-
Notifications
You must be signed in to change notification settings - Fork 10
/
Copy pathdocmap.c
executable file
·2491 lines (2221 loc) · 89.9 KB
/
docmap.c
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
/* docmap.c implements the interface declared in docmap.h. The docmap holds
* all per-document information in the search engine. As of the below date,
* the design has been changed and a new implementation written (also renamed
* ndocmap back to docmap).
*
* The new design is essentially a paged array. The docmap is split into pages
* of a given size, with each page consisting of a small header, and then some
* data. Pages may be either data or cache pages, where data pages hold the
* 'true' docmap record, and cache pages are used to load data into memory
* quickly between invocations.
* Each entry is serially encoded into a data page
* utilising simple inter- and intra-entry compression (so that
* they must be read as a stream). Specifically, all integer entries are
* vbyte-encoded, and all text entries associated with each entry are
* front-coded with respect to the previous entry.
* Unlike the previous design, this docmap holds all elements of an entry
* contiguously on disk.
*
* The docmap then has an append buffer, which holds new documents in main
* memory prior to them migrating to disk, and a read buffer.
* Both are an integral number of pages.
* In addition to these buffers, the user can request that certain quantities
* be cached in main memory, where that quantity is individually held as
* an array. Cache pages are read/written at the load/save of the docmap to
* move cached items in/out of memory quickly.
*
* The data pages are indexed by an in-memory sorted array, mapping the first
* docno contained within each page to its location. This information and
* other aggregate information is written to the end of the docmap on shutdown,
* in cache pages.
*
* The format of each data page is:
* - the constant byte '0xda', or '0xdf' for the final data page
* - number of entries in this page (4 byte big-endian integer)
* - for each entry:
* - fileno gap + 1 or 0 (vbyte). 0 indicates that this document
* immediately follows the previous, and no offset field is present
* - (optional, see above) offset field (vbyte)
* - docno gap (since docnos strictly ascend - vbyte)
* - flags (shifted onto docno gap vbyte)
* - distinct words in document (vbyte)
* - words - distinct_words (since words >= distinct_words, vbyte)
* - bytes - (2 * words - 1) (since each word must occupy at least two
* bytes, except the first. vbyte)
* - mime type of document
* - TREC docno (front-coded):
* - number of prefix bytes from previous TREC docno (vbyte)
* - length of suffix bytes (vbyte)
* - suffix bytes
* - weight (float)
*
* this implementation doesn't currently handle deletion, but has been designed
* so that adding it shouldn't be too much hassle. The in-memory arrays can be
* limited by knowing the smallest active entry. The map can be made a true
* search structure (such as rbtree) instead of just a sorted array. Adjacent
* pages can be combined when they become sufficiently sparse, and freed pages
* reused for new content.
*
* The format of each cache page is:
* - the constant byte '0xca', or constant byte '0xcf' for the final cache
* page
* - the last document number in the docmap (acts as a timestamp - vbyte)
* - the page number of the first cache page (vbyte)
* - a series of self-delimiting entries containing cache information:
* - byte 0x00 meaning end-of-page
* - byte 0x02, followed by the aggregate quantities (sum_bytes, sum_words,
* sum_dwords and sum_weight) written as a vbyte normalised fraction
* (multiplied by UINT32_MAX) and a vbyte exponent
* - byte 0x03, followed by a map cache entry:
* - vbyte number of entries in the map
* - vbyte offset of following entries in the map
* - uint32_t number of following entries from the map
* - a series of vbyte map entries
* - note that no entry shall exceed a page, and that the offsets
* encountered should form a sequence across pages
* - byte 0x4, followed by word array entry, same format as map array
* - byte 0x5, followed by dword array entry, same format as word array
* - byte 0x6, followed by bytes array entry, same format as word array
* - byte 0x7, followed by weight array entry, same format as word array,
* except that weights are stored like aggregate doubles
* - byte 0x8, followed by trecno offset array entry, same format as word
* array
* - byte 0x9, followed by trecno font-coded entry:
* - vbyte number of characters in array
* - vbyte offset of following entry in array
* - vbyte number of characters in following entry
* - a character array of front-coded trecno entries
*
* There are a few in-memory tricks which i've used to reduce the amount of
* space required to cache elements. The first is to use 3-in-4
* front-coding of TREC docno entries. This means that only every fourth entry
* is wholly represented, with three subsequent entries front-coded against it.
* All entries are stored in a character array, with an array of pointers into
* the char array for reasonably fast random access. Last time i checked, this
* saved 50% of the space required for sensibly ordered TREC docnos.
*
* A process similar to 3-in-4 front-coding is used for the location array.
* The array records vbyte'd byte counts for each document. Every 8 entries,
* an offset into the array is recorded for faster random access. In addition,
* if that record has a non-zero offset (which we can tell through the reposset
* structure) then the offset is recorded before the vbyte bytes entry.
* This provides access to the byte counts, but additionally allows the offset
* to be reconstructed for any record by decoding at most 8 bytes records.
*
* To complete the in-memory caching of location information, document types
* are recorded in a sorted in-memory array if they are not of type
* MIME_TYPE_APPLICATION_X_TREC (yes, this is a bit hacky).
*
* There are still some flaws in the current implementation (XXX). Firstly,
* the empty docmap has no valid representation.
* Also, the number of entries on a page can change during reading, which won't
* be picked up. Also, there's some confusion about whether buffers point to a
* series of pages or to a single page.
*
* written nml 2006-04-06
*
*/
#include "firstinclude.h"
#include "_docmap.h"
#include "docmap.h"
#include "def.h"
#include "binsearch.h"
#include "fdset.h"
#include "mem.h"
#include "reposset.h"
#include "_reposset.h"
#include "str.h"
#include "timings.h"
#include "vec.h"
#include "zstdint.h"
#include "zvalgrind.h"
#include <assert.h>
#include <stdlib.h>
#include <string.h>
/* frequency of front-coding of trecno's. For example, a value of 4 sets a
* policy of 3-in-4 front coding, where only every fourth entry is not
* front-encoded */
#define TRECNO_FRONT_FREQ 4
/* frequency of relative coding of locations. */
#define LOC_REL_FREQ 8
/* initial memory allocated to various arrays */
#define INIT_LEN 8
/* character constants for identifying pages */
#define FINAL_DATA_BYTE ((char) 0xdf)
#define DATA_BYTE ((char) 0xda)
#define CACHE_BYTE ((char) 0xca)
#define FINAL_CACHE_BYTE ((char) 0xcf)
#define DOCMAP_WEIGHT_PRECISION 7
enum cache_id {
CACHE_ID_END = 0,
CACHE_ID_AGG = 0x01,
CACHE_ID_MAP = 0x02,
CACHE_ID_WORDS = 0x03,
CACHE_ID_DWORDS = 0x04,
CACHE_ID_WEIGHT = 0x07,
CACHE_ID_TRECNO = 0x08,
CACHE_ID_TRECNO_CODE = 0x09,
CACHE_ID_REPOS_REC = 0x0a,
CACHE_ID_REPOS_CHECK = 0x0b,
CACHE_ID_LOC = 0x0c,
CACHE_ID_LOC_CODE = 0x0d,
CACHE_ID_TYPEEX = 0x0e
};
/* internal function to make a cursor point to nothing */
static void invalidate_cursor(struct docmap_cursor *cur) {
cur->buf = NULL;
cur->page = -1;
cur->entry.docno = -1;
cur->entry.fileno = -1;
cur->entry.offset = -1;
cur->entry.bytes = 0;
cur->first_docno = cur->last_docno = -1;
cur->pos.pos = cur->pos.end = NULL;
}
/* internal function to invalidate a buffer entry and the cursors that point to
* it */
static void invalidate_buffer(struct docmap *dm, struct docmap_buffer *buf) {
if (dm->read.buf == buf) {
invalidate_cursor(&dm->read);
}
if (dm->write.buf == buf) {
invalidate_cursor(&dm->write);
}
buf->page = -1;
buf->buflen = 0;
buf->dirty = 0;
}
/* internal function to initialise a docmap structure, used for both docmap_new
* and docmap_load */
static struct docmap *docmap_init(struct fdset *fdset,
int fd_type, unsigned int pagesize, unsigned int pages,
unsigned long int max_filesize, enum docmap_cache cache) {
struct docmap *dm = malloc(sizeof(*dm));
if (pages < 2) {
pages = 2;
}
if (dm && (dm->buf = malloc(pagesize * pages))
&& (dm->map = malloc(sizeof(*dm->map) * INIT_LEN))
&& (dm->rset = reposset_new())) {
dm->map_size = INIT_LEN;
dm->map_len = 0;
dm->map[dm->map_len] = ULONG_MAX;
dm->file_pages = max_filesize / pagesize;
dm->fdset = fdset;
dm->fd_type = fd_type;
dm->pagesize = pagesize;
dm->entries = 0;
/* assign all pages to reading, append buffering will steal them later
* if necessary */
invalidate_cursor(&dm->read);
invalidate_cursor(&dm->write);
invalidate_buffer(dm, &dm->readbuf);
invalidate_buffer(dm, &dm->appendbuf);
dm->read.entry.trecno = dm->write.entry.trecno = NULL;
dm->read.entry.trecno_len = dm->read.entry.trecno_size = 0;
dm->write.entry.trecno_len = dm->write.entry.trecno_size = 0;
dm->readbuf.buf = dm->buf;
dm->readbuf.bufsize = pages;
dm->appendbuf.page = 0; /* append gets page 0 because it has no
* buflen */
dm->appendbuf.buf = dm->buf;
dm->appendbuf.bufsize = 0;
dm->dirty = 0;
dm->cache.cache = cache;
dm->cache.len = dm->cache.size = 0;
dm->cache.words = NULL;
dm->cache.dwords = NULL;
dm->cache.weight = NULL;
dm->cache.loc_off = NULL;
dm->cache.loc.buf = NULL;
dm->cache.loc.len = dm->cache.loc.size = 0;
dm->cache.trecno_off = NULL;
dm->cache.trecno.buf = NULL;
dm->cache.trecno.len = dm->cache.trecno.size = 0;
dm->cache.typeex = NULL;
dm->cache.typeex_len = dm->cache.typeex_size = 0;
dm->agg.avg_bytes = dm->agg.sum_bytes
= dm->agg.avg_weight = dm->agg.sum_weight
= dm->agg.avg_words = dm->agg.sum_words
= dm->agg.avg_dwords = dm->agg.sum_dwords
= dm->agg.sum_trecno = 0;
} else {
if (dm) {
if (dm->buf) {
free(dm->buf);
dm->buf = NULL;
}
free(dm);
dm = NULL;
}
}
return dm;
}
/* internal function to encode an entry relative to a previous entry into
* some space. */
static enum docmap_ret encode(struct vec *v, struct docmap_entry *prev,
struct docmap_entry *curr) {
uintmax_t tmp[9] = {0, 0, 0, 0, 0, 0, 0, 0, 0};
unsigned int arrlen = sizeof(tmp) / sizeof(*tmp),
prefix,
len,
bytes,
i = 0;
char *pos = v->pos;
/* figure out the length of the front-coding prefix between trecnos */
if (curr->trecno_len < prev->trecno_len) {
len = curr->trecno_len;
} else {
len = prev->trecno_len;
}
for (prefix = 0;
prefix < len
&& curr->trecno[prefix] == prev->trecno[prefix]; prefix++) ;
assert(prefix <= len && prefix <= curr->trecno_len);
if (curr->fileno == prev->fileno
&& curr->offset == prev->offset + (off_t) prev->bytes) {
/* document immediately follows the previous */
tmp[i++] = 0;
arrlen--;
} else {
assert(curr->fileno >= prev->fileno || prev->fileno == -1);
tmp[i++] = 1 + curr->fileno - prev->fileno;
assert(tmp[i - 1] > 0); /* CANNOT be zero */
tmp[i++] = curr->offset;
}
/* encode docno gap and flags into one integer */
tmp[i] = curr->docno - prev->docno;
tmp[i] <<= 1;
tmp[i++] |= curr->flags;
tmp[i++] = curr->dwords;
tmp[i++] = curr->words - curr->dwords;
tmp[i++] = curr->bytes + 1 - 2 * curr->words;
tmp[i++] = curr->mtype;
tmp[i++] = prefix;
tmp[i++] = curr->trecno_len - prefix;
if ((vec_maxint_arr_write(v, tmp, arrlen, &bytes) == arrlen)
&& vec_byte_write(v, curr->trecno + prefix,
(unsigned int) tmp[arrlen - 1]) == tmp[arrlen - 1]
&& vec_flt_write(v, curr->weight, DOCMAP_WEIGHT_PRECISION)) {
/* Relations that we use to compress entries. These assertions go
* after encoding so that we don't assert things if encoding fails) */
assert(curr->words >= curr->dwords);
assert(curr->bytes + 1 >= 2 * curr->words);
return DOCMAP_OK;
} else {
v->pos = pos;
return DOCMAP_BUFSIZE_ERROR;
}
}
/* internal function to decode an entry from some space. Note that target must
* contain the previous entry for the next entry to be correctly decoded. */
static enum docmap_ret decode(struct docmap_cursor *cur) {
struct vec *v = &cur->pos;
struct docmap_entry *target = &cur->entry;
uintmax_t tmp[9] = {0, 0, 0, 0, 0, 0, 0, 0, 0};
unsigned int i,
bytes,
readlen,
arrlen = sizeof(tmp) / sizeof(*tmp);
char *pos = v->pos;
if (cur->past < cur->entries
&& vec_maxint_arr_read(v, tmp, 1, &readlen)) {
/* have to examine fileno before we can tell how many more numbers to
* read */
if (!tmp[0]) {
/* fileno indicates that this doc starts immediately after prev */
i = 2;
tmp[0] = target->fileno;
tmp[1] = target->offset + target->bytes;
} else {
tmp[0] += target->fileno - 1;
i = 1;
/* offset will be read below */
}
if (vec_maxint_arr_read(v, &tmp[i], arrlen - i, &bytes)) {
/* ensure that we've got enough space for the trecno */
if (target->trecno_size < tmp[arrlen - 2] + tmp[arrlen - 1] + 1) {
void *ptr = realloc(target->trecno, (unsigned int)
(tmp[arrlen - 2] + tmp[arrlen - 1] + 1));
if (ptr) {
target->trecno = ptr;
target->trecno_size = (unsigned int)
(tmp[arrlen - 2] + tmp[arrlen - 1] + 1);
} else {
assert(!CRASH);
v->pos = pos;
return DOCMAP_MEM_ERROR;
}
}
i = 0;
assert(tmp[0] != target->fileno || tmp[1] >= target->offset);
target->fileno = (unsigned int) tmp[i++];
target->offset = (unsigned long int) tmp[i++];
target->flags = (int) (tmp[i] & 1);
target->docno += (((unsigned int) tmp[i++]) >> 1);
target->dwords = (unsigned int) tmp[i++];
target->words = target->dwords + (unsigned int) tmp[i++];
target->bytes = (unsigned int) tmp[i++] + 2 * target->words - 1;
target->mtype = (unsigned int) tmp[i++];
if (vec_byte_read(v, target->trecno + tmp[i],
(unsigned int) tmp[i + 1]) == tmp[i + 1]
&& vec_flt_read(v, &target->weight, DOCMAP_WEIGHT_PRECISION)) {
target->trecno_len = (unsigned int) tmp[i]
+ (unsigned int) tmp[i + 1];
target->trecno[target->trecno_len] = '\0';
cur->past++;
return DOCMAP_OK;
} else {
/* bugger, we've destroyed the previous entry */
assert("can't get here" && 0);
return DOCMAP_BUFSIZE_ERROR;
}
}
return DOCMAP_OK;
}
v->pos = pos;
return DOCMAP_BUFSIZE_ERROR;
}
/* internal method to transfer a buffer from the read buffer to the append
* buffer */
static enum docmap_ret take_read_buffer(struct docmap *dm) {
assert(!dm->readbuf.dirty);
assert(dm->readbuf.bufsize > 1);
if (!dm->readbuf.bufsize) {
return DOCMAP_ARG_ERROR;
}
dm->readbuf.buf += dm->pagesize;
dm->readbuf.page++;
dm->readbuf.bufsize--;
if (dm->readbuf.buflen >= dm->readbuf.bufsize) {
dm->readbuf.buflen = dm->readbuf.bufsize;
}
if (dm->read.pos.pos >= dm->readbuf.buf && dm->read.pos.pos
< dm->readbuf.buf + dm->pagesize * dm->readbuf.buflen) {
/* do nothing */
} else {
invalidate_cursor(&dm->read);
}
dm->appendbuf.bufsize++;
return DOCMAP_OK;
}
static void zero_entry(struct docmap_entry *e) {
e->docno = 0;
e->fileno = 0;
e->offset = 0;
e->bytes = 0;
e->trecno_len = 0;
}
/* internal function to initialise the next page in the append buffer */
static enum docmap_ret init_append_buffer(struct docmap *dm,
unsigned long int docno) {
while (dm->map_len + 1 >= dm->map_size) {
void *ptr = realloc(dm->map,
sizeof(*dm->map) * (dm->map_size * 2 + 1));
if (ptr) {
dm->map = ptr;
dm->map_size *= 2;
dm->map_size++;
} else {
assert(!CRASH);
return DOCMAP_MEM_ERROR;
}
}
assert(dm->appendbuf.buflen < dm->appendbuf.bufsize);
dm->appendbuf.dirty = 1;
dm->write.pos.pos
= dm->appendbuf.buf + dm->appendbuf.buflen * dm->pagesize;
dm->write.pos.end = dm->write.pos.pos + dm->pagesize;
dm->write.entries = 0;
zero_entry(&dm->write.entry);
dm->appendbuf.buflen++;
dm->map[dm->map_len++] = docno;
dm->map[dm->map_len] = ULONG_MAX;
*dm->write.pos.pos++ = DATA_BYTE;
dm->write.pos.pos += sizeof(uint32_t); /* skip space for entries */
assert(vec_len(&dm->write.pos));
if (DEAR_DEBUG) {
memset(dm->write.pos.pos, 0, vec_len(&dm->write.pos));
}
return DOCMAP_OK;
}
/* internal function to commit a dirty buffer to disk */
static enum docmap_ret commit(struct docmap *dm, struct docmap_buffer *buf) {
unsigned int page = buf->page,
pages = buf->buflen,
pagesize = dm->pagesize,
fileno,
target,
bytes;
char *pos = buf->buf;
unsigned long int offset;
int fd;
assert(buf->dirty);
/* note that buffers may span multiple files */
do {
fileno = page / dm->file_pages;
offset = page % dm->file_pages;
target = pages;
if (offset + target > dm->file_pages) {
target = dm->file_pages - offset;
assert(target);
}
VALGRIND_CHECK_READABLE(pos, target * pagesize);
if (((fd = fdset_pin(dm->fdset, dm->fd_type, fileno, offset * pagesize,
SEEK_SET)) >= 0)
&& (bytes = write(fd, pos, target * pagesize)) == target * pagesize
&& fdset_unpin(dm->fdset, dm->fd_type, fileno, fd) == FDSET_OK) {
pages -= target;
pos += bytes;
} else {
if (fd >= 0) {
fdset_unpin(dm->fdset, dm->fd_type, fileno, fd);
}
return DOCMAP_IO_ERROR;
}
} while (pages);
buf->dirty = 0;
return DOCMAP_OK;
}
/* internal function to recalculate aggregate values */
static void aggregate(struct docmap *dm) {
if (dm->entries) {
dm->agg.avg_words = dm->agg.sum_words / dm->entries;
dm->agg.avg_dwords = dm->agg.sum_dwords / dm->entries;
dm->agg.avg_bytes = dm->agg.sum_bytes / dm->entries;
dm->agg.avg_weight = dm->agg.sum_weight / dm->entries;
} else {
dm->agg.avg_words = dm->agg.avg_dwords = dm->agg.avg_bytes
= dm->agg.avg_weight = 0;
}
}
static enum docmap_ret cbuf_realloc(struct docmap_cbuf *buf) {
void *ptr = realloc(buf->buf, buf->size * 2 + 1);
if (ptr) {
buf->buf = ptr;
buf->size *= 2;
buf->size++;
return DOCMAP_OK;
} else {
assert(!CRASH);
return DOCMAP_MEM_ERROR;
}
}
static enum docmap_ret cache(struct docmap *dm, struct docmap_entry *entry) {
enum docmap_ret dmret;
assert(dm->cache.len == entry->docno);
if (dm->cache.cache & DOCMAP_CACHE_WORDS) {
dm->cache.words[entry->docno] = entry->words;
}
if (dm->cache.cache & DOCMAP_CACHE_DISTINCT_WORDS) {
dm->cache.dwords[entry->docno] = entry->dwords;
}
if (dm->cache.cache & DOCMAP_CACHE_WEIGHT) {
dm->cache.weight[entry->docno] = entry->weight;
}
if (dm->cache.cache & DOCMAP_CACHE_LOCATION) {
/* look up fileno of this entry */
struct vec v;
struct reposset_record *rec = reposset_record(dm->rset, entry->docno);
enum docmap_ret dmret;
unsigned int bytes;
assert(rec);
v.pos = dm->cache.loc.buf + dm->cache.loc.len;
v.end = dm->cache.loc.buf + dm->cache.loc.size;
if (!(entry->docno % LOC_REL_FREQ)) {
/* need to add random access for this entry */
dm->cache.loc_off[entry->docno / LOC_REL_FREQ] = dm->cache.loc.len;
if (rec->rectype == REPOSSET_SINGLE_FILE
&& rec->docno != entry->docno) {
/* need to record offset for this entry, since it's in a
* many-docs file and it is not the first entry. */
uintmax_t arr = entry->offset;
while (!(vec_maxint_arr_write(&v, &arr, 1, &bytes))) {
if ((dmret = cbuf_realloc(&dm->cache.loc)) == DOCMAP_OK) {
v.pos = dm->cache.loc.buf + dm->cache.loc.len;
v.end = dm->cache.loc.buf + dm->cache.loc.size;
} else {
return dmret;
}
}
dm->cache.loc.len += bytes;
}
}
/* write the number of bytes for this doc into the loc buffer */
while (!(bytes = vec_vbyte_write(&v, entry->bytes))) {
if ((dmret = cbuf_realloc(&dm->cache.loc)) == DOCMAP_OK) {
v.pos = dm->cache.loc.buf + dm->cache.loc.len;
v.end = dm->cache.loc.buf + dm->cache.loc.size;
} else {
return dmret;
}
}
dm->cache.loc.len += bytes;
/* record type if it's not X_TREC */
if (entry->mtype != MIME_TYPE_APPLICATION_X_TREC) {
/* ensure that we have enough memory */
if (dm->cache.typeex_len >= dm->cache.typeex_size) {
void *ptr = realloc(dm->cache.typeex,
sizeof(*dm->cache.typeex) * (dm->cache.typeex_size * 2 + 1));
if (ptr) {
dm->cache.typeex = ptr;
dm->cache.typeex_size *= 2;
dm->cache.typeex_size++;
} else {
assert(!CRASH);
return DOCMAP_MEM_ERROR;
}
}
dm->cache.typeex[dm->cache.typeex_len].docno = entry->docno;
dm->cache.typeex[dm->cache.typeex_len++].mtype = entry->mtype;
}
}
if (dm->cache.cache & DOCMAP_CACHE_TRECNO) {
struct vec v;
struct vec readv;
unsigned int index = entry->docno / TRECNO_FRONT_FREQ;
unsigned int offset = entry->docno % TRECNO_FRONT_FREQ;
int encoded;
unsigned int front = 0;
v.pos = dm->cache.trecno.buf + dm->cache.trecno.len;
v.end = dm->cache.trecno.buf + dm->cache.trecno.size;
if (offset) {
unsigned int i,
ret;
unsigned long int len,
prefix;
encoded = 1;
/* front-code against unencoded entry */
readv.pos = dm->cache.trecno.buf + dm->cache.trecno_off[index];
readv.end = dm->cache.trecno.buf + dm->cache.trecno.len;
ret = vec_vbyte_read(&readv, &len);
assert(ret);
for (front = 0;
front < len && readv.pos[front] == entry->trecno[front];
front++) ;
readv.pos += len;
/* propagate front-code through similarly coded entries */
for (i = 0; i + 1 < offset; i++) {
ret = vec_vbyte_read(&readv, &prefix);
assert(ret);
ret = vec_vbyte_read(&readv, &len);
assert(ret);
assert(vec_len(&readv) >= len);
if (front >= prefix) {
/* front may only match the length of the prefix, but can
* be extended to match the rest of the string */
for (front = prefix;
front < len + prefix && readv.pos[front - prefix]
== entry->trecno[front]; front++) ;
}
readv.pos += len;
}
} else {
/* don't front-encode this entry, just write it in */
encoded = 0;
dm->cache.trecno_off[index] = dm->cache.trecno.len;
}
/* write either unencoded or front-coded entry */
while ((encoded && !vec_vbyte_write(&v, front))
|| !vec_vbyte_write(&v, entry->trecno_len - front)
|| vec_byte_write(&v, entry->trecno + front,
entry->trecno_len - front) != entry->trecno_len - front) {
if ((dmret = cbuf_realloc(&dm->cache.trecno)) == DOCMAP_OK) {
v.pos = dm->cache.trecno.buf + dm->cache.trecno.len;
v.end = dm->cache.trecno.buf + dm->cache.trecno.size;
} else {
return dmret;
}
}
dm->cache.trecno.len = v.pos - dm->cache.trecno.buf;
}
dm->cache.len++;
return DOCMAP_OK;
}
static enum docmap_ret cache_realloc(struct docmap *dm) {
void *ptr;
if (dm->cache.cache & DOCMAP_CACHE_WORDS) {
if ((ptr = realloc(dm->cache.words,
sizeof(*dm->cache.words) * dm->cache.size))) {
dm->cache.words = ptr;
} else {
assert(!CRASH);
return DOCMAP_MEM_ERROR;
}
}
if (dm->cache.cache & DOCMAP_CACHE_DISTINCT_WORDS) {
if ((ptr = realloc(dm->cache.dwords,
sizeof(*dm->cache.dwords) * dm->cache.size))) {
dm->cache.dwords = ptr;
} else {
assert(!CRASH);
return DOCMAP_MEM_ERROR;
}
}
if (dm->cache.cache & DOCMAP_CACHE_WEIGHT) {
if ((ptr = realloc(dm->cache.weight,
sizeof(*dm->cache.weight) * dm->cache.size))) {
dm->cache.weight = ptr;
} else {
assert(!CRASH);
return DOCMAP_MEM_ERROR;
}
}
if (dm->cache.cache & DOCMAP_CACHE_TRECNO) {
if ((ptr = realloc(dm->cache.trecno_off, sizeof(*dm->cache.trecno_off)
* (dm->cache.size / TRECNO_FRONT_FREQ + 1)))) {
dm->cache.trecno_off = ptr;
} else {
assert(!CRASH);
return DOCMAP_MEM_ERROR;
}
}
if (dm->cache.cache & DOCMAP_CACHE_LOCATION) {
if ((ptr = realloc(dm->cache.loc_off, sizeof(*dm->cache.loc_off)
* (dm->cache.size / LOC_REL_FREQ + 1)))) {
dm->cache.loc_off = ptr;
} else {
assert(!CRASH);
return DOCMAP_MEM_ERROR;
}
}
return DOCMAP_OK;
}
void update_append_entries(struct docmap *dm) {
/* first update entries in current append page */
mem_hton(dm->appendbuf.buf
+ (dm->appendbuf.buflen - 1) * dm->pagesize + 1,
&dm->write.entries, sizeof(dm->write.entries));
}
enum docmap_ret docmap_add(struct docmap *dm,
unsigned int fileno, off_t offset,
unsigned int bytes, enum docmap_flag flags,
unsigned int words, unsigned int distinct_words,
float weight, const char *trecno, unsigned trecno_len, enum mime_types mtype,
unsigned long int *docno) {
enum docmap_ret dmret;
struct docmap_entry entry;
char *tmp;
unsigned int tmplen,
reposno;
enum reposset_ret rret;
assert(offset >= 0);
/* copy parameters into entry */
entry.docno = dm->entries;
entry.fileno = fileno;
entry.offset = offset;
entry.bytes = bytes;
entry.words = words;
entry.dwords = distinct_words;
entry.flags = flags;
entry.mtype = mtype;
entry.weight = weight;
entry.trecno = (char *) trecno;
entry.trecno_size = entry.trecno_len = trecno_len;
/* fiddle reposset in response to new docno */
assert(!entry.docno
|| reposset_reposno(dm->rset, entry.docno - 1, &reposno) == REPOSSET_OK);
if (!offset) {
unsigned int reposno;
if ((rret = reposset_append(dm->rset, entry.docno, &reposno))
== REPOSSET_OK) {
assert(reposno == entry.fileno);
} else {
assert(!CRASH);
return DOCMAP_MEM_ERROR;
}
/* add a checkpoint at the start of compressed repositories */
if (entry.flags & DOCMAP_COMPRESSED) {
rret = reposset_add_checkpoint(dm->rset, entry.fileno,
MIME_TYPE_APPLICATION_X_GZIP, 0);
if (rret != REPOSSET_OK) {
assert(!CRASH);
return DOCMAP_MEM_ERROR;
}
}
} else if ((rret = reposset_append_docno(dm->rset, entry.docno, 1))
!= REPOSSET_OK) {
assert(!CRASH);
return DOCMAP_MEM_ERROR;
}
/* check that previous reposno hasn't changed */
assert(!entry.docno
|| (reposset_reposno(dm->rset, entry.docno - 1, &tmplen) == REPOSSET_OK
&& tmplen == reposno));
/* check that this reposno is as expected */
assert(reposset_reposno(dm->rset, entry.docno, &tmplen) == REPOSSET_OK
&& tmplen == entry.fileno);
/* ensure we've got enough space to store docno *before* we encode */
while (dm->write.entry.trecno_size <= entry.trecno_len) {
void *ptr = realloc(dm->write.entry.trecno,
dm->write.entry.trecno_size * 2 + 1);
if (ptr) {
dm->write.entry.trecno = ptr;
dm->write.entry.trecno_size *= 2;
dm->write.entry.trecno_size++;
} else {
assert(!CRASH);
return DOCMAP_MEM_ERROR;
}
}
/* ensure we've got enough space for cache */
if (dm->cache.cache && dm->cache.len >= dm->cache.size) {
dm->cache.size = dm->cache.size * 2 + 1;
if ((dmret = cache_realloc(dm)) != DOCMAP_OK) {
dm->cache.size = (dm->cache.size - 1) / 2;
return dmret;
}
}
while ((dmret = encode(&dm->write.pos, &dm->write.entry, &entry))
!= DOCMAP_OK) {
switch (dmret) {
case DOCMAP_OK: assert("can't get here" && 0); break;
default: return dmret;
case DOCMAP_BUFSIZE_ERROR:
/* need to move to the next page */
if (dm->appendbuf.buflen) {
update_append_entries(dm);
}
assert(dm->readbuf.bufsize >= 1);
if (dm->readbuf.bufsize == 1) {
/* no more buffer space. commit current contents and then
* give all buffer back to read */
if ((!dm->readbuf.dirty
|| (dmret = commit(dm, &dm->readbuf)) == DOCMAP_OK)
&& (!dm->appendbuf.dirty
|| ((dmret = commit(dm, &dm->appendbuf)) == DOCMAP_OK))) {
/* remove the append buffer, giving it all back to the read
* buffer */
assert(dm->appendbuf.buflen == dm->appendbuf.bufsize);
invalidate_buffer(dm, &dm->readbuf);
dm->appendbuf.page += dm->appendbuf.buflen;
dm->appendbuf.buflen = 0;
dm->readbuf.bufsize += dm->appendbuf.bufsize;
dm->readbuf.buf = dm->buf;
dm->appendbuf.bufsize = 0;
/* invalidate write cursor */
invalidate_cursor(&dm->write);
} else {
return dmret;
}
}
/* see if we need to create a new file for the append page we're
* about to create */
if (!((dm->appendbuf.page + dm->appendbuf.buflen)
% dm->file_pages) && dm->appendbuf.page) {
/* create new file (not for first file though - docmap_new
* handles that) */
unsigned int dmfileno = (dm->appendbuf.page
+ dm->appendbuf.buflen) / dm->file_pages;
/* need to initialise a new file for this page */
int fd = fdset_create(dm->fdset, dm->fd_type, dmfileno);
if (fd >= 0) {
fdset_unpin(dm->fdset, dm->fd_type, dmfileno, fd);
} else {
return DOCMAP_IO_ERROR;
}
}
/* steal a page from the read buffer */
if ((dmret = take_read_buffer(dm)) != DOCMAP_OK) {
return dmret;
}
/* initialise buffer */
if ((dmret = init_append_buffer(dm, dm->entries)) != DOCMAP_OK) {
return dmret;
}
aggregate(dm); /* recalculate aggregates */
break;
}
}
dm->dirty = 1;
dm->appendbuf.dirty = 1;
dm->write.entries++;
/* update stored entry in write cursor */
tmp = dm->write.entry.trecno;
tmplen = dm->write.entry.trecno_size;
dm->write.entry = entry;
dm->write.entry.trecno = tmp;
dm->write.entry.trecno_size = tmplen;
assert(dm->write.entry.trecno_size > entry.trecno_len);
memcpy(dm->write.entry.trecno, entry.trecno, entry.trecno_len);
dm->write.entry.trecno[entry.trecno_len] = '\0';
if (dm->cache.cache) {
assert(dm->cache.len < dm->cache.size);
if ((dmret = cache(dm, &dm->write.entry)) != DOCMAP_OK) {
return dmret;
}
}
/* update aggregate entries */
dm->agg.sum_bytes += bytes;
dm->agg.sum_words += words;
dm->agg.sum_dwords += distinct_words;
dm->agg.sum_weight += weight;
dm->agg.sum_trecno += trecno_len;
*docno = dm->entries++;
return DOCMAP_OK;
}
static int map_cmp(const void *vone, const void *vtwo) {
const unsigned long int *one = vone,
*two = vtwo;
if (*one < *two) {
return -1;
} else if (*one > *two) {
return 1;
} else {
return 0;
}
}
/* internal function to reset a cursor to the first entry of a new buffer */
static void reset_cursor(struct docmap *dm, struct docmap_cursor *cur,
struct docmap_buffer *buf, unsigned int page, unsigned long int last_docno) {
int ret;
assert(page >= buf->page && page < buf->page + buf->buflen);
cur->pos.pos = buf->buf + dm->pagesize * (page - buf->page);
cur->pos.end = cur->pos.pos + dm->pagesize;
cur->page = page;
cur->buf = buf;
cur->last_docno = last_docno;
cur->past = 0;
/* read header */
ret = *cur->pos.pos++;
assert(ret == DATA_BYTE || ret == FINAL_DATA_BYTE);
assert(vec_len(&cur->pos) > sizeof(cur->entries));
mem_ntoh(&cur->entries, cur->pos.pos, sizeof(cur->entries));
cur->pos.pos += sizeof(cur->entries);
zero_entry(&cur->entry);
ret = decode(cur);