fix(clone): reject --no-checkout for non-bare layouts and cd into root on success

[avihut]

Summary

Two related --no-checkout fixes for git-worktree-clone:

• Reject --no-checkout for non-bare layouts. The flag promises a bare clone with no worktree, but for non-bare layouts (sibling, nested, centralized, contained-classic) the resolved repo_path is the working tree — there's no separate bare directory to leave behind. The flag silently degraded to a regular clone with an empty working tree. Now bails after layout resolution with an error naming the resolved layout and pointing to bare alternatives (contained, contained-flat). Cleans up the bare clone (already on disk from clone_bare_phase) so a rejected clone leaves no orphan directory.
• Cd into project root after successful --no-checkout. Previously the cd-redirect was gated on worktree_dir.is_some() (or branch_not_found), so the bare-only success case fell through and left the user in their original directory. Now writes parent_dir to DAFT_CD_FILE so the user lands in the new project root immediately. --no-cd still suppresses via the existing autocd check inside output.cd_path.

Updated --no-checkout help text and regenerated man pages + CLI docs to describe the bare-layout constraint.

Fixes #443

Test plan

• Unit tests cover all 6 builtin layouts plus a custom non-bare layout (cargo test --lib commands::clone::tests)
• YAML scenario clone:no-checkout-non-bare-rejected covers each non-bare builtin and asserts no orphan directory is left after rejection
• clone:no-checkout extended to assert cd target is the project root via DAFT_CD_FILE
• clone:empty-repo-no-checkout regression check still passes
• mise run fmt, mise run clippy, cargo clippy --all-targets -- -D warnings clean
• Live binary test confirmed cleanup preserves unrelated files in cwd

🤖 Generated with Claude Code

[avihut]

/claude review

[claude]

Claude finished @avihut's task in 2m 29s —— View job

---

PR Review

• Gather context: read diff and changed files
• Review src/commands/clone.rs for Rust correctness and idioms
• Review test coverage and regression tests
• Review YAML scenarios
• Review docs/man pages

---

Overall this is a clean, correct fix with solid test coverage. The two changes are well-scoped and the implementation follows existing patterns throughout the codebase. A few observations below, roughly ordered by priority.

Correctness

Cleanup path is correct. The check_no_checkout_compat error handler at clone.rs:284-290 correctly:

1. Returns to original_dir before removing the clone directory (avoiding a dangling-cwd after remove_directory).
2. Uses .ok() on both cleanup calls — appropriate, since masking the real error with a cleanup failure would be worse.
3. Returns the original error so the caller sees the meaningful message.

--no-cd is respected. output.cd_path (cli.rs:268-278) gates on self.config.autocd, which is set from settings.autocd && !args.no_cd. The new --no-checkout cd branch at clone.rs:504-509 goes through the same output.cd_path, so --no-cd suppresses it correctly — consistent with the existing branches.

result.parent_dir is canonical at the cd call. The code at clone.rs:347-349 captures canonical_parent_dir from std::env::current_dir() immediately after Phase 3, and clone.rs:428 overwrites result.parent_dir with it before the final output block. So the --no-checkout cd call at line 508 uses an absolute path. No issue.

---

Minor Issues

1. Eager layout validation opportunity (UX improvement, not a bug)

When --layout is passed explicitly on the CLI, the resolved layout is known before any network or disk work begins. Currently the flow is:

1. clone_bare_phase — full network clone to disk
2. Read daft.yml from bare clone
3. Resolve layout
4. check_no_checkout_compat — reject, clean up

For the explicit-layout case (most common user error), all that disk/network work is wasted. You could call check_no_checkout_compat immediately after validate_arg_combinations when args.layout.is_some(), giving users instant feedback with no clone:

// In validate_arg_combinations or just after it in run():
if args.no_checkout {
    if let Some(ref layout_str) = args.layout {
        if let Some(builtin) = BuiltinLayout::from_name(layout_str) {
            let layout = builtin.to_layout();
            check_no_checkout_compat(true, &layout)?;
        }
    }
}

This is an enhancement, not a blocker — the current behaviour is correct, just slower for the reject path.

Fix this →

2. YAML scenario: orphan-check only covers sibling

no-checkout-non-bare-rejected.yml has test ! -e $WORK_DIR/test-repo-no-co-sibling && echo CLEAN after the sibling rejection, but no equivalent assertions for nested, centralized, or contained-classic. Since the cleanup logic is unconditional (same code path for all non-bare layouts), this is a minor coverage gap, not a behaviour issue.

Fix this →

3. Custom-layout test asserts fewer properties than the builtin test

no_checkout_rejects_custom_non_bare_layout (clone.rs:1499-1508) only asserts err.to_string().contains("'my-custom'"). The parallel builtin test at clone.rs:1471-1495 also verifies the message contains --no-checkout and the bare-layout suggestions. Since the error message is generated by the same check_no_checkout_compat function (no special-casing for builtins vs. custom), a custom layout will always produce the same message shape, but the test doesn't confirm it. Minor consistency gap.

---

Nits

• BuiltinLayout::ContainedFlat acceptance not covered in YAML — no-checkout.yml uses --layout contained for its success test. Adding a step with contained-flat would mirror the unit test's no_checkout_accepts_bare_layouts coverage at the integration level.

• Comment in check_no_checkout_compat — The doc comment references -n as an alias (-n in the comment, --no-checkout in the error message). The flag is actually declared short = 'n', so this is accurate, but the comment refers to the bare-clone concern generically. Fine as-is; just noting the comment is about the function's contract, not the implementation — consistent with project style.

---

Summary

Two correct fixes, well-scoped. Test coverage hits all 6 builtin layouts at the unit level and the four non-bare builtins at the YAML integration level. Man pages and CLI docs are updated. The suggestions above are improvements (faster failure, fuller YAML coverage) but nothing here blocks the merge.