1 /*-------------------------------------------------------------------------
3 * Copyright (c) 2006-2015, NTT DATA Corporation
8 * Update Ludia functions so that they are available with PostgreSQL9.1.
9 * Author: NTT DATA Corporation
11 *-------------------------------------------------------------------------
16 #include <sys/types.h>
20 #include "catalog/pg_type.h"
23 #include "ludia_funcs.h"
24 #include "mb/pg_wchar.h"
26 #include "storage/fd.h"
27 #include "utils/builtins.h"
28 #include "utils/guc.h"
29 #include "miscadmin.h"
31 #if PG_VERSION_NUM >= 90300
32 #include "access/htup_details.h"
37 /* Last update date of ludia_funcs */
38 #define PGS2_LAST_UPDATE "2015.09.10"
42 static bool pgs2_enable_debug = false;
44 static char *pgs2_last_update = NULL;
45 static int norm_cache_limit = -1;
46 static bool escape_snippet_keyword = false;
48 #define SEN_NORMALIZE_FLAGS 0
49 #define SEN_MAX_N_EXPRS 32
51 /* upper limit for GUC variables measured in kilobytes of memory */
52 /* note that various places assume the byte size fits in a "long" variable */
53 #if SIZEOF_SIZE_T > 4 && SIZEOF_LONG > 4
54 #define MAX_KILOBYTES INT_MAX
56 #define MAX_KILOBYTES (INT_MAX / 1024)
59 #define ISBACKSLASHCHAR(x) (*(x) == '\\')
60 #define ISDOUBLEQUOTECHAR(x) (*(x) == '"')
61 #define ISSENNAOPSCHAR(x) (*(x) == '+' || *(x) == '-' || *(x) == ' ')
63 PG_FUNCTION_INFO_V1(pgs2snippet1);
64 Datum pgs2snippet1(PG_FUNCTION_ARGS);
65 PG_FUNCTION_INFO_V1(pgs2norm);
66 Datum pgs2norm(PG_FUNCTION_ARGS);
67 PG_FUNCTION_INFO_V1(pgs2textporter1);
68 Datum pgs2textporter1(PG_FUNCTION_ARGS);
69 PG_FUNCTION_INFO_V1(pgs2seninfo);
70 Datum pgs2seninfo(PG_FUNCTION_ARGS);
72 static sen_encoding GetSennaEncoding(void);
73 static sen_query *GetSennaQuery(char *str, size_t len);
74 static bool EscapeSnippetKeyword(char **s, size_t *slen);
77 #define TEXTPORTER_TMPDIR "/tmp"
78 #define TEXTPORTER_MKSTEMP_UMASK 0177
79 #define TEXTPORTER_GROUPNAME "UTF-8"
80 #define TEXTPORTER_DEFLANGNAME "Japanese"
81 #define TEXTPORTER_BBIGENDIAN 1
82 #define TEXTPORTER_OPTION 0x00000020 /* DMC_GETTEXT_OPT_LF */
83 #define TEXTPORTER_OPTION_STRING "32"
84 #define TEXTPORTER_OPTION1 0x00010000 /* DMC_GETTEXT_OPT1_TXCONV */
85 #define TEXTPORTER_SIZE 0
86 #define TEXTPORTER_CSV_C 0
89 /* GUC variables for pgs2textpoter1 */
90 static int textporter_error = ERROR;
91 static unsigned int textporter_option = TEXTPORTER_OPTION;
94 * This variable is a dummy that doesn't do anything, except in some
95 * cases provides the value for SHOW to display. The real state is
96 * elsewhere and is kept in sync by assign_hooks.
98 static char *textporter_option_string;
100 static const struct config_enum_entry textporter_error_options[] = {
101 {"debug1", DEBUG1, false},
103 {"info", INFO, false},
104 {"notice", NOTICE, false},
105 {"warning", WARNING, false},
106 {"error", ERROR, false},
110 static void CleanupTextPorterTmpFiles(void);
112 static bool check_textporter_option(char **newval, void **extra, GucSource source);
113 static void assign_textporter_option(const char *newval, void *extra);
114 #endif /* TEXTPORTER */
125 /* Define custom GUC variable for debugging */
126 DefineCustomBoolVariable("ludia_funcs.enable_debug",
127 "Emit ludia_funcs debugging output.",
138 /* Can't be set in postgresql.conf */
139 DefineCustomStringVariable("ludia_funcs.last_update",
140 "Shows the last update date of ludia_funcs.",
145 GUC_REPORT | GUC_NOT_IN_SAMPLE | GUC_DISALLOW_IN_FILE,
151 /* Define custom GUC variables */
152 DefineCustomEnumVariable("ludia_funcs.textporter_error",
153 "Sets the message levels that are emitted "
154 "when textporter fails.",
158 textporter_error_options,
165 DefineCustomStringVariable("ludia_funcs.textporter_option",
166 "Sets the option used to get text data "
169 &textporter_option_string,
170 TEXTPORTER_OPTION_STRING,
173 check_textporter_option,
174 assign_textporter_option,
177 /* Clean up remaining textporter temporary files */
178 CleanupTextPorterTmpFiles();
179 #endif /* TEXTPORTER */
182 * A value of 0 means no limit on the cache size. A value of -1 means
183 * that work_mem is used as the upper size limit of the cache.
185 DefineCustomIntVariable("ludia_funcs.norm_cache_limit",
186 "Sets the maximum memory to be used for caching "
187 "the result of pgs2norm()",
199 DefineCustomBoolVariable("ludia_funcs.escape_snippet_keyword",
200 "Escapes snippet keyword string.",
202 &escape_snippet_keyword,
210 EmitWarningsOnPlaceholders("ludia_funcs");
212 /* Initialize Senna */
214 if (rc != sen_success)
216 (errmsg("sen_init() failed: %d", rc)));
225 #define REMOVE_TMPFILE(path) \
227 if (unlink(path) != 0) \
229 (errcode_for_file_access(), \
230 errmsg("could not remove temporary file \"%s\": %m", path))); \
234 pgs2textporter1(PG_FUNCTION_ARGS)
236 char *appfile = text_to_cstring(PG_GETARG_TEXT_P(0));
237 char txtfile[] = TEXTPORTER_TMPDIR "/ludia_funcs_XXXXXX";
243 bool return_null = false;
246 /* Confirm that database encoding is UTF-8 */
252 * Generate a unique temporary filename where text data gotten
253 * from application file by TextPorter is stored temporarily.
254 * Set the permission of a temporary file to 0600 to ensure that
255 * only the owner of PostgreSQL server can read and write the file.
257 oumask = umask(TEXTPORTER_MKSTEMP_UMASK);
258 tmpfd = mkstemp(txtfile);
263 (errcode_for_file_access(),
264 errmsg("could not generate a unique temporary filename: %m")));
265 if (close(tmpfd) != 0)
267 (errcode_for_file_access(),
268 errmsg("could not close temporary file \"%s\": %m", txtfile)));
271 * Run TextPorter to read text data from application file (appfile)
272 * to temporary file (txtfile).
274 ret = ExecTextPorter((unsigned char *)appfile,
275 (unsigned char *)txtfile,
276 (unsigned char *)TEXTPORTER_GROUPNAME,
277 (unsigned char *)TEXTPORTER_DEFLANGNAME,
278 TEXTPORTER_BBIGENDIAN, textporter_option,
279 TEXTPORTER_OPTION1, TEXTPORTER_SIZE,
283 ereport(textporter_error,
284 (errmsg("could not get text from application file \"%s\"",
286 errdetail("DMC_GetText_V5() failed with errcode %d",
289 /* Return NULL if textporter_error is set to other than ERROR */
294 /* Read text data from temporary file to memory */
295 if (stat(txtfile, &statbuf))
297 (errcode_for_file_access(),
298 errmsg("could not stat file \"%s\": %m", txtfile)));
299 result = (text *) palloc(statbuf.st_size + VARHDRSZ);
301 fp = AllocateFile(txtfile, "r");
304 (errcode_for_file_access(),
305 errmsg("could not open file \"%s\": %m", txtfile)));
307 if (fread(VARDATA(result), 1, statbuf.st_size, fp) != statbuf.st_size ||
310 (errcode_for_file_access(),
311 errmsg("could not read file \"%s\": %m", txtfile)));
316 REMOVE_TMPFILE(txtfile);
321 REMOVE_TMPFILE(txtfile);
329 SET_VARSIZE(result, statbuf.st_size + VARHDRSZ);
331 PG_RETURN_TEXT_P(result);
335 * Clean up remaining textporter temporary files
338 CleanupTextPorterTmpFiles(void)
342 char path[MAXPGPATH];
344 tpdir = AllocateDir(TEXTPORTER_TMPDIR);
347 (errcode_for_file_access(),
348 errmsg("could not open textporter temporary file directory \"%s\": %m",
349 TEXTPORTER_TMPDIR)));
351 while ((tpde = ReadDir(tpdir, TEXTPORTER_TMPDIR)) != NULL)
353 if (strlen(tpde->d_name) == 18 &&
354 strncmp(tpde->d_name, "ludia_funcs_", 12) == 0)
356 snprintf(path, MAXPGPATH, TEXTPORTER_TMPDIR "/%s", tpde->d_name);
357 REMOVE_TMPFILE(path);
365 check_textporter_option(char **newval, void **extra, GucSource source)
369 unsigned int *myextra;
372 val = strtoul(*newval, &endptr, 0);
377 if (errno == ERANGE || val != (unsigned long) ((unsigned int) val))
379 GUC_check_errhint("Value exceeds unsigned integer range.");
383 /* Set up the "extra" struct actually used by assign_textporter_option */
384 myextra = (unsigned int *) malloc(sizeof(unsigned int));
387 GUC_check_errcode(ERRCODE_OUT_OF_MEMORY);
388 GUC_check_errmsg("out of memory");
391 *myextra = (unsigned int) val;
392 *extra = (void *) myextra;
398 assign_textporter_option(const char *newval, void *extra)
400 textporter_option = *((unsigned int *) extra);
403 #else /* TEXTPORTER */
406 pgs2textporter1(PG_FUNCTION_ARGS)
411 #endif /* TEXTPORTER */
414 GetSennaEncoding(void)
416 static sen_encoding encoding = sen_enc_default;
418 if (encoding == sen_enc_default)
420 if (GetDatabaseEncoding() == PG_UTF8)
421 encoding = sen_enc_utf8;
424 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
425 errmsg("does not support database encoding \"%s\"",
426 GetDatabaseEncodingName())));
432 * Escape the backslash and double quote characters in the given string.
434 * Return false if the given string has no character which needs to be
435 * escaped. Otherwise, return true. In this case, **s points the palloc'd
436 * space storing the escaped keyword string and *slen is set to the size
437 * of that string. The caller needs to free the palloc'd space.
440 EscapeSnippetKeyword(char **s, size_t *slen)
447 bool in_doublequote = false;
448 bool in_sennaops = false;
449 bool need_escape = false;
452 * Skip the heading double quote character because it always doesn't
453 * need to be interpreted as a character itself and be escaped.
454 * Note that we must not skip the heading character if it's not a
458 if (ISDOUBLEQUOTECHAR(sp))
462 * Check whether the snippet keyword string has a character which
463 * needs to be escaped.
465 while ((sp - *s) < *slen)
467 mblen = pg_mblen(sp);
470 * Backslash in the keyword always needs to be escaped.
472 if (ISBACKSLASHCHAR(sp))
480 if (ISSENNAOPSCHAR(sp))
483 in_doublequote = false;
488 * Double quote in the keyword needs to be escaped if
489 * any Senna search operators are to neither its right
498 if (ISDOUBLEQUOTECHAR(sp) && !in_sennaops)
499 in_doublequote = true;
500 if (!ISSENNAOPSCHAR(sp))
508 * Quick exit if the keyword has no character which needs to be
515 * Allocate the buffer space to store the escaped snippet keyword string.
516 * The maximum size of escaped string is double the input keyword size.
517 * The size reaches the maximum when every character in the input keyword
518 * needs to be escaped.
520 ep = escaped = (char *) palloc(*slen * 2);
523 * Copy the characters which have been passed through in the above loop
524 * and don't need to be escaped, into the buffer. If in_doublequote is
525 * true, we don't copy the double quote in the previous position into the
526 * buffer because it might still need to be escaped.
528 copylen = sp - *s - ((in_doublequote) ? 1 : 0);
529 memcpy(ep, *s, copylen);
533 * Construct the escaped snippet keyword string.
535 while ((sp - *s) < *slen)
537 mblen = pg_mblen(sp);
542 * dqchar indicates the previous character, that is a double
543 * quote. We assume here that a double quote is single-byte
546 char dqchar = *(sp - 1);
548 if (ISSENNAOPSCHAR(sp))
551 * Don't escape the double quote which is just before Senna
557 in_doublequote = false;
562 * Escape the double quote if no Senna operator is next to it.
567 if (ISDOUBLEQUOTECHAR(sp))
568 in_doublequote = true;
571 if (ISBACKSLASHCHAR(sp))
573 memcpy(ep, sp, mblen);
575 in_doublequote = false;
581 if (ISDOUBLEQUOTECHAR(sp))
584 * Don't escape the double quote which is just after Senna
590 in_doublequote = true;
594 if (ISBACKSLASHCHAR(sp))
597 * We don't check ISSENNAOPSCHAR() here. We handle Senna
598 * operator character as a character itself instead of
599 * an operator if it doesn't follow a double quote.
601 memcpy(ep, sp, mblen);
605 if (!ISSENNAOPSCHAR(sp))
612 /* Add the tailing double quote into the buffer */
620 if (pgs2_enable_debug)
622 char *tmp = pnstrdup(*s, *slen);
624 elog(LOG, "escaped snippet keyword: %s", tmp);
633 GetSennaQuery(char *str, size_t len)
635 static sen_query *query_cache = NULL;
636 static char *key_cache = NULL;
637 static size_t len_cache = 0;
638 static bool guc_cache = false;
640 sen_encoding encoding;
644 bool needfree = false;
647 * Return the cached Senna query if the same keyword has
648 * been used the last time.
650 if (key_cache != NULL &&
652 strncmp(key_cache, str, len) == 0 &&
653 escape_snippet_keyword == guc_cache)
656 if (pgs2_enable_debug)
658 char *tmp = pnstrdup(str, len);
660 elog(LOG, "GetSennaQuery(): quick exit: %s", tmp);
667 encoding = GetSennaEncoding();
672 (errcode(ERRCODE_OUT_OF_MEMORY),
673 errmsg("out of memory")));
676 * We always cache the unescaped keyword. Which enables us
677 * to check whether we can use the cached Senna query before
678 * escaping the keyword.
680 memcpy(key, str, len);
684 * If the keyword has been escaped, 'str' points to the
685 * newly-palloc'd space storing the escaped keyword. This
686 * space needs to be freed later.
688 if (escape_snippet_keyword)
689 needfree = EscapeSnippetKeyword(&str, &len);
691 query = sen_query_open(str, len, sen_sel_or, SEN_MAX_N_EXPRS,
697 (errmsg("sen_query_open() failed")));
700 if ((rest = sen_query_rest(query, NULL)) != 0)
702 (errmsg("too many expressions (%d)", rest)));
704 if (query_cache != NULL)
706 sen_query_close(query_cache);
713 guc_cache = escape_snippet_keyword;
722 pgs2snippet1(PG_FUNCTION_ARGS)
724 int flags = PG_GETARG_INT32(0);
725 uint32 width = PG_GETARG_UINT32(1);
726 uint32 max_results = PG_GETARG_UINT32(2);
727 text *opentags = PG_GETARG_TEXT_P(3);
728 text *closetags = PG_GETARG_TEXT_P(4);
729 int mapping = PG_GETARG_INT32(5);
730 text *keywords = PG_GETARG_TEXT_P(6);
731 text *document = PG_GETARG_TEXT_P(7);
733 sen_snip *snip = NULL;
734 const char *opentags_str = VARDATA_ANY(opentags);
735 const char *closetags_str = VARDATA_ANY(closetags);
736 char *keywords_str = VARDATA_ANY(keywords);
737 char *document_str = VARDATA_ANY(document);
738 uint32 opentags_len = VARSIZE_ANY_EXHDR(opentags);
739 uint32 closetags_len = VARSIZE_ANY_EXHDR(closetags);
740 uint32 keywords_len = VARSIZE_ANY_EXHDR(keywords);
741 uint32 document_len = VARSIZE_ANY_EXHDR(document);
743 uint32 max_tagged_len = 0;
746 uint32 result_len = 0;
747 bool return_null = false;
749 query = GetSennaQuery(keywords_str, keywords_len);
751 snip = sen_query_snip(query, flags, width, max_results, 1,
752 &opentags_str, &opentags_len,
753 &closetags_str, &closetags_len,
754 mapping == 0 ? NULL : (sen_snip_mapping *)-1);
757 (errmsg("sen_query_snip() failed")));
761 rc = sen_snip_exec(snip, document_str, document_len,
762 &nresults, &max_tagged_len);
763 if (rc != sen_success)
765 (errmsg("sen_snip_exec() failed: %d", rc)));
767 result = (text *) palloc(max_tagged_len + VARHDRSZ);
769 rc = sen_snip_get_result(snip, 0, VARDATA(result), &result_len);
770 if (rc == sen_invalid_argument)
772 else if (rc != sen_success)
774 (errmsg("sen_snip_get_result() failed: %d", rc)));
778 sen_snip_close(snip);
783 sen_snip_close(snip);
788 SET_VARSIZE(result, max_tagged_len + VARHDRSZ);
790 PG_RETURN_TEXT_P(result);
794 * Make sure there is enough space for 'needed' more bytes.
796 * Sets **buf to the allocated space which can store the needed bytes if OK,
797 * NULL if failed to enlarge the space because 'needed' is larger than 'maxlen'.
800 pgs2malloc(void **buf, long *buflen, long needed, long maxlen)
803 if (pgs2_enable_debug)
804 elog(LOG, "pgs2malloc(): buflen %ld, needed %ld, maxlen %ld",
805 *buflen, needed, maxlen);
808 if (*buf != NULL && *buflen >= needed && (*buflen <= maxlen || maxlen == 0))
809 return; /* got enough space already */
812 * Release the already-allocated space since it's too small to
813 * store the needed bytes or larger than the upper limit.
823 * Don't allocate any space if the needed space is larger than
826 if (needed > maxlen && maxlen != 0)
830 * Allocate the space for the needed bytes.
832 * We don't want to allocate just a little more space with each enlarge;
833 * for efficiency, double the buffer size each time it overflows.
834 * Actually, we might need to more than double it if 'needed' is big...
836 * We check whether '*buflen' overflows each cycle to avoid infinite loop.
839 while (*buflen < needed && *buflen != 0)
843 * Clamp to maxlen in case we went past it. Note we are assuming
844 * here that maxlen <= LONG_MAX/2, else the above loop could
845 * overflow. We will still have *buflen >= needed.
847 if (*buflen > maxlen && maxlen != 0)
850 /* Guard against out-of-range '*buflen' value */
853 (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
854 errmsg("out of memory"),
855 errdetail("Cannot enlarge buffer by %ld more bytes.",
858 *buf = (void *) malloc(*buflen);
861 (errcode(ERRCODE_OUT_OF_MEMORY),
862 errmsg("out of memory")));
866 pgs2norm(PG_FUNCTION_ARGS)
868 text *str = PG_GETARG_TEXT_PP(0);
869 char *s = VARDATA_ANY(str);
870 long slen = VARSIZE_ANY_EXHDR(str);
878 * norm_cache is the cache memory storing both input and normalized strings
879 * as the result of pgs2norm(). norm_cache_size is the size of norm_cache
880 * and its upper limit is specified by norm_cache_limit parameter. norm_result
881 * is the pointer to the normalized string with the verlena header (i.e.,
882 * text type) stored in the latter half of the cache. norm_reslen is the size
883 * of norm_result. norm_slen is the size of the input string which is stored
884 * in the first half of the cache.
886 static char *norm_cache = NULL;
887 static long norm_cache_size = 0;
888 static long norm_slen = 0;
889 static char *norm_result = NULL;
890 static long norm_reslen = 0;
893 * Return the cached normalization result if the same string of
894 * the given one has been normalized the last time.
896 if (norm_cache != NULL &&
898 strncmp(norm_cache, s, slen) == 0)
901 if (pgs2_enable_debug)
903 char *tmp = text_to_cstring(str);
905 elog(LOG, "pgs2norm(): quick exit: %s", tmp);
910 PG_RETURN_TEXT_P(pnstrdup(norm_result, norm_reslen));
913 /* Confirm that database encoding is UTF-8 */
917 * Allocate the result buffer to store the normalized string. Since the size of
918 * normalized string can be larger than that of input one, the result buffer needs
919 * extra space. Problem is that, before calling sen_str_normalize, we need to
920 * allocate the result buffer but cannot know how large extra space is required.
921 * So we use RESULT_EXTRA_SIZE as the estimated size of extra space here.
923 #define RESULT_EXTRA_SIZE 64
924 buflen = slen + RESULT_EXTRA_SIZE;
927 result = (text *) palloc(buflen + VARHDRSZ);
929 #if defined(FAST_SENNA)
930 reslen = fast_sen_str_normalize(s, slen, VARDATA(result), buflen);
932 reslen = sen_str_normalize(s, slen, sen_enc_utf8,
934 VARDATA(result), buflen);
939 (errmsg("could not normalize the string")));
942 * If the result buffer size is too short to store the normalized string,
943 * we enlarge the buffer and retry the string normalization.
945 if (buflen <= reslen)
952 SET_VARSIZE(result, reslen + VARHDRSZ);
955 * Cache both input and normalized strings to accelerate the subsequent
956 * calls of pgs2norm() with the same input string. But we don't do that
957 * if the maximum allowed size of the cache is too small to store them.
959 needed = slen + reslen + VARHDRSZ;
960 maxlen = ((norm_cache_limit >= 0) ? norm_cache_limit : work_mem) * 1024L;
962 pgs2malloc((void **) &norm_cache, &norm_cache_size, needed, maxlen);
963 if (norm_cache != NULL)
965 /* Store the input string into the first half of the cache */
967 memcpy(norm_cache, s, slen);
970 * Store the normalized string with the varlena header (i.e., text type)
971 * into the latter half of the cache.
973 norm_result = norm_cache + slen;
974 norm_reslen = reslen + VARHDRSZ;
975 memcpy(norm_result, result, norm_reslen);
979 if (pgs2_enable_debug)
981 char *tmp = text_to_cstring(str);
983 elog(LOG, "pgs2norm(): complete (%s result cache): %s",
984 (norm_cache == NULL) ? "unset" : "set", tmp);
989 PG_RETURN_TEXT_P(result);
993 * Report the version and configure options of Senna which
994 * ludia_funcs depends on.
997 pgs2seninfo(PG_FUNCTION_ARGS)
999 char *version[MAXPGPATH];
1000 char *coptions[MAXPGPATH];
1007 * Get the version and configure options of Senna. Ignore the
1008 * return value of sen_info() because it always returns a success.
1010 sen_info((char **)&version, (char **)&coptions, NULL, NULL, NULL, NULL);
1013 * Construct a tuple descriptor for the result row. This must
1014 * match this function's ludia_funcs--x.x.sql entry.
1016 tupdesc = CreateTemplateTupleDesc(2, false);
1017 TupleDescInitEntry(tupdesc, (AttrNumber) 1,
1018 "version", TEXTOID, -1, 0);
1019 TupleDescInitEntry(tupdesc, (AttrNumber) 2,
1020 "configure_options", TEXTOID, -1, 0);
1021 tupdesc = BlessTupleDesc(tupdesc);
1024 values[0] = CStringGetTextDatum(*version);
1027 /* configure option */
1028 values[1] = CStringGetTextDatum(*coptions);
1031 tuple = heap_form_tuple(tupdesc, values, isnull);
1032 PG_RETURN_DATUM(HeapTupleGetDatum(tuple));