Re: GIT_WORK_TREE is not exported for post-checkout-hook

[Junio C Hamano <gitster@pobox.com>]

Phil Sainty <psainty@orcon.net.nz> writes:

> I have a post-checkout-hook shell script which indirectly calls
> "git rev-parse --show-toplevel", and I've found that when that hook
> runs for a checkout within a worktree directory, GIT_DIR is set but
> GIT_WORK_TREE is not set, which leads to the CWD being returned
> instead of the genuine toplevel value.
>
> The script is seeing a GIT_DIR value like this:
>
>      GIT_DIR=/path/to/repo/.git/worktrees/name
>
> The only other git variables in the environment are GIT_EXEC_PATH
> and GIT_PREFIX.
>
> The outcome I'm seeing is explained by
> https://git-scm.com/book/en/v2/Git-Internals-Environment-Variables#_repository_locations
> which says:
>
>      "If --git-dir or GIT_DIR is specified but none of --work-tree,
>      GIT_WORK_TREE or core.worktree is specified, the current working
>      directory is regarded as the top level of your working tree."
>
> The current githooks(5) man page talks about this somewhat:
>
>      "Environment variables, such as GIT_DIR, GIT_WORK_TREE, etc., are
>      exported so that Git commands run by the hook can correctly locate
>      the repository.  If your hook needs to invoke Git commands in a
>      foreign repository or in a different working tree of the same
>      repository, then it should clear these environment variables so
>      they do not interfere with Git operations at the foreign
>      location."
>
> Unsetting GIT_DIR certainly fixes the problem, but I don't think that
> ought to be necessary in my case, as my hook script is calling git in
> the *same* working copy where the checkout happens.  It is not dealing
> with "a foreign" repository [or] a different working tree".
>
> My impression is that GIT_WORK_TREE should also be set for this hook
> when dealing with a worktree.  I have tested this in git version 2.34
> and 2.55 (maint) and also 2.56.0.rc0.355.g945ea7012d (next).

That certainly would be a possible workaround.  I do not know why we
need to export GIT_DIR in the first place, though.

The attached patch would export GIT_WORK_TREE when worktree exists
and GIT_DIR would also be exported, but there is a large NEEDSWORK
comment in the precontext of the hunk, which merits consideration
before moving forward.

 setup.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git c/setup.c w/setup.c
index 0d157ac254..d0c0b2f68a 100644
--- c/setup.c
+++ w/setup.c
@@ -2064,8 +2064,11 @@ const char *setup_git_directory_gently(struct repository *repo, int *nongit_ok)
 	    startup_info->have_repository ||
 	    /* GIT_DIR_EXPLICIT */
 	    getenv(GIT_DIR_ENVIRONMENT)) {
-		if (discovery.worktree)
+		if (discovery.worktree) {
 			set_git_work_tree(repo, discovery.worktree);
+			if (discovery.gitdir || getenv(GIT_DIR_ENVIRONMENT))
+				xsetenv(GIT_WORK_TREE_ENVIRONMENT, discovery.worktree, 1);
+		}
 
 		if (discovery.gitdir) {
 			apply_and_export_relative_gitdir(repo, discovery.gitdir, 0);