ClaudeDo/src/ClaudeDo.Worker/Runner/WorktreeManager.cs

using ClaudeDo.Data;
using ClaudeDo.Data.Git;
using ClaudeDo.Data.Models;
using ClaudeDo.Data.Repositories;
using ClaudeDo.Worker.AzureDevOps;
using ClaudeDo.Worker.Config;
using ClaudeDo.Worker.Git;
using ClaudeDo.Worker.Hub;
using ClaudeDo.Worker.Lifecycle;
using ClaudeDo.Worker.Tickets;
using Microsoft.EntityFrameworkCore;

namespace ClaudeDo.Worker.Runner;

public sealed record WorktreeContext(string WorktreePath, string BranchName, string BaseCommit);

public sealed class WorktreeManager
{
    // Phase token for the OperationProgress broadcast while `git worktree add` (+ the
    // self-heal retry) is in flight -- stable identifier, localized by the UI via the
    // pre-existing "ops.worker.creatingWorktree" key. opKey is the task id.
    public const string PhaseCreatingWorktree = "creating_worktree";

    private readonly GitService _git;
    private readonly IDbContextFactory<ClaudeDoDbContext> _dbFactory;
    private readonly WorkerConfig _cfg;
    private readonly ILogger<WorktreeManager> _logger;
    private readonly HubBroadcaster? _broadcaster;
    private readonly RunningOperationRegistry _operations;
    private readonly AzureDevOpsClient? _azdo;

    public WorktreeManager(
        GitService git, IDbContextFactory<ClaudeDoDbContext> dbFactory, WorkerConfig cfg,
        ILogger<WorktreeManager> logger, HubBroadcaster? broadcaster = null,
        RunningOperationRegistry? operations = null, AzureDevOpsClient? azdo = null)
    {
        _git = git;
        _dbFactory = dbFactory;
        _cfg = cfg;
        _logger = logger;
        _broadcaster = broadcaster;
        _operations = operations ?? new RunningOperationRegistry();
        _azdo = azdo;
    }

    private Task BroadcastCreating(string taskId) =>
        _broadcaster?.OperationProgress(taskId, PhaseCreatingWorktree, 0, 0) ?? Task.CompletedTask;

    /// <summary>
    /// Branch leaf and folder name: <c>{nummer}-{slug}</c>. Nummer = Ticket-Id, sonst die Task-Nummer —
    /// außer im Firmen-Repo ohne Ticket, dort sähe sie wie eine Ticket-Id aus und entfällt.
    /// </summary>
    public static string NameFor(TaskEntity task, int? ticketId, bool isCompanyRepo)
    {
        var titleSlug = CommitMessageBuilder.ToSlug(task.Title);
        if (titleSlug.Length > 40) titleSlug = titleSlug[..40].TrimEnd('-');
        var number = ticketId?.ToString() ?? (isCompanyRepo ? null : task.Number.ToString());
        if (titleSlug.Length == 0) return number ?? task.Number.ToString();
        return number is null ? titleSlug : $"{number}-{titleSlug}";
    }

    /// <summary>Branch-Präfix nach work-conventions: <c>feat</c> → <c>feature</c>, Typen ohne eigenes Präfix → <c>chore</c>.</summary>
    public static string BranchTypeFor(string? commitType)
    {
        var t = (commitType ?? "").Trim().ToLowerInvariant();
        return t switch
        {
            "feat" => "feature",
            "fix" or "refactor" or "docs" or "test" or "chore" => t,
            _ => "chore",
        };
    }

    public async Task<WorktreeContext> CreateAsync(TaskEntity task, ListEntity list, CancellationToken ct)
    {
        var workingDir = list.WorkingDir
            ?? throw new InvalidOperationException("list.WorkingDir is null");

        if (!await _git.IsGitRepoAsync(workingDir, ct))
            throw new InvalidOperationException($"working_dir is not a git repository: {workingDir}");

        // A fresh `git init` repo has no HEAD commit to base a worktree on — bootstrap one.
        if (await _git.EnsureHeadCommitAsync(workingDir, ct))
            _logger.LogInformation("Repository at {Dir} had no commits; created an empty bootstrap commit", workingDir);

        var isCompany = await CompanyRepo.IsCompanyAsync(_git, workingDir, _cfg.CompanyRemotePattern, ct);
        var baseCommit = await ResolveBaseCommitAsync(task, workingDir, isCompany, ct);
        var name = NameFor(task, await ResolveTicketIdAsync(task, ct), isCompany);
        var branchType = BranchTypeFor(task.CommitType);
        var slug = CommitMessageBuilder.ToSlug(list.Name);

        string strategy;
        string? centralRoot;
        using (var settingsCtx = _dbFactory.CreateDbContext())
        {
            var settings = await new AppSettingsRepository(settingsCtx).GetAsync(ct);
            strategy = settings.WorktreeStrategy;
            centralRoot = !string.IsNullOrWhiteSpace(settings.CentralWorktreeRoot)
                ? settings.CentralWorktreeRoot
                : _cfg.CentralWorktreeRoot;
        }

        var parentDir = strategy.Equals("central", StringComparison.OrdinalIgnoreCase)
            ? Path.Combine(centralRoot ?? _cfg.CentralWorktreeRoot, slug)
            : Path.Combine(WorktreeRootResolver.ResolveSiblingRoot(workingDir), ".claudedo-worktrees", slug);

        // Numbers restart after a DB reset, and without the claudedo/ prefix a taken name may also be
        // the user's own branch — the self-heal below would force-delete it, the id suffix keeps it untouched.
        if (Directory.Exists(Path.Combine(parentDir, name))
            || (await _git.ListLocalBranchesAsync(workingDir, ct)).Contains($"{branchType}/{name}"))
        {
            var id = task.Id.Replace("-", "");
            name = $"{name}-{id[..Math.Min(8, id.Length)]}";
        }

        var branchName = $"{branchType}/{name}";
        var worktreePath = Path.GetFullPath(Path.Combine(parentDir, name));
        WorktreeRootResolver.EnsureOutsideWorkingDir(worktreePath, workingDir);

        // Ensure parent directory exists.
        Directory.CreateDirectory(Path.GetDirectoryName(worktreePath)!);

        // Create the worktree. If a stale branch from a previous run remains
        // (e.g. after force-remove), delete it and retry once.
        using var creating = _operations.Begin(OperationKinds.CreatingWorktree, task.Title, task.Id, task.Number, task.ListId);
        await BroadcastCreating(task.Id);
        try
        {
            await _git.WorktreeAddAsync(workingDir, branchName, worktreePath, baseCommit, ct);
        }
        catch (InvalidOperationException ex) when (ex.Message.Contains("already exists", StringComparison.OrdinalIgnoreCase))
        {
            _logger.LogWarning("Branch {Branch} already exists; cleaning phantom worktrees and retrying", branchName);
            // Self-heal is the slow section (list -> remove -> prune -> branch delete -> retry);
            // report it as its own section rather than staying silent through the retry.
            await BroadcastCreating(task.Id);

            // Find and forcefully remove any existing worktree registered against this branch.
            List<string> stalePaths;
            try { stalePaths = await _git.ListWorktreePathsForBranchAsync(workingDir, branchName, ct); }
            catch (Exception listEx)
            {
                _logger.LogWarning(listEx, "git worktree list failed during self-heal");
                stalePaths = new();
            }
            foreach (var stalePath in stalePaths)
            {
                try { await _git.WorktreeRemoveAsync(workingDir, stalePath, force: true, ct); }
                catch (Exception wrEx) { _logger.LogWarning(wrEx, "Failed to remove stale worktree at {Path}", stalePath); }
            }

            try { await _git.WorktreePruneAsync(workingDir, ct); }
            catch (Exception pruneEx) { _logger.LogWarning(pruneEx, "git worktree prune failed during self-heal"); }
            try { await _git.BranchDeleteAsync(workingDir, branchName, force: true, ct); }
            catch (Exception delEx) { _logger.LogWarning(delEx, "git branch -D failed during self-heal"); }

            await _git.WorktreeAddAsync(workingDir, branchName, worktreePath, baseCommit, ct);
        }

        // Insert worktrees row AFTER git succeeds — if git throws, no row is created.
        // If the insert itself fails, the worktree is already on disk with nothing
        // tracking it: remove it (best-effort, non-cancellable) before rethrowing.
        try
        {
            using var context = _dbFactory.CreateDbContext();
            var wtRepo = new WorktreeRepository(context);
            // Drop any stale row from a prior run (force-remove may have left the DB side behind).
            await wtRepo.DeleteAsync(task.Id, ct);
            await wtRepo.AddAsync(new WorktreeEntity
            {
                TaskId = task.Id,
                Path = worktreePath,
                BranchName = branchName,
                BaseCommit = baseCommit,
                HeadCommit = null,
                DiffStat = null,
                State = WorktreeState.Active,
                CreatedAt = DateTime.UtcNow,
            }, ct);
        }
        catch (Exception dbEx)
        {
            _logger.LogError(dbEx,
                "Failed to record worktree row for task {TaskId}; removing orphaned worktree at {Path}",
                task.Id, worktreePath);
            try { await _git.WorktreeRemoveAsync(workingDir, worktreePath, force: true, CancellationToken.None); }
            catch (Exception rmEx) { _logger.LogWarning(rmEx, "Failed to remove orphaned worktree at {Path}", worktreePath); }
            try { await _git.BranchDeleteAsync(workingDir, branchName, force: true, CancellationToken.None); }
            catch (Exception delEx) { _logger.LogWarning(delEx, "Failed to delete orphaned branch {Branch}", branchName); }
            throw;
        }

        _logger.LogInformation("Created worktree for task {TaskId} at {Path} (branch {Branch}, base {Base})",
            task.Id, worktreePath, branchName, baseCommit);

        return new WorktreeContext(worktreePath, branchName, baseCommit);
    }

    /// <returns>true if a commit was made; false if no changes.</returns>
    public async Task<bool> CommitIfChangedAsync(WorktreeContext ctx, TaskEntity task, ListEntity list, CancellationToken ct)
    {
        if (!await _git.HasChangesAsync(ctx.WorktreePath, ct))
        {
            _logger.LogInformation("No changes in worktree for task {TaskId}, skipping commit", task.Id);
            return false;
        }

        await _git.AddAllAsync(ctx.WorktreePath, ct);

        var isCompany = list.WorkingDir is not null
            && await CompanyRepo.IsCompanyAsync(_git, list.WorkingDir, _cfg.CompanyRemotePattern, ct);
        var message = CommitMessageBuilder.Build(
            task.CommitType, list.Name, task.Title, task.Description, task.Id, includeTrailer: !isCompany);

        await _git.CommitAsync(ctx.WorktreePath, message, ct);

        var head = await _git.RevParseHeadAsync(ctx.WorktreePath, ct);
        var diffStat = await _git.DiffStatAsync(ctx.WorktreePath, ctx.BaseCommit, head, ct: ct);

        using var context = _dbFactory.CreateDbContext();
        var wtRepo = new WorktreeRepository(context);
        await wtRepo.UpdateHeadAsync(task.Id, head, diffStat, ct);

        _logger.LogInformation("Committed changes for task {TaskId}: {Head}", task.Id, head);
        return true;
    }

    public async Task DiscardAsync(WorktreeEntity wt, string workingDir, CancellationToken ct)
    {
        try
        {
            await _git.WorktreeRemoveAsync(workingDir, wt.Path, force: true, ct);
        }
        catch (Exception ex)
        {
            _logger.LogError(ex, "git worktree remove failed for {Path}", wt.Path);
            throw;
        }

        try
        {
            await _git.BranchDeleteAsync(workingDir, wt.BranchName, force: true, ct);
        }
        catch (Exception ex)
        {
            _logger.LogWarning(ex, "git branch -D {Branch} failed after worktree removal; continuing", wt.BranchName);
        }

        using var context = _dbFactory.CreateDbContext();
        var wtRepo = new WorktreeRepository(context);
        await wtRepo.SetStateAsync(wt.TaskId, WorktreeState.Discarded, ct);

        _logger.LogInformation("Discarded worktree for task {TaskId} (branch {Branch})", wt.TaskId, wt.BranchName);
    }

    /// <summary>Ticket-Id des Tasks, sonst die seines Parents (Children erben das Ticket ihrer Unit).</summary>
    private async Task<int?> ResolveTicketIdAsync(TaskEntity task, CancellationToken ct)
    {
        var own = TicketStatusMap.ParseBandelTicketId(task.TicketRef);
        if (own is not null || task.ParentTaskId is null) return own;
        using var ctx = _dbFactory.CreateDbContext();
        var parentRef = await ctx.Tasks.AsNoTracking()
            .Where(t => t.Id == task.ParentTaskId)
            .Select(t => t.TicketRef)
            .FirstOrDefaultAsync(ct);
        return TicketStatusMap.ParseBandelTicketId(parentRef);
    }

    /// <summary>
    /// A child of a parent with an Active worktree starts from the unit branch's current head, so it
    /// sees every sibling absorbed before it. Unit roots (and legacy children) base off
    /// <c>list_config.base_branch</c>, else the current HEAD — in a company repo off a freshly fetched
    /// <c>origin/&lt;basis&gt;</c> (<see cref="FetchCompanyBaseAsync"/>).
    /// </summary>
    private async Task<string> ResolveBaseCommitAsync(TaskEntity task, string workingDir, bool isCompany, CancellationToken ct)
    {
        if (task.ParentTaskId is not null)
        {
            using var ctx = _dbFactory.CreateDbContext();
            var parentWt = await new WorktreeRepository(ctx).GetByTaskIdAsync(task.ParentTaskId, ct);
            if (parentWt is { State: WorktreeState.Active })
                return await _git.RevParseAsync(workingDir, parentWt.BranchName, ct);
        }
        using var cfgCtx = _dbFactory.CreateDbContext();
        var baseBranch = (await cfgCtx.ListConfigs.AsNoTracking()
            .Where(c => c.ListId == task.ListId)
            .Select(c => c.BaseBranch)
            .FirstOrDefaultAsync(ct)).NullIfBlank();

        if (isCompany)
        {
            try
            {
                if (await FetchCompanyBaseAsync(workingDir, baseBranch, ct) is { } fetched) return fetched;
            }
            catch (Exception ex) when (ex is not OperationCanceledException)
            {
                _logger.LogWarning(ex, "Fetching the company base in {working_dir} failed, continuing with the local branch", workingDir);
            }
        }

        return baseBranch is null
            ? await _git.RevParseHeadAsync(workingDir, ct)
            : await _git.RevParseAsync(workingDir, baseBranch, ct);
    }

    /// <summary>
    /// Fetches the basis (<c>base_branch</c>, else the ADO repo's default branch, else <c>origin/HEAD</c>) and
    /// returns <c>origin/&lt;basis&gt;</c>'s commit; the local basis is fast-forwarded best-effort. Null when no
    /// basis is known or the fetch failed (offline, no credentials) — the caller then uses the local basis,
    /// a fetch problem never fails the task.
    /// </summary>
    private async Task<string?> FetchCompanyBaseAsync(string workingDir, string? baseBranch, CancellationToken ct)
    {
        var remote = await _git.GetRemoteUrlAsync(workingDir, "origin", ct);
        var pat = _azdo?.Pat;
        var basis = baseBranch;
        if (basis is null && _azdo is not null && pat is not null && AzureDevOpsClient.ParseRemote(remote) is { } repo)
        {
            try { basis = await _azdo.GetDefaultBranchAsync(repo, ct); }
            catch (Exception ex) when (ex is AzureDevOpsException or InvalidOperationException) { _logger.LogWarning("Default branch from Azure DevOps unavailable: {reason}", ex.Message); }
        }
        basis ??= await _git.GetRemoteHeadBranchAsync(workingDir, ct);
        if (basis is null)
        {
            _logger.LogWarning("Company repo {working_dir}: no base branch known, worktree starts from HEAD without fetch", workingDir);
            return null;
        }

        var (fetchExit, fetchErr) = await _git.FetchBranchAsync(workingDir, basis, AzureDevOpsClient.GitAuthEnv(pat), ct);
        if (fetchExit != 0)
        {
            _logger.LogWarning("git fetch origin {branch} failed in {working_dir}, continuing with the local branch: {reason}",
                basis, workingDir, fetchErr);
            return baseBranch is null ? await TryRevParseAsync(workingDir, basis, ct) : null;
        }

        await FastForwardLocalBaseAsync(workingDir, basis, ct);
        return await _git.RevParseAsync(workingDir, $"refs/remotes/origin/{basis}", ct);
    }

    /// <summary>Pulls the local basis up to <c>origin/&lt;basis&gt;</c> without ever touching a dirty or diverged checkout.</summary>
    private async Task FastForwardLocalBaseAsync(string workingDir, string basis, CancellationToken ct)
    {
        string? current;
        try { current = await _git.GetCurrentBranchAsync(workingDir, ct); }
        catch (InvalidOperationException) { current = null; } // detached HEAD

        (int ExitCode, string Stderr) result;
        if (current == basis)
        {
            if (await _git.HasChangesAsync(workingDir, includeUntracked: false, ct))
            {
                _logger.LogWarning("Local {branch} in {working_dir} has uncommitted changes, not fast-forwarded to origin", basis, workingDir);
                return;
            }
            result = await _git.MergeFastForwardOnlyAsync(workingDir, $"refs/remotes/origin/{basis}", ct);
        }
        else
        {
            if (!(await _git.ListLocalBranchesAsync(workingDir, ct)).Contains(basis)) return;
            result = await _git.FastForwardLocalBranchAsync(workingDir, basis, ct);
        }
        if (result.ExitCode != 0)
            _logger.LogWarning("Local {branch} in {working_dir} not fast-forwarded to origin: {reason}", basis, workingDir, result.Stderr);
    }

    private async Task<string?> TryRevParseAsync(string workingDir, string revision, CancellationToken ct)
    {
        try { return await _git.RevParseAsync(workingDir, revision, ct); }
        catch (InvalidOperationException) { return null; }
    }
}