summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorRozhuk Ivan <rozhuk.im@gmail.com>2014-11-26 23:48:18 +0000
committerRozhuk Ivan <rozhuk.im@gmail.com>2014-11-26 23:48:18 +0000
commitf19ed36b66a807420c771a467a954114f027384c (patch)
tree1c61894875576ee9cae4b37d448a6a671585fa9d
parent08e744c5f087a8e693d160c527097a9f6a5bf712 (diff)
SpellChecker: code cleanup
git-svn-id: http://svn.miranda-ng.org/main/trunk@11110 1316c22d-e87f-b044-9b9b-93d7a3e3ba9c
-rw-r--r--plugins/SpellChecker/src/dictionary.cpp1929
-rw-r--r--plugins/SpellChecker/src/hunspell/csutil.cxx3
-rw-r--r--plugins/SpellChecker/src/hunspell/suggestmgr.cxx5
3 files changed, 970 insertions, 967 deletions
diff --git a/plugins/SpellChecker/src/dictionary.cpp b/plugins/SpellChecker/src/dictionary.cpp
index cd27a3fd78..0926273837 100644
--- a/plugins/SpellChecker/src/dictionary.cpp
+++ b/plugins/SpellChecker/src/dictionary.cpp
@@ -1,965 +1,964 @@
-/*
-Copyright (C) 2006-2010 Ricardo Pescuma Domenecci
-
-This is free software; you can redistribute it and/or
-modify it under the terms of the GNU Library General Public
-License as published by the Free Software Foundation; either
-version 2 of the License, or (at your option) any later version.
-
-This 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
-Library General Public License for more details.
-
-You should have received a copy of the GNU Library General Public
-License along with this file; see the file license.txt. If
-not, write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330,
-Boston, MA 02111-1307, USA.
-*/
-
-#include "commons.h"
-
-#define APPPATH _T("SOFTWARE\\Microsoft\\Windows\\CurrentVersion\\App Paths\\%s")
-#define MUICACHE _T("Software\\Microsoft\\Windows\\ShellNoRoam\\MUICache")
-
-
-// Additional languages that i could not find in Windows
-struct {
- TCHAR *language;
- TCHAR *localized_name;
-} aditionalLanguages[] = {
- { _T("tl_PH"), LPGENT("Tagalog (Philippines)") },
- { _T("de_frami_neu"), LPGENT("German (Germany)") }
-};
-
-struct {
- TCHAR *name;
- TCHAR *key;
-} otherHunspellApps[] = {
- { _T("Thunderbird"), _T("thunderbird.exe") },
- { _T("Firefox"), _T("firefox.exe") }
-};
-
-struct {
- char *name;
- UINT codepage;
-} codepages[] = {
- { "ISO8859-1", 28591 },
- { "UTF-7", CP_UTF7 },
- { "UTF-8", CP_UTF8 },
- { "UTF7", CP_UTF7 },
- { "UTF8", CP_UTF8 },
- { "ISO8859-2", 28592 },
- { "ISO8859-3", 28593 },
- { "ISO8859-4", 28594 },
- { "ISO8859-5", 28595 },
- { "ISO8859-6", 28596 },
- { "ISO8859-7", 28597 },
- { "ISO8859-8", 28598 },
- { "ISO8859-9", 28599 },
- { "ASMO-708", 708 },
- { "DOS-720", 720 },
- { "iso-8859-6", 28596 },
- { "arabic", 28596 },
- { "csISOLatinArabic", 28596 },
- { "ECMA-114", 28596 },
- { "ISO_8859-6", 28596 },
- { "ISO_8859-6:1987", 28596 },
- { "iso-ir-127", 28596 },
- { "x-mac-arabic", 10004 },
- { "windows-1256", 1256 },
- { "cp1256", 1256 },
- { "ibm775", 775 },
- { "CP500", 775 },
- { "iso-8859-4", 28594 },
- { "csISOLatin4", 28594 },
- { "ISO_8859-4", 28594 },
- { "ISO_8859-4:1988", 28594 },
- { "iso-ir-110", 28594 },
- { "l4", 28594 },
- { "latin4", 28594 },
- { "windows-1257", 1257 },
- { "ibm852", 852 },
- { "cp852", 852 },
- { "iso-8859-2", 28592 },
- { "csISOLatin2", 28592 },
- { "iso_8859-2", 28592 },
- { "iso_8859-2:1987", 28592 },
- { "iso-ir-101", 28592 },
- { "l2", 28592 },
- { "latin2", 28592 },
- { "x-mac-ce", 10029 },
- { "windows-1250", 1250 },
- { "x-cp1250", 1250 },
- { "EUC-CN", 51936 },
- { "x-euc-cn", 51936 },
- { "gb2312", 936 },
- { "chinese", 936 },
- { "CN-GB", 936 },
- { "csGB2312", 936 },
- { "csGB231280", 936 },
- { "csISO58GB231280", 936 },
- { "GB_2312-80", 936 },
- { "GB231280", 936 },
- { "GB2312-80", 936 },
- { "GBK", 936 },
- { "iso-ir-58", 936 },
- { "hz-gb-2312", 52936 },
- { "x-mac-chinesesimp", 10008 },
- { "big5", 950 },
- { "cn-big5", 950 },
- { "csbig5", 950 },
- { "x-x-big5", 950 },
- { "x-Chinese-CNS", 20000 },
- { "x-Chinese-Eten", 20002 },
- { "x-mac-chinesetrad", 10002 },
- { "cp866", 866 },
- { "ibm866", 866 },
- { "iso-8859-5", 28595 },
- { "csISOLatin5", 28595 },
- { "csISOLatinCyrillic", 28595 },
- { "cyrillic", 28595 },
- { "ISO_8859-5", 28595 },
- { "ISO_8859-5:1988", 28595 },
- { "iso-ir-144", 28595 },
- { "l5", 28595 },
- { "KOI8-R", 20866 },
- { "csKOI8R", 20866 },
- { "koi", 20866 },
- { "koi8", 20866 },
- { "koi8r", 20866 },
- { "KOI8-U", 21866 },
- { "koi8-ru", 21866 },
- { "x-mac-cyrillic", 10007 },
- { "windows-1251", 1251 },
- { "Win1251", 1251 },
- { "x-cp1251", 1251 },
- { "x-Europa", 29001 },
- { "x-IA5-German", 20106 },
- { "ibm737", 737 },
- { "iso-8859-7", 28597 },
- { "csISOLatinGreek", 28597 },
- { "ECMA-118", 28597 },
- { "ELOT_928", 28597 },
- { "greek", 28597 },
- { "greek8", 28597 },
- { "ISO_8859-7", 28597 },
- { "ISO_8859-7:1987", 28597 },
- { "iso-ir-126", 28597 },
- { "x-mac-greek", 10006 },
- { "windows-1253", 1253 },
- { "ibm869", 869 },
- { "DOS-862", 862 },
- { "iso-8859-8-i", 38598 },
- { "logical", 38598 },
- { "iso-8859-8", 28598 },
- { "csISOLatinHebrew", 28598 },
- { "hebrew", 28598 },
- { "ISO_8859-8", 28598 },
- { "ISO_8859-8:1988", 28598 },
- { "ISO-8859-8", 28598 },
- { "iso-ir-138", 28598 },
- { "visual", 28598 },
- { "x-mac-hebrew", 10005 },
- { "windows-1255", 1255 },
- { "ISO_8859-8-I", 1255 },
- { "ISO-8859-8", 1255 },
- { "x-EBCDIC-Arabic", 20420 },
- { "x-EBCDIC-CyrillicRussian", 20880 },
- { "x-EBCDIC-CyrillicSerbianBulgarian", 21025 },
- { "x-EBCDIC-DenmarkNorway", 20277 },
- { "x-ebcdic-denmarknorway-euro", 1142 },
- { "x-EBCDIC-FinlandSweden", 20278 },
- { "x-ebcdic-finlandsweden-euro", 1143 },
- { "X-EBCDIC-France", 1143 },
- { "X-EBCDIC-France", 1143 },
- { "x-ebcdic-france-euro", 1147 },
- { "x-EBCDIC-Germany", 20273 },
- { "x-ebcdic-germany-euro", 1141 },
- { "x-EBCDIC-GreekModern", 875 },
- { "x-EBCDIC-Greek", 20423 },
- { "x-EBCDIC-Hebrew", 20424 },
- { "x-EBCDIC-Icelandic", 20871 },
- { "x-ebcdic-icelandic-euro", 1149 },
- { "x-ebcdic-international-euro", 1148 },
- { "x-EBCDIC-Italy", 20280 },
- { "x-ebcdic-italy-euro", 1144 },
- { "x-EBCDIC-JapaneseAndKana", 50930 },
- { "x-EBCDIC-JapaneseAndJapaneseLatin", 50939 },
- { "x-EBCDIC-JapaneseAndUSCanada", 50931 },
- { "x-EBCDIC-JapaneseKatakana", 20290 },
- { "x-EBCDIC-KoreanAndKoreanExtended", 50933 },
- { "x-EBCDIC-KoreanExtended", 20833 },
- { "CP870", 870 },
- { "x-EBCDIC-SimplifiedChinese", 50935 },
- { "X-EBCDIC-Spain", 20284 },
- { "x-ebcdic-spain-euro", 1145 },
- { "x-EBCDIC-Thai", 20838 },
- { "x-EBCDIC-TraditionalChinese", 50937 },
- { "CP1026", 1026 },
- { "x-EBCDIC-Turkish", 20905 },
- { "x-EBCDIC-UK", 20285 },
- { "x-ebcdic-uk-euro", 1146 },
- { "ebcdic-cp-us", 37 },
- { "x-ebcdic-cp-us-euro", 1140 },
- { "ibm861", 861 },
- { "x-mac-icelandic", 10079 },
- { "x-iscii-as", 57006 },
- { "x-iscii-be", 57003 },
- { "x-iscii-de", 57002 },
- { "x-iscii-gu", 57010 },
- { "x-iscii-ka", 57008 },
- { "x-iscii-ma", 57009 },
- { "x-iscii-or", 57007 },
- { "x-iscii-pa", 57011 },
- { "x-iscii-ta", 57004 },
- { "x-iscii-te", 57005 },
- { "euc-jp", 51932 },
- { "csEUCPkdFmtJapanese", 51932 },
- { "Extended_UNIX_Code_Packed_Format_for_Japanese", 51932 },
- { "x-euc", 51932 },
- { "x-euc-jp", 51932 },
- { "iso-2022-jp", 50220 },
- { "iso-2022-jp", 50222 },
- { "_iso-2022-jp$SIO", 50222 },
- { "csISO2022JP", 50221 },
- { "_iso-2022-jp", 50221 },
- { "x-mac-japanese", 10001 },
- { "shift_jis", 932 },
- { "csShiftJIS", 932 },
- { "csWindows31J", 932 },
- { "ms_Kanji", 932 },
- { "shift-jis", 932 },
- { "x-ms-cp932", 932 },
- { "x-sjis", 932 },
- { "ks_c_5601-1987", 949 },
- { "csKSC56011987", 949 },
- { "euc-kr", 949 },
- { "iso-ir-149", 949 },
- { "korean", 949 },
- { "ks_c_5601", 949 },
- { "ks_c_5601_1987", 949 },
- { "ks_c_5601-1989", 949 },
- { "KSC_5601", 949 },
- { "KSC5601", 949 },
- { "euc-kr", 51949 },
- { "csEUCKR", 51949 },
- { "iso-2022-kr", 50225 },
- { "csISO2022KR", 50225 },
- { "Johab", 1361 },
- { "x-mac-korean", 10003 },
- { "iso-8859-3", 28593 },
- { "csISO", 28593 },
- { "Latin3", 28593 },
- { "ISO_8859-3", 28593 },
- { "ISO_8859-3:1988", 28593 },
- { "iso-ir-109", 28593 },
- { "l3", 28593 },
- { "latin3", 28593 },
- { "iso-8859-15", 28605 },
- { "csISO", 28605 },
- { "Latin9", 28605 },
- { "ISO_8859-15", 28605 },
- { "l9", 28605 },
- { "latin9", 28605 },
- { "x-IA5-Norwegian", 20108 },
- { "IBM437", 437 },
- { "437", 437 },
- { "cp437", 437 },
- { "csPC8", 437 },
- { "CodePage437", 437 },
- { "x-IA5-Swedish", 20107 },
- { "windows-874", 874 },
- { "DOS-874", 874 },
- { "iso-8859-11", 874 },
- { "TIS-620", 874 },
- { "ibm857", 857 },
- { "iso-8859-9", 28599 },
- { "csISO", 28599 },
- { "Latin5", 28599 },
- { "ISO_8859-9", 28599 },
- { "ISO_8859-9:1989", 28599 },
- { "iso-ir-148", 28599 },
- { "l5", 28599 },
- { "latin5", 28599 },
- { "x-mac-turkish", 10081 },
- { "windows-1254", 1254 },
- { "ISO_8859-9", 1254 },
- { "ISO_8859-9:1989", 1254 },
- { "iso-8859-9", 1254 },
- { "iso-ir-148", 1254 },
- { "latin5", 1254 },
- { "unicode", 1200 },
- { "utf-16", 1200 },
- { "unicodeFFFE", 1201 },
- { "utf-7", 65000 },
- { "csUnicode11UTF7", 65000 },
- { "unicode-1-1-utf-7", 65000 },
- { "x-unicode-2-0-utf-7", 65000 },
- { "utf-8", 65001 },
- { "unicode-1-1-utf-8", 65001 },
- { "unicode-2-0-utf-8", 65001 },
- { "x-unicode-2-0-utf-8", 65001 },
- { "us-ascii", 20127 },
- { "ANSI_X3.4-1968", 20127 },
- { "ANSI_X3.4-1986", 20127 },
- { "ascii", 20127 },
- { "cp367", 20127 },
- { "csASCII", 20127 },
- { "IBM367", 20127 },
- { "ISO_646.irv:1991", 20127 },
- { "ISO646-US", 20127 },
- { "iso-ir-6us", 20127 },
- { "windows-1258", 1258 },
- { "ibm850", 850 },
- { "x-IA5", 20105 },
- { "iso-8859-1", 28591 },
- { "cp819", 28591 },
- { "csISO", 28591 },
- { "Latin1", 28591 },
- { "ibm819", 28591 },
- { "iso_8859-1", 28591 },
- { "iso_8859-1:1987", 28591 },
- { "iso-ir-100", 28591 },
- { "l1", 28591 },
- { "latin1", 28591 },
- { "macintosh", 10000 },
- { "Windows-1252", 1252 },
- { "ANSI_X3.4-1968", 1252 },
- { "ANSI_X3.4-1986", 1252 },
- { "ascii", 1252 },
- { "cp367", 1252 },
- { "cp819", 1252 },
- { "csASCII", 1252 },
- { "IBM367", 1252 },
- { "ibm819", 1252 },
- { "ISO_646.irv:1991", 1252 },
- { "iso_8859-1", 1252 },
- { "iso_8859-1:1987", 1252 },
- { "ISO646-US", 1252 },
- { "iso-ir-100", 1252 },
- { "iso-ir-6", 1252 },
- { "latin1", 1252 },
- { "us", 1252 },
- { "us-ascii", 1252 },
- { "x-ansi", 1252 },
- { "microsoft-cp1251", 1251 }
-};
-
-#define LANGUAGE_NOT_LOADED 1
-#define LANGUAGE_LOADING -1
-#define LANGUAGE_LOADED 0
-
-void LoadThread(LPVOID hd);
-
-class HunspellDictionary : public Dictionary
-{
-protected:
- TCHAR fileWithoutExtension[1024];
- TCHAR userPath[1024];
- volatile int loaded;
- Hunspell *hunspell;
- TCHAR *wordChars;
- UINT codePage;
-
- void loadCustomDict()
- {
- TCHAR filename[1024];
- mir_sntprintf(filename, SIZEOF(filename), _T("%s\\%s.cdic"), userPath, language);
-
- FILE *file = _tfopen(filename, _T("rb"));
- if (file != NULL) {
- char tmp[1024];
- char c;
- int pos = 0;
- while ((c = fgetc(file)) != EOF) {
- if (c == '\n' || c == '\r' || pos >= SIZEOF(tmp) - 1) {
- if (pos > 0) {
- tmp[pos] = '\0';
- hunspell->add(tmp);
- }
-
- pos = 0;
- }
- else {
- tmp[pos] = c;
- pos++;
- }
- }
- fclose(file);
- }
- }
-
- void appendToCustomDict(const TCHAR *word)
- {
- CreateDirectoryTreeT(userPath);
-
- TCHAR filename[1024];
- mir_sntprintf(filename, SIZEOF(filename), _T("%s\\%s.cdic"), userPath, language);
-
- FILE *file = _tfopen(filename, _T("ab"));
- if (file != NULL) {
- char tmp[1024];
- toHunspell(tmp, word, SIZEOF(tmp));
- fprintf(file, "%s\n", tmp);
- fclose(file);
- }
- }
-
- virtual void addWordInternal(const TCHAR * word)
- {
- if (loaded != LANGUAGE_LOADED)
- return;
-
- char hunspell_word[1024];
- toHunspell(hunspell_word, word, SIZEOF(hunspell_word));
-
- hunspell->add(hunspell_word);
- }
-
- void toHunspell(char *hunspellWord, const TCHAR *word, size_t hunspellWordLen)
- {
- WideCharToMultiByte(codePage, 0, word, -1, hunspellWord, (int)hunspellWordLen, NULL, NULL);
- }
-
- TCHAR* fromHunspell(const char *hunspellWord)
- {
- int len = MultiByteToWideChar(codePage, 0, hunspellWord, -1, NULL, 0);
- WCHAR *ret = (WCHAR *)malloc((len + 1) * sizeof(WCHAR));
- MultiByteToWideChar(codePage, 0, hunspellWord, -1, ret, len + 1);
- return ret;
- }
-
- TCHAR* fromHunspellAndFree(char *hunspellWord)
- {
- if (hunspellWord == NULL)
- return NULL;
-
- TCHAR *ret = fromHunspell(hunspellWord);
- free(hunspellWord);
- return ret;
- }
-
-public:
- HunspellDictionary(TCHAR *aLanguage, TCHAR *aFileWithoutExtension, TCHAR *anUserPath, TCHAR *aSource)
- {
- lstrcpyn(language, aLanguage, SIZEOF(language));
- lstrcpyn(fileWithoutExtension, aFileWithoutExtension, SIZEOF(fileWithoutExtension));
- lstrcpyn(userPath, anUserPath, SIZEOF(userPath));
- if (aSource == NULL)
- source[0] = _T('\0');
- else
- lstrcpyn(source, aSource, SIZEOF(source));
-
- loaded = LANGUAGE_NOT_LOADED;
- localized_name[0] = _T('\0');
- english_name[0] = _T('\0');
- full_name[0] = _T('\0');
- hunspell = NULL;
- wordChars = NULL;
- codePage = CP_ACP;
- autoReplace = NULL;
- }
-
- virtual ~HunspellDictionary()
- {
- if (hunspell != NULL)
- delete hunspell;
- if (wordChars != NULL)
- free(wordChars);
- }
-
- TCHAR * merge(TCHAR * s1, TCHAR *s2)
- {
- int len1 = (s1 == NULL ? 0 : lstrlen(s1));
- int len2 = (s2 == NULL ? 0 : lstrlen(s2));
-
- TCHAR *ret;
- if (len1 > 0 && len2 > 0) {
- ret = (TCHAR *)malloc(sizeof(TCHAR) * (len1 + len2 + 1));
- lstrcpyn(ret, s1, len1 + 1);
- lstrcpyn(&ret[len1], s2, len2 + 1);
-
- FREE(s1);
- FREE(s2);
- }
- else if (len1 > 0) {
- ret = s1;
- FREE(s2);
- }
- else if (len2 > 0) {
- ret = s2;
- FREE(s1);
- }
- else {
- ret = (TCHAR *)malloc(sizeof(TCHAR));
- ret[0] = 0;
-
- FREE(s1);
- FREE(s2);
- }
-
- // Remove duplicated chars
- int last = lstrlen(ret) - 1;
- for (int i = 0; i <= last; i++) {
- TCHAR c = ret[i];
- for (int j = last; j > i; j--) {
- if (c != ret[j])
- continue;
- if (j != last)
- ret[j] = ret[last];
- ret[last] = _T('\0');
- last--;
- }
- }
-
- return ret;
- }
-
-
- void loadThread()
- {
- char dic[1024];
- char aff[1024];
-
-
- mir_snprintf(dic, SIZEOF(dic), "%S.dic", fileWithoutExtension);
- mir_snprintf(aff, SIZEOF(aff), "%S.aff", fileWithoutExtension);
-
-
- hunspell = new Hunspell(aff, dic);
-
- // Get codepage
- const char *dic_enc = hunspell->get_dic_encoding();
-
- TCHAR *hwordchars;
- if (strcmp(dic_enc, "UTF-8") == 0) {
- codePage = CP_UTF8;
-
-
- int wcs_len;
- hwordchars = fromHunspell((char *)hunspell->get_wordchars_utf16(&wcs_len));
-
- }
- else {
- for (int i = 0; i < SIZEOF(codepages); i++) {
- if (_strcmpi(codepages[i].name, dic_enc) == 0) {
- if (IsValidCodePage(codepages[i].codepage))
- codePage = codepages[i].codepage;
- break;
- }
- }
-
- hwordchars = fromHunspell(hunspell->get_wordchars());
- }
-
- TCHAR *casechars = fromHunspellAndFree(get_casechars(dic_enc));
- TCHAR *try_string = fromHunspellAndFree(hunspell->get_try_string());
-
- wordChars = merge(merge(casechars, hwordchars), try_string);
-
- // Make a suggestion to load hunspell internalls
- char ** words = NULL;
- int count = hunspell->suggest(&words, "asdf");
- for (int i = 0; i < count; i++)
- free(words[i]);
- if (words != NULL)
- free(words);
-
- loadCustomDict();
-
- loaded = LANGUAGE_LOADED;
- }
-
- // Return TRUE if the word is correct
- virtual BOOL spell(const TCHAR *word)
- {
- load();
- if (loaded != LANGUAGE_LOADED)
- return TRUE;
-
- // TODO Check if it was generated by auto-replacement
-
- char hunspell_word[1024];
- toHunspell(hunspell_word, word, SIZEOF(hunspell_word));
-
- return hunspell->spell(hunspell_word);
- }
-
- // Return a list of suggestions to a word
- virtual Suggestions suggest(const TCHAR * word)
- {
- Suggestions ret = { 0 };
-
- load();
- if (loaded != LANGUAGE_LOADED)
- return ret;
-
- char hunspell_word[1024];
- toHunspell(hunspell_word, word, SIZEOF(hunspell_word));
-
- char ** words = NULL;
- ret.count = hunspell->suggest(&words, hunspell_word);
-
- if (ret.count > 0) {
- // Oki, lets make our array
- ret.words = (TCHAR **)malloc(ret.count * sizeof(TCHAR *));
- for (unsigned i = 0; i < ret.count; i++) {
- ret.words[i] = fromHunspell(words[i]);
- free(words[i]);
- }
- }
-
- if (words != NULL)
- free(words);
-
- return ret;
- }
-
- // Return a list of auto suggestions to a word
- virtual Suggestions autoSuggest(const TCHAR * word)
- {
- Suggestions ret = { 0 };
-
- load();
- if (loaded != LANGUAGE_LOADED)
- return ret;
-
- char hunspell_word[1024];
- toHunspell(hunspell_word, word, SIZEOF(hunspell_word));
-
- char ** words;
- int count = hunspell->suggest_auto(&words, hunspell_word);
-
- if (count <= 0)
- return ret;
-
- // Oki, lets make our array
- ret.count = count;
- ret.words = (TCHAR **)malloc(ret.count * sizeof(TCHAR *));
- for (int i = 0; i < count; i++) {
- ret.words[i] = fromHunspell(words[i]);
- free(words[i]);
- }
- free(words);
-
- return ret;
- }
-
- // Return a list of auto suggestions to a word
- // You have to free the list AND each item
- virtual TCHAR * autoSuggestOne(const TCHAR * word)
- {
- load();
- if (loaded != LANGUAGE_LOADED)
- return NULL;
-
- char hunspell_word[1024];
- toHunspell(hunspell_word, word, SIZEOF(hunspell_word));
-
- char ** words;
- int count = hunspell->suggest_auto(&words, hunspell_word);
-
- if (count <= 0)
- return NULL;
-
- TCHAR *ret = fromHunspell(words[0]);
-
- // Oki, lets make our array
- for (int i = 0; i < count; i++)
- free(words[i]);
- free(words);
-
- return ret;
- }
-
- // Return TRUE if the char is a word char
- virtual BOOL isWordChar(TCHAR c)
- {
- if (c == 0)
- return FALSE;
-
- load();
- if (loaded != LANGUAGE_LOADED)
- return TRUE;
-
- return _tcschr(wordChars, (_TINT)c) != NULL;
- }
-
- // Assert that all needed data is loaded
- virtual void load()
- {
- if (loaded == LANGUAGE_NOT_LOADED) {
- loaded = LANGUAGE_LOADING;
- mir_forkthread(LoadThread, this);
- }
- }
-
- virtual BOOL isLoaded()
- {
- return loaded == LANGUAGE_LOADED;
- }
-
-
- // Add a word to the user custom dict
- virtual void addWord(const TCHAR * word)
- {
- addWordInternal(word);
- appendToCustomDict(word);
- }
-
- // Add a word to the list of ignored words
- virtual void ignoreWord(const TCHAR * word)
- {
- addWordInternal(word);
- }
-};
-
-void LoadThread(LPVOID hd)
-{
- HunspellDictionary *dict = (HunspellDictionary *)hd;
- dict->loadThread();
-}
-
-
-
-// To use with EnumLocalesProc :(
-LIST<Dictionary> *tmp_dicts;
-
-// To get the names of the languages
-BOOL CALLBACK EnumLocalesProc(LPTSTR lpLocaleString)
-{
- TCHAR *stopped = NULL;
- USHORT langID = (USHORT)_tcstol(lpLocaleString, &stopped, 16);
-
- TCHAR ini[32];
- TCHAR end[32];
- GetLocaleInfo(MAKELCID(langID, 0), LOCALE_SISO639LANGNAME, ini, SIZEOF(ini));
- GetLocaleInfo(MAKELCID(langID, 0), LOCALE_SISO3166CTRYNAME, end, SIZEOF(end));
-
- TCHAR name[64];
- mir_sntprintf(name, SIZEOF(name), _T("%s_%s"), ini, end);
-
- for (int i = 0; i < tmp_dicts->getCount(); i++) {
- Dictionary *dict = (*tmp_dicts)[i];
- if (lstrcmpi(dict->language, name) == 0) {
- GetLocaleInfo(MAKELCID(langID, 0), LOCALE_SENGLANGUAGE, dict->english_name, SIZEOF(dict->english_name));
-
- GetLocaleInfo(MAKELCID(langID, 0), LOCALE_SLANGUAGE, dict->localized_name, SIZEOF(dict->localized_name));
- if (dict->localized_name[0] == 0)
- GetLocaleInfo(MAKELCID(langID, 0), LOCALE_SLOCALIZEDLANGUAGENAME, dict->localized_name, SIZEOF(dict->localized_name));
- if (dict->localized_name[0] == 0)
- GetLocaleInfo(MAKELCID(langID, 0), LOCALE_SNATIVEDISPLAYNAME, dict->localized_name, SIZEOF(dict->localized_name));
- if (dict->localized_name[0] == 0 && dict->english_name[0] != 0) {
- TCHAR country[1024];
- GetLocaleInfo(MAKELCID(langID, 0), LOCALE_SENGCOUNTRY, country, SIZEOF(country));
-
- TCHAR name[1024];
- if (country[0] != 0)
- mir_sntprintf(name, SIZEOF(name), _T("%s (%s)"), dict->english_name, country);
- else
- lstrcpyn(name, dict->english_name, SIZEOF(name));
-
- lstrcpyn(dict->localized_name, TranslateTS(name), SIZEOF(dict->localized_name));
- }
-
- if (dict->localized_name[0] != 0) {
- mir_sntprintf(dict->full_name, SIZEOF(dict->full_name), _T("%s [%s]"), dict->localized_name, dict->language);
- }
- break;
- }
- }
- return TRUE;
-}
-
-
-void GetDictsInfo(LIST<Dictionary> &dicts)
-{
- tmp_dicts = &dicts;
- EnumSystemLocales(EnumLocalesProc, LCID_SUPPORTED);
-
- // Try to get name from DB
- for (int i = 0; i < dicts.getCount(); i++) {
- Dictionary *dict = dicts[i];
-
- if (dict->full_name[0] == _T('\0')) {
- DBVARIANT dbv;
-
- char lang[128];
- WideCharToMultiByte(CP_ACP, 0, dict->language, -1, lang, sizeof(lang), NULL, NULL);
- if (!db_get_ts(NULL, MODULE_NAME, lang, &dbv)) {
- lstrcpyn(dict->localized_name, dbv.ptszVal, SIZEOF(dict->localized_name));
- db_free(&dbv);
- }
-
- if (dict->localized_name[0] == _T('\0')) {
- for (size_t j = 0; j < SIZEOF(aditionalLanguages); j++) {
- if (!lstrcmp(aditionalLanguages[j].language, dict->language)) {
- lstrcpyn(dict->localized_name, TranslateTS(aditionalLanguages[j].localized_name), SIZEOF(dict->localized_name));
- break;
- }
- }
- }
-
- if (dict->localized_name[0] != _T('\0')) {
- mir_sntprintf(dict->full_name, SIZEOF(dict->full_name), _T("%s [%s]"), dict->localized_name, dict->language);
- }
- else {
- lstrcpyn(dict->full_name, dict->language, SIZEOF(dict->full_name));
- }
- }
- }
-}
-
-
-void GetHunspellDictionariesFromFolder(LIST<Dictionary> &dicts, TCHAR *path, TCHAR *user_path, TCHAR *source)
-{
- // Load the language files and create an array with then
- TCHAR file[1024];
- mir_sntprintf(file, SIZEOF(file), _T("%s\\*.dic"), path);
-
- BOOL found = FALSE;
-
- WIN32_FIND_DATA ffd = { 0 };
- HANDLE hFFD = FindFirstFile(file, &ffd);
- if (hFFD != INVALID_HANDLE_VALUE) {
- do {
- mir_sntprintf(file, SIZEOF(file), _T("%s\\%s"), path, ffd.cFileName);
-
- // Check .dic
- DWORD attrib = GetFileAttributes(file);
- if (attrib == 0xFFFFFFFF || (attrib & FILE_ATTRIBUTE_DIRECTORY))
- continue;
-
- // See if .aff exists too
- lstrcpy(&file[lstrlen(file) - 4], _T(".aff"));
- attrib = GetFileAttributes(file);
- if (attrib == 0xFFFFFFFF || (attrib & FILE_ATTRIBUTE_DIRECTORY))
- continue;
-
- ffd.cFileName[lstrlen(ffd.cFileName) - 4] = _T('\0');
-
- TCHAR *lang = ffd.cFileName;
-
- // Replace - for _
- for (int i = 0; i < lstrlen(lang); i++)
- if (lang[i] == _T('-'))
- lang[i] = _T('_');
-
- // Check if dict is new
- bool exists = false;
- for (int i = 0; i < dicts.getCount() && !exists; i++)
- if (lstrcmp(dicts[i]->language, lang) == 0)
- exists = true;
-
- if (!exists) {
- found = TRUE;
- file[lstrlen(file) - 4] = _T('\0');
- dicts.insert(new HunspellDictionary(lang, file, user_path, source));
- }
- }
- while (FindNextFile(hFFD, &ffd));
-
- FindClose(hFFD);
- }
-}
-
-
-// Return a list of avaible languages
-void GetAvaibleDictionaries(LIST<Dictionary> &dicts, TCHAR *path, TCHAR *user_path)
-{
- // Get miranda folder dicts
- GetHunspellDictionariesFromFolder(dicts, path, user_path, NULL);
-
- if (opts.use_other_apps_dicts) {
- // Get other apps dicts
- for (int i = 0; i < SIZEOF(otherHunspellApps); i++) {
- TCHAR key[1024];
- mir_sntprintf(key, SIZEOF(key), APPPATH, otherHunspellApps[i].key);
-
- HKEY hKey = 0;
- LONG lResult = 0;
- if (ERROR_SUCCESS == RegOpenKeyEx(HKEY_LOCAL_MACHINE, key, 0, KEY_QUERY_VALUE, &hKey)) {
- DWORD size = SIZEOF(key);
- lResult = RegQueryValueEx(hKey, _T("Path"), NULL, NULL, (LPBYTE)key, &size);
- RegCloseKey(hKey);
- }
- else {
- // Not found in installed apps - Try MUICache
- lResult = RegOpenKeyEx(HKEY_CURRENT_USER, MUICACHE, 0, KEY_QUERY_VALUE, &hKey);
- if (ERROR_SUCCESS == lResult) {
- DWORD numValues;
- if (ERROR_SUCCESS != RegQueryInfoKey(hKey, NULL, NULL, NULL, NULL, NULL, NULL, &numValues, NULL, NULL, NULL, NULL))
- numValues = 0;
-
- lResult = ERROR_NO_MORE_ITEMS;
- for (DWORD local = 0; local < numValues; local++) {
- DWORD cchValue = SIZEOF(key);
- if (ERROR_SUCCESS != RegEnumValue(hKey, local, key, &cchValue, NULL, NULL, NULL, NULL))
- break;
- key[cchValue] = 0;
- TCHAR *pos;
- if (pos = _tcsrchr(key, _T('\\'))) {
- if (!lstrcmpi(&pos[1], otherHunspellApps[i].key)) {
- pos[0] = 0;
- lResult = ERROR_SUCCESS;
- break;
- }
- }
- }
- RegCloseKey(hKey);
- }
- }
-
- if (ERROR_SUCCESS == lResult) {
- TCHAR folder[1024];
- mir_sntprintf(folder, SIZEOF(folder), _T("%s\\Dictionaries"), key);
-
- GetHunspellDictionariesFromFolder(languages, folder, user_path, otherHunspellApps[i].name);
- }
- }
- }
-
- GetDictsInfo(dicts);
-
- // Yeah, yeah, yeah, I know, but this is the easiest way...
- SortedList *sl = (SortedList *)&dicts;
-
- // Sort dicts
- for (int i = 0; i < dicts.getCount(); i++) {
- for (int j = i + 1; j < dicts.getCount(); j++) {
- if (lstrcmp(dicts[i]->full_name, dicts[j]->full_name) > 0) {
- Dictionary *dict = dicts[i];
- sl->items[i] = dicts[j];
- sl->items[j] = dict;
- }
- }
- }
-}
-
-
-// Free the list returned by GetAvaibleDictionaries
-void FreeDictionaries(LIST<Dictionary> &dicts)
-{
- for (int i = 0; i < dicts.getCount(); i++)
- delete dicts[i];
-
- dicts.destroy();
-}
-
-Dictionary::~Dictionary()
-{
- delete autoReplace;
-}
-
-// Free the list returned by GetAvaibleDictionaries
-void FreeSuggestions(Suggestions &suggestions)
-{
- for (size_t i = 0; i < suggestions.count; i++)
- free(suggestions.words[i]);
-
- free(suggestions.words);
-
- suggestions.words = NULL;
- suggestions.count = 0;
-}
+/*
+Copyright (C) 2006-2010 Ricardo Pescuma Domenecci
+
+This is free software; you can redistribute it and/or
+modify it under the terms of the GNU Library General Public
+License as published by the Free Software Foundation; either
+version 2 of the License, or (at your option) any later version.
+
+This 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
+Library General Public License for more details.
+
+You should have received a copy of the GNU Library General Public
+License along with this file; see the file license.txt. If
+not, write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+Boston, MA 02111-1307, USA.
+*/
+
+#include "commons.h"
+
+#define APPPATH _T("SOFTWARE\\Microsoft\\Windows\\CurrentVersion\\App Paths\\%s")
+#define MUICACHE _T("Software\\Microsoft\\Windows\\ShellNoRoam\\MUICache")
+
+
+// Additional languages that i could not find in Windows
+struct {
+ TCHAR *language;
+ TCHAR *localized_name;
+} aditionalLanguages[] = {
+ { _T("tl_PH"), LPGENT("Tagalog (Philippines)") },
+ { _T("de_frami_neu"), LPGENT("German (Germany)") }
+};
+
+struct {
+ TCHAR *name;
+ TCHAR *key;
+} otherHunspellApps[] = {
+ { _T("Thunderbird"), _T("thunderbird.exe") },
+ { _T("Firefox"), _T("firefox.exe") }
+};
+
+struct {
+ char *name;
+ UINT codepage;
+} codepages[] = {
+ { "ISO8859-1", 28591 },
+ { "UTF-7", CP_UTF7 },
+ { "UTF-8", CP_UTF8 },
+ { "UTF7", CP_UTF7 },
+ { "UTF8", CP_UTF8 },
+ { "ISO8859-2", 28592 },
+ { "ISO8859-3", 28593 },
+ { "ISO8859-4", 28594 },
+ { "ISO8859-5", 28595 },
+ { "ISO8859-6", 28596 },
+ { "ISO8859-7", 28597 },
+ { "ISO8859-8", 28598 },
+ { "ISO8859-9", 28599 },
+ { "ASMO-708", 708 },
+ { "DOS-720", 720 },
+ { "iso-8859-6", 28596 },
+ { "arabic", 28596 },
+ { "csISOLatinArabic", 28596 },
+ { "ECMA-114", 28596 },
+ { "ISO_8859-6", 28596 },
+ { "ISO_8859-6:1987", 28596 },
+ { "iso-ir-127", 28596 },
+ { "x-mac-arabic", 10004 },
+ { "windows-1256", 1256 },
+ { "cp1256", 1256 },
+ { "ibm775", 775 },
+ { "CP500", 775 },
+ { "iso-8859-4", 28594 },
+ { "csISOLatin4", 28594 },
+ { "ISO_8859-4", 28594 },
+ { "ISO_8859-4:1988", 28594 },
+ { "iso-ir-110", 28594 },
+ { "l4", 28594 },
+ { "latin4", 28594 },
+ { "windows-1257", 1257 },
+ { "ibm852", 852 },
+ { "cp852", 852 },
+ { "iso-8859-2", 28592 },
+ { "csISOLatin2", 28592 },
+ { "iso_8859-2", 28592 },
+ { "iso_8859-2:1987", 28592 },
+ { "iso-ir-101", 28592 },
+ { "l2", 28592 },
+ { "latin2", 28592 },
+ { "x-mac-ce", 10029 },
+ { "windows-1250", 1250 },
+ { "x-cp1250", 1250 },
+ { "EUC-CN", 51936 },
+ { "x-euc-cn", 51936 },
+ { "gb2312", 936 },
+ { "chinese", 936 },
+ { "CN-GB", 936 },
+ { "csGB2312", 936 },
+ { "csGB231280", 936 },
+ { "csISO58GB231280", 936 },
+ { "GB_2312-80", 936 },
+ { "GB231280", 936 },
+ { "GB2312-80", 936 },
+ { "GBK", 936 },
+ { "iso-ir-58", 936 },
+ { "hz-gb-2312", 52936 },
+ { "x-mac-chinesesimp", 10008 },
+ { "big5", 950 },
+ { "cn-big5", 950 },
+ { "csbig5", 950 },
+ { "x-x-big5", 950 },
+ { "x-Chinese-CNS", 20000 },
+ { "x-Chinese-Eten", 20002 },
+ { "x-mac-chinesetrad", 10002 },
+ { "cp866", 866 },
+ { "ibm866", 866 },
+ { "iso-8859-5", 28595 },
+ { "csISOLatin5", 28595 },
+ { "csISOLatinCyrillic", 28595 },
+ { "cyrillic", 28595 },
+ { "ISO_8859-5", 28595 },
+ { "ISO_8859-5:1988", 28595 },
+ { "iso-ir-144", 28595 },
+ { "l5", 28595 },
+ { "KOI8-R", 20866 },
+ { "csKOI8R", 20866 },
+ { "koi", 20866 },
+ { "koi8", 20866 },
+ { "koi8r", 20866 },
+ { "KOI8-U", 21866 },
+ { "koi8-ru", 21866 },
+ { "x-mac-cyrillic", 10007 },
+ { "windows-1251", 1251 },
+ { "Win1251", 1251 },
+ { "x-cp1251", 1251 },
+ { "x-Europa", 29001 },
+ { "x-IA5-German", 20106 },
+ { "ibm737", 737 },
+ { "iso-8859-7", 28597 },
+ { "csISOLatinGreek", 28597 },
+ { "ECMA-118", 28597 },
+ { "ELOT_928", 28597 },
+ { "greek", 28597 },
+ { "greek8", 28597 },
+ { "ISO_8859-7", 28597 },
+ { "ISO_8859-7:1987", 28597 },
+ { "iso-ir-126", 28597 },
+ { "x-mac-greek", 10006 },
+ { "windows-1253", 1253 },
+ { "ibm869", 869 },
+ { "DOS-862", 862 },
+ { "iso-8859-8-i", 38598 },
+ { "logical", 38598 },
+ { "iso-8859-8", 28598 },
+ { "csISOLatinHebrew", 28598 },
+ { "hebrew", 28598 },
+ { "ISO_8859-8", 28598 },
+ { "ISO_8859-8:1988", 28598 },
+ { "ISO-8859-8", 28598 },
+ { "iso-ir-138", 28598 },
+ { "visual", 28598 },
+ { "x-mac-hebrew", 10005 },
+ { "windows-1255", 1255 },
+ { "ISO_8859-8-I", 1255 },
+ { "ISO-8859-8", 1255 },
+ { "x-EBCDIC-Arabic", 20420 },
+ { "x-EBCDIC-CyrillicRussian", 20880 },
+ { "x-EBCDIC-CyrillicSerbianBulgarian", 21025 },
+ { "x-EBCDIC-DenmarkNorway", 20277 },
+ { "x-ebcdic-denmarknorway-euro", 1142 },
+ { "x-EBCDIC-FinlandSweden", 20278 },
+ { "x-ebcdic-finlandsweden-euro", 1143 },
+ { "X-EBCDIC-France", 1143 },
+ { "X-EBCDIC-France", 1143 },
+ { "x-ebcdic-france-euro", 1147 },
+ { "x-EBCDIC-Germany", 20273 },
+ { "x-ebcdic-germany-euro", 1141 },
+ { "x-EBCDIC-GreekModern", 875 },
+ { "x-EBCDIC-Greek", 20423 },
+ { "x-EBCDIC-Hebrew", 20424 },
+ { "x-EBCDIC-Icelandic", 20871 },
+ { "x-ebcdic-icelandic-euro", 1149 },
+ { "x-ebcdic-international-euro", 1148 },
+ { "x-EBCDIC-Italy", 20280 },
+ { "x-ebcdic-italy-euro", 1144 },
+ { "x-EBCDIC-JapaneseAndKana", 50930 },
+ { "x-EBCDIC-JapaneseAndJapaneseLatin", 50939 },
+ { "x-EBCDIC-JapaneseAndUSCanada", 50931 },
+ { "x-EBCDIC-JapaneseKatakana", 20290 },
+ { "x-EBCDIC-KoreanAndKoreanExtended", 50933 },
+ { "x-EBCDIC-KoreanExtended", 20833 },
+ { "CP870", 870 },
+ { "x-EBCDIC-SimplifiedChinese", 50935 },
+ { "X-EBCDIC-Spain", 20284 },
+ { "x-ebcdic-spain-euro", 1145 },
+ { "x-EBCDIC-Thai", 20838 },
+ { "x-EBCDIC-TraditionalChinese", 50937 },
+ { "CP1026", 1026 },
+ { "x-EBCDIC-Turkish", 20905 },
+ { "x-EBCDIC-UK", 20285 },
+ { "x-ebcdic-uk-euro", 1146 },
+ { "ebcdic-cp-us", 37 },
+ { "x-ebcdic-cp-us-euro", 1140 },
+ { "ibm861", 861 },
+ { "x-mac-icelandic", 10079 },
+ { "x-iscii-as", 57006 },
+ { "x-iscii-be", 57003 },
+ { "x-iscii-de", 57002 },
+ { "x-iscii-gu", 57010 },
+ { "x-iscii-ka", 57008 },
+ { "x-iscii-ma", 57009 },
+ { "x-iscii-or", 57007 },
+ { "x-iscii-pa", 57011 },
+ { "x-iscii-ta", 57004 },
+ { "x-iscii-te", 57005 },
+ { "euc-jp", 51932 },
+ { "csEUCPkdFmtJapanese", 51932 },
+ { "Extended_UNIX_Code_Packed_Format_for_Japanese", 51932 },
+ { "x-euc", 51932 },
+ { "x-euc-jp", 51932 },
+ { "iso-2022-jp", 50220 },
+ { "iso-2022-jp", 50222 },
+ { "_iso-2022-jp$SIO", 50222 },
+ { "csISO2022JP", 50221 },
+ { "_iso-2022-jp", 50221 },
+ { "x-mac-japanese", 10001 },
+ { "shift_jis", 932 },
+ { "csShiftJIS", 932 },
+ { "csWindows31J", 932 },
+ { "ms_Kanji", 932 },
+ { "shift-jis", 932 },
+ { "x-ms-cp932", 932 },
+ { "x-sjis", 932 },
+ { "ks_c_5601-1987", 949 },
+ { "csKSC56011987", 949 },
+ { "euc-kr", 949 },
+ { "iso-ir-149", 949 },
+ { "korean", 949 },
+ { "ks_c_5601", 949 },
+ { "ks_c_5601_1987", 949 },
+ { "ks_c_5601-1989", 949 },
+ { "KSC_5601", 949 },
+ { "KSC5601", 949 },
+ { "euc-kr", 51949 },
+ { "csEUCKR", 51949 },
+ { "iso-2022-kr", 50225 },
+ { "csISO2022KR", 50225 },
+ { "Johab", 1361 },
+ { "x-mac-korean", 10003 },
+ { "iso-8859-3", 28593 },
+ { "csISO", 28593 },
+ { "Latin3", 28593 },
+ { "ISO_8859-3", 28593 },
+ { "ISO_8859-3:1988", 28593 },
+ { "iso-ir-109", 28593 },
+ { "l3", 28593 },
+ { "latin3", 28593 },
+ { "iso-8859-15", 28605 },
+ { "csISO", 28605 },
+ { "Latin9", 28605 },
+ { "ISO_8859-15", 28605 },
+ { "l9", 28605 },
+ { "latin9", 28605 },
+ { "x-IA5-Norwegian", 20108 },
+ { "IBM437", 437 },
+ { "437", 437 },
+ { "cp437", 437 },
+ { "csPC8", 437 },
+ { "CodePage437", 437 },
+ { "x-IA5-Swedish", 20107 },
+ { "windows-874", 874 },
+ { "DOS-874", 874 },
+ { "iso-8859-11", 874 },
+ { "TIS-620", 874 },
+ { "ibm857", 857 },
+ { "iso-8859-9", 28599 },
+ { "csISO", 28599 },
+ { "Latin5", 28599 },
+ { "ISO_8859-9", 28599 },
+ { "ISO_8859-9:1989", 28599 },
+ { "iso-ir-148", 28599 },
+ { "l5", 28599 },
+ { "latin5", 28599 },
+ { "x-mac-turkish", 10081 },
+ { "windows-1254", 1254 },
+ { "ISO_8859-9", 1254 },
+ { "ISO_8859-9:1989", 1254 },
+ { "iso-8859-9", 1254 },
+ { "iso-ir-148", 1254 },
+ { "latin5", 1254 },
+ { "unicode", 1200 },
+ { "utf-16", 1200 },
+ { "unicodeFFFE", 1201 },
+ { "utf-7", 65000 },
+ { "csUnicode11UTF7", 65000 },
+ { "unicode-1-1-utf-7", 65000 },
+ { "x-unicode-2-0-utf-7", 65000 },
+ { "utf-8", 65001 },
+ { "unicode-1-1-utf-8", 65001 },
+ { "unicode-2-0-utf-8", 65001 },
+ { "x-unicode-2-0-utf-8", 65001 },
+ { "us-ascii", 20127 },
+ { "ANSI_X3.4-1968", 20127 },
+ { "ANSI_X3.4-1986", 20127 },
+ { "ascii", 20127 },
+ { "cp367", 20127 },
+ { "csASCII", 20127 },
+ { "IBM367", 20127 },
+ { "ISO_646.irv:1991", 20127 },
+ { "ISO646-US", 20127 },
+ { "iso-ir-6us", 20127 },
+ { "windows-1258", 1258 },
+ { "ibm850", 850 },
+ { "x-IA5", 20105 },
+ { "iso-8859-1", 28591 },
+ { "cp819", 28591 },
+ { "csISO", 28591 },
+ { "Latin1", 28591 },
+ { "ibm819", 28591 },
+ { "iso_8859-1", 28591 },
+ { "iso_8859-1:1987", 28591 },
+ { "iso-ir-100", 28591 },
+ { "l1", 28591 },
+ { "latin1", 28591 },
+ { "macintosh", 10000 },
+ { "Windows-1252", 1252 },
+ { "ANSI_X3.4-1968", 1252 },
+ { "ANSI_X3.4-1986", 1252 },
+ { "ascii", 1252 },
+ { "cp367", 1252 },
+ { "cp819", 1252 },
+ { "csASCII", 1252 },
+ { "IBM367", 1252 },
+ { "ibm819", 1252 },
+ { "ISO_646.irv:1991", 1252 },
+ { "iso_8859-1", 1252 },
+ { "iso_8859-1:1987", 1252 },
+ { "ISO646-US", 1252 },
+ { "iso-ir-100", 1252 },
+ { "iso-ir-6", 1252 },
+ { "latin1", 1252 },
+ { "us", 1252 },
+ { "us-ascii", 1252 },
+ { "x-ansi", 1252 },
+ { "microsoft-cp1251", 1251 }
+};
+
+#define LANGUAGE_NOT_LOADED 1
+#define LANGUAGE_LOADING -1
+#define LANGUAGE_LOADED 0
+
+void LoadThread(LPVOID hd);
+
+class HunspellDictionary : public Dictionary
+{
+protected:
+ TCHAR fileWithoutExtension[1024];
+ TCHAR userPath[1024];
+ volatile int loaded;
+ Hunspell *hunspell;
+ TCHAR *wordChars;
+ UINT codePage;
+
+ void loadCustomDict()
+ {
+ TCHAR filename[1024];
+ mir_sntprintf(filename, SIZEOF(filename), _T("%s\\%s.cdic"), userPath, language);
+
+ FILE *file = _tfopen(filename, _T("rb"));
+ if (file != NULL) {
+ char tmp[1024];
+ char c;
+ int pos = 0;
+ while ((c = fgetc(file)) != EOF) {
+ if (c == '\n' || c == '\r' || pos >= SIZEOF(tmp) - 1) {
+ if (pos > 0) {
+ tmp[pos] = '\0';
+ hunspell->add(tmp);
+ }
+
+ pos = 0;
+ }
+ else {
+ tmp[pos] = c;
+ pos++;
+ }
+ }
+ fclose(file);
+ }
+ }
+
+ void appendToCustomDict(const TCHAR *word)
+ {
+ CreateDirectoryTreeT(userPath);
+
+ TCHAR filename[1024];
+ mir_sntprintf(filename, SIZEOF(filename), _T("%s\\%s.cdic"), userPath, language);
+
+ FILE *file = _tfopen(filename, _T("ab"));
+ if (file != NULL) {
+ char tmp[1024];
+ toHunspell(tmp, word, SIZEOF(tmp));
+ fprintf(file, "%s\n", tmp);
+ fclose(file);
+ }
+ }
+
+ virtual void addWordInternal(const TCHAR * word)
+ {
+ if (loaded != LANGUAGE_LOADED)
+ return;
+
+ char hunspell_word[1024];
+ toHunspell(hunspell_word, word, SIZEOF(hunspell_word));
+
+ hunspell->add(hunspell_word);
+ }
+
+ void toHunspell(char *hunspellWord, const TCHAR *word, size_t hunspellWordLen)
+ {
+ WideCharToMultiByte(codePage, 0, word, -1, hunspellWord, (int)hunspellWordLen, NULL, NULL);
+ }
+
+ TCHAR* fromHunspell(const char *hunspellWord)
+ {
+ int len = MultiByteToWideChar(codePage, 0, hunspellWord, -1, NULL, 0);
+ WCHAR *ret = (WCHAR *)malloc((len + 1) * sizeof(WCHAR));
+ MultiByteToWideChar(codePage, 0, hunspellWord, -1, ret, len + 1);
+ return ret;
+ }
+
+ TCHAR* fromHunspellAndFree(char *hunspellWord)
+ {
+ if (hunspellWord == NULL)
+ return NULL;
+
+ TCHAR *ret = fromHunspell(hunspellWord);
+ free(hunspellWord);
+ return ret;
+ }
+
+public:
+ HunspellDictionary(TCHAR *aLanguage, TCHAR *aFileWithoutExtension, TCHAR *anUserPath, TCHAR *aSource)
+ {
+ lstrcpyn(language, aLanguage, SIZEOF(language));
+ lstrcpyn(fileWithoutExtension, aFileWithoutExtension, SIZEOF(fileWithoutExtension));
+ lstrcpyn(userPath, anUserPath, SIZEOF(userPath));
+ if (aSource == NULL)
+ source[0] = _T('\0');
+ else
+ lstrcpyn(source, aSource, SIZEOF(source));
+
+ loaded = LANGUAGE_NOT_LOADED;
+ localized_name[0] = _T('\0');
+ english_name[0] = _T('\0');
+ full_name[0] = _T('\0');
+ hunspell = NULL;
+ wordChars = NULL;
+ codePage = CP_ACP;
+ autoReplace = NULL;
+ }
+
+ virtual ~HunspellDictionary()
+ {
+ if (hunspell != NULL)
+ delete hunspell;
+ if (wordChars != NULL)
+ free(wordChars);
+ }
+
+ TCHAR * merge(TCHAR * s1, TCHAR *s2)
+ {
+ int len1 = (s1 == NULL ? 0 : lstrlen(s1));
+ int len2 = (s2 == NULL ? 0 : lstrlen(s2));
+
+ TCHAR *ret;
+ if (len1 > 0 && len2 > 0) {
+ ret = (TCHAR *)malloc(sizeof(TCHAR) * (len1 + len2 + 1));
+ lstrcpyn(ret, s1, len1 + 1);
+ lstrcpyn(&ret[len1], s2, len2 + 1);
+
+ FREE(s1);
+ FREE(s2);
+ }
+ else if (len1 > 0) {
+ ret = s1;
+ FREE(s2);
+ }
+ else if (len2 > 0) {
+ ret = s2;
+ FREE(s1);
+ }
+ else {
+ ret = (TCHAR *)malloc(sizeof(TCHAR));
+ ret[0] = 0;
+
+ FREE(s1);
+ FREE(s2);
+ }
+
+ // Remove duplicated chars
+ int last = lstrlen(ret) - 1;
+ for (int i = 0; i <= last; i++) {
+ TCHAR c = ret[i];
+ for (int j = last; j > i; j--) {
+ if (c != ret[j])
+ continue;
+ if (j != last)
+ ret[j] = ret[last];
+ ret[last] = _T('\0');
+ last--;
+ }
+ }
+
+ return ret;
+ }
+
+
+ void loadThread()
+ {
+ char dic[1024];
+ char aff[1024];
+
+
+ mir_snprintf(dic, SIZEOF(dic), "%S.dic", fileWithoutExtension);
+ mir_snprintf(aff, SIZEOF(aff), "%S.aff", fileWithoutExtension);
+
+
+ hunspell = new Hunspell(aff, dic);
+
+ // Get codepage
+ const char *dic_enc = hunspell->get_dic_encoding();
+
+ TCHAR *hwordchars;
+ if (strcmp(dic_enc, "UTF-8") == 0) {
+ codePage = CP_UTF8;
+
+
+ int wcs_len;
+ hwordchars = fromHunspell((char *)hunspell->get_wordchars_utf16(&wcs_len));
+
+ }
+ else {
+ for (int i = 0; i < SIZEOF(codepages); i++) {
+ if (_strcmpi(codepages[i].name, dic_enc) == 0) {
+ if (IsValidCodePage(codepages[i].codepage))
+ codePage = codepages[i].codepage;
+ break;
+ }
+ }
+
+ hwordchars = fromHunspell(hunspell->get_wordchars());
+ }
+
+ TCHAR *casechars = fromHunspellAndFree(get_casechars(dic_enc));
+ TCHAR *try_string = fromHunspellAndFree(hunspell->get_try_string());
+
+ wordChars = merge(merge(casechars, hwordchars), try_string);
+
+ // Make a suggestion to load hunspell internalls
+ char ** words = NULL;
+ int count = hunspell->suggest(&words, "asdf");
+ for (int i = 0; i < count; i++)
+ free(words[i]);
+ if (words != NULL)
+ free(words);
+
+ loadCustomDict();
+
+ loaded = LANGUAGE_LOADED;
+ }
+
+ // Return TRUE if the word is correct
+ virtual BOOL spell(const TCHAR *word)
+ {
+ load();
+ if (loaded != LANGUAGE_LOADED)
+ return TRUE;
+
+ // TODO Check if it was generated by auto-replacement
+
+ char hunspell_word[1024];
+ toHunspell(hunspell_word, word, SIZEOF(hunspell_word));
+
+ return hunspell->spell(hunspell_word);
+ }
+
+ // Return a list of suggestions to a word
+ virtual Suggestions suggest(const TCHAR * word)
+ {
+ Suggestions ret = { 0 };
+
+ load();
+ if (loaded != LANGUAGE_LOADED)
+ return ret;
+
+ char hunspell_word[1024];
+ toHunspell(hunspell_word, word, SIZEOF(hunspell_word));
+
+ char ** words = NULL;
+ ret.count = hunspell->suggest(&words, hunspell_word);
+
+ if (ret.count > 0 && words != NULL) {
+ // Oki, lets make our array
+ ret.words = (TCHAR **)malloc(ret.count * sizeof(TCHAR *));
+ for (unsigned i = 0; i < ret.count; i++) {
+ ret.words[i] = fromHunspell(words[i]);
+ free(words[i]);
+ }
+ }
+ if (words != NULL)
+ free(words);
+
+ return ret;
+ }
+
+ // Return a list of auto suggestions to a word
+ virtual Suggestions autoSuggest(const TCHAR * word)
+ {
+ Suggestions ret = { 0 };
+
+ load();
+ if (loaded != LANGUAGE_LOADED)
+ return ret;
+
+ char hunspell_word[1024];
+ toHunspell(hunspell_word, word, SIZEOF(hunspell_word));
+
+ char ** words;
+ int count = hunspell->suggest_auto(&words, hunspell_word);
+
+ if (count <= 0)
+ return ret;
+
+ // Oki, lets make our array
+ ret.count = count;
+ ret.words = (TCHAR **)malloc(ret.count * sizeof(TCHAR *));
+ for (int i = 0; i < count; i++) {
+ ret.words[i] = fromHunspell(words[i]);
+ free(words[i]);
+ }
+ free(words);
+
+ return ret;
+ }
+
+ // Return a list of auto suggestions to a word
+ // You have to free the list AND each item
+ virtual TCHAR * autoSuggestOne(const TCHAR * word)
+ {
+ load();
+ if (loaded != LANGUAGE_LOADED)
+ return NULL;
+
+ char hunspell_word[1024];
+ toHunspell(hunspell_word, word, SIZEOF(hunspell_word));
+
+ char ** words;
+ int count = hunspell->suggest_auto(&words, hunspell_word);
+
+ if (count <= 0)
+ return NULL;
+
+ TCHAR *ret = fromHunspell(words[0]);
+
+ // Oki, lets make our array
+ for (int i = 0; i < count; i++)
+ free(words[i]);
+ free(words);
+
+ return ret;
+ }
+
+ // Return TRUE if the char is a word char
+ virtual BOOL isWordChar(TCHAR c)
+ {
+ if (c == 0)
+ return FALSE;
+
+ load();
+ if (loaded != LANGUAGE_LOADED)
+ return TRUE;
+
+ return _tcschr(wordChars, (_TINT)c) != NULL;
+ }
+
+ // Assert that all needed data is loaded
+ virtual void load()
+ {
+ if (loaded == LANGUAGE_NOT_LOADED) {
+ loaded = LANGUAGE_LOADING;
+ mir_forkthread(LoadThread, this);
+ }
+ }
+
+ virtual BOOL isLoaded()
+ {
+ return loaded == LANGUAGE_LOADED;
+ }
+
+
+ // Add a word to the user custom dict
+ virtual void addWord(const TCHAR * word)
+ {
+ addWordInternal(word);
+ appendToCustomDict(word);
+ }
+
+ // Add a word to the list of ignored words
+ virtual void ignoreWord(const TCHAR * word)
+ {
+ addWordInternal(word);
+ }
+};
+
+void LoadThread(LPVOID hd)
+{
+ HunspellDictionary *dict = (HunspellDictionary *)hd;
+ dict->loadThread();
+}
+
+
+
+// To use with EnumLocalesProc :(
+LIST<Dictionary> *tmp_dicts;
+
+// To get the names of the languages
+BOOL CALLBACK EnumLocalesProc(LPTSTR lpLocaleString)
+{
+ TCHAR *stopped = NULL;
+ USHORT langID = (USHORT)_tcstol(lpLocaleString, &stopped, 16);
+
+ TCHAR ini[32];
+ TCHAR end[32];
+ GetLocaleInfo(MAKELCID(langID, 0), LOCALE_SISO639LANGNAME, ini, SIZEOF(ini));
+ GetLocaleInfo(MAKELCID(langID, 0), LOCALE_SISO3166CTRYNAME, end, SIZEOF(end));
+
+ TCHAR name[64];
+ mir_sntprintf(name, SIZEOF(name), _T("%s_%s"), ini, end);
+
+ for (int i = 0; i < tmp_dicts->getCount(); i++) {
+ Dictionary *dict = (*tmp_dicts)[i];
+ if (lstrcmpi(dict->language, name) == 0) {
+ GetLocaleInfo(MAKELCID(langID, 0), LOCALE_SENGLANGUAGE, dict->english_name, SIZEOF(dict->english_name));
+
+ GetLocaleInfo(MAKELCID(langID, 0), LOCALE_SLANGUAGE, dict->localized_name, SIZEOF(dict->localized_name));
+ if (dict->localized_name[0] == 0)
+ GetLocaleInfo(MAKELCID(langID, 0), LOCALE_SLOCALIZEDLANGUAGENAME, dict->localized_name, SIZEOF(dict->localized_name));
+ if (dict->localized_name[0] == 0)
+ GetLocaleInfo(MAKELCID(langID, 0), LOCALE_SNATIVEDISPLAYNAME, dict->localized_name, SIZEOF(dict->localized_name));
+ if (dict->localized_name[0] == 0 && dict->english_name[0] != 0) {
+ TCHAR country[1024];
+ GetLocaleInfo(MAKELCID(langID, 0), LOCALE_SENGCOUNTRY, country, SIZEOF(country));
+
+ TCHAR name[1024];
+ if (country[0] != 0)
+ mir_sntprintf(name, SIZEOF(name), _T("%s (%s)"), dict->english_name, country);
+ else
+ lstrcpyn(name, dict->english_name, SIZEOF(name));
+
+ lstrcpyn(dict->localized_name, TranslateTS(name), SIZEOF(dict->localized_name));
+ }
+
+ if (dict->localized_name[0] != 0) {
+ mir_sntprintf(dict->full_name, SIZEOF(dict->full_name), _T("%s [%s]"), dict->localized_name, dict->language);
+ }
+ break;
+ }
+ }
+ return TRUE;
+}
+
+
+void GetDictsInfo(LIST<Dictionary> &dicts)
+{
+ tmp_dicts = &dicts;
+ EnumSystemLocales(EnumLocalesProc, LCID_SUPPORTED);
+
+ // Try to get name from DB
+ for (int i = 0; i < dicts.getCount(); i++) {
+ Dictionary *dict = dicts[i];
+
+ if (dict->full_name[0] == _T('\0')) {
+ DBVARIANT dbv;
+
+ char lang[128];
+ WideCharToMultiByte(CP_ACP, 0, dict->language, -1, lang, sizeof(lang), NULL, NULL);
+ if (!db_get_ts(NULL, MODULE_NAME, lang, &dbv)) {
+ lstrcpyn(dict->localized_name, dbv.ptszVal, SIZEOF(dict->localized_name));
+ db_free(&dbv);
+ }
+
+ if (dict->localized_name[0] == _T('\0')) {
+ for (size_t j = 0; j < SIZEOF(aditionalLanguages); j++) {
+ if (!lstrcmp(aditionalLanguages[j].language, dict->language)) {
+ lstrcpyn(dict->localized_name, TranslateTS(aditionalLanguages[j].localized_name), SIZEOF(dict->localized_name));
+ break;
+ }
+ }
+ }
+
+ if (dict->localized_name[0] != _T('\0')) {
+ mir_sntprintf(dict->full_name, SIZEOF(dict->full_name), _T("%s [%s]"), dict->localized_name, dict->language);
+ }
+ else {
+ lstrcpyn(dict->full_name, dict->language, SIZEOF(dict->full_name));
+ }
+ }
+ }
+}
+
+
+void GetHunspellDictionariesFromFolder(LIST<Dictionary> &dicts, TCHAR *path, TCHAR *user_path, TCHAR *source)
+{
+ // Load the language files and create an array with then
+ TCHAR file[1024];
+ mir_sntprintf(file, SIZEOF(file), _T("%s\\*.dic"), path);
+
+ BOOL found = FALSE;
+
+ WIN32_FIND_DATA ffd = { 0 };
+ HANDLE hFFD = FindFirstFile(file, &ffd);
+ if (hFFD != INVALID_HANDLE_VALUE) {
+ do {
+ mir_sntprintf(file, SIZEOF(file), _T("%s\\%s"), path, ffd.cFileName);
+
+ // Check .dic
+ DWORD attrib = GetFileAttributes(file);
+ if (attrib == 0xFFFFFFFF || (attrib & FILE_ATTRIBUTE_DIRECTORY))
+ continue;
+
+ // See if .aff exists too
+ lstrcpy(&file[lstrlen(file) - 4], _T(".aff"));
+ attrib = GetFileAttributes(file);
+ if (attrib == 0xFFFFFFFF || (attrib & FILE_ATTRIBUTE_DIRECTORY))
+ continue;
+
+ ffd.cFileName[lstrlen(ffd.cFileName) - 4] = _T('\0');
+
+ TCHAR *lang = ffd.cFileName;
+
+ // Replace - for _
+ for (int i = 0; i < lstrlen(lang); i++)
+ if (lang[i] == _T('-'))
+ lang[i] = _T('_');
+
+ // Check if dict is new
+ bool exists = false;
+ for (int i = 0; i < dicts.getCount() && !exists; i++)
+ if (lstrcmp(dicts[i]->language, lang) == 0)
+ exists = true;
+
+ if (!exists) {
+ found = TRUE;
+ file[lstrlen(file) - 4] = _T('\0');
+ dicts.insert(new HunspellDictionary(lang, file, user_path, source));
+ }
+ }
+ while (FindNextFile(hFFD, &ffd));
+
+ FindClose(hFFD);
+ }
+}
+
+
+// Return a list of avaible languages
+void GetAvaibleDictionaries(LIST<Dictionary> &dicts, TCHAR *path, TCHAR *user_path)
+{
+ // Get miranda folder dicts
+ GetHunspellDictionariesFromFolder(dicts, path, user_path, NULL);
+
+ if (opts.use_other_apps_dicts) {
+ // Get other apps dicts
+ for (int i = 0; i < SIZEOF(otherHunspellApps); i++) {
+ TCHAR key[1024];
+ mir_sntprintf(key, SIZEOF(key), APPPATH, otherHunspellApps[i].key);
+
+ HKEY hKey = 0;
+ LONG lResult = 0;
+ if (ERROR_SUCCESS == RegOpenKeyEx(HKEY_LOCAL_MACHINE, key, 0, KEY_QUERY_VALUE, &hKey)) {
+ DWORD size = SIZEOF(key);
+ lResult = RegQueryValueEx(hKey, _T("Path"), NULL, NULL, (LPBYTE)key, &size);
+ RegCloseKey(hKey);
+ }
+ else {
+ // Not found in installed apps - Try MUICache
+ lResult = RegOpenKeyEx(HKEY_CURRENT_USER, MUICACHE, 0, KEY_QUERY_VALUE, &hKey);
+ if (ERROR_SUCCESS == lResult) {
+ DWORD numValues;
+ if (ERROR_SUCCESS != RegQueryInfoKey(hKey, NULL, NULL, NULL, NULL, NULL, NULL, &numValues, NULL, NULL, NULL, NULL))
+ numValues = 0;
+
+ lResult = ERROR_NO_MORE_ITEMS;
+ for (DWORD local = 0; local < numValues; local++) {
+ DWORD cchValue = SIZEOF(key);
+ if (ERROR_SUCCESS != RegEnumValue(hKey, local, key, &cchValue, NULL, NULL, NULL, NULL))
+ break;
+ key[cchValue] = 0;
+ TCHAR *pos;
+ if (pos = _tcsrchr(key, _T('\\'))) {
+ if (!lstrcmpi(&pos[1], otherHunspellApps[i].key)) {
+ pos[0] = 0;
+ lResult = ERROR_SUCCESS;
+ break;
+ }
+ }
+ }
+ RegCloseKey(hKey);
+ }
+ }
+
+ if (ERROR_SUCCESS == lResult) {
+ TCHAR folder[1024];
+ mir_sntprintf(folder, SIZEOF(folder), _T("%s\\Dictionaries"), key);
+
+ GetHunspellDictionariesFromFolder(languages, folder, user_path, otherHunspellApps[i].name);
+ }
+ }
+ }
+
+ GetDictsInfo(dicts);
+
+ // Yeah, yeah, yeah, I know, but this is the easiest way...
+ SortedList *sl = (SortedList *)&dicts;
+
+ // Sort dicts
+ for (int i = 0; i < dicts.getCount(); i++) {
+ for (int j = i + 1; j < dicts.getCount(); j++) {
+ if (lstrcmp(dicts[i]->full_name, dicts[j]->full_name) > 0) {
+ Dictionary *dict = dicts[i];
+ sl->items[i] = dicts[j];
+ sl->items[j] = dict;
+ }
+ }
+ }
+}
+
+
+// Free the list returned by GetAvaibleDictionaries
+void FreeDictionaries(LIST<Dictionary> &dicts)
+{
+ for (int i = 0; i < dicts.getCount(); i++)
+ delete dicts[i];
+
+ dicts.destroy();
+}
+
+Dictionary::~Dictionary()
+{
+ delete autoReplace;
+}
+
+// Free the list returned by GetAvaibleDictionaries
+void FreeSuggestions(Suggestions &suggestions)
+{
+ for (size_t i = 0; i < suggestions.count; i++)
+ free(suggestions.words[i]);
+
+ free(suggestions.words);
+
+ suggestions.words = NULL;
+ suggestions.count = 0;
+}
diff --git a/plugins/SpellChecker/src/hunspell/csutil.cxx b/plugins/SpellChecker/src/hunspell/csutil.cxx
index cf24bc06dd..ee78edbc40 100644
--- a/plugins/SpellChecker/src/hunspell/csutil.cxx
+++ b/plugins/SpellChecker/src/hunspell/csutil.cxx
@@ -343,6 +343,9 @@ char * line_uniq(char * text, char breakchar) {
char ** lines;
int linenum = line_tok(text, &lines, breakchar);
int i;
+
+ if (linenum == 0)
+ return NULL;
strcpy(text, lines[0]);
for ( i = 1; i < linenum; i++ ) {
int dup = 0;
diff --git a/plugins/SpellChecker/src/hunspell/suggestmgr.cxx b/plugins/SpellChecker/src/hunspell/suggestmgr.cxx
index d89630849c..843075e62f 100644
--- a/plugins/SpellChecker/src/hunspell/suggestmgr.cxx
+++ b/plugins/SpellChecker/src/hunspell/suggestmgr.cxx
@@ -1578,7 +1578,8 @@ char * SuggestMgr::suggest_morph_for_spelling_error(const char * word)
{
char * p = NULL;
char ** wlst = (char **) calloc(maxSug, sizeof(char *));
- if (!**wlst) return NULL;
+ if (!wlst)
+ return NULL;
// we will use only the first suggestion
for (int i = 0; i < maxSug - 1; i++) wlst[i] = "";
int ns = suggest(&wlst, word, maxSug - 1, NULL);
@@ -1586,7 +1587,7 @@ char * SuggestMgr::suggest_morph_for_spelling_error(const char * word)
p = suggest_morph(wlst[maxSug - 1]);
free(wlst[maxSug - 1]);
}
- if (wlst) free(wlst);
+ free(wlst);
return p;
}
#endif // END OF HUNSPELL_EXPERIMENTAL CODE