patchsetup.c: set workdir when gitdir is not default

> When gitfile is used, git sets GIT_DIR environment variable for
> subsequent commands, and that commands start working in mode "GIT_DIR
> set, workdir current", which is incorrect for the case when git runs
> from subdirectory of repository. This can be observed at least for
> running aliases - git fails with message "internal error: work tree has
> already been set"
>
> Fix by setting GIT_WORK_TREE environment also.
>
> Add test which demonstrates problem with alias.
>
> Signed-off-by: Max Kirillov <max@max630.net>
> ---
>  setup.c            | 4 +++-
>  t/t0002-gitfile.sh | 7 +++++++
>  2 files changed, 10 insertions(+), 1 deletion(-)
>
> diff --git a/setup.c b/setup.c
> index 0a22f8b..bcf4e31 100644
> --- a/setup.c
> +++ b/setup.c
> @@ -508,8 +508,10 @@ static const char *setup_discovered_git_dir(const char *gitdir,
>
>         /* #0, #1, #5, #8, #9, #12, #13 */
>         set_git_work_tree(".");
> -       if (strcmp(gitdir, DEFAULT_GIT_DIR_ENVIRONMENT))
> +       if (strcmp(gitdir, DEFAULT_GIT_DIR_ENVIRONMENT)) {
>                 set_git_dir(gitdir);
> +               setenv(GIT_WORK_TREE_ENVIRONMENT, get_git_work_tree(), 1);
> +       }
>         inside_git_dir = 0;
>         inside_work_tree = 1;
>         if (offset == len)
> diff --git a/t/t0002-gitfile.sh b/t/t0002-gitfile.sh
> index 37e9396..428cfdc 100755
> --- a/t/t0002-gitfile.sh
> +++ b/t/t0002-gitfile.sh
> @@ -99,4 +99,11 @@ test_expect_success 'check rev-list' '
>         test "$SHA" = "$(git rev-list HEAD)"
>  '
>
> +test_expect_success 'check alias call from subdirectory' '
> +       git config alias.testalias "rev-parse HEAD" &&

> +'
> +
>  test_done
> --
> 2.0.1.1697.g73c6810

> > +       mkdir -p subdir &&
> > +       cd subdir &&
> > +       git testalias
> 
> If a new test is added following this one, it will be run from within
> 'subdir', which might come as a surprise as the author of the new
> test. Wrapping the 'cd' and following code in a subshell is a good way
> to avoid the problem.

> diff --git a/setup.c b/setup.c
> index 0a22f8b..bcf4e31 100644
> --- a/setup.c
> +++ b/setup.c
> @@ -508,8 +508,10 @@ static const char *setup_discovered_git_dir(const char *gitdir,
>
>         /* #0, #1, #5, #8, #9, #12, #13 */
>         set_git_work_tree(".");

> -       if (strcmp(gitdir, DEFAULT_GIT_DIR_ENVIRONMENT))
> +       if (strcmp(gitdir, DEFAULT_GIT_DIR_ENVIRONMENT)) {
>                 set_git_dir(gitdir);
> +               setenv(GIT_WORK_TREE_ENVIRONMENT, get_git_work_tree(), 1);
> +       }
>         inside_git_dir = 0;
>         inside_work_tree = 1;
>         if (offset == len)

> On Thu, Sep 4, 2014 at 5:42 AM, Max Kirillov <max@max630.net> wrote:
>>         /* #0, #1, #5, #8, #9, #12, #13 */
>>         set_git_work_tree(".");
> 
> I wonder if we should setenv(GIT_WORK_TREE_) from inside this function
> instead. A quick glance over 'git grep set_git_work_tree' gives me the
> impression that it's safe to do so, and could cover future bugs
> similar to this.