postgresql/contrib/fuzzystrmatch/dmetaphone.c
Peter Eisentraut e7128e8dbb Create function prototype as part of PG_FUNCTION_INFO_V1 macro
Because of gcc -Wmissing-prototypes, all functions in dynamically
loadable modules must have a separate prototype declaration.  This is
meant to detect global functions that are not declared in header files,
but in cases where the function is called via dfmgr, this is redundant.
Besides filling up space with boilerplate, this is a frequent source of
compiler warnings in extension modules.

We can fix that by creating the function prototype as part of the
PG_FUNCTION_INFO_V1 macro, which such modules have to use anyway.  That
makes the code of modules cleaner, because there is one less place where
the entry points have to be listed, and creates an additional check that
functions have the right prototype.

Remove now redundant prototypes from contrib and other modules.
2014-04-18 00:03:19 -04:00

1440 lines
32 KiB
C

/*
* This is a port of the Double Metaphone algorithm for use in PostgreSQL.
*
* contrib/fuzzystrmatch/dmetaphone.c
*
* Double Metaphone computes 2 "sounds like" strings - a primary and an
* alternate. In most cases they are the same, but for foreign names
* especially they can be a bit different, depending on pronunciation.
*
* Information on using Double Metaphone can be found at
* http://www.codeproject.com/string/dmetaphone1.asp
* and the original article describing it can be found at
* http://drdobbs.com/184401251
*
* For PostgreSQL we provide 2 functions - one for the primary and one for
* the alternate. That way the functions are pure text->text mappings that
* are useful in functional indexes. These are 'dmetaphone' for the
* primary and 'dmetaphone_alt' for the alternate.
*
* Assuming that dmetaphone.so is in $libdir, the SQL to set up the
* functions looks like this:
*
* CREATE FUNCTION dmetaphone (text) RETURNS text
* LANGUAGE C IMMUTABLE STRICT
* AS '$libdir/dmetaphone', 'dmetaphone';
*
* CREATE FUNCTION dmetaphone_alt (text) RETURNS text
* LANGUAGE C IMMUTABLE STRICT
* AS '$libdir/dmetaphone', 'dmetaphone_alt';
*
* Note that you have to declare the functions IMMUTABLE if you want to
* use them in functional indexes, and you have to declare them as STRICT
* as they do not check for NULL input, and will segfault if given NULL input.
* (See below for alternative ) Declaring them as STRICT means PostgreSQL
* will never call them with NULL, but instead assume the result is NULL,
* which is what we (I) want.
*
* Alternatively, compile with -DDMETAPHONE_NOSTRICT and the functions
* will detect NULL input and return NULL. The you don't have to declare them
* as STRICT.
*
* There is a small inefficiency here - each function call actually computes
* both the primary and the alternate and then throws away the one it doesn't
* need. That's the way the perl module was written, because perl can handle
* a list return more easily than we can in PostgreSQL. The result has been
* fast enough for my needs, but it could maybe be optimized a bit to remove
* that behaviour.
*
*/
/***************************** COPYRIGHT NOTICES ***********************
Most of this code is directly from the Text::DoubleMetaphone perl module
version 0.05 available from http://www.cpan.org.
It bears this copyright notice:
Copyright 2000, Maurice Aubrey <maurice@hevanet.com>.
All rights reserved.
This code is based heavily on the C++ implementation by
Lawrence Philips and incorporates several bug fixes courtesy
of Kevin Atkinson <kevina@users.sourceforge.net>.
This module is free software; you may redistribute it and/or
modify it under the same terms as Perl itself.
The remaining code is authored by Andrew Dunstan <amdunstan@ncshp.org> and
<andrew@dunslane.net> and is covered this copyright:
Copyright 2003, North Carolina State Highway Patrol.
All rights reserved.
Permission to use, copy, modify, and distribute this software and its
documentation for any purpose, without fee, and without a written agreement
is hereby granted, provided that the above copyright notice and this
paragraph and the following two paragraphs appear in all copies.
IN NO EVENT SHALL THE NORTH CAROLINA STATE HIGHWAY PATROL BE LIABLE TO ANY
PARTY FOR DIRECT, INDIRECT, SPECIAL, INCIDENTAL, OR CONSEQUENTIAL DAMAGES,
INCLUDING LOST PROFITS, ARISING OUT OF THE USE OF THIS SOFTWARE AND ITS
DOCUMENTATION, EVEN IF THE NORTH CAROLINA STATE HIGHWAY PATROL HAS BEEN
ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
THE NORTH CAROLINA STATE HIGHWAY PATROL SPECIFICALLY DISCLAIMS ANY
WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. THE SOFTWARE PROVIDED
HEREUNDER IS ON AN "AS IS" BASIS, AND THE NORTH CAROLINA STATE HIGHWAY PATROL
HAS NO OBLIGATIONS TO PROVIDE MAINTENANCE, SUPPORT, UPDATES, ENHANCEMENTS, OR
MODIFICATIONS.
***********************************************************************/
/* include these first, according to the docs */
#ifndef DMETAPHONE_MAIN
#include "postgres.h"
#include "utils/builtins.h"
/* turn off assertions for embedded function */
#define NDEBUG
#endif
#include <stdio.h>
#include <ctype.h>
#include <stdlib.h>
#include <string.h>
#include <stdarg.h>
#include <assert.h>
/* prototype for the main function we got from the perl module */
static void DoubleMetaphone(char *, char **);
#ifndef DMETAPHONE_MAIN
/*
* The PostgreSQL visible dmetaphone function.
*/
PG_FUNCTION_INFO_V1(dmetaphone);
Datum
dmetaphone(PG_FUNCTION_ARGS)
{
text *arg;
char *aptr,
*codes[2],
*code;
#ifdef DMETAPHONE_NOSTRICT
if (PG_ARGISNULL(0))
PG_RETURN_NULL();
#endif
arg = PG_GETARG_TEXT_P(0);
aptr = text_to_cstring(arg);
DoubleMetaphone(aptr, codes);
code = codes[0];
if (!code)
code = "";
PG_RETURN_TEXT_P(cstring_to_text(code));
}
/*
* The PostgreSQL visible dmetaphone_alt function.
*/
PG_FUNCTION_INFO_V1(dmetaphone_alt);
Datum
dmetaphone_alt(PG_FUNCTION_ARGS)
{
text *arg;
char *aptr,
*codes[2],
*code;
#ifdef DMETAPHONE_NOSTRICT
if (PG_ARGISNULL(0))
PG_RETURN_NULL();
#endif
arg = PG_GETARG_TEXT_P(0);
aptr = text_to_cstring(arg);
DoubleMetaphone(aptr, codes);
code = codes[1];
if (!code)
code = "";
PG_RETURN_TEXT_P(cstring_to_text(code));
}
/* here is where we start the code imported from the perl module */
/* all memory handling is done with these macros */
#define META_MALLOC(v,n,t) \
(v = (t*)palloc(((n)*sizeof(t))))
#define META_REALLOC(v,n,t) \
(v = (t*)repalloc((v),((n)*sizeof(t))))
/*
* Don't do pfree - it seems to cause a segv sometimes - which might have just
* been caused by reloading the module in development.
* So we rely on context cleanup - Tom Lane says pfree shouldn't be necessary
* in a case like this.
*/
#define META_FREE(x) /* pfree((x)) */
#else /* not defined DMETAPHONE_MAIN */
/* use the standard malloc library when not running in PostgreSQL */
#define META_MALLOC(v,n,t) \
(v = (t*)malloc(((n)*sizeof(t))))
#define META_REALLOC(v,n,t) \
(v = (t*)realloc((v),((n)*sizeof(t))))
#define META_FREE(x) free((x))
#endif /* defined DMETAPHONE_MAIN */
/* this typedef was originally in the perl module's .h file */
typedef struct
{
char *str;
int length;
int bufsize;
int free_string_on_destroy;
}
metastring;
/*
* remaining perl module funcs unchanged except for declaring them static
* and reformatting to PostgreSQL indentation and to fit in 80 cols.
*
*/
static metastring *
NewMetaString(char *init_str)
{
metastring *s;
char empty_string[] = "";
META_MALLOC(s, 1, metastring);
assert(s != NULL);
if (init_str == NULL)
init_str = empty_string;
s->length = strlen(init_str);
/* preallocate a bit more for potential growth */
s->bufsize = s->length + 7;
META_MALLOC(s->str, s->bufsize, char);
assert(s->str != NULL);
strncpy(s->str, init_str, s->length + 1);
s->free_string_on_destroy = 1;
return s;
}
static void
DestroyMetaString(metastring *s)
{
if (s == NULL)
return;
if (s->free_string_on_destroy && (s->str != NULL))
META_FREE(s->str);
META_FREE(s);
}
static void
IncreaseBuffer(metastring *s, int chars_needed)
{
META_REALLOC(s->str, (s->bufsize + chars_needed + 10), char);
assert(s->str != NULL);
s->bufsize = s->bufsize + chars_needed + 10;
}
static void
MakeUpper(metastring *s)
{
char *i;
for (i = s->str; *i; i++)
*i = toupper((unsigned char) *i);
}
static int
IsVowel(metastring *s, int pos)
{
char c;
if ((pos < 0) || (pos >= s->length))
return 0;
c = *(s->str + pos);
if ((c == 'A') || (c == 'E') || (c == 'I') || (c == 'O') ||
(c == 'U') || (c == 'Y'))
return 1;
return 0;
}
static int
SlavoGermanic(metastring *s)
{
if ((char *) strstr(s->str, "W"))
return 1;
else if ((char *) strstr(s->str, "K"))
return 1;
else if ((char *) strstr(s->str, "CZ"))
return 1;
else if ((char *) strstr(s->str, "WITZ"))
return 1;
else
return 0;
}
static char
GetAt(metastring *s, int pos)
{
if ((pos < 0) || (pos >= s->length))
return '\0';
return ((char) *(s->str + pos));
}
static void
SetAt(metastring *s, int pos, char c)
{
if ((pos < 0) || (pos >= s->length))
return;
*(s->str + pos) = c;
}
/*
Caveats: the START value is 0 based
*/
static int
StringAt(metastring *s, int start, int length,...)
{
char *test;
char *pos;
va_list ap;
if ((start < 0) || (start >= s->length))
return 0;
pos = (s->str + start);
va_start(ap, length);
do
{
test = va_arg(ap, char *);
if (*test && (strncmp(pos, test, length) == 0))
return 1;
}
while (strcmp(test, "") != 0);
va_end(ap);
return 0;
}
static void
MetaphAdd(metastring *s, char *new_str)
{
int add_length;
if (new_str == NULL)
return;
add_length = strlen(new_str);
if ((s->length + add_length) > (s->bufsize - 1))
IncreaseBuffer(s, add_length);
strcat(s->str, new_str);
s->length += add_length;
}
static void
DoubleMetaphone(char *str, char **codes)
{
int length;
metastring *original;
metastring *primary;
metastring *secondary;
int current;
int last;
current = 0;
/* we need the real length and last prior to padding */
length = strlen(str);
last = length - 1;
original = NewMetaString(str);
/* Pad original so we can index beyond end */
MetaphAdd(original, " ");
primary = NewMetaString("");
secondary = NewMetaString("");
primary->free_string_on_destroy = 0;
secondary->free_string_on_destroy = 0;
MakeUpper(original);
/* skip these when at start of word */
if (StringAt(original, 0, 2, "GN", "KN", "PN", "WR", "PS", ""))
current += 1;
/* Initial 'X' is pronounced 'Z' e.g. 'Xavier' */
if (GetAt(original, 0) == 'X')
{
MetaphAdd(primary, "S"); /* 'Z' maps to 'S' */
MetaphAdd(secondary, "S");
current += 1;
}
/* main loop */
while ((primary->length < 4) || (secondary->length < 4))
{
if (current >= length)
break;
switch (GetAt(original, current))
{
case 'A':
case 'E':
case 'I':
case 'O':
case 'U':
case 'Y':
if (current == 0)
{
/* all init vowels now map to 'A' */
MetaphAdd(primary, "A");
MetaphAdd(secondary, "A");
}
current += 1;
break;
case 'B':
/* "-mb", e.g", "dumb", already skipped over... */
MetaphAdd(primary, "P");
MetaphAdd(secondary, "P");
if (GetAt(original, current + 1) == 'B')
current += 2;
else
current += 1;
break;
case '\xc7': /* C with cedilla */
MetaphAdd(primary, "S");
MetaphAdd(secondary, "S");
current += 1;
break;
case 'C':
/* various germanic */
if ((current > 1)
&& !IsVowel(original, current - 2)
&& StringAt(original, (current - 1), 3, "ACH", "")
&& ((GetAt(original, current + 2) != 'I')
&& ((GetAt(original, current + 2) != 'E')
|| StringAt(original, (current - 2), 6, "BACHER",
"MACHER", ""))))
{
MetaphAdd(primary, "K");
MetaphAdd(secondary, "K");
current += 2;
break;
}
/* special case 'caesar' */
if ((current == 0)
&& StringAt(original, current, 6, "CAESAR", ""))
{
MetaphAdd(primary, "S");
MetaphAdd(secondary, "S");
current += 2;
break;
}
/* italian 'chianti' */
if (StringAt(original, current, 4, "CHIA", ""))
{
MetaphAdd(primary, "K");
MetaphAdd(secondary, "K");
current += 2;
break;
}
if (StringAt(original, current, 2, "CH", ""))
{
/* find 'michael' */
if ((current > 0)
&& StringAt(original, current, 4, "CHAE", ""))
{
MetaphAdd(primary, "K");
MetaphAdd(secondary, "X");
current += 2;
break;
}
/* greek roots e.g. 'chemistry', 'chorus' */
if ((current == 0)
&& (StringAt(original, (current + 1), 5,
"HARAC", "HARIS", "")
|| StringAt(original, (current + 1), 3, "HOR",
"HYM", "HIA", "HEM", ""))
&& !StringAt(original, 0, 5, "CHORE", ""))
{
MetaphAdd(primary, "K");
MetaphAdd(secondary, "K");
current += 2;
break;
}
/* germanic, greek, or otherwise 'ch' for 'kh' sound */
if (
(StringAt(original, 0, 4, "VAN ", "VON ", "")
|| StringAt(original, 0, 3, "SCH", ""))
/* 'architect but not 'arch', 'orchestra', 'orchid' */
|| StringAt(original, (current - 2), 6, "ORCHES",
"ARCHIT", "ORCHID", "")
|| StringAt(original, (current + 2), 1, "T", "S",
"")
|| ((StringAt(original, (current - 1), 1,
"A", "O", "U", "E", "")
|| (current == 0))
/*
* e.g., 'wachtler', 'wechsler', but not 'tichner'
*/
&& StringAt(original, (current + 2), 1, "L", "R",
"N", "M", "B", "H", "F", "V", "W",
" ", "")))
{
MetaphAdd(primary, "K");
MetaphAdd(secondary, "K");
}
else
{
if (current > 0)
{
if (StringAt(original, 0, 2, "MC", ""))
{
/* e.g., "McHugh" */
MetaphAdd(primary, "K");
MetaphAdd(secondary, "K");
}
else
{
MetaphAdd(primary, "X");
MetaphAdd(secondary, "K");
}
}
else
{
MetaphAdd(primary, "X");
MetaphAdd(secondary, "X");
}
}
current += 2;
break;
}
/* e.g, 'czerny' */
if (StringAt(original, current, 2, "CZ", "")
&& !StringAt(original, (current - 2), 4, "WICZ", ""))
{
MetaphAdd(primary, "S");
MetaphAdd(secondary, "X");
current += 2;
break;
}
/* e.g., 'focaccia' */
if (StringAt(original, (current + 1), 3, "CIA", ""))
{
MetaphAdd(primary, "X");
MetaphAdd(secondary, "X");
current += 3;
break;
}
/* double 'C', but not if e.g. 'McClellan' */
if (StringAt(original, current, 2, "CC", "")
&& !((current == 1) && (GetAt(original, 0) == 'M')))
{
/* 'bellocchio' but not 'bacchus' */
if (StringAt(original, (current + 2), 1, "I", "E", "H", "")
&& !StringAt(original, (current + 2), 2, "HU", ""))
{
/* 'accident', 'accede' 'succeed' */
if (
((current == 1)
&& (GetAt(original, current - 1) == 'A'))
|| StringAt(original, (current - 1), 5, "UCCEE",
"UCCES", ""))
{
MetaphAdd(primary, "KS");
MetaphAdd(secondary, "KS");
/* 'bacci', 'bertucci', other italian */
}
else
{
MetaphAdd(primary, "X");
MetaphAdd(secondary, "X");
}
current += 3;
break;
}
else
{ /* Pierce's rule */
MetaphAdd(primary, "K");
MetaphAdd(secondary, "K");
current += 2;
break;
}
}
if (StringAt(original, current, 2, "CK", "CG", "CQ", ""))
{
MetaphAdd(primary, "K");
MetaphAdd(secondary, "K");
current += 2;
break;
}
if (StringAt(original, current, 2, "CI", "CE", "CY", ""))
{
/* italian vs. english */
if (StringAt
(original, current, 3, "CIO", "CIE", "CIA", ""))
{
MetaphAdd(primary, "S");
MetaphAdd(secondary, "X");
}
else
{
MetaphAdd(primary, "S");
MetaphAdd(secondary, "S");
}
current += 2;
break;
}
/* else */
MetaphAdd(primary, "K");
MetaphAdd(secondary, "K");
/* name sent in 'mac caffrey', 'mac gregor */
if (StringAt(original, (current + 1), 2, " C", " Q", " G", ""))
current += 3;
else if (StringAt(original, (current + 1), 1, "C", "K", "Q", "")
&& !StringAt(original, (current + 1), 2,
"CE", "CI", ""))
current += 2;
else
current += 1;
break;
case 'D':
if (StringAt(original, current, 2, "DG", ""))
{
if (StringAt(original, (current + 2), 1,
"I", "E", "Y", ""))
{
/* e.g. 'edge' */
MetaphAdd(primary, "J");
MetaphAdd(secondary, "J");
current += 3;
break;
}
else
{
/* e.g. 'edgar' */
MetaphAdd(primary, "TK");
MetaphAdd(secondary, "TK");
current += 2;
break;
}
}
if (StringAt(original, current, 2, "DT", "DD", ""))
{
MetaphAdd(primary, "T");
MetaphAdd(secondary, "T");
current += 2;
break;
}
/* else */
MetaphAdd(primary, "T");
MetaphAdd(secondary, "T");
current += 1;
break;
case 'F':
if (GetAt(original, current + 1) == 'F')
current += 2;
else
current += 1;
MetaphAdd(primary, "F");
MetaphAdd(secondary, "F");
break;
case 'G':
if (GetAt(original, current + 1) == 'H')
{
if ((current > 0) && !IsVowel(original, current - 1))
{
MetaphAdd(primary, "K");
MetaphAdd(secondary, "K");
current += 2;
break;
}
if (current < 3)
{
/* 'ghislane', ghiradelli */
if (current == 0)
{
if (GetAt(original, current + 2) == 'I')
{
MetaphAdd(primary, "J");
MetaphAdd(secondary, "J");
}
else
{
MetaphAdd(primary, "K");
MetaphAdd(secondary, "K");
}
current += 2;
break;
}
}
/*
* Parker's rule (with some further refinements) - e.g.,
* 'hugh'
*/
if (
((current > 1)
&& StringAt(original, (current - 2), 1,
"B", "H", "D", ""))
/* e.g., 'bough' */
|| ((current > 2)
&& StringAt(original, (current - 3), 1,
"B", "H", "D", ""))
/* e.g., 'broughton' */
|| ((current > 3)
&& StringAt(original, (current - 4), 1,
"B", "H", "")))
{
current += 2;
break;
}
else
{
/*
* e.g., 'laugh', 'McLaughlin', 'cough', 'gough',
* 'rough', 'tough'
*/
if ((current > 2)
&& (GetAt(original, current - 1) == 'U')
&& StringAt(original, (current - 3), 1, "C",
"G", "L", "R", "T", ""))
{
MetaphAdd(primary, "F");
MetaphAdd(secondary, "F");
}
else if ((current > 0)
&& GetAt(original, current - 1) != 'I')
{
MetaphAdd(primary, "K");
MetaphAdd(secondary, "K");
}
current += 2;
break;
}
}
if (GetAt(original, current + 1) == 'N')
{
if ((current == 1) && IsVowel(original, 0)
&& !SlavoGermanic(original))
{
MetaphAdd(primary, "KN");
MetaphAdd(secondary, "N");
}
else
/* not e.g. 'cagney' */
if (!StringAt(original, (current + 2), 2, "EY", "")
&& (GetAt(original, current + 1) != 'Y')
&& !SlavoGermanic(original))
{
MetaphAdd(primary, "N");
MetaphAdd(secondary, "KN");
}
else
{
MetaphAdd(primary, "KN");
MetaphAdd(secondary, "KN");
}
current += 2;
break;
}
/* 'tagliaro' */
if (StringAt(original, (current + 1), 2, "LI", "")
&& !SlavoGermanic(original))
{
MetaphAdd(primary, "KL");
MetaphAdd(secondary, "L");
current += 2;
break;
}
/* -ges-,-gep-,-gel-, -gie- at beginning */
if ((current == 0)
&& ((GetAt(original, current + 1) == 'Y')
|| StringAt(original, (current + 1), 2, "ES", "EP",
"EB", "EL", "EY", "IB", "IL", "IN", "IE",
"EI", "ER", "")))
{
MetaphAdd(primary, "K");
MetaphAdd(secondary, "J");
current += 2;
break;
}
/* -ger-, -gy- */
if (
(StringAt(original, (current + 1), 2, "ER", "")
|| (GetAt(original, current + 1) == 'Y'))
&& !StringAt(original, 0, 6,
"DANGER", "RANGER", "MANGER", "")
&& !StringAt(original, (current - 1), 1, "E", "I", "")
&& !StringAt(original, (current - 1), 3, "RGY", "OGY",
""))
{
MetaphAdd(primary, "K");
MetaphAdd(secondary, "J");
current += 2;
break;
}
/* italian e.g, 'biaggi' */
if (StringAt(original, (current + 1), 1, "E", "I", "Y", "")
|| StringAt(original, (current - 1), 4,
"AGGI", "OGGI", ""))
{
/* obvious germanic */
if (
(StringAt(original, 0, 4, "VAN ", "VON ", "")
|| StringAt(original, 0, 3, "SCH", ""))
|| StringAt(original, (current + 1), 2, "ET", ""))
{
MetaphAdd(primary, "K");
MetaphAdd(secondary, "K");
}
else
{
/* always soft if french ending */
if (StringAt
(original, (current + 1), 4, "IER ", ""))
{
MetaphAdd(primary, "J");
MetaphAdd(secondary, "J");
}
else
{
MetaphAdd(primary, "J");
MetaphAdd(secondary, "K");
}
}
current += 2;
break;
}
if (GetAt(original, current + 1) == 'G')
current += 2;
else
current += 1;
MetaphAdd(primary, "K");
MetaphAdd(secondary, "K");
break;
case 'H':
/* only keep if first & before vowel or btw. 2 vowels */
if (((current == 0) || IsVowel(original, current - 1))
&& IsVowel(original, current + 1))
{
MetaphAdd(primary, "H");
MetaphAdd(secondary, "H");
current += 2;
}
else
/* also takes care of 'HH' */
current += 1;
break;
case 'J':
/* obvious spanish, 'jose', 'san jacinto' */
if (StringAt(original, current, 4, "JOSE", "")
|| StringAt(original, 0, 4, "SAN ", ""))
{
if (((current == 0)
&& (GetAt(original, current + 4) == ' '))
|| StringAt(original, 0, 4, "SAN ", ""))
{
MetaphAdd(primary, "H");
MetaphAdd(secondary, "H");
}
else
{
MetaphAdd(primary, "J");
MetaphAdd(secondary, "H");
}
current += 1;
break;
}
if ((current == 0)
&& !StringAt(original, current, 4, "JOSE", ""))
{
MetaphAdd(primary, "J"); /* Yankelovich/Jankelowicz */
MetaphAdd(secondary, "A");
}
else
{
/* spanish pron. of e.g. 'bajador' */
if (IsVowel(original, current - 1)
&& !SlavoGermanic(original)
&& ((GetAt(original, current + 1) == 'A')
|| (GetAt(original, current + 1) == 'O')))
{
MetaphAdd(primary, "J");
MetaphAdd(secondary, "H");
}
else
{
if (current == last)
{
MetaphAdd(primary, "J");
MetaphAdd(secondary, "");
}
else
{
if (!StringAt(original, (current + 1), 1, "L", "T",
"K", "S", "N", "M", "B", "Z", "")
&& !StringAt(original, (current - 1), 1,
"S", "K", "L", ""))
{
MetaphAdd(primary, "J");
MetaphAdd(secondary, "J");
}
}
}
}
if (GetAt(original, current + 1) == 'J') /* it could happen! */
current += 2;
else
current += 1;
break;
case 'K':
if (GetAt(original, current + 1) == 'K')
current += 2;
else
current += 1;
MetaphAdd(primary, "K");
MetaphAdd(secondary, "K");
break;
case 'L':
if (GetAt(original, current + 1) == 'L')
{
/* spanish e.g. 'cabrillo', 'gallegos' */
if (((current == (length - 3))
&& StringAt(original, (current - 1), 4, "ILLO",
"ILLA", "ALLE", ""))
|| ((StringAt(original, (last - 1), 2, "AS", "OS", "")
|| StringAt(original, last, 1, "A", "O", ""))
&& StringAt(original, (current - 1), 4,
"ALLE", "")))
{
MetaphAdd(primary, "L");
MetaphAdd(secondary, "");
current += 2;
break;
}
current += 2;
}
else
current += 1;
MetaphAdd(primary, "L");
MetaphAdd(secondary, "L");
break;
case 'M':
if ((StringAt(original, (current - 1), 3, "UMB", "")
&& (((current + 1) == last)
|| StringAt(original, (current + 2), 2, "ER", "")))
/* 'dumb','thumb' */
|| (GetAt(original, current + 1) == 'M'))
current += 2;
else
current += 1;
MetaphAdd(primary, "M");
MetaphAdd(secondary, "M");
break;
case 'N':
if (GetAt(original, current + 1) == 'N')
current += 2;
else
current += 1;
MetaphAdd(primary, "N");
MetaphAdd(secondary, "N");
break;
case '\xd1': /* N with tilde */
current += 1;
MetaphAdd(primary, "N");
MetaphAdd(secondary, "N");
break;
case 'P':
if (GetAt(original, current + 1) == 'H')
{
MetaphAdd(primary, "F");
MetaphAdd(secondary, "F");
current += 2;
break;
}
/* also account for "campbell", "raspberry" */
if (StringAt(original, (current + 1), 1, "P", "B", ""))
current += 2;
else
current += 1;
MetaphAdd(primary, "P");
MetaphAdd(secondary, "P");
break;
case 'Q':
if (GetAt(original, current + 1) == 'Q')
current += 2;
else
current += 1;
MetaphAdd(primary, "K");
MetaphAdd(secondary, "K");
break;
case 'R':
/* french e.g. 'rogier', but exclude 'hochmeier' */
if ((current == last)
&& !SlavoGermanic(original)
&& StringAt(original, (current - 2), 2, "IE", "")
&& !StringAt(original, (current - 4), 2, "ME", "MA", ""))
{
MetaphAdd(primary, "");
MetaphAdd(secondary, "R");
}
else
{
MetaphAdd(primary, "R");
MetaphAdd(secondary, "R");
}
if (GetAt(original, current + 1) == 'R')
current += 2;
else
current += 1;
break;
case 'S':
/* special cases 'island', 'isle', 'carlisle', 'carlysle' */
if (StringAt(original, (current - 1), 3, "ISL", "YSL", ""))
{
current += 1;
break;
}
/* special case 'sugar-' */
if ((current == 0)
&& StringAt(original, current, 5, "SUGAR", ""))
{
MetaphAdd(primary, "X");
MetaphAdd(secondary, "S");
current += 1;
break;
}
if (StringAt(original, current, 2, "SH", ""))
{
/* germanic */
if (StringAt
(original, (current + 1), 4, "HEIM", "HOEK", "HOLM",
"HOLZ", ""))
{
MetaphAdd(primary, "S");
MetaphAdd(secondary, "S");
}
else
{
MetaphAdd(primary, "X");
MetaphAdd(secondary, "X");
}
current += 2;
break;
}
/* italian & armenian */
if (StringAt(original, current, 3, "SIO", "SIA", "")
|| StringAt(original, current, 4, "SIAN", ""))
{
if (!SlavoGermanic(original))
{
MetaphAdd(primary, "S");
MetaphAdd(secondary, "X");
}
else
{
MetaphAdd(primary, "S");
MetaphAdd(secondary, "S");
}
current += 3;
break;
}
/*
* german & anglicisations, e.g. 'smith' match 'schmidt',
* 'snider' match 'schneider' also, -sz- in slavic language
* although in hungarian it is pronounced 's'
*/
if (((current == 0)
&& StringAt(original, (current + 1), 1,
"M", "N", "L", "W", ""))
|| StringAt(original, (current + 1), 1, "Z", ""))
{
MetaphAdd(primary, "S");
MetaphAdd(secondary, "X");
if (StringAt(original, (current + 1), 1, "Z", ""))
current += 2;
else
current += 1;
break;
}
if (StringAt(original, current, 2, "SC", ""))
{
/* Schlesinger's rule */
if (GetAt(original, current + 2) == 'H')
{
/* dutch origin, e.g. 'school', 'schooner' */
if (StringAt(original, (current + 3), 2,
"OO", "ER", "EN",
"UY", "ED", "EM", ""))
{
/* 'schermerhorn', 'schenker' */
if (StringAt(original, (current + 3), 2,
"ER", "EN", ""))
{
MetaphAdd(primary, "X");
MetaphAdd(secondary, "SK");
}
else
{
MetaphAdd(primary, "SK");
MetaphAdd(secondary, "SK");
}
current += 3;
break;
}
else
{
if ((current == 0) && !IsVowel(original, 3)
&& (GetAt(original, 3) != 'W'))
{
MetaphAdd(primary, "X");
MetaphAdd(secondary, "S");
}
else
{
MetaphAdd(primary, "X");
MetaphAdd(secondary, "X");
}
current += 3;
break;
}
}
if (StringAt(original, (current + 2), 1,
"I", "E", "Y", ""))
{
MetaphAdd(primary, "S");
MetaphAdd(secondary, "S");
current += 3;
break;
}
/* else */
MetaphAdd(primary, "SK");
MetaphAdd(secondary, "SK");
current += 3;
break;
}
/* french e.g. 'resnais', 'artois' */
if ((current == last)
&& StringAt(original, (current - 2), 2, "AI", "OI", ""))
{
MetaphAdd(primary, "");
MetaphAdd(secondary, "S");
}
else
{
MetaphAdd(primary, "S");
MetaphAdd(secondary, "S");
}
if (StringAt(original, (current + 1), 1, "S", "Z", ""))
current += 2;
else
current += 1;
break;
case 'T':
if (StringAt(original, current, 4, "TION", ""))
{
MetaphAdd(primary, "X");
MetaphAdd(secondary, "X");
current += 3;
break;
}
if (StringAt(original, current, 3, "TIA", "TCH", ""))
{
MetaphAdd(primary, "X");
MetaphAdd(secondary, "X");
current += 3;
break;
}
if (StringAt(original, current, 2, "TH", "")
|| StringAt(original, current, 3, "TTH", ""))
{
/* special case 'thomas', 'thames' or germanic */
if (StringAt(original, (current + 2), 2, "OM", "AM", "")
|| StringAt(original, 0, 4, "VAN ", "VON ", "")
|| StringAt(original, 0, 3, "SCH", ""))
{
MetaphAdd(primary, "T");
MetaphAdd(secondary, "T");
}
else
{
MetaphAdd(primary, "0");
MetaphAdd(secondary, "T");
}
current += 2;
break;
}
if (StringAt(original, (current + 1), 1, "T", "D", ""))
current += 2;
else
current += 1;
MetaphAdd(primary, "T");
MetaphAdd(secondary, "T");
break;
case 'V':
if (GetAt(original, current + 1) == 'V')
current += 2;
else
current += 1;
MetaphAdd(primary, "F");
MetaphAdd(secondary, "F");
break;
case 'W':
/* can also be in middle of word */
if (StringAt(original, current, 2, "WR", ""))
{
MetaphAdd(primary, "R");
MetaphAdd(secondary, "R");
current += 2;
break;
}
if ((current == 0)
&& (IsVowel(original, current + 1)
|| StringAt(original, current, 2, "WH", "")))
{
/* Wasserman should match Vasserman */
if (IsVowel(original, current + 1))
{
MetaphAdd(primary, "A");
MetaphAdd(secondary, "F");
}
else
{
/* need Uomo to match Womo */
MetaphAdd(primary, "A");
MetaphAdd(secondary, "A");
}
}
/* Arnow should match Arnoff */
if (((current == last) && IsVowel(original, current - 1))
|| StringAt(original, (current - 1), 5, "EWSKI", "EWSKY",
"OWSKI", "OWSKY", "")
|| StringAt(original, 0, 3, "SCH", ""))
{
MetaphAdd(primary, "");
MetaphAdd(secondary, "F");
current += 1;
break;
}
/* polish e.g. 'filipowicz' */
if (StringAt(original, current, 4, "WICZ", "WITZ", ""))
{
MetaphAdd(primary, "TS");
MetaphAdd(secondary, "FX");
current += 4;
break;
}
/* else skip it */
current += 1;
break;
case 'X':
/* french e.g. breaux */
if (!((current == last)
&& (StringAt(original, (current - 3), 3,
"IAU", "EAU", "")
|| StringAt(original, (current - 2), 2,
"AU", "OU", ""))))
{
MetaphAdd(primary, "KS");
MetaphAdd(secondary, "KS");
}
if (StringAt(original, (current + 1), 1, "C", "X", ""))
current += 2;
else
current += 1;
break;
case 'Z':
/* chinese pinyin e.g. 'zhao' */
if (GetAt(original, current + 1) == 'H')
{
MetaphAdd(primary, "J");
MetaphAdd(secondary, "J");
current += 2;
break;
}
else if (StringAt(original, (current + 1), 2,
"ZO", "ZI", "ZA", "")
|| (SlavoGermanic(original)
&& ((current > 0)
&& GetAt(original, current - 1) != 'T')))
{
MetaphAdd(primary, "S");
MetaphAdd(secondary, "TS");
}
else
{
MetaphAdd(primary, "S");
MetaphAdd(secondary, "S");
}
if (GetAt(original, current + 1) == 'Z')
current += 2;
else
current += 1;
break;
default:
current += 1;
}
/*
* printf("PRIMARY: %s\n", primary->str); printf("SECONDARY: %s\n",
* secondary->str);
*/
}
if (primary->length > 4)
SetAt(primary, 4, '\0');
if (secondary->length > 4)
SetAt(secondary, 4, '\0');
*codes = primary->str;
*++codes = secondary->str;
DestroyMetaString(original);
DestroyMetaString(primary);
DestroyMetaString(secondary);
}
#ifdef DMETAPHONE_MAIN
/* just for testing - not part of the perl code */
main(int argc, char **argv)
{
char *codes[2];
if (argc > 1)
{
DoubleMetaphone(argv[1], codes);
printf("%s|%s\n", codes[0], codes[1]);
}
}
#endif