vct_core/utils/

directory.rs

use crate::constants::FastHashSet;
use crate::models::TimeRange;
use anyhow::Result;
use std::ffi::OsStr;
use std::fs;
use std::path::{Path, PathBuf};
use walkdir::WalkDir;

/// A file matched during directory traversal, paired with its modification date.
pub struct FileInfo {
    /// Absolute path to the matched file.
    pub path: PathBuf,
    /// Local modification date formatted as `YYYY-MM-DD`, used for grouping.
    pub modified_date: String,
}

/// One directory traversal or metadata error encountered during discovery.
#[derive(Debug)]
pub(crate) struct FileDiscoveryFailure {
    /// Path whose directory entry or metadata could not be read.
    pub path: PathBuf,
    /// Content-safe error summary.
    pub error: String,
}

/// Files discovered successfully plus any partial traversal failures.
pub(crate) struct FileDiscovery {
    /// Matched files that could be inspected successfully.
    pub files: Vec<FileInfo>,
    /// Traversal, metadata, or mtime failures in deterministic path order.
    pub failures: Vec<FileDiscoveryFailure>,
}

/// Recursively collects the files under `dir` that pass `filter_fn`, tagging
/// each with its modification date.
///
/// `filter_fn` decides whether a given path is included; `time_range` drops
/// files whose modification date falls before its cutoff. Equivalent to
/// [`collect_files_with_max_depth`] with no depth cap.
///
/// # Errors
///
/// Returns `Result` for caller ergonomics, but the current implementation
/// never produces an error: a missing directory yields an empty list, and
/// per-entry traversal or metadata failures are silently skipped.
pub fn collect_files_with_dates<P, F>(
    dir: P,
    filter_fn: F,
    time_range: TimeRange,
) -> Result<Vec<FileInfo>>
where
    P: AsRef<Path>,
    F: Fn(&Path) -> bool,
{
    collect_files_with_max_depth(dir, filter_fn, time_range, None)
}

/// Same as [`collect_files_with_dates`] but with an optional traversal depth cap.
///
/// `max_depth` is counted from the root directory: `None` walks the whole
/// tree, `Some(2)` only descends two levels. Callers that know the exact
/// nesting of the file they want (e.g. Copilot's
/// `session-state/<sessionId>/events.jsonl` is always at depth 2) can use
/// this to avoid paying the cost of `WalkDir` visiting large sibling
/// subtrees such as `rewind-snapshots/backups/` that never contain a
/// match but can hold hundreds of files per session.
///
/// # Errors
///
/// Returns `Result` for caller ergonomics, but the current implementation
/// never produces an error: a missing directory yields an empty list, and
/// per-entry traversal or metadata failures are silently skipped.
pub fn collect_files_with_max_depth<P, F>(
    dir: P,
    filter_fn: F,
    time_range: TimeRange,
    max_depth: Option<usize>,
) -> Result<Vec<FileInfo>>
where
    P: AsRef<Path>,
    F: Fn(&Path) -> bool,
{
    Ok(collect_files_with_max_depth_diagnostics(dir, filter_fn, time_range, max_depth).files)
}

/// Discovers matching files while retaining partial traversal failures.
pub(crate) fn collect_files_with_max_depth_diagnostics<P, F>(
    dir: P,
    filter_fn: F,
    time_range: TimeRange,
    max_depth: Option<usize>,
) -> FileDiscovery
where
    P: AsRef<Path>,
    F: Fn(&Path) -> bool,
{
    let dir = dir.as_ref();
    match fs::metadata(dir) {
        Ok(_) => {}
        Err(error) if error.kind() == std::io::ErrorKind::NotFound => {
            return FileDiscovery {
                files: Vec::new(),
                failures: Vec::new(),
            };
        }
        Err(error) => {
            return FileDiscovery {
                files: Vec::new(),
                failures: vec![FileDiscoveryFailure {
                    path: dir.to_path_buf(),
                    error: error.to_string(),
                }],
            };
        }
    }

    let cutoff = time_range
        .cutoff_date()
        .map(|d| d.format("%Y-%m-%d").to_string());

    // Pre-allocate Vec with estimated capacity (typical: 10-50 session files)
    let mut results = Vec::with_capacity(20);
    let mut failures = Vec::new();

    let mut walker = WalkDir::new(dir);
    if let Some(depth) = max_depth {
        walker = walker.max_depth(depth);
    }

    for entry in walker {
        let entry = match entry {
            Ok(entry) => entry,
            Err(error) => {
                failures.push(FileDiscoveryFailure {
                    path: error.path().unwrap_or(dir).to_path_buf(),
                    error: error.to_string(),
                });
                continue;
            }
        };
        if !entry.file_type().is_file() {
            continue;
        }

        let path = entry.path();

        // Apply filter
        if !filter_fn(path) {
            continue;
        }

        // Get file modification time for date grouping
        let metadata = match entry.metadata() {
            Ok(metadata) => metadata,
            Err(error) => {
                failures.push(FileDiscoveryFailure {
                    path: path.to_path_buf(),
                    error: error.to_string(),
                });
                continue;
            }
        };

        let modified = match metadata.modified() {
            Ok(modified) => modified,
            Err(error) => {
                failures.push(FileDiscoveryFailure {
                    path: path.to_path_buf(),
                    error: error.to_string(),
                });
                continue;
            }
        };
        let datetime: chrono::DateTime<chrono::Local> = modified.into();
        let date_key = datetime.format("%Y-%m-%d").to_string();

        // Apply time range filter
        if let Some(ref cutoff_str) = cutoff
            && date_key.as_str() < cutoff_str.as_str()
        {
            continue;
        }

        results.push(FileInfo {
            path: path.to_path_buf(),
            modified_date: date_key,
        });
    }

    failures.sort_by(|left, right| {
        left.path
            .cmp(&right.path)
            .then_with(|| left.error.cmp(&right.error))
    });
    FileDiscovery {
        files: results,
        failures,
    }
}

/// Discovers matching files across every root one provider's sessions live in.
///
/// Roots are walked in order and their results concatenated, so a provider with
/// a single root behaves exactly as [`collect_files_with_max_depth_diagnostics`]
/// does. With more than one root, a file name already found in an **earlier**
/// root is dropped from a later one: Codex archives a session by moving its
/// rollout log from `sessions/` to `archived_sessions/`, and a scan that races
/// that move would otherwise fold the same session twice.
///
/// Declaring more than one root therefore asserts that a repeated file name means
/// a repeated session. Codex satisfies that: its rollout logs are named
/// `rollout-<timestamp>-<uuid>.jsonl` and keep that name across the move. Names
/// are compared only across roots, never within one, because a provider may
/// legitimately give every session the same file name (Copilot writes
/// `events.jsonl` for all of them) — such a provider has to stay single-root.
/// Each drop is logged at debug level so a wrong assertion is diagnosable rather
/// than a silently missing session.
pub(crate) fn collect_provider_files_diagnostics<F>(
    dirs: &[&Path],
    filter_fn: F,
    time_range: TimeRange,
    max_depth: Option<usize>,
) -> FileDiscovery
where
    F: Fn(&Path) -> bool,
{
    let Some((first, rest)) = dirs.split_first() else {
        return FileDiscovery {
            files: Vec::new(),
            failures: Vec::new(),
        };
    };
    let mut discovery =
        collect_files_with_max_depth_diagnostics(first, &filter_fn, time_range, max_depth);
    if rest.is_empty() {
        return discovery;
    }

    for dir in rest {
        let mut found =
            collect_files_with_max_depth_diagnostics(dir, &filter_fn, time_range, max_depth);
        discovery.failures.append(&mut found.failures);
        // The overwhelmingly common shape is an absent or empty later root (no
        // session has been archived yet), so skip building the name index for it.
        if found.files.is_empty() {
            continue;
        }
        // Scoped so the borrow of the already-merged names ends before the append.
        {
            let seen: FastHashSet<&OsStr> = discovery
                .files
                .iter()
                .filter_map(|file| file.path.file_name())
                .collect();
            found.files.retain(|file| {
                let duplicate = file
                    .path
                    .file_name()
                    .is_some_and(|name| seen.contains(name));
                if duplicate {
                    log::debug!(
                        "skipping {}: already discovered under an earlier session root",
                        file.path.display()
                    );
                }
                !duplicate
            });
        }
        discovery.files.append(&mut found.files);
    }

    discovery.failures.sort_by(|left, right| {
        left.path
            .cmp(&right.path)
            .then_with(|| left.error.cmp(&right.error))
    });
    discovery
}

/// Maximum traversal depth for Copilot CLI session scans.
///
/// Copilot writes `~/.copilot/session-state/<sessionId>/events.jsonl`, so
/// the event log is always exactly two directory levels below
/// `session-state/`. Anything deeper belongs to companion subtrees we
/// intentionally ignore (`rewind-snapshots/backups/`, `checkpoints/`,
/// `files/`, `research/`) — some of which can hold dozens of files per
/// session and would otherwise make `WalkDir` visit hundreds of entries
/// per session just to have `is_copilot_session_file` reject them.
pub const COPILOT_SESSION_MAX_DEPTH: usize = 2;

/// Maximum traversal depth for Grok CLI session scans.
///
/// Grok writes `$GROK_HOME/sessions/<workspace>/<session>/signals.json`, so
/// each signals file is exactly three levels below the sessions root.
pub const GROK_SESSION_MAX_DEPTH: usize = 3;

/// Filter for Codex session files under `~/.codex/sessions/YYYY/MM/DD/`.
///
/// Codex writes `rollout-*.jsonl` files directly into the dated sub-folders.
/// We also accept `.json` defensively in case an older dump ever gets dropped
/// into the tree, but reject Claude Code's `*.meta.json` / `*.meta.jsonl`
/// subagent sidecar files (those live under `~/.claude/projects/` in
/// practice, but the filter still guards against the unlikely collision).
///
/// Intentionally separated from the other per-provider filters
/// (`is_claude_session_file`, `is_copilot_session_file`,
/// `is_gemini_session_file`) even though the body is currently trivial:
/// keeping one filter per provider means future format changes on one
/// provider do not require teasing apart a shared implementation.
pub fn is_codex_session_file(path: &Path) -> bool {
    if is_meta_sidecar_file(path) {
        return false;
    }
    if let Some(ext) = path.extension() {
        ext == "jsonl" || ext == "json"
    } else {
        false
    }
}

/// Filter for Claude Code session files (`.jsonl` only).
///
/// Matches both top-level sessions (`~/.claude/projects/<project>/<session>.jsonl`)
/// and subagent sessions (`~/.claude/projects/<project>/<session>/subagents/agent-*.jsonl`).
/// Rejects meta sidecars (`*.meta.json` / `*.meta.jsonl`) and any non-JSONL artifact
/// that ends up under the projects directory (e.g. screenshots pasted into prompts).
pub fn is_claude_session_file(path: &Path) -> bool {
    if is_meta_sidecar_file(path) {
        return false;
    }
    path.extension().is_some_and(|ext| ext == "jsonl")
}

/// Filter for Gemini CLI session files.
///
/// Current Gemini CLI stores each chat as a line-delimited event stream at
/// `~/.gemini/tmp/<project>/chats/session-*.jsonl`. Subagent sessions sit one
/// level deeper at `chats/<parent-session>/<subagent>.jsonl`. Sibling artifacts
/// (`discordbot/logs.json`, the `bin/rg` binary, `.project_root`) are rejected.
///
/// Legacy single-object exports (`chats/<session>.json`) are intentionally
/// not matched: the JSONL format is the only shape the analyzer understands
/// today, and silently scanning `.json` files we can no longer parse just
/// yields `Warning: Failed to analyze ...` noise on every run.
pub fn is_gemini_session_file(path: &Path) -> bool {
    if path.extension() != Some(std::ffi::OsStr::new("jsonl")) {
        false
    } else {
        path.parent().is_some_and(|parent| {
            parent.file_name() == Some(std::ffi::OsStr::new("chats"))
                || parent.parent().is_some_and(|grandparent| {
                    grandparent.file_name() == Some(std::ffi::OsStr::new("chats"))
                })
        })
    }
}

/// Filter for Copilot CLI session files.
///
/// Copilot CLI stores each session as a directory under
/// `~/.copilot/session-state/<sessionId>/` containing the event log
/// (`events.jsonl`) plus sibling subdirectories for file snapshots
/// (`rewind-snapshots/`, `files/`, `research/`, `checkpoints/`) and a
/// per-workspace YAML file. Only `events.jsonl` carries the conversation
/// stream we want to analyze — if we fell back to the generic JSON filter,
/// `rewind-snapshots/index.json` would be mis-picked up as a session log
/// and fail to parse.
///
/// The historical single-file layout
/// (`~/.copilot/history-session-state/<sessionId>.json`) is not matched
/// and no longer supported by the analyzer at all.
pub fn is_copilot_session_file(path: &Path) -> bool {
    // Compare raw `OsStr` rather than going through `to_str()` so paths with
    // non-UTF-8 bytes elsewhere in the tree do not silently reject the file.
    // The chosen constants (`events.jsonl`, `session-state`) are pure ASCII,
    // so the comparison is safe on every platform we care about and keeps
    // the style consistent with `is_gemini_session_file`'s `OsStr::new("chats")`
    // check above.
    if path.file_name() != Some(std::ffi::OsStr::new("events.jsonl")) {
        return false;
    }

    // Must sit one level under `session-state/<sessionId>/events.jsonl` —
    // reject anything that just happens to be called `events.jsonl` in a
    // nested subfolder (e.g. `rewind-snapshots/events.jsonl`).
    path.parent()
        .and_then(|p| p.parent())
        .map(|pp| pp.file_name() == Some(std::ffi::OsStr::new("session-state")))
        .unwrap_or(false)
}

/// Filter for Grok CLI session signal files.
///
/// Only `$GROK_HOME/sessions/<workspace>/<session>/signals.json` is a session
/// entry point. Sibling files such as `summary.json` and `updates.jsonl` are
/// consumed by the Grok parser after the signals file has been selected.
pub fn is_grok_session_file(path: &Path) -> bool {
    if path.file_name() != Some(std::ffi::OsStr::new("signals.json")) {
        return false;
    }

    path.parent()
        .and_then(Path::parent)
        .and_then(Path::parent)
        .is_some_and(|sessions| sessions.file_name() == Some(std::ffi::OsStr::new("sessions")))
}

/// Returns true if the path is a Claude Code meta sidecar file.
///
/// Claude Code writes these next to subagent session logs with metadata like
/// `agentType` / `description`. Today they're `*.meta.json`; we also reject
/// `*.meta.jsonl` pre-emptively so the filter stays correct if the format
/// ever switches to line-delimited JSON.
fn is_meta_sidecar_file(path: &Path) -> bool {
    path.file_name()
        .and_then(|n| n.to_str())
        .is_some_and(|name| name.ends_with(".meta.json") || name.ends_with(".meta.jsonl"))
}

#[cfg(test)]
mod tests {
    use super::*;
    use std::fs::{self, File};
    use std::io::Write;
    use tempfile::tempdir;

    #[test]
    fn test_is_codex_session_file_jsonl() {
        // Test JSONL file extension
        let path = std::path::Path::new("test.jsonl");
        assert!(is_codex_session_file(path));
    }

    #[test]
    fn test_is_codex_session_file_json() {
        // Test JSON file extension
        let path = std::path::Path::new("test.json");
        assert!(is_codex_session_file(path));
    }

    #[test]
    fn test_is_codex_session_file_txt() {
        // Test non-JSON file extension
        let path = std::path::Path::new("test.txt");
        assert!(!is_codex_session_file(path));
    }

    #[test]
    fn test_is_codex_session_file_no_extension() {
        // Test file without extension
        let path = std::path::Path::new("test");
        assert!(!is_codex_session_file(path));
    }

    #[test]
    fn test_is_codex_session_file_uppercase() {
        // Test uppercase extension
        let path = std::path::Path::new("test.JSON");
        assert!(!is_codex_session_file(path)); // Case-sensitive
    }

    #[test]
    fn test_is_gemini_session_file_rejects_legacy_json() {
        // Legacy single-object exports (`chats/<session>.json`) are intentionally
        // no longer matched — the analyzer only handles the JSONL event stream.
        let path = std::path::Path::new("/home/user/.gemini/tmp/hash/chats/chat.json");
        assert!(!is_gemini_session_file(path));
    }

    #[test]
    fn test_is_gemini_session_file_accepts_jsonl() {
        // Current Gemini CLI writes each event as a JSONL line under `chats/`.
        let path = std::path::Path::new(
            "/home/user/.gemini/tmp/proj/chats/session-2026-04-23T12-52.jsonl",
        );
        assert!(is_gemini_session_file(path));
    }

    #[test]
    fn test_is_gemini_session_file_accepts_nested_subagent() {
        let path =
            std::path::Path::new("/home/user/.gemini/tmp/proj/chats/parent-session/subagent.jsonl");
        assert!(is_gemini_session_file(path));
    }

    #[test]
    fn test_is_gemini_session_file_rejects_deeper_jsonl() {
        let path = std::path::Path::new(
            "/home/user/.gemini/tmp/proj/chats/parent-session/artifacts/data.jsonl",
        );
        assert!(!is_gemini_session_file(path));
    }

    #[test]
    fn test_is_gemini_session_file_wrong_parent() {
        // Test file not in chats directory
        let path = std::path::Path::new("/home/user/.gemini/tmp/hash/other/file.json");
        assert!(!is_gemini_session_file(path));
    }

    #[test]
    fn test_is_gemini_session_file_wrong_extension() {
        // Test file in chats directory but wrong extension
        let path = std::path::Path::new("/home/user/.gemini/tmp/hash/chats/file.txt");
        assert!(!is_gemini_session_file(path));
    }

    #[test]
    fn test_is_gemini_session_file_no_parent() {
        // Test file without parent
        let path = std::path::Path::new("file.json");
        assert!(!is_gemini_session_file(path));
    }

    #[test]
    fn test_is_gemini_session_file_excludes_sibling_dirs() {
        // `tmp/discordbot/logs.json` lives next to the `chats/` folder but
        // must not be picked up because its parent is not `chats`.
        let sibling = std::path::Path::new("/home/user/.gemini/tmp/discordbot/logs.json");
        assert!(!is_gemini_session_file(sibling));

        // The `bin/rg` binary that Gemini CLI drops alongside session data has no
        // JSON extension at all.
        let binary = std::path::Path::new("/home/user/.gemini/tmp/bin/rg");
        assert!(!is_gemini_session_file(binary));
    }

    #[test]
    fn test_is_copilot_session_file_accepts_events_jsonl() {
        // Current layout: `session-state/<sessionId>/events.jsonl`
        let path = std::path::Path::new(
            "/home/user/.copilot/session-state/d2e098d0-e0d6-4d6b-914b-c4c5543b17e3/events.jsonl",
        );
        assert!(is_copilot_session_file(path));
    }

    #[test]
    fn test_is_copilot_session_file_rejects_snapshots() {
        // Rewind snapshots also emit JSON files; they must not be picked up as
        // session logs.
        let snapshot_index = std::path::Path::new(
            "/home/user/.copilot/session-state/d2e098d0/rewind-snapshots/index.json",
        );
        assert!(!is_copilot_session_file(snapshot_index));

        let snapshot_backup = std::path::Path::new(
            "/home/user/.copilot/session-state/d2e098d0/rewind-snapshots/backups/2ee575c19132c8bd-1776949007337",
        );
        assert!(!is_copilot_session_file(snapshot_backup));

        let workspace =
            std::path::Path::new("/home/user/.copilot/session-state/d2e098d0/workspace.yaml");
        assert!(!is_copilot_session_file(workspace));
    }

    #[test]
    fn test_is_copilot_session_file_rejects_nested_events_jsonl() {
        // A stray `events.jsonl` inside a rewind snapshot must not count as a
        // session log — the parent of the parent must be `session-state`.
        let nested = std::path::Path::new(
            "/home/user/.copilot/session-state/d2e098d0/rewind-snapshots/events.jsonl",
        );
        assert!(!is_copilot_session_file(nested));
    }

    #[test]
    fn test_is_copilot_session_file_rejects_other_files() {
        // Plain JSON / JSONL files outside the `session-state/<uuid>/` layout
        // should never pass.
        let path1 = std::path::Path::new("/tmp/events.jsonl");
        assert!(!is_copilot_session_file(path1));

        let path2 = std::path::Path::new("/home/user/.copilot/logs.json");
        assert!(!is_copilot_session_file(path2));
    }

    #[test]
    fn test_is_grok_session_file_accepts_exact_layout() {
        let path =
            std::path::Path::new("/home/user/.grok/sessions/workspace/session-id/signals.json");
        assert!(is_grok_session_file(path));
    }

    #[test]
    fn test_is_grok_session_file_rejects_siblings_and_nested_files() {
        let summary =
            std::path::Path::new("/home/user/.grok/sessions/workspace/session-id/summary.json");
        let nested = std::path::Path::new(
            "/home/user/.grok/sessions/workspace/session-id/snapshots/signals.json",
        );
        let shallow = std::path::Path::new("/home/user/.grok/sessions/session-id/signals.json");

        assert!(!is_grok_session_file(summary));
        assert!(!is_grok_session_file(nested));
        assert!(!is_grok_session_file(shallow));
    }

    #[test]
    fn test_grok_max_depth_collects_only_signals_entry_point() {
        let dir = tempdir().unwrap();
        let sessions = dir.path().join("sessions");
        let session = sessions.join("workspace").join("session-id");
        let nested = session.join("snapshots");
        fs::create_dir_all(&nested).unwrap();
        File::create(session.join("signals.json")).unwrap();
        File::create(session.join("summary.json")).unwrap();
        File::create(nested.join("signals.json")).unwrap();

        let files = collect_files_with_max_depth(
            &sessions,
            is_grok_session_file,
            TimeRange::All,
            Some(GROK_SESSION_MAX_DEPTH),
        )
        .unwrap();

        assert_eq!(files.len(), 1);
        assert!(files[0].path.ends_with("workspace/session-id/signals.json"));
    }

    #[test]
    fn test_collect_files_with_max_depth_respects_bound() {
        // Layout mirrors a real Copilot `session-state/` tree: `events.jsonl` sits
        // at depth 2 and must be collected, while snapshot artifacts deeper in
        // the tree (`rewind-snapshots/backups/<hash>`) must be skipped entirely
        // by a max_depth(2) walk so they never even hit `is_copilot_session_file`.
        let dir = tempdir().unwrap();
        let session = dir.path().join("session-state").join("sess-abc");
        let backups = session.join("rewind-snapshots").join("backups");
        fs::create_dir_all(&backups).unwrap();

        File::create(session.join("events.jsonl")).unwrap();
        File::create(backups.join("deadbeef-123")).unwrap();
        File::create(backups.join("events.jsonl")).unwrap(); // still valid JSONL
        File::create(session.join("workspace.yaml")).unwrap();

        // Unbounded walk picks up every file whose filter passes...
        let unbounded = collect_files_with_dates(
            dir.path().join("session-state"),
            is_copilot_session_file,
            TimeRange::All,
        )
        .unwrap();
        let unbounded_names: Vec<&str> = unbounded
            .iter()
            .filter_map(|f| f.path.file_name()?.to_str())
            .collect();
        // `rewind-snapshots/backups/events.jsonl` fails the filter (parent-of-parent
        // is `rewind-snapshots`, not `session-state`), so the filter already blocks
        // it. But we must have picked up the real one at depth 2.
        assert!(unbounded_names.contains(&"events.jsonl"));

        // Bounded walk (depth 2) must not even descend into backups/.
        let bounded = collect_files_with_max_depth(
            dir.path().join("session-state"),
            is_copilot_session_file,
            TimeRange::All,
            Some(2),
        )
        .unwrap();
        assert_eq!(bounded.len(), 1);
        assert!(bounded[0].path.ends_with("sess-abc/events.jsonl"));
    }

    #[test]
    fn test_collect_provider_files_deduplicates_only_across_roots() {
        // Mirrors Codex: a dated active root plus the flat archived root the
        // rollout log is moved into, with one session momentarily in both.
        let dir = tempdir().unwrap();
        let active = dir
            .path()
            .join("sessions")
            .join("2026")
            .join("06")
            .join("06");
        let archived = dir.path().join("archived_sessions");
        fs::create_dir_all(&active).unwrap();
        fs::create_dir_all(&archived).unwrap();
        File::create(active.join("rollout-live.jsonl")).unwrap();
        File::create(active.join("rollout-moving.jsonl")).unwrap();
        File::create(archived.join("rollout-moving.jsonl")).unwrap();
        File::create(archived.join("rollout-old.jsonl")).unwrap();

        let roots = [dir.path().join("sessions"), archived.clone()];
        let roots: Vec<&Path> = roots.iter().map(PathBuf::as_path).collect();
        let files =
            collect_provider_files_diagnostics(&roots, is_codex_session_file, TimeRange::All, None)
                .files;

        let mut found: Vec<PathBuf> = files.into_iter().map(|file| file.path).collect();
        found.sort();
        assert_eq!(found.len(), 3, "the moving session must be counted once");
        // The active copy wins, so the archived duplicate is the one dropped.
        assert!(found.contains(&active.join("rollout-moving.jsonl")));
        assert!(!found.contains(&archived.join("rollout-moving.jsonl")));
        assert!(found.contains(&archived.join("rollout-old.jsonl")));
    }

    #[test]
    fn test_collect_provider_files_keeps_same_named_files_in_one_root() {
        // Every Copilot session log is named `events.jsonl`; a single-root
        // provider must never have those collapsed into one.
        let dir = tempdir().unwrap();
        let sessions = dir.path().join("session-state");
        for session in ["sess-a", "sess-b"] {
            let path = sessions.join(session);
            fs::create_dir_all(&path).unwrap();
            File::create(path.join("events.jsonl")).unwrap();
        }

        let roots = [sessions.as_path()];
        let files = collect_provider_files_diagnostics(
            &roots,
            is_copilot_session_file,
            TimeRange::All,
            Some(COPILOT_SESSION_MAX_DEPTH),
        )
        .files;

        assert_eq!(files.len(), 2);
    }

    #[test]
    fn test_collect_provider_files_skips_missing_roots() {
        let dir = tempdir().unwrap();
        let archived = dir.path().join("archived_sessions");
        fs::create_dir_all(&archived).unwrap();
        File::create(archived.join("rollout-old.jsonl")).unwrap();

        // A Codex home that has only ever archived: the active root is absent.
        let roots = [dir.path().join("sessions"), archived];
        let roots: Vec<&Path> = roots.iter().map(PathBuf::as_path).collect();
        let discovery =
            collect_provider_files_diagnostics(&roots, is_codex_session_file, TimeRange::All, None);

        assert_eq!(discovery.files.len(), 1);
        assert!(discovery.failures.is_empty());
        assert!(
            collect_provider_files_diagnostics(&[], is_codex_session_file, TimeRange::All, None)
                .files
                .is_empty()
        );
    }

    #[test]
    fn test_collect_files_with_dates_empty_dir() {
        // Test collecting files from empty directory
        let dir = tempdir().unwrap();

        let results =
            collect_files_with_dates(dir.path(), is_codex_session_file, TimeRange::All).unwrap();
        assert_eq!(results.len(), 0);
    }

    #[test]
    fn test_collect_files_with_dates_nonexistent_dir() {
        // Test collecting files from non-existent directory
        let results =
            collect_files_with_dates("/nonexistent/path", is_codex_session_file, TimeRange::All)
                .unwrap();
        assert_eq!(results.len(), 0);
    }

    #[test]
    fn test_collect_files_with_dates_with_files() {
        // Test collecting JSON files from directory
        let dir = tempdir().unwrap();

        // Create some JSON files
        File::create(dir.path().join("file1.json")).unwrap();
        File::create(dir.path().join("file2.jsonl")).unwrap();
        File::create(dir.path().join("file3.txt")).unwrap(); // Should be filtered out

        let results =
            collect_files_with_dates(dir.path(), is_codex_session_file, TimeRange::All).unwrap();
        assert_eq!(results.len(), 2);

        // Check that date fields are set
        for file_info in &results {
            assert!(!file_info.modified_date.is_empty());
            assert!(file_info.modified_date.contains('-')); // Should be YYYY-MM-DD format
        }
    }

    #[test]
    fn test_collect_files_with_dates_nested_directories() {
        // Test collecting files from nested directories
        let dir = tempdir().unwrap();

        // Create nested structure
        fs::create_dir_all(dir.path().join("subdir1")).unwrap();
        fs::create_dir_all(dir.path().join("subdir2")).unwrap();

        File::create(dir.path().join("file1.json")).unwrap();
        File::create(dir.path().join("subdir1/file2.json")).unwrap();
        File::create(dir.path().join("subdir2/file3.jsonl")).unwrap();

        let results =
            collect_files_with_dates(dir.path(), is_codex_session_file, TimeRange::All).unwrap();
        assert_eq!(results.len(), 3);
    }

    #[test]
    fn test_collect_files_with_dates_filter_function() {
        // Test that filter function works correctly
        let dir = tempdir().unwrap();

        File::create(dir.path().join("file1.json")).unwrap();
        File::create(dir.path().join("file2.jsonl")).unwrap();
        File::create(dir.path().join("file3.txt")).unwrap();

        // Custom filter: only .txt files
        let results = collect_files_with_dates(
            dir.path(),
            |p| p.extension().is_some_and(|e| e == "txt"),
            TimeRange::All,
        )
        .unwrap();

        assert_eq!(results.len(), 1);
    }

    #[test]
    fn test_collect_files_with_dates_no_matching_files() {
        // Test when no files match filter
        let dir = tempdir().unwrap();

        File::create(dir.path().join("file1.txt")).unwrap();
        File::create(dir.path().join("file2.md")).unwrap();

        let results =
            collect_files_with_dates(dir.path(), is_codex_session_file, TimeRange::All).unwrap();
        assert_eq!(results.len(), 0);
    }

    #[test]
    fn test_file_info_path() {
        // Test that FileInfo contains correct paths
        let dir = tempdir().unwrap();
        let file_path = dir.path().join("test.json");
        File::create(&file_path).unwrap();

        let results =
            collect_files_with_dates(dir.path(), is_codex_session_file, TimeRange::All).unwrap();
        assert_eq!(results.len(), 1);
        assert_eq!(results[0].path, file_path);
    }

    #[test]
    fn test_file_info_date_format() {
        // Test that date format is YYYY-MM-DD
        let dir = tempdir().unwrap();
        File::create(dir.path().join("test.json")).unwrap();

        let results =
            collect_files_with_dates(dir.path(), is_codex_session_file, TimeRange::All).unwrap();
        assert_eq!(results.len(), 1);

        let date = &results[0].modified_date;
        assert_eq!(date.len(), 10); // YYYY-MM-DD is 10 chars
        assert_eq!(date.chars().filter(|&c| c == '-').count(), 2); // Two dashes
    }

    #[test]
    fn test_collect_files_ignores_directories() {
        // Test that directories are not included in results
        let dir = tempdir().unwrap();

        // Create a directory with .json in name
        fs::create_dir(dir.path().join("test.json")).unwrap();

        // Create an actual file
        File::create(dir.path().join("real.json")).unwrap();

        let results =
            collect_files_with_dates(dir.path(), is_codex_session_file, TimeRange::All).unwrap();
        assert_eq!(results.len(), 1); // Only the file, not the directory
    }

    #[test]
    fn test_is_codex_session_file_with_dots_in_name() {
        // Test files with dots in name
        let path = std::path::Path::new("my.test.file.json");
        assert!(is_codex_session_file(path));

        let path2 = std::path::Path::new("my.test.file.jsonl");
        assert!(is_codex_session_file(path2));
    }

    #[test]
    fn test_is_gemini_session_file_multiple_levels() {
        // Depth of the enclosing directory does not matter as long as the
        // *immediate* parent is `chats/` and the extension is `.jsonl`.
        let path = std::path::Path::new("/a/b/c/d/chats/file.jsonl");
        assert!(is_gemini_session_file(path));
    }

    #[test]
    fn test_collect_files_with_content() {
        // Test that files with content are collected
        let dir = tempdir().unwrap();
        let file_path = dir.path().join("test.json");

        let mut file = File::create(&file_path).unwrap();
        writeln!(file, r#"{{"key": "value"}}"#).unwrap();

        let results =
            collect_files_with_dates(dir.path(), is_codex_session_file, TimeRange::All).unwrap();
        assert_eq!(results.len(), 1);
        assert!(results[0].path.exists());
    }

    #[test]
    fn test_is_codex_session_file_excludes_meta_sidecars() {
        // *.meta.json sidecars live next to Claude Code subagent sessions and must not
        // be picked up by the generic json filter (they would otherwise be parsed and
        // mis-detected as Codex logs).
        let path = std::path::Path::new("agent-afda1991051a0eb93.meta.json");
        assert!(!is_codex_session_file(path));

        let nested = std::path::Path::new(
            "/home/user/.claude/projects/proj/sess/subagents/agent-x.meta.json",
        );
        assert!(!is_codex_session_file(nested));

        // Pre-emptive defense: reject `.meta.jsonl` too, in case Claude Code ever
        // switches the sidecar format to line-delimited JSON.
        let meta_jsonl = std::path::Path::new(
            "/home/user/.claude/projects/proj/sess/subagents/agent-x.meta.jsonl",
        );
        assert!(!is_codex_session_file(meta_jsonl));
    }

    #[test]
    fn test_is_claude_session_file_accepts_jsonl() {
        // Top-level Claude session logs
        let top_level = std::path::Path::new("/home/user/.claude/projects/proj/sess.jsonl");
        assert!(is_claude_session_file(top_level));

        // Subagent session logs sit one extra level deeper
        let subagent = std::path::Path::new(
            "/home/user/.claude/projects/proj/sess/subagents/agent-afda1991051a0eb93.jsonl",
        );
        assert!(is_claude_session_file(subagent));
    }

    #[test]
    fn test_is_claude_session_file_rejects_non_jsonl() {
        // Metadata sidecars — current format (`.meta.json`)
        let meta = std::path::Path::new(
            "/home/user/.claude/projects/proj/sess/subagents/agent-afda1991051a0eb93.meta.json",
        );
        assert!(!is_claude_session_file(meta));

        // Metadata sidecars — hypothetical future format (`.meta.jsonl`). Has `.jsonl`
        // extension, so without the sidecar check it would slip through.
        let meta_jsonl = std::path::Path::new(
            "/home/user/.claude/projects/proj/sess/subagents/agent-afda1991051a0eb93.meta.jsonl",
        );
        assert!(!is_claude_session_file(meta_jsonl));

        // Plain .json files (not something Claude Code writes in this tree)
        let plain_json = std::path::Path::new("/home/user/.claude/projects/proj/notes.json");
        assert!(!is_claude_session_file(plain_json));

        // Other pasted artifacts
        let image = std::path::Path::new("/home/user/.claude/projects/proj/sess/paste.png");
        assert!(!is_claude_session_file(image));
    }

    #[test]
    fn test_collect_claude_session_files_includes_subagents() {
        // Simulates the `~/.claude/projects/<project>/<session>/subagents/` layout:
        //   projects/<project>/
        //     sess-a.jsonl                          <- top-level session
        //     sess-a/subagents/agent-one.jsonl      <- subagent session
        //     sess-a/subagents/agent-one.meta.json  <- metadata sidecar (ignored)
        //     sess-a/screenshot.png                 <- pasted artifact (ignored)
        let dir = tempdir().unwrap();
        let project = dir.path().join("-home-user-repo");
        let session_subdir = project.join("sess-a");
        let subagents = session_subdir.join("subagents");
        fs::create_dir_all(&subagents).unwrap();

        File::create(project.join("sess-a.jsonl")).unwrap();
        File::create(subagents.join("agent-one.jsonl")).unwrap();
        File::create(subagents.join("agent-one.meta.json")).unwrap();
        File::create(subagents.join("agent-two.meta.jsonl")).unwrap();
        File::create(session_subdir.join("screenshot.png")).unwrap();

        let results =
            collect_files_with_dates(dir.path(), is_claude_session_file, TimeRange::All).unwrap();
        let names: Vec<String> = results
            .iter()
            .filter_map(|f| f.path.file_name()?.to_str().map(String::from))
            .collect();

        assert_eq!(results.len(), 2, "collected: {:?}", names);
        assert!(names.contains(&"sess-a.jsonl".to_string()));
        assert!(names.contains(&"agent-one.jsonl".to_string()));
        assert!(!names.iter().any(|n| n.contains(".meta.")));
        assert!(!names.iter().any(|n| n.ends_with(".png")));
    }
}