avatarr_parser/language/

subtitle.rs

// Ported from Sonarr v4.0.17.2952 (97e85a90):
//   src/NzbDrone.Core/Parser/LanguageParser.cs (lines 217-335 + regex
//   declarations 32-34, 36).
//   src/NzbDrone.Core/Parser/Model/SubtitleTitleInfo.cs (return-type fields).
//
// Public entry points:
//   - [`parse_subtitle_language`] mirrors C# `ParseSubtitleLanguage`.
//   - [`parse_basic_subtitle`] mirrors C# `ParseBasicSubtitle`.
//   - [`parse_subtitle_language_information`] mirrors C# `ParseSubtitleLanguageInformation`.
//   - [`parse_language_tags`] mirrors C# `ParseLanguageTags`.
//
// **Plan-spec / Rust-regex divergences resolved in favour of C# semantics
// (per standing rule #1):**
//
// 1. Rust's `regex` crate forbids duplicate named groups. C#
//    `SubtitleLanguageRegex` declares `(?<tags>...)` twice (once before
//    iso_code, once after) and relies on `match.Groups["tags"].Captures`
//    enumerating BOTH. The Rust port drops both `(?<tags>...)` captures
//    from the structural regex; tag-token positions are recovered by
//    walking outward from the iso_code position post-match, applying
//    the same vocabulary (`forced|foreign|default|cc|psdh|sdh`). The
//    walking algorithm is faithful to C#'s greedy `(?:[-_. ]<tag>)*` /
//    lazy `.+?` interaction: tags are only collected when they are
//    contiguous with the iso_code via tag-or-iso-only segments.
//
// 2. Same problem in `SubtitleLanguageTitleRegex`: the C# pattern names
//    `iso_code` twice (pre-title and post-title) AND `tags1`/`tags2`
//    repeat inside their respective `\.(...)*` clusters. The Rust port
//    captures only the structural anchor (the title group) and walks
//    the pre-title and post-title segment lists in code to enumerate
//    iso_codes + tags. This mirrors C#'s `Captures.Count` semantics
//    (the "exactly one iso_code" gate at line 267).
//
// 3. C# `SubtitleTitleRegex` (line 36) uses `(?<!\d+)\d{1,3}(?!\d+)`
//    for the copy-number tail. Rust regex supports neither lookahead
//    nor lookbehind; the port relaxes the regex to `\d{1,3}` and
//    rejects matches whose copy-number is adjacent to other digits.
//
// 4. C# uses `Path.GetFileNameWithoutExtension` to strip the final
//    extension before matching. We reproduce that with a string-based
//    helper that takes everything before the last `.` (or the whole
//    string if no `.`). This avoids `std::path::Path` rewrites of
//    forward/backward slashes that would conflict with C# `Path`
//    semantics on POSIX hosts.

use crate::language::{Language, iso_languages};
use once_cell::sync::Lazy;
use regex::Regex;

/// Tag vocabulary matching the alternation in `SubtitleLanguageRegex`
/// and `SubtitleLanguageTitleRegex`. Order matters only for negative
/// regression checks; lookups are case-insensitive in all consumers.
const TAG_VOCAB: &[&str] = &["forced", "foreign", "default", "cc", "psdh", "sdh"];

/// Outcome of [`parse_basic_subtitle`] / mirrors C# `SubtitleTitleInfo`
/// minimum surface (the fields `ParseBasicSubtitle` actually populates).
///
/// `ParseBasicSubtitle` only sets `TitleFirst` (always `false`),
/// `LanguageTags`, and `Language`. `RawTitle`, `Title`, `Copy` are
/// left at their default values (`None`, `None`, `0`).
#[derive(Debug, Clone, PartialEq, Eq, Default)]
pub struct SubtitleTitleInfo {
    /// Mirrors C# `LanguageTags`. Always lowercased per
    /// `tag.Value.ToLower()` in `ParseSubtitleLanguageInformation`.
    pub language_tags: Vec<String>,
    /// Mirrors C# `Language`. `Language::Unknown` when the file
    /// carried no resolvable language code.
    pub language: Language,
    /// Mirrors C# `RawTitle`. `None` when no title segment was parsed
    /// (e.g. `ParseBasicSubtitle` short-circuit, or
    /// `ParseSubtitleLanguageInformation` falling back to
    /// `ParseBasicSubtitle`).
    pub raw_title: Option<String>,
    /// Mirrors C# `Title`. Either the bare `RawTitle` or the
    /// `(?<title>.+)` capture from `SubtitleTitleRegex` once the
    /// trailing copy-number is stripped.
    pub title: Option<String>,
    /// Mirrors C# `Copy`. Zero when no trailing copy-number suffix
    /// was found.
    pub copy: i32,
    /// Mirrors C# `TitleFirst`. `true` when the parsed title segment
    /// precedes any tag/iso_code segment in the filename (i.e. when
    /// `tags1` is empty in C#'s regex).
    pub title_first: bool,
}

/// Regex constants ported from `LanguageParser.cs:32-34, 36`. Each
/// `Lazy<Regex>` corresponds to a `private static readonly Regex`
/// in C#. Where the C# pattern uses a feature Rust regex lacks
/// (duplicate named groups, lookarounds), the Rust regex is the
/// structural superset and a sibling helper applies the missing
/// constraint post-match. See the file header for the complete list.
pub mod regexes {
    use super::{Lazy, Regex};

    /// `SubtitleLanguageRegex` from `LanguageParser.cs:32`.
    ///
    /// C# pattern (with two `(?<tags>...)` groups whose `Captures`
    /// collection C# concatenates):
    /// `.+?([-_. ](?<tags>forced|foreign|default|cc|psdh|sdh))*[-_. ](?<iso_code>[a-z]{2,3})([-_. ](?<tags>forced|foreign|default|cc|psdh|sdh))*$`
    ///
    /// Rust port drops the inner `(?<tags>...)` captures (so the regex
    /// is just structural validation) and lets [`super::collect_tags_around`]
    /// recover the tag tokens from positions adjacent to `iso_code`.
    pub static SUBTITLE_LANGUAGE_REGEX: Lazy<Regex> = Lazy::new(|| {
        Regex::new(
            r"(?i)^.+?(?:[-_. ](?:forced|foreign|default|cc|psdh|sdh))*[-_. ](?P<iso_code>[a-z]{2,3})(?:[-_. ](?:forced|foreign|default|cc|psdh|sdh))*$",
        )
        .expect("SUBTITLE_LANGUAGE_REGEX must compile")
    });

    /// `SubtitleLanguageTitleRegex` from `LanguageParser.cs:34`.
    ///
    /// C# pattern (with duplicate `iso_code` groups):
    /// `.+?(\.((?<tags1>forced|...)|(?<iso_code>[a-z]{2,3})))*[-_. ](?<title>[^.]*)(\.((?<tags2>forced|...)|(?<iso_code>[a-z]{2,3})))*$`
    ///
    /// Rust port retains the structure but drops the inner alternation
    /// captures; consumers walk the segments to enumerate iso_codes
    /// (gated to exactly one) and tags. The `title` group is the
    /// only structural capture.
    pub static SUBTITLE_LANGUAGE_TITLE_REGEX: Lazy<Regex> = Lazy::new(|| {
        Regex::new(
            r"(?i)^.+?(?:\.(?:forced|foreign|default|cc|psdh|sdh|[a-z]{2,3}))*[-_. ](?P<title>[^.]*)(?:\.(?:forced|foreign|default|cc|psdh|sdh|[a-z]{2,3}))*$",
        )
        .expect("SUBTITLE_LANGUAGE_TITLE_REGEX must compile")
    });

    /// `SubtitleTitleRegex` from `LanguageParser.cs:36`.
    ///
    /// C# pattern: `^((?<title>.+) - )?(?<copy>(?<!\d+)\d{1,3}(?!\d+))$`.
    /// The lookarounds gate against multi-digit-adjacency (e.g. a
    /// 4-digit year, where `\d{1,3}` would otherwise capture three of
    /// the four digits). Rust regex supports neither lookahead nor
    /// lookbehind; the port relaxes the regex and applies the gate
    /// in [`super::is_valid_copy_match`].
    pub static SUBTITLE_TITLE_REGEX: Lazy<Regex> = Lazy::new(|| {
        Regex::new(r"^(?:(?P<title>.+) - )?(?P<copy>\d{1,3})$")
            .expect("SUBTITLE_TITLE_REGEX must compile")
    });
}

/// Public entry: mirror C# `LanguageParser.ParseSubtitleLanguage`
/// (lines 217-250).
///
/// Algorithm:
/// 1. `simple_filename = strip_last_extension(file_name)`.
/// 2. Match `SUBTITLE_LANGUAGE_REGEX` against `simple_filename`.
///    On success, look up `iso_code` via `iso_languages::find`; that
///    maps to `Language::Unknown` if the code is unknown.
/// 3. Otherwise, walk `Language::All` checking
///    `simple_filename.ends_with(language.name())` case-insensitively.
///    First hit wins.
/// 4. Otherwise return `Language::Unknown`.
pub fn parse_subtitle_language(file_name: &str) -> Language {
    let simple = strip_last_extension(file_name);

    if let Some(caps) = regexes::SUBTITLE_LANGUAGE_REGEX.captures(simple)
        && let Some(iso) = caps.name("iso_code")
    {
        // C# `IsoLanguages.Find` is the same lookup wired in T10; a
        // miss yields `Language::Unknown` (matching C#'s
        // `isoLanguage?.Language ?? Language.Unknown`).
        return iso_languages::find(&iso.as_str().to_ascii_lowercase())
            .unwrap_or(Language::Unknown);
    }

    // Fallback: ends_with(language.name()) for every Language variant,
    // case-insensitive. C# loops over `Language.All`; we mirror that
    // ordering (C# sets `All` to the contiguous list plus `Original`).
    for language in ALL_LANGUAGES_FOR_ENDS_WITH {
        let name = language.name();
        if ends_with_ignore_ascii_case(simple, name) {
            return *language;
        }
    }

    Language::Unknown
}

/// Public entry: mirror C# `LanguageParser.ParseBasicSubtitle`
/// (lines 252-260).
///
/// Composes `parse_language_tags` and `parse_subtitle_language` into a
/// minimal `SubtitleTitleInfo` (only `language_tags`, `language`, and
/// `title_first = false` are populated; `raw_title`/`title`/`copy`
/// stay at defaults).
pub fn parse_basic_subtitle(file_name: &str) -> SubtitleTitleInfo {
    SubtitleTitleInfo {
        title_first: false,
        language_tags: parse_language_tags(file_name),
        language: parse_subtitle_language(file_name),
        ..Default::default()
    }
}

/// Public entry: mirror C# `LanguageParser.ParseSubtitleLanguageInformation`
/// (lines 262-297).
///
/// Tries the richer `SUBTITLE_LANGUAGE_TITLE_REGEX` first. On match the
/// `iso_code` count must be exactly one (C# line 267 gate). On failure
/// or gate-violation, falls back to `parse_basic_subtitle`.
pub fn parse_subtitle_language_information(file_name: &str) -> SubtitleTitleInfo {
    let simple = strip_last_extension(file_name);

    let Some(caps) = regexes::SUBTITLE_LANGUAGE_TITLE_REGEX.captures(simple) else {
        return parse_basic_subtitle(file_name);
    };

    let title_match = match caps.name("title") {
        Some(m) => m,
        None => return parse_basic_subtitle(file_name),
    };

    // Walk the dot-prefixed segments BEFORE the title separator and
    // AFTER the title to recover iso_codes and tags. The title's own
    // separator is `[-_. ]` (any of dash/underscore/dot/space) but the
    // surrounding segments are `\.<token>` (literal dot prefix).
    let title_start = title_match.start();
    let title_end = title_match.end();

    // C# `[-_. ](?<title>...)` consumes one separator immediately
    // before the title's first byte. Find its byte index so the
    // pre-segments scan stops at the right place.
    let title_sep_idx = title_start.saturating_sub(1);
    let pre_section = &simple[..title_sep_idx];
    let post_section = &simple[title_end..];

    let (pre_iso_codes, pre_tags) = scan_dot_segments_from_right(pre_section);
    let (post_iso_codes, post_tags) = scan_dot_segments_from_left(post_section);

    let iso_count = pre_iso_codes.len() + post_iso_codes.len();
    if iso_count != 1 {
        return parse_basic_subtitle(file_name);
    }

    let iso_code = pre_iso_codes
        .first()
        .or_else(|| post_iso_codes.first())
        .expect("iso_count==1 guarantees at least one iso_code");

    let language = iso_languages::find(&iso_code.to_ascii_lowercase()).unwrap_or(Language::Unknown);

    // C# `tags1.Captures.Union(tags2.Captures).Cast<Capture>()` uses
    // `Capture`'s default reference-equality comparer (Capture has no
    // `Equals`/`GetHashCode` override), so distinct Capture objects
    // are NEVER deduped even when their `Value` matches. The Rust
    // port mirrors that with a plain concatenate-and-lowercase. (No
    // current test fixture exercises the dup case, but
    // standing rule #1 says C# fidelity wins.)
    let language_tags: Vec<String> = pre_tags
        .iter()
        .chain(post_tags.iter())
        .filter(|s| !s.is_empty())
        .map(|s| s.to_ascii_lowercase())
        .collect();

    // C# `RawTitle = matchTitle.Groups["title"].Value`. Always
    // `Some(...)` (possibly empty string) when the outer regex
    // matched; `[^.]*` always succeeds (zero or more), so
    // `Group.Success` is true and `.Value` is at minimum `""`.
    let raw_title = caps.name("title").map(|m| m.as_str().to_string());
    // C# `TitleFirst = matchTitle.Groups["tags1"].Captures.Empty()` -- it
    // checks tags1 only, NOT iso_code. A pre-iso segment (e.g.
    // `.eng.testtitle.forced.ass`) leaves tags1 empty and so still has
    // `TitleFirst = true` in C#.
    let title_first = pre_tags.is_empty();

    let mut info = SubtitleTitleInfo {
        title_first,
        language_tags,
        raw_title,
        language,
        ..Default::default()
    };
    update_title_and_copy_from_title(&mut info);
    info
}

/// Public entry: mirror C# `LanguageParser.ParseLanguageTags`
/// (lines 318-335).
///
/// Returns the lowercased tag tokens captured around `iso_code` by
/// `SUBTITLE_LANGUAGE_REGEX`. Returns an empty list when the regex
/// does not match (matching C#'s empty `Captures` enumeration).
pub fn parse_language_tags(file_name: &str) -> Vec<String> {
    let simple = strip_last_extension(file_name);

    let Some(caps) = regexes::SUBTITLE_LANGUAGE_REGEX.captures(simple) else {
        return Vec::new();
    };
    let Some(iso) = caps.name("iso_code") else {
        return Vec::new();
    };

    // The iso_code group span doesn't include its leading separator;
    // `iso_start` points at the iso_code's first byte. Walk from
    // `iso_start - 1` leftward (the separator chars are 1 byte each)
    // collecting tag tokens, then from `iso.end()` rightward.
    collect_tags_around(simple, iso.start(), iso.end())
}

/// Mirror C# `UpdateTitleAndCopyFromTitle` (lines 299-316). Operates
/// in-place on the partially-filled `SubtitleTitleInfo`.
///
/// Matches `SUBTITLE_TITLE_REGEX` against `info.raw_title`. On match
/// the optional `(?<title>.+)` capture (may be `None` when only a
/// copy-number is present) becomes `info.title`, and `(?<copy>\d{1,3})`
/// becomes `info.copy`. On no-match, `info.title = info.raw_title`
/// and `info.copy = 0`.
fn update_title_and_copy_from_title(info: &mut SubtitleTitleInfo) {
    let Some(raw) = info.raw_title.as_deref() else {
        info.title = None;
        info.copy = 0;
        return;
    };

    if let Some(caps) = regexes::SUBTITLE_TITLE_REGEX.captures(raw)
        && let Some(copy) = caps.name("copy")
    {
        // Reapply the C# lookarounds: copy must NOT be adjacent
        // to other digits inside `raw`. With the regex anchored
        // start-to-end, "adjacency" can only mean: the digit run
        // is longer than 3 (the regex would still match since
        // `\d{1,3}` is greedy at end with $). Concretely, if
        // RawTitle contains a 4+ digit run, the regex either
        // fails (because part of the run lands inside `title`
        // and breaks the ` - ` separator) or backtracks to a
        // valid 1-3 digit copy with title preceding.
        //
        // For robustness, validate the C# lookaround logic
        // explicitly: characters immediately before `copy.start()`
        // and after `copy.end()` (within `raw`) must not be ASCII
        // digits.
        if !is_valid_copy_match(raw, copy.start(), copy.end()) {
            info.title = Some(raw.to_string());
            info.copy = 0;
            return;
        }
        info.title = caps.name("title").map(|m| m.as_str().to_string());
        info.copy = copy.as_str().parse().unwrap_or(0);
        return;
    }

    info.title = Some(raw.to_string());
    info.copy = 0;
}

/// Reapplies the C# `(?<!\d+)\d{1,3}(?!\d+)` lookarounds. Returns
/// `true` if `raw[copy_start..copy_end]` is a valid copy-number
/// span (no immediately adjacent digit on either side).
fn is_valid_copy_match(raw: &str, copy_start: usize, copy_end: usize) -> bool {
    let bytes = raw.as_bytes();
    if copy_start > 0 && bytes[copy_start - 1].is_ascii_digit() {
        return false;
    }
    if copy_end < bytes.len() && bytes[copy_end].is_ascii_digit() {
        return false;
    }
    true
}

/// Strip the last `.<ext>` from a filename, mirroring
/// `Path.GetFileNameWithoutExtension`. First strips any leading
/// directory components (everything up to and including the final
/// `/` or `\`), then strips the last `.<ext>`. Returns the input
/// unchanged when no `.` is present; returns `""` for a leading-dot
/// name like `.bashrc` (C# treats those as having an empty stem).
fn strip_last_extension(file_name: &str) -> &str {
    // Step 1: drop any leading directory portion. C# `Path.GetFileName`
    // splits on either `/` or `\` (the latter for legacy Windows paths).
    let bare = match file_name.rfind(['/', '\\']) {
        Some(idx) => &file_name[idx + 1..],
        None => file_name,
    };
    // Step 2: drop the last `.<ext>` if any.
    match bare.rfind('.') {
        None => bare,
        Some(idx) => &bare[..idx],
    }
}

/// Case-insensitive `ends_with` over ASCII bytes.
///
/// C# uses `StringComparison.OrdinalIgnoreCase` which is byte-equal
/// modulo ASCII case. Language names are pure ASCII (e.g. "Portuguese
/// (Brazil)") so a byte-level compare suffices.
fn ends_with_ignore_ascii_case(haystack: &str, needle: &str) -> bool {
    if needle.len() > haystack.len() {
        return false;
    }
    let tail = &haystack.as_bytes()[haystack.len() - needle.len()..];
    tail.eq_ignore_ascii_case(needle.as_bytes())
}

/// Walk leftward from `iso_start - 1` (i.e. the iso_code's leading
/// separator byte) collecting `[-_. ]<tag>` segments while each
/// segment's token is in the tag vocabulary. Then walk rightward
/// from `iso_end` collecting the same. Returns concatenated tag
/// tokens lowercased, in left-to-right source order.
///
/// This recovers the `(?<tags>...)*` captures C# would have collected
/// before and after iso_code, despite Rust regex's no-duplicate-name
/// constraint. Faithfulness check: stops at the first non-tag-vocab
/// segment because C#'s `(?:[-_. ]<tag>)*` would also stop the loop
/// at that point (the tag alternation would fail, ending the `*`
/// repetition).
fn collect_tags_around(simple: &str, iso_start: usize, iso_end: usize) -> Vec<String> {
    let mut left_tags = Vec::new();
    let mut cursor = iso_start;
    while cursor > 0 {
        let Some((tag, new_cursor)) = pop_left_tag_segment(simple, cursor) else {
            break;
        };
        left_tags.push(tag);
        cursor = new_cursor;
    }
    left_tags.reverse();

    let mut right_tags = Vec::new();
    let mut cursor = iso_end;
    while cursor < simple.len() {
        let Some((tag, new_cursor)) = pop_right_tag_segment(simple, cursor) else {
            break;
        };
        right_tags.push(tag);
        cursor = new_cursor;
    }

    left_tags.append(&mut right_tags);
    left_tags
}

/// If the bytes ending at `cursor` form `[-_. ]<tag>`, return
/// `(tag.lowercased(), new_cursor)` where `new_cursor` points one
/// byte past the separator (i.e. the position the next leftward step
/// should start from). Returns `None` if the segment isn't a tag.
fn pop_left_tag_segment(simple: &str, cursor: usize) -> Option<(String, usize)> {
    // Look back for a separator. The C# regex uses `[-_. ]` as
    // separator both before and after iso_code in
    // `SubtitleLanguageRegex`.
    let bytes = simple.as_bytes();
    if cursor == 0 {
        return None;
    }
    // `cursor` is currently the byte index of the iso_code's first
    // byte (or the previously-popped tag's separator byte). The
    // separator immediately before `cursor` is at `cursor - 1`.
    let sep_idx = cursor.checked_sub(1)?;
    let sep = bytes[sep_idx];
    if !is_subtitle_separator(sep) {
        return None;
    }
    // The token starts at the byte AFTER the previous separator (or 0)
    // and runs up to `sep_idx`. Walk leftward from `sep_idx - 1` until
    // we find another separator or reach the start.
    let mut tok_start = sep_idx;
    while tok_start > 0 && !is_subtitle_separator(bytes[tok_start - 1]) {
        tok_start -= 1;
    }
    let token = &simple[tok_start..sep_idx];
    if !is_tag_token(token) {
        return None;
    }
    Some((token.to_ascii_lowercase(), tok_start))
}

/// Mirror of [`pop_left_tag_segment`] for rightward walks. Returns
/// `(tag, new_cursor)` where `new_cursor` is the byte index one past
/// the consumed token.
fn pop_right_tag_segment(simple: &str, cursor: usize) -> Option<(String, usize)> {
    let bytes = simple.as_bytes();
    if cursor >= bytes.len() {
        return None;
    }
    let sep = bytes[cursor];
    if !is_subtitle_separator(sep) {
        return None;
    }
    // Token starts after the separator, runs until next separator or
    // end of string.
    let tok_start = cursor + 1;
    let mut tok_end = tok_start;
    while tok_end < bytes.len() && !is_subtitle_separator(bytes[tok_end]) {
        tok_end += 1;
    }
    let token = &simple[tok_start..tok_end];
    if !is_tag_token(token) {
        return None;
    }
    Some((token.to_ascii_lowercase(), tok_end))
}

/// Walk a pre-title section `\.<token>` rightward. Returns
/// `(iso_codes, tags)` accumulated. The last segment ends at the
/// section's end (the title separator is consumed by the caller).
///
/// Ordering: tokens appear left-to-right (which is also the order
/// C# would enumerate `Captures`).
fn scan_dot_segments_from_right(section: &str) -> (Vec<String>, Vec<String>) {
    // Walk the section breaking on every `.` boundary.
    let mut iso_codes = Vec::new();
    let mut tags = Vec::new();
    if section.is_empty() {
        return (iso_codes, tags);
    }
    let bytes = section.as_bytes();

    // Walk RIGHT-TO-LEFT collecting trailing `\.<token>` segments
    // until a segment fails to match either tag vocab or 2-3 letter
    // ISO. This mirrors C#'s greedy `(\.(...))*` cluster anchored
    // against the title separator: the cluster stops at the first
    // segment that doesn't fit either alternation.
    let mut consumed: Vec<(String, bool)> = Vec::new();
    let mut cursor = bytes.len();
    while cursor > 0 {
        // Find the rightmost `.` strictly inside [..cursor].
        let Some(dot) = section[..cursor].rfind('.') else {
            break;
        };
        let token = &section[dot + 1..cursor];
        if token.is_empty() {
            break;
        }
        let is_iso = is_iso_code_token(token);
        let is_tag = is_tag_token(token);
        if !is_iso && !is_tag {
            break;
        }
        // Tags take precedence over iso_code in C# (alternation order).
        if is_tag {
            consumed.push((token.to_ascii_lowercase(), false));
        } else {
            consumed.push((token.to_string(), true));
        }
        cursor = dot;
    }
    // We collected right-to-left; reverse for source order.
    consumed.reverse();
    for (tok, is_iso_flag) in consumed {
        if is_iso_flag {
            iso_codes.push(tok);
        } else {
            tags.push(tok);
        }
    }
    (iso_codes, tags)
}

/// Walk a post-title section `\.<token>` left-to-right. Returns
/// `(iso_codes, tags)` accumulated.
fn scan_dot_segments_from_left(section: &str) -> (Vec<String>, Vec<String>) {
    let mut iso_codes = Vec::new();
    let mut tags = Vec::new();
    if section.is_empty() {
        return (iso_codes, tags);
    }
    if !section.starts_with('.') {
        // C# regex requires `\.<token>`; without a leading dot the
        // cluster matches zero times.
        return (iso_codes, tags);
    }
    // Split on '.' and walk left-to-right. The leading empty part
    // before the first `.` is dropped.
    for token in section.split('.').skip(1) {
        if token.is_empty() {
            break;
        }
        let is_iso = is_iso_code_token(token);
        let is_tag = is_tag_token(token);
        if !is_iso && !is_tag {
            break;
        }
        if is_tag {
            tags.push(token.to_ascii_lowercase());
        } else {
            iso_codes.push(token.to_string());
        }
    }
    (iso_codes, tags)
}

/// `[a-z]{2,3}` case-insensitive ASCII test mirroring the iso_code
/// branch of the alternation. Pure-ASCII because the C# pattern is
/// `[a-z]` not `[\p{L}]`.
fn is_iso_code_token(token: &str) -> bool {
    let n = token.len();
    if n != 2 && n != 3 {
        return false;
    }
    token.bytes().all(|b| b.is_ascii_alphabetic())
}

/// Tag vocabulary membership test (case-insensitive ASCII).
fn is_tag_token(token: &str) -> bool {
    TAG_VOCAB
        .iter()
        .any(|t| t.len() == token.len() && t.eq_ignore_ascii_case(token))
}

/// `[-_. ]` separator class matching the C# alternation. ASCII bytes
/// only (matches the C# class verbatim).
fn is_subtitle_separator(b: u8) -> bool {
    matches!(b, b'-' | b'_' | b'.' | b' ')
}

/// `Language.All` ordered list for the `EndsWith` fallback in
/// `parse_subtitle_language`. Mirrors C# `Language.All` ordering
/// (the contiguous-ID list plus `Original` last). `Unknown` is
/// excluded because the fallback should produce `Unknown` only via
/// the final `else`.
const ALL_LANGUAGES_FOR_ENDS_WITH: &[Language] = &[
    Language::English,
    Language::French,
    Language::Spanish,
    Language::German,
    Language::Italian,
    Language::Danish,
    Language::Dutch,
    Language::Japanese,
    Language::Icelandic,
    Language::Chinese,
    Language::Russian,
    Language::Polish,
    Language::Vietnamese,
    Language::Swedish,
    Language::Norwegian,
    Language::Finnish,
    Language::Turkish,
    Language::Portuguese,
    Language::Flemish,
    Language::Greek,
    Language::Korean,
    Language::Hungarian,
    Language::Hebrew,
    Language::Lithuanian,
    Language::Czech,
    Language::Arabic,
    Language::Hindi,
    Language::Bulgarian,
    Language::Malayalam,
    Language::Ukrainian,
    Language::Slovak,
    Language::Thai,
    Language::PortugueseBrazil,
    Language::SpanishLatino,
    Language::Romanian,
    Language::Latvian,
    Language::Persian,
    Language::Catalan,
    Language::Croatian,
    Language::Serbian,
    Language::Bosnian,
    Language::Estonian,
    Language::Tamil,
    Language::Indonesian,
    Language::Macedonian,
    Language::Slovenian,
    Language::Original,
];

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

    // ----- Plan-required tests -------------------------------------------

    #[test]
    fn parse_subtitle_extracts_eng_iso_code() {
        let lang = parse_subtitle_language("Show.S01E01.eng.srt");
        assert_eq!(lang, Language::English);
    }

    #[test]
    fn parse_subtitle_extracts_fre_iso_code() {
        // "fre" -> "fra" via the B-to-T map wired in T10.
        assert_eq!(
            parse_subtitle_language("Show.S01E01.fre.srt"),
            Language::French
        );
    }

    #[test]
    fn parse_basic_subtitle_returns_lang_and_tags() {
        let r = parse_basic_subtitle("Show.S01E01.eng.forced.srt");
        assert_eq!(r.language, Language::English);
        assert!(r.language_tags.contains(&"forced".to_string()));
    }

    #[test]
    fn parse_language_tags_extracts_sdh() {
        let tags = parse_language_tags("Show.S01E01.eng.sdh.srt");
        assert!(tags.contains(&"sdh".to_string()));
    }

    // ----- C# LanguageParserFixture parity -------------------------------

    #[test]
    fn fixture_subtitle_unknown_when_no_iso_code() {
        // C# fixture line 12: `"Series Title - S01E01 - Pilot.sub"` -> Unknown.
        assert_eq!(
            parse_subtitle_language("Series Title - S01E01 - Pilot.sub"),
            Language::Unknown
        );
    }

    #[test]
    fn fixture_subtitle_english_via_iso_codes_and_endswith() {
        // C# fixture lines 19-29: every variant resolves to English.
        let cases = [
            "Series Title - S01E01 - Pilot.en.sub",
            "Series Title - S01E01 - Pilot.EN.sub",
            "Series Title - S01E01 - Pilot.eng.sub",
            "Series Title - S01E01 - Pilot.ENG.sub",
            "Series Title - S01E01 - Pilot.English.sub",
            "Series Title - S01E01 - Pilot.english.sub",
            "Series Title - S01E01 - Pilot.en.cc.sub",
            "Series Title - S01E01 - Pilot.en.sdh.sub",
            "Series Title - S01E01 - Pilot.en.forced.sub",
            "Series Title - S01E01 - Pilot.en.sdh.forced.sub",
        ];
        for case in cases {
            assert_eq!(
                parse_subtitle_language(case),
                Language::English,
                "case = {case}"
            );
        }
    }

    #[test]
    fn fixture_parse_subtitle_language_information_eng_with_default_forced() {
        // C# fixture line 465.
        let info = parse_subtitle_language_information(
            "Name (2020) - S01E20 - [AAC 2.0].testtitle.default.eng.forced.ass",
        );
        assert_eq!(info.language, Language::English);
        assert_eq!(
            info.language_tags,
            vec!["default".to_string(), "forced".to_string()]
        );
        assert_eq!(info.title.as_deref(), Some("testtitle"));
    }

    #[test]
    fn fixture_parse_subtitle_language_information_fra_with_default_forced() {
        // C# fixture line 471.
        let info = parse_subtitle_language_information(
            "Name (2020) - S01E20 - [AAC 2.0].testtitle.default.fra.forced.ass",
        );
        assert_eq!(info.language, Language::French);
        assert_eq!(
            info.language_tags,
            vec!["default".to_string(), "forced".to_string()]
        );
        assert_eq!(info.title.as_deref(), Some("testtitle"));
    }

    #[test]
    fn fixture_parse_subtitle_language_information_ru_dashed_title() {
        // C# fixture line 477: `ru-something-else` -> language Russian,
        // title `something-else`, tags empty.
        let info = parse_subtitle_language_information(
            "Name (2020) - S01E20 - [AAC 2.0].ru-something-else.srt",
        );
        assert_eq!(info.language, Language::Russian);
        assert!(
            info.language_tags.is_empty(),
            "got {:?}",
            info.language_tags
        );
        assert_eq!(info.title.as_deref(), Some("something-else"));
    }

    #[test]
    fn fixture_parse_subtitle_language_information_full_subtitles_title() {
        // C# fixture line 478.
        let info = parse_subtitle_language_information(
            "Name (2020) - S01E20 - [AAC 2.0].Full Subtitles.eng.ass",
        );
        assert_eq!(info.language, Language::English);
        assert!(
            info.language_tags.is_empty(),
            "got {:?}",
            info.language_tags
        );
        assert_eq!(info.title.as_deref(), Some("Full Subtitles"));
    }

    #[test]
    fn fixture_parse_subtitle_language_information_dash_one_copy_strip() {
        // C# fixture line 479: title "mytitle - 1" => after
        // SubtitleTitleRegex => Title="mytitle", Copy=1.
        let info = parse_subtitle_language_information(
            "Name (2020) - S01E20 - [AAC 2.0].mytitle - 1.en.ass",
        );
        assert_eq!(info.language, Language::English);
        assert_eq!(info.title.as_deref(), Some("mytitle"));
        assert_eq!(info.copy, 1);
    }

    #[test]
    fn fixture_parse_subtitle_language_information_space_one_no_strip() {
        // C# fixture line 480: title "mytitle 1" => Title="mytitle 1",
        // Copy=0 (no ` - ` separator before the digit).
        let info = parse_subtitle_language_information(
            "Name (2020) - S01E20 - [AAC 2.0].mytitle 1.en.ass",
        );
        assert_eq!(info.language, Language::English);
        assert_eq!(info.title.as_deref(), Some("mytitle 1"));
        assert_eq!(info.copy, 0);
    }

    #[test]
    fn fixture_parse_subtitle_language_information_no_copy() {
        // C# fixture line 481: title "mytitle" -> Title="mytitle", Copy=0.
        let info =
            parse_subtitle_language_information("Name (2020) - S01E20 - [AAC 2.0].mytitle.en.ass");
        assert_eq!(info.language, Language::English);
        assert_eq!(info.title.as_deref(), Some("mytitle"));
        assert_eq!(info.copy, 0);
    }

    #[test]
    fn fixture_parse_subtitle_language_information_no_iso_code_falls_back_to_basic() {
        // C# fixture line 491-494: zero or two iso_codes => fallback
        // to ParseBasicSubtitle => Language=Unknown, LanguageTags=empty,
        // RawTitle=null.
        let cases = [
            "Name (2020) - S01E20 - [AAC 2.0].default.forced.ass",
            "Name (2020) - S01E20 - [AAC 2.0].default.ass",
            "Name (2020) - S01E20 - [AAC 2.0].ass",
            "Name (2020) - S01E20 - [AAC 2.0].testtitle.ass",
        ];
        for case in cases {
            let info = parse_subtitle_language_information(case);
            assert_eq!(info.language, Language::Unknown, "case = {case}");
            assert!(info.language_tags.is_empty(), "case = {case}");
            assert!(info.raw_title.is_none(), "case = {case}");
        }
    }

    // ----- Edge cases that pin our helper-function logic -------------------

    #[test]
    fn parse_subtitle_endswith_after_extension_strip() {
        // C# `Path.GetFileNameWithoutExtension` strips one extension.
        // `Show - Pilot.English.sub` -> simple `Show - Pilot.English`
        // which neither matches `SUBTITLE_LANGUAGE_REGEX` (no `[a-z]{2,3}`
        // tail) nor ends with a Language name except via a trailing
        // language-name component. Fixture line 23 already covers the
        // `.English.sub` -> English case via the EndsWith fallback.
        // Pin the same path via `Pilot.German.sub` as a positive test
        // and `Pilot.NotALanguage.sub` as a negative.
        assert_eq!(
            parse_subtitle_language("Show - Pilot.German.sub"),
            Language::German
        );
        assert_eq!(
            parse_subtitle_language("Show - Pilot.NotALanguage.sub"),
            Language::Unknown
        );
    }

    #[test]
    fn parse_subtitle_endswith_portuguese_brazil() {
        // `Language::PortugueseBrazil.name() == "Portuguese (Brazil)"`.
        // The EndsWith fallback uses the literal Display name. Strip
        // one extension first, so the test's simple filename ends in
        // `Portuguese (Brazil)`.
        assert_eq!(
            parse_subtitle_language("Show - Pilot.Portuguese (Brazil).sub"),
            Language::PortugueseBrazil
        );
    }

    #[test]
    fn parse_language_tags_empty_when_no_match() {
        assert!(parse_language_tags("Pilot.sub").is_empty());
        assert!(parse_language_tags("file_with_no_extension").is_empty());
    }

    #[test]
    fn parse_language_tags_handles_pre_iso_tags() {
        // `Pilot.forced.eng.srt` -> simple `Pilot.forced.eng`.
        // Regex matches: pre-tag `forced`, iso `eng`, post-tags none.
        let tags = parse_language_tags("Pilot.forced.eng.srt");
        assert_eq!(tags, vec!["forced".to_string()]);
    }

    #[test]
    fn parse_language_tags_handles_post_iso_tags() {
        let tags = parse_language_tags("Pilot.eng.sdh.forced.srt");
        assert_eq!(tags, vec!["sdh".to_string(), "forced".to_string()]);
    }

    #[test]
    fn parse_language_tags_drops_non_vocab_words() {
        // `Pilot.junk.eng.srt` -> simple `Pilot.junk.eng`. The C#
        // regex would match `iso=eng` with `.+? = Pilot.junk`. Tags
        // would be empty (junk is not a tag and gets absorbed by .+?).
        let tags = parse_language_tags("Pilot.junk.eng.srt");
        assert!(tags.is_empty(), "got {tags:?}");
    }

    #[test]
    fn parse_basic_subtitle_title_first_is_false() {
        // C# `ParseBasicSubtitle` always sets TitleFirst = false.
        let r = parse_basic_subtitle("Show.S01E01.eng.srt");
        assert!(!r.title_first);
    }

    #[test]
    fn parse_subtitle_language_information_title_first_true_when_no_pre_segments() {
        // `testtitle.eng.forced` -> title is the first segment after
        // the title separator, no pre-iso/pre-tag => TitleFirst=true.
        let info = parse_subtitle_language_information(
            "Name (2020) - S01E20 - [AAC 2.0].testtitle.eng.forced.ass",
        );
        assert!(info.title_first, "got {info:?}");
    }

    #[test]
    fn parse_subtitle_language_information_title_first_true_with_pre_iso_only() {
        // `eng.testtitle.forced` -> pre-iso fires but tags1 stays
        // empty. C# `TitleFirst = tags1.Captures.Empty()` is true.
        let info = parse_subtitle_language_information(
            "Name (2020) - S01E20 - [AAC 2.0].eng.testtitle.forced.ass",
        );
        assert!(info.title_first, "got {info:?}");
    }

    #[test]
    fn parse_subtitle_language_information_title_first_false_with_pre_tag() {
        // `default.eng.testtitle.forced` -> pre-tag fires (tags1 has
        // `default`), TitleFirst=false.
        let info = parse_subtitle_language_information(
            "Name (2020) - S01E20 - [AAC 2.0].default.eng.testtitle.forced.ass",
        );
        assert!(!info.title_first, "got {info:?}");
    }

    #[test]
    fn parse_subtitle_unknown_for_pure_garbage() {
        assert_eq!(
            parse_subtitle_language("garbage_no_match"),
            Language::Unknown
        );
    }

    #[test]
    fn strip_last_extension_handles_empty_and_dotfiles() {
        assert_eq!(strip_last_extension(""), "");
        // C# `Path.GetFileNameWithoutExtension(".bashrc")` returns "":
        // a leading-dot name has an empty stem.
        assert_eq!(strip_last_extension(".bashrc"), "");
        assert_eq!(strip_last_extension("a.b.c"), "a.b");
        assert_eq!(strip_last_extension("foo"), "foo");
        assert_eq!(strip_last_extension("foo.bar"), "foo");
    }

    #[test]
    fn strip_last_extension_strips_directories() {
        // C# `Path.GetFileNameWithoutExtension` first calls `GetFileName`
        // (drops directory portion), then strips the last extension.
        assert_eq!(strip_last_extension("/foo/bar.zip/baz.eng.srt"), "baz.eng");
        assert_eq!(strip_last_extension("/tmp/Pilot.eng.srt"), "Pilot.eng");
        assert_eq!(strip_last_extension(r"C:\tmp\Pilot.eng.srt"), "Pilot.eng");
    }

    #[test]
    fn ends_with_ignore_ascii_case_basic() {
        assert!(ends_with_ignore_ascii_case("FooEnglish", "english"));
        assert!(ends_with_ignore_ascii_case("FooEnglish", "ENGLISH"));
        assert!(!ends_with_ignore_ascii_case("Foo", "FooBar"));
    }
}