Skip to content

Commit 4fd1359

Browse files
committed
Merge branch 'jk/pack-revindex'
In-core storage of the reverse index for .pack files (which lets you go from a pack offset to an object name) has been streamlined. * jk/pack-revindex: pack-revindex: store entries directly in packed_git pack-revindex: drop hash table
2 parents b4e8e0e + 9d98bbf commit 4fd1359

File tree

4 files changed

+35
-90
lines changed

4 files changed

+35
-90
lines changed

cache.h

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -9,6 +9,7 @@
99
#include "convert.h"
1010
#include "trace.h"
1111
#include "string-list.h"
12+
#include "pack-revindex.h"
1213

1314
#include SHA1_HEADER
1415
#ifndef platform_SHA_CTX
@@ -1299,6 +1300,7 @@ extern struct packed_git {
12991300
freshened:1,
13001301
do_not_close:1;
13011302
unsigned char sha1[20];
1303+
struct revindex_entry *revindex;
13021304
/* something like ".git/objects/pack/xxxxx.pack" */
13031305
char pack_name[FLEX_ARRAY]; /* more */
13041306
} *packed_git;

pack-bitmap.c

Lines changed: 5 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -33,9 +33,6 @@ static struct bitmap_index {
3333
/* Packfile to which this bitmap index belongs to */
3434
struct packed_git *pack;
3535

36-
/* reverse index for the packfile */
37-
struct pack_revindex *reverse_index;
38-
3936
/*
4037
* Mark the first `reuse_objects` in the packfile as reused:
4138
* they will be sent as-is without using them for repacking
@@ -308,7 +305,7 @@ static int load_pack_bitmap(void)
308305

309306
bitmap_git.bitmaps = kh_init_sha1();
310307
bitmap_git.ext_index.positions = kh_init_sha1_pos();
311-
bitmap_git.reverse_index = revindex_for_pack(bitmap_git.pack);
308+
load_pack_revindex(bitmap_git.pack);
312309

313310
if (!(bitmap_git.commits = read_bitmap_1(&bitmap_git)) ||
314311
!(bitmap_git.trees = read_bitmap_1(&bitmap_git)) ||
@@ -380,7 +377,7 @@ static inline int bitmap_position_packfile(const unsigned char *sha1)
380377
if (!offset)
381378
return -1;
382379

383-
return find_revindex_position(bitmap_git.reverse_index, offset);
380+
return find_revindex_position(bitmap_git.pack, offset);
384381
}
385382

386383
static int bitmap_position(const unsigned char *sha1)
@@ -630,7 +627,7 @@ static void show_objects_for_type(
630627
if (pos + offset < bitmap_git.reuse_objects)
631628
continue;
632629

633-
entry = &bitmap_git.reverse_index->revindex[pos + offset];
630+
entry = &bitmap_git.pack->revindex[pos + offset];
634631
sha1 = nth_packed_object_sha1(bitmap_git.pack, entry->nr);
635632

636633
if (bitmap_git.hashes)
@@ -804,7 +801,7 @@ int reuse_partial_packfile_from_bitmap(struct packed_git **packfile,
804801
return -1;
805802

806803
bitmap_git.reuse_objects = *entries = reuse_objects;
807-
*up_to = bitmap_git.reverse_index->revindex[reuse_objects].offset;
804+
*up_to = bitmap_git.pack->revindex[reuse_objects].offset;
808805
*packfile = bitmap_git.pack;
809806

810807
return 0;
@@ -1038,7 +1035,7 @@ int rebuild_existing_bitmaps(struct packing_data *mapping,
10381035
struct revindex_entry *entry;
10391036
struct object_entry *oe;
10401037

1041-
entry = &bitmap_git.reverse_index->revindex[i];
1038+
entry = &bitmap_git.pack->revindex[i];
10421039
sha1 = nth_packed_object_sha1(bitmap_git.pack, entry->nr);
10431040
oe = packlist_find(mapping, sha1, NULL);
10441041

pack-revindex.c

Lines changed: 24 additions & 75 deletions
Original file line numberDiff line numberDiff line change
@@ -8,52 +8,13 @@
88
* size is easily available by examining the pack entry header). It is
99
* also rather expensive to find the sha1 for an object given its offset.
1010
*
11-
* We build a hashtable of existing packs (pack_revindex), and keep reverse
12-
* index here -- pack index file is sorted by object name mapping to offset;
13-
* this pack_revindex[].revindex array is a list of offset/index_nr pairs
11+
* The pack index file is sorted by object name mapping to offset;
12+
* this revindex array is a list of offset/index_nr pairs
1413
* ordered by offset, so if you know the offset of an object, next offset
1514
* is where its packed representation ends and the index_nr can be used to
1615
* get the object sha1 from the main index.
1716
*/
1817

19-
static struct pack_revindex *pack_revindex;
20-
static int pack_revindex_hashsz;
21-
22-
static int pack_revindex_ix(struct packed_git *p)
23-
{
24-
unsigned long ui = (unsigned long)(intptr_t)p;
25-
int i;
26-
27-
ui = ui ^ (ui >> 16); /* defeat structure alignment */
28-
i = (int)(ui % pack_revindex_hashsz);
29-
while (pack_revindex[i].p) {
30-
if (pack_revindex[i].p == p)
31-
return i;
32-
if (++i == pack_revindex_hashsz)
33-
i = 0;
34-
}
35-
return -1 - i;
36-
}
37-
38-
static void init_pack_revindex(void)
39-
{
40-
int num;
41-
struct packed_git *p;
42-
43-
for (num = 0, p = packed_git; p; p = p->next)
44-
num++;
45-
if (!num)
46-
return;
47-
pack_revindex_hashsz = num * 11;
48-
pack_revindex = xcalloc(pack_revindex_hashsz, sizeof(*pack_revindex));
49-
for (p = packed_git; p; p = p->next) {
50-
num = pack_revindex_ix(p);
51-
num = - 1 - num;
52-
pack_revindex[num].p = p;
53-
}
54-
/* revindex elements are lazily initialized */
55-
}
56-
5718
/*
5819
* This is a least-significant-digit radix sort.
5920
*
@@ -154,14 +115,13 @@ static void sort_revindex(struct revindex_entry *entries, unsigned n, off_t max)
154115
/*
155116
* Ordered list of offsets of objects in the pack.
156117
*/
157-
static void create_pack_revindex(struct pack_revindex *rix)
118+
static void create_pack_revindex(struct packed_git *p)
158119
{
159-
struct packed_git *p = rix->p;
160120
unsigned num_ent = p->num_objects;
161121
unsigned i;
162122
const char *index = p->index_data;
163123

164-
rix->revindex = xmalloc(sizeof(*rix->revindex) * (num_ent + 1));
124+
p->revindex = xmalloc(sizeof(*p->revindex) * (num_ent + 1));
165125
index += 4 * 256;
166126

167127
if (p->index_version > 1) {
@@ -171,55 +131,42 @@ static void create_pack_revindex(struct pack_revindex *rix)
171131
for (i = 0; i < num_ent; i++) {
172132
uint32_t off = ntohl(*off_32++);
173133
if (!(off & 0x80000000)) {
174-
rix->revindex[i].offset = off;
134+
p->revindex[i].offset = off;
175135
} else {
176-
rix->revindex[i].offset =
136+
p->revindex[i].offset =
177137
((uint64_t)ntohl(*off_64++)) << 32;
178-
rix->revindex[i].offset |=
138+
p->revindex[i].offset |=
179139
ntohl(*off_64++);
180140
}
181-
rix->revindex[i].nr = i;
141+
p->revindex[i].nr = i;
182142
}
183143
} else {
184144
for (i = 0; i < num_ent; i++) {
185145
uint32_t hl = *((uint32_t *)(index + 24 * i));
186-
rix->revindex[i].offset = ntohl(hl);
187-
rix->revindex[i].nr = i;
146+
p->revindex[i].offset = ntohl(hl);
147+
p->revindex[i].nr = i;
188148
}
189149
}
190150

191151
/* This knows the pack format -- the 20-byte trailer
192152
* follows immediately after the last object data.
193153
*/
194-
rix->revindex[num_ent].offset = p->pack_size - 20;
195-
rix->revindex[num_ent].nr = -1;
196-
sort_revindex(rix->revindex, num_ent, p->pack_size);
154+
p->revindex[num_ent].offset = p->pack_size - 20;
155+
p->revindex[num_ent].nr = -1;
156+
sort_revindex(p->revindex, num_ent, p->pack_size);
197157
}
198158

199-
struct pack_revindex *revindex_for_pack(struct packed_git *p)
159+
void load_pack_revindex(struct packed_git *p)
200160
{
201-
int num;
202-
struct pack_revindex *rix;
203-
204-
if (!pack_revindex_hashsz)
205-
init_pack_revindex();
206-
207-
num = pack_revindex_ix(p);
208-
if (num < 0)
209-
die("internal error: pack revindex fubar");
210-
211-
rix = &pack_revindex[num];
212-
if (!rix->revindex)
213-
create_pack_revindex(rix);
214-
215-
return rix;
161+
if (!p->revindex)
162+
create_pack_revindex(p);
216163
}
217164

218-
int find_revindex_position(struct pack_revindex *pridx, off_t ofs)
165+
int find_revindex_position(struct packed_git *p, off_t ofs)
219166
{
220167
int lo = 0;
221-
int hi = pridx->p->num_objects + 1;
222-
struct revindex_entry *revindex = pridx->revindex;
168+
int hi = p->num_objects + 1;
169+
struct revindex_entry *revindex = p->revindex;
223170

224171
do {
225172
unsigned mi = lo + (hi - lo) / 2;
@@ -237,11 +184,13 @@ int find_revindex_position(struct pack_revindex *pridx, off_t ofs)
237184

238185
struct revindex_entry *find_pack_revindex(struct packed_git *p, off_t ofs)
239186
{
240-
struct pack_revindex *pridx = revindex_for_pack(p);
241-
int pos = find_revindex_position(pridx, ofs);
187+
int pos;
188+
189+
load_pack_revindex(p);
190+
pos = find_revindex_position(p, ofs);
242191

243192
if (pos < 0)
244193
return NULL;
245194

246-
return pridx->revindex + pos;
195+
return p->revindex + pos;
247196
}

pack-revindex.h

Lines changed: 4 additions & 7 deletions
Original file line numberDiff line numberDiff line change
@@ -1,18 +1,15 @@
11
#ifndef PACK_REVINDEX_H
22
#define PACK_REVINDEX_H
33

4+
struct packed_git;
5+
46
struct revindex_entry {
57
off_t offset;
68
unsigned int nr;
79
};
810

9-
struct pack_revindex {
10-
struct packed_git *p;
11-
struct revindex_entry *revindex;
12-
};
13-
14-
struct pack_revindex *revindex_for_pack(struct packed_git *p);
15-
int find_revindex_position(struct pack_revindex *pridx, off_t ofs);
11+
void load_pack_revindex(struct packed_git *p);
12+
int find_revindex_position(struct packed_git *p, off_t ofs);
1613

1714
struct revindex_entry *find_pack_revindex(struct packed_git *p, off_t ofs);
1815

0 commit comments

Comments
 (0)