date(b"18446744073709551617", b"-0700", &DateMode::Default),
            "Thu Jan 1 00:00:00 1970 +0000"
        );
        assert_eq!(
            ident_render_date(b"18446744073709551614", b"-0700", &DateMode::Default),
            "Thu Jan 1 00:00:00 1970 +0000"
        );
    }

    #[test]
    fn ident_render_date_valid_value_uses_original_timezone() {
        assert_eq!(
            ident_render_date(b"0", b"+0000", &DateMode::Default),
            "Thu Jan 1 00:00:00 1970 +0000"
        );
    }

    #[test]
    fn redact_url_for_display_strips_https_userinfo() {
        assert_eq!(
            redact_url_for_display("https://user:pass@host/repo.git"),
            "https://<redacted>@host/repo.git"
        );
    }

    #[test]
    fn redact_url_for_display_leaves_urls_without_userinfo_unchanged() {
        assert_eq!(
            redact_url_for_display("https://host/repo.git"),
            "https://host/repo.git"
        );
        assert_eq!(redact_url_for_display("origin"), "origin");
    }
}