feat(git): support GIT_DIR/GIT_WORK_TREE for bare-repo dotfile managers by jdx · Pull Request #847 · jdx/hk

jdx · 2026-04-19T00:19:00Z

Summary

Honor GIT_DIR / GIT_WORK_TREE env vars during repository discovery so hk works with YADM and other bare-repo dotfile managers where there is no .git in the work tree
Fall back to shell-git when libgit2 opens a bare repo (libgit2 refuses status/diff on bare repos even with a GIT_WORK_TREE override)
hk builtins no longer loads project settings, so it works outside a repo instead of panicking

Fixes #831.

Changes

src/git.rs — Git::new() uses Repository::open_from_env() when either env var is set; bare repos fall back to shell git
src/git_util.rs — find_git_path() returns $GIT_DIR when set; new find_work_tree_root() helper
src/tera.rs, src/test_runner.rs — use find_work_tree_root() instead of walking up for .git
src/cli/mod.rs — add Commands::Builtins to the settings-skip list

Test plan

hk check, hk fix, hk install, hk uninstall, hk builtins all work against a bare repo + GIT_WORK_TREE setup
Verified on both HK_LIBGIT2=1 (default) and HK_LIBGIT2=0
Regression-tested against a normal (non-bare) repo

🤖 Generated with Claude Code

Note

Medium Risk
Moderate risk because it changes repository discovery and working-directory behavior in Git::new(), which can affect path resolution and git operations across many commands; mitigated by explicit fallbacks and new regression tests.

Overview
Enables hk to operate in bare-repo dotfile-manager setups (e.g., YADM) by honoring GIT_DIR/GIT_WORK_TREE, normalizing relative git env vars, and always cding to the effective work-tree root during Git::new().

When libgit2 opens a bare repo, git status/diff now falls back to shell git to keep operations working. find_git_path() and a new find_work_tree_root() helper are added/used to resolve roots consistently, and hk builtins now skips project settings loading so it runs outside a repo. Adds bats regression tests covering check/install/uninstall and subdirectory execution under bare-repo env vars.

^{Reviewed by Cursor Bugbot for commit 3799316. Bugbot is set up for automated code reviews on this repo. Configure here.}

Honor the standard git environment variables during repository discovery so hk works with YADM and other bare-repo dotfile managers where there is no `.git` in the work tree. - `Git::new()` uses `Repository::open_from_env()` when either var is set, and falls back to the shell-git path when the opened repo is bare (libgit2 refuses status/diff on bare repos even when a work tree is provided via `GIT_WORK_TREE`). - `find_git_path()` returns `$GIT_DIR` directly when set so hooks resolve to the bare repository's hooks dir. - `tera.rs` and `test_runner.rs` use a new `find_work_tree_root()` helper that honors `GIT_WORK_TREE`. - `hk builtins` no longer loads project settings, so it works outside a repo instead of panicking. Fixes #831 Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>

Regression test for the GIT_DIR/GIT_WORK_TREE support, exercising both the libgit2 and shell-git code paths. Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>

gemini-code-assist

Code Review

This pull request introduces support for GIT_DIR and GIT_WORK_TREE environment variables, enabling compatibility with bare-repository managers like YADM. It centralizes work-tree resolution logic in git_util and updates the Git initialization to handle bare repositories by falling back to shell-based Git when libgit2 is insufficient. Feedback was provided regarding the handling of current directory changes when Git environment variables are present, specifically the need to absolute-ify relative paths and ensure the process remains at the repository root to avoid breaking internal operations. Additionally, a suggestion was made to improve error handling when retrieving the current working directory to avoid issues with empty paths.

gemini-code-assist · 2026-04-19T00:20:53Z

+        if !has_git_env {
+            let cwd = std::env::current_dir()?;
+            let root = xx::file::find_up(&cwd, &[".git"])
+                .and_then(|p| p.parent().map(|p| p.to_path_buf()))
+                .ok_or(eyre!("failed to find git repository"))?;
+            std::env::set_current_dir(&root)?;
+        }


By skipping set_current_dir when GIT_DIR or GIT_WORK_TREE is set, hk remains in the user's current working directory. This breaks many internal operations (like status or all_files) that expect the process to be at the repository root to correctly resolve relative paths returned by Git.

Additionally, if we do change the directory, any relative paths in GIT_DIR or GIT_WORK_TREE must be converted to absolute paths first, otherwise they will point to the wrong location after the cd.

let root = crate::git_util::find_work_tree_root(); if !has_git_env { let cwd = std::env::current_dir()?; if xx::file::find_up(&cwd, &[".git"]).is_none() { return Err(eyre!("failed to find git repository")); } } // Absolute-ify relative git env vars before changing directory for var in ["GIT_DIR", "GIT_WORK_TREE"] { if let Some(val) = std::env::var_os(var) { let p = std::path::Path::new(&val); if p.is_relative() { std::env::set_var(var, std::env::current_dir()?.join(p)); } } } std::env::set_current_dir(&root)?;

gemini-code-assist · 2026-04-19T00:20:53Z

+/// (for bare-repo setups like YADM). Falls back to walking up for `.git`, and
+/// finally to `cwd` if no repository is found.
+pub fn find_work_tree_root() -> PathBuf {
+    let cwd = std::env::current_dir().unwrap_or_default();


Using unwrap_or_default() on current_dir() returns an empty PathBuf if the call fails. This can lead to confusing behavior or incorrect path joining later. It's safer to handle the error or at least fall back to . explicitly.

Suggested change

let cwd = std::env::current_dir().unwrap_or_default();

let cwd = std::env::current_dir().unwrap_or_else(|_| PathBuf::from("."));

greptile-apps · 2026-04-19T00:23:03Z

Greptile Summary

This PR adds support for bare-repo dotfile managers (YADM, etc.) by honoring GIT_DIR/GIT_WORK_TREE env vars during repository discovery, falling back to shell git when libgit2 encounters a bare repo, and skipping project-settings loading for hk builtins. The previous reviewer concern about CWD not being set for non-bare env-var repos is addressed by the fix commit (always cd to work tree root when GIT_DIR/GIT_WORK_TREE is set) and covered by a new bats regression test.

Confidence Score: 5/5

Safe to merge; the one remaining concern is a narrow theoretical race with set_var in a live tokio runtime, which has very low practical impact.

All changes are well-scoped: the previously flagged CWD issue is addressed and regression-tested, the bare-repo fallback logic is clean, and the builtins settings-skip is a safe no-op for existing users. The only outstanding finding is a P2 note about set_var soundness in an async context — not a production blocker.

src/git.rs lines 119–129 (unsafe set_var in live tokio runtime)

Important Files Changed

Filename	Overview
src/git.rs	Git::new() now absolutizes GIT_DIR/GIT_WORK_TREE/GIT_INDEX_FILE, uses open_from_env() when env vars are set, and falls back to shell git for bare repos; uses set_var in a live tokio runtime (technically unsound).
src/git_util.rs	New find_work_tree_root() helper correctly honors GIT_WORK_TREE and falls back to walking up for .git; find_git_path() updated to return GIT_DIR when set.
src/tera.rs	BASE_CONTEXT now derives root via find_work_tree_root() which correctly handles GIT_WORK_TREE; LazyLock initialized after Git::new() sets CWD so ordering is safe in practice.
src/cli/mod.rs	Commands::Builtins added to the settings-skip list so it no longer tries to load project config, allowing it to run outside a repo without panicking.
test/bare_repo_env_vars.bats	New Bats regression suite covering builtins (outside repo), check (libgit2 on/off), install/uninstall to bare hooks dir, and the previously-flagged subdirectory path-resolution bug.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[Git::new] --> B{GIT_DIR or\nGIT_WORK_TREE set?}
    B -- yes --> C[Absolutize relative\nGIT_DIR / GIT_WORK_TREE\n/ GIT_INDEX_FILE via set_var]
    C --> D[root = find_work_tree_root\nhonors GIT_WORK_TREE]
    B -- no --> E[root = walk up for .git\nparent dir]
    D --> F[set_current_dir to root]
    E --> F
    F --> G{HK_LIBGIT2?}
    G -- yes --> H{has_git_env?}
    H -- yes --> I[Repository::open_from_env]
    H -- no --> J[Repository::open dot]
    I --> K{repo.is_bare?}
    J --> K
    K -- yes --> L[repo = None\nshell-git fallback]
    K -- no --> M[Apply GIT_INDEX_FILE\nif set, repo = Some]
    G -- no --> L
    L --> N[Git struct created]
    M --> N

_{Reviews (2): Last reviewed commit: "fix(git): always cd to work tree root wh..." | Re-trigger Greptile}

cursor

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 4d3c390. Configure here.}

Addresses code-review feedback on #847. Skipping set_current_dir in the env-var branch left cwd pointing at a subdirectory of the work tree, which broke path.exists() filtering in status() — modified files silently disappeared from hk's file list when running from a subdir. Always cd to the work tree root after absolutizing relative GIT_DIR / GIT_WORK_TREE / GIT_INDEX_FILE values (so libgit2 and shell-git resolve them correctly after the cwd change). Adds a regression test that modifies a file at the work-tree root, runs hk check from a subdirectory, and asserts the file is processed. The test fails on both HK_LIBGIT2=1 and HK_LIBGIT2=0 without this fix. Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>

@jdx

jdx and others added 2 commits April 18, 2026 19:17

test(git): cover bare-repo dotfile manager scenario

4d3c390

Regression test for the GIT_DIR/GIT_WORK_TREE support, exercising both the libgit2 and shell-git code paths. Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>

gemini-code-assist Bot reviewed Apr 19, 2026

View reviewed changes

greptile-apps Bot reviewed Apr 19, 2026

View reviewed changes

Comment thread src/git.rs Outdated

cursor Bot reviewed Apr 19, 2026

View reviewed changes

Comment thread src/git.rs Outdated

jdx merged commit ad8cf88 into main Apr 19, 2026
20 checks passed

jdx deleted the fix/git-dir-work-tree-env-vars branch April 19, 2026 00:58

jdx mentioned this pull request Apr 19, 2026

chore: release v1.44.0 #837

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

feat(git): support GIT_DIR/GIT_WORK_TREE for bare-repo dotfile managers#847

feat(git): support GIT_DIR/GIT_WORK_TREE for bare-repo dotfile managers#847
jdx merged 3 commits intomainfrom
fix/git-dir-work-tree-env-vars

jdx commented Apr 19, 2026 •

edited by cursor Bot

Loading

Uh oh!

gemini-code-assist Bot left a comment

Uh oh!

gemini-code-assist Bot Apr 19, 2026

Uh oh!

gemini-code-assist Bot Apr 19, 2026

Uh oh!

greptile-apps Bot commented Apr 19, 2026 •

edited

Loading

Uh oh!

Uh oh!

cursor Bot left a comment

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

	let cwd = std::env::current_dir().unwrap_or_default();
	let cwd = std::env::current_dir().unwrap_or_else(\|_\| PathBuf::from("."));

Uh oh!

Conversation

jdx commented Apr 19, 2026 • edited by cursor Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Changes

Test plan

Uh oh!

gemini-code-assist Bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

gemini-code-assist Bot Apr 19, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist Bot Apr 19, 2026

Choose a reason for hiding this comment

Uh oh!

greptile-apps Bot commented Apr 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Greptile Summary

Confidence Score: 5/5

Important Files Changed

Flowchart

Uh oh!

Uh oh!

cursor Bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

jdx commented Apr 19, 2026 •

edited by cursor Bot

Loading

greptile-apps Bot commented Apr 19, 2026 •

edited

Loading