What causes "jj git export retry loop should always return" in nikivdev/code?

Verify the repo marker path shown in the message exists (ls ); recreate/re-clone the repo if deleted.

jj git export retry loop should always return

Error message

jj git export retry loop should always return

What it means

This error is thrown by a repository-marker resolution routine in src/workflow.rs (near line 784). Given a repo marker path (e.g. a .git or .jj marker), it tries git/jj resolution branches; if none applies and the marker file simply does not exist on disk, it bails with "expected <marker path> to exist". It is a precondition assertion that the marker file representing the repo was present.

Source

Thrown at src/sync.rs:4653

            std::thread::sleep(delay);
            continue;
        }

        jj_print_output(&output);
        let failure = jj_failure_message(&args, &output);
        if is_git_index_lock_error(&failure_text) {
            let retries = JJ_GIT_EXPORT_LOCK_RETRY_DELAYS_MS.len();
            bail!(
                "{}. Git index stayed locked after {} retr{}; close competing git/jj processes or remove stale .git/index.lock, then retry.",
                failure,
                retries,
                if retries == 1 { "y" } else { "ies" }
            );
        }
        bail!("{}", failure);
    }

    unreachable!("jj git export retry loop should always return");
}

fn jj_run_in(repo_root: &Path, args: &[&str]) -> Result<()> {
    let output = jj_run_output_in(repo_root, args)?;
    jj_print_output(&output);
    if !output.status.success() {
        bail!("{}", jj_failure_message(args, &output));
    }
    Ok(())
}

fn jj_preferred_binary() -> std::path::PathBuf {
    if let Ok(path) = std::env::var("FLOW_JJ_BIN") {
        let candidate = PathBuf::from(path);
        if candidate.exists() {
            return candidate;
        }
    }

View on GitHub (pinned to a747e741ae)

Solutions

Verify the repo marker path shown in the message exists (ls <path>); recreate/re-clone the repo if deleted.
Fix the workspace/config entry that references the stale repository path.
Run the workflow from the correct repo root so the marker resolves.

Example fix

// before (stale path in workspace config)
repo_marker = /home/user/work/old-repo/.git
// after
+ git clone git@host:work/old-repo.git /home/user/work/old-repo
+ # or update config to the new path /home/user/work/new-repo/.git

Defensive patterns

Strategy: validation

Validate before calling

if !std::path::Path::new(&repo_marker).exists() {
    return Err(anyhow!("repo marker {} missing; fix workspace config", repo_marker.display()));
}

Type guard

fn repo_marker_exists(p: &std::path::Path) -> bool { p.exists() }

Try / catch

match resolve_repo_marker(marker) {
    Err(e) if e.to_string().contains("expected") && e.to_string().contains("to exist") => {
        eprintln!("re-clone or fix the workspace path: {}", marker.display());
    }
    r => r?,
}

Prevention

Keep workspace config paths in sync when moving/cloning repos.
Check repo existence before running multi-repo workflows.
Avoid deleting repo directories referenced by active workspaces.

When it happens

Trigger: Calling the repo-resolution helper with a repo_marker path whose file/directory does not exist on disk, and the path is not under a .jj repo (so the jj parent-resolution branch is skipped).

Common situations: A workspace entry pointing at a moved/deleted repo; config referencing a repo path that was never cloned; a stale cache entry after the repository directory was removed; running from a subpath after the repo was relocated.

Related errors

AI-assisted analysis of nikivdev/code@a747e741ae (2026-09-01). Data as JSON: /api/errors/f522f5202394b111. Report an issue: GitHub.