aboutsummaryrefslogtreecommitdiffstats
path: root/e-util/ename/e-name-western.c
diff options
context:
space:
mode:
authornobody <nobody@localhost>2001-06-23 13:14:10 +0800
committernobody <nobody@localhost>2001-06-23 13:14:10 +0800
commitdd277dbef9a0108419304f19b25e3f5dc13bc11a (patch)
treea4eb5e010ba601de261dc3f9da4dd4b86fcc46fe /e-util/ename/e-name-western.c
parent72ca61d74b9942040638b628a6f1358e02d23d42 (diff)
downloadgsoc2013-evolution-dd277dbef9a0108419304f19b25e3f5dc13bc11a.tar
gsoc2013-evolution-dd277dbef9a0108419304f19b25e3f5dc13bc11a.tar.gz
gsoc2013-evolution-dd277dbef9a0108419304f19b25e3f5dc13bc11a.tar.bz2
gsoc2013-evolution-dd277dbef9a0108419304f19b25e3f5dc13bc11a.tar.lz
gsoc2013-evolution-dd277dbef9a0108419304f19b25e3f5dc13bc11a.tar.xz
gsoc2013-evolution-dd277dbef9a0108419304f19b25e3f5dc13bc11a.tar.zst
gsoc2013-evolution-dd277dbef9a0108419304f19b25e3f5dc13bc11a.zip
This commit was manufactured by cvs2svn to create tagPRE_WDFCHANGES
'PRE_WDFCHANGES'. svn path=/tags/PRE_WDFCHANGES/; revision=10430
Diffstat (limited to 'e-util/ename/e-name-western.c')
-rw-r--r--e-util/ename/e-name-western.c870
1 files changed, 0 insertions, 870 deletions
diff --git a/e-util/ename/e-name-western.c b/e-util/ename/e-name-western.c
deleted file mode 100644
index 3848e7249b..0000000000
--- a/e-util/ename/e-name-western.c
+++ /dev/null
@@ -1,870 +0,0 @@
-/*
- * A simple Western name parser.
- *
- * <Nat> Jamie, do you know anything about name parsing?
- * <jwz> Are you going down that rat hole? Bring a flashlight.
- *
- * Author:
- * Nat Friedman (nat@helixcode.com)
- *
- * Copyright 1999, Helix Code, Inc.
- */
-
-#include <ctype.h>
-#include <string.h>
-#include <glib.h>
-
-#include <ename/e-name-western.h>
-#include <ename/e-name-western-tables.h>
-
-typedef struct {
- int prefix_idx;
- int first_idx;
- int middle_idx;
- int nick_idx;
- int last_idx;
- int suffix_idx;
-} ENameWesternIdxs;
-
-static int
-e_name_western_str_count_words (char *str)
-{
- int word_count;
- char *p;
-
- word_count = 0;
-
- for (p = str; p != NULL; p = strchr (p, ' ')) {
- word_count ++;
- p ++;
- }
-
- return word_count;
-}
-
-static void
-e_name_western_cleanup_string (char **str)
-{
- char *newstr;
- char *p;
-
- if (*str == NULL)
- return;
-
- /* skip any spaces and commas at the start of the string */
- p = *str;
- while (isspace (*p) || *p == ',')
- p ++;
-
- /* make the copy we're going to return */
- newstr = g_strdup (p);
-
- if ( strlen(newstr) > 0) {
- /* now search from the back, skipping over any spaces and commas */
- p = newstr + strlen (newstr) - 1;
- while (isspace (*p) || *p == ',')
- p --;
- /* advance p to after the character that caused us to exit the
- previous loop, and end the string. */
- if ((! isspace (*p)) && *p != ',')
- p ++;
- *p = '\0';
- }
-
- g_free (*str);
- *str = newstr;
-}
-
-static char *
-e_name_western_get_words_at_idx (char *str, int idx, int num_words)
-{
- char *words;
- char *p;
- int word_count;
- int words_len;
-
- /*
- * Walk to the end of the words.
- */
- word_count = 0;
- p = str + idx;
- while (word_count < num_words && *p != '\0') {
- while (! isspace (*p) && *p != '\0')
- p ++;
-
- while (isspace (*p) && *p != '\0')
- p ++;
-
- word_count ++;
- }
-
- words_len = p - str - idx - 1;
-
- if (*p == '\0')
- words_len ++;
-
- words = g_malloc0 (1 + words_len);
- strncpy (words, str + idx, words_len);
-
- return words;
-}
-
-/*
- * What the fuck is wrong with glib's MAX macro.
- */
-static int
-e_name_western_max (const int a, const int b)
-{
- if (a > b)
- return a;
-
- return b;
-}
-
-static gboolean
-e_name_western_word_is_suffix (char *word)
-{
- int i;
-
- for (i = 0; e_name_western_sfx_table [i] != NULL; i ++) {
- if (g_strcasecmp (word, e_name_western_sfx_table [i]))
- continue;
-
- return TRUE;
- }
-
- return FALSE;
-}
-
-static char *
-e_name_western_get_one_prefix_at_str (char *str)
-{
- char *word;
- int i;
-
- /*
- * Check for prefixes from our table.
- */
- for (i = 0; e_name_western_pfx_table [i] != NULL; i ++) {
- int pfx_words;
- char *words;
-
- pfx_words = e_name_western_str_count_words (e_name_western_pfx_table [i]);
- words = e_name_western_get_words_at_idx (str, 0, pfx_words);
-
- if (! g_strcasecmp (words, e_name_western_pfx_table [i]))
- return words;
-
- g_free (words);
- }
-
- /*
- * Check for prefixes we don't know about. These are always a
- * sequence of more than one letters followed by a period.
- */
- word = e_name_western_get_words_at_idx (str, 0, 1);
-
- if (strlen (word) > 2 &&
- isalpha ((unsigned char) word [0]) &&
- isalpha ((unsigned char) word [1]) &&
- word [strlen (word) - 1] == '.')
- return word;
-
- g_free (word);
-
- return NULL;
-}
-
-static char *
-e_name_western_get_prefix_at_str (char *str)
-{
- char *pfx;
- char *pfx1;
- char *pfx2;
- char *p;
-
- /* Get the first prefix. */
- pfx1 = e_name_western_get_one_prefix_at_str (str);
-
- if (pfx1 == NULL)
- return NULL;
-
- /* Check for a second prefix. */
- p = str + strlen (pfx1);
- while (isspace (*p) && *p != '\0')
- p ++;
-
- pfx2 = e_name_western_get_one_prefix_at_str (p);
-
- if (pfx2 != NULL) {
- int pfx_len;
-
- pfx_len = (p + strlen (pfx2)) - str;
- pfx = g_malloc0 (pfx_len + 1);
- strncpy (pfx, str, pfx_len);
- } else {
- pfx = g_strdup (pfx1);
- }
-
- g_free (pfx1);
- g_free (pfx2);
-
- return pfx;
-}
-
-static void
-e_name_western_extract_prefix (ENameWestern *name, ENameWesternIdxs *idxs)
-{
- char *pfx;
-
- pfx = e_name_western_get_prefix_at_str (name->full);
-
- if (pfx == NULL)
- return;
-
- idxs->prefix_idx = 0;
- name->prefix = pfx;
-}
-
-static gboolean
-e_name_western_is_complex_last_beginning (char *word)
-{
- int i;
-
- for (i = 0; e_name_western_complex_last_table [i] != NULL; i ++) {
-
- if (! g_strcasecmp (
- word, e_name_western_complex_last_table [i]))
- return TRUE;
- }
-
- return FALSE;
-}
-
-static void
-e_name_western_extract_first (ENameWestern *name, ENameWesternIdxs *idxs)
-{
- /*
- * If there's a prefix, then the first name is right after it.
- */
- if (idxs->prefix_idx != -1) {
- int first_idx;
- char *p;
-
- first_idx = idxs->prefix_idx + strlen (name->prefix);
-
- /* Skip past white space. */
- p = name->full + first_idx;
- while (isspace (*p) && *p != '\0')
- p++;
-
- if (*p == '\0')
- return;
-
- idxs->first_idx = p - name->full;
- name->first = e_name_western_get_words_at_idx (
- name->full, idxs->first_idx, 1);
-
- } else {
-
- /*
- * Otherwise, the first name is probably the first string.
- */
- idxs->first_idx = 0;
- name->first = e_name_western_get_words_at_idx (
- name->full, idxs->first_idx, 1);
- }
-
- /*
- * Check that we didn't just assign the beginning of a
- * compound last name to the first name.
- */
- if (name->first != NULL) {
- if (e_name_western_is_complex_last_beginning (name->first)) {
- g_free (name->first);
- name->first = NULL;
- idxs->first_idx = -1;
- }
- }
-}
-
-static void
-e_name_western_extract_middle (ENameWestern *name, ENameWesternIdxs *idxs)
-{
- char *word;
- int middle_idx;
-
- /*
- * Middle names can only exist if you have a first name.
- */
- if (idxs->first_idx == -1)
- return;
-
- middle_idx = idxs->first_idx + strlen (name->first) + 1;
-
- if (middle_idx > strlen (name->full))
- return;
-
- /*
- * Search for the first space (or the terminating \0)
- */
- while (isspace (name->full [middle_idx]) &&
- name->full [middle_idx] != '\0')
- middle_idx ++;
-
- if (name->full [middle_idx] == '\0')
- return;
-
- /*
- * Skip past the nickname, if it's there.
- */
- if (name->full [middle_idx] == '\"') {
- if (idxs->nick_idx == -1)
- return;
-
- middle_idx = idxs->nick_idx + strlen (name->nick) + 1;
-
- while (isspace (name->full [middle_idx]) &&
- name->full [middle_idx] != '\0')
- middle_idx ++;
-
- if (name->full [middle_idx] == '\0')
- return;
- }
-
- /*
- * Make sure this isn't the beginning of a complex last name.
- */
- word = e_name_western_get_words_at_idx (name->full, middle_idx, 1);
- if (e_name_western_is_complex_last_beginning (word)) {
- g_free (word);
- return;
- }
-
- /*
- * Make sure this isn't a suffix.
- */
- e_name_western_cleanup_string (& word);
- if (e_name_western_word_is_suffix (word)) {
- g_free (word);
- return;
- }
-
- /*
- * Make sure we didn't just grab a cute nickname.
- */
- if (word [0] == '\"') {
- g_free (word);
- return;
- }
-
- idxs->middle_idx = middle_idx;
- name->middle = word;
-}
-
-static void
-e_name_western_extract_nickname (ENameWestern *name, ENameWesternIdxs *idxs)
-{
- int idx;
- int start_idx;
- char *str;
-
- if (idxs->first_idx == -1)
- return;
-
- if (idxs->middle_idx > idxs->first_idx)
- idx = idxs->middle_idx + strlen (name->middle);
- else
- idx = idxs->first_idx + strlen (name->first);
-
- while (name->full [idx] != '\"' && name->full [idx] != '\0')
- idx ++;
-
- if (name->full [idx] != '\"')
- return;
-
- start_idx = idx;
-
- /*
- * Advance to the next double quote.
- */
- idx ++;
-
- while (name->full [idx] != '\"' && name->full [idx] != '\0')
- idx ++;
-
- if (name->full [idx] == '\0')
- return;
-
- str = g_malloc0 (idx - start_idx + 2);
- strncpy (str, name->full + start_idx, idx - start_idx + 1);
-
- name->nick = str;
- idxs->nick_idx = start_idx;
-}
-
-static int
-e_name_western_last_get_max_idx (ENameWestern *name, ENameWesternIdxs *idxs)
-{
- int max_idx = -1;
-
- if (name->prefix != NULL)
- max_idx = e_name_western_max (
- max_idx, idxs->prefix_idx + strlen (name->prefix));
-
- if (name->first != NULL)
- max_idx = e_name_western_max (
- max_idx, idxs->first_idx + strlen (name->first));
-
- if (name->middle != NULL)
- max_idx = e_name_western_max (
- max_idx, idxs->middle_idx + strlen (name->middle));
-
- if (name->nick != NULL)
- max_idx = e_name_western_max (
- max_idx, idxs->nick_idx + strlen (name->nick));
-
- return max_idx;
-}
-
-static void
-e_name_western_extract_last (ENameWestern *name, ENameWesternIdxs *idxs)
-{
- char *word;
- int idx = -1;
-
- idx = e_name_western_last_get_max_idx (name, idxs);
-
- /*
- * In the case where there is no preceding name element, the
- * name is either just a first name ("Nat", "John"), is a
- * single-element name ("Cher", which we treat as a first
- * name), or is just a last name. The only time we can
- * differentiate a last name alone from a single-element name
- * or a first name alone is if it's a complex last name ("de
- * Icaza", "van Josephsen"). So if there is no preceding name
- * element, we check to see whether or not the first part of
- * the name is the beginning of a complex name. If it is,
- * we subsume the entire string. If we accidentally subsume
- * the suffix, this will get fixed in the fixup routine.
- */
- if (idx == -1) {
- word = e_name_western_get_words_at_idx (name->full, 0, 1);
- if (! e_name_western_is_complex_last_beginning (word)) {
- g_free (word);
- return;
- }
-
- name->last = g_strdup (name->full);
- idxs->last_idx = 0;
- return;
- }
-
- /* Skip past the white space. */
- while (isspace (name->full [idx]) && name->full [idx] != '\0')
- idx ++;
-
- if (name->full [idx] == '\0')
- return;
-
- word = e_name_western_get_words_at_idx (name->full, idx, 1);
- e_name_western_cleanup_string (& word);
- if (e_name_western_word_is_suffix (word)) {
- g_free (word);
- return;
- }
- g_free (word);
-
- /*
- * Subsume the rest of the string into the last name. If we
- * accidentally include the prefix, it will get fixed later.
- * This is the only way to handle things like "Miguel de Icaza
- * Amozorrutia" without dropping data and forcing the user
- * to retype it.
- */
- name->last = g_strdup (name->full + idx);
- idxs->last_idx = idx;
-}
-
-static char *
-e_name_western_get_preceding_word (char *str, int idx)
-{
- int word_len;
- char *word;
- char *p;
-
- p = str + idx;
-
- while (isspace (*p) && p > str)
- p --;
-
- while (! isspace (*p) && p > str)
- p --;
-
- if (isspace (*p))
- p ++;
-
- word_len = (str + idx) - p;
- word = g_malloc0 (word_len + 1);
- if (word_len > 0)
- strncpy (word, p, word_len);
-
- return word;
-}
-
-static char *
-e_name_western_get_suffix_at_str_end (char *str)
-{
- char *suffix;
- char *p;
-
- /*
- * Walk backwards till we reach the beginning of the
- * (potentially-comma-separated) list of suffixes.
- */
- p = str + strlen (str);
- while (1) {
- char *nextp;
- char *word;
-
- word = e_name_western_get_preceding_word (str, p - str);
- nextp = p - strlen (word) - 1;
-
- e_name_western_cleanup_string (& word);
-
- if (e_name_western_word_is_suffix (word)) {
- p = nextp;
- g_free (word);
- } else {
- g_free (word);
- break;
- }
- }
-
- if (p == (str + strlen (str)))
- return NULL;
-
- suffix = g_strdup (p);
- e_name_western_cleanup_string (& suffix);
-
- if (strlen (suffix) == 0) {
- g_free (suffix);
- return NULL;
- }
-
- return suffix;
-}
-
-static void
-e_name_western_extract_suffix (ENameWestern *name, ENameWesternIdxs *idxs)
-{
-
- name->suffix = e_name_western_get_suffix_at_str_end (name->full);
-
- if (name->suffix == NULL)
- return;
-
- idxs->suffix_idx = strlen (name->full) - strlen (name->suffix);
-}
-
-static gboolean
-e_name_western_detect_backwards (ENameWestern *name, ENameWesternIdxs *idxs)
-{
- char *comma;
- char *word;
-
- comma = strchr (name->full, ',');
-
- if (comma == NULL)
- return FALSE;
-
- /*
- * If there's a comma, we need to detect whether it's
- * separating the last name from the first or just separating
- * suffixes. So we grab the word which comes before the
- * comma and check if it's a suffix.
- */
- word = e_name_western_get_preceding_word (name->full, comma - name->full);
-
- if (e_name_western_word_is_suffix (word)) {
- g_free (word);
- return FALSE;
- }
-
- g_free (word);
- return TRUE;
-}
-
-static void
-e_name_western_reorder_asshole (ENameWestern *name, ENameWesternIdxs *idxs)
-{
- char *prefix;
- char *last;
- char *suffix;
- char *firstmidnick;
- char *newfull;
-
- char *comma;
- char *p;
-
- if (! e_name_western_detect_backwards (name, idxs))
- return;
-
- /*
- * Convert
- * <Prefix> <Last name>, <First name> <Middle[+nick] name> <Suffix>
- * to
- * <Prefix> <First name> <Middle[+nick] name> <Last name> <Suffix>
- */
-
- /*
- * Grab the prefix from the beginning.
- */
- prefix = e_name_western_get_prefix_at_str (name->full);
-
- /*
- * Everything from the end of the prefix to the comma is the
- * last name.
- */
- comma = strchr (name->full, ',');
- if (comma == NULL)
- return;
-
- p = name->full + (prefix == NULL ? 0 : strlen (prefix));
-
- while (isspace (*p) && *p != '\0')
- p ++;
-
- last = g_malloc0 (comma - p + 1);
- strncpy (last, p, comma - p);
-
- /*
- * Get the suffix off the end.
- */
- suffix = e_name_western_get_suffix_at_str_end (name->full);
-
- /*
- * Firstmidnick is everything from the comma to the beginning
- * of the suffix.
- */
- p = comma + 1;
-
- while (isspace (*p) && *p != '\0')
- p ++;
-
- if (suffix != NULL) {
- char *q;
-
- /*
- * Point q at the beginning of the suffix.
- */
- q = name->full + strlen (name->full) - strlen (suffix) - 1;
-
- /*
- * Walk backwards until we hit the space which
- * separates the suffix from firstmidnick.
- */
- while (! isspace (*q) && q > comma)
- q --;
-
- if ((q - p + 1) > 0) {
- firstmidnick = g_malloc0 (q - p + 1);
- strncpy (firstmidnick, p, q - p);
- } else
- firstmidnick = NULL;
- } else {
- firstmidnick = g_strdup (p);
- }
-
- /*
- * Create our new reordered version of the name.
- */
-#define NULLSTR(a) ((a) == NULL ? "" : (a))
- newfull = g_strdup_printf ("%s %s %s %s", NULLSTR (prefix), NULLSTR (firstmidnick),
- NULLSTR (last), NULLSTR (suffix));
- g_strstrip (newfull);
- g_free (name->full);
- name->full = newfull;
-
-
- g_free (prefix);
- g_free (firstmidnick);
- g_free (last);
- g_free (suffix);
-}
-
-static void
-e_name_western_zap_nil (char **str, int *idx)
-{
- if (*str == NULL)
- return;
-
- if (strlen (*str) != 0)
- return;
-
- *idx = -1;
- g_free (*str);
- *str = NULL;
-}
-
-static void
-e_name_western_fixup (ENameWestern *name, ENameWesternIdxs *idxs)
-{
- /*
- * The middle and last names cannot be the same.
- */
- if (idxs->middle_idx != -1 && idxs->middle_idx == idxs->last_idx) {
- idxs->middle_idx = -1;
- g_free (name->middle);
- name->middle = NULL;
- }
-
- /*
- * If we have a middle name and no last name, then we mistook
- * the last name for the middle name.
- */
- if (idxs->last_idx == -1 && idxs->middle_idx != -1) {
- idxs->last_idx = idxs->middle_idx;
- name->last = name->middle;
- name->middle = NULL;
- idxs->middle_idx = -1;
- }
-
- /*
- * Check to see if we accidentally included the suffix in the
- * last name.
- */
- if (idxs->suffix_idx != -1 && idxs->last_idx != -1 &&
- idxs->suffix_idx < (idxs->last_idx + strlen (name->last))) {
- char *sfx;
-
- sfx = name->last + (idxs->suffix_idx - idxs->last_idx);
- if (sfx != NULL) {
- char *newlast;
- char *p;
-
- p = sfx - 1;
- while (isspace (*p) && p > name->last)
- p --;
- p ++;
-
- newlast = g_malloc0 (p - name->last + 1);
- strncpy (newlast, name->last, p - name->last);
- g_free (name->last);
- name->last = newlast;
- }
- }
-
- /*
- * If we have a prefix and a first name, but no last name,
- * then we need to assign the first name to the last name.
- * This way we get things like "Mr Friedman" correctly.
- */
- if (idxs->first_idx != -1 && idxs->prefix_idx != -1 &&
- idxs->last_idx == -1) {
- name->last = name->first;
- idxs->last_idx = idxs->first_idx;
- idxs->first_idx = -1;
- name->first = NULL;
- }
-
- /*
- * Remove stray spaces and commas (although there don't seem
- * to be any in the test cases, they might show up later).
- */
- e_name_western_cleanup_string (& name->prefix);
- e_name_western_cleanup_string (& name->first);
- e_name_western_cleanup_string (& name->middle);
- e_name_western_cleanup_string (& name->nick);
- e_name_western_cleanup_string (& name->last);
- e_name_western_cleanup_string (& name->suffix);
-
- /*
- * Make zero-length strings just NULL.
- */
- e_name_western_zap_nil (& name->prefix, & idxs->prefix_idx);
- e_name_western_zap_nil (& name->first, & idxs->first_idx);
- e_name_western_zap_nil (& name->middle, & idxs->middle_idx);
- e_name_western_zap_nil (& name->nick, & idxs->nick_idx);
- e_name_western_zap_nil (& name->last, & idxs->last_idx);
- e_name_western_zap_nil (& name->suffix, & idxs->suffix_idx);
-}
-
-/**
- * e_name_western_western_parse_fullname:
- * @full_name: A string containing a Western name.
- *
- * Parses @full_name and returns an #ENameWestern object filled with
- * the component parts of the name.
- */
-ENameWestern *
-e_name_western_parse (const char *full_name)
-{
- ENameWesternIdxs *idxs;
- ENameWestern *wname;
-
- wname = g_new0 (ENameWestern, 1);
-
- wname->full = g_strdup (full_name);
-
- idxs = g_new0 (ENameWesternIdxs, 1);
-
- idxs->prefix_idx = -1;
- idxs->first_idx = -1;
- idxs->middle_idx = -1;
- idxs->nick_idx = -1;
- idxs->last_idx = -1;
- idxs->suffix_idx = -1;
-
- /*
- * An extremely simple algorithm.
- *
- * The goal here is to get it right 95% of the time for
- * Western names.
- *
- * First we check to see if this is an ass-backwards name
- * ("Prefix Last, First Middle Suffix"). These names really
- * suck (imagine "Dr von Johnson, Albert Roderick Jr"), so
- * we reorder them first and then parse them.
- *
- * Next, we grab the most obvious assignments for the various
- * parts of the name. Once this is done, we check for stupid
- * errors and fix them up.
- */
- e_name_western_reorder_asshole (wname, idxs);
-
- e_name_western_extract_prefix (wname, idxs);
- e_name_western_extract_first (wname, idxs);
- e_name_western_extract_nickname (wname, idxs);
- e_name_western_extract_middle (wname, idxs);
- e_name_western_extract_last (wname, idxs);
- e_name_western_extract_suffix (wname, idxs);
-
- e_name_western_fixup (wname, idxs);
-
- g_free (idxs);
-
- return wname;
-}
-
-/**
- * e_name_western_free:
- * @name: An ENameWestern object which needs to be freed.
- *
- * Deep-frees @name
- */
-void
-e_name_western_free (ENameWestern *w)
-{
-
- g_free (w->prefix);
- g_free (w->first);
- g_free (w->middle);
- g_free (w->nick);
- g_free (w->last);
- g_free (w->suffix);
-
- g_free (w->full);
-
- g_free (w);
-}