test-tool: add a helper to synthesize large packfiles
To test Git's behavior with very large pack files, we need a way to generate such files quickly. A naive approach using only readily-available Git commands would take over 10 hours for a 4GB pack file, which is prohibitive. Side-stepping Git's machinery and actual zlib compression by writing uncompressed content with the appropriate zlib header makes things much faster. The fastest method using this approach generates many small, unreachable blob objects and takes about 1.5 minutes for 4GB. However, this cannot be used because we need to test git clone, which requires a reachable commit history. Generating many reachable commits with small, uncompressed blobs takes about 4 minutes for 4GB. But this approach 1) does not reproduce the issues we want to fix (which require individual objects larger than 4GB) and 2) is comparatively slow because of the many SHA-1 calculations. The approach taken here generates a single large blob (filled with NUL bytes), along with the trees and commits needed to make it reachable. This takes about 2.5 minutes for 4.5GB, which is the fastest option that produces a valid, clonable repository with an object large enough to trigger the bugs we want to test. Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Johannes Schindelin committed
May 8, 2026 at 08:16 UTC
438886aecb7546e6c5ba68cf1afd7535136f8bb4
6 files changed
+256
Makefile
+1
@@ -872,6 +872,7 @@ TEST_BUILTINS_OBJS += test-submodule-config.o
872
TEST_BUILTINS_OBJS += test-submodule-nested-repo-config.o
873
TEST_BUILTINS_OBJS += test-submodule.o
874
TEST_BUILTINS_OBJS += test-subprocess.o
875
+TEST_BUILTINS_OBJS += test-synthesize.o
876
TEST_BUILTINS_OBJS += test-trace2.o
877
TEST_BUILTINS_OBJS += test-truncate.o
878
TEST_BUILTINS_OBJS += test-userdiff.o
compat/zlib-compat.h
+2
@@ -7,6 +7,8 @@
7
# define z_stream_s zng_stream_s
8
# define gz_header_s zng_gz_header_s
9
10
+# define adler32(adler, buf, len) zng_adler32(adler, buf, len)
11
+
12
# define crc32(crc, buf, len) zng_crc32(crc, buf, len)
13
14
# define inflate(strm, bits) zng_inflate(strm, bits)
t/helper/meson.build
+1
@@ -69,6 +69,7 @@ test_tool_sources = [
69
'test-submodule-nested-repo-config.c',
70
'test-submodule.c',
71
'test-subprocess.c',
72
+ 'test-synthesize.c',
73
'test-tool.c',
74
'test-trace2.c',
75
'test-truncate.c',
t/helper/test-synthesize.c
new
+250
@@ -0,0 +1,250 @@
1
+#define USE_THE_REPOSITORY_VARIABLE
2
+
3
+#include "test-tool.h"
4
+#include "git-compat-util.h"
5
+#include "git-zlib.h"
6
+#include "hash.h"
7
+#include "hex.h"
8
+#include "object-file.h"
9
+#include "object.h"
10
+#include "pack.h"
11
+#include "parse-options.h"
12
+#include "parse.h"
13
+#include "repository.h"
14
+#include "setup.h"
15
+#include "strbuf.h"
16
+#include "write-or-die.h"
17
+
18
+#define BLOCK_SIZE 0xffff
19
+static const unsigned char zeros[BLOCK_SIZE];
20
+
21
+/*
22
+ * Write data as an uncompressed zlib stream.
23
+ * For data larger than 64KB, writes multiple uncompressed blocks.
24
+ * If data is NULL, writes zeros.
25
+ * Updates the pack checksum context.
26
+ */
27
+static void write_uncompressed_zlib(FILE *f, struct git_hash_ctx *pack_ctx,
28
+ const void *data, size_t len,
29
+ const struct git_hash_algo *algo)
30
+{
31
+ unsigned char zlib_header[2] = { 0x78, 0x01 }; /* CMF, FLG */
32
+ unsigned char block_header[5];
33
+ const unsigned char *p = data;
34
+ size_t remaining = len;
35
+ uint32_t adler = 1L; /* adler32 initial value */
36
+ unsigned char adler_buf[4];
37
+
38
+ /* Write zlib header */
39
+ fwrite_or_die(f, zlib_header, sizeof(zlib_header));
40
+ algo->update_fn(pack_ctx, zlib_header, 2);
41
+
42
+ /* Write uncompressed blocks (max 64KB each) */
43
+ do {
44
+ size_t block_len = remaining > BLOCK_SIZE ? BLOCK_SIZE : remaining;
45
+ int is_final = (block_len == remaining);
46
+ const unsigned char *block_data = data ? p : zeros;
47
+
48
+ block_header[0] = is_final ? 0x01 : 0x00;
49
+ block_header[1] = block_len & 0xff;
50
+ block_header[2] = (block_len >> 8) & 0xff;
51
+ block_header[3] = block_header[1] ^ 0xff;
52
+ block_header[4] = block_header[2] ^ 0xff;
53
+
54
+ fwrite_or_die(f, block_header, sizeof(block_header));
55
+ algo->update_fn(pack_ctx, block_header, 5);
56
+
57
+ if (block_len) {
58
+ fwrite_or_die(f, block_data, block_len);
59
+ algo->update_fn(pack_ctx, block_data, block_len);
60
+ adler = adler32(adler, block_data, block_len);
61
+ }
62
+
63
+ if (data)
64
+ p += block_len;
65
+ remaining -= block_len;
66
+ } while (remaining > 0);
67
+
68
+ /* Write adler32 checksum */
69
+ put_be32(adler_buf, adler);
70
+ fwrite_or_die(f, adler_buf, sizeof(adler_buf));
71
+ algo->update_fn(pack_ctx, adler_buf, 4);
72
+}
73
+
74
+/*
75
+ * Write an uncompressed object to the pack file.
76
+ * If `data == NULL`, it is treated like a buffer to NUL bytes.
77
+ * Updates the pack checksum context.
78
+ */
79
+static void write_pack_object(FILE *f, struct git_hash_ctx *pack_ctx,
80
+ enum object_type type,
81
+ const void *data, size_t len,
82
+ struct object_id *oid,
83
+ const struct git_hash_algo *algo)
84
+{
85
+ unsigned char pack_header[MAX_PACK_OBJECT_HEADER];
86
+ char object_header[32];
87
+ int pack_header_len, object_header_len;
88
+ struct git_hash_ctx ctx;
89
+
90
+ /* Write pack object header */
91
+ pack_header_len = encode_in_pack_object_header(pack_header,
92
+ sizeof(pack_header),
93
+ type, len);
94
+ fwrite_or_die(f, pack_header, pack_header_len);
95
+ algo->update_fn(pack_ctx, pack_header, pack_header_len);
96
+
97
+ /* Write the data as uncompressed zlib */
98
+ write_uncompressed_zlib(f, pack_ctx, data, len, algo);
99
+
100
+ algo->init_fn(&ctx);
101
+ object_header_len = format_object_header(object_header,
102
+ sizeof(object_header),
103
+ type, len);
104
+ algo->update_fn(&ctx, object_header, object_header_len);
105
+ if (data)
106
+ algo->update_fn(&ctx, data, len);
107
+ else {
108
+ for (size_t i = len / BLOCK_SIZE; i; i--)
109
+ algo->update_fn(&ctx, zeros, BLOCK_SIZE);
110
+ algo->update_fn(&ctx, zeros, len % BLOCK_SIZE);
111
+ }
112
+ algo->final_oid_fn(oid, &ctx);
113
+}
114
+
115
+/*
116
+ * Generate a pack file with a single large (>4GB) reachable object.
117
+ *
118
+ * Creates:
119
+ * 1. A large blob (all NUL bytes)
120
+ * 2. A tree containing that blob as "file"
121
+ * 3. A commit using that tree
122
+ * 4. The empty tree
123
+ * 5. A child commit using the empty tree
124
+ *
125
+ * This is useful for testing that Git can handle objects larger than 4GB.
126
+ */
127
+static int generate_pack_with_large_object(const char *path, size_t blob_size,
128
+ const struct git_hash_algo *algo)
129
+{
130
+ FILE *f = xfopen(path, "wb");
131
+ struct git_hash_ctx pack_ctx;
132
+ unsigned char pack_hash[GIT_MAX_RAWSZ];
133
+ struct object_id blob_oid, tree_oid, commit_oid, empty_tree_oid, final_commit_oid;
134
+ struct strbuf buf = STRBUF_INIT;
135
+ const uint32_t object_count = 5;
136
+ struct pack_header pack_header = {
137
+ .hdr_signature = htonl(PACK_SIGNATURE),
138
+ .hdr_version = htonl(PACK_VERSION),
139
+ .hdr_entries = htonl(object_count),
140
+ };
141
+
142
+ algo->init_fn(&pack_ctx);
143
+
144
+ /* Write pack header */
145
+ fwrite_or_die(f, &pack_header, sizeof(pack_header));
146
+ algo->update_fn(&pack_ctx, &pack_header, sizeof(pack_header));
147
+
148
+ /* 1. Write the large blob */
149
+ write_pack_object(f, &pack_ctx, OBJ_BLOB, NULL, blob_size, &blob_oid, algo);
150
+
151
+ /* 2. Write tree containing the blob as "file" */
152
+ strbuf_addf(&buf, "100644 file%c", '\0');
153
+ strbuf_add(&buf, blob_oid.hash, algo->rawsz);
154
+ write_pack_object(f, &pack_ctx, OBJ_TREE, buf.buf, buf.len, &tree_oid, algo);
155
+
156
+ /* 3. Write commit using that tree */
157
+ strbuf_reset(&buf);
158
+ strbuf_addf(&buf,
159
+ "tree %s\n"
160
+ "author A U Thor <author@example.com> 1234567890 +0000\n"
161
+ "committer C O Mitter <committer@example.com> 1234567890 +0000\n"
162
+ "\n"
163
+ "Large blob commit\n",
164
+ oid_to_hex(&tree_oid));
165
+ write_pack_object(f, &pack_ctx, OBJ_COMMIT, buf.buf, buf.len, &commit_oid, algo);
166
+
167
+ /* 4. Write the empty tree */
168
+ write_pack_object(f, &pack_ctx, OBJ_TREE, "", 0, &empty_tree_oid, algo);
169
+
170
+ /* 5. Write final commit using empty tree, with previous commit as parent */
171
+ strbuf_reset(&buf);
172
+ strbuf_addf(&buf,
173
+ "tree %s\n"
174
+ "parent %s\n"
175
+ "author A U Thor <author@example.com> 1234567890 +0000\n"
176
+ "committer C O Mitter <committer@example.com> 1234567890 +0000\n"
177
+ "\n"
178
+ "Empty tree commit\n",
179
+ oid_to_hex(&empty_tree_oid),
180
+ oid_to_hex(&commit_oid));
181
+ write_pack_object(f, &pack_ctx, OBJ_COMMIT, buf.buf, buf.len, &final_commit_oid, algo);
182
+
183
+ /* Write pack trailer (checksum) */
184
+ algo->final_fn(pack_hash, &pack_ctx);
185
+ fwrite_or_die(f, pack_hash, algo->rawsz);
186
+ if (fclose(f))
187
+ die_errno(_("could not close '%s'"), path);
188
+
189
+ strbuf_release(&buf);
190
+
191
+ /* Print the final commit OID so caller can set up refs */
192
+ printf("%s\n", oid_to_hex(&final_commit_oid));
193
+
194
+ return 0;
195
+}
196
+
197
+static int cmd__synthesize__pack(int argc, const char **argv,
198
+ const char *prefix UNUSED,
199
+ struct repository *repo)
200
+{
201
+ int non_git;
202
+ int reachable_large = 0;
203
+ const struct git_hash_algo *algo;
204
+ size_t blob_size;
205
+ uintmax_t blob_size_u;
206
+ const char *path;
207
+ const char * const usage[] = {
208
+ "test-tool synthesize pack "
209
+ "--reachable-large <blob-size> <filename>",
210
+ NULL
211
+ };
212
+ struct option options[] = {
213
+ OPT_BOOL(0, "reachable-large", &reachable_large,
214
+ N_("write a pack with a single reachable large blob")),
215
+ OPT_END()
216
+ };
217
+
218
+ setup_git_directory_gently(&non_git);
219
+ repo = the_repository;
220
+ algo = repo->hash_algo;
221
+
222
+ argc = parse_options(argc, argv, NULL, options, usage,
223
+ PARSE_OPT_KEEP_ARGV0);
224
+ if (argc != 3 || !reachable_large)
225
+ usage_with_options(usage, options);
226
+
227
+ if (!git_parse_unsigned(argv[1], &blob_size_u,
228
+ maximum_unsigned_value_of_type(size_t)))
229
+ die(_("'%s' is not a valid blob size"), argv[1]);
230
+ blob_size = blob_size_u;
231
+ path = argv[2];
232
+
233
+ return !!generate_pack_with_large_object(path, blob_size, algo);
234
+}
235
+
236
+int cmd__synthesize(int argc, const char **argv)
237
+{
238
+ const char *prefix = NULL;
239
+ char const * const synthesize_usage[] = {
240
+ "test-tool synthesize pack <options>",
241
+ NULL,
242
+ };
243
+ parse_opt_subcommand_fn *fn = NULL;
244
+ struct option options[] = {
245
+ OPT_SUBCOMMAND("pack", &fn, cmd__synthesize__pack),
246
+ OPT_END()
247
+ };
248
+ argc = parse_options(argc, argv, prefix, options, synthesize_usage, 0);
249
+ return !!fn(argc, argv, prefix, NULL);
250
+}
t/helper/test-tool.c
+1
@@ -82,6 +82,7 @@ static struct test_cmd cmds[] = {
82
{ "submodule-config", cmd__submodule_config },
83
{ "submodule-nested-repo-config", cmd__submodule_nested_repo_config },
84
{ "subprocess", cmd__subprocess },
85
+ { "synthesize", cmd__synthesize },
86
{ "trace2", cmd__trace2 },
87
{ "truncate", cmd__truncate },
88
{ "userdiff", cmd__userdiff },
t/helper/test-tool.h
+1
@@ -75,6 +75,7 @@ int cmd__submodule(int argc, const char **argv);
75
int cmd__submodule_config(int argc, const char **argv);
76
int cmd__submodule_nested_repo_config(int argc, const char **argv);
77
int cmd__subprocess(int argc, const char **argv);
78
+int cmd__synthesize(int argc, const char **argv);
79
int cmd__trace2(int argc, const char **argv);
80
int cmd__truncate(int argc, const char **argv);
81
int cmd__userdiff(int argc, const char **argv);