On Fri, Sep 24, 2021 at 3:43 AM Ævar Arnfjörð Bjarmason
<avarab@xxxxxxxxx> wrote:
>
>
> On Wed, Sep 22 2021, David Aguilar wrote:
>
> > @@ -580,7 +574,7 @@ static int run_dir_diff(const char *extcmd, int symlinks, const char *prefix,
> > flags = 0;
> > } else
> > setenv("GIT_DIFFTOOL_DIRDIFF", "true", 1);
> > - rc = run_command_v_opt(helper_argv, flags);
> > + ret = run_command_v_opt(helper_argv, flags);
> >
> > /* TODO: audit for interaction with sparse-index. */
> > ensure_full_index(&wtindex);
>
> Just on this part: There was already a logic error in the pre-image
> where we'd return error() return values up to cmd_main() et al. This
> means that (on POSIX) we return 255, not 1, per the C standard doing
> this is "implementation defined".
>
> I think you want the "rc" variable back, and either in another commit or
> in some other cleanup this on top or ahead of this patch:
Good catch. I submitted an incremental follow-up patch that addresses
this issue.
> diff --git a/builtin/difftool.c b/builtin/difftool.c
> index a867c49067c..6605d17e6bc 100644
> --- a/builtin/difftool.c
> +++ b/builtin/difftool.c
> @@ -345,6 +345,7 @@ static int run_dir_diff(const char *extcmd, int symlinks, const char *prefix,
> const char *helper_argv[] = { "difftool--helper", NULL, NULL, NULL };
> struct hashmap wt_modified, tmp_modified;
> int indices_loaded = 0;
> + int rc;
>
> workdir = get_git_work_tree();
>
> @@ -574,7 +575,7 @@ static int run_dir_diff(const char *extcmd, int symlinks, const char *prefix,
> flags = 0;
> } else
> setenv("GIT_DIFFTOOL_DIRDIFF", "true", 1);
> - ret = run_command_v_opt(helper_argv, flags);
> + rc = run_command_v_opt(helper_argv, flags);
>
> /* TODO: audit for interaction with sparse-index. */
> ensure_full_index(&wtindex);
> @@ -660,7 +661,7 @@ static int run_dir_diff(const char *extcmd, int symlinks, const char *prefix,
> strbuf_release(&tmpdir);
> UNLEAK(working_tree_dups);
>
> - return ret;
> + return ret < 0 ? 1 : rc;
> }
"return ret < 0 ? 1 : rc;" seems rather subtle.
In the submitted patch we're able to keep this as "return ret;" by
never assigning -1 in the first place.
My rationale is that it's better to not have to apply this fixup after the fact.
Thanks for reviewing.
--
David
