X-Git-Url: http://git.indexdata.com/?p=idzebra-moved-to-github.git;a=blobdiff_plain;f=index%2Fkcompare.c;h=83f5af58e1550111cde6891ea43dde79e15be626;hp=fa407ecbdb0ebb608cfb52fdc46ccc007ddae6b0;hb=ecb3935e78cd9bcfdebafdee0834cfb1060d7b5e;hpb=e16099ccf0ad1aab3ff7930bc4ea542357af0e38 diff --git a/index/kcompare.c b/index/kcompare.c index fa407ec..83f5af5 100644 --- a/index/kcompare.c +++ b/index/kcompare.c @@ -1,24 +1,24 @@ -/* - * Copyright (C) 1994-1995, Index Data I/S - * All rights reserved. - * Sebastian Hammer, Adam Dickmeiss - * - * $Log: kcompare.c,v $ - * Revision 1.3 1995-09-07 13:58:36 adam - * New parameter: result-set file descriptor (RSFD) to support multiple - * positions within the same result-set. - * Boolean operators: and, or, not implemented. - * Result-set references. - * - * Revision 1.2 1995/09/06 16:11:17 adam - * Option: only one word key per file. - * - * Revision 1.1 1995/09/04 09:10:36 adam - * More work on index add/del/update. - * Merge sort implemented. - * Initial work on z39 server. - * - */ +/* $Id: kcompare.c,v 1.58 2006-05-10 08:13:21 adam Exp $ + Copyright (C) 1995-2005 + Index Data ApS + +This file is part of the Zebra server. + +Zebra is free software; you can redistribute it and/or modify it under +the terms of the GNU General Public License as published by the Free +Software Foundation; either version 2, or (at your option) any later +version. + +Zebra is distributed in the hope that it will be useful, but WITHOUT ANY +WARRANTY; without even the implied warranty of MERCHANTABILITY or +FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License +for more details. + +You should have received a copy of the GNU General Public License +along with Zebra; see the file LICENSE.zebra. If not, write to the +Free Software Foundation, 59 Temple Place - Suite 330, Boston, MA +02111-1307, USA. +*/ #include #include @@ -27,47 +27,322 @@ #include "index.h" +#ifdef __GNUC__ +#define CODEC_INLINE inline +#else +#define CODEC_INLINE +#endif + +void key_logdump_txt(int logmask, const void *p, const char *txt) +{ + struct it_key key; + if (!txt) + txt = "(none)"; + if (p) + { + char formstr[128]; + int i; + + memcpy (&key, p, sizeof(key)); + assert(key.len > 0 && key.len <= IT_KEY_LEVEL_MAX); + *formstr = '\0'; + for (i = 0; ilen; + if (((struct it_key *) p2)->len > l) + l = ((struct it_key *) p2)->len; + assert (l <= 4 && l > 0); + for (i = 0; i < l; i++) + { + if (((struct it_key *) p1)->mem[i] != ((struct it_key *) p2)->mem[i]) + { + if (((struct it_key *) p1)->mem[i] > ((struct it_key *) p2)->mem[i]) + return l-i; + else + return i-l; + } + } + return 0; +} + +char *key_print_it (const void *p, char *buf) +{ + strcpy(buf, ""); + return buf; +} + int key_compare (const void *p1, const void *p2) { struct it_key i1, i2; + int i, l; memcpy (&i1, p1, sizeof(i1)); memcpy (&i2, p2, sizeof(i2)); - if (i1.sysno != i2.sysno) + l = i1.len; + if (i2.len > l) + l = i2.len; + assert (l <= 4 && l > 0); + for (i = 0; i < l; i++) + { + if (i1.mem[i] != i2.mem[i]) + { + if (i1.mem[i] > i2.mem[i]) + return l-i; + else + return i-l; + } + } + return 0; +} + +zint key_get_seq(const void *p) +{ + struct it_key k; + memcpy (&k, p, sizeof(k)); + return k.mem[k.len-1]; +} + +int key_qsort_compare (const void *p1, const void *p2) +{ + int r; + size_t l; + char *cp1 = *(char **) p1; + char *cp2 = *(char **) p2; + + if ((r = strcmp (cp1, cp2))) + return r; + l = strlen(cp1)+1; + if ((r = key_compare (cp1+l+1, cp2+l+1))) + return r; + return cp1[l] - cp2[l]; +} + +struct iscz1_code_info { + struct it_key key; +}; + +void *iscz1_start (void) +{ + struct iscz1_code_info *p = (struct iscz1_code_info *) + xmalloc (sizeof(*p)); + iscz1_reset(p); + return p; +} + +void key_init(struct it_key *key) +{ + int i; + key->len = 0; + for (i = 0; imem[i] = 0; +} + +void iscz1_reset (void *vp) +{ + struct iscz1_code_info *p = (struct iscz1_code_info *) vp; + int i; + p->key.len = 0; + for (i = 0; i< IT_KEY_LEVEL_MAX; i++) + p->key.mem[i] = 0; +} + +void iscz1_stop (void *p) +{ + xfree (p); +} + +/* small encoder that works with unsigneds of any length */ +static CODEC_INLINE void iscz1_encode_int (zint d, char **dst) +{ + unsigned char *bp = (unsigned char*) *dst; + + while (d > 127) + { + *bp++ = (unsigned) (128 | (d & 127)); + d = d >> 7; + } + *bp++ = (unsigned) d; + *dst = (char *) bp; +} + +/* small decoder that works with unsigneds of any length */ +static CODEC_INLINE zint iscz1_decode_int (unsigned char **src) +{ + zint d = 0; + unsigned char c; + unsigned r = 0; + + while (((c = *(*src)++) & 128)) { - if (i1.sysno > i2.sysno) - return 2; - else - return -2; + d += ((zint) (c&127) << r); + r += 7; } - if (i1.seqno != i2.seqno) + d += ((zint) c << r); + return d; +} + +void iscz1_encode (void *vp, char **dst, const char **src) +{ + struct iscz1_code_info *p = (struct iscz1_code_info *) vp; + struct it_key tkey; + zint d; + int i; + + /* 1 + 3, 2, 9, 12 + 3, 2, 10, 2 + 4, 1 + + if diff is 0, then there is more ... + if diff is non-zero, then _may_ be more + */ + memcpy (&tkey, *src, sizeof(struct it_key)); + + /* deal with leader + delta encoding .. */ + d = 0; + assert(tkey.len > 0 && tkey.len <= 4); + for (i = 0; i < tkey.len; i++) { - if (i1.seqno > i2.seqno) - return 1; - else - return -1; + d = tkey.mem[i] - p->key.mem[i]; + if (d || i == tkey.len-1) + { /* all have been equal until now, now make delta .. */ + p->key.mem[i] = tkey.mem[i]; + if (d > 0) + { + iscz1_encode_int (i + (tkey.len << 3) + 64, dst); + i++; + iscz1_encode_int (d, dst); + } + else + { + iscz1_encode_int (i + (tkey.len << 3), dst); + } + break; + } } -#if IT_KEY_HAVE_FIELD - if (i1.field != i2.field) + /* rest uses absolute encoding ... */ + for (; i < tkey.len; i++) { - if (i1.field > i2.field) - return 1; - else - return -1; + iscz1_encode_int (tkey.mem[i], dst); + p->key.mem[i] = tkey.mem[i]; } -#endif - return 0; + (*src) += sizeof(struct it_key); } -int key_compare_x (const struct it_key *i1, const struct it_key *i2) +void iscz1_decode (void *vp, char **dst, const char **src) { - if ( i1->sysno != i2->sysno) - return i1->sysno - i2->sysno; -#if IT_KEY_HAVE_FIELD - if ( i1->seqno != i2->seqno) - return i1->seqno - i2->seqno; - return i1->field - i2->field; -#else - return i1->seqno - i2->seqno; -#endif + struct iscz1_code_info *p = (struct iscz1_code_info *) vp; + int i; + + int leader = (int) iscz1_decode_int ((unsigned char **) src); + i = leader & 7; + if (leader & 64) + p->key.mem[i] += iscz1_decode_int ((unsigned char **) src); + else + p->key.mem[i] = iscz1_decode_int ((unsigned char **) src); + p->key.len = (leader >> 3) & 7; + while (++i < p->key.len) + p->key.mem[i] = iscz1_decode_int ((unsigned char **) src); + memcpy (*dst, &p->key, sizeof(struct it_key)); + (*dst) += sizeof(struct it_key); +} + +ISAMS_M *key_isams_m (Res res, ISAMS_M *me) +{ + isams_getmethod (me); + + me->compare_item = key_compare; + me->log_item = key_logdump_txt; + + me->codec.start = iscz1_start; + me->codec.decode = iscz1_decode; + me->codec.encode = iscz1_encode; + me->codec.stop = iscz1_stop; + me->codec.reset = iscz1_reset; + + me->debug = atoi(res_get_def (res, "isamsDebug", "0")); + + return me; +} + +ISAMC_M *key_isamc_m (Res res, ISAMC_M *me) +{ + isamc_getmethod (me); + + me->compare_item = key_compare; + me->log_item = key_logdump_txt; + + me->codec.start = iscz1_start; + me->codec.decode = iscz1_decode; + me->codec.encode = iscz1_encode; + me->codec.stop = iscz1_stop; + me->codec.reset = iscz1_reset; + + me->debug = atoi(res_get_def (res, "isamcDebug", "0")); + + return me; +} + +int key_SU_encode (int ch, char *out) +{ + int i; + for (i = 0; ch; i++) + { + if (ch >= 64) + out[i] = 65 + (ch & 63); + else + out[i] = 1 + ch; + ch = ch >> 6; + } + return i; + /* in out + 0 1 + 1 2 + 63 64 + 64 65, 2 + 65 66, 2 + 127 128, 2 + 128 65, 3 + 191 128, 3 + 192 65, 4 + */ +} + +int key_SU_decode (int *ch, const unsigned char *out) +{ + int len = 1; + int fact = 1; + *ch = 0; + for (len = 1; *out >= 65; len++, out++) + { + *ch += (*out - 65) * fact; + fact <<= 6; + } + *ch += (*out - 1) * fact; + return len; } +/* + * Local variables: + * c-basic-offset: 4 + * indent-tabs-mode: nil + * End: + * vim: shiftwidth=4 tabstop=8 expandtab + */ +