git/list[1] front-page[2] threads[3] people[4] search[5] about
 

[PATCH v2 12/16] pack-objects: prepare SHA-1 table in v4

From
Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Date
Sep 9, 2013, 13:58 UTC
Message-ID
<1378735087-4813-13-git-send-email-pclouds@gmail.com>
In-Reply-To
<1378735087-4813-1-git-send-email-pclouds@gmail.com>

SHA-1 table is trickier than ident or path tables because it must contains exactly the number entries in pack. In the thin pack case it must also cover bases that will be appended by index-pack.

The problem is not all preferred_base entries end up becoming actually needed. So we do a fake write_one() round just to get what is written and what is not. It also helps the case when the multiple packs are written.

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
---
 builtin/pack-objects.c | 55 +++++++++++++++++++++++++++++++++++++++++++++++---
 1 file changed, 52 insertions(+), 3 deletions(-)
diff --git a/builtin/pack-objects.c b/builtin/pack-objects.c
index 60ea5a7..055b59d 100644
--- a/builtin/pack-objects.c
+++ b/builtin/pack-objects.c
@@ -434,7 +434,7 @@ static unsigned long write_object(struct sha1file *f,
 	unsigned long limit, len;
 	int usable_delta, to_reuse;
 
-	if (!pack_to_stdout)
+	if (f && !pack_to_stdout)
 		crc32_begin(f);
 
 	/* apply size limit if limited packsize and not first object */
@@ -477,6 +477,12 @@ static unsigned long write_object(struct sha1file *f,
 				 * and we do not need to deltify it.
 				 */
 
+	if (!f) {
+		if (usable_delta && entry->delta->idx.offset < 2)
+			entry->delta->idx.offset = 2;
+		return 2;
+	}
+
 	if (!to_reuse)
 		len = write_no_reuse_object(f, entry, limit, usable_delta);
 	else
@@ -543,10 +549,14 @@ static enum write_one_status write_one(struct sha1file *f,
 		e->idx.offset = recursing;
 		return WRITE_ONE_BREAK;
 	}
+	if (!f) {
+		*offset += size;
+		return WRITE_ONE_WRITTEN;
+	}
 	written_list[nr_written++] = &e->idx;
 
 	/* make sure off_t is sufficiently large not to wrap */
-	if (signed_add_overflows(*offset, size))
+	if (f && signed_add_overflows(*offset, size))
 		die("pack too large for current definition of off_t");
 	*offset += size;
 	return WRITE_ONE_WRITTEN;
@@ -716,6 +726,39 @@ static struct object_entry **compute_write_order(void)
 	return wo;
 }
 
+static int sha1_idx_sort(const void *a_, const void *b_)
+{
+	const struct pack_idx_entry *a = a_;
+	const struct pack_idx_entry *b = b_;
+	return hashcmp(a->sha1, b->sha1);
+}
+
+/*
+ * Do a fake writting round to detemine what's in the SHA-1 table.
+ */
+static void prepare_sha1_table(uint32_t start, struct object_entry **write_order)
+{
+	int i = start;
+	off_t fake_offset = 2;
+	for (; i < nr_objects; i++) {
+		struct object_entry *e = write_order[i];
+		if (write_one(NULL, e, &fake_offset) == WRITE_ONE_BREAK)
+			break;
+	}
+
+	v4.all_objs_nr = 0;
+	for (i = 0; i < nr_objects; i++) {
+		struct object_entry *e = write_order[i];
+		if (e->idx.offset > 0) {
+			v4.all_objs[v4.all_objs_nr++] = e->idx;
+			fprintf(stderr, "%s in\n", sha1_to_hex(e->idx.sha1));
+			e->idx.offset = 0;
+		}
+	}
+	qsort(v4.all_objs, v4.all_objs_nr, sizeof(*v4.all_objs),
+	      sha1_idx_sort);
+}
+
 static void write_pack_file(void)
 {
 	uint32_t i = 0, j;
@@ -739,7 +782,12 @@ static void write_pack_file(void)
 		else
 			f = create_tmp_packfile(&pack_tmp_name);
 
-		offset = write_pack_header(f, pack_version, nr_remaining);
+		if (pack_version == 4)
+			prepare_sha1_table(i, write_order);
+
+		offset = write_pack_header(f, pack_version,
+					   pack_version < 4 ? nr_remaining : v4.all_objs_nr);
+
 		if (!offset)
 			die_errno("unable to write pack header");
 		nr_written = 0;
@@ -2058,6 +2106,7 @@ static void prepare_pack(int window, int depth)
 	if (pack_version == 4) {
 		sort_dict_entries_by_hits(v4.commit_ident_table);
 		sort_dict_entries_by_hits(v4.tree_path_table);
+		v4.all_objs = xmalloc(nr_objects * sizeof(*v4.all_objs));
 	}
 
 	get_object_details();
-- 
1.8.2.83.gc99314b
Previous: Nguyễn Thái Ngọc DuyNext: Nguyễn Thái Ngọc Duy
Message 111 of 124 in “pack version 4 basic functionalities”
  1. 00/38 pack version 4 basic functionalitiesNicolas Pitre, Sep 5, 2013
  2. 01/38 pack v4: initial pack dictionary structure and codeNicolas Pitre, Sep 5, 2013
  3. 02/38 export packed_object_info()Nicolas Pitre, Sep 5, 2013
  4. 03/38 pack v4: scan tree objectsNicolas Pitre, Sep 5, 2013
  5. 04/38 pack v4: add tree entry mode support to dictionary entriesNicolas Pitre, Sep 5, 2013
  6. 05/38 pack v4: add commit object parsingNicolas Pitre, Sep 5, 2013
  7. SZEDER GáborSep 5, 2013
  8. Nicolas PitreSep 5, 2013
  9. 06/38 pack v4: split the object list and dictionary creationNicolas Pitre, Sep 5, 2013
  10. 07/38 pack v4: move to struct pack_idx_entry and get rid of our own struct idx_entryNicolas Pitre, Sep 5, 2013
  11. 08/38 pack v4: basic SHA1 reference encodingNicolas Pitre, Sep 5, 2013
  12. 09/38 introduce get_sha1_lowhex()Nicolas Pitre, Sep 5, 2013
  13. 10/38 pack v4: commit object encodingNicolas Pitre, Sep 5, 2013
  14. Junio C HamanoSep 6, 2013
  15. Nicolas PitreSep 6, 2013
  16. Junio C HamanoSep 6, 2013
  17. Nicolas PitreSep 7, 2013
  18. 11/38 pack v4: tree object encodingNicolas Pitre, Sep 5, 2013
  19. 12/38 pack v4: dictionary table outputNicolas Pitre, Sep 5, 2013
  20. 13/38 pack v4: creation codeNicolas Pitre, Sep 5, 2013
  21. 14/38 pack v4: object headersNicolas Pitre, Sep 5, 2013
  22. 15/38 pack v4: object data copyNicolas Pitre, Sep 5, 2013
  23. 16/38 pack v4: object writingNicolas Pitre, Sep 5, 2013
  24. 17/38 pack v4: tree object delta encodingNicolas Pitre, Sep 5, 2013
  25. 18/38 pack v4: load delta candidate for encoding tree objectsNicolas Pitre, Sep 5, 2013
  26. 19/38 packv4-create: optimize delta encodingNicolas Pitre, Sep 5, 2013
  27. 20/38 pack v4: honor pack.compression config optionNicolas Pitre, Sep 5, 2013
  28. 21/38 pack v4: relax commit parsing a bitNicolas Pitre, Sep 5, 2013
  29. 22/38 pack index v3Nicolas Pitre, Sep 5, 2013
  30. 23/38 packv4-create: normalize pack name to properly generate the pack index file nameNicolas Pitre, Sep 5, 2013
  31. 24/38 packv4-create: add progress displayNicolas Pitre, Sep 5, 2013
  32. 25/38 pack v4: initial pack index v3 support on the read sideNicolas Pitre, Sep 5, 2013
  33. 26/38 pack v4: object header decodeNicolas Pitre, Sep 5, 2013
  34. 27/38 pack v4: code to obtain a SHA1 from a sha1refNicolas Pitre, Sep 5, 2013
  35. 28/38 pack v4: code to load and prepare a pack dictionary table for useNicolas Pitre, Sep 5, 2013
  36. 29/38 pack v4: code to retrieve a nameNicolas Pitre, Sep 5, 2013
  37. 30/38 pack v4: code to recreate a canonical commit objectNicolas Pitre, Sep 5, 2013
  38. 31/38 sha1_file.c: make use of decode_varint()Nicolas Pitre, Sep 5, 2013
  39. SZEDER GáborSep 5, 2013
  40. 32/38 pack v4: parse delta base referenceNicolas Pitre, Sep 5, 2013
  41. 33/38 pack v4: we can read commit objects nowNicolas Pitre, Sep 5, 2013
  42. 34/38 pack v4: code to retrieve a path componentNicolas Pitre, Sep 5, 2013
  43. 35/38 pack v4: decode tree objectsNicolas Pitre, Sep 5, 2013
  44. 36/38 pack v4: get tree objectsNicolas Pitre, Sep 5, 2013
  45. 37/38 pack v4: introduce "escape hatches" in the name and path indexesNicolas Pitre, Sep 5, 2013
  46. Nicolas PitreSep 5, 2013
  47. Nicolas PitreSep 5, 2013
  48. Duy NguyenSep 5, 2013
  49. 38/38 packv4-create: add a command line argument to limit tree copy sequencesNicolas Pitre, Sep 5, 2013
  50. 00/12 pack v4 support in index-packNguyễn Thái Ngọc Duy, Sep 7, 2013
  51. 01/12 pack v4: split pv4_create_dict() out of load_dict()Nguyễn Thái Ngọc Duy, Sep 7, 2013
  52. 02/12 index-pack: split out varint decoding codeNguyễn Thái Ngọc Duy, Sep 7, 2013
  53. 03/12 index-pack: do not allocate buffer for unpacking deltas in the first passNguyễn Thái Ngọc Duy, Sep 7, 2013
  54. 04/12 index-pack: split inflate/digest code out of unpack_entry_dataNguyễn Thái Ngọc Duy, Sep 7, 2013
  55. 05/12 index-pack: parse v4 header and dictionariesNguyễn Thái Ngọc Duy, Sep 7, 2013
  56. Nicolas PitreSep 8, 2013
  57. 06/12 index-pack: make sure all objects are registered in v4's SHA-1 tableNguyễn Thái Ngọc Duy, Sep 7, 2013
  58. 07/12 index-pack: parse v4 commit formatNguyễn Thái Ngọc Duy, Sep 7, 2013
  59. 08/12 index-pack: parse v4 tree formatNguyễn Thái Ngọc Duy, Sep 7, 2013
  60. Nicolas PitreSep 8, 2013
  61. 09/12 index-pack: move delta base queuing code to unpack_raw_entryNguyễn Thái Ngọc Duy, Sep 7, 2013
  62. 10/12 index-pack: record all delta bases in v4 (tree and ref-delta)Nguyễn Thái Ngọc Duy, Sep 7, 2013
  63. 11/12 index-pack: skip looking for ofs-deltas in v4 as they are not allowedNguyễn Thái Ngọc Duy, Sep 7, 2013
  64. 12/12 index-pack: resolve v4 one-base treesNguyễn Thái Ngọc Duy, Sep 7, 2013
  65. Nicolas PitreSep 8, 2013
  66. Duy NguyenSep 8, 2013
  67. 00/14 pack v4 support in index-packNguyễn Thái Ngọc Duy, Sep 8, 2013
  68. 01/14 pack v4: split pv4_create_dict() out of load_dict()Nguyễn Thái Ngọc Duy, Sep 8, 2013
  69. 02/14 pack v4: add pv4_free_dict()Nguyễn Thái Ngọc Duy, Sep 8, 2013
  70. 03/14 index-pack: add more comments on some big functionsNguyễn Thái Ngọc Duy, Sep 8, 2013
  71. 04/14 index-pack: split out varint decoding codeNguyễn Thái Ngọc Duy, Sep 8, 2013
  72. 05/14 index-pack: do not allocate buffer for unpacking deltas in the first passNguyễn Thái Ngọc Duy, Sep 8, 2013
  73. 06/14 index-pack: split inflate/digest code out of unpack_entry_dataNguyễn Thái Ngọc Duy, Sep 8, 2013
  74. 07/14 index-pack: parse v4 header and dictionariesNguyễn Thái Ngọc Duy, Sep 8, 2013
  75. 08/14 index-pack: make sure all objects are registered in v4's SHA-1 tableNguyễn Thái Ngọc Duy, Sep 8, 2013
  76. 09/14 index-pack: parse v4 commit formatNguyễn Thái Ngọc Duy, Sep 8, 2013
  77. 10/14 index-pack: parse v4 tree formatNguyễn Thái Ngọc Duy, Sep 8, 2013
  78. 11/14 index-pack: move delta base queuing code to unpack_raw_entryNguyễn Thái Ngọc Duy, Sep 8, 2013
  79. 12/14 index-pack: record all delta bases in v4 (tree and ref-delta)Nguyễn Thái Ngọc Duy, Sep 8, 2013
  80. 13/14 index-pack: skip looking for ofs-deltas in v4 as they are not allowedNguyễn Thái Ngọc Duy, Sep 8, 2013
  81. 14/14 index-pack: resolve v4 one-base treesNguyễn Thái Ngọc Duy, Sep 8, 2013
  82. 00/11 pack v4 support in pack-objectsNguyễn Thái Ngọc Duy, Sep 8, 2013
  83. 01/11 pack v4: allocate dicts from the beginningNguyễn Thái Ngọc Duy, Sep 8, 2013
  84. 02/11 pack v4: stop using static/global variables in packv4-create.cNguyễn Thái Ngọc Duy, Sep 8, 2013
  85. 03/11 pack v4: move packv4-create.c to libgit.aNguyễn Thái Ngọc Duy, Sep 8, 2013
  86. Nicolas PitreSep 8, 2013
  87. 04/11 pack v4: add version argument to write_pack_headerNguyễn Thái Ngọc Duy, Sep 8, 2013
  88. 05/11 pack-write.c: add pv4_encode_in_pack_object_headerNguyễn Thái Ngọc Duy, Sep 8, 2013
  89. Nicolas PitreSep 8, 2013
  90. 06/11 pack-objects: add --version to specify written pack versionNguyễn Thái Ngọc Duy, Sep 8, 2013
  91. 07/11 list-objects.c: add show_tree_entry callback to traverse_commit_listNguyễn Thái Ngọc Duy, Sep 8, 2013
  92. 08/11 pack-objects: create pack v4 tablesNguyễn Thái Ngọc Duy, Sep 8, 2013
  93. Duy NguyenSep 9, 2013
  94. Nicolas PitreSep 9, 2013
  95. Junio C HamanoSep 9, 2013
  96. 09/11 pack-objects: do not cache delta for v4 treesNguyễn Thái Ngọc Duy, Sep 8, 2013
  97. 10/11 pack-objects: exclude commits out of delta objects in v4Nguyễn Thái Ngọc Duy, Sep 8, 2013
  98. 11/11 pack-objects: support writing pack v4Nguyễn Thái Ngọc Duy, Sep 8, 2013
  99. 00/16 pack v4 support in pack-objectsNguyễn Thái Ngọc Duy, Sep 9, 2013
  100. 01/16 pack v4: allocate dicts from the beginningNguyễn Thái Ngọc Duy, Sep 9, 2013
  101. 02/16 pack v4: stop using static/global variables in packv4-create.cNguyễn Thái Ngọc Duy, Sep 9, 2013
  102. 03/16 pack v4: move packv4-create.c to libgit.aNguyễn Thái Ngọc Duy, Sep 9, 2013
  103. 04/16 pack v4: add version argument to write_pack_headerNguyễn Thái Ngọc Duy, Sep 9, 2013
  104. 05/16 pack_write: tighten valid object type check in encode_in_pack_object_headerNguyễn Thái Ngọc Duy, Sep 9, 2013
  105. 06/16 pack-write.c: add pv4_encode_object_headerNguyễn Thái Ngọc Duy, Sep 9, 2013
  106. 07/16 pack-objects: add --version to specify written pack versionNguyễn Thái Ngọc Duy, Sep 9, 2013
  107. 08/16 list-objects.c: add show_tree_entry callback to traverse_commit_listNguyễn Thái Ngọc Duy, Sep 9, 2013
  108. 09/16 pack-objects: do not cache delta for v4 treesNguyễn Thái Ngọc Duy, Sep 9, 2013
  109. 10/16 pack-objects: exclude commits out of delta objects in v4Nguyễn Thái Ngọc Duy, Sep 9, 2013
  110. 11/16 pack-objects: create pack v4 tablesNguyễn Thái Ngọc Duy, Sep 9, 2013
  111. 12/16 pack-objects: prepare SHA-1 table in v4Nguyễn Thái Ngọc Duy, Sep 9, 2013
  112. 13/16 pack-objects: support writing pack v4Nguyễn Thái Ngọc Duy, Sep 9, 2013
  113. 14/16 pack v4: support "end-of-pack" indicator in index-pack and pack-objectsNguyễn Thái Ngọc Duy, Sep 9, 2013
  114. 15/16 index-pack: use nr_objects_final as sha1_table sizeNguyễn Thái Ngọc Duy, Sep 9, 2013
  115. Nicolas PitreSep 9, 2013
  116. Junio C HamanoSep 9, 2013
  117. Nicolas PitreSep 9, 2013
  118. Junio C HamanoSep 9, 2013
  119. Nicolas PitreSep 9, 2013
  120. Junio C HamanoSep 9, 2013
  121. Nicolas PitreSep 9, 2013
  122. Duy NguyenSep 10, 2013
  123. Nicolas PitreSep 12, 2013
  124. 16/16 index-pack: support completing thin packs v4Nguyễn Thái Ngọc Duy, Sep 9, 2013

Read the whole thread, see it on lore, or plain text.

$ cat FOOTERMessages come from the public archive at lore.kernel.org/git, fetched every hour. The front page is chosen and written each morning by an AI editor and can be wrong; the threads themselves are the record. About and API. For agents: an MCP server at https://gitlist.dev/mcp, and any thread, story or person page as Markdown by adding .md to its URL (or sending Accept: text/markdown). Details in /llms.txt.