hooks: skip rebuild in linked git worktrees

[Claude-Madera]

Problem

The post-commit and post-checkout hooks rebuild graphify-out/ on every commit and branch switch. Inside a linked worktree (git worktree add) this misbehaves:

• The canonical graphify-out/ belongs to the primary checkout; rebuilding it from every worktree is wasteful.

• Deploy/CI flows that create a worktree, switch/checkout, then git clean it, race the detached background rebuild against the cleanup — producing:

  failed to remove graphify-out/: Directory not empty
  

This surfaced in a deploy "cockpit" that pins release worktrees of three repos: every pin fired a pointless full rebuild and intermittently failed the git clean.

Fix

Guard both hooks: a linked worktree has git-dir != git-common-dir, so short-circuit before launching the rebuild.

Key correctness detail: the compare uses absolute paths —

_GFY_GITDIR=$(git rev-parse --absolute-git-dir 2>/dev/null)
_GFY_COMMONDIR=$(cd "$(git rev-parse --git-common-dir 2>/dev/null)" 2>/dev/null && pwd)
if [ -n "$_GFY_COMMONDIR" ] && [ "$_GFY_GITDIR" != "$_GFY_COMMONDIR" ]; then
    exit 0
fi

Git exports GIT_DIR to hooks and it is sometimes absolute; comparing a raw $GIT_DIR against a relative --git-common-dir (.git) would false-positive and wrongly skip the primary checkout. Resolving both to absolute paths avoids that. Factored into a shared _WORKTREE_GUARD constant, mirroring the existing _PYTHON_DETECT pattern.

Tests

• test_hooks_skip_linked_worktrees — both generated scripts contain the guard exactly once and use the absolute-path compare.
• test_worktree_guard_skips_only_linked_worktree — end-to-end against a real git worktree: the guard runs through on the primary checkout and exits early in a linked worktree.

Both pass; no previously-passing tests affected.

🤖 Generated with Claude Code

[safishamsi]

Thanks for this, @Claude-Madera — the worktree diagnosis and especially the absolute-path detail (raw $GIT_DIR can be absolute while --git-common-dir is the relative .git, which would false-positive the primary checkout) were exactly right, and I've carried that guard into 5f57c46 (shipping in 0.9.14).

I landed it directly rather than merging so the same commit could also close the two related gaps from #1809 that this PR didn't cover: post-checkout not honoring GRAPHIFY_SKIP_HOOK, and the .git/info/exclude bloat in #1810. A couple of small deltas from your version: the guard normalizes both paths with cd ... && pwd instead of git rev-parse --absolute-git-dir (works on older git without the newer flag), and I left the existing comment em-dashes alone to keep the diff focused on behavior. Credited you in the commit trailer. Really appreciate the PR.