"Johannes Schindelin via GitGitGadget" <gitgitgadget@xxxxxxxxx> writes: > diff --git a/builtin/stash.c b/builtin/stash.c > index b5a301f24d..a1e2e7ae7e 100644 > --- a/builtin/stash.c > +++ b/builtin/stash.c > @@ -497,6 +497,8 @@ static int do_apply_stash(const char *prefix, struct stash_info *info, > */ > cp.git_cmd = 1; > cp.dir = prefix; > + argv_array_pushf(&cp.env_array, GIT_WORK_TREE_ENVIRONMENT"=%s", > + absolute_path(get_git_work_tree())); > argv_array_push(&cp.args, "status"); > run_command(&cp); > } Nicely spotted. Exporting GIT_WORK_TREE alone without GIT_DIR feels a bit disturbing, at least to me, though. I wondered if this misbehaves when the end user has GIT_WORK_TREE environment exported, but in such a case, get_git_work_tree() would return that directory, and by re-exporting it to the child process, we would honor the end user's intention, so all is good, I think. Thanks. > diff --git a/t/t3908-stash-in-worktree.sh b/t/t3908-stash-in-worktree.sh > new file mode 100755 > index 0000000000..2b2b366ef9 > --- /dev/null > +++ b/t/t3908-stash-in-worktree.sh > @@ -0,0 +1,27 @@ > +#!/bin/sh > +# > +# Copyright (c) 2019 Johannes E Schindelin > +# > + > +test_description='Test git stash in a worktree' > + > +. ./test-lib.sh > + > +test_expect_success 'setup' ' > + test_commit initial && > + git worktree add wt && > + test_commit -C wt in-worktree > +' > + > +test_expect_success 'apply in subdirectory' ' > + mkdir wt/subdir && > + ( > + cd wt/subdir && > + echo modified >../initial.t && > + git stash && > + git stash apply >out > + ) && > + grep "\.\.\/initial\.t" wt/subdir/out > +' > + > +test_done
