transport: add client support for object-info
Sometimes, it is beneficial to retrieve information about an object without downloading it entirely. The server-side logic for this functionality was implemented in commit "a2ba162cda (object-info: support for retrieving object info, 2021-04-20)." And the wire format is documented at https://git-scm.com/docs/protocol-v2#_object_info. Introduce client-side support for the object-info capability. Add its own function for object-info separate from existing fetch infrastructure. Currently, the client supports requesting a list of OIDs with the size attribute from a v2 server. If the server does not advertise this feature (i.e., transfer.advertiseobjectinfo is set to false), the client returns an error and exits. Note that: 1. The entire request is written into req_buf before being sent to the remote. This approach follows the pattern used in the send_fetch_request() logic within 'fetch-pack.c'. Streaming the request is not addressed in this patch. 2. A new field 'unrecognized' has been added to object_info. This new field is set at fetch_object_info() when the object is unrecognized by the server. Helped-by: Jonathan Tan <jonathantanmy@google.com> Helped-by: Christian Couder <chriscool@tuxfamily.org> Signed-off-by: Calvin Wan <calvinwan@google.com> Signed-off-by: Eric Ju <eric.peijian@gmail.com> Signed-off-by: Pablo Sabater <pabloosabaterr@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Calvin Wan committed
Jul 24, 2026 at 12:54 UTC
12a19eec1d7c0121e0c82c95a2300fdbad1f18f2
9 files changed
+240
Makefile
+1
@@ -1158,6 +1158,7 @@ LIB_OBJS += ewah/ewah_io.o
1158
LIB_OBJS += ewah/ewah_rlw.o
1159
LIB_OBJS += exec-cmd.o
1160
LIB_OBJS += fetch-negotiator.o
1161
+LIB_OBJS += fetch-object-info.o
1162
LIB_OBJS += fetch-pack.o
1163
LIB_OBJS += fmt-merge-msg.o
1164
LIB_OBJS += fsck.o
fetch-object-info.c
new
+138
@@ -0,0 +1,138 @@
1
+#include "git-compat-util.h"
2
+#include "gettext.h"
3
+#include "hex.h"
4
+#include "pkt-line.h"
5
+#include "connect.h"
6
+#include "oid-array.h"
7
+#include "odb.h"
8
+#include "fetch-object-info.h"
9
+#include "string-list.h"
10
+
11
+/* Sends object-info command and its arguments into the request buffer. */
12
+static void send_object_info_request(const int fd_out, struct object_info_args *args)
13
+{
14
+ struct strbuf req_buf = STRBUF_INIT;
15
+
16
+ write_command_and_capabilities(&req_buf, "object-info", args->server_options);
17
+
18
+ if (unsorted_string_list_has_string(args->object_info_options, "size"))
19
+ packet_buf_write(&req_buf, "size");
20
+ else if (args->object_info_options->nr)
21
+ BUG("only size should be in object_info_options");
22
+
23
+ if (args->oids)
24
+ for (size_t i = 0; i < args->oids->nr; i++)
25
+ packet_buf_write(&req_buf, "oid %s", oid_to_hex(&args->oids->oid[i]));
26
+
27
+ packet_buf_flush(&req_buf);
28
+ if (write_in_full(fd_out, req_buf.buf, req_buf.len) < 0)
29
+ die_errno(_("unable to write request to remote"));
30
+
31
+ strbuf_release(&req_buf);
32
+}
33
+
34
+static int parse_object_size(const char *s, size_t *res)
35
+{
36
+ uintmax_t uim;
37
+
38
+ if (!s[0] || s[strspn(s, "0123456789")])
39
+ return -1;
40
+ errno = 0;
41
+ uim = strtoumax(s, NULL, 10);
42
+ if (errno || uim > SIZE_MAX)
43
+ return -1;
44
+ *res = uim;
45
+ return 0;
46
+}
47
+
48
+int fetch_object_info(const enum protocol_version version, struct object_info_args *args,
49
+ struct packet_reader *reader, struct object_info *object_info_data,
50
+ const int stateless_rpc, const int fd_out)
51
+{
52
+ int size_index = -1;
53
+
54
+ switch (version) {
55
+ case protocol_v2:
56
+ if (!server_supports_v2("object-info"))
57
+ die(_("object-info capability is not enabled on the server"));
58
+ send_object_info_request(fd_out, args);
59
+ break;
60
+ case protocol_v1:
61
+ case protocol_v0:
62
+ die(_("object-info requires protocol v2"));
63
+ case protocol_unknown_version:
64
+ BUG("unknown protocol version");
65
+ }
66
+
67
+ for (size_t i = 0; i < args->object_info_options->nr; i++) {
68
+ if (packet_reader_read(reader) != PACKET_READ_NORMAL) {
69
+ check_stateless_delimiter(stateless_rpc, reader,
70
+ "stateless delimiter expected");
71
+ return -1;
72
+ }
73
+
74
+ if (!string_list_has_string(args->object_info_options, reader->line))
75
+ return -1;
76
+
77
+ if (!strcmp(reader->line, "size")) {
78
+ /*
79
+ * i is the number of supported options which currently
80
+ * is only size. No risk of overflow.
81
+ */
82
+ size_index = (int)i;
83
+ for (size_t j = 0; j < args->oids->nr; j++)
84
+ object_info_data[j].sizep =
85
+ xcalloc(1, sizeof(*object_info_data[j].sizep));
86
+ } else {
87
+ BUG("only size is supported");
88
+ }
89
+ }
90
+
91
+ for (size_t i = 0;
92
+ packet_reader_read(reader) == PACKET_READ_NORMAL &&
93
+ i < args->oids->nr;
94
+ i++) {
95
+ struct string_list object_info_values = STRING_LIST_INIT_DUP;
96
+
97
+ string_list_split(&object_info_values, reader->line, " ", -1);
98
+
99
+ if (strcmp(object_info_values.items[0].string,
100
+ oid_to_hex(&args->oids->oid[i])))
101
+ die(_("object-info: expected OID: %s, got %s"),
102
+ oid_to_hex(&args->oids->oid[i]),
103
+ object_info_values.items[0].string);
104
+
105
+ /*
106
+ * If the response is two elements but the second one is an
107
+ * empty string, that means that the OID is unrecognized by the
108
+ * server.
109
+ */
110
+ if (object_info_values.nr >= 2 &&
111
+ !strcmp(object_info_values.items[1].string, "")) {
112
+ object_info_data[i].unrecognized = 1;
113
+ string_list_clear(&object_info_values, 0);
114
+ continue;
115
+ }
116
+
117
+ /*
118
+ * Because we filter the options to be only the supported by
119
+ * the server we expect the server to answer with the same
120
+ * number of attributes requested.
121
+ */
122
+ if (args->object_info_options->nr + 1 != object_info_values.nr)
123
+ die("object-info: unexpected number of attributes: %s",
124
+ reader->line);
125
+
126
+ if (size_index >= 0 &&
127
+ parse_object_size(object_info_values.items[size_index + 1].string,
128
+ object_info_data[i].sizep))
129
+ die("object-info: ref %s has invalid size %s",
130
+ object_info_values.items[0].string,
131
+ object_info_values.items[size_index + 1].string);
132
+
133
+ string_list_clear(&object_info_values, 0);
134
+ }
135
+ check_stateless_delimiter(stateless_rpc, reader, "stateless delimiter expected");
136
+
137
+ return 0;
138
+}
fetch-object-info.h
new
+22
@@ -0,0 +1,22 @@
1
+#ifndef FETCH_OBJECT_INFO_H
2
+#define FETCH_OBJECT_INFO_H
3
+
4
+#include "pkt-line.h"
5
+#include "protocol.h"
6
+
7
+struct object_info_args {
8
+ struct string_list *object_info_options;
9
+ const struct string_list *server_options;
10
+ struct oid_array *oids;
11
+};
12
+
13
+struct object_info;
14
+/*
15
+ * Sends git-cat-file object-info command into the request buf and read the
16
+ * results from packets.
17
+ */
18
+int fetch_object_info(enum protocol_version version, struct object_info_args *args,
19
+ struct packet_reader *reader, struct object_info *object_info_data,
20
+ int stateless_rpc, int fd_out);
21
+
22
+#endif /* FETCH_OBJECT_INFO_H */
meson.build
+1
@@ -359,6 +359,7 @@ libgit_sources = [
359
'ewah/ewah_rlw.c',
360
'exec-cmd.c',
361
'fetch-negotiator.c',
362
+ 'fetch-object-info.c',
363
'fetch-pack.c',
364
'fmt-merge-msg.c',
365
'fsck.c',
odb.h
+6
@@ -339,6 +339,12 @@ struct object_info {
339
* or multiple times in the same source.
340
*/
341
struct odb_source_info *source_infop;
342
+
343
+ /*
344
+ * object-info protocol specific. Set by the protocol when the remote
345
+ * does not recognize the requested object.
346
+ */
347
+ unsigned int unrecognized:1;
348
};
349
350
/*
transport-helper.c
+10
@@ -784,6 +784,15 @@ static int fetch_refs(struct transport *transport,
784
return -1;
785
}
786
787
+static int fetch_object_info_helper(struct transport *transport)
788
+{
789
+ get_helper(transport);
790
+ if (process_connect(transport, 0))
791
+ return transport->vtable->fetch_object_info(transport);
792
+
793
+ die(_("object-info requires protocol v2"));
794
+}
795
+
796
struct push_update_ref_state {
797
struct ref *hint;
798
struct ref_push_report *report;
@@ -1330,6 +1339,7 @@ static struct transport_vtable vtable = {
1339
.get_refs_list = get_refs_list,
1340
.get_bundle_uri = get_bundle_uri,
1341
.fetch_refs = fetch_refs,
1342
+ .fetch_object_info = fetch_object_info_helper,
1343
.push_refs = push_refs,
1344
.connect = connect_helper,
1345
.disconnect = release_helper
transport-internal.h
+8
@@ -45,6 +45,14 @@ struct transport_vtable {
45
**/
46
int (*fetch_refs)(struct transport *transport, int refs_nr, struct ref **refs);
47
48
+ /*
49
+ * Fetch object info (only size currently) from remote without
50
+ * downloading the objects.
51
+ *
52
+ * Uses object-info capability of v2 protocol.
53
+ */
54
+ int (*fetch_object_info)(struct transport *transport);
55
+
56
/**
57
* Push the objects and refs. Send the necessary objects, and
58
* then, for any refs where peer_ref is set and
transport.c
+45
@@ -9,6 +9,7 @@
9
#include "hook.h"
10
#include "pkt-line.h"
11
#include "fetch-pack.h"
12
+#include "fetch-object-info.h"
13
#include "remote.h"
14
#include "connect.h"
15
#include "send-pack.h"
@@ -432,6 +433,48 @@ static int get_bundle_uri(struct transport *transport)
433
transport->bundles, stateless_rpc);
434
}
435
436
+static int fetch_object_info_via_pack(struct transport *transport)
437
+{
438
+ int ret = 0;
439
+ struct git_transport_data *data = transport->data;
440
+ struct packet_reader reader;
441
+ struct object_info_args args = { 0 };
442
+
443
+ args.server_options = transport->server_options;
444
+ args.oids = transport->smart_options->object_info_oids;
445
+ args.object_info_options = transport->smart_options->object_info_options;
446
+ string_list_sort(args.object_info_options);
447
+
448
+ connect_setup(transport, 0);
449
+ packet_reader_init(&reader, data->fd[0], NULL, 0,
450
+ PACKET_READ_CHOMP_NEWLINE |
451
+ PACKET_READ_GENTLE_ON_EOF |
452
+ PACKET_READ_DIE_ON_ERR_PACKET);
453
+
454
+ data->version = discover_version(&reader);
455
+ transport->hash_algo = reader.hash_algo;
456
+
457
+ ret = fetch_object_info(data->version, &args, &reader,
458
+ data->options.object_info_data,
459
+ transport->stateless_rpc, data->fd[1]);
460
+
461
+ close(data->fd[0]);
462
+ if (data->fd[1] >= 0)
463
+ close(data->fd[1]);
464
+ if (finish_connect(data->conn))
465
+ ret = -1;
466
+ data->conn = NULL;
467
+
468
+ return ret;
469
+}
470
+
471
+int transport_fetch_object_info(struct transport *transport)
472
+{
473
+ if (!transport->vtable->fetch_object_info)
474
+ die(_("remote does not support object-info"));
475
+ return transport->vtable->fetch_object_info(transport);
476
+}
477
+
478
static int fetch_refs_via_pack(struct transport *transport,
479
int nr_heads, struct ref **to_fetch)
480
{
@@ -1004,6 +1047,7 @@ static struct transport_vtable taken_over_vtable = {
1047
.get_refs_list = get_refs_via_connect,
1048
.get_bundle_uri = get_bundle_uri,
1049
.fetch_refs = fetch_refs_via_pack,
1050
+ .fetch_object_info = fetch_object_info_via_pack,
1051
.push_refs = git_transport_push,
1052
.disconnect = disconnect_git
1053
};
@@ -1169,6 +1213,7 @@ static struct transport_vtable builtin_smart_vtable = {
1213
.get_refs_list = get_refs_via_connect,
1214
.get_bundle_uri = get_bundle_uri,
1215
.fetch_refs = fetch_refs_via_pack,
1216
+ .fetch_object_info = fetch_object_info_via_pack,
1217
.push_refs = git_transport_push,
1218
.connect = connect_git,
1219
.disconnect = disconnect_git
transport.h
+9
@@ -55,6 +55,10 @@ struct git_transport_options {
55
* common commits to this oidset instead of fetching any packfiles.
56
*/
57
struct oidset *acked_commits;
58
+
59
+ struct oid_array *object_info_oids;
60
+ struct object_info *object_info_data;
61
+ struct string_list *object_info_options;
62
};
63
64
enum transport_family {
@@ -309,6 +313,11 @@ int transport_get_remote_bundle_uri(struct transport *transport);
313
const struct git_hash_algo *transport_get_hash_algo(struct transport *transport);
314
int transport_fetch_refs(struct transport *transport, struct ref *refs);
315
316
+/*
317
+ * Fetch the object info from remote
318
+ */
319
+int transport_fetch_object_info(struct transport *transport);
320
+
321
/*
322
* If this flag is set, unlocking will avoid to call non-async-signal-safe
323
* functions. This will necessarily leave behind some data structures which