On Tue, Dec 10, 2024 at 07:36:26PM +0800, Jiang Xin wrote:
> diff --git a/send-pack.c b/send-pack.c
> index 6677c44e8a..f1556dd53c 100644
> --- a/send-pack.c
> +++ b/send-pack.c
> @@ -630,7 +630,7 @@ int send_pack(struct send_pack_args *args,
> reject_atomic_push(remote_refs, args->send_mirror);
> error("atomic push failed for ref %s. status: %d",
> ref->name, ref->status);
> - ret = args->porcelain ? 0 : -1;
> + ret = ERROR_SEND_PACK_BAD_REF_STATUS;
> goto out;
> }
> /* else fallthrough */
Okay, this change looks sensible -- instead of having different
behaviour with `--porcelain` we now have the exact same behaviour with
and without that flag set, which is good.
> @@ -761,11 +761,6 @@ int send_pack(struct send_pack_args *args,
> if (ret < 0)
> goto out;
>
> - if (args->porcelain) {
> - ret = 0;
> - goto out;
> - }
We don't reset the error code here anymore.
> for (ref = remote_refs; ref; ref = ref->next) {
> switch (ref->status) {
> case REF_STATUS_NONE:
> @@ -773,7 +768,7 @@ int send_pack(struct send_pack_args *args,
> case REF_STATUS_OK:
> break;
> default:
> - ret = -1;
> + ret = ERROR_SEND_PACK_BAD_REF_STATUS;
> goto out;
> }
> }
And this is the equivalent change without `--atomic`.
Also, none of the existing code paths used to set `ret = 1`. We do call
`ret = receive_status()` and bubble that up, but neither that nor any of
its transitive calls set `ret = 1`, either. So this should be fine and
we don't seem to have conflicting error codes.
> diff --git a/send-pack.h b/send-pack.h
> index 7edb80596c..ee88f9fe9f 100644
> --- a/send-pack.h
> +++ b/send-pack.h
> @@ -12,6 +12,9 @@ struct ref;
> #define SEND_PACK_PUSH_CERT_IF_ASKED 1
> #define SEND_PACK_PUSH_CERT_ALWAYS 2
>
> +/* Custom exit code from send_pack. */
> +#define ERROR_SEND_PACK_BAD_REF_STATUS 1
> +
> struct send_pack_args {
> const char *url;
> unsigned verbose:1,
It might help to document `send_pack()` accordingly and point out the
special return code.
> diff --git a/transport.c b/transport.c
> index 47fda6a773..454d7f21a9 100644
> --- a/transport.c
> +++ b/transport.c
> @@ -914,6 +914,13 @@ static int git_transport_push(struct transport *transport, struct ref *remote_re
> case protocol_v0:
> ret = send_pack(&args, data->fd, data->conn, remote_refs,
> &data->extra_have);
> + /*
> + * Ignore the specific error code to maintain consistent behavior
> + * with the "push_refs()" function across different transports,
> + * such as "push_refs_with_push()" for HTTP protocol.
> + */
> + if (ret == ERROR_SEND_PACK_BAD_REF_STATUS)
> + ret = 0;
> break;
> case protocol_unknown_version:
> BUG("unknown protocol version");
And here we ignore the error code to retain compatibility.
Is there any case where this causes a user-visible change in behaviour
already? If so, we can add a test for this?
Patrick
