grit_lib/

crlf.rs

//! CRLF / EOL conversion and clean/smudge filter support.
//!
//! This module handles line-ending conversion when staging files (`git add`)
//! and checking out files (`git checkout`, `read-tree -u`, `checkout-index`).
//!
//! Config knobs:
//!   - `core.autocrlf` (true / input / false)
//!   - `core.eol` (lf / crlf / native)
//!   - `core.safecrlf` (true / warn / false)
//!
//! Gitattributes:
//!   - `text` / `text=auto` / `-text` / `binary`
//!   - `eol=lf` / `eol=crlf`
//!   - `filter=<name>` (with `filter.<name>.clean` / `filter.<name>.smudge`)
//!   - `ident` keyword expansion

use std::collections::HashMap;
use std::path::{Path, PathBuf};
use std::sync::Arc;

use crate::attributes::{
    attr_rule_matches as git_attr_rule_matches, collect_attrs_for_path,
    parse_gitattributes_file_content, parse_gitattributes_file_content_with_base, AttrValue,
    MacroTable, ParsedGitAttributes, MAX_ATTR_FILE_BYTES,
};
use crate::command_runner::{system_command_runner, CommandRunner, CommandSpec, CommandStdin};
use crate::config::ConfigSet;
use crate::error::{FilterError, FilterPhase};
use crate::filter_process::{
    apply_process_clean, apply_process_smudge, FilterProcessState, FilterSmudgeMeta,
};
use encoding_rs::UTF_8;

/// Working-tree encoding conversion failure (Git `reencode_string_len` returning NULL).
#[derive(Debug, Clone, PartialEq, Eq, thiserror::Error)]
pub enum WorkTreeEncodingError {
    /// The encoding label is not supported by `encoding_rs` / the commit encoding helpers.
    #[error("unsupported encoding '{label}'")]
    UnsupportedEncoding {
        /// Config or attribute encoding name.
        label: String,
    },
}

/// Error from internal encoding helpers (not public convert_to_git surface).
#[derive(Debug, Clone, PartialEq, Eq, thiserror::Error)]
pub enum ConversionError {
    /// Working-tree encoding could not be resolved or converted.
    #[error(transparent)]
    Encoding(WorkTreeEncodingError),
    /// Filter, safecrlf, BOM validation, or other conversion failure (Git-compatible message).
    #[error("{0}")]
    Message(String),
}

impl From<WorkTreeEncodingError> for ConversionError {
    fn from(value: WorkTreeEncodingError) -> Self {
        Self::Encoding(value)
    }
}

impl From<String> for ConversionError {
    fn from(value: String) -> Self {
        Self::Message(value)
    }
}

impl From<&str> for ConversionError {
    fn from(value: &str) -> Self {
        Self::Message(value.to_owned())
    }
}

fn filter_detail(rel_path: &str, detail: String) -> FilterError {
    FilterError::NotFilteredProperly {
        path: rel_path.to_owned(),
        detail,
    }
}

fn external_filter_failed(rel_path: &str, driver: &str, phase: FilterPhase) -> FilterError {
    FilterError::ExternalFilterFailed {
        driver: driver.to_owned(),
        path: rel_path.to_owned(),
        phase,
    }
}

/// What `core.autocrlf` is set to.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum AutoCrlf {
    /// `True` variant.
    True,
    /// `Input` variant.
    Input,
    /// `False` variant.
    False,
}

/// What `core.eol` is set to.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum CoreEol {
    /// `Lf` variant.
    Lf,
    /// `Crlf` variant.
    Crlf,
    /// `Native` variant.
    Native,
}

/// What `core.safecrlf` is set to.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum SafeCrlf {
    /// `True` variant.
    True,
    /// `Warn` variant.
    Warn,
    /// `False` variant.
    False,
}

/// Per-file text attribute from .gitattributes.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum TextAttr {
    /// `text` — always treat as text.
    Set,
    /// `text=auto` — auto-detect.
    Auto,
    /// `-text` or `binary` — never convert.
    Unset,
    /// No text attribute specified.
    Unspecified,
}

/// Per-file eol attribute from .gitattributes.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum EolAttr {
    /// `Lf` variant.
    Lf,
    /// `Crlf` variant.
    Crlf,
    /// `Unspecified` variant.
    Unspecified,
}

/// Legacy `crlf` gitattribute (deprecated in Git; still honored for EOL conversion).
#[derive(Debug, Clone, Copy, PartialEq, Eq, Default)]
pub enum CrlfLegacyAttr {
    #[default]
    /// `Unspecified` variant.
    Unspecified,
    /// `-crlf` — disable CRLF conversion.
    Unset,
    /// `crlf=input` — normalize to LF in the object database; no CRLF on checkout.
    Input,
    /// Bare `crlf` (set) — force CRLF on checkout for text files.
    Crlf,
}

/// Per-file merge attribute from .gitattributes.
#[derive(Debug, Clone, PartialEq, Eq)]
pub enum MergeAttr {
    /// No merge attribute specified.
    Unspecified,
    /// `-merge` — treat as binary/non-text merge.
    Unset,
    /// `merge=<driver>` — use named merge driver.
    Driver(String),
}

/// How the `diff` gitattribute affects diff output.
#[derive(Debug, Clone, PartialEq, Eq)]
pub enum DiffAttr {
    /// No `diff` attribute (use heuristics / default).
    Unspecified,
    /// `-diff` / `diff=unset` / `binary` — treat as binary for diff purposes.
    Unset,
    /// Bare `diff` (set) — force textual diff even when the blob contains NUL.
    Set,
    /// `diff=<driver>` — use named driver (e.g. for textconv).
    Driver(String),
}

/// Per-file attributes relevant to conversion.
#[derive(Debug, Clone)]
pub struct FileAttrs {
    /// `text` field.
    pub text: TextAttr,
    /// `eol` field.
    pub eol: EolAttr,
    /// Effect of the `diff` gitattribute on diff output.
    pub diff_attr: DiffAttr,
    /// `filter_clean` field.
    pub filter_clean: Option<String>,
    /// `filter_smudge` field.
    pub filter_smudge: Option<String>,
    /// `filter.<name>.process` — long-running filter (takes precedence over clean/smudge commands).
    pub filter_process: Option<String>,
    /// Driver name from the active `filter=<name>` gitattribute (for error messages).
    pub filter_driver_name: Option<String>,
    /// Whether `filter.<name>.required` is set for this path's filter driver.
    pub filter_smudge_required: bool,
    /// Same config key as smudge; clean direction fails when unset if true.
    pub filter_clean_required: bool,
    /// `ident` field.
    pub ident: bool,
    /// `merge` field.
    pub merge: MergeAttr,
    /// `conflict_marker_size` field.
    pub conflict_marker_size: Option<String>,
    /// Working tree encoding (e.g. "utf-16") — content is converted to UTF-8 on add.
    pub working_tree_encoding: Option<String>,
    /// Legacy `crlf` / `-crlf` / `crlf=input` from `.gitattributes`.
    pub crlf_legacy: CrlfLegacyAttr,
    /// `whitespace` attribute value: `None` if unset, `Some("set")` for bare `whitespace`,
    /// `Some("unset")` for `-whitespace`, or `Some("trailing,...")` for `whitespace=...`.
    pub whitespace: Option<String>,
}

impl Default for FileAttrs {
    fn default() -> Self {
        FileAttrs {
            text: TextAttr::Unspecified,
            eol: EolAttr::Unspecified,
            diff_attr: DiffAttr::Unspecified,
            filter_clean: None,
            filter_smudge: None,
            filter_process: None,
            filter_driver_name: None,
            filter_smudge_required: false,
            filter_clean_required: false,
            ident: false,
            merge: MergeAttr::Unspecified,
            conflict_marker_size: None,
            working_tree_encoding: None,
            crlf_legacy: CrlfLegacyAttr::Unspecified,
            whitespace: None,
        }
    }
}

impl FileAttrs {
    /// Width of conflict marker lines for this path (`conflict-marker-size` attribute).
    ///
    /// Matches Git: unset or invalid values default to 7; explicit `0` also maps to 7.
    #[must_use]
    pub(crate) fn effective_conflict_marker_size(&self) -> usize {
        match self
            .conflict_marker_size
            .as_ref()
            .and_then(|value| value.parse::<usize>().ok())
        {
            None | Some(0) => 7,
            Some(n) => n,
        }
    }
}

/// Global conversion settings derived from config.
#[derive(Clone)]
pub struct ConversionConfig {
    /// `autocrlf` field.
    pub autocrlf: AutoCrlf,
    /// `eol` field.
    pub eol: CoreEol,
    /// `safecrlf` field.
    pub safecrlf: SafeCrlf,
    /// Subprocess runner for clean/smudge filters and iconv helpers.
    pub command_runner: Arc<dyn CommandRunner>,
}

impl ConversionConfig {
    /// Load conversion settings from a ConfigSet.
    pub fn from_config(config: &ConfigSet) -> Self {
        let autocrlf = match config.get("core.autocrlf") {
            Some(v) => match v.to_lowercase().as_str() {
                "true" | "yes" | "on" | "1" => AutoCrlf::True,
                "input" => AutoCrlf::Input,
                _ => AutoCrlf::False,
            },
            None => AutoCrlf::False,
        };

        let eol = match config.get("core.eol") {
            Some(v) => match v.to_lowercase().as_str() {
                "crlf" => CoreEol::Crlf,
                "lf" => CoreEol::Lf,
                "native" => CoreEol::Native,
                _ => CoreEol::Native,
            },
            None => CoreEol::Native,
        };

        let safecrlf = match config.get("core.safecrlf") {
            Some(v) => match v.to_lowercase().as_str() {
                "true" | "yes" | "on" | "1" => SafeCrlf::True,
                "warn" => SafeCrlf::Warn,
                _ => SafeCrlf::False,
            },
            // Git warns on round-trip EOL issues by default when unset.
            None => SafeCrlf::Warn,
        };

        ConversionConfig {
            autocrlf,
            eol,
            safecrlf,
            command_runner: system_command_runner(),
        }
    }

    /// Like [`Self::from_config`] with an explicit filter/hook subprocess runner.
    #[must_use]
    pub fn from_config_with_runner(
        config: &ConfigSet,
        command_runner: Arc<dyn CommandRunner>,
    ) -> Self {
        let mut cfg = Self::from_config(config);
        cfg.command_runner = command_runner;
        cfg
    }
}

impl std::fmt::Debug for ConversionConfig {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        f.debug_struct("ConversionConfig")
            .field("autocrlf", &self.autocrlf)
            .field("eol", &self.eol)
            .field("safecrlf", &self.safecrlf)
            .field("command_runner", &"<CommandRunner>")
            .finish()
    }
}

pub use crate::attributes::AttrRule;

fn merge_parsed_into(
    merged: &mut ParsedGitAttributes,
    mut parsed: ParsedGitAttributes,
    take_macros: bool,
) {
    merged.rules.append(&mut parsed.rules);
    if take_macros {
        merged.macros.defs.extend(parsed.macros.defs.drain());
    }
    merged.warnings.append(&mut parsed.warnings);
}

fn merge_gitattributes_text(
    merged: &mut ParsedGitAttributes,
    content: &str,
    display: &str,
    attr_base: &str,
) {
    if content.len() > MAX_ATTR_FILE_BYTES {
        return;
    }
    let parsed = if attr_base.is_empty() {
        parse_gitattributes_file_content(content, display)
    } else {
        parse_gitattributes_file_content_with_base(content, display, attr_base)
    };
    merge_parsed_into(merged, parsed, attr_base.is_empty());
}

pub(crate) fn config_ignorecase(config: &ConfigSet) -> bool {
    config
        .get_bool("core.ignorecase")
        .unwrap_or(Ok(false))
        .unwrap_or(false)
}

/// Load root and `info/attributes` gitattributes from the worktree.
#[must_use]
pub fn load_gitattributes_parsed(work_tree: &Path) -> ParsedGitAttributes {
    let mut merged = ParsedGitAttributes::default();
    let root_attrs = work_tree.join(".gitattributes");
    if let Ok(content) = std::fs::read_to_string(&root_attrs) {
        merge_gitattributes_text(&mut merged, &content, ".gitattributes", "");
    }
    let info_attrs = work_tree.join(".git/info/attributes");
    if let Ok(content) = std::fs::read_to_string(&info_attrs) {
        merge_gitattributes_text(&mut merged, &content, "info/attributes", "");
    }
    merged
}

/// Load .gitattributes from the worktree root.
#[must_use]
pub fn load_gitattributes(work_tree: &Path) -> Vec<AttrRule> {
    load_gitattributes_parsed(work_tree).rules
}

/// Parse gitattributes content into attribute rules (repository root scope).
///
/// This is useful when attributes are sourced from non-worktree inputs
/// (for example, tree objects selected by `--attr-source`).
#[must_use]
pub fn parse_gitattributes_content(content: &str) -> Vec<AttrRule> {
    parse_gitattributes_file_content(content, ".gitattributes").rules
}

/// Load `.gitattributes` from the index (stage 0 paths only).
#[must_use]
pub fn load_gitattributes_from_index(
    index: &crate::index::Index,
    odb: &crate::odb::Odb,
) -> Vec<AttrRule> {
    crate::attributes::load_gitattributes_from_index(index, odb, Path::new("."))
        .map(|p| p.rules)
        .unwrap_or_default()
}

/// Load merged `.gitattributes` for `rel_path` (root, nested dirs, info; index fallback).
#[must_use]
pub fn load_gitattributes_for_checkout_parsed(
    work_tree: &Path,
    rel_path: &str,
    index: &crate::index::Index,
    odb: &crate::odb::Odb,
) -> ParsedGitAttributes {
    let mut merged = load_gitattributes_parsed(work_tree);

    if !work_tree.join(".gitattributes").exists() {
        if let Some(entry) = index.get(b".gitattributes", 0) {
            if let Ok(obj) = odb.read(&entry.oid) {
                if let Ok(content) = String::from_utf8(obj.data) {
                    merge_gitattributes_text(&mut merged, &content, ".gitattributes", "");
                }
            }
        }
    }

    let path = Path::new(rel_path);
    if let Some(parent) = path.parent() {
        let mut accum = PathBuf::new();
        for comp in parent.components() {
            accum.push(comp);
            let ga_rel = accum.join(".gitattributes");
            let ga_stored = ga_rel.to_string_lossy().replace('\\', "/");
            let attr_base = Path::new(&ga_stored)
                .parent()
                .map(|p| p.to_string_lossy().replace('\\', "/"))
                .unwrap_or_default();
            let wt_ga = work_tree.join(&ga_rel);
            if let Ok(content) = std::fs::read_to_string(&wt_ga) {
                merge_gitattributes_text(&mut merged, &content, &ga_stored, &attr_base);
            } else {
                let key = crate::repo_relpath::storage_bytes(&ga_stored).into_owned();
                if let Some(entry) = index.get(&key, 0) {
                    if let Ok(obj) = odb.read(&entry.oid) {
                        if let Ok(content) = String::from_utf8(obj.data) {
                            merge_gitattributes_text(&mut merged, &content, &ga_stored, &attr_base);
                        }
                    }
                }
            }
        }
    }

    merged
}

/// Load `.gitattributes` rules that apply to `rel_path`, including root and
/// nested `dir/.gitattributes` along parent directories (Git-consistent order:
/// root first, then each ancestor directory; later rules win in [`get_file_attrs`]).
///
/// Reads from the working tree when present, otherwise from a stage-0 index entry.
#[must_use]
pub fn load_gitattributes_for_checkout(
    work_tree: &Path,
    rel_path: &str,
    index: &crate::index::Index,
    odb: &crate::odb::Odb,
) -> Vec<AttrRule> {
    load_gitattributes_for_checkout_parsed(work_tree, rel_path, index, odb).rules
}

fn config_bool_truthy(value: &str) -> bool {
    matches!(
        value.trim().to_ascii_lowercase().as_str(),
        "true" | "yes" | "on" | "1"
    )
}

fn attr_path_for_match(rel_path: &str, is_dir: bool) -> String {
    if is_dir && !rel_path.is_empty() && !rel_path.ends_with('/') {
        format!("{rel_path}/")
    } else {
        rel_path.to_owned()
    }
}

/// Map merged gitattributes (Git `collect_attrs`) to CRLF/filter [`FileAttrs`].
#[must_use]
pub fn file_attrs_from_attr_map(map: &HashMap<String, AttrValue>, config: &ConfigSet) -> FileAttrs {
    let mut fa = FileAttrs::default();
    for (name, value) in map {
        match name.as_str() {
            "text" => {
                fa.text = match value {
                    AttrValue::Set => TextAttr::Set,
                    AttrValue::Unset => TextAttr::Unset,
                    AttrValue::Value(v) if v == "auto" => TextAttr::Auto,
                    AttrValue::Clear | AttrValue::Value(_) => TextAttr::Unspecified,
                };
            }
            "eol" => {
                fa.eol = match value {
                    AttrValue::Value(v) if v == "lf" => EolAttr::Lf,
                    AttrValue::Value(v) if v == "crlf" => EolAttr::Crlf,
                    _ => EolAttr::Unspecified,
                };
            }
            "filter" => {
                let driver = match value {
                    AttrValue::Unset => {
                        fa.filter_clean = None;
                        fa.filter_smudge = None;
                        fa.filter_process = None;
                        fa.filter_driver_name = None;
                        fa.filter_smudge_required = false;
                        fa.filter_clean_required = false;
                        continue;
                    }
                    AttrValue::Set => continue,
                    AttrValue::Clear => continue,
                    AttrValue::Value(d) => d.as_str(),
                };
                let clean_key = format!("filter.{driver}.clean");
                let smudge_key = format!("filter.{driver}.smudge");
                let process_key = format!("filter.{driver}.process");
                let req_key = format!("filter.{driver}.required");
                fa.filter_driver_name = Some(driver.to_owned());
                fa.filter_process = config.get(&process_key).filter(|s| !s.is_empty());
                if fa.filter_process.is_some() {
                    fa.filter_clean = None;
                    fa.filter_smudge = None;
                } else {
                    fa.filter_clean = config.get(&clean_key);
                    fa.filter_smudge = config.get(&smudge_key);
                }
                let required = config
                    .get(&req_key)
                    .is_some_and(|v| config_bool_truthy(v.as_str()));
                fa.filter_smudge_required = required;
                fa.filter_clean_required = required;
            }
            "diff" => {
                fa.diff_attr = match value {
                    AttrValue::Unset => DiffAttr::Unset,
                    AttrValue::Set => DiffAttr::Set,
                    AttrValue::Value(v) if !v.is_empty() => DiffAttr::Driver(v.clone()),
                    AttrValue::Clear => DiffAttr::Unspecified,
                    _ => DiffAttr::Unspecified,
                };
            }
            "ident" => {
                fa.ident = matches!(value, AttrValue::Set);
            }
            "merge" => {
                fa.merge = match value {
                    AttrValue::Unset => MergeAttr::Unset,
                    AttrValue::Set => MergeAttr::Unspecified,
                    AttrValue::Value(v) => MergeAttr::Driver(v.clone()),
                    AttrValue::Clear => MergeAttr::Unspecified,
                };
            }
            "conflict-marker-size" => match value {
                AttrValue::Unset | AttrValue::Clear => fa.conflict_marker_size = None,
                AttrValue::Value(v) => fa.conflict_marker_size = Some(v.clone()),
                AttrValue::Set => {}
            },
            "working-tree-encoding" => {
                if let AttrValue::Value(v) = value {
                    if !v.is_empty() {
                        fa.working_tree_encoding = Some(v.clone());
                    }
                }
            }
            "crlf" => {
                fa.crlf_legacy = match value {
                    AttrValue::Unset => CrlfLegacyAttr::Unset,
                    AttrValue::Value(v) if v == "input" => CrlfLegacyAttr::Input,
                    AttrValue::Set => CrlfLegacyAttr::Crlf,
                    AttrValue::Value(v) if v == "true" || v == "set" => CrlfLegacyAttr::Crlf,
                    _ => CrlfLegacyAttr::Unspecified,
                };
            }
            "whitespace" => match value {
                AttrValue::Unset => fa.whitespace = Some("unset".to_owned()),
                AttrValue::Set => fa.whitespace = Some("set".to_owned()),
                AttrValue::Value(v) if !v.is_empty() => fa.whitespace = Some(v.clone()),
                _ => {}
            },
            _ => {}
        }
    }
    fa
}

/// Resolve [`FileAttrs`] using Git-compatible attribute matching and macro expansion.
#[must_use]
pub fn file_attrs_from_gitattributes(
    rules: &[AttrRule],
    macros: &MacroTable,
    rel_path: &str,
    is_dir: bool,
    config: &ConfigSet,
) -> FileAttrs {
    let path = attr_path_for_match(rel_path, is_dir);
    let map = collect_attrs_for_path(rules, macros, &path, config_ignorecase(config));
    file_attrs_from_attr_map(&map, config)
}

/// Get file attributes for a given path from .gitattributes rules and config.
///
/// `is_dir` should be true when `rel_path` names a directory (Git passes a trailing `/` for
/// directory paths in some call sites; we accept either trailing `/` or this flag from tree walks).
#[must_use]
pub fn get_file_attrs(
    rules: &[AttrRule],
    rel_path: &str,
    is_dir: bool,
    config: &ConfigSet,
) -> FileAttrs {
    file_attrs_from_gitattributes(rules, &MacroTable::default(), rel_path, is_dir, config)
}

/// Every attribute assigned to `path` after macro expansion, with `icase` selecting
/// case-insensitive pattern matching (`core.ignorecase`).
pub(crate) fn path_gitattribute_values(
    rules: &[AttrRule],
    macros: &MacroTable,
    icase: bool,
    path: &str,
    is_dir: bool,
) -> std::collections::HashMap<String, crate::attributes::AttrValue> {
    collect_attrs_for_path(rules, macros, &attr_path_for_match(path, is_dir), icase)
}

/// Whether `rule` matches `rel_path` given directory vs file context (Git `path_matches`).
#[must_use]
pub fn attr_rule_matches(rule: &AttrRule, rel_path: &str, is_dir: bool) -> bool {
    let match_path = attr_path_for_match(rel_path, is_dir);
    git_attr_rule_matches(rule, &match_path, false)
}

/// Returns true if the data looks binary (contains NUL bytes in the first 8000 bytes).
pub fn is_binary(data: &[u8]) -> bool {
    let check_len = data.len().min(8000);
    data[..check_len].contains(&0)
}

// Git `convert.c` `CONVERT_STAT_BITS_*` / `gather_convert_stats_ascii` (for `ls-files --eol`).
const CONVERT_STAT_BITS_TXT_LF: u32 = 0x1;
const CONVERT_STAT_BITS_TXT_CRLF: u32 = 0x2;
const CONVERT_STAT_BITS_BIN: u32 = 0x4;

#[derive(Default, Clone)]
struct TextStat {
    nul: u32,
    lonecr: u32,
    lonelf: u32,
    crlf: u32,
    printable: u32,
    nonprintable: u32,
}

fn gather_text_stat(data: &[u8]) -> TextStat {
    let mut s = TextStat::default();
    let mut i = 0usize;
    while i < data.len() {
        let c = data[i];
        if c == b'\r' {
            if i + 1 < data.len() && data[i + 1] == b'\n' {
                s.crlf += 1;
                i += 2;
            } else {
                s.lonecr += 1;
                i += 1;
            }
            continue;
        }
        if c == b'\n' {
            s.lonelf += 1;
            i += 1;
            continue;
        }
        if c == 127 {
            s.nonprintable += 1;
        } else if c < 32 {
            match c {
                b'\t' | b'\x08' | b'\x1b' | b'\x0c' => s.printable += 1,
                0 => {
                    s.nul += 1;
                    s.nonprintable += 1;
                }
                _ => s.nonprintable += 1,
            }
        } else {
            s.printable += 1;
        }
        i += 1;
    }
    s
}

fn convert_is_binary(stats: &TextStat) -> bool {
    stats.lonecr > 0 || stats.nul > 0 || (stats.printable >> 7) < stats.nonprintable
}

fn git_text_stat(data: &[u8]) -> TextStat {
    let mut stats = gather_text_stat(data);
    if !data.is_empty() && data[data.len() - 1] == 0x1a {
        stats.nonprintable = stats.nonprintable.saturating_sub(1);
    }
    stats
}

/// Git `will_convert_lf_to_crlf` using [`TextStat`] (same rules as [`should_convert_to_crlf`] on bytes).
fn will_convert_lf_to_crlf_from_stats(
    stats: &TextStat,
    conv: &ConversionConfig,
    attrs: &FileAttrs,
) -> bool {
    let has_lone_lf = stats.lonelf > 0;
    let is_bin = convert_is_binary(stats);

    match attrs.crlf_legacy {
        CrlfLegacyAttr::Unset | CrlfLegacyAttr::Input => return false,
        CrlfLegacyAttr::Crlf => {
            if attrs.text == TextAttr::Unset {
                return false;
            }
            return has_lone_lf;
        }
        CrlfLegacyAttr::Unspecified => {}
    }

    if attrs.text == TextAttr::Unset {
        return false;
    }

    if attrs.eol != EolAttr::Unspecified {
        if attrs.text == TextAttr::Auto && is_bin {
            return false;
        }
        if attrs.eol != EolAttr::Crlf {
            return false;
        }
        if attrs.text == TextAttr::Auto {
            return auto_crlf_should_smudge_lf_to_crlf_from_stats(stats);
        }
        return has_lone_lf;
    }

    if attrs.text == TextAttr::Set {
        if !output_eol_is_crlf(conv) {
            return false;
        }
        return has_lone_lf;
    }

    if attrs.text == TextAttr::Auto {
        if is_bin || !output_eol_is_crlf(conv) {
            return false;
        }
        return auto_crlf_should_smudge_lf_to_crlf_from_stats(stats);
    }

    match conv.autocrlf {
        AutoCrlf::True => {
            if is_bin {
                return false;
            }
            auto_crlf_should_smudge_lf_to_crlf_from_stats(stats)
        }
        AutoCrlf::Input | AutoCrlf::False => false,
    }
}

fn auto_crlf_should_smudge_lf_to_crlf_from_stats(stats: &TextStat) -> bool {
    if stats.lonelf == 0 {
        return false;
    }
    if stats.lonecr > 0 || stats.crlf > 0 {
        return false;
    }
    !convert_is_binary(stats)
}

fn gather_convert_stats(data: &[u8]) -> u32 {
    if data.is_empty() {
        return 0;
    }
    let mut stats = gather_text_stat(data);
    if !data.is_empty() && data[data.len() - 1] == 0x1a {
        stats.nonprintable = stats.nonprintable.saturating_sub(1);
    }
    let mut ret = 0u32;
    if convert_is_binary(&stats) {
        ret |= CONVERT_STAT_BITS_BIN;
    }
    if stats.crlf > 0 {
        ret |= CONVERT_STAT_BITS_TXT_CRLF;
    }
    if stats.lonelf > 0 {
        ret |= CONVERT_STAT_BITS_TXT_LF;
    }
    ret
}

/// Git `convert.c` `gather_convert_stats_ascii` — worktree/index blob EOL stats for `ls-files --eol`.
#[must_use]
pub fn gather_convert_stats_ascii(data: &[u8]) -> &'static str {
    let convert_stats = gather_convert_stats(data);
    if convert_stats & CONVERT_STAT_BITS_BIN != 0 {
        return "-text";
    }
    match convert_stats {
        CONVERT_STAT_BITS_TXT_LF => "lf",
        CONVERT_STAT_BITS_TXT_CRLF => "crlf",
        x if x == (CONVERT_STAT_BITS_TXT_LF | CONVERT_STAT_BITS_TXT_CRLF) => "mixed",
        _ => "none",
    }
}

/// Returns true if data contains any CRLF sequences.
pub fn has_crlf(data: &[u8]) -> bool {
    data.windows(2).any(|w| w == b"\r\n")
}

/// Returns true if data contains any lone LF (not preceded by CR).
pub fn has_lone_lf(data: &[u8]) -> bool {
    for i in 0..data.len() {
        if data[i] == b'\n' && (i == 0 || data[i - 1] != b'\r') {
            return true;
        }
    }
    false
}

/// Returns true if data contains a bare CR not followed by LF (Git `text_stat.lonecr`).
fn has_lone_cr(data: &[u8]) -> bool {
    for i in 0..data.len() {
        if data[i] == b'\r' && (i + 1 >= data.len() || data[i + 1] != b'\n') {
            return true;
        }
    }
    false
}

/// Git `convert.c` `will_convert_lf_to_crlf` for `CRLF_AUTO` / `CRLF_AUTO_INPUT` / `CRLF_AUTO_CRLF`:
/// if the blob already has CRLF pairs or lone CRs, do not convert lone LFs to CRLF on checkout.
fn auto_crlf_should_smudge_lf_to_crlf(data: &[u8]) -> bool {
    if !has_lone_lf(data) {
        return false;
    }
    if has_lone_cr(data) || has_crlf(data) {
        return false;
    }
    if is_binary(data) {
        return false;
    }
    true
}

/// Returns true if ALL line endings are CRLF (no lone LF).
pub fn is_all_crlf(data: &[u8]) -> bool {
    has_crlf(data) && !has_lone_lf(data)
}

/// Returns true if ALL line endings are LF (no CRLF).
pub fn is_all_lf(data: &[u8]) -> bool {
    has_lone_lf(data) && !has_crlf(data)
}

/// Git `convert.c` `has_crlf_in_index`: index blob already contains CRLF pairs (non-binary).
#[must_use]
pub fn has_crlf_in_index_blob(data: &[u8]) -> bool {
    if !data.contains(&b'\r') {
        return false;
    }
    let st = gather_convert_stats(data);
    st & CONVERT_STAT_BITS_BIN == 0 && (st & CONVERT_STAT_BITS_TXT_CRLF) != 0
}

/// Whether clean conversion uses Git's `has_crlf_in_index` guard (`convert.c` only for
/// `CRLF_AUTO`, `CRLF_AUTO_INPUT`, `CRLF_AUTO_CRLF`). Bare `eol=` without `text=auto` becomes
/// `CRLF_TEXT_*` and must not use this guard.
#[must_use]
pub fn clean_uses_autocrlf_index_guard(attrs: &FileAttrs, conv: &ConversionConfig) -> bool {
    if attrs.text == TextAttr::Unset || attrs.crlf_legacy == CrlfLegacyAttr::Unset {
        return false;
    }
    if attrs.eol != EolAttr::Unspecified && attrs.text != TextAttr::Auto {
        return false;
    }
    attrs.text == TextAttr::Auto
        || (attrs.text == TextAttr::Unspecified
            && matches!(conv.autocrlf, AutoCrlf::True | AutoCrlf::Input))
}

/// Optional inputs for [`convert_to_git_with_opts`] (Git `CONV_EOL_RENORMALIZE` / index blob).
#[derive(Clone, Copy)]
pub struct ConvertToGitOpts<'a> {
    /// Stage-0 blob bytes for this path before the current add (for safer-autocrlf).
    pub index_blob: Option<&'a [u8]>,
    /// When true, always apply CRLF→LF when configured (merge/cherry-pick renormalize).
    pub renormalize: bool,
    /// When false, skip `core.safecrlf` simulation (used for internal diff/hashing — must not spam stderr).
    pub check_safecrlf: bool,
    /// Long-running filter registry for this repository handle.
    pub filter_process: Option<&'a FilterProcessState>,
    /// Subprocess runner for clean/smudge shell filters (defaults to [`ConversionConfig::command_runner`]).
    pub command_runner: Option<&'a dyn CommandRunner>,
}

impl Default for ConvertToGitOpts<'_> {
    fn default() -> Self {
        Self {
            index_blob: None,
            renormalize: false,
            check_safecrlf: true,
            filter_process: None,
            command_runner: None,
        }
    }
}

impl std::fmt::Debug for ConvertToGitOpts<'_> {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        f.debug_struct("ConvertToGitOpts")
            .field("index_blob", &self.index_blob.map(|b| b.len()))
            .field("renormalize", &self.renormalize)
            .field("check_safecrlf", &self.check_safecrlf)
            .field("command_runner", &self.command_runner.is_some())
            .finish()
    }
}

// ---------------------------------------------------------------------------
// working-tree-encoding (Git `convert.c` `encode_to_git` / `encode_to_worktree`)
// ---------------------------------------------------------------------------

// BOM byte sequences (Git `utf8.c`).
const UTF16_BE_BOM: &[u8] = &[0xFE, 0xFF];
const UTF16_LE_BOM: &[u8] = &[0xFF, 0xFE];
const UTF32_BE_BOM: &[u8] = &[0x00, 0x00, 0xFE, 0xFF];
const UTF32_LE_BOM: &[u8] = &[0xFF, 0xFE, 0x00, 0x00];

/// Canonical lowercase UTF label for a `working-tree-encoding` value, or `None` if the label is
/// not a UTF-16/UTF-32/UTF-8 variant Git treats specially. Mirrors Git's `same_utf_encoding`
/// (strip a leading `utf` then an optional `-`, case-insensitive), so `utf16`, `UTF-16`,
/// `Utf16Le-Bom` all normalize.
fn canonical_utf_label(label: &str) -> Option<String> {
    let trimmed = label.trim();
    let lower = trimmed.to_ascii_lowercase();
    let rest = lower.strip_prefix("utf")?;
    let rest = rest.strip_prefix('-').unwrap_or(rest);
    match rest {
        "8" => Some("utf-8".to_string()),
        "16" => Some("utf-16".to_string()),
        "16be" => Some("utf-16be".to_string()),
        "16le" => Some("utf-16le".to_string()),
        "16be-bom" => Some("utf-16be-bom".to_string()),
        "16le-bom" => Some("utf-16le-bom".to_string()),
        "32" => Some("utf-32".to_string()),
        "32be" => Some("utf-32be".to_string()),
        "32le" => Some("utf-32le".to_string()),
        _ => None,
    }
}

fn has_bom_prefix(data: &[u8], bom: &[u8]) -> bool {
    data.len() >= bom.len() && &data[..bom.len()] == bom
}

/// Git `has_prohibited_utf_bom`: UTF-16BE/LE and UTF-32BE/LE must not begin with a BOM.
fn has_prohibited_utf_bom(canon: &str, data: &[u8]) -> bool {
    match canon {
        "utf-16be" | "utf-16le" => {
            has_bom_prefix(data, UTF16_BE_BOM) || has_bom_prefix(data, UTF16_LE_BOM)
        }
        "utf-32be" | "utf-32le" => {
            has_bom_prefix(data, UTF32_BE_BOM) || has_bom_prefix(data, UTF32_LE_BOM)
        }
        _ => false,
    }
}

/// Git `is_missing_required_utf_bom`: bare UTF-16 / UTF-32 must begin with a BOM.
fn is_missing_required_utf_bom(canon: &str, data: &[u8]) -> bool {
    match canon {
        "utf-16" => !(has_bom_prefix(data, UTF16_BE_BOM) || has_bom_prefix(data, UTF16_LE_BOM)),
        "utf-32" => !(has_bom_prefix(data, UTF32_BE_BOM) || has_bom_prefix(data, UTF32_LE_BOM)),
        _ => false,
    }
}

/// Git `validate_encoding`: emit the advice line to stderr and return an error body when the BOM
/// presence is wrong for a UTF-16/UTF-32 encoding.
///
/// `label` is the original attribute spelling (preserved in messages, like Git). When
/// `die_on_error` is true (`CONV_WRITE_OBJECT`) the body is prefixed `fatal:` so the top-level
/// printer surfaces it verbatim; otherwise the `error:` line is printed here (Git `error()` returns
/// "content unmodified") and the same body is returned for the caller to swallow.
fn validate_utf_bom(
    canon: &str,
    label: &str,
    rel_path: &str,
    data: &[u8],
    die_on_error: bool,
) -> Result<(), String> {
    if has_prohibited_utf_bom(canon, data) {
        // Advice cuts the trailing "be"/"le" so the user sees the BOM-capable name (UTF-16/UTF-32).
        let stripped = label
            .strip_prefix("utf")
            .or_else(|| label.strip_prefix("UTF"));
        let _utf_num = stripped
            .map(|s| s.trim_start_matches('-'))
            .and_then(|s| s.get(..s.len().saturating_sub(2)))
            .unwrap_or("");
        let body = format!("BOM is prohibited in '{rel_path}' if encoded as {label}");
        if die_on_error {
            return Err(body);
        }
        return Err(body);
    }
    if is_missing_required_utf_bom(canon, data) {
        let _utf_num = label
            .strip_prefix("utf")
            .or_else(|| label.strip_prefix("UTF"))
            .map(|s| s.trim_start_matches('-'))
            .unwrap_or("");
        let body = format!("BOM is required in '{rel_path}' if encoded as {label}");
        if die_on_error {
            return Err(body);
        }
        return Err(body);
    }
    Ok(())
}

/// Decode raw working-tree bytes (`enc_label`) into UTF-8 for the object DB (Git `encode_to_git`).
///
/// When `validate` is true (writing to the object DB), enforce Git's UTF BOM rules and surface the
/// matching fatal message + advice (`die_on_error`). For internal diff/status reads it is false.
fn decode_working_tree_bytes_to_utf8(
    src: &[u8],
    rel_path: &str,
    enc_label: &str,
    validate: bool,
) -> Result<Vec<u8>, ConversionError> {
    let label = enc_label.trim();
    if label.is_empty() {
        return Ok(src.to_vec());
    }

    let canon = canonical_utf_label(label);

    // BOM validation (only the UTF-16/UTF-32 family). Git validates on every `encode_to_git`; when
    // writing to the object DB it dies, otherwise (diff/status reads) it prints `error:` and treats
    // the content as unmodified — `validate` here is Git's `die_on_error` (`CONV_WRITE_OBJECT`).
    if let Some(ref c) = canon {
        validate_utf_bom(c, label, rel_path, src, validate)?;
    }

    // UTF-8 is the default encoding: no conversion (Git `git_path_check_encoding`).
    if canon.as_deref() == Some("utf-8") {
        return Ok(src.to_vec());
    }

    // The `*-BOM` aliases decode like the matching raw encoding once the BOM is stripped.
    let (iconv_from, body): (&str, &[u8]) = match canon.as_deref() {
        Some("utf-16le-bom") => {
            let body = if has_bom_prefix(src, UTF16_LE_BOM) {
                &src[2..]
            } else {
                src
            };
            ("UTF-16LE", body)
        }
        Some("utf-16be-bom") => {
            let body = if has_bom_prefix(src, UTF16_BE_BOM) {
                &src[2..]
            } else {
                src
            };
            ("UTF-16BE", body)
        }
        // Bare UTF-16/UTF-32 keep their BOM; iconv consumes it to pick the byte order.
        Some(c) => (utf_canon_to_iconv_name(c), src),
        None => {
            if crate::commit_encoding::is_iso_8859_1(label) {
                return Ok(crate::commit_encoding::worktree_latin1_to_utf8_bytes(src));
            }
            // Unknown / unsupported label (Git `reencode_string_len` returns NULL →
            // `failed to encode '%s' from %s to %s`).
            let Some(enc) = crate::commit_encoding::resolve(label) else {
                return Err(WorkTreeEncodingError::UnsupportedEncoding {
                    label: label.to_owned(),
                }
                .into());
            };
            if enc == UTF_8 {
                return Ok(src.to_vec());
            }
            let (cow, _, had_errors) = enc.decode(src);
            if had_errors {
                return Err(ConversionError::Message(format!(
                    "failed to encode '{rel_path}' from {label} to UTF-8"
                )));
            }
            return Ok(cow.into_owned().into_bytes());
        }
    };

    decode_utf_bytes_with_encoding_rs(body, rel_path, label, iconv_from)
}

/// Decode UTF-16/UTF-32 family bytes to UTF-8 (`encoding_rs` or manual UTF-32).
fn decode_utf_bytes_with_encoding_rs(
    body: &[u8],
    rel_path: &str,
    label: &str,
    iconv_from: &str,
) -> Result<Vec<u8>, ConversionError> {
    let fail = || {
        ConversionError::Message(format!(
            "failed to encode '{rel_path}' from {label} to UTF-8"
        ))
    };
    match iconv_from {
        "UTF-16BE" => {
            let (cow, _, had_errors) = encoding_rs::UTF_16BE.decode(body);
            if had_errors {
                return Err(fail());
            }
            Ok(cow.into_owned().into_bytes())
        }
        "UTF-16LE" => {
            let (cow, _, had_errors) = encoding_rs::UTF_16LE.decode(body);
            if had_errors {
                return Err(fail());
            }
            Ok(cow.into_owned().into_bytes())
        }
        "UTF-16" => {
            if has_bom_prefix(body, UTF16_BE_BOM) {
                decode_utf_bytes_with_encoding_rs(&body[2..], rel_path, label, "UTF-16BE")
            } else if has_bom_prefix(body, UTF16_LE_BOM) {
                decode_utf_bytes_with_encoding_rs(&body[2..], rel_path, label, "UTF-16LE")
            } else {
                Err(fail())
            }
        }
        "UTF-32" => {
            if has_bom_prefix(body, UTF32_BE_BOM) {
                decode_utf32_body_to_utf8_bytes(&body[4..], rel_path, true)
            } else if has_bom_prefix(body, UTF32_LE_BOM) {
                decode_utf32_body_to_utf8_bytes(&body[4..], rel_path, false)
            } else {
                Err(fail())
            }
        }
        "UTF-32BE" => decode_utf32_body_to_utf8_bytes(body, rel_path, true),
        "UTF-32LE" => decode_utf32_body_to_utf8_bytes(body, rel_path, false),
        _ => Err(fail()),
    }
}

fn decode_utf32_body_to_utf8_bytes(
    body: &[u8],
    rel_path: &str,
    big_endian: bool,
) -> Result<Vec<u8>, ConversionError> {
    let fail = || {
        ConversionError::Message(format!(
            "failed to encode '{rel_path}' from UTF-32 to UTF-8"
        ))
    };
    if !body.len().is_multiple_of(4) {
        return Err(fail());
    }
    let mut s = String::new();
    for chunk in body.as_chunks::<4>().0 {
        let cp = if big_endian {
            u32::from_be_bytes([chunk[0], chunk[1], chunk[2], chunk[3]])
        } else {
            u32::from_le_bytes([chunk[0], chunk[1], chunk[2], chunk[3]])
        };
        let Some(ch) = char::from_u32(cp) else {
            return Err(fail());
        };
        s.push(ch);
    }
    Ok(s.into_bytes())
}

/// iconv encoding name for a canonical UTF label (raw encodings only; `*-bom` handled separately).
fn utf_canon_to_iconv_name(canon: &str) -> &'static str {
    match canon {
        "utf-16" => "UTF-16",
        "utf-16be" => "UTF-16BE",
        "utf-16le" => "UTF-16LE",
        "utf-32" => "UTF-32",
        "utf-32be" => "UTF-32BE",
        "utf-32le" => "UTF-32LE",
        _ => "UTF-8",
    }
}

/// Encode a UTF-8 blob into raw working-tree bytes for `enc_label` (Git `encode_to_worktree`).
///
/// Bare `UTF-16`/`UTF-32` and the `*-BOM` aliases get a BOM (Git relies on libiconv / explicit
/// BOM handling); the raw `UTF-16BE`/`UTF-16LE`/`UTF-32BE`/`UTF-32LE` encodings produce no BOM.
fn encode_utf8_blob_to_working_tree_bytes(
    src: &[u8],
    rel_path: &str,
    enc_label: &str,
) -> Result<Vec<u8>, ConversionError> {
    let label = enc_label.trim();
    if label.is_empty() {
        return Ok(src.to_vec());
    }

    let canon = canonical_utf_label(label);
    if canon.as_deref() == Some("utf-8") {
        return Ok(src.to_vec());
    }

    let fail = || {
        ConversionError::Message(format!(
            "failed to encode '{rel_path}' from UTF-8 to {label}"
        ))
    };

    // The `*-BOM` aliases: encode to the raw form, then prepend the requested BOM.
    match canon.as_deref() {
        Some("utf-16le-bom") => {
            let body = encode_utf_with_encoding_rs(src, "utf-16le").ok_or_else(fail)?;
            let mut out = UTF16_LE_BOM.to_vec();
            out.extend(body);
            return Ok(out);
        }
        Some("utf-16be-bom") => {
            let body = encode_utf_with_encoding_rs(src, "utf-16be").ok_or_else(fail)?;
            let mut out = UTF16_BE_BOM.to_vec();
            out.extend(body);
            return Ok(out);
        }
        // Bare UTF-16/UTF-32: Git/libiconv emits a BOM then little-endian code units (see `iconv`).
        Some("utf-16") => {
            let body = encode_utf_with_encoding_rs(src, "utf-16le").ok_or_else(fail)?;
            let mut out = UTF16_LE_BOM.to_vec();
            out.extend(body);
            return Ok(out);
        }
        Some("utf-32") => {
            let body = encode_utf_with_encoding_rs(src, "utf-32le").ok_or_else(fail)?;
            let mut out = UTF32_LE_BOM.to_vec();
            out.extend(body);
            return Ok(out);
        }
        Some(c) => {
            return encode_utf_with_encoding_rs(src, c).ok_or_else(fail);
        }
        None => {}
    }

    if crate::commit_encoding::is_iso_8859_1(label) {
        return crate::commit_encoding::utf8_bytes_to_worktree_latin1(src).ok_or_else(fail);
    }

    let s = std::str::from_utf8(src).map_err(|_| fail())?;
    let Some(enc) = crate::commit_encoding::resolve(label) else {
        return Err(WorkTreeEncodingError::UnsupportedEncoding {
            label: label.to_owned(),
        }
        .into());
    };
    if enc == UTF_8 {
        return Ok(src.to_vec());
    }
    let (cow, _, had_errors) = enc.encode(s);
    if had_errors {
        return Err(fail());
    }
    Ok(cow.into_owned())
}

/// UTF encode via `encoding_rs`/manual UTF-32. `target` is a canonical label (`utf-16le` etc.).
/// Produces raw bytes (no BOM).
fn encode_utf_with_encoding_rs(src: &[u8], target: &str) -> Option<Vec<u8>> {
    let s = std::str::from_utf8(src).ok()?;
    let lower = target.to_ascii_lowercase();
    let mut out = Vec::new();
    match lower.as_str() {
        "utf-16" | "utf-16be" => {
            for u in s.encode_utf16() {
                out.extend_from_slice(&u.to_be_bytes());
            }
        }
        "utf-16le" => {
            for u in s.encode_utf16() {
                out.extend_from_slice(&u.to_le_bytes());
            }
        }
        "utf-32" | "utf-32be" => {
            for ch in s.chars() {
                out.extend_from_slice(&(ch as u32).to_be_bytes());
            }
        }
        "utf-32le" => {
            for ch in s.chars() {
                out.extend_from_slice(&(ch as u32).to_le_bytes());
            }
        }
        _ => return None,
    }
    Some(out)
}

// ---------------------------------------------------------------------------
// Input (add / clean) direction
// ---------------------------------------------------------------------------

/// Convert data for storage in the index/object database (the "clean" direction).
///
/// This handles:
/// 1. Clean filter execution
/// 2. `working-tree-encoding` decode to UTF-8
/// 3. CRLF → LF conversion based on config + attributes
/// 4. `$Id: … $` → `$Id$` collapse when `ident` is set
/// 5. safecrlf checking
///
/// Returns `Ok(data)` on success, or an error if safecrlf rejects it.
pub fn convert_to_git(
    data: &[u8],
    rel_path: &str,
    conv: &ConversionConfig,
    file_attrs: &FileAttrs,
) -> Result<Vec<u8>, FilterError> {
    convert_to_git_with_opts(
        data,
        rel_path,
        conv,
        file_attrs,
        ConvertToGitOpts::default(),
    )
}

/// Like [`convert_to_git`] with Git-compatible safer-autocrlf index handling.
pub fn convert_to_git_with_opts(
    data: &[u8],
    rel_path: &str,
    conv: &ConversionConfig,
    file_attrs: &FileAttrs,
    opts: ConvertToGitOpts<'_>,
) -> Result<Vec<u8>, FilterError> {
    convert_to_git_owned(data.to_vec(), rel_path, conv, file_attrs, opts)
}

/// [`convert_to_git_with_opts`] taking ownership of `data`, so content that needs no
/// conversion is returned without a copy.
pub(crate) fn convert_to_git_owned(
    data: Vec<u8>,
    rel_path: &str,
    conv: &ConversionConfig,
    file_attrs: &FileAttrs,
    opts: ConvertToGitOpts<'_>,
) -> Result<Vec<u8>, FilterError> {
    let mut buf = data;
    let owned_filter_state;
    let filter_state = if let Some(s) = opts.filter_process {
        s
    } else {
        owned_filter_state = FilterProcessState::new(Arc::clone(&conv.command_runner));
        &owned_filter_state
    };
    let runner = opts.command_runner.unwrap_or(conv.command_runner.as_ref());

    // 1. Run clean filter if configured (long-running `process` overrides clean command)
    if let Some(ref proc_cmd) = file_attrs.filter_process {
        let name = file_attrs.filter_driver_name.as_deref().unwrap_or_default();
        if file_attrs.filter_clean_required {
            start_required_process_filter(
                filter_state,
                proc_cmd,
                name,
                rel_path,
                FilterPhase::Clean,
            )?;
        }
        match apply_process_clean(filter_state, proc_cmd, rel_path, &buf) {
            Ok(filtered) => buf = filtered,
            Err(e) => {
                if file_attrs.filter_clean_required {
                    if e.contains("expected git-filter-server") {
                        return Err(filter_detail(rel_path, e));
                    }
                    return Err(FilterError::ExternalFilterFailed {
                        driver: name.to_owned(),
                        path: rel_path.to_owned(),
                        phase: FilterPhase::Clean,
                    });
                }
                if e.starts_with("filter status: abort") {
                    filter_state.disable_process_filter(proc_cmd);
                }
            }
        }
    } else {
        match file_attrs.filter_clean.as_ref() {
            Some(clean_cmd) => {
                match run_filter_command(runner, clean_cmd, &buf, rel_path, Some(filter_state)) {
                    Ok(filtered) => buf = filtered,
                    Err(e) if file_attrs.filter_clean_required => {
                        let name = file_attrs.filter_driver_name.as_deref().unwrap_or_default();
                        return Err(e.into_required_error(name, rel_path, FilterPhase::Clean));
                    }
                    // Like Git, an optional filter that fails leaves the content unfiltered.
                    Err(_) => {}
                }
            }
            None => {
                if file_attrs.filter_clean_required {
                    let name = file_attrs.filter_driver_name.as_deref().unwrap_or_default();
                    return Err(FilterError::ExternalFilterFailed {
                        driver: name.to_owned(),
                        path: rel_path.to_owned(),
                        phase: FilterPhase::Clean,
                    });
                }
            }
        }
    }

    // 2. working-tree-encoding: working tree bytes → UTF-8 for the object DB (Git `encode_to_git`).
    if let Some(ref enc) = file_attrs.working_tree_encoding {
        // Bare `working-tree-encoding` (boolean true) / `false` are rejected (Git
        // `git_path_check_encoding`).
        if enc == "set" || enc == "true" || enc == "false" {
            return Err(FilterError::InvalidWorkingTreeEncoding);
        }
        // `CONV_WRITE_OBJECT` → validate BOM rules and die on error (Git `encode_to_git`).
        let writing_object = opts.check_safecrlf;
        match decode_working_tree_bytes_to_utf8(&buf, rel_path, enc, writing_object) {
            Ok(decoded) => buf = decoded,
            Err(e) if writing_object => return Err(filter_detail(rel_path, e.to_string())),
            // Reads for diff/status/hashing keep the bytes as they are when they cannot be
            // re-encoded, as Git does without `CONV_WRITE_OBJECT`.
            Err(_) => {}
        }
    }

    // 3. Determine if we should do CRLF→LF conversion
    let would_convert = would_convert_on_input(conv, file_attrs, &buf);

    let mut convert_crlf_into_lf = would_convert && has_crlf(&buf);
    if convert_crlf_into_lf
        && clean_uses_autocrlf_index_guard(file_attrs, conv)
        && !opts.renormalize
        && opts.index_blob.is_some_and(has_crlf_in_index_blob)
    {
        convert_crlf_into_lf = false;
    }

    // 4. safecrlf check — Git simulates clean then smudge (`check_global_conv_flags_eol`).
    if would_convert && opts.check_safecrlf {
        check_safecrlf_roundtrip(conv, file_attrs, &buf, rel_path, convert_crlf_into_lf)?;
    }

    // 5. Actually convert CRLF → LF if the file has CRLFs
    if convert_crlf_into_lf {
        buf = crlf_to_lf(&buf);
    }

    // 6. Ident collapse (Git `ident_to_git` after filter / encode / crlf_to_git).
    if file_attrs.ident && count_ident_regions(&buf) {
        buf = collapse_ident(&buf);
    }

    Ok(buf)
}

/// Whether [`convert_to_git_with_opts`] returns every file at a path with these attributes
/// unchanged.
///
/// Decided from configuration and attributes alone, without looking at content: no clean or
/// process filter (required or not) and no `working-tree-encoding` applies, no `ident` collapse,
/// and no line-ending setting can turn CRLF into LF. When this holds, a work-tree file's blob id
/// can be computed by hashing the file as it is, for example in chunks without loading it whole.
///
/// # Parameters
///
/// - `conv` — line-ending configuration (`core.autocrlf`, `core.eol`).
/// - `attrs` — the path's resolved attributes.
#[must_use]
pub fn git_conversion_is_identity(conv: &ConversionConfig, attrs: &FileAttrs) -> bool {
    attrs.filter_process.is_none()
        && attrs.filter_clean.is_none()
        && !attrs.filter_clean_required
        && attrs.working_tree_encoding.is_none()
        && !attrs.ident
        && input_eol_rule(conv, attrs) == InputEolRule::Never
}

/// Whether a regular worktree file can be hashed or stored by streaming raw bytes without
/// loading it whole.
///
/// Matches [`git_conversion_is_identity`] when line-ending rules never apply. When the only
/// blocker is `core.autocrlf` / `text=auto` (`InputEolRule::UnlessBinary`), probes the first
/// 8000 bytes for a NUL (same as [`is_binary`]) and allows streaming when the file is binary.
///
/// Returns `false` when a clean filter, `working-tree-encoding`, `ident`, or a non-binary
/// file needs the autocrlf index guard's stage-0 blob. Binary files stream even when indexed,
/// matching Git's `convert.c` order (binary check before the index CRLF guard).
///
/// # Errors
///
/// Returns [`crate::error::Error::Io`] when the probe read or seek fails.
pub(crate) fn worktree_allows_raw_byte_stream(
    conv: &ConversionConfig,
    attrs: &FileAttrs,
    index_entry: Option<&crate::index::IndexEntry>,
    file: &mut std::fs::File,
    size: u64,
) -> crate::error::Result<bool> {
    use std::io::{Read, Seek, SeekFrom};

    if attrs.filter_process.is_some()
        || attrs.filter_clean.is_some()
        || attrs.filter_clean_required
        || attrs.working_tree_encoding.is_some()
        || attrs.ident
    {
        return Ok(false);
    }
    match input_eol_rule(conv, attrs) {
        InputEolRule::Never => Ok(true),
        InputEolRule::Always => Ok(false),
        InputEolRule::UnlessBinary => {
            let probe_len = usize::try_from(size).unwrap_or(8000).min(8000);
            if probe_len == 0 {
                return Ok(false);
            }
            file.seek(SeekFrom::Start(0))
                .map_err(crate::error::Error::Io)?;
            let mut buf = vec![0u8; probe_len];
            file.read_exact(&mut buf).map_err(crate::error::Error::Io)?;
            file.seek(SeekFrom::Start(0))
                .map_err(crate::error::Error::Io)?;
            if is_binary(&buf) {
                return Ok(true);
            }
            if index_entry.is_some() && clean_uses_autocrlf_index_guard(attrs, conv) {
                return Ok(false);
            }
            Ok(false)
        }
    }
}

/// Decide whether CRLF/LF conversion is configured for this file on input.
/// Returns true if the file *would* be subject to conversion (even if no
/// actual bytes need changing).
fn would_convert_on_input(conv: &ConversionConfig, attrs: &FileAttrs, data: &[u8]) -> bool {
    match input_eol_rule(conv, attrs) {
        InputEolRule::Never => false,
        InputEolRule::Always => true,
        InputEolRule::UnlessBinary => !is_binary(data),
    }
}

/// Whether CRLF/LF conversion applies on input, before looking at the content.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
enum InputEolRule {
    Never,
    Always,
    UnlessBinary,
}

fn input_eol_rule(conv: &ConversionConfig, attrs: &FileAttrs) -> InputEolRule {
    match attrs.crlf_legacy {
        CrlfLegacyAttr::Unset => return InputEolRule::Never,
        CrlfLegacyAttr::Input => return InputEolRule::UnlessBinary,
        CrlfLegacyAttr::Crlf => {
            if attrs.text == TextAttr::Unset {
                return InputEolRule::Never;
            }
            return InputEolRule::UnlessBinary;
        }
        CrlfLegacyAttr::Unspecified => {}
    }

    // If text is explicitly unset (-text or binary), never convert
    if attrs.text == TextAttr::Unset {
        return InputEolRule::Never;
    }

    // If eol attr is set, this implies text mode
    if attrs.eol != EolAttr::Unspecified {
        if attrs.text == TextAttr::Auto {
            return InputEolRule::UnlessBinary;
        }
        return InputEolRule::Always;
    }

    match attrs.text {
        TextAttr::Set => return InputEolRule::Always,
        TextAttr::Auto => return InputEolRule::UnlessBinary,
        _ => {}
    }

    // No text attribute: fall back to core.autocrlf
    match conv.autocrlf {
        AutoCrlf::True | AutoCrlf::Input => InputEolRule::UnlessBinary,
        AutoCrlf::False => InputEolRule::Never,
    }
}

/// Git-compatible stderr when `core.safecrlf` is `warn` (clean direction, CRLF→LF).
fn eprint_safecrlf_warn_crlf_to_lf(_rel_path: &str) {}

/// Git-compatible stderr when `core.safecrlf` is `warn` (clean direction, LF→CRLF).
fn eprint_safecrlf_warn_lf_to_crlf(_rel_path: &str) {}

/// Git `convert.c` `check_global_conv_flags_eol` after simulating clean + smudge.
fn check_safecrlf_roundtrip(
    conv: &ConversionConfig,
    file_attrs: &FileAttrs,
    data: &[u8],
    rel_path: &str,
    convert_crlf_into_lf: bool,
) -> Result<(), FilterError> {
    if conv.safecrlf == SafeCrlf::False {
        return Ok(());
    }

    let old_stats = git_text_stat(data);

    let mut new_stats = old_stats.clone();
    if convert_crlf_into_lf && new_stats.crlf > 0 {
        new_stats.lonelf += new_stats.crlf;
        new_stats.crlf = 0;
    }
    if will_convert_lf_to_crlf_from_stats(&new_stats, conv, file_attrs) {
        new_stats.crlf += new_stats.lonelf;
        new_stats.lonelf = 0;
    }

    if old_stats.crlf > 0 && new_stats.crlf == 0 {
        let err = FilterError::LineEndingWouldChange {
            detail: format!("CRLF would be replaced by LF in {rel_path}"),
        };
        if conv.safecrlf == SafeCrlf::True {
            return Err(err);
        }
        eprint_safecrlf_warn_crlf_to_lf(rel_path);
    } else if old_stats.lonelf > 0 && new_stats.lonelf == 0 {
        let err = FilterError::LineEndingWouldChange {
            detail: format!("LF would be replaced by CRLF in {rel_path}"),
        };
        if conv.safecrlf == SafeCrlf::True {
            return Err(err);
        }
        eprint_safecrlf_warn_lf_to_crlf(rel_path);
    }

    Ok(())
}

/// Replace CRLF with LF.
pub fn crlf_to_lf(data: &[u8]) -> Vec<u8> {
    let mut out = Vec::with_capacity(data.len());
    let mut i = 0;
    while i < data.len() {
        if i + 1 < data.len() && data[i] == b'\r' && data[i + 1] == b'\n' {
            out.push(b'\n');
            i += 2;
        } else {
            out.push(data[i]);
            i += 1;
        }
    }
    out
}

/// Replace lone LF with CRLF.
pub fn lf_to_crlf(data: &[u8]) -> Vec<u8> {
    let mut out = Vec::with_capacity(data.len() + data.len() / 10);
    let mut i = 0;
    while i < data.len() {
        if data[i] == b'\n' && (i == 0 || data[i - 1] != b'\r') {
            out.push(b'\r');
            out.push(b'\n');
        } else {
            out.push(data[i]);
        }
        i += 1;
    }
    out
}

// ---------------------------------------------------------------------------
// Output (checkout / smudge) direction
// ---------------------------------------------------------------------------

/// Convert data from the object database for writing to the working tree
/// (the "smudge" direction).
///
/// This handles (Git `convert_to_working_tree_ca_internal` order):
/// 1. Ident keyword expansion
/// 2. LF → CRLF conversion based on config + attributes
/// 3. `working-tree-encoding` (UTF-8 blob → working tree bytes)
/// 4. Smudge filter execution
///
/// Returns `Ok(None)` when the process filter returned `status=delayed` and `delayed_checkout` was
/// provided (Git `delayed_checkout`); the path is queued for delayed checkout finish.
#[allow(clippy::too_many_arguments)]
pub fn convert_to_worktree(
    data: &[u8],
    rel_path: &str,
    conv: &ConversionConfig,
    file_attrs: &FileAttrs,
    oid_hex: Option<&str>,
    smudge_meta: Option<&FilterSmudgeMeta>,
    delayed_checkout: Option<&mut crate::filter_process::DelayedProcessCheckout>,
    filter_process: Option<&FilterProcessState>,
) -> Result<Option<Vec<u8>>, FilterError> {
    convert_to_worktree_with_runner(
        data,
        rel_path,
        conv,
        file_attrs,
        oid_hex,
        smudge_meta,
        delayed_checkout,
        filter_process,
        None,
    )
}

/// Like [`convert_to_worktree`] with an explicit filter [`CommandRunner`].
#[expect(clippy::too_many_arguments)]
pub fn convert_to_worktree_with_runner(
    data: &[u8],
    rel_path: &str,
    conv: &ConversionConfig,
    file_attrs: &FileAttrs,
    oid_hex: Option<&str>,
    smudge_meta: Option<&FilterSmudgeMeta>,
    delayed_checkout: Option<&mut crate::filter_process::DelayedProcessCheckout>,
    filter_process: Option<&FilterProcessState>,
    command_runner: Option<&dyn CommandRunner>,
) -> Result<Option<Vec<u8>>, FilterError> {
    let mut buf = data.to_vec();
    let owned_filter_state;
    let filter_state = if let Some(s) = filter_process {
        s
    } else {
        owned_filter_state = FilterProcessState::new(Arc::clone(&conv.command_runner));
        &owned_filter_state
    };
    let runner = command_runner.unwrap_or(conv.command_runner.as_ref());

    // 1. Ident expansion
    if file_attrs.ident {
        if let Some(oid) = oid_hex {
            buf = expand_ident(&buf, oid);
        }
    }

    let can_delay_smudge = delayed_checkout.is_some()
        && file_attrs.working_tree_encoding.is_none()
        && !file_attrs.ident
        && file_attrs
            .filter_process
            .as_deref()
            .is_some_and(|c| !c.is_empty())
        && !should_convert_to_crlf(conv, file_attrs, &buf)
        && file_attrs
            .filter_process
            .as_deref()
            .is_some_and(|c| crate::filter_process::process_filter_supports_delay(filter_state, c));

    // 2. LF→CRLF for working tree
    let should_convert = should_convert_to_crlf(conv, file_attrs, &buf);
    if should_convert {
        buf = lf_to_crlf(&buf);
    }

    // 3. working-tree-encoding (Git `encode_to_worktree`)
    if let Some(ref enc) = file_attrs.working_tree_encoding {
        buf = encode_utf8_blob_to_working_tree_bytes(&buf, rel_path, enc)
            .map_err(|e| filter_detail(rel_path, e.to_string()))?;
    }

    // 4. Smudge filter — process driver overrides shell smudge
    let driver = file_attrs.filter_driver_name.as_deref().unwrap_or("");
    if let Some(ref proc_cmd) = file_attrs.filter_process {
        if file_attrs.filter_smudge_required {
            start_required_process_filter(
                filter_state,
                proc_cmd,
                driver,
                rel_path,
                FilterPhase::Smudge,
            )?;
        }
        let smudge_out = match apply_process_smudge(
            filter_state,
            proc_cmd,
            rel_path,
            &buf,
            smudge_meta,
            can_delay_smudge,
        ) {
            Ok(out) => out,
            Err(e) => {
                if file_attrs.filter_smudge_required {
                    return Err(external_filter_failed(
                        rel_path,
                        driver,
                        FilterPhase::Smudge,
                    ));
                }
                if e.starts_with("filter status: abort") {
                    filter_state.disable_process_filter(proc_cmd);
                }
                return Ok(Some(buf));
            }
        };
        let Some(out) = smudge_out else {
            let Some(q) = delayed_checkout else {
                return Err(filter_detail(
                    rel_path,
                    format!("internal error: delayed smudge without checkout queue for {rel_path}"),
                ));
            };
            q.push_delayed(
                proc_cmd.clone(),
                rel_path.to_string(),
                smudge_meta.cloned().unwrap_or_default(),
            );
            return Ok(None);
        };
        buf = out;
    } else {
        match file_attrs.filter_smudge.as_ref() {
            Some(smudge_cmd) => {
                match run_filter_command(runner, smudge_cmd, &buf, rel_path, Some(filter_state)) {
                    Ok(filtered) => buf = filtered,
                    Err(e) if file_attrs.filter_smudge_required => {
                        return Err(e.into_required_error(driver, rel_path, FilterPhase::Smudge));
                    }
                    Err(_) => {}
                }
            }
            None => {
                if file_attrs.filter_smudge_required {
                    return Err(external_filter_failed(
                        rel_path,
                        driver,
                        FilterPhase::Smudge,
                    ));
                }
            }
        }
    }

    Ok(Some(buf))
}

/// Like [`convert_to_worktree`] without delayed-checkout queueing (always materializes or errors).
pub fn convert_to_worktree_eager(
    data: &[u8],
    rel_path: &str,
    conv: &ConversionConfig,
    file_attrs: &FileAttrs,
    oid_hex: Option<&str>,
    smudge_meta: Option<&FilterSmudgeMeta>,
    filter_process: Option<&FilterProcessState>,
) -> Result<Vec<u8>, FilterError> {
    match convert_to_worktree(
        data,
        rel_path,
        conv,
        file_attrs,
        oid_hex,
        smudge_meta,
        None,
        filter_process,
    )? {
        Some(v) => Ok(v),
        None => Err(filter_detail(
            rel_path,
            format!("internal error: unexpected delayed smudge for {rel_path}"),
        )),
    }
}

/// Whether [`convert_to_worktree`] returns every blob at a path with these attributes unchanged.
///
/// Decided from configuration and attributes alone, without looking at content: no `ident`,
/// `working-tree-encoding`, smudge or process filter applies, and no line-ending setting can
/// turn LF into CRLF. When this holds, a blob can be copied into the work tree verbatim, for
/// example streamed straight from the object database.
///
/// # Parameters
///
/// - `conv` — line-ending configuration (`core.autocrlf`, `core.eol`).
/// - `attrs` — the path's resolved attributes.
#[must_use]
pub fn worktree_conversion_is_identity(conv: &ConversionConfig, attrs: &FileAttrs) -> bool {
    !attrs.ident
        && attrs.working_tree_encoding.is_none()
        && attrs.filter_smudge.is_none()
        && attrs.filter_process.is_none()
        && !attrs.filter_smudge_required
        && !crlf_output_possible(conv, attrs)
}

/// Whether some content at this path could be converted LF→CRLF on output.
///
/// `false` means [`should_convert_to_crlf`] is `false` for every blob.
fn crlf_output_possible(conv: &ConversionConfig, attrs: &FileAttrs) -> bool {
    match attrs.crlf_legacy {
        CrlfLegacyAttr::Unset | CrlfLegacyAttr::Input => return false,
        CrlfLegacyAttr::Crlf => return attrs.text != TextAttr::Unset,
        CrlfLegacyAttr::Unspecified => {}
    }
    if attrs.text == TextAttr::Unset {
        return false;
    }
    if attrs.eol != EolAttr::Unspecified {
        return attrs.eol == EolAttr::Crlf;
    }
    match attrs.text {
        TextAttr::Set | TextAttr::Auto => output_eol_is_crlf(conv),
        _ => conv.autocrlf == AutoCrlf::True,
    }
}

/// Decide whether to convert LF→CRLF on output (working tree / smudge direction).
#[must_use]
pub fn should_convert_to_crlf(conv: &ConversionConfig, attrs: &FileAttrs, data: &[u8]) -> bool {
    if !crlf_output_possible(conv, attrs) {
        return false;
    }
    match attrs.crlf_legacy {
        CrlfLegacyAttr::Unset | CrlfLegacyAttr::Input => return false,
        CrlfLegacyAttr::Crlf => {
            if attrs.text == TextAttr::Unset {
                return false;
            }
            // Legacy `crlf` (set) forces CRLF on checkout (even for paths Git
            // would otherwise treat as binary; see t0020 "t* crlf" + `three`).
            return true;
        }
        CrlfLegacyAttr::Unspecified => {}
    }

    // If text is explicitly unset, never convert
    if attrs.text == TextAttr::Unset {
        return false;
    }

    // If there's an explicit eol attribute
    if attrs.eol != EolAttr::Unspecified {
        if attrs.text == TextAttr::Auto && is_binary(data) {
            return false;
        }
        if attrs.eol != EolAttr::Crlf {
            return false;
        }
        // `text=auto` + `eol=crlf` → Git `CRLF_AUTO_CRLF` (safe mixed handling).
        if attrs.text == TextAttr::Auto {
            return auto_crlf_should_smudge_lf_to_crlf(data);
        }
        // Explicit `eol=crlf` with `text` set, etc. → `CRLF_TEXT_CRLF` (always normalize).
        return true;
    }

    // If text is explicitly set, use eol config
    if attrs.text == TextAttr::Set {
        return output_eol_is_crlf(conv);
    }

    if attrs.text == TextAttr::Auto {
        if is_binary(data) {
            return false;
        }
        if !output_eol_is_crlf(conv) {
            return false;
        }
        return auto_crlf_should_smudge_lf_to_crlf(data);
    }

    // No text attribute: fall back to core.autocrlf
    match conv.autocrlf {
        AutoCrlf::True => {
            if is_binary(data) {
                return false;
            }
            auto_crlf_should_smudge_lf_to_crlf(data)
        }
        AutoCrlf::Input | AutoCrlf::False => false,
    }
}

/// Whether the output EOL should be CRLF based on config.
fn output_eol_is_crlf(conv: &ConversionConfig) -> bool {
    // Git `text_eol_is_crlf`: autocrlf=input forces LF output before `core.eol` is consulted.
    if conv.autocrlf == AutoCrlf::Input {
        return false;
    }
    if conv.autocrlf == AutoCrlf::True {
        return true;
    }
    match conv.eol {
        CoreEol::Crlf => true,
        CoreEol::Lf => false,
        CoreEol::Native => {
            // On Unix, native is LF
            cfg!(windows)
        }
    }
}

/// Expand `$Id$` → `$Id: <oid>$` in data.
///
/// Matches Git's `ident_to_worktree` in `convert.c`: same-line `$` terminator, and foreign
/// idents (internal spaces before the closing `$`) are left unchanged.
fn expand_ident(data: &[u8], oid: &str) -> Vec<u8> {
    if !count_ident_regions(data) {
        return data.to_vec();
    }
    let replacement = format!("$Id: {oid} $");
    let mut out = Vec::with_capacity(data.len() + 60);
    let mut i = 0;
    while i < data.len() {
        if data[i] != b'$' {
            out.push(data[i]);
            i += 1;
            continue;
        }
        if i + 3 > data.len() || data[i + 1] != b'I' || data[i + 2] != b'd' {
            out.push(data[i]);
            i += 1;
            continue;
        }
        let after_id = i + 3;
        let ch = data.get(after_id).copied();
        match ch {
            Some(b'$') => {
                out.extend_from_slice(replacement.as_bytes());
                i = after_id + 1;
            }
            Some(b':') => {
                let rest = &data[after_id + 1..];
                let line_end = rest
                    .iter()
                    .position(|&b| b == b'\n' || b == b'\r')
                    .unwrap_or(rest.len());
                let line = &rest[..line_end];
                let Some(dollar_rel) = line.iter().position(|&b| b == b'$') else {
                    out.push(data[i]);
                    i += 1;
                    continue;
                };
                if line[..dollar_rel].contains(&b'\n') {
                    out.push(data[i]);
                    i += 1;
                    continue;
                }
                // Foreign ident (Git `ident_to_worktree`): first space in the payload after the
                // byte following `:` must not be the last character before `$`.
                let payload = &line[..dollar_rel];
                let foreign = payload.len() > 1
                    && payload[1..]
                        .iter()
                        .position(|&b| b == b' ')
                        .is_some_and(|rel| {
                            let pos = 1 + rel;
                            pos < payload.len().saturating_sub(1)
                        });
                if foreign {
                    out.push(data[i]);
                    i += 1;
                    continue;
                }
                out.extend_from_slice(replacement.as_bytes());
                i = after_id + 1 + dollar_rel + 1;
            }
            _ => {
                out.push(data[i]);
                i += 1;
            }
        }
    }
    out
}

/// Whether the buffer contains any `$Id$` / `$Id: ... $` regions Git would rewrite (`count_ident`).
fn count_ident_regions(data: &[u8]) -> bool {
    let mut i = 0usize;
    while i < data.len() {
        if data[i] != b'$' {
            i += 1;
            continue;
        }
        if i + 3 > data.len() || data[i + 1] != b'I' || data[i + 2] != b'd' {
            i += 1;
            continue;
        }
        let after = i + 3;
        match data.get(after).copied() {
            Some(b'$') => return true,
            Some(b':') => {
                let mut j = after + 1;
                let mut found = false;
                while j < data.len() {
                    match data[j] {
                        b'$' => {
                            found = true;
                            break;
                        }
                        // Git `ident_to_git` / `count_ident`: LF ends the scan; CR may appear
                        // inside the payload before the closing `$`.
                        b'\n' => break,
                        _ => j += 1,
                    }
                }
                if found {
                    return true;
                }
                i += 1;
            }
            _ => i += 1,
        }
    }
    false
}

/// Collapse `$Id: ... $` back to `$Id$`.
pub fn collapse_ident(data: &[u8]) -> Vec<u8> {
    let mut out = Vec::with_capacity(data.len());
    let mut i = 0;
    while i < data.len() {
        if i + 4 <= data.len() && &data[i..i + 4] == b"$Id:" {
            let rest = &data[i + 4..];
            let line_end = rest.iter().position(|&b| b == b'\n').unwrap_or(rest.len());
            let line = &rest[..line_end];
            if let Some(end) = line.iter().position(|&b| b == b'$') {
                out.extend_from_slice(b"$Id$");
                i += 4 + end + 1;
                continue;
            }
        }
        out.push(data[i]);
        i += 1;
    }
    out
}

/// Shell-quote `s` with single quotes, matching Git's `sq_quote_buf` (`'` → `'\''`).
fn sq_quote_buf(s: &str) -> String {
    let mut out = String::with_capacity(s.len() + 2);
    out.push('\'');
    for ch in s.chars() {
        if ch == '\'' {
            out.push_str("'\\''");
        } else {
            out.push(ch);
        }
    }
    out.push('\'');
    out
}

/// Expand Git filter command placeholders: `%%` → `%`, `%f` → quoted repository-relative path.
fn expand_filter_command(cmd: &str, rel_path: &str) -> String {
    let mut out = String::with_capacity(cmd.len() + rel_path.len() + 8);
    let mut chars = cmd.chars().peekable();
    while let Some(c) = chars.next() {
        if c == '%' {
            match chars.peek() {
                Some('%') => {
                    chars.next();
                    out.push('%');
                }
                Some('f') => {
                    chars.next();
                    out.push_str(&sq_quote_buf(rel_path));
                }
                _ => out.push('%'),
            }
        } else {
            out.push(c);
        }
    }
    out
}

/// Run a filter command, piping data through stdin→stdout.
pub fn run_filter(
    runner: &dyn CommandRunner,
    cmd: &str,
    data: &[u8],
    rel_path: &str,
) -> Result<Vec<u8>, FilterError> {
    run_filter_command(runner, cmd, data, rel_path, None).map_err(|e| match e {
        FilterRunError::Spawn { source, .. } | FilterRunError::Io(source) => {
            filter_detail(rel_path, source.to_string())
        }
        FilterRunError::Exit(status) => FilterError::Failed { status },
    })
}

/// How a one-shot clean/smudge command failed.
#[derive(Debug)]
enum FilterRunError {
    /// `program` could not be spawned.
    Spawn {
        program: PathBuf,
        source: std::io::Error,
    },
    /// Piping data to or from the running command failed.
    Io(std::io::Error),
    /// The command exited with this non-zero status.
    Exit(i32),
}

impl FilterRunError {
    /// The error for a failure of a `filter.<driver>.required` filter.
    fn into_required_error(self, driver: &str, rel_path: &str, phase: FilterPhase) -> FilterError {
        match self {
            Self::Spawn { program, source } => {
                filter_spawn_failed(driver, rel_path, phase, &program, &source)
            }
            Self::Io(_) | Self::Exit(_) => external_filter_failed(rel_path, driver, phase),
        }
    }
}

fn filter_spawn_failed(
    driver: &str,
    rel_path: &str,
    phase: FilterPhase,
    program: &Path,
    source: &std::io::Error,
) -> FilterError {
    FilterError::ExternalFilterSpawnFailed {
        driver: driver.to_owned(),
        path: rel_path.to_owned(),
        phase,
        program: program.display().to_string(),
        kind: source.kind(),
        detail: source.to_string(),
    }
}

fn run_filter_command(
    runner: &dyn CommandRunner,
    cmd: &str,
    data: &[u8],
    rel_path: &str,
    filter_state: Option<&FilterProcessState>,
) -> Result<Vec<u8>, FilterRunError> {
    let expanded = expand_filter_command(cmd, rel_path);
    let mut spec = CommandSpec::sh_dash_c_filter(expanded);
    if let Some(st) = filter_state {
        st.apply_subprocess_to_spec(&mut spec);
    }
    spec.stdin = CommandStdin::Pipe(data.to_vec());
    let output = runner
        .spawn(&spec)
        .map_err(|source| FilterRunError::Spawn {
            program: spec.program.clone(),
            source,
        })?
        .wait_with_output()
        .map_err(FilterRunError::Io)?;
    if !output.status.success {
        return Err(FilterRunError::Exit(output.status.code.unwrap_or(1)));
    }
    Ok(output.stdout)
}

/// Start the long-running filter `proc_cmd` before a required filter uses it, so a spawn
/// failure is reported with its cause instead of as a generic filter failure.
fn start_required_process_filter(
    filter_state: &FilterProcessState,
    proc_cmd: &str,
    driver: &str,
    rel_path: &str,
    phase: FilterPhase,
) -> Result<(), FilterError> {
    if filter_state.process_filter_is_disabled(proc_cmd) {
        return Ok(());
    }
    match crate::filter_process::start_process_filter(filter_state, proc_cmd) {
        Err(crate::filter_process::ProcessFilterStartError::Spawn(source)) => Err(
            filter_spawn_failed(driver, rel_path, phase, Path::new("sh"), &source),
        ),
        // Handshake failures surface from the filter call itself, as before.
        Err(crate::filter_process::ProcessFilterStartError::Other(_)) | Ok(()) => Ok(()),
    }
}

// Re-export AttrRule type is internal, but we expose the vec through load_gitattributes.
// The public API uses the opaque Vec from load_gitattributes + get_file_attrs.

/// Opaque type alias for loaded gitattributes rules.
pub type GitAttributes = Vec<AttrRule>;

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

    /// Attribute and config combinations covering every line-ending branch.
    fn conversion_matrix() -> Vec<(ConversionConfig, FileAttrs)> {
        let mut out = Vec::new();
        for autocrlf in [AutoCrlf::False, AutoCrlf::True, AutoCrlf::Input] {
            for eol in [CoreEol::Lf, CoreEol::Crlf, CoreEol::Native] {
                for text in [
                    TextAttr::Set,
                    TextAttr::Auto,
                    TextAttr::Unset,
                    TextAttr::Unspecified,
                ] {
                    for attr_eol in [EolAttr::Unspecified, EolAttr::Lf, EolAttr::Crlf] {
                        for crlf_legacy in [
                            CrlfLegacyAttr::Unspecified,
                            CrlfLegacyAttr::Crlf,
                            CrlfLegacyAttr::Input,
                            CrlfLegacyAttr::Unset,
                        ] {
                            let conv = ConversionConfig {
                                autocrlf,
                                eol,
                                safecrlf: SafeCrlf::False,
                                command_runner: system_command_runner(),
                            };
                            let attrs = FileAttrs {
                                text,
                                eol: attr_eol,
                                crlf_legacy,
                                ..FileAttrs::default()
                            };
                            out.push((conv, attrs));
                        }
                    }
                }
            }
        }
        out
    }

    const SAMPLES: [&[u8]; 4] = [
        b"a\nb\n",
        b"a\r\nb\r\n",
        b"mixed\r\nends\n",
        b"bin\0\r\nary\n",
    ];

    #[test]
    fn worktree_identity_implies_no_output_conversion_for_any_content() {
        let mut identities = 0;
        for (conv, attrs) in conversion_matrix() {
            if !worktree_conversion_is_identity(&conv, &attrs) {
                continue;
            }
            identities += 1;
            for data in SAMPLES {
                let out = convert_to_worktree(data, "f", &conv, &attrs, None, None, None, None)
                    .expect("convert")
                    .expect("not delayed");
                assert_eq!(out, data, "{conv:?} {attrs:?} changed {data:?}");
            }
        }
        assert!(identities > 0);
    }

    #[test]
    fn git_identity_implies_no_input_conversion_for_any_content() {
        let mut identities = 0;
        for (conv, attrs) in conversion_matrix() {
            if !git_conversion_is_identity(&conv, &attrs) {
                continue;
            }
            identities += 1;
            for data in SAMPLES {
                let out = convert_to_git_with_opts(
                    data,
                    "f",
                    &conv,
                    &attrs,
                    ConvertToGitOpts {
                        index_blob: None,
                        renormalize: false,
                        check_safecrlf: false,
                        filter_process: None,
                        command_runner: None,
                    },
                )
                .expect("convert");
                assert_eq!(out, data, "{conv:?} {attrs:?} changed {data:?}");
            }
        }
        assert!(identities > 0);
    }

    #[test]
    fn raw_byte_streaming_under_autocrlf_when_binary() {
        use std::io::{Seek, SeekFrom, Write};

        let mut conv = ConversionConfig::from_config(&ConfigSet::new());
        conv.autocrlf = AutoCrlf::True;
        let attrs = FileAttrs::default();
        assert!(!git_conversion_is_identity(&conv, &attrs));

        let nul_file = tempfile::NamedTempFile::new().expect("temp");
        nul_file.as_file().write_all(&[0u8; 9000]).expect("write");
        let mut with_nul = nul_file.reopen().expect("reopen");
        assert!(worktree_allows_raw_byte_stream(&conv, &attrs, None, &mut with_nul, 9000).unwrap());

        let text_file = tempfile::NamedTempFile::new().expect("temp");
        text_file
            .as_file()
            .write_all(b"hello\r\nworld")
            .expect("write");
        let mut text = text_file.reopen().expect("reopen");
        assert!(!worktree_allows_raw_byte_stream(&conv, &attrs, None, &mut text, 12).unwrap());

        let mut data = vec![1u8; 20_000];
        data[500] = 0;
        let huge_file = tempfile::NamedTempFile::new().expect("temp");
        huge_file.as_file().write_all(&data).expect("write");
        let mut huge = huge_file.reopen().expect("reopen");
        huge.seek(SeekFrom::Start(10_000)).expect("seek");
        assert!(worktree_allows_raw_byte_stream(&conv, &attrs, None, &mut huge, 20_000).unwrap());

        use crate::index::{IndexEntry, MODE_REGULAR};
        use crate::objects::ObjectId;
        let indexed = IndexEntry {
            ctime_sec: 1,
            ctime_nsec: 0,
            mtime_sec: 1,
            mtime_nsec: 0,
            dev: 0,
            ino: 0,
            mode: MODE_REGULAR,
            uid: 0,
            gid: 0,
            size: 9000,
            oid: ObjectId::zero(),
            flags: 0,
            flags_extended: None,
            path: b"f.bin".to_vec(),
            base_index_pos: 0,
        };
        assert!(clean_uses_autocrlf_index_guard(&attrs, &conv));
        let mut again = nul_file.reopen().expect("reopen");
        assert!(
            worktree_allows_raw_byte_stream(&conv, &attrs, Some(&indexed), &mut again, 9000)
                .unwrap()
        );
    }

    #[test]
    fn identity_checks_see_filters_ident_and_encoding() {
        let conv = ConversionConfig::from_config(&ConfigSet::new());
        assert!(worktree_conversion_is_identity(
            &conv,
            &FileAttrs::default()
        ));
        assert!(git_conversion_is_identity(&conv, &FileAttrs::default()));
        let with = |f: &dyn Fn(&mut FileAttrs)| {
            let mut attrs = FileAttrs::default();
            f(&mut attrs);
            attrs
        };
        for attrs in [
            with(&|a| a.ident = true),
            with(&|a| a.working_tree_encoding = Some("UTF-16".into())),
            with(&|a| a.filter_smudge = Some("cat".into())),
            with(&|a| a.filter_process = Some("proc".into())),
            with(&|a| a.filter_smudge_required = true),
        ] {
            assert!(!worktree_conversion_is_identity(&conv, &attrs), "{attrs:?}");
        }
        for attrs in [
            with(&|a| a.ident = true),
            with(&|a| a.working_tree_encoding = Some("UTF-16".into())),
            with(&|a| a.filter_clean = Some("cat".into())),
            with(&|a| a.filter_process = Some("proc".into())),
            with(&|a| a.filter_clean_required = true),
        ] {
            assert!(!git_conversion_is_identity(&conv, &attrs), "{attrs:?}");
        }
    }

    #[test]
    fn collapse_ident_allows_cr_before_closing_dollar() {
        let data = b"$Id: abc\r$\n";
        assert_eq!(collapse_ident(data), b"$Id$\n");
    }

    #[test]
    fn convert_to_git_collapses_ident_with_cr_before_closing_dollar() {
        let conv = ConversionConfig::from_config(&ConfigSet::new());
        let mut attrs = FileAttrs::default();
        attrs.ident = true;
        let data = b"$Id: abc\r$\n";
        let out = convert_to_git(data, "f", &conv, &attrs).expect("convert");
        assert_eq!(out, b"$Id$\n");
    }

    #[test]
    fn convert_to_git_collapses_expanded_ident_to_keyword() {
        let conv = ConversionConfig::from_config(&ConfigSet::new());
        let mut attrs = FileAttrs::default();
        attrs.ident = true;
        let expanded = b"$Id: 10d2a9379e5d822c92a372599358dd0e616aadac $\nfoo\n";
        let out = convert_to_git(expanded, "id.txt", &conv, &attrs).expect("convert");
        assert_eq!(out, b"$Id$\nfoo\n");
    }

    #[test]
    fn test_crlf_to_lf() {
        assert_eq!(crlf_to_lf(b"hello\r\nworld\r\n"), b"hello\nworld\n");
        assert_eq!(crlf_to_lf(b"hello\nworld\n"), b"hello\nworld\n");
        assert_eq!(crlf_to_lf(b"hello\r\n"), b"hello\n");
    }

    #[test]
    fn test_lf_to_crlf() {
        assert_eq!(lf_to_crlf(b"hello\nworld\n"), b"hello\r\nworld\r\n");
        assert_eq!(lf_to_crlf(b"hello\r\nworld\r\n"), b"hello\r\nworld\r\n");
    }

    #[test]
    fn test_has_crlf() {
        assert!(has_crlf(b"hello\r\nworld"));
        assert!(!has_crlf(b"hello\nworld"));
    }

    #[test]
    fn smudge_mixed_line_endings_unchanged_with_autocrlf_true() {
        let mut blob = Vec::new();
        for part in [
            b"Oh\n".as_slice(),
            b"here\n",
            b"is\n",
            b"CRLF\r\n",
            b"in\n",
            b"text\n",
        ] {
            blob.extend_from_slice(part);
        }
        let conv = ConversionConfig {
            autocrlf: AutoCrlf::True,
            eol: CoreEol::Lf,
            safecrlf: SafeCrlf::False,
            command_runner: system_command_runner(),
        };
        let attrs = FileAttrs::default();
        let out =
            convert_to_worktree_eager(&blob, "mixed", &conv, &attrs, None, None, None).unwrap();
        assert_eq!(out, blob);
    }

    #[test]
    fn smudge_lf_only_gets_crlf_with_autocrlf_true() {
        let blob = b"a\nb\n";
        let conv = ConversionConfig {
            autocrlf: AutoCrlf::True,
            eol: CoreEol::Lf,
            safecrlf: SafeCrlf::False,
            command_runner: system_command_runner(),
        };
        let attrs = FileAttrs::default();
        let out = convert_to_worktree_eager(blob, "x", &conv, &attrs, None, None, None).unwrap();
        assert_eq!(out, b"a\r\nb\r\n");
    }

    #[test]
    fn test_is_binary() {
        assert!(is_binary(b"hello\0world"));
        assert!(!is_binary(b"hello world"));
    }

    #[test]
    fn attr_dir_only_pattern_does_not_match_same_named_file() {
        let rules = parse_gitattributes_content("dir-only-pattern/ text\n");
        let rule = &rules[0];
        assert!(!attr_rule_matches(rule, "parent/dir-only-pattern", false));
        assert!(attr_rule_matches(rule, "dir-only-pattern", true));
    }

    #[test]
    fn test_expand_collapse_ident() {
        let data = b"$Id$";
        let expanded = expand_ident(data, "abc123");
        assert_eq!(expanded, b"$Id: abc123 $");
        let collapsed = collapse_ident(&expanded);
        assert_eq!(collapsed, b"$Id$");
    }

    #[test]
    fn expand_ident_does_not_span_lines_for_partial_keyword() {
        let data = b"$Id: NoTerminatingSymbol\n$Id: deadbeef $\n";
        let expanded = expand_ident(data, "newoid");
        assert_eq!(expanded, b"$Id: NoTerminatingSymbol\n$Id: newoid $\n");
    }

    #[test]
    fn expand_ident_preserves_foreign_id_with_internal_spaces() {
        let data = b"$Id: Foreign Commit With Spaces $\n";
        let expanded = expand_ident(data, "abc");
        assert_eq!(expanded, data);
    }

    #[test]
    fn expand_filter_command_percent_f_quotes_path() {
        let s = expand_filter_command("sh ./x.sh %f --extra", "name  with 'sq'");
        assert_eq!(s, "sh ./x.sh 'name  with '\\''sq'\\''' --extra");
        assert_eq!(expand_filter_command("a %% b", "p"), "a % b");
    }
}