xref: /vim-8.2.3635/src/mbyte.c (revision 899dddf8)
1 /* vi:set ts=8 sts=4 sw=4:
2  *
3  * VIM - Vi IMproved	by Bram Moolenaar
4  * Multibyte extensions partly by Sung-Hoon Baek
5  *
6  * Do ":help uganda"  in Vim to read copying and usage conditions.
7  * Do ":help credits" in Vim to see a list of people who contributed.
8  * See README.txt for an overview of the Vim source code.
9  */
10 /*
11  * mbyte.c: Code specifically for handling multi-byte characters.
12  *
13  * The encoding used in the core is set with 'encoding'.  When 'encoding' is
14  * changed, the following four variables are set (for speed).
15  * Currently these types of character encodings are supported:
16  *
17  * "enc_dbcs"	    When non-zero it tells the type of double byte character
18  *		    encoding (Chinese, Korean, Japanese, etc.).
19  *		    The cell width on the display is equal to the number of
20  *		    bytes.  (exception: DBCS_JPNU with first byte 0x8e)
21  *		    Recognizing the first or second byte is difficult, it
22  *		    requires checking a byte sequence from the start.
23  * "enc_utf8"	    When TRUE use Unicode characters in UTF-8 encoding.
24  *		    The cell width on the display needs to be determined from
25  *		    the character value.
26  *		    Recognizing bytes is easy: 0xxx.xxxx is a single-byte
27  *		    char, 10xx.xxxx is a trailing byte, 11xx.xxxx is a leading
28  *		    byte of a multi-byte character.
29  *		    To make things complicated, up to two composing characters
30  *		    are allowed.  These are drawn on top of the first char.
31  *		    For most editing the sequence of bytes with composing
32  *		    characters included is considered to be one character.
33  * "enc_unicode"    When 2 use 16-bit Unicode characters (or UTF-16).
34  *		    When 4 use 32-but Unicode characters.
35  *		    Internally characters are stored in UTF-8 encoding to
36  *		    avoid NUL bytes.  Conversion happens when doing I/O.
37  *		    "enc_utf8" will also be TRUE.
38  *
39  * "has_mbyte" is set when "enc_dbcs" or "enc_utf8" is non-zero.
40  *
41  * If none of these is TRUE, 8-bit bytes are used for a character.  The
42  * encoding isn't currently specified (TODO).
43  *
44  * 'encoding' specifies the encoding used in the core.  This is in registers,
45  * text manipulation, buffers, etc.  Conversion has to be done when characters
46  * in another encoding are received or send:
47  *
48  *		       clipboard
49  *			   ^
50  *			   | (2)
51  *			   V
52  *		   +---------------+
53  *	      (1)  |		   | (3)
54  *  keyboard ----->|	 core	   |-----> display
55  *		   |		   |
56  *		   +---------------+
57  *			   ^
58  *			   | (4)
59  *			   V
60  *			 file
61  *
62  * (1) Typed characters arrive in the current locale.  Conversion is to be
63  *     done when 'encoding' is different from 'termencoding'.
64  * (2) Text will be made available with the encoding specified with
65  *     'encoding'.  If this is not sufficient, system-specific conversion
66  *     might be required.
67  * (3) For the GUI the correct font must be selected, no conversion done.
68  *     Otherwise, conversion is to be done when 'encoding' differs from
69  *     'termencoding'.  (Different in the GTK+ 2 port -- 'termencoding'
70  *     is always used for both input and output and must always be set to
71  *     "utf-8".  gui_mch_init() does this automatically.)
72  * (4) The encoding of the file is specified with 'fileencoding'.  Conversion
73  *     is to be done when it's different from 'encoding'.
74  *
75  * The viminfo file is a special case: Only text is converted, not file names.
76  * Vim scripts may contain an ":encoding" command.  This has an effect for
77  * some commands, like ":menutrans"
78  */
79 
80 #include "vim.h"
81 
82 #ifdef WIN32UNIX
83 # ifndef WIN32_LEAN_AND_MEAN
84 #  define WIN32_LEAN_AND_MEAN
85 # endif
86 # include <windows.h>
87 # ifdef WIN32
88 #  undef WIN32	    /* Some windows.h define WIN32, we don't want that here. */
89 # endif
90 #endif
91 
92 #if (defined(WIN3264) || defined(WIN32UNIX)) && !defined(__MINGW32__)
93 # include <winnls.h>
94 #endif
95 
96 #ifdef FEAT_GUI_X11
97 # include <X11/Intrinsic.h>
98 #endif
99 #ifdef X_LOCALE
100 #include <X11/Xlocale.h>
101 #endif
102 
103 #if defined(FEAT_GUI_GTK) && defined(FEAT_XIM) && defined(HAVE_GTK2)
104 # include <gdk/gdkkeysyms.h>
105 # ifdef WIN3264
106 #  include <gdk/gdkwin32.h>
107 # else
108 #  include <gdk/gdkx.h>
109 # endif
110 #endif
111 
112 #ifdef HAVE_WCHAR_H
113 # include <wchar.h>
114 #endif
115 
116 #if 0
117 /* This has been disabled, because several people reported problems with the
118  * wcwidth() and iswprint() library functions, esp. for Hebrew. */
119 # ifdef __STDC_ISO_10646__
120 #  define USE_WCHAR_FUNCTIONS
121 # endif
122 #endif
123 
124 #if defined(FEAT_MBYTE) || defined(PROTO)
125 
126 static int enc_canon_search __ARGS((char_u *name));
127 static int dbcs_char2len __ARGS((int c));
128 static int dbcs_char2bytes __ARGS((int c, char_u *buf));
129 static int dbcs_ptr2len __ARGS((char_u *p));
130 static int dbcs_char2cells __ARGS((int c));
131 static int dbcs_ptr2char __ARGS((char_u *p));
132 
133 /* Lookup table to quickly get the length in bytes of a UTF-8 character from
134  * the first byte of a UTF-8 string.  Bytes which are illegal when used as the
135  * first byte have a one, because these will be used separately. */
136 static char utf8len_tab[256] =
137 {
138     1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,
139     1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,
140     1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,
141     1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,
142     1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /*bogus*/
143     1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /*bogus*/
144     2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,
145     3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,4,4,4,4,4,4,4,4,5,5,5,5,6,6,1,1,
146 };
147 
148 /*
149  * XIM often causes trouble.  Define XIM_DEBUG to get a log of XIM callbacks
150  * in the "xim.log" file.
151  */
152 /* #define XIM_DEBUG */
153 #ifdef XIM_DEBUG
154     static void
155 xim_log(char *s, ...)
156 {
157     va_list arglist;
158     static FILE *fd = NULL;
159 
160     if (fd == (FILE *)-1)
161 	return;
162     if (fd == NULL)
163     {
164 	fd = mch_fopen("xim.log", "w");
165 	if (fd == NULL)
166 	{
167 	    EMSG("Cannot open xim.log");
168 	    fd = (FILE *)-1;
169 	    return;
170 	}
171     }
172 
173     va_start(arglist, s);
174     vfprintf(fd, s, arglist);
175     va_end(arglist);
176 }
177 #endif
178 
179 #endif
180 
181 #if defined(FEAT_MBYTE) || defined(FEAT_POSTSCRIPT) || defined(PROTO)
182 /*
183  * Canonical encoding names and their properties.
184  * "iso-8859-n" is handled by enc_canonize() directly.
185  */
186 static struct
187 {   char *name;		int prop;		int codepage;}
188 enc_canon_table[] =
189 {
190 #define IDX_LATIN_1	0
191     {"latin1",		ENC_8BIT + ENC_LATIN1,	1252},
192 #define IDX_ISO_2	1
193     {"iso-8859-2",	ENC_8BIT,		0},
194 #define IDX_ISO_3	2
195     {"iso-8859-3",	ENC_8BIT,		0},
196 #define IDX_ISO_4	3
197     {"iso-8859-4",	ENC_8BIT,		0},
198 #define IDX_ISO_5	4
199     {"iso-8859-5",	ENC_8BIT,		0},
200 #define IDX_ISO_6	5
201     {"iso-8859-6",	ENC_8BIT,		0},
202 #define IDX_ISO_7	6
203     {"iso-8859-7",	ENC_8BIT,		0},
204 #define IDX_ISO_8	7
205     {"iso-8859-8",	ENC_8BIT,		0},
206 #define IDX_ISO_9	8
207     {"iso-8859-9",	ENC_8BIT,		0},
208 #define IDX_ISO_10	9
209     {"iso-8859-10",	ENC_8BIT,		0},
210 #define IDX_ISO_11	10
211     {"iso-8859-11",	ENC_8BIT,		0},
212 #define IDX_ISO_13	11
213     {"iso-8859-13",	ENC_8BIT,		0},
214 #define IDX_ISO_14	12
215     {"iso-8859-14",	ENC_8BIT,		0},
216 #define IDX_ISO_15	13
217     {"iso-8859-15",	ENC_8BIT + ENC_LATIN9,	0},
218 #define IDX_KOI8_R	14
219     {"koi8-r",		ENC_8BIT,		0},
220 #define IDX_KOI8_U	15
221     {"koi8-u",		ENC_8BIT,		0},
222 #define IDX_UTF8	16
223     {"utf-8",		ENC_UNICODE,		0},
224 #define IDX_UCS2	17
225     {"ucs-2",		ENC_UNICODE + ENC_ENDIAN_B + ENC_2BYTE, 0},
226 #define IDX_UCS2LE	18
227     {"ucs-2le",		ENC_UNICODE + ENC_ENDIAN_L + ENC_2BYTE, 0},
228 #define IDX_UTF16	19
229     {"utf-16",		ENC_UNICODE + ENC_ENDIAN_B + ENC_2WORD, 0},
230 #define IDX_UTF16LE	20
231     {"utf-16le",	ENC_UNICODE + ENC_ENDIAN_L + ENC_2WORD, 0},
232 #define IDX_UCS4	21
233     {"ucs-4",		ENC_UNICODE + ENC_ENDIAN_B + ENC_4BYTE, 0},
234 #define IDX_UCS4LE	22
235     {"ucs-4le",		ENC_UNICODE + ENC_ENDIAN_L + ENC_4BYTE, 0},
236 
237     /* For debugging DBCS encoding on Unix. */
238 #define IDX_DEBUG	23
239     {"debug",		ENC_DBCS,		DBCS_DEBUG},
240 #define IDX_EUC_JP	24
241     {"euc-jp",		ENC_DBCS,		DBCS_JPNU},
242 #define IDX_SJIS	25
243     {"sjis",		ENC_DBCS,		DBCS_JPN},
244 #define IDX_EUC_KR	26
245     {"euc-kr",		ENC_DBCS,		DBCS_KORU},
246 #define IDX_EUC_CN	27
247     {"euc-cn",		ENC_DBCS,		DBCS_CHSU},
248 #define IDX_EUC_TW	28
249     {"euc-tw",		ENC_DBCS,		DBCS_CHTU},
250 #define IDX_BIG5	29
251     {"big5",		ENC_DBCS,		DBCS_CHT},
252 
253     /* MS-DOS and MS-Windows codepages are included here, so that they can be
254      * used on Unix too.  Most of them are similar to ISO-8859 encodings, but
255      * not exactly the same. */
256 #define IDX_CP437	30
257     {"cp437",		ENC_8BIT,		437}, /* like iso-8859-1 */
258 #define IDX_CP737	31
259     {"cp737",		ENC_8BIT,		737}, /* like iso-8859-7 */
260 #define IDX_CP775	32
261     {"cp775",		ENC_8BIT,		775}, /* Baltic */
262 #define IDX_CP850	33
263     {"cp850",		ENC_8BIT,		850}, /* like iso-8859-4 */
264 #define IDX_CP852	34
265     {"cp852",		ENC_8BIT,		852}, /* like iso-8859-1 */
266 #define IDX_CP855	35
267     {"cp855",		ENC_8BIT,		855}, /* like iso-8859-2 */
268 #define IDX_CP857	36
269     {"cp857",		ENC_8BIT,		857}, /* like iso-8859-5 */
270 #define IDX_CP860	37
271     {"cp860",		ENC_8BIT,		860}, /* like iso-8859-9 */
272 #define IDX_CP861	38
273     {"cp861",		ENC_8BIT,		861}, /* like iso-8859-1 */
274 #define IDX_CP862	39
275     {"cp862",		ENC_8BIT,		862}, /* like iso-8859-1 */
276 #define IDX_CP863	40
277     {"cp863",		ENC_8BIT,		863}, /* like iso-8859-8 */
278 #define IDX_CP865	41
279     {"cp865",		ENC_8BIT,		865}, /* like iso-8859-1 */
280 #define IDX_CP866	42
281     {"cp866",		ENC_8BIT,		866}, /* like iso-8859-5 */
282 #define IDX_CP869	43
283     {"cp869",		ENC_8BIT,		869}, /* like iso-8859-7 */
284 #define IDX_CP874	44
285     {"cp874",		ENC_8BIT,		874}, /* Thai */
286 #define IDX_CP932	45
287     {"cp932",		ENC_DBCS,		DBCS_JPN},
288 #define IDX_CP936	46
289     {"cp936",		ENC_DBCS,		DBCS_CHS},
290 #define IDX_CP949	47
291     {"cp949",		ENC_DBCS,		DBCS_KOR},
292 #define IDX_CP950	48
293     {"cp950",		ENC_DBCS,		DBCS_CHT},
294 #define IDX_CP1250	49
295     {"cp1250",		ENC_8BIT,		1250}, /* Czech, Polish, etc. */
296 #define IDX_CP1251	50
297     {"cp1251",		ENC_8BIT,		1251}, /* Cyrillic */
298     /* cp1252 is considered to be equal to latin1 */
299 #define IDX_CP1253	51
300     {"cp1253",		ENC_8BIT,		1253}, /* Greek */
301 #define IDX_CP1254	52
302     {"cp1254",		ENC_8BIT,		1254}, /* Turkish */
303 #define IDX_CP1255	53
304     {"cp1255",		ENC_8BIT,		1255}, /* Hebrew */
305 #define IDX_CP1256	54
306     {"cp1256",		ENC_8BIT,		1256}, /* Arabic */
307 #define IDX_CP1257	55
308     {"cp1257",		ENC_8BIT,		1257}, /* Baltic */
309 #define IDX_CP1258	56
310     {"cp1258",		ENC_8BIT,		1258}, /* Vietnamese */
311 
312 #define IDX_MACROMAN	57
313     {"macroman",	ENC_8BIT + ENC_MACROMAN, 0},	/* Mac OS */
314 #define IDX_COUNT	58
315 };
316 
317 /*
318  * Aliases for encoding names.
319  */
320 static struct
321 {   char *name;		int canon;}
322 enc_alias_table[] =
323 {
324     {"ansi",		IDX_LATIN_1},
325     {"iso-8859-1",	IDX_LATIN_1},
326     {"latin2",		IDX_ISO_2},
327     {"latin3",		IDX_ISO_3},
328     {"latin4",		IDX_ISO_4},
329     {"cyrillic",	IDX_ISO_5},
330     {"arabic",		IDX_ISO_6},
331     {"greek",		IDX_ISO_7},
332 #ifdef WIN3264
333     {"hebrew",		IDX_CP1255},
334 #else
335     {"hebrew",		IDX_ISO_8},
336 #endif
337     {"latin5",		IDX_ISO_9},
338     {"turkish",		IDX_ISO_9}, /* ? */
339     {"latin6",		IDX_ISO_10},
340     {"nordic",		IDX_ISO_10}, /* ? */
341     {"thai",		IDX_ISO_11}, /* ? */
342     {"latin7",		IDX_ISO_13},
343     {"latin8",		IDX_ISO_14},
344     {"latin9",		IDX_ISO_15},
345     {"utf8",		IDX_UTF8},
346     {"unicode",		IDX_UCS2},
347     {"ucs2",		IDX_UCS2},
348     {"ucs2be",		IDX_UCS2},
349     {"ucs-2be",		IDX_UCS2},
350     {"ucs2le",		IDX_UCS2LE},
351     {"utf16",		IDX_UTF16},
352     {"utf16be",		IDX_UTF16},
353     {"utf-16be",	IDX_UTF16},
354     {"utf16le",		IDX_UTF16LE},
355     {"ucs4",		IDX_UCS4},
356     {"ucs4be",		IDX_UCS4},
357     {"ucs-4be",		IDX_UCS4},
358     {"ucs4le",		IDX_UCS4LE},
359     {"932",		IDX_CP932},
360     {"949",		IDX_CP949},
361     {"936",		IDX_CP936},
362     {"950",		IDX_CP950},
363     {"eucjp",		IDX_EUC_JP},
364     {"unix-jis",	IDX_EUC_JP},
365     {"ujis",		IDX_EUC_JP},
366     {"shift-jis",	IDX_SJIS},
367     {"euckr",		IDX_EUC_KR},
368     {"5601",		IDX_EUC_KR},	/* Sun: KS C 5601 */
369     {"euccn",		IDX_EUC_CN},
370     {"gb2312",		IDX_EUC_CN},
371     {"euctw",		IDX_EUC_TW},
372 #if defined(WIN3264) || defined(WIN32UNIX) || defined(MACOS)
373     {"japan",		IDX_CP932},
374     {"korea",		IDX_CP949},
375     {"prc",		IDX_CP936},
376     {"chinese",		IDX_CP936},
377     {"taiwan",		IDX_CP950},
378     {"big5",		IDX_CP950},
379 #else
380     {"japan",		IDX_EUC_JP},
381     {"korea",		IDX_EUC_KR},
382     {"prc",		IDX_EUC_CN},
383     {"chinese",		IDX_EUC_CN},
384     {"taiwan",		IDX_EUC_TW},
385     {"cp950",		IDX_BIG5},
386     {"950",		IDX_BIG5},
387 #endif
388     {"mac",		IDX_MACROMAN},
389     {NULL,		0}
390 };
391 
392 #ifndef CP_UTF8
393 # define CP_UTF8 65001	/* magic number from winnls.h */
394 #endif
395 
396 /*
397  * Find encoding "name" in the list of canonical encoding names.
398  * Returns -1 if not found.
399  */
400     static int
401 enc_canon_search(name)
402     char_u	*name;
403 {
404     int		i;
405 
406     for (i = 0; i < IDX_COUNT; ++i)
407 	if (STRCMP(name, enc_canon_table[i].name) == 0)
408 	    return i;
409     return -1;
410 }
411 
412 #endif
413 
414 #if defined(FEAT_MBYTE) || defined(PROTO)
415 
416 /*
417  * Find canonical encoding "name" in the list and return its properties.
418  * Returns 0 if not found.
419  */
420     int
421 enc_canon_props(name)
422     char_u	*name;
423 {
424     int		i;
425 
426     i = enc_canon_search(name);
427     if (i >= 0)
428 	return enc_canon_table[i].prop;
429 #ifdef WIN3264
430     if (name[0] == 'c' && name[1] == 'p' && VIM_ISDIGIT(name[2]))
431     {
432 	CPINFO	cpinfo;
433 
434 	/* Get info on this codepage to find out what it is. */
435 	if (GetCPInfo(atoi(name + 2), &cpinfo) != 0)
436 	{
437 	    if (cpinfo.MaxCharSize == 1) /* some single-byte encoding */
438 		return ENC_8BIT;
439 	    if (cpinfo.MaxCharSize == 2
440 		    && (cpinfo.LeadByte[0] != 0 || cpinfo.LeadByte[1] != 0))
441 		/* must be a DBCS encoding */
442 		return ENC_DBCS;
443 	}
444 	return 0;
445     }
446 #endif
447     if (STRNCMP(name, "2byte-", 6) == 0)
448 	return ENC_DBCS;
449     if (STRNCMP(name, "8bit-", 5) == 0 || STRNCMP(name, "iso-8859-", 9) == 0)
450 	return ENC_8BIT;
451     return 0;
452 }
453 
454 /*
455  * Set up for using multi-byte characters.
456  * Called in three cases:
457  * - by main() to initialize (p_enc == NULL)
458  * - by set_init_1() after 'encoding' was set to its default.
459  * - by do_set() when 'encoding' has been set.
460  * p_enc must have been passed through enc_canonize() already.
461  * Sets the "enc_unicode", "enc_utf8", "enc_dbcs" and "has_mbyte" flags.
462  * Fills mb_bytelen_tab[] and returns NULL when there are no problems.
463  * When there is something wrong: Returns an error message and doesn't change
464  * anything.
465  */
466     char_u *
467 mb_init()
468 {
469     int		i;
470     int		idx;
471     int		n;
472     int		enc_dbcs_new = 0;
473 #if defined(USE_ICONV) && !defined(WIN3264) && !defined(WIN32UNIX) \
474 	&& !defined(MACOS)
475 # define LEN_FROM_CONV
476     vimconv_T	vimconv;
477     char_u	*p;
478 #endif
479 
480     if (p_enc == NULL)
481     {
482 	/* Just starting up: set the whole table to one's. */
483 	for (i = 0; i < 256; ++i)
484 	    mb_bytelen_tab[i] = 1;
485 	input_conv.vc_type = CONV_NONE;
486 	input_conv.vc_factor = 1;
487 	output_conv.vc_type = CONV_NONE;
488 	return NULL;
489     }
490 
491 #ifdef WIN3264
492     if (p_enc[0] == 'c' && p_enc[1] == 'p' && VIM_ISDIGIT(p_enc[2]))
493     {
494 	CPINFO	cpinfo;
495 
496 	/* Get info on this codepage to find out what it is. */
497 	if (GetCPInfo(atoi(p_enc + 2), &cpinfo) != 0)
498 	{
499 	    if (cpinfo.MaxCharSize == 1)
500 	    {
501 		/* some single-byte encoding */
502 		enc_unicode = 0;
503 		enc_utf8 = FALSE;
504 	    }
505 	    else if (cpinfo.MaxCharSize == 2
506 		    && (cpinfo.LeadByte[0] != 0 || cpinfo.LeadByte[1] != 0))
507 	    {
508 		/* must be a DBCS encoding, check below */
509 		enc_dbcs_new = atoi(p_enc + 2);
510 	    }
511 	    else
512 		goto codepage_invalid;
513 	}
514 	else if (GetLastError() == ERROR_INVALID_PARAMETER)
515 	{
516 codepage_invalid:
517 	    return (char_u *)N_("E543: Not a valid codepage");
518 	}
519     }
520 #endif
521     else if (STRNCMP(p_enc, "8bit-", 5) == 0
522 	    || STRNCMP(p_enc, "iso-8859-", 9) == 0)
523     {
524 	/* Accept any "8bit-" or "iso-8859-" name. */
525 	enc_unicode = 0;
526 	enc_utf8 = FALSE;
527     }
528     else if (STRNCMP(p_enc, "2byte-", 6) == 0)
529     {
530 #ifdef WIN3264
531 	/* Windows: accept only valid codepage numbers, check below. */
532 	if (p_enc[6] != 'c' || p_enc[7] != 'p'
533 				      || (enc_dbcs_new = atoi(p_enc + 8)) == 0)
534 	    return e_invarg;
535 #else
536 	/* Unix: accept any "2byte-" name, assume current locale. */
537 	enc_dbcs_new = DBCS_2BYTE;
538 #endif
539     }
540     else if ((idx = enc_canon_search(p_enc)) >= 0)
541     {
542 	i = enc_canon_table[idx].prop;
543 	if (i & ENC_UNICODE)
544 	{
545 	    /* Unicode */
546 	    enc_utf8 = TRUE;
547 	    if (i & (ENC_2BYTE | ENC_2WORD))
548 		enc_unicode = 2;
549 	    else if (i & ENC_4BYTE)
550 		enc_unicode = 4;
551 	    else
552 		enc_unicode = 0;
553 	}
554 	else if (i & ENC_DBCS)
555 	{
556 	    /* 2byte, handle below */
557 	    enc_dbcs_new = enc_canon_table[idx].codepage;
558 	}
559 	else
560 	{
561 	    /* Must be 8-bit. */
562 	    enc_unicode = 0;
563 	    enc_utf8 = FALSE;
564 	}
565     }
566     else    /* Don't know what encoding this is, reject it. */
567 	return e_invarg;
568 
569     if (enc_dbcs_new != 0)
570     {
571 #ifdef WIN3264
572 	/* Check if the DBCS code page is OK. */
573 	if (!IsValidCodePage(enc_dbcs_new))
574 	    goto codepage_invalid;
575 #endif
576 	enc_unicode = 0;
577 	enc_utf8 = FALSE;
578     }
579     enc_dbcs = enc_dbcs_new;
580     has_mbyte = (enc_dbcs != 0 || enc_utf8);
581 
582 #ifdef WIN3264
583     enc_codepage = encname2codepage(p_enc);
584     enc_latin9 = (STRCMP(p_enc, "iso-8859-15") == 0);
585 #endif
586 
587     /* Detect an encoding that uses latin1 characters. */
588     enc_latin1like = (enc_utf8 || STRCMP(p_enc, "latin1") == 0
589 					|| STRCMP(p_enc, "iso-8859-15") == 0);
590 
591     /*
592      * Set the function pointers.
593      */
594     if (enc_utf8)
595     {
596 	mb_ptr2len = utfc_ptr2len;
597 	mb_char2len = utf_char2len;
598 	mb_char2bytes = utf_char2bytes;
599 	mb_ptr2cells = utf_ptr2cells;
600 	mb_char2cells = utf_char2cells;
601 	mb_off2cells = utf_off2cells;
602 	mb_ptr2char = utf_ptr2char;
603 	mb_head_off = utf_head_off;
604     }
605     else if (enc_dbcs != 0)
606     {
607 	mb_ptr2len = dbcs_ptr2len;
608 	mb_char2len = dbcs_char2len;
609 	mb_char2bytes = dbcs_char2bytes;
610 	mb_ptr2cells = dbcs_ptr2cells;
611 	mb_char2cells = dbcs_char2cells;
612 	mb_off2cells = dbcs_off2cells;
613 	mb_ptr2char = dbcs_ptr2char;
614 	mb_head_off = dbcs_head_off;
615     }
616     else
617     {
618 	mb_ptr2len = latin_ptr2len;
619 	mb_char2len = latin_char2len;
620 	mb_char2bytes = latin_char2bytes;
621 	mb_ptr2cells = latin_ptr2cells;
622 	mb_char2cells = latin_char2cells;
623 	mb_off2cells = latin_off2cells;
624 	mb_ptr2char = latin_ptr2char;
625 	mb_head_off = latin_head_off;
626     }
627 
628     /*
629      * Fill the mb_bytelen_tab[] for MB_BYTE2LEN().
630      */
631 #ifdef LEN_FROM_CONV
632     /* When 'encoding' is different from the current locale mblen() won't
633      * work.  Use conversion to "utf-8" instead. */
634     vimconv.vc_type = CONV_NONE;
635     if (enc_dbcs)
636     {
637 	p = enc_locale();
638 	if (p == NULL || STRCMP(p, p_enc) != 0)
639 	{
640 	    convert_setup(&vimconv, p_enc, (char_u *)"utf-8");
641 	    vimconv.vc_fail = TRUE;
642 	}
643 	vim_free(p);
644     }
645 #endif
646 
647     for (i = 0; i < 256; ++i)
648     {
649 	/* Our own function to reliably check the length of UTF-8 characters,
650 	 * independent of mblen(). */
651 	if (enc_utf8)
652 	    n = utf8len_tab[i];
653 	else if (enc_dbcs == 0)
654 	    n = 1;
655 	else
656 	{
657 #if defined(WIN3264) || defined(WIN32UNIX)
658 	    /* enc_dbcs is set by setting 'fileencoding'.  It becomes a Windows
659 	     * CodePage identifier, which we can pass directly in to Windows
660 	     * API */
661 	    n = IsDBCSLeadByteEx(enc_dbcs, (BYTE)i) ? 2 : 1;
662 #else
663 # ifdef MACOS
664 	    /*
665 	     * if mblen() is not available, character which MSB is turned on
666 	     * are treated as leading byte character. (note : This assumption
667 	     * is not always true.)
668 	     */
669 	    n = (i & 0x80) ? 2 : 1;
670 # else
671 	    char buf[MB_MAXBYTES];
672 # ifdef X_LOCALE
673 #  ifndef mblen
674 #   define mblen _Xmblen
675 #  endif
676 # endif
677 	    if (i == NUL)	/* just in case mblen() can't handle "" */
678 		n = 1;
679 	    else
680 	    {
681 		buf[0] = i;
682 		buf[1] = 0;
683 #ifdef LEN_FROM_CONV
684 		if (vimconv.vc_type != CONV_NONE)
685 		{
686 		    /*
687 		     * string_convert() should fail when converting the first
688 		     * byte of a double-byte character.
689 		     */
690 		    p = string_convert(&vimconv, (char_u *)buf, NULL);
691 		    if (p != NULL)
692 		    {
693 			vim_free(p);
694 			n = 1;
695 		    }
696 		    else
697 			n = 2;
698 		}
699 		else
700 #endif
701 		{
702 		    /*
703 		     * mblen() should return -1 for invalid (means the leading
704 		     * multibyte) character.  However there are some platforms
705 		     * where mblen() returns 0 for invalid character.
706 		     * Therefore, following condition includes 0.
707 		     */
708 		    (void)mblen(NULL, 0);	/* First reset the state. */
709 		    if (mblen(buf, (size_t)1) <= 0)
710 			n = 2;
711 		    else
712 			n = 1;
713 		}
714 	    }
715 # endif
716 #endif
717 	}
718 
719 	mb_bytelen_tab[i] = n;
720     }
721 
722 #ifdef LEN_FROM_CONV
723     convert_setup(&vimconv, NULL, NULL);
724 #endif
725 
726     /* The cell width depends on the type of multi-byte characters. */
727     (void)init_chartab();
728 
729     /* When enc_utf8 is set or reset, (de)allocate ScreenLinesUC[] */
730     screenalloc(FALSE);
731 
732     /* When using Unicode, set default for 'fileencodings'. */
733     if (enc_utf8 && !option_was_set((char_u *)"fencs"))
734 	set_string_option_direct((char_u *)"fencs", -1,
735 		       (char_u *)"ucs-bom,utf-8,default,latin1", OPT_FREE, 0);
736 
737 #if defined(HAVE_BIND_TEXTDOMAIN_CODESET) && defined(FEAT_GETTEXT)
738     /* GNU gettext 0.10.37 supports this feature: set the codeset used for
739      * translated messages independently from the current locale. */
740     (void)bind_textdomain_codeset(VIMPACKAGE,
741 					  enc_utf8 ? "utf-8" : (char *)p_enc);
742 #endif
743 
744 #ifdef WIN32
745     /* When changing 'encoding' while starting up, then convert the command
746      * line arguments from the active codepage to 'encoding'. */
747     if (starting != 0)
748 	fix_arg_enc();
749 #endif
750 
751 #ifdef FEAT_AUTOCMD
752     /* Fire an autocommand to let people do custom font setup. This must be
753      * after Vim has been setup for the new encoding. */
754     apply_autocmds(EVENT_ENCODINGCHANGED, NULL, (char_u *)"", FALSE, curbuf);
755 #endif
756 
757 #ifdef FEAT_SPELL
758     /* Need to reload spell dictionaries */
759     spell_reload();
760 #endif
761 
762     return NULL;
763 }
764 
765 /*
766  * Return the size of the BOM for the current buffer:
767  * 0 - no BOM
768  * 2 - UCS-2 or UTF-16 BOM
769  * 4 - UCS-4 BOM
770  * 3 - UTF-8 BOM
771  */
772     int
773 bomb_size()
774 {
775     int n = 0;
776 
777     if (curbuf->b_p_bomb && !curbuf->b_p_bin)
778     {
779 	if (*curbuf->b_p_fenc == NUL)
780 	{
781 	    if (enc_utf8)
782 	    {
783 		if (enc_unicode != 0)
784 		    n = enc_unicode;
785 		else
786 		    n = 3;
787 	    }
788 	}
789 	else if (STRCMP(curbuf->b_p_fenc, "utf-8") == 0)
790 	    n = 3;
791 	else if (STRNCMP(curbuf->b_p_fenc, "ucs-2", 5) == 0
792 		|| STRNCMP(curbuf->b_p_fenc, "utf-16", 6) == 0)
793 	    n = 2;
794 	else if (STRNCMP(curbuf->b_p_fenc, "ucs-4", 5) == 0)
795 	    n = 4;
796     }
797     return n;
798 }
799 
800 /*
801  * Get class of pointer:
802  * 0 for blank or NUL
803  * 1 for punctuation
804  * 2 for an (ASCII) word character
805  * >2 for other word characters
806  */
807     int
808 mb_get_class(p)
809     char_u	*p;
810 {
811     if (MB_BYTE2LEN(p[0]) == 1)
812     {
813 	if (p[0] == NUL || vim_iswhite(p[0]))
814 	    return 0;
815 	if (vim_iswordc(p[0]))
816 	    return 2;
817 	return 1;
818     }
819     if (enc_dbcs != 0 && p[0] != NUL && p[1] != NUL)
820 	return dbcs_class(p[0], p[1]);
821     if (enc_utf8)
822 	return utf_class(utf_ptr2char(p));
823     return 0;
824 }
825 
826 /*
827  * Get class of a double-byte character.  This always returns 3 or bigger.
828  * TODO: Should return 1 for punctuation.
829  */
830     int
831 dbcs_class(lead, trail)
832     unsigned	lead;
833     unsigned	trail;
834 {
835     switch (enc_dbcs)
836     {
837 	/* please add classfy routine for your language in here */
838 
839 	case DBCS_JPNU:	/* ? */
840 	case DBCS_JPN:
841 	    {
842 		/* JIS code classification */
843 		unsigned char lb = lead;
844 		unsigned char tb = trail;
845 
846 		/* convert process code to JIS */
847 # if defined(WIN3264) || defined(WIN32UNIX) || defined(MACOS)
848 		/* process code is SJIS */
849 		if (lb <= 0x9f)
850 		    lb = (lb - 0x81) * 2 + 0x21;
851 		else
852 		    lb = (lb - 0xc1) * 2 + 0x21;
853 		if (tb <= 0x7e)
854 		    tb -= 0x1f;
855 		else if (tb <= 0x9e)
856 		    tb -= 0x20;
857 		else
858 		{
859 		    tb -= 0x7e;
860 		    lb += 1;
861 		}
862 # else
863 		/*
864 		 * XXX: Code page identification can not use with all
865 		 *	    system! So, some other encoding information
866 		 *	    will be needed.
867 		 *	    In japanese: SJIS,EUC,UNICODE,(JIS)
868 		 *	    Note that JIS-code system don't use as
869 		 *	    process code in most system because it uses
870 		 *	    escape sequences(JIS is context depend encoding).
871 		 */
872 		/* assume process code is JAPANESE-EUC */
873 		lb &= 0x7f;
874 		tb &= 0x7f;
875 # endif
876 		/* exceptions */
877 		switch (lb << 8 | tb)
878 		{
879 		    case 0x2121: /* ZENKAKU space */
880 			return 0;
881 		    case 0x2122: /* KU-TEN (Japanese comma) */
882 		    case 0x2123: /* TOU-TEN (Japanese period) */
883 		    case 0x2124: /* ZENKAKU comma */
884 		    case 0x2125: /* ZENKAKU period */
885 			return 1;
886 		    case 0x213c: /* prolongedsound handled as KATAKANA */
887 			return 13;
888 		}
889 		/* sieved by KU code */
890 		switch (lb)
891 		{
892 		    case 0x21:
893 		    case 0x22:
894 			/* special symbols */
895 			return 10;
896 		    case 0x23:
897 			/* alpha-numeric */
898 			return 11;
899 		    case 0x24:
900 			/* hiragana */
901 			return 12;
902 		    case 0x25:
903 			/* katakana */
904 			return 13;
905 		    case 0x26:
906 			/* greek */
907 			return 14;
908 		    case 0x27:
909 			/* russian */
910 			return 15;
911 		    case 0x28:
912 			/* lines */
913 			return 16;
914 		    default:
915 			/* kanji */
916 			return 17;
917 		}
918 	    }
919 
920 	case DBCS_KORU:	/* ? */
921 	case DBCS_KOR:
922 	    {
923 		/* KS code classification */
924 		unsigned char c1 = lead;
925 		unsigned char c2 = trail;
926 
927 		/*
928 		 * 20 : Hangul
929 		 * 21 : Hanja
930 		 * 22 : Symbols
931 		 * 23 : Alpha-numeric/Roman Letter (Full width)
932 		 * 24 : Hangul Letter(Alphabet)
933 		 * 25 : Roman Numeral/Greek Letter
934 		 * 26 : Box Drawings
935 		 * 27 : Unit Symbols
936 		 * 28 : Circled/Parenthesized Letter
937 		 * 29 : Hirigana/Katakana
938 		 * 30 : Cyrillic Letter
939 		 */
940 
941 		if (c1 >= 0xB0 && c1 <= 0xC8)
942 		    /* Hangul */
943 		    return 20;
944 #if defined(WIN3264) || defined(WIN32UNIX)
945 		else if (c1 <= 0xA0 || c2 <= 0xA0)
946 		    /* Extended Hangul Region : MS UHC(Unified Hangul Code) */
947 		    /* c1: 0x81-0xA0 with c2: 0x41-0x5A, 0x61-0x7A, 0x81-0xFE
948 		     * c1: 0xA1-0xC6 with c2: 0x41-0x5A, 0x61-0x7A, 0x81-0xA0
949 		     */
950 		    return 20;
951 #endif
952 
953 		else if (c1 >= 0xCA && c1 <= 0xFD)
954 		    /* Hanja */
955 		    return 21;
956 		else switch (c1)
957 		{
958 		    case 0xA1:
959 		    case 0xA2:
960 			/* Symbols */
961 			return 22;
962 		    case 0xA3:
963 			/* Alpha-numeric */
964 			return 23;
965 		    case 0xA4:
966 			/* Hangul Letter(Alphabet) */
967 			return 24;
968 		    case 0xA5:
969 			/* Roman Numeral/Greek Letter */
970 			return 25;
971 		    case 0xA6:
972 			/* Box Drawings */
973 			return 26;
974 		    case 0xA7:
975 			/* Unit Symbols */
976 			return 27;
977 		    case 0xA8:
978 		    case 0xA9:
979 			if (c2 <= 0xAF)
980 			    return 25;  /* Roman Letter */
981 			else if (c2 >= 0xF6)
982 			    return 22;  /* Symbols */
983 			else
984 			    /* Circled/Parenthesized Letter */
985 			    return 28;
986 		    case 0xAA:
987 		    case 0xAB:
988 			/* Hirigana/Katakana */
989 			return 29;
990 		    case 0xAC:
991 			/* Cyrillic Letter */
992 			return 30;
993 		}
994 	    }
995 	default:
996 	    break;
997     }
998     return 3;
999 }
1000 
1001 /*
1002  * mb_char2len() function pointer.
1003  * Return length in bytes of character "c".
1004  * Returns 1 for a single-byte character.
1005  */
1006 /* ARGSUSED */
1007     int
1008 latin_char2len(c)
1009     int		c;
1010 {
1011     return 1;
1012 }
1013 
1014     static int
1015 dbcs_char2len(c)
1016     int		c;
1017 {
1018     if (c >= 0x100)
1019 	return 2;
1020     return 1;
1021 }
1022 
1023 /*
1024  * mb_char2bytes() function pointer.
1025  * Convert a character to its bytes.
1026  * Returns the length in bytes.
1027  */
1028     int
1029 latin_char2bytes(c, buf)
1030     int		c;
1031     char_u	*buf;
1032 {
1033     buf[0] = c;
1034     return 1;
1035 }
1036 
1037     static int
1038 dbcs_char2bytes(c, buf)
1039     int		c;
1040     char_u	*buf;
1041 {
1042     if (c >= 0x100)
1043     {
1044 	buf[0] = (unsigned)c >> 8;
1045 	buf[1] = c;
1046 	/* Never use a NUL byte, it causes lots of trouble.  It's an invalid
1047 	 * character anyway. */
1048 	if (buf[1] == NUL)
1049 	    buf[1] = '\n';
1050 	return 2;
1051     }
1052     buf[0] = c;
1053     return 1;
1054 }
1055 
1056 /*
1057  * mb_ptr2len() function pointer.
1058  * Get byte length of character at "*p" but stop at a NUL.
1059  * For UTF-8 this includes following composing characters.
1060  * Returns 0 when *p is NUL.
1061  *
1062  */
1063     int
1064 latin_ptr2len(p)
1065     char_u	*p;
1066 {
1067     return MB_BYTE2LEN(*p);
1068 }
1069 
1070     static int
1071 dbcs_ptr2len(p)
1072     char_u	*p;
1073 {
1074     int		len;
1075 
1076     /* Check if second byte is not missing. */
1077     len = MB_BYTE2LEN(*p);
1078     if (len == 2 && p[1] == NUL)
1079 	len = 1;
1080     return len;
1081 }
1082 
1083 struct interval
1084 {
1085     unsigned short first;
1086     unsigned short last;
1087 };
1088 static int intable __ARGS((struct interval *table, size_t size, int c));
1089 
1090 /*
1091  * Return TRUE if "c" is in "table[size / sizeof(struct interval)]".
1092  */
1093     static int
1094 intable(table, size, c)
1095     struct interval	*table;
1096     size_t		size;
1097     int			c;
1098 {
1099     int mid, bot, top;
1100 
1101     /* first quick check for Latin1 etc. characters */
1102     if (c < table[0].first)
1103 	return FALSE;
1104 
1105     /* binary search in table */
1106     bot = 0;
1107     top = size / sizeof(struct interval) - 1;
1108     while (top >= bot)
1109     {
1110 	mid = (bot + top) / 2;
1111 	if (table[mid].last < c)
1112 	    bot = mid + 1;
1113 	else if (table[mid].first > c)
1114 	    top = mid - 1;
1115 	else
1116 	    return TRUE;
1117     }
1118     return FALSE;
1119 }
1120 
1121 /*
1122  * For UTF-8 character "c" return 2 for a double-width character, 1 for others.
1123  * Returns 4 or 6 for an unprintable character.
1124  * Is only correct for characters >= 0x80.
1125  * When p_ambw is "double", return 2 for a character with East Asian Width
1126  * class 'A'(mbiguous).
1127  */
1128     int
1129 utf_char2cells(c)
1130     int		c;
1131 {
1132     /* sorted list of non-overlapping intervals of East Asian Ambiguous
1133      * characters, generated with:
1134      * "uniset +WIDTH-A -cat=Me -cat=Mn -cat=Cf c" */
1135     static struct interval ambiguous[] = {
1136 	{0x00A1, 0x00A1}, {0x00A4, 0x00A4}, {0x00A7, 0x00A8},
1137 	{0x00AA, 0x00AA}, {0x00AE, 0x00AE}, {0x00B0, 0x00B4},
1138 	{0x00B6, 0x00BA}, {0x00BC, 0x00BF}, {0x00C6, 0x00C6},
1139 	{0x00D0, 0x00D0}, {0x00D7, 0x00D8}, {0x00DE, 0x00E1},
1140 	{0x00E6, 0x00E6}, {0x00E8, 0x00EA}, {0x00EC, 0x00ED},
1141 	{0x00F0, 0x00F0}, {0x00F2, 0x00F3}, {0x00F7, 0x00FA},
1142 	{0x00FC, 0x00FC}, {0x00FE, 0x00FE}, {0x0101, 0x0101},
1143 	{0x0111, 0x0111}, {0x0113, 0x0113}, {0x011B, 0x011B},
1144 	{0x0126, 0x0127}, {0x012B, 0x012B}, {0x0131, 0x0133},
1145 	{0x0138, 0x0138}, {0x013F, 0x0142}, {0x0144, 0x0144},
1146 	{0x0148, 0x014B}, {0x014D, 0x014D}, {0x0152, 0x0153},
1147 	{0x0166, 0x0167}, {0x016B, 0x016B}, {0x01CE, 0x01CE},
1148 	{0x01D0, 0x01D0}, {0x01D2, 0x01D2}, {0x01D4, 0x01D4},
1149 	{0x01D6, 0x01D6}, {0x01D8, 0x01D8}, {0x01DA, 0x01DA},
1150 	{0x01DC, 0x01DC}, {0x0251, 0x0251}, {0x0261, 0x0261},
1151 	{0x02C4, 0x02C4}, {0x02C7, 0x02C7}, {0x02C9, 0x02CB},
1152 	{0x02CD, 0x02CD}, {0x02D0, 0x02D0}, {0x02D8, 0x02DB},
1153 	{0x02DD, 0x02DD}, {0x02DF, 0x02DF}, {0x0391, 0x03A1},
1154 	{0x03A3, 0x03A9}, {0x03B1, 0x03C1}, {0x03C3, 0x03C9},
1155 	{0x0401, 0x0401}, {0x0410, 0x044F}, {0x0451, 0x0451},
1156 	{0x2010, 0x2010}, {0x2013, 0x2016}, {0x2018, 0x2019},
1157 	{0x201C, 0x201D}, {0x2020, 0x2022}, {0x2024, 0x2027},
1158 	{0x2030, 0x2030}, {0x2032, 0x2033}, {0x2035, 0x2035},
1159 	{0x203B, 0x203B}, {0x203E, 0x203E}, {0x2074, 0x2074},
1160 	{0x207F, 0x207F}, {0x2081, 0x2084}, {0x20AC, 0x20AC},
1161 	{0x2103, 0x2103}, {0x2105, 0x2105}, {0x2109, 0x2109},
1162 	{0x2113, 0x2113}, {0x2116, 0x2116}, {0x2121, 0x2122},
1163 	{0x2126, 0x2126}, {0x212B, 0x212B}, {0x2153, 0x2154},
1164 	{0x215B, 0x215E}, {0x2160, 0x216B}, {0x2170, 0x2179},
1165 	{0x2190, 0x2199}, {0x21B8, 0x21B9}, {0x21D2, 0x21D2},
1166 	{0x21D4, 0x21D4}, {0x21E7, 0x21E7}, {0x2200, 0x2200},
1167 	{0x2202, 0x2203}, {0x2207, 0x2208}, {0x220B, 0x220B},
1168 	{0x220F, 0x220F}, {0x2211, 0x2211}, {0x2215, 0x2215},
1169 	{0x221A, 0x221A}, {0x221D, 0x2220}, {0x2223, 0x2223},
1170 	{0x2225, 0x2225}, {0x2227, 0x222C}, {0x222E, 0x222E},
1171 	{0x2234, 0x2237}, {0x223C, 0x223D}, {0x2248, 0x2248},
1172 	{0x224C, 0x224C}, {0x2252, 0x2252}, {0x2260, 0x2261},
1173 	{0x2264, 0x2267}, {0x226A, 0x226B}, {0x226E, 0x226F},
1174 	{0x2282, 0x2283}, {0x2286, 0x2287}, {0x2295, 0x2295},
1175 	{0x2299, 0x2299}, {0x22A5, 0x22A5}, {0x22BF, 0x22BF},
1176 	{0x2312, 0x2312}, {0x2460, 0x24E9}, {0x24EB, 0x254B},
1177 	{0x2550, 0x2573}, {0x2580, 0x258F}, {0x2592, 0x2595},
1178 	{0x25A0, 0x25A1}, {0x25A3, 0x25A9}, {0x25B2, 0x25B3},
1179 	{0x25B6, 0x25B7}, {0x25BC, 0x25BD}, {0x25C0, 0x25C1},
1180 	{0x25C6, 0x25C8}, {0x25CB, 0x25CB}, {0x25CE, 0x25D1},
1181 	{0x25E2, 0x25E5}, {0x25EF, 0x25EF}, {0x2605, 0x2606},
1182 	{0x2609, 0x2609}, {0x260E, 0x260F}, {0x2614, 0x2615},
1183 	{0x261C, 0x261C}, {0x261E, 0x261E}, {0x2640, 0x2640},
1184 	{0x2642, 0x2642}, {0x2660, 0x2661}, {0x2663, 0x2665},
1185 	{0x2667, 0x266A}, {0x266C, 0x266D}, {0x266F, 0x266F},
1186 	{0x273D, 0x273D}, {0x2776, 0x277F}, {0xE000, 0xF8FF},
1187 	{0xFFFD, 0xFFFD}, /* {0xF0000, 0xFFFFD}, {0x100000, 0x10FFFD} */
1188     };
1189 
1190     if (c >= 0x100)
1191     {
1192 #ifdef USE_WCHAR_FUNCTIONS
1193 	/*
1194 	 * Assume the library function wcwidth() works better than our own
1195 	 * stuff.  It should return 1 for ambiguous width chars!
1196 	 */
1197 	int	n = wcwidth(c);
1198 
1199 	if (n < 0)
1200 	    return 6;		/* unprintable, displays <xxxx> */
1201 	if (n > 1)
1202 	    return n;
1203 #else
1204 	if (!utf_printable(c))
1205 	    return 6;		/* unprintable, displays <xxxx> */
1206 	if (c >= 0x1100
1207 	    && (c <= 0x115f			/* Hangul Jamo */
1208 		|| c == 0x2329
1209 		|| c == 0x232a
1210 		|| (c >= 0x2e80 && c <= 0xa4cf
1211 		    && c != 0x303f)		/* CJK ... Yi */
1212 		|| (c >= 0xac00 && c <= 0xd7a3)	/* Hangul Syllables */
1213 		|| (c >= 0xf900 && c <= 0xfaff)	/* CJK Compatibility
1214 						   Ideographs */
1215 		|| (c >= 0xfe30 && c <= 0xfe6f)	/* CJK Compatibility Forms */
1216 		|| (c >= 0xff00 && c <= 0xff60)	/* Fullwidth Forms */
1217 		|| (c >= 0xffe0 && c <= 0xffe6)
1218 		|| (c >= 0x20000 && c <= 0x2fffd)
1219 		|| (c >= 0x30000 && c <= 0x3fffd)))
1220 	    return 2;
1221 #endif
1222     }
1223 
1224     /* Characters below 0x100 are influenced by 'isprint' option */
1225     else if (c >= 0x80 && !vim_isprintc(c))
1226 	return 4;		/* unprintable, displays <xx> */
1227 
1228     if (c >= 0x80 && *p_ambw == 'd' && intable(ambiguous, sizeof(ambiguous), c))
1229 	return 2;
1230 
1231     return 1;
1232 }
1233 
1234 /*
1235  * mb_ptr2cells() function pointer.
1236  * Return the number of display cells character at "*p" occupies.
1237  * This doesn't take care of unprintable characters, use ptr2cells() for that.
1238  */
1239 /*ARGSUSED*/
1240     int
1241 latin_ptr2cells(p)
1242     char_u	*p;
1243 {
1244     return 1;
1245 }
1246 
1247     int
1248 utf_ptr2cells(p)
1249     char_u	*p;
1250 {
1251     int		c;
1252 
1253     /* Need to convert to a wide character. */
1254     if (*p >= 0x80)
1255     {
1256 	c = utf_ptr2char(p);
1257 	/* An illegal byte is displayed as <xx>. */
1258 	if (utf_ptr2len(p) == 1 || c == NUL)
1259 	    return 4;
1260 	/* If the char is ASCII it must be an overlong sequence. */
1261 	if (c < 0x80)
1262 	    return char2cells(c);
1263 	return utf_char2cells(c);
1264     }
1265     return 1;
1266 }
1267 
1268     int
1269 dbcs_ptr2cells(p)
1270     char_u	*p;
1271 {
1272     /* Number of cells is equal to number of bytes, except for euc-jp when
1273      * the first byte is 0x8e. */
1274     if (enc_dbcs == DBCS_JPNU && *p == 0x8e)
1275 	return 1;
1276     return MB_BYTE2LEN(*p);
1277 }
1278 
1279 /*
1280  * mb_char2cells() function pointer.
1281  * Return the number of display cells character "c" occupies.
1282  * Only takes care of multi-byte chars, not "^C" and such.
1283  */
1284 /*ARGSUSED*/
1285     int
1286 latin_char2cells(c)
1287     int		c;
1288 {
1289     return 1;
1290 }
1291 
1292     static int
1293 dbcs_char2cells(c)
1294     int		c;
1295 {
1296     /* Number of cells is equal to number of bytes, except for euc-jp when
1297      * the first byte is 0x8e. */
1298     if (enc_dbcs == DBCS_JPNU && ((unsigned)c >> 8) == 0x8e)
1299 	return 1;
1300     /* use the first byte */
1301     return MB_BYTE2LEN((unsigned)c >> 8);
1302 }
1303 
1304 /*
1305  * mb_off2cells() function pointer.
1306  * Return number of display cells for char at ScreenLines[off].
1307  * Caller must make sure "off" and "off + 1" are valid!
1308  */
1309 /*ARGSUSED*/
1310     int
1311 latin_off2cells(off)
1312     unsigned	off;
1313 {
1314     return 1;
1315 }
1316 
1317     int
1318 dbcs_off2cells(off)
1319     unsigned	off;
1320 {
1321     /* Number of cells is equal to number of bytes, except for euc-jp when
1322      * the first byte is 0x8e. */
1323     if (enc_dbcs == DBCS_JPNU && ScreenLines[off] == 0x8e)
1324 	return 1;
1325     return MB_BYTE2LEN(ScreenLines[off]);
1326 }
1327 
1328     int
1329 utf_off2cells(off)
1330     unsigned	off;
1331 {
1332     return ScreenLines[off + 1] == 0 ? 2 : 1;
1333 }
1334 
1335 /*
1336  * mb_ptr2char() function pointer.
1337  * Convert a byte sequence into a character.
1338  */
1339     int
1340 latin_ptr2char(p)
1341     char_u	*p;
1342 {
1343     return *p;
1344 }
1345 
1346     static int
1347 dbcs_ptr2char(p)
1348     char_u	*p;
1349 {
1350     if (MB_BYTE2LEN(*p) > 1 && p[1] != NUL)
1351 	return (p[0] << 8) + p[1];
1352     return *p;
1353 }
1354 
1355 /*
1356  * Convert a UTF-8 byte sequence to a wide character.
1357  * If the sequence is illegal or truncated by a NUL the first byte is
1358  * returned.
1359  * Does not include composing characters, of course.
1360  */
1361     int
1362 utf_ptr2char(p)
1363     char_u	*p;
1364 {
1365     int		len;
1366 
1367     if (p[0] < 0x80)	/* be quick for ASCII */
1368 	return p[0];
1369 
1370     len = utf8len_tab[p[0]];
1371     if ((p[1] & 0xc0) == 0x80)
1372     {
1373 	if (len == 2)
1374 	    return ((p[0] & 0x1f) << 6) + (p[1] & 0x3f);
1375 	if ((p[2] & 0xc0) == 0x80)
1376 	{
1377 	    if (len == 3)
1378 		return ((p[0] & 0x0f) << 12) + ((p[1] & 0x3f) << 6)
1379 		    + (p[2] & 0x3f);
1380 	    if ((p[3] & 0xc0) == 0x80)
1381 	    {
1382 		if (len == 4)
1383 		    return ((p[0] & 0x07) << 18) + ((p[1] & 0x3f) << 12)
1384 			+ ((p[2] & 0x3f) << 6) + (p[3] & 0x3f);
1385 		if ((p[4] & 0xc0) == 0x80)
1386 		{
1387 		    if (len == 5)
1388 			return ((p[0] & 0x03) << 24) + ((p[1] & 0x3f) << 18)
1389 			    + ((p[2] & 0x3f) << 12) + ((p[3] & 0x3f) << 6)
1390 			    + (p[4] & 0x3f);
1391 		    if ((p[5] & 0xc0) == 0x80 && len == 6)
1392 			return ((p[0] & 0x01) << 30) + ((p[1] & 0x3f) << 24)
1393 			    + ((p[2] & 0x3f) << 18) + ((p[3] & 0x3f) << 12)
1394 			    + ((p[4] & 0x3f) << 6) + (p[5] & 0x3f);
1395 		}
1396 	    }
1397 	}
1398     }
1399     /* Illegal value, just return the first byte */
1400     return p[0];
1401 }
1402 
1403 /*
1404  * Get character at **pp and advance *pp to the next character.
1405  * Note: composing characters are skipped!
1406  */
1407     int
1408 mb_ptr2char_adv(pp)
1409     char_u	**pp;
1410 {
1411     int		c;
1412 
1413     c = (*mb_ptr2char)(*pp);
1414     *pp += (*mb_ptr2len)(*pp);
1415     return c;
1416 }
1417 
1418 /*
1419  * Get character at **pp and advance *pp to the next character.
1420  * Note: composing characters are returned as separate characters.
1421  */
1422     int
1423 mb_cptr2char_adv(pp)
1424     char_u	**pp;
1425 {
1426     int		c;
1427 
1428     c = (*mb_ptr2char)(*pp);
1429     if (enc_utf8)
1430 	*pp += utf_ptr2len(*pp);
1431     else
1432 	*pp += (*mb_ptr2len)(*pp);
1433     return c;
1434 }
1435 
1436 #if defined(FEAT_ARABIC) || defined(PROTO)
1437 /*
1438  * Check whether we are dealing with Arabic combining characters.
1439  * Note: these are NOT really composing characters!
1440  */
1441     int
1442 arabic_combine(one, two)
1443     int		one;	    /* first character */
1444     int		two;	    /* character just after "one" */
1445 {
1446     if (one == a_LAM)
1447 	return arabic_maycombine(two);
1448     return FALSE;
1449 }
1450 
1451 /*
1452  * Check whether we are dealing with a character that could be regarded as an
1453  * Arabic combining character, need to check the character before this.
1454  */
1455     int
1456 arabic_maycombine(two)
1457     int		two;
1458 {
1459     if (p_arshape && !p_tbidi)
1460 	return (two == a_ALEF_MADDA
1461 		    || two == a_ALEF_HAMZA_ABOVE
1462 		    || two == a_ALEF_HAMZA_BELOW
1463 		    || two == a_ALEF);
1464     return FALSE;
1465 }
1466 
1467 /*
1468  * Check if the character pointed to by "p2" is a composing character when it
1469  * comes after "p1".  For Arabic sometimes "ab" is replaced with "c", which
1470  * behaves like a composing character.
1471  */
1472     int
1473 utf_composinglike(p1, p2)
1474     char_u	*p1;
1475     char_u	*p2;
1476 {
1477     int		c2;
1478 
1479     c2 = utf_ptr2char(p2);
1480     if (utf_iscomposing(c2))
1481 	return TRUE;
1482     if (!arabic_maycombine(c2))
1483 	return FALSE;
1484     return arabic_combine(utf_ptr2char(p1), c2);
1485 }
1486 #endif
1487 
1488 /*
1489  * Convert a UTF-8 byte string to a wide chararacter.  Also get up to MAX_MCO
1490  * composing characters.
1491  */
1492     int
1493 utfc_ptr2char(p, pcc)
1494     char_u	*p;
1495     int		*pcc;	/* return: composing chars, last one is 0 */
1496 {
1497     int		len;
1498     int		c;
1499     int		cc;
1500     int		i = 0;
1501 
1502     c = utf_ptr2char(p);
1503     len = utf_ptr2len(p);
1504 
1505     /* Only accept a composing char when the first char isn't illegal. */
1506     if ((len > 1 || *p < 0x80)
1507 	    && p[len] >= 0x80
1508 	    && UTF_COMPOSINGLIKE(p, p + len))
1509     {
1510 	cc = utf_ptr2char(p + len);
1511 	for (;;)
1512 	{
1513 	    pcc[i++] = cc;
1514 	    if (i == MAX_MCO)
1515 		break;
1516 	    len += utf_ptr2len(p + len);
1517 	    if (p[len] < 0x80 || !utf_iscomposing(cc = utf_ptr2char(p + len)))
1518 		break;
1519 	}
1520     }
1521 
1522     if (i < MAX_MCO)	/* last composing char must be 0 */
1523 	pcc[i] = 0;
1524 
1525     return c;
1526 }
1527 
1528 /*
1529  * Convert a UTF-8 byte string to a wide chararacter.  Also get up to MAX_MCO
1530  * composing characters.  Use no more than p[maxlen].
1531  */
1532     int
1533 utfc_ptr2char_len(p, pcc, maxlen)
1534     char_u	*p;
1535     int		*pcc;	/* return: composing chars, last one is 0 */
1536     int		maxlen;
1537 {
1538     int		len;
1539     int		c;
1540     int		cc;
1541     int		i = 0;
1542 
1543     c = utf_ptr2char(p);
1544     len = utf_ptr2len_len(p, maxlen);
1545     /* Only accept a composing char when the first char isn't illegal. */
1546     if ((len > 1 || *p < 0x80)
1547 	    && len < maxlen
1548 	    && p[len] >= 0x80
1549 	    && UTF_COMPOSINGLIKE(p, p + len))
1550     {
1551 	cc = utf_ptr2char(p + len);
1552 	for (;;)
1553 	{
1554 	    pcc[i++] = cc;
1555 	    if (i == MAX_MCO)
1556 		break;
1557 	    len += utf_ptr2len_len(p + len, maxlen - len);
1558 	    if (len >= maxlen
1559 		    || p[len] < 0x80
1560 		    || !utf_iscomposing(cc = utf_ptr2char(p + len)))
1561 		break;
1562 	}
1563     }
1564 
1565     if (i < MAX_MCO)	/* last composing char must be 0 */
1566 	pcc[i] = 0;
1567 
1568     return c;
1569 }
1570 
1571 /*
1572  * Convert the character at screen position "off" to a sequence of bytes.
1573  * Includes the composing characters.
1574  * "buf" must at least have the length MB_MAXBYTES.
1575  * Returns the produced number of bytes.
1576  */
1577     int
1578 utfc_char2bytes(off, buf)
1579     int		off;
1580     char_u	*buf;
1581 {
1582     int		len;
1583     int		i;
1584 
1585     len = utf_char2bytes(ScreenLinesUC[off], buf);
1586     for (i = 0; i < Screen_mco; ++i)
1587     {
1588 	if (ScreenLinesC[i][off] == 0)
1589 	    break;
1590 	len += utf_char2bytes(ScreenLinesC[i][off], buf + len);
1591     }
1592     return len;
1593 }
1594 
1595 /*
1596  * Get the length of a UTF-8 byte sequence, not including any following
1597  * composing characters.
1598  * Returns 0 for "".
1599  * Returns 1 for an illegal byte sequence.
1600  */
1601     int
1602 utf_ptr2len(p)
1603     char_u	*p;
1604 {
1605     int		len;
1606     int		i;
1607 
1608     if (*p == NUL)
1609 	return 0;
1610     len = utf8len_tab[*p];
1611     for (i = 1; i < len; ++i)
1612 	if ((p[i] & 0xc0) != 0x80)
1613 	    return 1;
1614     return len;
1615 }
1616 
1617 /*
1618  * Return length of UTF-8 character, obtained from the first byte.
1619  * "b" must be between 0 and 255!
1620  */
1621     int
1622 utf_byte2len(b)
1623     int		b;
1624 {
1625     return utf8len_tab[b];
1626 }
1627 
1628 /*
1629  * Get the length of UTF-8 byte sequence "p[size]".  Does not include any
1630  * following composing characters.
1631  * Returns 1 for "".
1632  * Returns 1 for an illegal byte sequence.
1633  * Returns number > "size" for an incomplete byte sequence.
1634  */
1635     int
1636 utf_ptr2len_len(p, size)
1637     char_u	*p;
1638     int		size;
1639 {
1640     int		len;
1641     int		i;
1642 
1643     if (*p == NUL)
1644 	return 1;
1645     len = utf8len_tab[*p];
1646     if (len > size)
1647 	return len;	/* incomplete byte sequence. */
1648     for (i = 1; i < len; ++i)
1649 	if ((p[i] & 0xc0) != 0x80)
1650 	    return 1;
1651     return len;
1652 }
1653 
1654 /*
1655  * Return the number of bytes the UTF-8 encoding of the character at "p" takes.
1656  * This includes following composing characters.
1657  */
1658     int
1659 utfc_ptr2len(p)
1660     char_u	*p;
1661 {
1662     int		len;
1663     int		b0 = *p;
1664 #ifdef FEAT_ARABIC
1665     int		prevlen;
1666 #endif
1667 
1668     if (b0 == NUL)
1669 	return 0;
1670     if (b0 < 0x80 && p[1] < 0x80)	/* be quick for ASCII */
1671 	return 1;
1672 
1673     /* Skip over first UTF-8 char, stopping at a NUL byte. */
1674     len = utf_ptr2len(p);
1675 
1676     /* Check for illegal byte. */
1677     if (len == 1 && b0 >= 0x80)
1678 	return 1;
1679 
1680     /*
1681      * Check for composing characters.  We can handle only the first two, but
1682      * skip all of them (otherwise the cursor would get stuck).
1683      */
1684 #ifdef FEAT_ARABIC
1685     prevlen = 0;
1686 #endif
1687     for (;;)
1688     {
1689 	if (p[len] < 0x80 || !UTF_COMPOSINGLIKE(p + prevlen, p + len))
1690 	    return len;
1691 
1692 	/* Skip over composing char */
1693 #ifdef FEAT_ARABIC
1694 	prevlen = len;
1695 #endif
1696 	len += utf_ptr2len(p + len);
1697     }
1698 }
1699 
1700 /*
1701  * Return the number of bytes the UTF-8 encoding of the character at "p[size]"
1702  * takes.  This includes following composing characters.
1703  * Returns 1 for an illegal char or an incomplete byte sequence.
1704  */
1705     int
1706 utfc_ptr2len_len(p, size)
1707     char_u	*p;
1708     int		size;
1709 {
1710     int		len;
1711 #ifdef FEAT_ARABIC
1712     int		prevlen;
1713 #endif
1714 
1715     if (*p == NUL)
1716 	return 0;
1717     if (p[0] < 0x80 && (size == 1 || p[1] < 0x80)) /* be quick for ASCII */
1718 	return 1;
1719 
1720     /* Skip over first UTF-8 char, stopping at a NUL byte. */
1721     len = utf_ptr2len_len(p, size);
1722 
1723     /* Check for illegal byte and incomplete byte sequence. */
1724     if ((len == 1 && p[0] >= 0x80) || len > size)
1725 	return 1;
1726 
1727     /*
1728      * Check for composing characters.  We can handle only the first two, but
1729      * skip all of them (otherwise the cursor would get stuck).
1730      */
1731 #ifdef FEAT_ARABIC
1732     prevlen = 0;
1733 #endif
1734     while (len < size)
1735     {
1736 	if (p[len] < 0x80 || !UTF_COMPOSINGLIKE(p + prevlen, p + len))
1737 	    break;
1738 
1739 	/* Skip over composing char */
1740 #ifdef FEAT_ARABIC
1741 	prevlen = len;
1742 #endif
1743 	len += utf_ptr2len_len(p + len, size - len);
1744     }
1745     return len;
1746 }
1747 
1748 /*
1749  * Return the number of bytes the UTF-8 encoding of character "c" takes.
1750  * This does not include composing characters.
1751  */
1752     int
1753 utf_char2len(c)
1754     int		c;
1755 {
1756     if (c < 0x80)
1757 	return 1;
1758     if (c < 0x800)
1759 	return 2;
1760     if (c < 0x10000)
1761 	return 3;
1762     if (c < 0x200000)
1763 	return 4;
1764     if (c < 0x4000000)
1765 	return 5;
1766     return 6;
1767 }
1768 
1769 /*
1770  * Convert Unicode character "c" to UTF-8 string in "buf[]".
1771  * Returns the number of bytes.
1772  * This does not include composing characters.
1773  */
1774     int
1775 utf_char2bytes(c, buf)
1776     int		c;
1777     char_u	*buf;
1778 {
1779     if (c < 0x80)		/* 7 bits */
1780     {
1781 	buf[0] = c;
1782 	return 1;
1783     }
1784     if (c < 0x800)		/* 11 bits */
1785     {
1786 	buf[0] = 0xc0 + ((unsigned)c >> 6);
1787 	buf[1] = 0x80 + (c & 0x3f);
1788 	return 2;
1789     }
1790     if (c < 0x10000)		/* 16 bits */
1791     {
1792 	buf[0] = 0xe0 + ((unsigned)c >> 12);
1793 	buf[1] = 0x80 + (((unsigned)c >> 6) & 0x3f);
1794 	buf[2] = 0x80 + (c & 0x3f);
1795 	return 3;
1796     }
1797     if (c < 0x200000)		/* 21 bits */
1798     {
1799 	buf[0] = 0xf0 + ((unsigned)c >> 18);
1800 	buf[1] = 0x80 + (((unsigned)c >> 12) & 0x3f);
1801 	buf[2] = 0x80 + (((unsigned)c >> 6) & 0x3f);
1802 	buf[3] = 0x80 + (c & 0x3f);
1803 	return 4;
1804     }
1805     if (c < 0x4000000)		/* 26 bits */
1806     {
1807 	buf[0] = 0xf8 + ((unsigned)c >> 24);
1808 	buf[1] = 0x80 + (((unsigned)c >> 18) & 0x3f);
1809 	buf[2] = 0x80 + (((unsigned)c >> 12) & 0x3f);
1810 	buf[3] = 0x80 + (((unsigned)c >> 6) & 0x3f);
1811 	buf[4] = 0x80 + (c & 0x3f);
1812 	return 5;
1813     }
1814 				/* 31 bits */
1815     buf[0] = 0xfc + ((unsigned)c >> 30);
1816     buf[1] = 0x80 + (((unsigned)c >> 24) & 0x3f);
1817     buf[2] = 0x80 + (((unsigned)c >> 18) & 0x3f);
1818     buf[3] = 0x80 + (((unsigned)c >> 12) & 0x3f);
1819     buf[4] = 0x80 + (((unsigned)c >> 6) & 0x3f);
1820     buf[5] = 0x80 + (c & 0x3f);
1821     return 6;
1822 }
1823 
1824 /*
1825  * Return TRUE if "c" is a composing UTF-8 character.  This means it will be
1826  * drawn on top of the preceding character.
1827  * Based on code from Markus Kuhn.
1828  */
1829     int
1830 utf_iscomposing(c)
1831     int		c;
1832 {
1833     /* sorted list of non-overlapping intervals */
1834     static struct interval combining[] =
1835     {
1836 	{0x0300, 0x034f}, {0x0360, 0x036f}, {0x0483, 0x0486}, {0x0488, 0x0489},
1837 	{0x0591, 0x05a1}, {0x05a3, 0x05b9}, {0x05bb, 0x05bd}, {0x05bf, 0x05bf},
1838 	{0x05c1, 0x05c2}, {0x05c4, 0x05c4}, {0x0610, 0x0615}, {0x064b, 0x0658},
1839 	{0x0670, 0x0670}, {0x06d6, 0x06dc}, {0x06de, 0x06e4}, {0x06e7, 0x06e8},
1840 	{0x06ea, 0x06ed}, {0x0711, 0x0711}, {0x0730, 0x074a}, {0x07a6, 0x07b0},
1841 	{0x0901, 0x0903}, {0x093c, 0x093c}, {0x093e, 0x094d}, {0x0951, 0x0954},
1842 	{0x0962, 0x0963}, {0x0981, 0x0983}, {0x09bc, 0x09bc}, {0x09be, 0x09c4},
1843 	{0x09c7, 0x09c8}, {0x09cb, 0x09cd}, {0x09d7, 0x09d7}, {0x09e2, 0x09e3},
1844 	{0x0a01, 0x0a03}, {0x0a3c, 0x0a3c}, {0x0a3e, 0x0a42}, {0x0a47, 0x0a48},
1845 	{0x0a4b, 0x0a4d}, {0x0a70, 0x0a71}, {0x0a81, 0x0a83}, {0x0abc, 0x0abc},
1846 	{0x0abe, 0x0ac5}, {0x0ac7, 0x0ac9}, {0x0acb, 0x0acd}, {0x0ae2, 0x0ae3},
1847 	{0x0b01, 0x0b03}, {0x0b3c, 0x0b3c}, {0x0b3e, 0x0b43}, {0x0b47, 0x0b48},
1848 	{0x0b4b, 0x0b4d}, {0x0b56, 0x0b57}, {0x0b82, 0x0b82}, {0x0bbe, 0x0bc2},
1849 	{0x0bc6, 0x0bc8}, {0x0bca, 0x0bcd}, {0x0bd7, 0x0bd7}, {0x0c01, 0x0c03},
1850 	{0x0c3e, 0x0c44}, {0x0c46, 0x0c48}, {0x0c4a, 0x0c4d}, {0x0c55, 0x0c56},
1851 	{0x0c82, 0x0c83}, {0x0cbc, 0x0cbc}, {0x0cbe, 0x0cc4}, {0x0cc6, 0x0cc8},
1852 	{0x0cca, 0x0ccd}, {0x0cd5, 0x0cd6}, {0x0d02, 0x0d03}, {0x0d3e, 0x0d43},
1853 	{0x0d46, 0x0d48}, {0x0d4a, 0x0d4d}, {0x0d57, 0x0d57}, {0x0d82, 0x0d83},
1854 	{0x0dca, 0x0dca}, {0x0dcf, 0x0dd4}, {0x0dd6, 0x0dd6}, {0x0dd8, 0x0ddf},
1855 	{0x0df2, 0x0df3}, {0x0e31, 0x0e31}, {0x0e34, 0x0e3a}, {0x0e47, 0x0e4e},
1856 	{0x0eb1, 0x0eb1}, {0x0eb4, 0x0eb9}, {0x0ebb, 0x0ebc}, {0x0ec8, 0x0ecd},
1857 	{0x0f18, 0x0f19}, {0x0f35, 0x0f35}, {0x0f37, 0x0f37}, {0x0f39, 0x0f39},
1858 	{0x0f3e, 0x0f3f}, {0x0f71, 0x0f84}, {0x0f86, 0x0f87}, {0x0f90, 0x0f97},
1859 	{0x0f99, 0x0fbc}, {0x0fc6, 0x0fc6}, {0x102c, 0x1032}, {0x1036, 0x1039},
1860 	{0x1056, 0x1059}, {0x1712, 0x1714}, {0x1732, 0x1734}, {0x1752, 0x1753},
1861 	{0x1772, 0x1773}, {0x17b6, 0x17d3}, {0x17dd, 0x17dd}, {0x180b, 0x180d},
1862 	{0x18a9, 0x18a9}, {0x1920, 0x192b}, {0x1930, 0x193b}, {0x20d0, 0x20ea},
1863 	{0x302a, 0x302f}, {0x3099, 0x309a}, {0xfb1e, 0xfb1e}, {0xfe00, 0xfe0f},
1864 	{0xfe20, 0xfe23},
1865     };
1866 
1867     return intable(combining, sizeof(combining), c);
1868 }
1869 
1870 /*
1871  * Return TRUE for characters that can be displayed in a normal way.
1872  * Only for characters of 0x100 and above!
1873  */
1874     int
1875 utf_printable(c)
1876     int		c;
1877 {
1878 #ifdef USE_WCHAR_FUNCTIONS
1879     /*
1880      * Assume the iswprint() library function works better than our own stuff.
1881      */
1882     return iswprint(c);
1883 #else
1884     /* Sorted list of non-overlapping intervals.
1885      * 0xd800-0xdfff is reserved for UTF-16, actually illegal. */
1886     static struct interval nonprint[] =
1887     {
1888 	{0x070f, 0x070f}, {0x180b, 0x180e}, {0x200b, 0x200f}, {0x202a, 0x202e},
1889 	{0x206a, 0x206f}, {0xd800, 0xdfff}, {0xfeff, 0xfeff}, {0xfff9, 0xfffb},
1890 	{0xfffe, 0xffff}
1891     };
1892 
1893     return !intable(nonprint, sizeof(nonprint), c);
1894 #endif
1895 }
1896 
1897 /*
1898  * Get class of a Unicode character.
1899  * 0: white space
1900  * 1: punctuation
1901  * 2 or bigger: some class of word character.
1902  */
1903     int
1904 utf_class(c)
1905     int		c;
1906 {
1907     /* sorted list of non-overlapping intervals */
1908     static struct clinterval
1909     {
1910 	unsigned short first;
1911 	unsigned short last;
1912 	unsigned short class;
1913     } classes[] =
1914     {
1915 	{0x037e, 0x037e, 1},		/* Greek question mark */
1916 	{0x0387, 0x0387, 1},		/* Greek ano teleia */
1917 	{0x055a, 0x055f, 1},		/* Armenian punctuation */
1918 	{0x0589, 0x0589, 1},		/* Armenian full stop */
1919 	{0x05be, 0x05be, 1},
1920 	{0x05c0, 0x05c0, 1},
1921 	{0x05c3, 0x05c3, 1},
1922 	{0x05f3, 0x05f4, 1},
1923 	{0x060c, 0x060c, 1},
1924 	{0x061b, 0x061b, 1},
1925 	{0x061f, 0x061f, 1},
1926 	{0x066a, 0x066d, 1},
1927 	{0x06d4, 0x06d4, 1},
1928 	{0x0700, 0x070d, 1},		/* Syriac punctuation */
1929 	{0x0964, 0x0965, 1},
1930 	{0x0970, 0x0970, 1},
1931 	{0x0df4, 0x0df4, 1},
1932 	{0x0e4f, 0x0e4f, 1},
1933 	{0x0e5a, 0x0e5b, 1},
1934 	{0x0f04, 0x0f12, 1},
1935 	{0x0f3a, 0x0f3d, 1},
1936 	{0x0f85, 0x0f85, 1},
1937 	{0x104a, 0x104f, 1},		/* Myanmar punctuation */
1938 	{0x10fb, 0x10fb, 1},		/* Georgian punctuation */
1939 	{0x1361, 0x1368, 1},		/* Ethiopic punctuation */
1940 	{0x166d, 0x166e, 1},		/* Canadian Syl. punctuation */
1941 	{0x1680, 0x1680, 0},
1942 	{0x169b, 0x169c, 1},
1943 	{0x16eb, 0x16ed, 1},
1944 	{0x1735, 0x1736, 1},
1945 	{0x17d4, 0x17dc, 1},		/* Khmer punctuation */
1946 	{0x1800, 0x180a, 1},		/* Mongolian punctuation */
1947 	{0x2000, 0x200b, 0},		/* spaces */
1948 	{0x200c, 0x2027, 1},		/* punctuation and symbols */
1949 	{0x2028, 0x2029, 0},
1950 	{0x202a, 0x202e, 1},		/* punctuation and symbols */
1951 	{0x202f, 0x202f, 0},
1952 	{0x2030, 0x205e, 1},		/* punctuation and symbols */
1953 	{0x205f, 0x205f, 0},
1954 	{0x2060, 0x27ff, 1},		/* punctuation and symbols */
1955 	{0x2070, 0x207f, 0x2070},	/* superscript */
1956 	{0x2080, 0x208f, 0x2080},	/* subscript */
1957 	{0x2983, 0x2998, 1},
1958 	{0x29d8, 0x29db, 1},
1959 	{0x29fc, 0x29fd, 1},
1960 	{0x3000, 0x3000, 0},		/* ideographic space */
1961 	{0x3001, 0x3020, 1},		/* ideographic punctuation */
1962 	{0x3030, 0x3030, 1},
1963 	{0x303d, 0x303d, 1},
1964 	{0x3040, 0x309f, 0x3040},	/* Hiragana */
1965 	{0x30a0, 0x30ff, 0x30a0},	/* Katakana */
1966 	{0x3300, 0x9fff, 0x4e00},	/* CJK Ideographs */
1967 	{0xac00, 0xd7a3, 0xac00},	/* Hangul Syllables */
1968 	{0xf900, 0xfaff, 0x4e00},	/* CJK Ideographs */
1969 	{0xfd3e, 0xfd3f, 1},
1970 	{0xfe30, 0xfe6b, 1},		/* punctuation forms */
1971 	{0xff00, 0xff0f, 1},		/* half/fullwidth ASCII */
1972 	{0xff1a, 0xff20, 1},		/* half/fullwidth ASCII */
1973 	{0xff3b, 0xff40, 1},		/* half/fullwidth ASCII */
1974 	{0xff5b, 0xff65, 1},		/* half/fullwidth ASCII */
1975     };
1976     int bot = 0;
1977     int top = sizeof(classes) / sizeof(struct clinterval) - 1;
1978     int mid;
1979 
1980     /* First quick check for Latin1 characters, use 'iskeyword'. */
1981     if (c < 0x100)
1982     {
1983 	if (c == ' ' || c == '\t' || c == NUL || c == 0xa0)
1984 	    return 0;	    /* blank */
1985 	if (vim_iswordc(c))
1986 	    return 2;	    /* word character */
1987 	return 1;	    /* punctuation */
1988     }
1989 
1990     /* binary search in table */
1991     while (top >= bot)
1992     {
1993 	mid = (bot + top) / 2;
1994 	if (classes[mid].last < c)
1995 	    bot = mid + 1;
1996 	else if (classes[mid].first > c)
1997 	    top = mid - 1;
1998 	else
1999 	    return (int)classes[mid].class;
2000     }
2001 
2002     /* most other characters are "word" characters */
2003     return 2;
2004 }
2005 
2006 /*
2007  * Code for Unicode case-dependent operations.  Based on notes in
2008  * http://www.unicode.org/Public/UNIDATA/CaseFolding.txt
2009  * This code uses simple case folding, not full case folding.
2010  */
2011 
2012 /*
2013  * The following table is built by foldExtract.pl < CaseFolding.txt .
2014  * It must be in numeric order, because we use binary search on it.
2015  * An entry such as {0x41,0x5a,1,32} means that UCS-4 characters in the range
2016  * from 0x41 to 0x5a inclusive, stepping by 1, are folded by adding 32.
2017  */
2018 
2019 typedef struct
2020 {
2021     int rangeStart;
2022     int rangeEnd;
2023     int step;
2024     int offset;
2025 } convertStruct;
2026 
2027 static convertStruct foldCase[] =
2028 {
2029 	{0x41,0x5a,1,32}, {0xc0,0xd6,1,32}, {0xd8,0xde,1,32},
2030 	{0x100,0x12e,2,1}, {0x130,0x130,-1,-199}, {0x132,0x136,2,1},
2031 	{0x139,0x147,2,1}, {0x14a,0x176,2,1}, {0x178,0x178,-1,-121},
2032 	{0x179,0x17d,2,1}, {0x181,0x181,-1,210}, {0x182,0x184,2,1},
2033 	{0x186,0x186,-1,206}, {0x187,0x187,-1,1}, {0x189,0x18a,1,205},
2034 	{0x18b,0x18b,-1,1}, {0x18e,0x18e,-1,79}, {0x18f,0x18f,-1,202},
2035 	{0x190,0x190,-1,203}, {0x191,0x191,-1,1}, {0x193,0x193,-1,205},
2036 	{0x194,0x194,-1,207}, {0x196,0x196,-1,211}, {0x197,0x197,-1,209},
2037 	{0x198,0x198,-1,1}, {0x19c,0x19c,-1,211}, {0x19d,0x19d,-1,213},
2038 	{0x19f,0x19f,-1,214}, {0x1a0,0x1a4,2,1}, {0x1a6,0x1a6,-1,218},
2039 	{0x1a7,0x1a7,-1,1}, {0x1a9,0x1a9,-1,218}, {0x1ac,0x1ac,-1,1},
2040 	{0x1ae,0x1ae,-1,218}, {0x1af,0x1af,-1,1}, {0x1b1,0x1b2,1,217},
2041 	{0x1b3,0x1b5,2,1}, {0x1b7,0x1b7,-1,219}, {0x1b8,0x1bc,4,1},
2042 	{0x1c4,0x1c4,-1,2}, {0x1c5,0x1c5,-1,1}, {0x1c7,0x1c7,-1,2},
2043 	{0x1c8,0x1c8,-1,1}, {0x1ca,0x1ca,-1,2}, {0x1cb,0x1db,2,1},
2044 	{0x1de,0x1ee,2,1}, {0x1f1,0x1f1,-1,2}, {0x1f2,0x1f4,2,1},
2045 	{0x1f6,0x1f6,-1,-97}, {0x1f7,0x1f7,-1,-56}, {0x1f8,0x21e,2,1},
2046 	{0x220,0x220,-1,-130}, {0x222,0x232,2,1}, {0x386,0x386,-1,38},
2047 	{0x388,0x38a,1,37}, {0x38c,0x38c,-1,64}, {0x38e,0x38f,1,63},
2048 	{0x391,0x3a1,1,32}, {0x3a3,0x3ab,1,32}, {0x3d8,0x3ee,2,1},
2049 	{0x3f4,0x3f4,-1,-60}, {0x3f7,0x3f7,-1,1}, {0x3f9,0x3f9,-1,-7},
2050 	{0x3fa,0x3fa,-1,1}, {0x400,0x40f,1,80}, {0x410,0x42f,1,32},
2051 	{0x460,0x480,2,1}, {0x48a,0x4be,2,1}, {0x4c1,0x4cd,2,1},
2052 	{0x4d0,0x4f4,2,1}, {0x4f8,0x500,8,1}, {0x502,0x50e,2,1},
2053 	{0x531,0x556,1,48}, {0x1e00,0x1e94,2,1}, {0x1ea0,0x1ef8,2,1},
2054 	{0x1f08,0x1f0f,1,-8}, {0x1f18,0x1f1d,1,-8}, {0x1f28,0x1f2f,1,-8},
2055 	{0x1f38,0x1f3f,1,-8}, {0x1f48,0x1f4d,1,-8}, {0x1f59,0x1f5f,2,-8},
2056 	{0x1f68,0x1f6f,1,-8}, {0x1f88,0x1f8f,1,-8}, {0x1f98,0x1f9f,1,-8},
2057 	{0x1fa8,0x1faf,1,-8}, {0x1fb8,0x1fb9,1,-8}, {0x1fba,0x1fbb,1,-74},
2058 	{0x1fbc,0x1fbc,-1,-9}, {0x1fc8,0x1fcb,1,-86}, {0x1fcc,0x1fcc,-1,-9},
2059 	{0x1fd8,0x1fd9,1,-8}, {0x1fda,0x1fdb,1,-100}, {0x1fe8,0x1fe9,1,-8},
2060 	{0x1fea,0x1feb,1,-112}, {0x1fec,0x1fec,-1,-7}, {0x1ff8,0x1ff9,1,-128},
2061 	{0x1ffa,0x1ffb,1,-126}, {0x1ffc,0x1ffc,-1,-9}, {0x2126,0x2126,-1,-7517},
2062 	{0x212a,0x212a,-1,-8383}, {0x212b,0x212b,-1,-8262},
2063 	{0x2160,0x216f,1,16}, {0x24b6,0x24cf,1,26}, {0xff21,0xff3a,1,32},
2064 	{0x10400,0x10427,1,40}
2065 };
2066 
2067 static int utf_convert(int a, convertStruct table[], int tableSize);
2068 
2069 /*
2070  * Generic conversion function for case operations.
2071  * Return the converted equivalent of "a", which is a UCS-4 character.  Use
2072  * the given conversion "table".  Uses binary search on "table".
2073  */
2074     static int
2075 utf_convert(a, table, tableSize)
2076     int			a;
2077     convertStruct	table[];
2078     int			tableSize;
2079 {
2080     int start, mid, end; /* indices into table */
2081 
2082     start = 0;
2083     end = tableSize / sizeof(convertStruct);
2084     while (start < end)
2085     {
2086 	/* need to search further */
2087 	mid = (end + start) /2;
2088 	if (table[mid].rangeEnd < a)
2089 	    start = mid + 1;
2090 	else
2091 	    end = mid;
2092     }
2093     if (table[start].rangeStart <= a && a <= table[start].rangeEnd
2094 	    && (a - table[start].rangeStart) % table[start].step == 0)
2095 	return (a + table[start].offset);
2096     else
2097 	return a;
2098 }
2099 
2100 /*
2101  * Return the folded-case equivalent of "a", which is a UCS-4 character.  Uses
2102  * simple case folding.
2103  */
2104     int
2105 utf_fold(a)
2106     int		a;
2107 {
2108     return utf_convert(a, foldCase, sizeof(foldCase));
2109 }
2110 
2111 /*
2112  * The following tables are built by upperLowerExtract.pl < UnicodeData.txt .
2113  * They must be in numeric order, because we use binary search on them.
2114  * An entry such as {0x41,0x5a,1,32} means that UCS-4 characters in the range
2115  * from 0x41 to 0x5a inclusive, stepping by 1, are switched to lower (for
2116  * example) by adding 32.
2117  */
2118 static convertStruct toLower[] =
2119 {
2120 	{0x41,0x5a,1,32}, {0xc0,0xd6,1,32}, {0xd8,0xde,1,32},
2121 	{0x100,0x12e,2,1}, {0x130,0x130,-1,-199}, {0x132,0x136,2,1},
2122 	{0x139,0x147,2,1}, {0x14a,0x176,2,1}, {0x178,0x178,-1,-121},
2123 	{0x179,0x17d,2,1}, {0x181,0x181,-1,210}, {0x182,0x184,2,1},
2124 	{0x186,0x186,-1,206}, {0x187,0x187,-1,1}, {0x189,0x18a,1,205},
2125 	{0x18b,0x18b,-1,1}, {0x18e,0x18e,-1,79}, {0x18f,0x18f,-1,202},
2126 	{0x190,0x190,-1,203}, {0x191,0x191,-1,1}, {0x193,0x193,-1,205},
2127 	{0x194,0x194,-1,207}, {0x196,0x196,-1,211}, {0x197,0x197,-1,209},
2128 	{0x198,0x198,-1,1}, {0x19c,0x19c,-1,211}, {0x19d,0x19d,-1,213},
2129 	{0x19f,0x19f,-1,214}, {0x1a0,0x1a4,2,1}, {0x1a6,0x1a6,-1,218},
2130 	{0x1a7,0x1a7,-1,1}, {0x1a9,0x1a9,-1,218}, {0x1ac,0x1ac,-1,1},
2131 	{0x1ae,0x1ae,-1,218}, {0x1af,0x1af,-1,1}, {0x1b1,0x1b2,1,217},
2132 	{0x1b3,0x1b5,2,1}, {0x1b7,0x1b7,-1,219}, {0x1b8,0x1bc,4,1},
2133 	{0x1c4,0x1ca,3,2}, {0x1cd,0x1db,2,1}, {0x1de,0x1ee,2,1},
2134 	{0x1f1,0x1f1,-1,2}, {0x1f4,0x1f4,-1,1}, {0x1f6,0x1f6,-1,-97},
2135 	{0x1f7,0x1f7,-1,-56}, {0x1f8,0x21e,2,1}, {0x220,0x220,-1,-130},
2136 	{0x222,0x232,2,1}, {0x386,0x386,-1,38}, {0x388,0x38a,1,37},
2137 	{0x38c,0x38c,-1,64}, {0x38e,0x38f,1,63}, {0x391,0x3a1,1,32},
2138 	{0x3a3,0x3ab,1,32}, {0x3d8,0x3ee,2,1}, {0x3f4,0x3f4,-1,-60},
2139 	{0x3f7,0x3f7,-1,1}, {0x3f9,0x3f9,-1,-7}, {0x3fa,0x3fa,-1,1},
2140 	{0x400,0x40f,1,80}, {0x410,0x42f,1,32}, {0x460,0x480,2,1},
2141 	{0x48a,0x4be,2,1}, {0x4c1,0x4cd,2,1}, {0x4d0,0x4f4,2,1},
2142 	{0x4f8,0x500,8,1}, {0x502,0x50e,2,1}, {0x531,0x556,1,48},
2143 	{0x1e00,0x1e94,2,1}, {0x1ea0,0x1ef8,2,1}, {0x1f08,0x1f0f,1,-8},
2144 	{0x1f18,0x1f1d,1,-8}, {0x1f28,0x1f2f,1,-8}, {0x1f38,0x1f3f,1,-8},
2145 	{0x1f48,0x1f4d,1,-8}, {0x1f59,0x1f5f,2,-8}, {0x1f68,0x1f6f,1,-8},
2146 	{0x1fb8,0x1fb9,1,-8}, {0x1fba,0x1fbb,1,-74}, {0x1fc8,0x1fcb,1,-86},
2147 	{0x1fd8,0x1fd9,1,-8}, {0x1fda,0x1fdb,1,-100}, {0x1fe8,0x1fe9,1,-8},
2148 	{0x1fea,0x1feb,1,-112}, {0x1fec,0x1fec,-1,-7}, {0x1ff8,0x1ff9,1,-128},
2149 	{0x1ffa,0x1ffb,1,-126}, {0x2126,0x2126,-1,-7517}, {0x212a,0x212a,-1,-8383},
2150 	{0x212b,0x212b,-1,-8262}, {0xff21,0xff3a,1,32}, {0x10400,0x10427,1,40}
2151 };
2152 
2153 static convertStruct toUpper[] =
2154 {
2155 	{0x61,0x7a,1,-32}, {0xb5,0xb5,-1,743}, {0xe0,0xf6,1,-32},
2156 	{0xf8,0xfe,1,-32}, {0xff,0xff,-1,121}, {0x101,0x12f,2,-1},
2157 	{0x131,0x131,-1,-232}, {0x133,0x137,2,-1}, {0x13a,0x148,2,-1},
2158 	{0x14b,0x177,2,-1}, {0x17a,0x17e,2,-1}, {0x17f,0x17f,-1,-300},
2159 	{0x183,0x185,2,-1}, {0x188,0x18c,4,-1}, {0x192,0x192,-1,-1},
2160 	{0x195,0x195,-1,97}, {0x199,0x199,-1,-1}, {0x19e,0x19e,-1,130},
2161 	{0x1a1,0x1a5,2,-1}, {0x1a8,0x1ad,5,-1}, {0x1b0,0x1b4,4,-1},
2162 	{0x1b6,0x1b9,3,-1}, {0x1bd,0x1bd,-1,-1}, {0x1bf,0x1bf,-1,56},
2163 	{0x1c5,0x1c6,1,-1}, {0x1c8,0x1c9,1,-1}, {0x1cb,0x1cc,1,-1},
2164 	{0x1ce,0x1dc,2,-1}, {0x1dd,0x1dd,-1,-79}, {0x1df,0x1ef,2,-1},
2165 	{0x1f2,0x1f3,1,-1}, {0x1f5,0x1f9,4,-1}, {0x1fb,0x21f,2,-1},
2166 	{0x223,0x233,2,-1}, {0x253,0x253,-1,-210}, {0x254,0x254,-1,-206},
2167 	{0x256,0x257,1,-205}, {0x259,0x259,-1,-202}, {0x25b,0x25b,-1,-203},
2168 	{0x260,0x260,-1,-205}, {0x263,0x263,-1,-207}, {0x268,0x268,-1,-209},
2169 	{0x269,0x26f,6,-211}, {0x272,0x272,-1,-213}, {0x275,0x275,-1,-214},
2170 	{0x280,0x283,3,-218}, {0x288,0x288,-1,-218}, {0x28a,0x28b,1,-217},
2171 	{0x292,0x292,-1,-219}, {0x3ac,0x3ac,-1,-38}, {0x3ad,0x3af,1,-37},
2172 	{0x3b1,0x3c1,1,-32}, {0x3c2,0x3c2,-1,-31}, {0x3c3,0x3cb,1,-32},
2173 	{0x3cc,0x3cc,-1,-64}, {0x3cd,0x3ce,1,-63}, {0x3d0,0x3d0,-1,-62},
2174 	{0x3d1,0x3d1,-1,-57}, {0x3d5,0x3d5,-1,-47}, {0x3d6,0x3d6,-1,-54},
2175 	{0x3d9,0x3ef,2,-1}, {0x3f0,0x3f0,-1,-86}, {0x3f1,0x3f1,-1,-80},
2176 	{0x3f2,0x3f2,-1,7}, {0x3f5,0x3f5,-1,-96}, {0x3f8,0x3fb,3,-1},
2177 	{0x430,0x44f,1,-32}, {0x450,0x45f,1,-80}, {0x461,0x481,2,-1},
2178 	{0x48b,0x4bf,2,-1}, {0x4c2,0x4ce,2,-1}, {0x4d1,0x4f5,2,-1},
2179 	{0x4f9,0x501,8,-1}, {0x503,0x50f,2,-1}, {0x561,0x586,1,-48},
2180 	{0x1e01,0x1e95,2,-1}, {0x1e9b,0x1e9b,-1,-59}, {0x1ea1,0x1ef9,2,-1},
2181 	{0x1f00,0x1f07,1,8}, {0x1f10,0x1f15,1,8}, {0x1f20,0x1f27,1,8},
2182 	{0x1f30,0x1f37,1,8}, {0x1f40,0x1f45,1,8}, {0x1f51,0x1f57,2,8},
2183 	{0x1f60,0x1f67,1,8}, {0x1f70,0x1f71,1,74}, {0x1f72,0x1f75,1,86},
2184 	{0x1f76,0x1f77,1,100}, {0x1f78,0x1f79,1,128}, {0x1f7a,0x1f7b,1,112},
2185 	{0x1f7c,0x1f7d,1,126}, {0x1f80,0x1f87,1,8}, {0x1f90,0x1f97,1,8},
2186 	{0x1fa0,0x1fa7,1,8}, {0x1fb0,0x1fb1,1,8}, {0x1fb3,0x1fb3,-1,9},
2187 	{0x1fbe,0x1fbe,-1,-7205}, {0x1fc3,0x1fc3,-1,9}, {0x1fd0,0x1fd1,1,8},
2188 	{0x1fe0,0x1fe1,1,8}, {0x1fe5,0x1fe5,-1,7}, {0x1ff3,0x1ff3,-1,9},
2189 	{0xff41,0xff5a,1,-32}, {0x10428,0x1044f,1,-40}
2190 };
2191 
2192 /*
2193  * Return the upper-case equivalent of "a", which is a UCS-4 character.  Use
2194  * simple case folding.
2195  */
2196     int
2197 utf_toupper(a)
2198     int		a;
2199 {
2200     /* If 'casemap' contains "keepascii" use ASCII style toupper(). */
2201     if (a < 128 && (cmp_flags & CMP_KEEPASCII))
2202 	return TOUPPER_ASC(a);
2203 
2204 #if defined(HAVE_TOWUPPER) && defined(__STDC__ISO_10646__)
2205     /* If towupper() is availble and handles Unicode, use it. */
2206     if (!(cmp_flags & CMP_INTERNAL))
2207 	return towupper(a);
2208 #endif
2209 
2210     /* For characters below 128 use locale sensitive toupper(). */
2211     if (a < 128)
2212 	return TOUPPER_LOC(a);
2213 
2214     /* For any other characters use the above mapping table. */
2215     return utf_convert(a, toUpper, sizeof(toUpper));
2216 }
2217 
2218     int
2219 utf_islower(a)
2220     int		a;
2221 {
2222     return (utf_toupper(a) != a);
2223 }
2224 
2225 /*
2226  * Return the lower-case equivalent of "a", which is a UCS-4 character.  Use
2227  * simple case folding.
2228  */
2229     int
2230 utf_tolower(a)
2231     int		a;
2232 {
2233     /* If 'casemap' contains "keepascii" use ASCII style tolower(). */
2234     if (a < 128 && (cmp_flags & CMP_KEEPASCII))
2235 	return TOLOWER_ASC(a);
2236 
2237 #if defined(HAVE_TOWLOWER) && defined(__STDC__ISO_10646__)
2238     /* If towlower() is available and handles Unicode, use it. */
2239     if (!(cmp_flags & CMP_INTERNAL))
2240 	return towlower(a);
2241 #endif
2242 
2243     /* For characters below 128 use locale sensitive tolower(). */
2244     if (a < 128)
2245 	return TOLOWER_LOC(a);
2246 
2247     /* For any other characters use the above mapping table. */
2248     return utf_convert(a, toLower, sizeof(toLower));
2249 }
2250 
2251     int
2252 utf_isupper(a)
2253     int		a;
2254 {
2255     return (utf_tolower(a) != a);
2256 }
2257 
2258 /*
2259  * Version of strnicmp() that handles multi-byte characters.
2260  * Needed for Big5, Sjift-JIS and UTF-8 encoding.  Other DBCS encodings can
2261  * probably use strnicmp(), because there are no ASCII characters in the
2262  * second byte.
2263  * Returns zero if s1 and s2 are equal (ignoring case), the difference between
2264  * two characters otherwise.
2265  */
2266     int
2267 mb_strnicmp(s1, s2, nn)
2268     char_u	*s1, *s2;
2269     size_t	nn;
2270 {
2271     int		i, j, l;
2272     int		cdiff;
2273     int		incomplete = FALSE;
2274     int		n = nn;
2275 
2276     for (i = 0; i < n; i += l)
2277     {
2278 	if (s1[i] == NUL && s2[i] == NUL)   /* both strings end */
2279 	    return 0;
2280 	if (enc_utf8)
2281 	{
2282 	    l = utf_byte2len(s1[i]);
2283 	    if (l > n - i)
2284 	    {
2285 		l = n - i;		    /* incomplete character */
2286 		incomplete = TRUE;
2287 	    }
2288 	    /* Check directly first, it's faster. */
2289 	    for (j = 0; j < l; ++j)
2290 		if (s1[i + j] != s2[i + j])
2291 		    break;
2292 	    if (j < l)
2293 	    {
2294 		/* If one of the two characters is incomplete return -1. */
2295 		if (incomplete || i + utf_byte2len(s2[i]) > n)
2296 		    return -1;
2297 		cdiff = utf_fold(utf_ptr2char(s1 + i))
2298 					     - utf_fold(utf_ptr2char(s2 + i));
2299 		if (cdiff != 0)
2300 		    return cdiff;
2301 	    }
2302 	}
2303 	else
2304 	{
2305 	    l = (*mb_ptr2len)(s1 + i);
2306 	    if (l <= 1)
2307 	    {
2308 		/* Single byte: first check normally, then with ignore case. */
2309 		if (s1[i] != s2[i])
2310 		{
2311 		    cdiff = TOLOWER_LOC(s1[i]) - TOLOWER_LOC(s2[i]);
2312 		    if (cdiff != 0)
2313 			return cdiff;
2314 		}
2315 	    }
2316 	    else
2317 	    {
2318 		/* For non-Unicode multi-byte don't ignore case. */
2319 		if (l > n - i)
2320 		    l = n - i;
2321 		cdiff = STRNCMP(s1 + i, s2 + i, l);
2322 		if (cdiff != 0)
2323 		    return cdiff;
2324 	    }
2325 	}
2326     }
2327     return 0;
2328 }
2329 
2330 /*
2331  * "g8": show bytes of the UTF-8 char under the cursor.  Doesn't matter what
2332  * 'encoding' has been set to.
2333  */
2334     void
2335 show_utf8()
2336 {
2337     int		len;
2338     int		rlen = 0;
2339     char_u	*line;
2340     int		clen;
2341     int		i;
2342 
2343     /* Get the byte length of the char under the cursor, including composing
2344      * characters. */
2345     line = ml_get_cursor();
2346     len = utfc_ptr2len(line);
2347     if (len == 0)
2348     {
2349 	MSG("NUL");
2350 	return;
2351     }
2352 
2353     clen = 0;
2354     for (i = 0; i < len; ++i)
2355     {
2356 	if (clen == 0)
2357 	{
2358 	    /* start of (composing) character, get its length */
2359 	    if (i > 0)
2360 	    {
2361 		STRCPY(IObuff + rlen, "+ ");
2362 		rlen += 2;
2363 	    }
2364 	    clen = utf_ptr2len(line + i);
2365 	}
2366 	sprintf((char *)IObuff + rlen, "%02x ", line[i]);
2367 	--clen;
2368 	rlen += STRLEN(IObuff + rlen);
2369 	if (rlen > IOSIZE - 20)
2370 	    break;
2371     }
2372 
2373     msg(IObuff);
2374 }
2375 
2376 /*
2377  * mb_head_off() function pointer.
2378  * Return offset from "p" to the first byte of the character it points into.
2379  * Returns 0 when already at the first byte of a character.
2380  */
2381 /*ARGSUSED*/
2382     int
2383 latin_head_off(base, p)
2384     char_u	*base;
2385     char_u	*p;
2386 {
2387     return 0;
2388 }
2389 
2390     int
2391 dbcs_head_off(base, p)
2392     char_u	*base;
2393     char_u	*p;
2394 {
2395     char_u	*q;
2396 
2397     /* It can't be a trailing byte when not using DBCS, at the start of the
2398      * string or the previous byte can't start a double-byte. */
2399     if (p <= base || MB_BYTE2LEN(p[-1]) == 1)
2400 	return 0;
2401 
2402     /* This is slow: need to start at the base and go forward until the
2403      * byte we are looking for.  Return 1 when we went past it, 0 otherwise. */
2404     q = base;
2405     while (q < p)
2406 	q += dbcs_ptr2len(q);
2407     return (q == p) ? 0 : 1;
2408 }
2409 
2410 #if defined(FEAT_CLIPBOARD) || defined(FEAT_GUI) || defined(FEAT_RIGHTLEFT) \
2411 	|| defined(PROTO)
2412 /*
2413  * Special version of dbcs_head_off() that works for ScreenLines[], where
2414  * single-width DBCS_JPNU characters are stored separately.
2415  */
2416     int
2417 dbcs_screen_head_off(base, p)
2418     char_u	*base;
2419     char_u	*p;
2420 {
2421     char_u	*q;
2422 
2423     /* It can't be a trailing byte when not using DBCS, at the start of the
2424      * string or the previous byte can't start a double-byte.
2425      * For euc-jp an 0x8e byte in the previous cell always means we have a
2426      * lead byte in the current cell. */
2427     if (p <= base
2428 	    || (enc_dbcs == DBCS_JPNU && p[-1] == 0x8e)
2429 	    || MB_BYTE2LEN(p[-1]) == 1)
2430 	return 0;
2431 
2432     /* This is slow: need to start at the base and go forward until the
2433      * byte we are looking for.  Return 1 when we went past it, 0 otherwise.
2434      * For DBCS_JPNU look out for 0x8e, which means the second byte is not
2435      * stored as the next byte. */
2436     q = base;
2437     while (q < p)
2438     {
2439 	if (enc_dbcs == DBCS_JPNU && *q == 0x8e)
2440 	    ++q;
2441 	else
2442 	    q += dbcs_ptr2len(q);
2443     }
2444     return (q == p) ? 0 : 1;
2445 }
2446 #endif
2447 
2448     int
2449 utf_head_off(base, p)
2450     char_u	*base;
2451     char_u	*p;
2452 {
2453     char_u	*q;
2454     char_u	*s;
2455     int		c;
2456 #ifdef FEAT_ARABIC
2457     char_u	*j;
2458 #endif
2459 
2460     if (*p < 0x80)		/* be quick for ASCII */
2461 	return 0;
2462 
2463     /* Skip backwards over trailing bytes: 10xx.xxxx
2464      * Skip backwards again if on a composing char. */
2465     for (q = p; ; --q)
2466     {
2467 	/* Move s to the last byte of this char. */
2468 	for (s = q; (s[1] & 0xc0) == 0x80; ++s)
2469 	    ;
2470 	/* Move q to the first byte of this char. */
2471 	while (q > base && (*q & 0xc0) == 0x80)
2472 	    --q;
2473 	/* Check for illegal sequence. Do allow an illegal byte after where we
2474 	 * started. */
2475 	if (utf8len_tab[*q] != (int)(s - q + 1)
2476 				       && utf8len_tab[*q] != (int)(p - q + 1))
2477 	    return 0;
2478 
2479 	if (q <= base)
2480 	    break;
2481 
2482 	c = utf_ptr2char(q);
2483 	if (utf_iscomposing(c))
2484 	    continue;
2485 
2486 #ifdef FEAT_ARABIC
2487 	if (arabic_maycombine(c))
2488 	{
2489 	    /* Advance to get a sneak-peak at the next char */
2490 	    j = q;
2491 	    --j;
2492 	    /* Move j to the first byte of this char. */
2493 	    while (j > base && (*j & 0xc0) == 0x80)
2494 		--j;
2495 	    if (arabic_combine(utf_ptr2char(j), c))
2496 		continue;
2497 	}
2498 #endif
2499 	break;
2500     }
2501 
2502     return (int)(p - q);
2503 }
2504 
2505 #if defined(FEAT_EVAL) || defined(PROTO)
2506 /*
2507  * Copy a character from "*fp" to "*tp" and advance the pointers.
2508  */
2509     void
2510 mb_copy_char(fp, tp)
2511     char_u	**fp;
2512     char_u	**tp;
2513 {
2514     int	    l = (*mb_ptr2len)(*fp);
2515 
2516     mch_memmove(*tp, *fp, (size_t)l);
2517     *tp += l;
2518     *fp += l;
2519 }
2520 #endif
2521 
2522 /*
2523  * Return the offset from "p" to the first byte of a character.  When "p" is
2524  * at the start of a character 0 is returned, otherwise the offset to the next
2525  * character.  Can start anywhere in a stream of bytes.
2526  */
2527     int
2528 mb_off_next(base, p)
2529     char_u	*base;
2530     char_u	*p;
2531 {
2532     int		i;
2533     int		j;
2534 
2535     if (enc_utf8)
2536     {
2537 	if (*p < 0x80)		/* be quick for ASCII */
2538 	    return 0;
2539 
2540 	/* Find the next character that isn't 10xx.xxxx */
2541 	for (i = 0; (p[i] & 0xc0) == 0x80; ++i)
2542 	    ;
2543 	if (i > 0)
2544 	{
2545 	    /* Check for illegal sequence. */
2546 	    for (j = 0; p - j > base; ++j)
2547 		if ((p[-j] & 0xc0) != 0x80)
2548 		    break;
2549 	    if (utf8len_tab[p[-j]] != i + j)
2550 		return 0;
2551 	}
2552 	return i;
2553     }
2554 
2555     /* Only need to check if we're on a trail byte, it doesn't matter if we
2556      * want the offset to the next or current character. */
2557     return (*mb_head_off)(base, p);
2558 }
2559 
2560 /*
2561  * Return the offset from "p" to the last byte of the character it points
2562  * into.  Can start anywhere in a stream of bytes.
2563  */
2564     int
2565 mb_tail_off(base, p)
2566     char_u	*base;
2567     char_u	*p;
2568 {
2569     int		i;
2570     int		j;
2571 
2572     if (*p == NUL)
2573 	return 0;
2574 
2575     if (enc_utf8)
2576     {
2577 	/* Find the last character that is 10xx.xxxx */
2578 	for (i = 0; (p[i + 1] & 0xc0) == 0x80; ++i)
2579 	    ;
2580 	/* Check for illegal sequence. */
2581 	for (j = 0; p - j > base; ++j)
2582 	    if ((p[-j] & 0xc0) != 0x80)
2583 		break;
2584 	if (utf8len_tab[p[-j]] != i + j + 1)
2585 	    return 0;
2586 	return i;
2587     }
2588 
2589     /* It can't be the first byte if a double-byte when not using DBCS, at the
2590      * end of the string or the byte can't start a double-byte. */
2591     if (enc_dbcs == 0 || p[1] == NUL || MB_BYTE2LEN(*p) == 1)
2592 	return 0;
2593 
2594     /* Return 1 when on the lead byte, 0 when on the tail byte. */
2595     return 1 - dbcs_head_off(base, p);
2596 }
2597 
2598 /*
2599  * Find the next illegal byte sequence.
2600  */
2601     void
2602 utf_find_illegal()
2603 {
2604     pos_T	pos = curwin->w_cursor;
2605     char_u	*p;
2606     int		len;
2607     vimconv_T	vimconv;
2608     char_u	*tofree = NULL;
2609 
2610     vimconv.vc_type = CONV_NONE;
2611     if (enc_utf8 && (enc_canon_props(curbuf->b_p_fenc) & ENC_8BIT))
2612     {
2613 	/* 'encoding' is "utf-8" but we are editing a 8-bit encoded file,
2614 	 * possibly a utf-8 file with illegal bytes.  Setup for conversion
2615 	 * from utf-8 to 'fileencoding'. */
2616 	convert_setup(&vimconv, p_enc, curbuf->b_p_fenc);
2617     }
2618 
2619 #ifdef FEAT_VIRTUALEDIT
2620     curwin->w_cursor.coladd = 0;
2621 #endif
2622     for (;;)
2623     {
2624 	p = ml_get_cursor();
2625 	if (vimconv.vc_type != CONV_NONE)
2626 	{
2627 	    vim_free(tofree);
2628 	    tofree = string_convert(&vimconv, p, NULL);
2629 	    if (tofree == NULL)
2630 		break;
2631 	    p = tofree;
2632 	}
2633 
2634 	while (*p != NUL)
2635 	{
2636 	    /* Illegal means that there are not enough trail bytes (checked by
2637 	     * utf_ptr2len()) or too many of them (overlong sequence). */
2638 	    len = utf_ptr2len(p);
2639 	    if (*p >= 0x80 && (len == 1
2640 				     || utf_char2len(utf_ptr2char(p)) != len))
2641 	    {
2642 		if (vimconv.vc_type == CONV_NONE)
2643 		    curwin->w_cursor.col += p - ml_get_cursor();
2644 		else
2645 		{
2646 		    int	    l;
2647 
2648 		    len = p - tofree;
2649 		    for (p = ml_get_cursor(); *p != NUL && len-- > 0; p += l)
2650 		    {
2651 			l = utf_ptr2len(p);
2652 			curwin->w_cursor.col += l;
2653 		    }
2654 		}
2655 		goto theend;
2656 	    }
2657 	    p += len;
2658 	}
2659 	if (curwin->w_cursor.lnum == curbuf->b_ml.ml_line_count)
2660 	    break;
2661 	++curwin->w_cursor.lnum;
2662 	curwin->w_cursor.col = 0;
2663     }
2664 
2665     /* didn't find it: don't move and beep */
2666     curwin->w_cursor = pos;
2667     beep_flush();
2668 
2669 theend:
2670     vim_free(tofree);
2671     convert_setup(&vimconv, NULL, NULL);
2672 }
2673 
2674 #if defined(HAVE_GTK2) || defined(PROTO)
2675 /*
2676  * Return TRUE if string "s" is a valid utf-8 string.
2677  * When "end" is NULL stop at the first NUL.
2678  * When "end" is positive stop there.
2679  */
2680     int
2681 utf_valid_string(s, end)
2682     char_u	*s;
2683     char_u	*end;
2684 {
2685     int		l;
2686     char_u	*p = s;
2687 
2688     while (end == NULL ? *p != NUL : p < end)
2689     {
2690 	if ((*p & 0xc0) == 0x80)
2691 	    return FALSE;	/* invalid lead byte */
2692 	l = utf8len_tab[*p];
2693 	if (end != NULL && p + l > end)
2694 	    return FALSE;	/* incomplete byte sequence */
2695 	++p;
2696 	while (--l > 0)
2697 	    if ((*p++ & 0xc0) != 0x80)
2698 		return FALSE;	/* invalid trail byte */
2699     }
2700     return TRUE;
2701 }
2702 #endif
2703 
2704 #if defined(FEAT_GUI) || defined(PROTO)
2705 /*
2706  * Special version of mb_tail_off() for use in ScreenLines[].
2707  */
2708     int
2709 dbcs_screen_tail_off(base, p)
2710     char_u	*base;
2711     char_u	*p;
2712 {
2713     /* It can't be the first byte if a double-byte when not using DBCS, at the
2714      * end of the string or the byte can't start a double-byte.
2715      * For euc-jp an 0x8e byte always means we have a lead byte in the current
2716      * cell. */
2717     if (*p == NUL || p[1] == NUL
2718 	    || (enc_dbcs == DBCS_JPNU && *p == 0x8e)
2719 	    || MB_BYTE2LEN(*p) == 1)
2720 	return 0;
2721 
2722     /* Return 1 when on the lead byte, 0 when on the tail byte. */
2723     return 1 - dbcs_screen_head_off(base, p);
2724 }
2725 #endif
2726 
2727 /*
2728  * If the cursor moves on an trail byte, set the cursor on the lead byte.
2729  * Thus it moves left if necessary.
2730  * Return TRUE when the cursor was adjusted.
2731  */
2732     void
2733 mb_adjust_cursor()
2734 {
2735     mb_adjustpos(&curwin->w_cursor);
2736 }
2737 
2738 /*
2739  * Adjust position "*lp" to point to the first byte of a multi-byte character.
2740  * If it points to a tail byte it's moved backwards to the head byte.
2741  */
2742     void
2743 mb_adjustpos(lp)
2744     pos_T	*lp;
2745 {
2746     char_u	*p;
2747 
2748     if (lp->col > 0
2749 #ifdef FEAT_VIRTUALEDIT
2750 	    || lp->coladd > 1
2751 #endif
2752 	    )
2753     {
2754 	p = ml_get(lp->lnum);
2755 	lp->col -= (*mb_head_off)(p, p + lp->col);
2756 #ifdef FEAT_VIRTUALEDIT
2757 	/* Reset "coladd" when the cursor would be on the right half of a
2758 	 * double-wide character. */
2759 	if (lp->coladd == 1
2760 		&& p[lp->col] != TAB
2761 		&& vim_isprintc((*mb_ptr2char)(p + lp->col))
2762 		&& ptr2cells(p + lp->col) > 1)
2763 	    lp->coladd = 0;
2764 #endif
2765     }
2766 }
2767 
2768 /*
2769  * Return a pointer to the character before "*p", if there is one.
2770  */
2771     char_u *
2772 mb_prevptr(line, p)
2773     char_u *line;	/* start of the string */
2774     char_u *p;
2775 {
2776     if (p > line)
2777 	mb_ptr_back(line, p);
2778     return p;
2779 }
2780 
2781 /*
2782  * Return the character length of "str".  Each multi-byte character (with
2783  * following composing characters) counts as one.
2784  */
2785     int
2786 mb_charlen(str)
2787     char_u	*str;
2788 {
2789     char_u	*p = str;
2790     int		count;
2791 
2792     if (p == NULL)
2793 	return 0;
2794 
2795     for (count = 0; *p != NUL; count++)
2796 	p += (*mb_ptr2len)(p);
2797 
2798     return count;
2799 }
2800 
2801 #if defined(FEAT_SPELL) || defined(PROTO)
2802 /*
2803  * Like mb_charlen() but for a string with specified length.
2804  */
2805     int
2806 mb_charlen_len(str, len)
2807     char_u	*str;
2808     int		len;
2809 {
2810     char_u	*p = str;
2811     int		count;
2812 
2813     for (count = 0; *p != NUL && p < str + len; count++)
2814 	p += (*mb_ptr2len)(p);
2815 
2816     return count;
2817 }
2818 #endif
2819 
2820 /*
2821  * Try to un-escape a multi-byte character.
2822  * Used for the "to" and "from" part of a mapping.
2823  * Return the un-escaped string if it is a multi-byte character, and advance
2824  * "pp" to just after the bytes that formed it.
2825  * Return NULL if no multi-byte char was found.
2826  */
2827     char_u *
2828 mb_unescape(pp)
2829     char_u **pp;
2830 {
2831     static char_u	buf[MB_MAXBYTES + 1];
2832     int			n, m = 0;
2833     char_u		*str = *pp;
2834 
2835     /* Must translate K_SPECIAL KS_SPECIAL KE_FILLER to K_SPECIAL and CSI
2836      * KS_EXTRA KE_CSI to CSI. */
2837     for (n = 0; str[n] != NUL && m <= MB_MAXBYTES; ++n)
2838     {
2839 	if (str[n] == K_SPECIAL
2840 		&& str[n + 1] == KS_SPECIAL
2841 		&& str[n + 2] == KE_FILLER)
2842 	{
2843 	    buf[m++] = K_SPECIAL;
2844 	    n += 2;
2845 	}
2846 # ifdef FEAT_GUI
2847 	else if (str[n] == CSI
2848 		&& str[n + 1] == KS_EXTRA
2849 		&& str[n + 2] == (int)KE_CSI)
2850 	{
2851 	    buf[m++] = CSI;
2852 	    n += 2;
2853 	}
2854 # endif
2855 	else if (str[n] == K_SPECIAL
2856 # ifdef FEAT_GUI
2857 		|| str[n] == CSI
2858 # endif
2859 		)
2860 	    break;		/* a special key can't be a multibyte char */
2861 	else
2862 	    buf[m++] = str[n];
2863 	buf[m] = NUL;
2864 
2865 	/* Return a multi-byte character if it's found.  An illegal sequence
2866 	 * will result in a 1 here. */
2867 	if ((*mb_ptr2len)(buf) > 1)
2868 	{
2869 	    *pp = str + n + 1;
2870 	    return buf;
2871 	}
2872     }
2873     return NULL;
2874 }
2875 
2876 /*
2877  * Return TRUE if the character at "row"/"col" on the screen is the left side
2878  * of a double-width character.
2879  * Caller must make sure "row" and "col" are not invalid!
2880  */
2881     int
2882 mb_lefthalve(row, col)
2883     int	    row;
2884     int	    col;
2885 {
2886 #ifdef FEAT_HANGULIN
2887     if (composing_hangul)
2888 	return TRUE;
2889 #endif
2890     if (enc_dbcs != 0)
2891 	return dbcs_off2cells(LineOffset[row] + col) > 1;
2892     if (enc_utf8)
2893 	return (col + 1 < Columns
2894 		&& ScreenLines[LineOffset[row] + col + 1] == 0);
2895     return FALSE;
2896 }
2897 
2898 # if defined(FEAT_CLIPBOARD) || defined(FEAT_GUI) || defined(FEAT_RIGHTLEFT) \
2899 	|| defined(PROTO)
2900 /*
2901  * Correct a position on the screen, if it's the right halve of a double-wide
2902  * char move it to the left halve.  Returns the corrected column.
2903  */
2904     int
2905 mb_fix_col(col, row)
2906     int		col;
2907     int		row;
2908 {
2909     col = check_col(col);
2910     row = check_row(row);
2911     if (has_mbyte && ScreenLines != NULL && col > 0
2912 	    && ((enc_dbcs
2913 		    && ScreenLines[LineOffset[row] + col] != NUL
2914 		    && dbcs_screen_head_off(ScreenLines + LineOffset[row],
2915 					 ScreenLines + LineOffset[row] + col))
2916 		|| (enc_utf8 && ScreenLines[LineOffset[row] + col] == 0)))
2917 	--col;
2918     return col;
2919 }
2920 # endif
2921 #endif
2922 
2923 #if defined(FEAT_MBYTE) || defined(FEAT_POSTSCRIPT) || defined(PROTO)
2924 static int enc_alias_search __ARGS((char_u *name));
2925 
2926 /*
2927  * Skip the Vim specific head of a 'encoding' name.
2928  */
2929     char_u *
2930 enc_skip(p)
2931     char_u	*p;
2932 {
2933     if (STRNCMP(p, "2byte-", 6) == 0)
2934 	return p + 6;
2935     if (STRNCMP(p, "8bit-", 5) == 0)
2936 	return p + 5;
2937     return p;
2938 }
2939 
2940 /*
2941  * Find the canonical name for encoding "enc".
2942  * When the name isn't recognized, returns "enc" itself, but with all lower
2943  * case characters and '_' replaced with '-'.
2944  * Returns an allocated string.  NULL for out-of-memory.
2945  */
2946     char_u *
2947 enc_canonize(enc)
2948     char_u	*enc;
2949 {
2950     char_u	*r;
2951     char_u	*p, *s;
2952     int		i;
2953 
2954 # ifdef FEAT_MBYTE
2955     if (STRCMP(enc, "default") == 0)
2956     {
2957 	/* Use the default encoding as it's found by set_init_1(). */
2958 	r = get_encoding_default();
2959 	if (r == NULL)
2960 	    r = (char_u *)"latin1";
2961 	return vim_strsave(r);
2962     }
2963 # endif
2964 
2965     /* copy "enc" to allocted memory, with room for two '-' */
2966     r = alloc((unsigned)(STRLEN(enc) + 3));
2967     if (r != NULL)
2968     {
2969 	/* Make it all lower case and replace '_' with '-'. */
2970 	p = r;
2971 	for (s = enc; *s != NUL; ++s)
2972 	{
2973 	    if (*s == '_')
2974 		*p++ = '-';
2975 	    else
2976 		*p++ = TOLOWER_ASC(*s);
2977 	}
2978 	*p = NUL;
2979 
2980 	/* Skip "2byte-" and "8bit-". */
2981 	p = enc_skip(r);
2982 
2983 	/* Change "microsoft-cp" to "cp".  Used in some spell files. */
2984 	if (STRNCMP(p, "microsoft-cp", 12) == 0)
2985 	    mch_memmove(p, p + 10, STRLEN(p + 10) + 1);
2986 
2987 	/* "iso8859" -> "iso-8859" */
2988 	if (STRNCMP(p, "iso8859", 7) == 0)
2989 	{
2990 	    mch_memmove(p + 4, p + 3, STRLEN(p + 2));
2991 	    p[3] = '-';
2992 	}
2993 
2994 	/* "iso-8859n" -> "iso-8859-n" */
2995 	if (STRNCMP(p, "iso-8859", 8) == 0 && p[8] != '-')
2996 	{
2997 	    mch_memmove(p + 9, p + 8, STRLEN(p + 7));
2998 	    p[8] = '-';
2999 	}
3000 
3001 	/* "latin-N" -> "latinN" */
3002 	if (STRNCMP(p, "latin-", 6) == 0)
3003 	    mch_memmove(p + 5, p + 6, STRLEN(p + 5));
3004 
3005 	if (enc_canon_search(p) >= 0)
3006 	{
3007 	    /* canonical name can be used unmodified */
3008 	    if (p != r)
3009 		mch_memmove(r, p, STRLEN(p) + 1);
3010 	}
3011 	else if ((i = enc_alias_search(p)) >= 0)
3012 	{
3013 	    /* alias recognized, get canonical name */
3014 	    vim_free(r);
3015 	    r = vim_strsave((char_u *)enc_canon_table[i].name);
3016 	}
3017     }
3018     return r;
3019 }
3020 
3021 /*
3022  * Search for an encoding alias of "name".
3023  * Returns -1 when not found.
3024  */
3025     static int
3026 enc_alias_search(name)
3027     char_u	*name;
3028 {
3029     int		i;
3030 
3031     for (i = 0; enc_alias_table[i].name != NULL; ++i)
3032 	if (STRCMP(name, enc_alias_table[i].name) == 0)
3033 	    return enc_alias_table[i].canon;
3034     return -1;
3035 }
3036 #endif
3037 
3038 #if defined(FEAT_MBYTE) || defined(PROTO)
3039 
3040 #ifdef HAVE_LANGINFO_H
3041 # include <langinfo.h>
3042 #endif
3043 
3044 /*
3045  * Get the canonicalized encoding of the current locale.
3046  * Returns an allocated string when successful, NULL when not.
3047  */
3048     char_u *
3049 enc_locale()
3050 {
3051 #ifndef WIN3264
3052     char	*s;
3053     char	*p;
3054     int		i;
3055 #endif
3056     char	buf[50];
3057 #ifdef WIN3264
3058     long	acp = GetACP();
3059 
3060     if (acp == 1200)
3061 	STRCPY(buf, "ucs-2le");
3062     else if (acp == 1252)	    /* cp1252 is used as latin1 */
3063 	STRCPY(buf, "latin1");
3064     else
3065 	sprintf(buf, "cp%ld", acp);
3066 #else
3067 # ifdef HAVE_NL_LANGINFO_CODESET
3068     if ((s = nl_langinfo(CODESET)) == NULL || *s == NUL)
3069 # endif
3070 #  if defined(HAVE_LOCALE_H) || defined(X_LOCALE)
3071 	if ((s = setlocale(LC_CTYPE, NULL)) == NULL || *s == NUL)
3072 #  endif
3073 	    if ((s = getenv("LC_ALL")) == NULL || *s == NUL)
3074 		if ((s = getenv("LC_CTYPE")) == NULL || *s == NUL)
3075 		    s = getenv("LANG");
3076 
3077     if (s == NULL || *s == NUL)
3078 	return FAIL;
3079 
3080     /* The most generic locale format is:
3081      * language[_territory][.codeset][@modifier][+special][,[sponsor][_revision]]
3082      * If there is a '.' remove the part before it.
3083      * if there is something after the codeset, remove it.
3084      * Make the name lowercase and replace '_' with '-'.
3085      * Exception: "ja_JP.EUC" == "euc-jp", "zh_CN.EUC" = "euc-cn",
3086      * "ko_KR.EUC" == "euc-kr"
3087      */
3088     if ((p = (char *)vim_strchr((char_u *)s, '.')) != NULL)
3089     {
3090 	if (p > s + 2 && STRNICMP(p + 1, "EUC", 3) == 0
3091 			&& !isalnum((int)p[4]) && p[4] != '-' && p[-3] == '_')
3092 	{
3093 	    /* copy "XY.EUC" to "euc-XY" to buf[10] */
3094 	    STRCPY(buf + 10, "euc-");
3095 	    buf[14] = p[-2];
3096 	    buf[15] = p[-1];
3097 	    buf[16] = 0;
3098 	    s = buf + 10;
3099 	}
3100 	else
3101 	    s = p + 1;
3102     }
3103     for (i = 0; s[i] != NUL && i < sizeof(buf) - 1; ++i)
3104     {
3105 	if (s[i] == '_' || s[i] == '-')
3106 	    buf[i] = '-';
3107 	else if (isalnum((int)s[i]))
3108 	    buf[i] = TOLOWER_ASC(s[i]);
3109 	else
3110 	    break;
3111     }
3112     buf[i] = NUL;
3113 #endif
3114 
3115     return enc_canonize((char_u *)buf);
3116 }
3117 
3118 #if defined(WIN3264) || defined(PROTO)
3119 /*
3120  * Convert an encoding name to an MS-Windows codepage.
3121  * Returns zero if no codepage can be figured out.
3122  */
3123     int
3124 encname2codepage(name)
3125     char_u	*name;
3126 {
3127     int		cp;
3128     char_u	*p = name;
3129     int		idx;
3130 
3131     if (STRNCMP(p, "8bit-", 5) == 0)
3132 	p += 5;
3133     else if (STRNCMP(p_enc, "2byte-", 6) == 0)
3134 	p += 6;
3135 
3136     if (p[0] == 'c' && p[1] == 'p')
3137 	cp = atoi(p + 2);
3138     else if ((idx = enc_canon_search(p)) >= 0)
3139 	cp = enc_canon_table[idx].codepage;
3140     else
3141 	return 0;
3142     if (IsValidCodePage(cp))
3143 	return cp;
3144     return 0;
3145 }
3146 #endif
3147 
3148 # if defined(USE_ICONV) || defined(PROTO)
3149 
3150 static char_u *iconv_string __ARGS((vimconv_T *vcp, char_u *str, int slen, int *unconvlenp));
3151 
3152 /*
3153  * Call iconv_open() with a check if iconv() works properly (there are broken
3154  * versions).
3155  * Returns (void *)-1 if failed.
3156  * (should return iconv_t, but that causes problems with prototypes).
3157  */
3158     void *
3159 my_iconv_open(to, from)
3160     char_u	*to;
3161     char_u	*from;
3162 {
3163     iconv_t	fd;
3164 #define ICONV_TESTLEN 400
3165     char_u	tobuf[ICONV_TESTLEN];
3166     char	*p;
3167     size_t	tolen;
3168     static int	iconv_ok = -1;
3169 
3170     if (iconv_ok == FALSE)
3171 	return (void *)-1;	/* detected a broken iconv() previously */
3172 
3173 #ifdef DYNAMIC_ICONV
3174     /* Check if the iconv.dll can be found. */
3175     if (!iconv_enabled(TRUE))
3176 	return (void *)-1;
3177 #endif
3178 
3179     fd = iconv_open((char *)enc_skip(to), (char *)enc_skip(from));
3180 
3181     if (fd != (iconv_t)-1 && iconv_ok == -1)
3182     {
3183 	/*
3184 	 * Do a dummy iconv() call to check if it actually works.  There is a
3185 	 * version of iconv() on Linux that is broken.  We can't ignore it,
3186 	 * because it's wide-spread.  The symptoms are that after outputting
3187 	 * the initial shift state the "to" pointer is NULL and conversion
3188 	 * stops for no apparent reason after about 8160 characters.
3189 	 */
3190 	p = (char *)tobuf;
3191 	tolen = ICONV_TESTLEN;
3192 	(void)iconv(fd, NULL, NULL, &p, &tolen);
3193 	if (p == NULL)
3194 	{
3195 	    iconv_ok = FALSE;
3196 	    iconv_close(fd);
3197 	    fd = (iconv_t)-1;
3198 	}
3199 	else
3200 	    iconv_ok = TRUE;
3201     }
3202 
3203     return (void *)fd;
3204 }
3205 
3206 /*
3207  * Convert the string "str[slen]" with iconv().
3208  * If "unconvlenp" is not NULL handle the string ending in an incomplete
3209  * sequence and set "*unconvlenp" to the length of it.
3210  * Returns the converted string in allocated memory.  NULL for an error.
3211  */
3212     static char_u *
3213 iconv_string(vcp, str, slen, unconvlenp)
3214     vimconv_T	*vcp;
3215     char_u	*str;
3216     int		slen;
3217     int		*unconvlenp;
3218 {
3219     const char	*from;
3220     size_t	fromlen;
3221     char	*to;
3222     size_t	tolen;
3223     size_t	len = 0;
3224     size_t	done = 0;
3225     char_u	*result = NULL;
3226     char_u	*p;
3227     int		l;
3228 
3229     from = (char *)str;
3230     fromlen = slen;
3231     for (;;)
3232     {
3233 	if (len == 0 || ICONV_ERRNO == ICONV_E2BIG)
3234 	{
3235 	    /* Allocate enough room for most conversions.  When re-allocating
3236 	     * increase the buffer size. */
3237 	    len = len + fromlen * 2 + 40;
3238 	    p = alloc((unsigned)len);
3239 	    if (p != NULL && done > 0)
3240 		mch_memmove(p, result, done);
3241 	    vim_free(result);
3242 	    result = p;
3243 	    if (result == NULL)	/* out of memory */
3244 		break;
3245 	}
3246 
3247 	to = (char *)result + done;
3248 	tolen = len - done - 2;
3249 	/* Avoid a warning for systems with a wrong iconv() prototype by
3250 	 * casting the second argument to void *. */
3251 	if (iconv(vcp->vc_fd, (void *)&from, &fromlen, &to, &tolen)
3252 								!= (size_t)-1)
3253 	{
3254 	    /* Finished, append a NUL. */
3255 	    *to = NUL;
3256 	    break;
3257 	}
3258 
3259 	/* Check both ICONV_EINVAL and EINVAL, because the dynamically loaded
3260 	 * iconv library may use one of them. */
3261 	if (!vcp->vc_fail && unconvlenp != NULL
3262 		&& (ICONV_ERRNO == ICONV_EINVAL || ICONV_ERRNO == EINVAL))
3263 	{
3264 	    /* Handle an incomplete sequence at the end. */
3265 	    *to = NUL;
3266 	    *unconvlenp = fromlen;
3267 	    break;
3268 	}
3269 
3270 	/* Check both ICONV_EILSEQ and EILSEQ, because the dynamically loaded
3271 	 * iconv library may use one of them. */
3272 	else if (!vcp->vc_fail
3273 		&& (ICONV_ERRNO == ICONV_EILSEQ || ICONV_ERRNO == EILSEQ
3274 		    || ICONV_ERRNO == ICONV_EINVAL || ICONV_ERRNO == EINVAL))
3275 	{
3276 	    /* Can't convert: insert a '?' and skip a character.  This assumes
3277 	     * conversion from 'encoding' to something else.  In other
3278 	     * situations we don't know what to skip anyway. */
3279 	    *to++ = '?';
3280 	    if ((*mb_ptr2cells)((char_u *)from) > 1)
3281 		*to++ = '?';
3282 	    if (enc_utf8)
3283 		l = utfc_ptr2len_len((char_u *)from, fromlen);
3284 	    else
3285 	    {
3286 		l = (*mb_ptr2len)((char_u *)from);
3287 		if (l > (int)fromlen)
3288 		    l = fromlen;
3289 	    }
3290 	    from += l;
3291 	    fromlen -= l;
3292 	}
3293 	else if (ICONV_ERRNO != ICONV_E2BIG)
3294 	{
3295 	    /* conversion failed */
3296 	    vim_free(result);
3297 	    result = NULL;
3298 	    break;
3299 	}
3300 	/* Not enough room or skipping illegal sequence. */
3301 	done = to - (char *)result;
3302     }
3303     return result;
3304 }
3305 
3306 #  if defined(DYNAMIC_ICONV) || defined(PROTO)
3307 /*
3308  * Dynamically load the "iconv.dll" on Win32.
3309  */
3310 
3311 #ifndef DYNAMIC_ICONV	    /* just generating prototypes */
3312 # define HINSTANCE int
3313 #endif
3314 static HINSTANCE hIconvDLL = 0;
3315 static HINSTANCE hMsvcrtDLL = 0;
3316 
3317 #  ifndef DYNAMIC_ICONV_DLL
3318 #   define DYNAMIC_ICONV_DLL "iconv.dll"
3319 #   define DYNAMIC_ICONV_DLL_ALT "libiconv.dll"
3320 #  endif
3321 #  ifndef DYNAMIC_MSVCRT_DLL
3322 #   define DYNAMIC_MSVCRT_DLL "msvcrt.dll"
3323 #  endif
3324 
3325 /*
3326  * Try opening the iconv.dll and return TRUE if iconv() can be used.
3327  */
3328     int
3329 iconv_enabled(verbose)
3330     int		verbose;
3331 {
3332     if (hIconvDLL != 0 && hMsvcrtDLL != 0)
3333 	return TRUE;
3334     hIconvDLL = LoadLibrary(DYNAMIC_ICONV_DLL);
3335     if (hIconvDLL == 0)		/* sometimes it's called libiconv.dll */
3336 	hIconvDLL = LoadLibrary(DYNAMIC_ICONV_DLL_ALT);
3337     if (hIconvDLL != 0)
3338 	hMsvcrtDLL = LoadLibrary(DYNAMIC_MSVCRT_DLL);
3339     if (hIconvDLL == 0 || hMsvcrtDLL == 0)
3340     {
3341 	/* Only give the message when 'verbose' is set, otherwise it might be
3342 	 * done whenever a conversion is attempted. */
3343 	if (verbose && p_verbose > 0)
3344 	{
3345 	    verbose_enter();
3346 	    EMSG2(_(e_loadlib),
3347 		    hIconvDLL == 0 ? DYNAMIC_ICONV_DLL : DYNAMIC_MSVCRT_DLL);
3348 	    verbose_leave();
3349 	}
3350 	iconv_end();
3351 	return FALSE;
3352     }
3353 
3354     iconv	= (void *)GetProcAddress(hIconvDLL, "libiconv");
3355     iconv_open	= (void *)GetProcAddress(hIconvDLL, "libiconv_open");
3356     iconv_close	= (void *)GetProcAddress(hIconvDLL, "libiconv_close");
3357     iconvctl	= (void *)GetProcAddress(hIconvDLL, "libiconvctl");
3358     iconv_errno	= (void *)GetProcAddress(hMsvcrtDLL, "_errno");
3359     if (iconv == NULL || iconv_open == NULL || iconv_close == NULL
3360 	    || iconvctl == NULL || iconv_errno == NULL)
3361     {
3362 	iconv_end();
3363 	if (verbose && p_verbose > 0)
3364 	{
3365 	    verbose_enter();
3366 	    EMSG2(_(e_loadfunc), "for libiconv");
3367 	    verbose_leave();
3368 	}
3369 	return FALSE;
3370     }
3371     return TRUE;
3372 }
3373 
3374     void
3375 iconv_end()
3376 {
3377     /* Don't use iconv() when inputting or outputting characters. */
3378     if (input_conv.vc_type == CONV_ICONV)
3379 	convert_setup(&input_conv, NULL, NULL);
3380     if (output_conv.vc_type == CONV_ICONV)
3381 	convert_setup(&output_conv, NULL, NULL);
3382 
3383     if (hIconvDLL != 0)
3384 	FreeLibrary(hIconvDLL);
3385     if (hMsvcrtDLL != 0)
3386 	FreeLibrary(hMsvcrtDLL);
3387     hIconvDLL = 0;
3388     hMsvcrtDLL = 0;
3389 }
3390 #  endif /* DYNAMIC_ICONV */
3391 # endif /* USE_ICONV */
3392 
3393 #endif /* FEAT_MBYTE */
3394 
3395 #if defined(FEAT_XIM) || defined(PROTO)
3396 
3397 # ifdef FEAT_GUI_GTK
3398 static int xim_has_preediting INIT(= FALSE);  /* IM current status */
3399 
3400 /*
3401  * Set preedit_start_col to the current cursor position.
3402  */
3403     static void
3404 init_preedit_start_col(void)
3405 {
3406     if (State & CMDLINE)
3407 	preedit_start_col = cmdline_getvcol_cursor();
3408     else if (curwin != NULL)
3409 	getvcol(curwin, &curwin->w_cursor, &preedit_start_col, NULL, NULL);
3410     /* Prevent that preediting marks the buffer as changed. */
3411     xim_changed_while_preediting = curbuf->b_changed;
3412 }
3413 # endif
3414 
3415 # if defined(HAVE_GTK2) && !defined(PROTO)
3416 
3417 static int im_is_active	       = FALSE;	/* IM is enabled for current mode    */
3418 static int im_preedit_cursor   = 0;	/* cursor offset in characters       */
3419 static int im_preedit_trailing = 0;	/* number of characters after cursor */
3420 
3421 static unsigned long im_commit_handler_id  = 0;
3422 static unsigned int  im_activatekey_keyval = GDK_VoidSymbol;
3423 static unsigned int  im_activatekey_state  = 0;
3424 
3425     void
3426 im_set_active(int active)
3427 {
3428     int was_active;
3429 
3430     was_active = !!im_is_active;
3431     im_is_active = (active && !p_imdisable);
3432 
3433     if (im_is_active != was_active)
3434 	xim_reset();
3435 }
3436 
3437     void
3438 xim_set_focus(int focus)
3439 {
3440     if (xic != NULL)
3441     {
3442 	if (focus)
3443 	    gtk_im_context_focus_in(xic);
3444 	else
3445 	    gtk_im_context_focus_out(xic);
3446     }
3447 }
3448 
3449     void
3450 im_set_position(int row, int col)
3451 {
3452     if (xic != NULL)
3453     {
3454 	GdkRectangle area;
3455 
3456 	area.x = FILL_X(col);
3457 	area.y = FILL_Y(row);
3458 	area.width  = gui.char_width * (mb_lefthalve(row, col) ? 2 : 1);
3459 	area.height = gui.char_height;
3460 
3461 	gtk_im_context_set_cursor_location(xic, &area);
3462     }
3463 }
3464 
3465 #  if 0 || defined(PROTO) /* apparently only used in gui_x11.c */
3466     void
3467 xim_set_preedit(void)
3468 {
3469     im_set_position(gui.row, gui.col);
3470 }
3471 #  endif
3472 
3473     static void
3474 im_add_to_input(char_u *str, int len)
3475 {
3476     /* Convert from 'termencoding' (always "utf-8") to 'encoding' */
3477     if (input_conv.vc_type != CONV_NONE)
3478     {
3479 	str = string_convert(&input_conv, str, &len);
3480 	g_return_if_fail(str != NULL);
3481     }
3482 
3483     add_to_input_buf_csi(str, len);
3484 
3485     if (input_conv.vc_type != CONV_NONE)
3486 	vim_free(str);
3487 
3488     if (p_mh) /* blank out the pointer if necessary */
3489 	gui_mch_mousehide(TRUE);
3490 }
3491 
3492     static void
3493 im_delete_preedit(void)
3494 {
3495     char_u bskey[]  = {CSI, 'k', 'b'};
3496     char_u delkey[] = {CSI, 'k', 'D'};
3497 
3498     if (State & NORMAL)
3499     {
3500 	im_preedit_cursor = 0;
3501 	return;
3502     }
3503     for (; im_preedit_cursor > 0; --im_preedit_cursor)
3504 	add_to_input_buf(bskey, (int)sizeof(bskey));
3505 
3506     for (; im_preedit_trailing > 0; --im_preedit_trailing)
3507 	add_to_input_buf(delkey, (int)sizeof(delkey));
3508 }
3509 
3510     static void
3511 im_correct_cursor(int num_move_back)
3512 {
3513     char_u backkey[] = {CSI, 'k', 'l'};
3514 
3515     if (State & NORMAL)
3516 	return;
3517 #  ifdef FEAT_RIGHTLEFT
3518     if ((State & CMDLINE) == 0 && curwin != NULL && curwin->w_p_rl)
3519 	backkey[2] = 'r';
3520 #  endif
3521     for (; num_move_back > 0; --num_move_back)
3522 	add_to_input_buf(backkey, (int)sizeof(backkey));
3523 }
3524 
3525 static int xim_expected_char = NUL;
3526 static int xim_ignored_char = FALSE;
3527 
3528 /*
3529  * Update the mode and cursor while in an IM callback.
3530  */
3531     static void
3532 im_show_info(void)
3533 {
3534     int	    old_vgetc_busy;
3535     old_vgetc_busy = vgetc_busy;
3536     vgetc_busy = TRUE;
3537     showmode();
3538     vgetc_busy = old_vgetc_busy;
3539     setcursor();
3540     out_flush();
3541 }
3542 
3543 /*
3544  * Callback invoked when the user finished preediting.
3545  * Put the final string into the input buffer.
3546  */
3547 /*ARGSUSED0*/
3548     static void
3549 im_commit_cb(GtkIMContext *context, const gchar *str, gpointer data)
3550 {
3551     int	slen = (int)STRLEN(str);
3552     int	add_to_input = TRUE;
3553     int	clen;
3554     int	len = slen;
3555     int	commit_with_preedit = TRUE;
3556     char_u	*im_str, *p;
3557 
3558 #ifdef XIM_DEBUG
3559     xim_log("im_commit_cb(): %s\n", str);
3560 #endif
3561 
3562     /* The imhangul module doesn't reset the preedit string before
3563      * committing.  Call im_delete_preedit() to work around that. */
3564     im_delete_preedit();
3565 
3566     /* Indicate that preediting has finished. */
3567     if (preedit_start_col == MAXCOL)
3568     {
3569 	init_preedit_start_col();
3570 	commit_with_preedit = FALSE;
3571     }
3572 
3573     /* The thing which setting "preedit_start_col" to MAXCOL means that
3574      * "preedit_start_col" will be set forcely when calling
3575      * preedit_changed_cb() next time.
3576      * "preedit_start_col" should not reset with MAXCOL on this part. Vim
3577      * is simulating the preediting by using add_to_input_str(). when
3578      * preedit begin immediately before committed, the typebuf is not
3579      * flushed to screen, then it can't get correct "preedit_start_col".
3580      * Thus, it should calculate the cells by adding cells of the committed
3581      * string. */
3582     if (input_conv.vc_type != CONV_NONE)
3583     {
3584 	im_str = string_convert(&input_conv, (char_u *)str, &len);
3585 	g_return_if_fail(im_str != NULL);
3586     }
3587     else
3588 	im_str = (char_u *)str;
3589     clen = 0;
3590     for (p = im_str; p < im_str + len; p += (*mb_ptr2len)(p))
3591 	clen += (*mb_ptr2cells)(p);
3592     if (input_conv.vc_type != CONV_NONE)
3593 	vim_free(im_str);
3594     preedit_start_col += clen;
3595 
3596     /* Is this a single character that matches a keypad key that's just
3597      * been pressed?  If so, we don't want it to be entered as such - let
3598      * us carry on processing the raw keycode so that it may be used in
3599      * mappings as <kSomething>. */
3600     if (xim_expected_char != NUL)
3601     {
3602 	/* We're currently processing a keypad or other special key */
3603 	if (slen == 1 && str[0] == xim_expected_char)
3604 	{
3605 	    /* It's a match - don't do it here */
3606 	    xim_ignored_char = TRUE;
3607 	    add_to_input = FALSE;
3608 	}
3609 	else
3610 	{
3611 	    /* Not a match */
3612 	    xim_ignored_char = FALSE;
3613 	}
3614     }
3615 
3616     if (add_to_input)
3617 	im_add_to_input((char_u *)str, slen);
3618 
3619     /* Inserting chars while "im_is_active" is set does not cause a change of
3620      * buffer.  When the chars are committed the buffer must be marked as
3621      * changed. */
3622     if (!commit_with_preedit)
3623 	preedit_start_col = MAXCOL;
3624 
3625     /* This flag is used in changed() at next call. */
3626     xim_changed_while_preediting = TRUE;
3627 
3628     if (gtk_main_level() > 0)
3629 	gtk_main_quit();
3630 }
3631 
3632 /*
3633  * Callback invoked after start to the preedit.
3634  */
3635 /*ARGSUSED*/
3636     static void
3637 im_preedit_start_cb(GtkIMContext *context, gpointer data)
3638 {
3639 #ifdef XIM_DEBUG
3640     xim_log("im_preedit_start_cb()\n");
3641 #endif
3642 
3643     im_is_active = TRUE;
3644     gui_update_cursor(TRUE, FALSE);
3645 }
3646 
3647 /*
3648  * Callback invoked after end to the preedit.
3649  */
3650 /*ARGSUSED*/
3651     static void
3652 im_preedit_end_cb(GtkIMContext *context, gpointer data)
3653 {
3654 #ifdef XIM_DEBUG
3655     xim_log("im_preedit_end_cb()\n");
3656 #endif
3657     im_delete_preedit();
3658 
3659     /* Indicate that preediting has finished */
3660     preedit_start_col = MAXCOL;
3661     xim_has_preediting = FALSE;
3662 
3663     im_is_active = FALSE;
3664     gui_update_cursor(TRUE, FALSE);
3665     im_show_info();
3666 }
3667 
3668 /*
3669  * Callback invoked after changes to the preedit string.  If the preedit
3670  * string was empty before, remember the preedit start column so we know
3671  * where to apply feedback attributes.  Delete the previous preedit string
3672  * if there was one, save the new preedit cursor offset, and put the new
3673  * string into the input buffer.
3674  *
3675  * TODO: The pragmatic "put into input buffer" approach used here has
3676  *       several fundamental problems:
3677  *
3678  * - The characters in the preedit string are subject to remapping.
3679  *   That's broken, only the finally committed string should be remapped.
3680  *
3681  * - There is a race condition involved:  The retrieved value for the
3682  *   current cursor position will be wrong if any unprocessed characters
3683  *   are still queued in the input buffer.
3684  *
3685  * - Due to the lack of synchronization between the file buffer in memory
3686  *   and any typed characters, it's practically impossible to implement the
3687  *   "retrieve_surrounding" and "delete_surrounding" signals reliably.  IM
3688  *   modules for languages such as Thai are likely to rely on this feature
3689  *   for proper operation.
3690  *
3691  * Conclusions:  I think support for preediting needs to be moved to the
3692  * core parts of Vim.  Ideally, until it has been committed, the preediting
3693  * string should only be displayed and not affect the buffer content at all.
3694  * The question how to deal with the synchronization issue still remains.
3695  * Circumventing the input buffer is probably not desirable.  Anyway, I think
3696  * implementing "retrieve_surrounding" is the only hard problem.
3697  *
3698  * One way to solve all of this in a clean manner would be to queue all key
3699  * press/release events "as is" in the input buffer, and apply the IM filtering
3700  * at the receiving end of the queue.  This, however, would have a rather large
3701  * impact on the code base.  If there is an easy way to force processing of all
3702  * remaining input from within the "retrieve_surrounding" signal handler, this
3703  * might not be necessary.  Gotta ask on vim-dev for opinions.
3704  */
3705 /*ARGSUSED1*/
3706     static void
3707 im_preedit_changed_cb(GtkIMContext *context, gpointer data)
3708 {
3709     char    *preedit_string = NULL;
3710     int	    cursor_index    = 0;
3711     int	    num_move_back   = 0;
3712     char_u  *str;
3713     char_u  *p;
3714     int	    i;
3715 
3716     gtk_im_context_get_preedit_string(context,
3717 				      &preedit_string, NULL,
3718 				      &cursor_index);
3719 
3720 #ifdef XIM_DEBUG
3721     xim_log("im_preedit_changed_cb(): %s\n", preedit_string);
3722 #endif
3723 
3724     g_return_if_fail(preedit_string != NULL); /* just in case */
3725 
3726     /* If preedit_start_col is MAXCOL set it to the current cursor position. */
3727     if (preedit_start_col == MAXCOL && preedit_string[0] != '\0')
3728     {
3729 	xim_has_preediting = TRUE;
3730 
3731 	/* Urgh, this breaks if the input buffer isn't empty now */
3732 	init_preedit_start_col();
3733     }
3734     else if (cursor_index == 0 && preedit_string[0] == '\0')
3735     {
3736 	if (preedit_start_col == MAXCOL)
3737 	    xim_has_preediting = FALSE;
3738 
3739 	/* If at the start position (after typing backspace)
3740 	 * preedit_start_col must be reset. */
3741 	preedit_start_col = MAXCOL;
3742     }
3743 
3744     im_delete_preedit();
3745 
3746     /*
3747      * Compute the end of the preediting area: "preedit_end_col".
3748      * According to the documentation of gtk_im_context_get_preedit_string(),
3749      * the cursor_pos output argument returns the offset in bytes.  This is
3750      * unfortunately not true -- real life shows the offset is in characters,
3751      * and the GTK+ source code agrees with me.  Will file a bug later.
3752      */
3753     if (preedit_start_col != MAXCOL)
3754 	preedit_end_col = preedit_start_col;
3755     str = (char_u *)preedit_string;
3756     for (p = str, i = 0; *p != NUL; p += utf_byte2len(*p), ++i)
3757     {
3758 	int is_composing;
3759 
3760 	is_composing = ((*p & 0x80) != 0 && utf_iscomposing(utf_ptr2char(p)));
3761 	/*
3762 	 * These offsets are used as counters when generating <BS> and <Del>
3763 	 * to delete the preedit string.  So don't count composing characters
3764 	 * unless 'delcombine' is enabled.
3765 	 */
3766 	if (!is_composing || p_deco)
3767 	{
3768 	    if (i < cursor_index)
3769 		++im_preedit_cursor;
3770 	    else
3771 		++im_preedit_trailing;
3772 	}
3773 	if (!is_composing && i >= cursor_index)
3774 	{
3775 	    /* This is essentially the same as im_preedit_trailing, except
3776 	     * composing characters are not counted even if p_deco is set. */
3777 	    ++num_move_back;
3778 	}
3779 	if (preedit_start_col != MAXCOL)
3780 	    preedit_end_col += utf_ptr2cells(p);
3781     }
3782 
3783     if (p > str)
3784     {
3785 	im_add_to_input(str, (int)(p - str));
3786 	im_correct_cursor(num_move_back);
3787     }
3788 
3789     g_free(preedit_string);
3790 
3791     if (gtk_main_level() > 0)
3792 	gtk_main_quit();
3793 }
3794 
3795 /*
3796  * Translate the Pango attributes at iter to Vim highlighting attributes.
3797  * Ignore attributes not supported by Vim highlighting.  This shouldn't have
3798  * too much impact -- right now we handle even more attributes than necessary
3799  * for the IM modules I tested with.
3800  */
3801     static int
3802 translate_pango_attributes(PangoAttrIterator *iter)
3803 {
3804     PangoAttribute  *attr;
3805     int		    char_attr = HL_NORMAL;
3806 
3807     attr = pango_attr_iterator_get(iter, PANGO_ATTR_UNDERLINE);
3808     if (attr != NULL && ((PangoAttrInt *)attr)->value
3809 						 != (int)PANGO_UNDERLINE_NONE)
3810 	char_attr |= HL_UNDERLINE;
3811 
3812     attr = pango_attr_iterator_get(iter, PANGO_ATTR_WEIGHT);
3813     if (attr != NULL && ((PangoAttrInt *)attr)->value >= (int)PANGO_WEIGHT_BOLD)
3814 	char_attr |= HL_BOLD;
3815 
3816     attr = pango_attr_iterator_get(iter, PANGO_ATTR_STYLE);
3817     if (attr != NULL && ((PangoAttrInt *)attr)->value
3818 						   != (int)PANGO_STYLE_NORMAL)
3819 	char_attr |= HL_ITALIC;
3820 
3821     attr = pango_attr_iterator_get(iter, PANGO_ATTR_BACKGROUND);
3822     if (attr != NULL)
3823     {
3824 	const PangoColor *color = &((PangoAttrColor *)attr)->color;
3825 
3826 	/* Assume inverse if black background is requested */
3827 	if ((color->red | color->green | color->blue) == 0)
3828 	    char_attr |= HL_INVERSE;
3829     }
3830 
3831     return char_attr;
3832 }
3833 
3834 /*
3835  * Retrieve the highlighting attributes at column col in the preedit string.
3836  * Return -1 if not in preediting mode or if col is out of range.
3837  */
3838     int
3839 im_get_feedback_attr(int col)
3840 {
3841     char	    *preedit_string = NULL;
3842     PangoAttrList   *attr_list	    = NULL;
3843     int		    char_attr	    = -1;
3844 
3845     if (xic == NULL)
3846 	return char_attr;
3847 
3848     gtk_im_context_get_preedit_string(xic, &preedit_string, &attr_list, NULL);
3849 
3850     if (preedit_string != NULL && attr_list != NULL)
3851     {
3852 	int index;
3853 
3854 	/* Get the byte index as used by PangoAttrIterator */
3855 	for (index = 0; col > 0 && preedit_string[index] != '\0'; --col)
3856 	    index += utfc_ptr2len((char_u *)preedit_string + index);
3857 
3858 	if (preedit_string[index] != '\0')
3859 	{
3860 	    PangoAttrIterator	*iter;
3861 	    int			start, end;
3862 
3863 	    char_attr = HL_NORMAL;
3864 	    iter = pango_attr_list_get_iterator(attr_list);
3865 
3866 	    /* Extract all relevant attributes from the list. */
3867 	    do
3868 	    {
3869 		pango_attr_iterator_range(iter, &start, &end);
3870 
3871 		if (index >= start && index < end)
3872 		    char_attr |= translate_pango_attributes(iter);
3873 	    }
3874 	    while (pango_attr_iterator_next(iter));
3875 
3876 	    pango_attr_iterator_destroy(iter);
3877 	}
3878     }
3879 
3880     if (attr_list != NULL)
3881 	pango_attr_list_unref(attr_list);
3882     g_free(preedit_string);
3883 
3884     return char_attr;
3885 }
3886 
3887     void
3888 xim_init(void)
3889 {
3890 #ifdef XIM_DEBUG
3891     xim_log("xim_init()\n");
3892 #endif
3893 
3894     g_return_if_fail(gui.drawarea != NULL);
3895     g_return_if_fail(gui.drawarea->window != NULL);
3896 
3897     xic = gtk_im_multicontext_new();
3898     g_object_ref(xic);
3899 
3900     im_commit_handler_id = g_signal_connect(G_OBJECT(xic), "commit",
3901 					    G_CALLBACK(&im_commit_cb), NULL);
3902     g_signal_connect(G_OBJECT(xic), "preedit_changed",
3903 		     G_CALLBACK(&im_preedit_changed_cb), NULL);
3904     g_signal_connect(G_OBJECT(xic), "preedit_start",
3905 		     G_CALLBACK(&im_preedit_start_cb), NULL);
3906     g_signal_connect(G_OBJECT(xic), "preedit_end",
3907 		     G_CALLBACK(&im_preedit_end_cb), NULL);
3908 
3909     gtk_im_context_set_client_window(xic, gui.drawarea->window);
3910 }
3911 
3912     void
3913 im_shutdown(void)
3914 {
3915 #ifdef XIM_DEBUG
3916     xim_log("im_shutdown()\n");
3917 #endif
3918 
3919     if (xic != NULL)
3920     {
3921 	gtk_im_context_focus_out(xic);
3922 	g_object_unref(xic);
3923 	xic = NULL;
3924     }
3925     im_is_active = FALSE;
3926     im_commit_handler_id = 0;
3927     preedit_start_col = MAXCOL;
3928     xim_has_preediting = FALSE;
3929 }
3930 
3931 /*
3932  * Convert the string argument to keyval and state for GdkEventKey.
3933  * If str is valid return TRUE, otherwise FALSE.
3934  *
3935  * See 'imactivatekey' for documentation of the format.
3936  */
3937     static int
3938 im_string_to_keyval(const char *str, unsigned int *keyval, unsigned int *state)
3939 {
3940     const char	    *mods_end;
3941     unsigned	    tmp_keyval;
3942     unsigned	    tmp_state = 0;
3943 
3944     mods_end = strrchr(str, '-');
3945     mods_end = (mods_end != NULL) ? mods_end + 1 : str;
3946 
3947     /* Parse modifier keys */
3948     while (str < mods_end)
3949 	switch (*str++)
3950 	{
3951 	    case '-':							break;
3952 	    case 'S': case 's': tmp_state |= (unsigned)GDK_SHIFT_MASK;	break;
3953 	    case 'L': case 'l': tmp_state |= (unsigned)GDK_LOCK_MASK;	break;
3954 	    case 'C': case 'c': tmp_state |= (unsigned)GDK_CONTROL_MASK;break;
3955 	    case '1':		tmp_state |= (unsigned)GDK_MOD1_MASK;	break;
3956 	    case '2':		tmp_state |= (unsigned)GDK_MOD2_MASK;	break;
3957 	    case '3':		tmp_state |= (unsigned)GDK_MOD3_MASK;	break;
3958 	    case '4':		tmp_state |= (unsigned)GDK_MOD4_MASK;	break;
3959 	    case '5':		tmp_state |= (unsigned)GDK_MOD5_MASK;	break;
3960 	    default:
3961 		return FALSE;
3962 	}
3963 
3964     tmp_keyval = gdk_keyval_from_name(str);
3965 
3966     if (tmp_keyval == 0 || tmp_keyval == GDK_VoidSymbol)
3967 	return FALSE;
3968 
3969     if (keyval != NULL)
3970 	*keyval = tmp_keyval;
3971     if (state != NULL)
3972 	*state = tmp_state;
3973 
3974     return TRUE;
3975 }
3976 
3977 /*
3978  * Return TRUE if p_imak is valid, otherwise FALSE.  As a special case, an
3979  * empty string is also regarded as valid.
3980  *
3981  * Note: The numerical key value of p_imak is cached if it was valid; thus
3982  * boldly assuming im_xim_isvalid_imactivate() will always be called whenever
3983  * 'imak' changes.  This is currently the case but not obvious -- should
3984  * probably rename the function for clarity.
3985  */
3986     int
3987 im_xim_isvalid_imactivate(void)
3988 {
3989     if (p_imak[0] == NUL)
3990     {
3991 	im_activatekey_keyval = GDK_VoidSymbol;
3992 	im_activatekey_state  = 0;
3993 	return TRUE;
3994     }
3995 
3996     return im_string_to_keyval((const char *)p_imak,
3997 			       &im_activatekey_keyval,
3998 			       &im_activatekey_state);
3999 }
4000 
4001     static void
4002 im_synthesize_keypress(unsigned int keyval, unsigned int state)
4003 {
4004     GdkEventKey *event;
4005 
4006 #  ifdef HAVE_GTK_MULTIHEAD
4007     event = (GdkEventKey *)gdk_event_new(GDK_KEY_PRESS);
4008     g_object_ref(gui.drawarea->window); /* unreffed by gdk_event_free() */
4009 #  else
4010     event = (GdkEventKey *)g_malloc0((gulong)sizeof(GdkEvent));
4011     event->type = GDK_KEY_PRESS;
4012 #  endif
4013     event->window = gui.drawarea->window;
4014     event->send_event = TRUE;
4015     event->time = GDK_CURRENT_TIME;
4016     event->state  = state;
4017     event->keyval = keyval;
4018     event->hardware_keycode = /* needed for XIM */
4019 	XKeysymToKeycode(GDK_WINDOW_XDISPLAY(event->window), (KeySym)keyval);
4020     event->length = 0;
4021     event->string = NULL;
4022 
4023     gtk_im_context_filter_keypress(xic, event);
4024 
4025     /* For consistency, also send the corresponding release event. */
4026     event->type = GDK_KEY_RELEASE;
4027     event->send_event = FALSE;
4028     gtk_im_context_filter_keypress(xic, event);
4029 
4030 #  ifdef HAVE_GTK_MULTIHEAD
4031     gdk_event_free((GdkEvent *)event);
4032 #  else
4033     g_free(event);
4034 #  endif
4035 }
4036 
4037     void
4038 xim_reset(void)
4039 {
4040     if (xic != NULL)
4041     {
4042 	/*
4043 	 * The third-party imhangul module (and maybe others too) ignores
4044 	 * gtk_im_context_reset() or at least doesn't reset the active state.
4045 	 * Thus sending imactivatekey would turn it off if it was on before,
4046 	 * which is clearly not what we want.  Fortunately we can work around
4047 	 * that for imhangul by sending GDK_Escape, but I don't know if it
4048 	 * works with all IM modules that support an activation key :/
4049 	 *
4050 	 * An alternative approach would be to destroy the IM context and
4051 	 * recreate it.  But that means loading/unloading the IM module on
4052 	 * every mode switch, which causes a quite noticable delay even on
4053 	 * my rather fast box...
4054 	 * *
4055 	 * Moreover, there are some XIM which cannot respond to
4056 	 * im_synthesize_keypress(). we hope that they reset by
4057 	 * xim_shutdown().
4058 	 */
4059 	if (im_activatekey_keyval != GDK_VoidSymbol && im_is_active)
4060 	    im_synthesize_keypress(GDK_Escape, 0U);
4061 
4062 	gtk_im_context_reset(xic);
4063 
4064 	/*
4065 	 * HACK for Ami: This sequence of function calls makes Ami handle
4066 	 * the IM reset gratiously, without breaking loads of other stuff.
4067 	 * It seems to force English mode as well, which is exactly what we
4068 	 * want because it makes the Ami status display work reliably.
4069 	 */
4070 	gtk_im_context_set_use_preedit(xic, FALSE);
4071 
4072 	if (p_imdisable)
4073 	    im_shutdown();
4074 	else
4075 	{
4076 	    gtk_im_context_set_use_preedit(xic, TRUE);
4077 	    xim_set_focus(gui.in_focus);
4078 
4079 	    if (im_activatekey_keyval != GDK_VoidSymbol)
4080 	    {
4081 		if (im_is_active)
4082 		{
4083 		    g_signal_handler_block(xic, im_commit_handler_id);
4084 		    im_synthesize_keypress(im_activatekey_keyval,
4085 						    im_activatekey_state);
4086 		    g_signal_handler_unblock(xic, im_commit_handler_id);
4087 		}
4088 	    }
4089 	    else
4090 	    {
4091 		im_shutdown();
4092 		xim_init();
4093 		xim_set_focus(gui.in_focus);
4094 	    }
4095 	}
4096     }
4097 
4098     preedit_start_col = MAXCOL;
4099     xim_has_preediting = FALSE;
4100 }
4101 
4102     int
4103 xim_queue_key_press_event(GdkEventKey *event, int down)
4104 {
4105     if (down)
4106     {
4107 	/*
4108 	 * Workaround GTK2 XIM 'feature' that always converts keypad keys to
4109 	 * chars., even when not part of an IM sequence (ref. feature of
4110 	 * gdk/gdkkeyuni.c).
4111 	 * Flag any keypad keys that might represent a single char.
4112 	 * If this (on its own - i.e., not part of an IM sequence) is
4113 	 * committed while we're processing one of these keys, we can ignore
4114 	 * that commit and go ahead & process it ourselves.  That way we can
4115 	 * still distinguish keypad keys for use in mappings.
4116 	 */
4117 	switch (event->keyval)
4118 	{
4119 	    case GDK_KP_Add:      xim_expected_char = '+';  break;
4120 	    case GDK_KP_Subtract: xim_expected_char = '-';  break;
4121 	    case GDK_KP_Divide:   xim_expected_char = '/';  break;
4122 	    case GDK_KP_Multiply: xim_expected_char = '*';  break;
4123 	    case GDK_KP_Decimal:  xim_expected_char = '.';  break;
4124 	    case GDK_KP_Equal:    xim_expected_char = '=';  break;
4125 	    case GDK_KP_0:	  xim_expected_char = '0';  break;
4126 	    case GDK_KP_1:	  xim_expected_char = '1';  break;
4127 	    case GDK_KP_2:	  xim_expected_char = '2';  break;
4128 	    case GDK_KP_3:	  xim_expected_char = '3';  break;
4129 	    case GDK_KP_4:	  xim_expected_char = '4';  break;
4130 	    case GDK_KP_5:	  xim_expected_char = '5';  break;
4131 	    case GDK_KP_6:	  xim_expected_char = '6';  break;
4132 	    case GDK_KP_7:	  xim_expected_char = '7';  break;
4133 	    case GDK_KP_8:	  xim_expected_char = '8';  break;
4134 	    case GDK_KP_9:	  xim_expected_char = '9';  break;
4135 	    default:		  xim_expected_char = NUL;
4136 	}
4137 	xim_ignored_char = FALSE;
4138     }
4139 
4140     /*
4141      * When typing fFtT, XIM may be activated. Thus it must pass
4142      * gtk_im_context_filter_keypress() in Normal mode.
4143      * And while doing :sh too.
4144      */
4145     if (xic != NULL && !p_imdisable
4146 		    && (State & (INSERT | CMDLINE | NORMAL | EXTERNCMD)) != 0)
4147     {
4148 	/*
4149 	 * Filter 'imactivatekey' and map it to CTRL-^.  This way, Vim is
4150 	 * always aware of the current status of IM, and can even emulate
4151 	 * the activation key for modules that don't support one.
4152 	 */
4153 	if (event->keyval == im_activatekey_keyval
4154 	     && (event->state & im_activatekey_state) == im_activatekey_state)
4155 	{
4156 	    unsigned int state_mask;
4157 
4158 	    /* Require the state of the 3 most used modifiers to match exactly.
4159 	     * Otherwise e.g. <S-C-space> would be unusable for other purposes
4160 	     * if the IM activate key is <S-space>. */
4161 	    state_mask  = im_activatekey_state;
4162 	    state_mask |= ((int)GDK_SHIFT_MASK | (int)GDK_CONTROL_MASK
4163 							| (int)GDK_MOD1_MASK);
4164 
4165 	    if ((event->state & state_mask) != im_activatekey_state)
4166 		return FALSE;
4167 
4168 	    /* Don't send it a second time on GDK_KEY_RELEASE. */
4169 	    if (event->type != GDK_KEY_PRESS)
4170 		return TRUE;
4171 
4172 	    if (map_to_exists_mode((char_u *)"", LANGMAP, FALSE))
4173 	    {
4174 		im_set_active(FALSE);
4175 
4176 		/* ":lmap" mappings exists, toggle use of mappings. */
4177 		State ^= LANGMAP;
4178 		if (State & LANGMAP)
4179 		{
4180 		    curbuf->b_p_iminsert = B_IMODE_NONE;
4181 		    State &= ~LANGMAP;
4182 		}
4183 		else
4184 		{
4185 		    curbuf->b_p_iminsert = B_IMODE_LMAP;
4186 		    State |= LANGMAP;
4187 		}
4188 		return TRUE;
4189 	    }
4190 
4191 	    return gtk_im_context_filter_keypress(xic, event);
4192 	}
4193 
4194 	/* Don't filter events through the IM context if IM isn't active
4195 	 * right now.  Unlike with GTK+ 1.2 we cannot rely on the IM module
4196 	 * not doing anything before the activation key was sent. */
4197 	if (im_activatekey_keyval == GDK_VoidSymbol || im_is_active)
4198 	{
4199 	    int imresult = gtk_im_context_filter_keypress(xic, event);
4200 
4201 	    /* Some XIM send following sequence:
4202 	     * 1. preedited string.
4203 	     * 2. committed string.
4204 	     * 3. line changed key.
4205 	     * 4. preedited string.
4206 	     * 5. remove preedited string.
4207 	     * if 3, Vim can't move back the above line for 5.
4208 	     * thus, this part should not parse the key. */
4209 	    if (!imresult && preedit_start_col != MAXCOL
4210 					       && event->keyval == GDK_Return)
4211 	    {
4212 		im_synthesize_keypress(GDK_Return, 0U);
4213 		return FALSE;
4214 	    }
4215 
4216 	    /* If XIM tried to commit a keypad key as a single char.,
4217 	     * ignore it so we can use the keypad key 'raw', for mappings. */
4218 	    if (xim_expected_char != NUL && xim_ignored_char)
4219 		/* We had a keypad key, and XIM tried to thieve it */
4220 		return FALSE;
4221 
4222 	    /* Normal processing */
4223 	    return imresult;
4224 	}
4225     }
4226 
4227     return FALSE;
4228 }
4229 
4230     int
4231 im_get_status(void)
4232 {
4233     return im_is_active;
4234 }
4235 
4236 # else /* !HAVE_GTK2 */
4237 
4238 static int	xim_is_active = FALSE;  /* XIM should be active in the current
4239 					   mode */
4240 static int	xim_has_focus = FALSE;	/* XIM is really being used for Vim */
4241 #ifdef FEAT_GUI_X11
4242 static XIMStyle	input_style;
4243 static int	status_area_enabled = TRUE;
4244 #endif
4245 
4246 #ifdef FEAT_GUI_GTK
4247 # ifdef WIN3264
4248 #  include <gdk/gdkwin32.h>
4249 # else
4250 #  include <gdk/gdkx.h>
4251 # endif
4252 #else
4253 # ifdef PROTO
4254 /* Define a few things to be able to generate prototypes while not configured
4255  * for GTK. */
4256 #  define GSList int
4257 #  define gboolean int
4258    typedef int GdkEvent;
4259    typedef int GdkEventKey;
4260 #  define GdkIC int
4261 # endif
4262 #endif
4263 
4264 #if defined(FEAT_GUI_GTK) || defined(PROTO)
4265 static int	preedit_buf_len = 0;
4266 static int	xim_can_preediting INIT(= FALSE);	/* XIM in showmode() */
4267 static int	xim_input_style;
4268 #ifndef FEAT_GUI_GTK
4269 # define gboolean int
4270 #endif
4271 static gboolean	use_status_area = 0;
4272 
4273 static int im_xim_str2keycode __ARGS((unsigned int *code, unsigned int *state));
4274 static void im_xim_send_event_imactivate __ARGS((void));
4275 
4276 /*
4277  * Convert string to keycode and state for XKeyEvent.
4278  * When string is valid return OK, when invalid return FAIL.
4279  *
4280  * See 'imactivatekey' documentation for the format.
4281  */
4282     static int
4283 im_xim_str2keycode(code, state)
4284     unsigned int *code;
4285     unsigned int *state;
4286 {
4287     int		retval = OK;
4288     int		len;
4289     unsigned	keycode = 0, keystate = 0;
4290     Window	window;
4291     Display	*display;
4292     char_u	*flag_end;
4293     char_u	*str;
4294 
4295     if (*p_imak != NUL)
4296     {
4297 	len = STRLEN(p_imak);
4298 	for (flag_end = p_imak + len - 1;
4299 			    flag_end > p_imak && *flag_end != '-'; --flag_end)
4300 	    ;
4301 
4302 	/* Parse modifier keys */
4303 	for (str = p_imak; str < flag_end; ++str)
4304 	{
4305 	    switch (*str)
4306 	    {
4307 		case 's': case 'S':
4308 		    keystate |= ShiftMask;
4309 		    break;
4310 		case 'l': case 'L':
4311 		    keystate |= LockMask;
4312 		    break;
4313 		case 'c': case 'C':
4314 		    keystate |= ControlMask;
4315 		    break;
4316 		case '1':
4317 		    keystate |= Mod1Mask;
4318 		    break;
4319 		case '2':
4320 		    keystate |= Mod2Mask;
4321 		    break;
4322 		case '3':
4323 		    keystate |= Mod3Mask;
4324 		    break;
4325 		case '4':
4326 		    keystate |= Mod4Mask;
4327 		    break;
4328 		case '5':
4329 		    keystate |= Mod5Mask;
4330 		    break;
4331 		case '-':
4332 		    break;
4333 		default:
4334 		    retval = FAIL;
4335 	    }
4336 	}
4337 	if (*str == '-')
4338 	    ++str;
4339 
4340 	/* Get keycode from string. */
4341 	gui_get_x11_windis(&window, &display);
4342 	if (display)
4343 	    keycode = XKeysymToKeycode(display, XStringToKeysym((char *)str));
4344 	if (keycode == 0)
4345 	    retval = FAIL;
4346 
4347 	if (code != NULL)
4348 	    *code = keycode;
4349 	if (state != NULL)
4350 	    *state = keystate;
4351     }
4352     return retval;
4353 }
4354 
4355     static void
4356 im_xim_send_event_imactivate()
4357 {
4358     /* Force turn on preedit state by symulate keypress event.
4359      * Keycode and state is specified by 'imactivatekey'.
4360      */
4361     XKeyEvent ev;
4362 
4363     gui_get_x11_windis(&ev.window, &ev.display);
4364     ev.root = RootWindow(ev.display, DefaultScreen(ev.display));
4365     ev.subwindow = None;
4366     ev.time = CurrentTime;
4367     ev.x = 1;
4368     ev.y = 1;
4369     ev.x_root = 1;
4370     ev.y_root = 1;
4371     ev.same_screen = 1;
4372     ev.type = KeyPress;
4373     if (im_xim_str2keycode(&ev.keycode, &ev.state) == OK)
4374 	XSendEvent(ev.display, ev.window, 1, KeyPressMask, (XEvent*)&ev);
4375 }
4376 
4377 /*
4378  * Return TRUE if 'imactivatekey' has a valid value.
4379  */
4380     int
4381 im_xim_isvalid_imactivate()
4382 {
4383     return im_xim_str2keycode(NULL, NULL) == OK;
4384 }
4385 #endif /* FEAT_GUI_GTK */
4386 
4387 /*
4388  * Switch using XIM on/off.  This is used by the code that changes "State".
4389  */
4390     void
4391 im_set_active(active)
4392     int		active;
4393 {
4394     if (xic == NULL)
4395 	return;
4396 
4397     /* If 'imdisable' is set, XIM is never active. */
4398     if (p_imdisable)
4399 	active = FALSE;
4400 #if !defined (FEAT_GUI_GTK)
4401     else if (input_style & XIMPreeditPosition)
4402 	/* There is a problem in switching XIM off when preediting is used,
4403 	 * and it is not clear how this can be solved.  For now, keep XIM on
4404 	 * all the time, like it was done in Vim 5.8. */
4405 	active = TRUE;
4406 #endif
4407 
4408     /* Remember the active state, it is needed when Vim gets keyboard focus. */
4409     xim_is_active = active;
4410 
4411 #ifdef FEAT_GUI_GTK
4412     /* When 'imactivatekey' has valid key-string, try to control XIM preedit
4413      * state.  When 'imactivatekey' has no or invalid string, try old XIM
4414      * focus control.
4415      */
4416     if (*p_imak != NUL)
4417     {
4418 	/* BASIC STRATEGY:
4419 	 * Destroy old Input Context (XIC), and create new one.  New XIC
4420 	 * would have a state of preedit that is off.  When argument:active
4421 	 * is false, that's all.  Else argument:active is true, send a key
4422 	 * event specified by 'imactivatekey' to activate XIM preedit state.
4423 	 */
4424 
4425 	xim_is_active = TRUE; /* Disable old XIM focus control */
4426 	/* If we can monitor preedit state with preedit callback functions,
4427 	 * try least creation of new XIC.
4428 	 */
4429 	if (xim_input_style & (int)GDK_IM_PREEDIT_CALLBACKS)
4430 	{
4431 	    if (xim_can_preediting && !active)
4432 	    {
4433 		/* Force turn off preedit state.  With some IM
4434 		 * implementations, we cannot turn off preedit state by
4435 		 * symulate keypress event.  It is why using such a method
4436 		 * that destroy old IC (input context), and create new one.
4437 		 * When create new IC, its preedit state is usually off.
4438 		 */
4439 		xim_reset();
4440 		xim_set_focus(FALSE);
4441 		gdk_ic_destroy(xic);
4442 		xim_init();
4443 		xim_can_preediting = FALSE;
4444 	    }
4445 	    else if (!xim_can_preediting && active)
4446 		im_xim_send_event_imactivate();
4447 	}
4448 	else
4449 	{
4450 	    /* First, force destroy old IC, and create new one.  It
4451 	     * symulates "turning off preedit state".
4452 	     */
4453 	    xim_set_focus(FALSE);
4454 	    gdk_ic_destroy(xic);
4455 	    xim_init();
4456 	    xim_can_preediting = FALSE;
4457 
4458 	    /* 2nd, when requested to activate IM, symulate this by sending
4459 	     * the event.
4460 	     */
4461 	    if (active)
4462 	    {
4463 		im_xim_send_event_imactivate();
4464 		xim_can_preediting = TRUE;
4465 	    }
4466 	}
4467     }
4468     else
4469     {
4470 # ifndef XIMPreeditUnKnown
4471 	/* X11R5 doesn't have these, it looks safe enough to define here. */
4472 	typedef unsigned long XIMPreeditState;
4473 #  define XIMPreeditUnKnown	0L
4474 #  define XIMPreeditEnable	1L
4475 #  define XIMPreeditDisable	(1L<<1)
4476 #  define XNPreeditState	"preeditState"
4477 # endif
4478 	XIMPreeditState preedit_state = XIMPreeditUnKnown;
4479 	XVaNestedList preedit_attr;
4480 	XIC pxic;
4481 
4482 	preedit_attr = XVaCreateNestedList(0,
4483 				XNPreeditState, &preedit_state,
4484 				NULL);
4485 	pxic = ((GdkICPrivate *)xic)->xic;
4486 
4487 	if (!XGetICValues(pxic, XNPreeditAttributes, preedit_attr, NULL))
4488 	{
4489 	    XFree(preedit_attr);
4490 	    preedit_attr = XVaCreateNestedList(0,
4491 				XNPreeditState,
4492 				active ? XIMPreeditEnable : XIMPreeditDisable,
4493 				NULL);
4494 	    XSetICValues(pxic, XNPreeditAttributes, preedit_attr, NULL);
4495 	    xim_can_preediting = active;
4496 	    xim_is_active = active;
4497 	}
4498 	XFree(preedit_attr);
4499     }
4500     if (xim_input_style & XIMPreeditCallbacks)
4501     {
4502 	preedit_buf_len = 0;
4503 	init_preedit_start_col();
4504     }
4505 #else
4506 # if 0
4507 	/* When had tested kinput2 + canna + Athena GUI version with
4508 	 * 'imactivatekey' is "s-space", im_xim_send_event_imactivate() did not
4509 	 * work correctly.  It just inserted one space.  I don't know why we
4510 	 * couldn't switch state of XIM preediting.  This is reason why these
4511 	 * codes are commented out.
4512 	 */
4513 	/* First, force destroy old IC, and create new one.  It symulates
4514 	 * "turning off preedit state".
4515 	 */
4516 	xim_set_focus(FALSE);
4517 	XDestroyIC(xic);
4518 	xic = NULL;
4519 	xim_init();
4520 
4521 	/* 2nd, when requested to activate IM, symulate this by sending the
4522 	 * event.
4523 	 */
4524 	if (active)
4525 	    im_xim_send_event_imactivate();
4526 # endif
4527 #endif
4528     xim_set_preedit();
4529 }
4530 
4531 /*
4532  * Adjust using XIM for gaining or losing keyboard focus.  Also called when
4533  * "xim_is_active" changes.
4534  */
4535     void
4536 xim_set_focus(focus)
4537     int		focus;
4538 {
4539     if (xic == NULL)
4540 	return;
4541 
4542     /*
4543      * XIM only gets focus when the Vim window has keyboard focus and XIM has
4544      * been set active for the current mode.
4545      */
4546     if (focus && xim_is_active)
4547     {
4548 	if (!xim_has_focus)
4549 	{
4550 	    xim_has_focus = TRUE;
4551 #ifdef FEAT_GUI_GTK
4552 	    gdk_im_begin(xic, gui.drawarea->window);
4553 #else
4554 	    XSetICFocus(xic);
4555 #endif
4556 	}
4557     }
4558     else
4559     {
4560 	if (xim_has_focus)
4561 	{
4562 	    xim_has_focus = FALSE;
4563 #ifdef FEAT_GUI_GTK
4564 	    gdk_im_end();
4565 #else
4566 	    XUnsetICFocus(xic);
4567 #endif
4568 	}
4569     }
4570 }
4571 
4572 /*ARGSUSED*/
4573     void
4574 im_set_position(row, col)
4575     int		row;
4576     int		col;
4577 {
4578     xim_set_preedit();
4579 }
4580 
4581 /*
4582  * Set the XIM to the current cursor position.
4583  */
4584     void
4585 xim_set_preedit()
4586 {
4587     if (xic == NULL)
4588 	return;
4589 
4590     xim_set_focus(TRUE);
4591 
4592 #ifdef FEAT_GUI_GTK
4593     if (gdk_im_ready())
4594     {
4595 	int		attrmask;
4596 	GdkICAttr	*attr;
4597 
4598 	if (!xic_attr)
4599 	    return;
4600 
4601 	attr = xic_attr;
4602 	attrmask = 0;
4603 
4604 # ifdef FEAT_XFONTSET
4605 	if ((xim_input_style & (int)GDK_IM_PREEDIT_POSITION)
4606 		&& gui.fontset != NOFONTSET
4607 		&& gui.fontset->type == GDK_FONT_FONTSET)
4608 	{
4609 	    if (!xim_has_focus)
4610 	    {
4611 		if (attr->spot_location.y >= 0)
4612 		{
4613 		    attr->spot_location.x = 0;
4614 		    attr->spot_location.y = -100;
4615 		    attrmask |= (int)GDK_IC_SPOT_LOCATION;
4616 		}
4617 	    }
4618 	    else
4619 	    {
4620 		gint	width, height;
4621 
4622 		if (attr->spot_location.x != TEXT_X(gui.col)
4623 		    || attr->spot_location.y != TEXT_Y(gui.row))
4624 		{
4625 		    attr->spot_location.x = TEXT_X(gui.col);
4626 		    attr->spot_location.y = TEXT_Y(gui.row);
4627 		    attrmask |= (int)GDK_IC_SPOT_LOCATION;
4628 		}
4629 
4630 		gdk_window_get_size(gui.drawarea->window, &width, &height);
4631 		width -= 2 * gui.border_offset;
4632 		height -= 2 * gui.border_offset;
4633 		if (xim_input_style & (int)GDK_IM_STATUS_AREA)
4634 		    height -= gui.char_height;
4635 		if (attr->preedit_area.width != width
4636 		    || attr->preedit_area.height != height)
4637 		{
4638 		    attr->preedit_area.x = gui.border_offset;
4639 		    attr->preedit_area.y = gui.border_offset;
4640 		    attr->preedit_area.width = width;
4641 		    attr->preedit_area.height = height;
4642 		    attrmask |= (int)GDK_IC_PREEDIT_AREA;
4643 		}
4644 
4645 		if (attr->preedit_fontset != gui.current_font)
4646 		{
4647 		    attr->preedit_fontset = gui.current_font;
4648 		    attrmask |= (int)GDK_IC_PREEDIT_FONTSET;
4649 		}
4650 	    }
4651 	}
4652 # endif /* FEAT_XFONTSET */
4653 
4654 	if (xim_fg_color == INVALCOLOR)
4655 	{
4656 	    xim_fg_color = gui.def_norm_pixel;
4657 	    xim_bg_color = gui.def_back_pixel;
4658 	}
4659 	if (attr->preedit_foreground.pixel != xim_fg_color)
4660 	{
4661 	    attr->preedit_foreground.pixel = xim_fg_color;
4662 	    attrmask |= (int)GDK_IC_PREEDIT_FOREGROUND;
4663 	}
4664 	if (attr->preedit_background.pixel != xim_bg_color)
4665 	{
4666 	    attr->preedit_background.pixel = xim_bg_color;
4667 	    attrmask |= (int)GDK_IC_PREEDIT_BACKGROUND;
4668 	}
4669 
4670 	if (attrmask != 0)
4671 	    gdk_ic_set_attr(xic, attr, (GdkICAttributesType)attrmask);
4672     }
4673 #else /* FEAT_GUI_GTK */
4674     {
4675 	XVaNestedList attr_list;
4676 	XRectangle spot_area;
4677 	XPoint over_spot;
4678 	int line_space;
4679 
4680 	if (!xim_has_focus)
4681 	{
4682 	    /* hide XIM cursor */
4683 	    over_spot.x = 0;
4684 	    over_spot.y = -100; /* arbitrary invisible position */
4685 	    attr_list = (XVaNestedList) XVaCreateNestedList(0,
4686 							    XNSpotLocation,
4687 							    &over_spot,
4688 							    NULL);
4689 	    XSetICValues(xic, XNPreeditAttributes, attr_list, NULL);
4690 	    XFree(attr_list);
4691 	    return;
4692 	}
4693 
4694 	if (input_style & XIMPreeditPosition)
4695 	{
4696 	    if (xim_fg_color == INVALCOLOR)
4697 	    {
4698 		xim_fg_color = gui.def_norm_pixel;
4699 		xim_bg_color = gui.def_back_pixel;
4700 	    }
4701 	    over_spot.x = TEXT_X(gui.col);
4702 	    over_spot.y = TEXT_Y(gui.row);
4703 	    spot_area.x = 0;
4704 	    spot_area.y = 0;
4705 	    spot_area.height = gui.char_height * Rows;
4706 	    spot_area.width  = gui.char_width * Columns;
4707 	    line_space = gui.char_height;
4708 	    attr_list = (XVaNestedList) XVaCreateNestedList(0,
4709 					    XNSpotLocation, &over_spot,
4710 					    XNForeground, (Pixel) xim_fg_color,
4711 					    XNBackground, (Pixel) xim_bg_color,
4712 					    XNArea, &spot_area,
4713 					    XNLineSpace, line_space,
4714 					    NULL);
4715 	    if (XSetICValues(xic, XNPreeditAttributes, attr_list, NULL))
4716 		EMSG(_("E284: Cannot set IC values"));
4717 	    XFree(attr_list);
4718 	}
4719     }
4720 #endif /* FEAT_GUI_GTK */
4721 }
4722 
4723 /*
4724  * Set up the status area.
4725  *
4726  * This should use a separate Widget, but that seems not possible, because
4727  * preedit_area and status_area should be set to the same window as for the
4728  * text input.  Unfortunately this means the status area pollutes the text
4729  * window...
4730  */
4731     void
4732 xim_set_status_area()
4733 {
4734     if (xic == NULL)
4735 	return;
4736 
4737 #ifdef FEAT_GUI_GTK
4738 # if defined(FEAT_XFONTSET)
4739     if (use_status_area)
4740     {
4741 	GdkICAttr	*attr;
4742 	int		style;
4743 	gint		width, height;
4744 	GtkWidget	*widget;
4745 	int		attrmask;
4746 
4747 	if (!xic_attr)
4748 	    return;
4749 
4750 	attr = xic_attr;
4751 	attrmask = 0;
4752 	style = (int)gdk_ic_get_style(xic);
4753 	if ((style & (int)GDK_IM_STATUS_MASK) == (int)GDK_IM_STATUS_AREA)
4754 	{
4755 	    if (gui.fontset != NOFONTSET
4756 		    && gui.fontset->type == GDK_FONT_FONTSET)
4757 	    {
4758 		widget = gui.mainwin;
4759 		gdk_window_get_size(widget->window, &width, &height);
4760 
4761 		attrmask |= (int)GDK_IC_STATUS_AREA;
4762 		attr->status_area.x = 0;
4763 		attr->status_area.y = height - gui.char_height - 1;
4764 		attr->status_area.width = width;
4765 		attr->status_area.height = gui.char_height;
4766 	    }
4767 	}
4768 	if (attrmask != 0)
4769 	    gdk_ic_set_attr(xic, attr, (GdkICAttributesType)attrmask);
4770     }
4771 # endif
4772 #else
4773     {
4774 	XVaNestedList preedit_list = 0, status_list = 0, list = 0;
4775 	XRectangle pre_area, status_area;
4776 
4777 	if (input_style & XIMStatusArea)
4778 	{
4779 	    if (input_style & XIMPreeditArea)
4780 	    {
4781 		XRectangle *needed_rect;
4782 
4783 		/* to get status_area width */
4784 		status_list = XVaCreateNestedList(0, XNAreaNeeded,
4785 						  &needed_rect, NULL);
4786 		XGetICValues(xic, XNStatusAttributes, status_list, NULL);
4787 		XFree(status_list);
4788 
4789 		status_area.width = needed_rect->width;
4790 	    }
4791 	    else
4792 		status_area.width = gui.char_width * Columns;
4793 
4794 	    status_area.x = 0;
4795 	    status_area.y = gui.char_height * Rows + gui.border_offset;
4796 	    if (gui.which_scrollbars[SBAR_BOTTOM])
4797 		status_area.y += gui.scrollbar_height;
4798 #ifdef FEAT_MENU
4799 	    if (gui.menu_is_active)
4800 		status_area.y += gui.menu_height;
4801 #endif
4802 	    status_area.height = gui.char_height;
4803 	    status_list = XVaCreateNestedList(0, XNArea, &status_area, NULL);
4804 	}
4805 	else
4806 	{
4807 	    status_area.x = 0;
4808 	    status_area.y = gui.char_height * Rows + gui.border_offset;
4809 	    if (gui.which_scrollbars[SBAR_BOTTOM])
4810 		status_area.y += gui.scrollbar_height;
4811 #ifdef FEAT_MENU
4812 	    if (gui.menu_is_active)
4813 		status_area.y += gui.menu_height;
4814 #endif
4815 	    status_area.width = 0;
4816 	    status_area.height = gui.char_height;
4817 	}
4818 
4819 	if (input_style & XIMPreeditArea)   /* off-the-spot */
4820 	{
4821 	    pre_area.x = status_area.x + status_area.width;
4822 	    pre_area.y = gui.char_height * Rows + gui.border_offset;
4823 	    pre_area.width = gui.char_width * Columns - pre_area.x;
4824 	    if (gui.which_scrollbars[SBAR_BOTTOM])
4825 		pre_area.y += gui.scrollbar_height;
4826 #ifdef FEAT_MENU
4827 	    if (gui.menu_is_active)
4828 		pre_area.y += gui.menu_height;
4829 #endif
4830 	    pre_area.height = gui.char_height;
4831 	    preedit_list = XVaCreateNestedList(0, XNArea, &pre_area, NULL);
4832 	}
4833 	else if (input_style & XIMPreeditPosition)   /* over-the-spot */
4834 	{
4835 	    pre_area.x = 0;
4836 	    pre_area.y = 0;
4837 	    pre_area.height = gui.char_height * Rows;
4838 	    pre_area.width = gui.char_width * Columns;
4839 	    preedit_list = XVaCreateNestedList(0, XNArea, &pre_area, NULL);
4840 	}
4841 
4842 	if (preedit_list && status_list)
4843 	    list = XVaCreateNestedList(0, XNPreeditAttributes, preedit_list,
4844 				       XNStatusAttributes, status_list, NULL);
4845 	else if (preedit_list)
4846 	    list = XVaCreateNestedList(0, XNPreeditAttributes, preedit_list,
4847 				       NULL);
4848 	else if (status_list)
4849 	    list = XVaCreateNestedList(0, XNStatusAttributes, status_list,
4850 				       NULL);
4851 	else
4852 	    list = NULL;
4853 
4854 	if (list)
4855 	{
4856 	    XSetICValues(xic, XNVaNestedList, list, NULL);
4857 	    XFree(list);
4858 	}
4859 	if (status_list)
4860 	    XFree(status_list);
4861 	if (preedit_list)
4862 	    XFree(preedit_list);
4863     }
4864 #endif
4865 }
4866 
4867 #if defined(FEAT_GUI_X11) || defined(FEAT_GUI_GTK)
4868 static char e_xim[] = N_("E285: Failed to create input context");
4869 #endif
4870 
4871 #if defined(FEAT_GUI_X11) || defined(PROTO)
4872 # if defined(XtSpecificationRelease) && XtSpecificationRelease >= 6 && !defined(sun)
4873 #  define USE_X11R6_XIM
4874 # endif
4875 
4876 static int xim_real_init __ARGS((Window x11_window, Display *x11_display));
4877 
4878 
4879 #ifdef USE_X11R6_XIM
4880 static void xim_instantiate_cb __ARGS((Display *display, XPointer client_data, XPointer	call_data));
4881 static void xim_destroy_cb __ARGS((XIM im, XPointer client_data, XPointer call_data));
4882 
4883 /*ARGSUSED*/
4884     static void
4885 xim_instantiate_cb(display, client_data, call_data)
4886     Display	*display;
4887     XPointer	client_data;
4888     XPointer	call_data;
4889 {
4890     Window	x11_window;
4891     Display	*x11_display;
4892 
4893 #ifdef XIM_DEBUG
4894     xim_log("xim_instantiate_cb()\n");
4895 #endif
4896 
4897     gui_get_x11_windis(&x11_window, &x11_display);
4898     if (display != x11_display)
4899 	return;
4900 
4901     xim_real_init(x11_window, x11_display);
4902     gui_set_shellsize(FALSE, FALSE);
4903     if (xic != NULL)
4904 	XUnregisterIMInstantiateCallback(x11_display, NULL, NULL, NULL,
4905 					 xim_instantiate_cb, NULL);
4906 }
4907 
4908 /*ARGSUSED*/
4909     static void
4910 xim_destroy_cb(im, client_data, call_data)
4911     XIM		im;
4912     XPointer	client_data;
4913     XPointer	call_data;
4914 {
4915     Window	x11_window;
4916     Display	*x11_display;
4917 
4918 #ifdef XIM_DEBUG
4919     xim_log("xim_destroy_cb()\n");
4920 #endif
4921     gui_get_x11_windis(&x11_window, &x11_display);
4922 
4923     xic = NULL;
4924     status_area_enabled = FALSE;
4925 
4926     gui_set_shellsize(FALSE, FALSE);
4927 
4928     XRegisterIMInstantiateCallback(x11_display, NULL, NULL, NULL,
4929 				   xim_instantiate_cb, NULL);
4930 }
4931 #endif
4932 
4933     void
4934 xim_init()
4935 {
4936     Window	x11_window;
4937     Display	*x11_display;
4938 
4939 #ifdef XIM_DEBUG
4940     xim_log("xim_init()\n");
4941 #endif
4942 
4943     gui_get_x11_windis(&x11_window, &x11_display);
4944 
4945     xic = NULL;
4946 
4947     if (xim_real_init(x11_window, x11_display))
4948 	return;
4949 
4950     gui_set_shellsize(FALSE, FALSE);
4951 
4952 #ifdef USE_X11R6_XIM
4953     XRegisterIMInstantiateCallback(x11_display, NULL, NULL, NULL,
4954 				   xim_instantiate_cb, NULL);
4955 #endif
4956 }
4957 
4958     static int
4959 xim_real_init(x11_window, x11_display)
4960     Window	x11_window;
4961     Display	*x11_display;
4962 {
4963     int		i;
4964     char	*p,
4965 		*s,
4966 		*ns,
4967 		*end,
4968 		tmp[1024];
4969 #define IMLEN_MAX 40
4970     char	buf[IMLEN_MAX + 7];
4971     XIM		xim = NULL;
4972     XIMStyles	*xim_styles;
4973     XIMStyle	this_input_style = 0;
4974     Boolean	found;
4975     XPoint	over_spot;
4976     XVaNestedList preedit_list, status_list;
4977 
4978     input_style = 0;
4979     status_area_enabled = FALSE;
4980 
4981     if (xic != NULL)
4982 	return FALSE;
4983 
4984     if (gui.rsrc_input_method != NULL && *gui.rsrc_input_method != NUL)
4985     {
4986 	strcpy(tmp, gui.rsrc_input_method);
4987 	for (ns = s = tmp; ns != NULL && *s != NUL;)
4988 	{
4989 	    s = (char *)skipwhite((char_u *)s);
4990 	    if (*s == NUL)
4991 		break;
4992 	    if ((ns = end = strchr(s, ',')) == NULL)
4993 		end = s + strlen(s);
4994 	    while (isspace(((char_u *)end)[-1]))
4995 		end--;
4996 	    *end = NUL;
4997 
4998 	    if (strlen(s) <= IMLEN_MAX)
4999 	    {
5000 		strcpy(buf, "@im=");
5001 		strcat(buf, s);
5002 		if ((p = XSetLocaleModifiers(buf)) != NULL && *p != NUL
5003 			&& (xim = XOpenIM(x11_display, NULL, NULL, NULL))
5004 								      != NULL)
5005 		    break;
5006 	    }
5007 
5008 	    s = ns + 1;
5009 	}
5010     }
5011 
5012     if (xim == NULL && (p = XSetLocaleModifiers("")) != NULL && *p != NUL)
5013 	xim = XOpenIM(x11_display, NULL, NULL, NULL);
5014 
5015     /* This is supposed to be useful to obtain characters through
5016      * XmbLookupString() without really using a XIM. */
5017     if (xim == NULL && (p = XSetLocaleModifiers("@im=none")) != NULL
5018 								 && *p != NUL)
5019 	xim = XOpenIM(x11_display, NULL, NULL, NULL);
5020 
5021     if (xim == NULL)
5022     {
5023 	/* Only give this message when verbose is set, because too many people
5024 	 * got this message when they didn't want to use a XIM. */
5025 	if (p_verbose > 0)
5026 	{
5027 	    verbose_enter();
5028 	    EMSG(_("E286: Failed to open input method"));
5029 	    verbose_leave();
5030 	}
5031 	return FALSE;
5032     }
5033 
5034 #ifdef USE_X11R6_XIM
5035     {
5036 	XIMCallback destroy_cb;
5037 
5038 	destroy_cb.callback = xim_destroy_cb;
5039 	destroy_cb.client_data = NULL;
5040 	if (XSetIMValues(xim, XNDestroyCallback, &destroy_cb, NULL))
5041 	    EMSG(_("E287: Warning: Could not set destroy callback to IM"));
5042     }
5043 #endif
5044 
5045     if (XGetIMValues(xim, XNQueryInputStyle, &xim_styles, NULL) || !xim_styles)
5046     {
5047 	EMSG(_("E288: input method doesn't support any style"));
5048 	XCloseIM(xim);
5049 	return FALSE;
5050     }
5051 
5052     found = False;
5053     strcpy(tmp, gui.rsrc_preedit_type_name);
5054     for (s = tmp; s && !found; )
5055     {
5056 	while (*s && isspace((unsigned char)*s))
5057 	    s++;
5058 	if (!*s)
5059 	    break;
5060 	if ((ns = end = strchr(s, ',')) != 0)
5061 	    ns++;
5062 	else
5063 	    end = s + strlen(s);
5064 	while (isspace((unsigned char)*end))
5065 	    end--;
5066 	*end = '\0';
5067 
5068 	if (!strcmp(s, "OverTheSpot"))
5069 	    this_input_style = (XIMPreeditPosition | XIMStatusArea);
5070 	else if (!strcmp(s, "OffTheSpot"))
5071 	    this_input_style = (XIMPreeditArea | XIMStatusArea);
5072 	else if (!strcmp(s, "Root"))
5073 	    this_input_style = (XIMPreeditNothing | XIMStatusNothing);
5074 
5075 	for (i = 0; (unsigned short)i < xim_styles->count_styles; i++)
5076 	{
5077 	    if (this_input_style == xim_styles->supported_styles[i])
5078 	    {
5079 		found = True;
5080 		break;
5081 	    }
5082 	}
5083 	if (!found)
5084 	    for (i = 0; (unsigned short)i < xim_styles->count_styles; i++)
5085 	    {
5086 		if ((xim_styles->supported_styles[i] & this_input_style)
5087 			== (this_input_style & ~XIMStatusArea))
5088 		{
5089 		    this_input_style &= ~XIMStatusArea;
5090 		    found = True;
5091 		    break;
5092 		}
5093 	    }
5094 
5095 	s = ns;
5096     }
5097     XFree(xim_styles);
5098 
5099     if (!found)
5100     {
5101 	/* Only give this message when verbose is set, because too many people
5102 	 * got this message when they didn't want to use a XIM. */
5103 	if (p_verbose > 0)
5104 	{
5105 	    verbose_enter();
5106 	    EMSG(_("E289: input method doesn't support my preedit type"));
5107 	    verbose_leave();
5108 	}
5109 	XCloseIM(xim);
5110 	return FALSE;
5111     }
5112 
5113     over_spot.x = TEXT_X(gui.col);
5114     over_spot.y = TEXT_Y(gui.row);
5115     input_style = this_input_style;
5116 
5117     /* A crash was reported when trying to pass gui.norm_font as XNFontSet,
5118      * thus that has been removed.  Hopefully the default works... */
5119 #ifdef FEAT_XFONTSET
5120     if (gui.fontset != NOFONTSET)
5121     {
5122 	preedit_list = XVaCreateNestedList(0,
5123 				XNSpotLocation, &over_spot,
5124 				XNForeground, (Pixel)gui.def_norm_pixel,
5125 				XNBackground, (Pixel)gui.def_back_pixel,
5126 				XNFontSet, (XFontSet)gui.fontset,
5127 				NULL);
5128 	status_list = XVaCreateNestedList(0,
5129 				XNForeground, (Pixel)gui.def_norm_pixel,
5130 				XNBackground, (Pixel)gui.def_back_pixel,
5131 				XNFontSet, (XFontSet)gui.fontset,
5132 				NULL);
5133     }
5134     else
5135 #endif
5136     {
5137 	preedit_list = XVaCreateNestedList(0,
5138 				XNSpotLocation, &over_spot,
5139 				XNForeground, (Pixel)gui.def_norm_pixel,
5140 				XNBackground, (Pixel)gui.def_back_pixel,
5141 				NULL);
5142 	status_list = XVaCreateNestedList(0,
5143 				XNForeground, (Pixel)gui.def_norm_pixel,
5144 				XNBackground, (Pixel)gui.def_back_pixel,
5145 				NULL);
5146     }
5147 
5148     xic = XCreateIC(xim,
5149 		    XNInputStyle, input_style,
5150 		    XNClientWindow, x11_window,
5151 		    XNFocusWindow, gui.wid,
5152 		    XNPreeditAttributes, preedit_list,
5153 		    XNStatusAttributes, status_list,
5154 		    NULL);
5155     XFree(status_list);
5156     XFree(preedit_list);
5157     if (xic != NULL)
5158     {
5159 	if (input_style & XIMStatusArea)
5160 	{
5161 	    xim_set_status_area();
5162 	    status_area_enabled = TRUE;
5163 	}
5164 	else
5165 	    gui_set_shellsize(FALSE, FALSE);
5166     }
5167     else
5168     {
5169 	EMSG(_(e_xim));
5170 	XCloseIM(xim);
5171 	return FALSE;
5172     }
5173 
5174     return TRUE;
5175 }
5176 
5177 #endif /* FEAT_GUI_X11 */
5178 
5179 #if defined(FEAT_GUI_GTK) || defined(PROTO)
5180 
5181 # ifdef FEAT_XFONTSET
5182 static char e_overthespot[] = N_("E290: over-the-spot style requires fontset");
5183 # endif
5184 
5185 # ifdef PROTO
5186 typedef int GdkIC;
5187 # endif
5188 
5189     void
5190 xim_decide_input_style()
5191 {
5192     /* GDK_IM_STATUS_CALLBACKS was disabled, enabled it to allow Japanese
5193      * OverTheSpot. */
5194     int supported_style = (int)GDK_IM_PREEDIT_NONE |
5195 				 (int)GDK_IM_PREEDIT_NOTHING |
5196 				 (int)GDK_IM_PREEDIT_POSITION |
5197 				 (int)GDK_IM_PREEDIT_CALLBACKS |
5198 				 (int)GDK_IM_STATUS_CALLBACKS |
5199 				 (int)GDK_IM_STATUS_AREA |
5200 				 (int)GDK_IM_STATUS_NONE |
5201 				 (int)GDK_IM_STATUS_NOTHING;
5202 
5203 #ifdef XIM_DEBUG
5204     xim_log("xim_decide_input_style()\n");
5205 #endif
5206 
5207     if (!gdk_im_ready())
5208 	xim_input_style = 0;
5209     else
5210     {
5211 	if (gtk_major_version > 1
5212 		|| (gtk_major_version == 1
5213 		    && (gtk_minor_version > 2
5214 			|| (gtk_minor_version == 2 && gtk_micro_version >= 3))))
5215 	    use_status_area = TRUE;
5216 	else
5217 	{
5218 	    EMSG(_("E291: Your GTK+ is older than 1.2.3. Status area disabled"));
5219 	    use_status_area = FALSE;
5220 	}
5221 #ifdef FEAT_XFONTSET
5222 	if (gui.fontset == NOFONTSET || gui.fontset->type != GDK_FONT_FONTSET)
5223 #endif
5224 	    supported_style &= ~((int)GDK_IM_PREEDIT_POSITION
5225 						   | (int)GDK_IM_STATUS_AREA);
5226 	if (!use_status_area)
5227 	    supported_style &= ~(int)GDK_IM_STATUS_AREA;
5228 	xim_input_style = (int)gdk_im_decide_style((GdkIMStyle)supported_style);
5229     }
5230 }
5231 
5232 /*ARGSUSED*/
5233     static void
5234 preedit_start_cbproc(XIC xic, XPointer client_data, XPointer call_data)
5235 {
5236 #ifdef XIM_DEBUG
5237     xim_log("xim_decide_input_style()\n");
5238 #endif
5239 
5240     draw_feedback = NULL;
5241     xim_can_preediting = TRUE;
5242     xim_has_preediting = TRUE;
5243     gui_update_cursor(TRUE, FALSE);
5244     if (showmode() > 0)
5245     {
5246 	setcursor();
5247 	out_flush();
5248     }
5249 }
5250 
5251     static void
5252 xim_back_delete(int n)
5253 {
5254     char_u str[3];
5255 
5256     str[0] = CSI;
5257     str[1] = 'k';
5258     str[2] = 'b';
5259     while (n-- > 0)
5260 	add_to_input_buf(str, 3);
5261 }
5262 
5263 static GSList *key_press_event_queue = NULL;
5264 static gboolean processing_queued_event = FALSE;
5265 
5266 /*ARGSUSED*/
5267     static void
5268 preedit_draw_cbproc(XIC xic, XPointer client_data, XPointer call_data)
5269 {
5270     XIMPreeditDrawCallbackStruct *draw_data;
5271     XIMText	*text;
5272     char	*src;
5273     GSList	*event_queue;
5274 
5275 #ifdef XIM_DEBUG
5276     xim_log("preedit_draw_cbproc()\n");
5277 #endif
5278 
5279     draw_data = (XIMPreeditDrawCallbackStruct *) call_data;
5280     text = (XIMText *) draw_data->text;
5281 
5282     if ((text == NULL && draw_data->chg_length == preedit_buf_len)
5283 						      || preedit_buf_len == 0)
5284     {
5285 	init_preedit_start_col();
5286 	vim_free(draw_feedback);
5287 	draw_feedback = NULL;
5288     }
5289     if (draw_data->chg_length > 0)
5290     {
5291 	int bs_cnt;
5292 
5293 	if (draw_data->chg_length > preedit_buf_len)
5294 	    bs_cnt = preedit_buf_len;
5295 	else
5296 	    bs_cnt = draw_data->chg_length;
5297 	xim_back_delete(bs_cnt);
5298 	preedit_buf_len -= bs_cnt;
5299     }
5300     if (text != NULL)
5301     {
5302 	int		len;
5303 #ifdef FEAT_MBYTE
5304 	char_u		*buf = NULL;
5305 	unsigned int	nfeedback = 0;
5306 #endif
5307 	char_u		*ptr;
5308 
5309 	src = text->string.multi_byte;
5310 	if (src != NULL && !text->encoding_is_wchar)
5311 	{
5312 	    len = strlen(src);
5313 	    ptr = (char_u *)src;
5314 	    /* Avoid the enter for decision */
5315 	    if (*ptr == '\n')
5316 		return;
5317 
5318 #ifdef FEAT_MBYTE
5319 	    if (input_conv.vc_type != CONV_NONE
5320 		    && (buf = string_convert(&input_conv,
5321 						 (char_u *)src, &len)) != NULL)
5322 	    {
5323 		/* Converted from 'termencoding' to 'encoding'. */
5324 		add_to_input_buf_csi(buf, len);
5325 		ptr = buf;
5326 	    }
5327 	    else
5328 #endif
5329 		add_to_input_buf_csi((char_u *)src, len);
5330 	    /* Add count of character to preedit_buf_len  */
5331 	    while (*ptr != NUL)
5332 	    {
5333 #ifdef FEAT_MBYTE
5334 		if (draw_data->text->feedback != NULL)
5335 		{
5336 		    if (draw_feedback == NULL)
5337 			draw_feedback = (char *)alloc(draw_data->chg_first
5338 							      + text->length);
5339 		    else
5340 			draw_feedback = realloc(draw_feedback,
5341 					 draw_data->chg_first + text->length);
5342 		    if (draw_feedback != NULL)
5343 		    {
5344 			draw_feedback[nfeedback + draw_data->chg_first]
5345 				       = draw_data->text->feedback[nfeedback];
5346 			nfeedback++;
5347 		    }
5348 		}
5349 		if (has_mbyte)
5350 		    ptr += (*mb_ptr2len)(ptr);
5351 		else
5352 #endif
5353 		    ptr++;
5354 		preedit_buf_len++;
5355 	    }
5356 #ifdef FEAT_MBYTE
5357 	    vim_free(buf);
5358 #endif
5359 	    preedit_end_col = MAXCOL;
5360 	}
5361     }
5362     if (text != NULL || draw_data->chg_length > 0)
5363     {
5364 	event_queue = key_press_event_queue;
5365 	processing_queued_event = TRUE;
5366 	while (event_queue != NULL && processing_queued_event)
5367 	{
5368 	    GdkEvent *ev = event_queue->data;
5369 
5370 	    gboolean *ret;
5371 	    gtk_signal_emit_by_name((GtkObject*)gui.mainwin, "key_press_event",
5372 								    ev, &ret);
5373 	    gdk_event_free(ev);
5374 	    event_queue = event_queue->next;
5375 	}
5376 	processing_queued_event = FALSE;
5377 	if (key_press_event_queue)
5378 	{
5379 	    g_slist_free(key_press_event_queue);
5380 	    key_press_event_queue = NULL;
5381 	}
5382     }
5383     if (gtk_main_level() > 0)
5384 	gtk_main_quit();
5385 }
5386 
5387 /*
5388  * Retrieve the highlighting attributes at column col in the preedit string.
5389  * Return -1 if not in preediting mode or if col is out of range.
5390  */
5391     int
5392 im_get_feedback_attr(int col)
5393 {
5394     if (draw_feedback != NULL && col < preedit_buf_len)
5395     {
5396 	if (draw_feedback[col] & XIMReverse)
5397 	    return HL_INVERSE;
5398 	else if (draw_feedback[col] & XIMUnderline)
5399 	    return HL_UNDERLINE;
5400 	else
5401 	    return hl_attr(HLF_V);
5402     }
5403 
5404     return -1;
5405 }
5406 
5407 /*ARGSUSED*/
5408     static void
5409 preedit_caret_cbproc(XIC xic, XPointer client_data, XPointer call_data)
5410 {
5411 #ifdef XIM_DEBUG
5412     xim_log("preedit_caret_cbproc()\n");
5413 #endif
5414 }
5415 
5416 /*ARGSUSED*/
5417     static void
5418 preedit_done_cbproc(XIC xic, XPointer client_data, XPointer call_data)
5419 {
5420 #ifdef XIM_DEBUG
5421     xim_log("preedit_done_cbproc()\n");
5422 #endif
5423 
5424     vim_free(draw_feedback);
5425     draw_feedback = NULL;
5426     xim_can_preediting = FALSE;
5427     xim_has_preediting = FALSE;
5428     gui_update_cursor(TRUE, FALSE);
5429     if (showmode() > 0)
5430     {
5431 	setcursor();
5432 	out_flush();
5433     }
5434 }
5435 
5436     void
5437 xim_reset(void)
5438 {
5439     char *text;
5440 
5441 #ifdef XIM_DEBUG
5442     xim_log("xim_reset()\n");
5443 #endif
5444 
5445     if (xic != NULL)
5446     {
5447 	text = XmbResetIC(((GdkICPrivate *)xic)->xic);
5448 	if (text != NULL && !(xim_input_style & (int)GDK_IM_PREEDIT_CALLBACKS))
5449 	    add_to_input_buf_csi((char_u *)text, strlen(text));
5450 	else
5451 	    preedit_buf_len = 0;
5452 	if (text != NULL)
5453 	    XFree(text);
5454     }
5455 }
5456 
5457 /*ARGSUSED*/
5458     int
5459 xim_queue_key_press_event(GdkEventKey *event, int down)
5460 {
5461 #ifdef XIM_DEBUG
5462     xim_log("xim_queue_key_press_event()\n");
5463 #endif
5464 
5465     if (preedit_buf_len <= 0)
5466 	return FALSE;
5467     if (processing_queued_event)
5468 	processing_queued_event = FALSE;
5469 
5470     key_press_event_queue = g_slist_append(key_press_event_queue,
5471 					   gdk_event_copy((GdkEvent *)event));
5472     return TRUE;
5473 }
5474 
5475 /*ARGSUSED*/
5476     static void
5477 preedit_callback_setup(GdkIC *ic)
5478 {
5479     XIC xxic;
5480     XVaNestedList preedit_attr;
5481     XIMCallback preedit_start_cb;
5482     XIMCallback preedit_draw_cb;
5483     XIMCallback preedit_caret_cb;
5484     XIMCallback preedit_done_cb;
5485 
5486     xxic = ((GdkICPrivate*)xic)->xic;
5487     preedit_start_cb.callback = (XIMProc)preedit_start_cbproc;
5488     preedit_draw_cb.callback = (XIMProc)preedit_draw_cbproc;
5489     preedit_caret_cb.callback = (XIMProc)preedit_caret_cbproc;
5490     preedit_done_cb.callback = (XIMProc)preedit_done_cbproc;
5491     preedit_attr
5492 	= XVaCreateNestedList (0,
5493 			       XNPreeditStartCallback, &preedit_start_cb,
5494 			       XNPreeditDrawCallback, &preedit_draw_cb,
5495 			       XNPreeditCaretCallback, &preedit_caret_cb,
5496 			       XNPreeditDoneCallback, &preedit_done_cb,
5497 			       0);
5498     XSetICValues (xxic, XNPreeditAttributes, preedit_attr, 0);
5499     XFree(preedit_attr);
5500 }
5501 
5502 /*ARGSUSED*/
5503     static void
5504 reset_state_setup(GdkIC *ic)
5505 {
5506 #ifdef USE_X11R6_XIM
5507     /* don't change the input context when we call reset */
5508     XSetICValues(((GdkICPrivate*)ic)->xic, XNResetState, XIMPreserveState, 0);
5509 #endif
5510 }
5511 
5512     void
5513 xim_init(void)
5514 {
5515 #ifdef XIM_DEBUG
5516     xim_log("xim_init()\n");
5517 #endif
5518 
5519     xic = NULL;
5520     xic_attr = NULL;
5521 
5522     if (!gdk_im_ready())
5523     {
5524 	if (p_verbose > 0)
5525 	{
5526 	    verbose_enter();
5527 	    EMSG(_("E292: Input Method Server is not running"));
5528 	    verbose_leave();
5529 	}
5530 	return;
5531     }
5532     if ((xic_attr = gdk_ic_attr_new()) != NULL)
5533     {
5534 #ifdef FEAT_XFONTSET
5535 	gint width, height;
5536 #endif
5537 	int		mask;
5538 	GdkColormap	*colormap;
5539 	GdkICAttr	*attr = xic_attr;
5540 	int		attrmask = (int)GDK_IC_ALL_REQ;
5541 	GtkWidget	*widget = gui.drawarea;
5542 
5543 	attr->style = (GdkIMStyle)xim_input_style;
5544 	attr->client_window = gui.mainwin->window;
5545 
5546 	if ((colormap = gtk_widget_get_colormap(widget)) !=
5547 	    gtk_widget_get_default_colormap())
5548 	{
5549 	    attrmask |= (int)GDK_IC_PREEDIT_COLORMAP;
5550 	    attr->preedit_colormap = colormap;
5551 	}
5552 	attrmask |= (int)GDK_IC_PREEDIT_FOREGROUND;
5553 	attrmask |= (int)GDK_IC_PREEDIT_BACKGROUND;
5554 	attr->preedit_foreground = widget->style->fg[GTK_STATE_NORMAL];
5555 	attr->preedit_background = widget->style->base[GTK_STATE_NORMAL];
5556 
5557 #ifdef FEAT_XFONTSET
5558 	if ((xim_input_style & (int)GDK_IM_PREEDIT_MASK)
5559 					      == (int)GDK_IM_PREEDIT_POSITION)
5560 	{
5561 	    if (gui.fontset == NOFONTSET
5562 		    || gui.fontset->type != GDK_FONT_FONTSET)
5563 	    {
5564 		EMSG(_(e_overthespot));
5565 	    }
5566 	    else
5567 	    {
5568 		gdk_window_get_size(widget->window, &width, &height);
5569 
5570 		attrmask |= (int)GDK_IC_PREEDIT_POSITION_REQ;
5571 		attr->spot_location.x = TEXT_X(0);
5572 		attr->spot_location.y = TEXT_Y(0);
5573 		attr->preedit_area.x = gui.border_offset;
5574 		attr->preedit_area.y = gui.border_offset;
5575 		attr->preedit_area.width = width - 2*gui.border_offset;
5576 		attr->preedit_area.height = height - 2*gui.border_offset;
5577 		attr->preedit_fontset = gui.fontset;
5578 	    }
5579 	}
5580 
5581 	if ((xim_input_style & (int)GDK_IM_STATUS_MASK)
5582 						   == (int)GDK_IM_STATUS_AREA)
5583 	{
5584 	    if (gui.fontset == NOFONTSET
5585 		    || gui.fontset->type != GDK_FONT_FONTSET)
5586 	    {
5587 		EMSG(_(e_overthespot));
5588 	    }
5589 	    else
5590 	    {
5591 		gdk_window_get_size(gui.mainwin->window, &width, &height);
5592 		attrmask |= (int)GDK_IC_STATUS_AREA_REQ;
5593 		attr->status_area.x = 0;
5594 		attr->status_area.y = height - gui.char_height - 1;
5595 		attr->status_area.width = width;
5596 		attr->status_area.height = gui.char_height;
5597 		attr->status_fontset = gui.fontset;
5598 	    }
5599 	}
5600 	else if ((xim_input_style & (int)GDK_IM_STATUS_MASK)
5601 					      == (int)GDK_IM_STATUS_CALLBACKS)
5602 	{
5603 	    /* FIXME */
5604 	}
5605 #endif
5606 
5607 	xic = gdk_ic_new(attr, (GdkICAttributesType)attrmask);
5608 
5609 	if (xic == NULL)
5610 	    EMSG(_(e_xim));
5611 	else
5612 	{
5613 	    mask = (int)gdk_window_get_events(widget->window);
5614 	    mask |= (int)gdk_ic_get_events(xic);
5615 	    gdk_window_set_events(widget->window, (GdkEventMask)mask);
5616 	    if (xim_input_style & (int)GDK_IM_PREEDIT_CALLBACKS)
5617 		preedit_callback_setup(xic);
5618 	    reset_state_setup(xic);
5619 	}
5620     }
5621 }
5622 
5623     void
5624 im_shutdown(void)
5625 {
5626 #ifdef XIM_DEBUG
5627     xim_log("im_shutdown()\n");
5628 #endif
5629 
5630     if (xic != NULL)
5631     {
5632 	gdk_im_end();
5633 	gdk_ic_destroy(xic);
5634 	xic = NULL;
5635     }
5636     xim_is_active = FALSE;
5637     xim_can_preediting = FALSE;
5638     preedit_start_col = MAXCOL;
5639     xim_has_preediting = FALSE;
5640 }
5641 
5642 #endif /* FEAT_GUI_GTK */
5643 
5644     int
5645 xim_get_status_area_height()
5646 {
5647 #ifdef FEAT_GUI_GTK
5648     if (xim_input_style & (int)GDK_IM_STATUS_AREA)
5649 	return gui.char_height;
5650 #else
5651     if (status_area_enabled)
5652 	return gui.char_height;
5653 #endif
5654     return 0;
5655 }
5656 
5657 /*
5658  * Get IM status.  When IM is on, return TRUE.  Else return FALSE.
5659  * FIXME: This doesn't work correctly: Having focus doesn't always mean XIM is
5660  * active, when not having focus XIM may still be active (e.g., when using a
5661  * tear-off menu item).
5662  */
5663     int
5664 im_get_status()
5665 {
5666 #  ifdef FEAT_GUI_GTK
5667     if (xim_input_style & (int)GDK_IM_PREEDIT_CALLBACKS)
5668 	return xim_can_preediting;
5669 #  endif
5670     return xim_has_focus;
5671 }
5672 
5673 # endif /* !HAVE_GTK2 */
5674 
5675 # if defined(FEAT_GUI_GTK) || defined(PROTO)
5676     int
5677 im_is_preediting()
5678 {
5679     return xim_has_preediting;
5680 }
5681 # endif
5682 #endif /* FEAT_XIM */
5683 
5684 #if defined(FEAT_MBYTE) || defined(PROTO)
5685 
5686 /*
5687  * Setup "vcp" for conversion from "from" to "to".
5688  * The names must have been made canonical with enc_canonize().
5689  * vcp->vc_type must have been initialized to CONV_NONE.
5690  * Note: cannot be used for conversion from/to ucs-2 and ucs-4 (will use utf-8
5691  * instead).
5692  * Afterwards invoke with "from" and "to" equal to NULL to cleanup.
5693  * Return FAIL when conversion is not supported, OK otherwise.
5694  */
5695     int
5696 convert_setup(vcp, from, to)
5697     vimconv_T	*vcp;
5698     char_u	*from;
5699     char_u	*to;
5700 {
5701     int		from_prop;
5702     int		to_prop;
5703 
5704     /* Reset to no conversion. */
5705 # ifdef USE_ICONV
5706     if (vcp->vc_type == CONV_ICONV && vcp->vc_fd != (iconv_t)-1)
5707 	iconv_close(vcp->vc_fd);
5708 # endif
5709     vcp->vc_type = CONV_NONE;
5710     vcp->vc_factor = 1;
5711     vcp->vc_fail = FALSE;
5712 
5713     /* No conversion when one of the names is empty or they are equal. */
5714     if (from == NULL || *from == NUL || to == NULL || *to == NUL
5715 						     || STRCMP(from, to) == 0)
5716 	return OK;
5717 
5718     from_prop = enc_canon_props(from);
5719     to_prop = enc_canon_props(to);
5720     if ((from_prop & ENC_LATIN1) && (to_prop & ENC_UNICODE))
5721     {
5722 	/* Internal latin1 -> utf-8 conversion. */
5723 	vcp->vc_type = CONV_TO_UTF8;
5724 	vcp->vc_factor = 2;	/* up to twice as long */
5725     }
5726     else if ((from_prop & ENC_LATIN9) && (to_prop & ENC_UNICODE))
5727     {
5728 	/* Internal latin9 -> utf-8 conversion. */
5729 	vcp->vc_type = CONV_9_TO_UTF8;
5730 	vcp->vc_factor = 3;	/* up to three as long (euro sign) */
5731     }
5732     else if ((from_prop & ENC_UNICODE) && (to_prop & ENC_LATIN1))
5733     {
5734 	/* Internal utf-8 -> latin1 conversion. */
5735 	vcp->vc_type = CONV_TO_LATIN1;
5736     }
5737     else if ((from_prop & ENC_UNICODE) && (to_prop & ENC_LATIN9))
5738     {
5739 	/* Internal utf-8 -> latin9 conversion. */
5740 	vcp->vc_type = CONV_TO_LATIN9;
5741     }
5742 #ifdef WIN3264
5743     /* Win32-specific codepage <-> codepage conversion without iconv. */
5744     else if (((from_prop & ENC_UNICODE) || encname2codepage(from) > 0)
5745 	    && ((to_prop & ENC_UNICODE) || encname2codepage(to) > 0))
5746     {
5747 	vcp->vc_type = CONV_CODEPAGE;
5748 	vcp->vc_factor = 2;	/* up to twice as long */
5749 	vcp->vc_cpfrom = (from_prop & ENC_UNICODE) ? 0 : encname2codepage(from);
5750 	vcp->vc_cpto = (to_prop & ENC_UNICODE) ? 0 : encname2codepage(to);
5751     }
5752 #endif
5753 #ifdef MACOS_X
5754     else if ((from_prop & ENC_MACROMAN) && (to_prop & ENC_LATIN1))
5755     {
5756 	vcp->vc_type = CONV_MAC_LATIN1;
5757     }
5758     else if ((from_prop & ENC_MACROMAN) && (to_prop & ENC_UNICODE))
5759     {
5760 	vcp->vc_type = CONV_MAC_UTF8;
5761 	vcp->vc_factor = 2;	/* up to twice as long */
5762     }
5763     else if ((from_prop & ENC_LATIN1) && (to_prop & ENC_MACROMAN))
5764     {
5765 	vcp->vc_type = CONV_LATIN1_MAC;
5766     }
5767     else if ((from_prop & ENC_UNICODE) && (to_prop & ENC_MACROMAN))
5768     {
5769 	vcp->vc_type = CONV_UTF8_MAC;
5770     }
5771 #endif
5772 # ifdef USE_ICONV
5773     else
5774     {
5775 	/* Use iconv() for conversion. */
5776 	vcp->vc_fd = (iconv_t)my_iconv_open(
5777 		(to_prop & ENC_UNICODE) ? (char_u *)"utf-8" : to,
5778 		(from_prop & ENC_UNICODE) ? (char_u *)"utf-8" : from);
5779 	if (vcp->vc_fd != (iconv_t)-1)
5780 	{
5781 	    vcp->vc_type = CONV_ICONV;
5782 	    vcp->vc_factor = 4;	/* could be longer too... */
5783 	}
5784     }
5785 # endif
5786     if (vcp->vc_type == CONV_NONE)
5787 	return FAIL;
5788 
5789     return OK;
5790 }
5791 
5792 #if defined(FEAT_GUI) || defined(AMIGA) || defined(WIN3264) \
5793 	|| defined(MSDOS) || defined(PROTO)
5794 /*
5795  * Do conversion on typed input characters in-place.
5796  * The input and output are not NUL terminated!
5797  * Returns the length after conversion.
5798  */
5799     int
5800 convert_input(ptr, len, maxlen)
5801     char_u	*ptr;
5802     int		len;
5803     int		maxlen;
5804 {
5805     return convert_input_safe(ptr, len, maxlen, NULL, NULL);
5806 }
5807 #endif
5808 
5809 /*
5810  * Like convert_input(), but when there is an incomplete byte sequence at the
5811  * end return that as an allocated string in "restp" and set "*restlenp" to
5812  * the length.  If "restp" is NULL it is not used.
5813  */
5814     int
5815 convert_input_safe(ptr, len, maxlen, restp, restlenp)
5816     char_u	*ptr;
5817     int		len;
5818     int		maxlen;
5819     char_u	**restp;
5820     int		*restlenp;
5821 {
5822     char_u	*d;
5823     int		dlen = len;
5824     int		unconvertlen = 0;
5825 
5826     d = string_convert_ext(&input_conv, ptr, &dlen,
5827 					restp == NULL ? NULL : &unconvertlen);
5828     if (d != NULL)
5829     {
5830 	if (dlen <= maxlen)
5831 	{
5832 	    if (unconvertlen > 0)
5833 	    {
5834 		/* Move the unconverted characters to allocated memory. */
5835 		*restp = alloc(unconvertlen);
5836 		if (*restp != NULL)
5837 		    mch_memmove(*restp, ptr + len - unconvertlen, unconvertlen);
5838 		*restlenp = unconvertlen;
5839 	    }
5840 	    mch_memmove(ptr, d, dlen);
5841 	}
5842 	else
5843 	    /* result is too long, keep the unconverted text (the caller must
5844 	     * have done something wrong!) */
5845 	    dlen = len;
5846 	vim_free(d);
5847     }
5848     return dlen;
5849 }
5850 
5851 /*
5852  * Convert text "ptr[*lenp]" according to "vcp".
5853  * Returns the result in allocated memory and sets "*lenp".
5854  * When "lenp" is NULL, use NUL terminated strings.
5855  * Illegal chars are often changed to "?", unless vcp->vc_fail is set.
5856  * When something goes wrong, NULL is returned and "*lenp" is unchanged.
5857  */
5858     char_u *
5859 string_convert(vcp, ptr, lenp)
5860     vimconv_T	*vcp;
5861     char_u	*ptr;
5862     int		*lenp;
5863 {
5864     return string_convert_ext(vcp, ptr, lenp, NULL);
5865 }
5866 
5867 /*
5868  * Like string_convert(), but when "unconvlenp" is not NULL and there are is
5869  * an incomplete sequence at the end it is not converted and "*unconvlenp" is
5870  * set to the number of remaining bytes.
5871  */
5872     char_u *
5873 string_convert_ext(vcp, ptr, lenp, unconvlenp)
5874     vimconv_T	*vcp;
5875     char_u	*ptr;
5876     int		*lenp;
5877     int		*unconvlenp;
5878 {
5879     char_u	*retval = NULL;
5880     char_u	*d;
5881     int		len;
5882     int		i;
5883     int		l;
5884     int		c;
5885 
5886     if (lenp == NULL)
5887 	len = (int)STRLEN(ptr);
5888     else
5889 	len = *lenp;
5890     if (len == 0)
5891 	return vim_strsave((char_u *)"");
5892 
5893     switch (vcp->vc_type)
5894     {
5895 	case CONV_TO_UTF8:	/* latin1 to utf-8 conversion */
5896 	    retval = alloc(len * 2 + 1);
5897 	    if (retval == NULL)
5898 		break;
5899 	    d = retval;
5900 	    for (i = 0; i < len; ++i)
5901 	    {
5902 		c = ptr[i];
5903 		if (c < 0x80)
5904 		    *d++ = c;
5905 		else
5906 		{
5907 		    *d++ = 0xc0 + ((unsigned)c >> 6);
5908 		    *d++ = 0x80 + (c & 0x3f);
5909 		}
5910 	    }
5911 	    *d = NUL;
5912 	    if (lenp != NULL)
5913 		*lenp = (int)(d - retval);
5914 	    break;
5915 
5916 	case CONV_9_TO_UTF8:	/* latin9 to utf-8 conversion */
5917 	    retval = alloc(len * 3 + 1);
5918 	    if (retval == NULL)
5919 		break;
5920 	    d = retval;
5921 	    for (i = 0; i < len; ++i)
5922 	    {
5923 		c = ptr[i];
5924 		switch (c)
5925 		{
5926 		    case 0xa4: c = 0x20ac; break;   /* euro */
5927 		    case 0xa6: c = 0x0160; break;   /* S hat */
5928 		    case 0xa8: c = 0x0161; break;   /* S -hat */
5929 		    case 0xb4: c = 0x017d; break;   /* Z hat */
5930 		    case 0xb8: c = 0x017e; break;   /* Z -hat */
5931 		    case 0xbc: c = 0x0152; break;   /* OE */
5932 		    case 0xbd: c = 0x0153; break;   /* oe */
5933 		    case 0xbe: c = 0x0178; break;   /* Y */
5934 		}
5935 		d += utf_char2bytes(c, d);
5936 	    }
5937 	    *d = NUL;
5938 	    if (lenp != NULL)
5939 		*lenp = (int)(d - retval);
5940 	    break;
5941 
5942 	case CONV_TO_LATIN1:	/* utf-8 to latin1 conversion */
5943 	case CONV_TO_LATIN9:	/* utf-8 to latin9 conversion */
5944 	    retval = alloc(len + 1);
5945 	    if (retval == NULL)
5946 		break;
5947 	    d = retval;
5948 	    for (i = 0; i < len; ++i)
5949 	    {
5950 		l = utf_ptr2len(ptr + i);
5951 		if (l == 0)
5952 		    *d++ = NUL;
5953 		else if (l == 1)
5954 		{
5955 		    if (unconvlenp != NULL && utf8len_tab[ptr[i]] > len - i)
5956 		    {
5957 			/* Incomplete sequence at the end. */
5958 			*unconvlenp = len - i;
5959 			break;
5960 		    }
5961 		    *d++ = ptr[i];
5962 		}
5963 		else
5964 		{
5965 		    c = utf_ptr2char(ptr + i);
5966 		    if (vcp->vc_type == CONV_TO_LATIN9)
5967 			switch (c)
5968 			{
5969 			    case 0x20ac: c = 0xa4; break;   /* euro */
5970 			    case 0x0160: c = 0xa6; break;   /* S hat */
5971 			    case 0x0161: c = 0xa8; break;   /* S -hat */
5972 			    case 0x017d: c = 0xb4; break;   /* Z hat */
5973 			    case 0x017e: c = 0xb8; break;   /* Z -hat */
5974 			    case 0x0152: c = 0xbc; break;   /* OE */
5975 			    case 0x0153: c = 0xbd; break;   /* oe */
5976 			    case 0x0178: c = 0xbe; break;   /* Y */
5977 			    case 0xa4:
5978 			    case 0xa6:
5979 			    case 0xa8:
5980 			    case 0xb4:
5981 			    case 0xb8:
5982 			    case 0xbc:
5983 			    case 0xbd:
5984 			    case 0xbe: c = 0x100; break; /* not in latin9 */
5985 			}
5986 		    if (!utf_iscomposing(c))	/* skip composing chars */
5987 		    {
5988 			if (c < 0x100)
5989 			    *d++ = c;
5990 			else if (vcp->vc_fail)
5991 			{
5992 			    vim_free(retval);
5993 			    return NULL;
5994 			}
5995 			else
5996 			{
5997 			    *d++ = 0xbf;
5998 			    if (utf_char2cells(c) > 1)
5999 				*d++ = '?';
6000 			}
6001 		    }
6002 		    i += l - 1;
6003 		}
6004 	    }
6005 	    *d = NUL;
6006 	    if (lenp != NULL)
6007 		*lenp = (int)(d - retval);
6008 	    break;
6009 
6010 # ifdef MACOS_CONVERT
6011 	case CONV_MAC_LATIN1:
6012 	    retval = mac_string_convert(ptr, len, lenp, vcp->vc_fail,
6013 					'm', 'l', unconvlenp);
6014 	    break;
6015 
6016 	case CONV_LATIN1_MAC:
6017 	    retval = mac_string_convert(ptr, len, lenp, vcp->vc_fail,
6018 					'l', 'm', unconvlenp);
6019 	    break;
6020 
6021 	case CONV_MAC_UTF8:
6022 	    retval = mac_string_convert(ptr, len, lenp, vcp->vc_fail,
6023 					'm', 'u', unconvlenp);
6024 	    break;
6025 
6026 	case CONV_UTF8_MAC:
6027 	    retval = mac_string_convert(ptr, len, lenp, vcp->vc_fail,
6028 					'u', 'm', unconvlenp);
6029 	    break;
6030 # endif
6031 
6032 # ifdef USE_ICONV
6033 	case CONV_ICONV:	/* conversion with output_conv.vc_fd */
6034 	    retval = iconv_string(vcp, ptr, len, unconvlenp);
6035 	    if (retval != NULL && lenp != NULL)
6036 		*lenp = (int)STRLEN(retval);
6037 	    break;
6038 # endif
6039 # ifdef WIN3264
6040 	case CONV_CODEPAGE:		/* codepage -> codepage */
6041 	{
6042 	    int		retlen;
6043 	    int		tmp_len;
6044 	    short_u	*tmp;
6045 
6046 	    /* 1. codepage/UTF-8  ->  ucs-2. */
6047 	    if (vcp->vc_cpfrom == 0)
6048 		tmp_len = utf8_to_ucs2(ptr, len, NULL, NULL);
6049 	    else
6050 		tmp_len = MultiByteToWideChar(vcp->vc_cpfrom, 0,
6051 							      ptr, len, 0, 0);
6052 	    tmp = (short_u *)alloc(sizeof(short_u) * tmp_len);
6053 	    if (tmp == NULL)
6054 		break;
6055 	    if (vcp->vc_cpfrom == 0)
6056 		utf8_to_ucs2(ptr, len, tmp, unconvlenp);
6057 	    else
6058 		MultiByteToWideChar(vcp->vc_cpfrom, 0, ptr, len, tmp, tmp_len);
6059 
6060 	    /* 2. ucs-2  ->  codepage/UTF-8. */
6061 	    if (vcp->vc_cpto == 0)
6062 		retlen = ucs2_to_utf8(tmp, tmp_len, NULL);
6063 	    else
6064 		retlen = WideCharToMultiByte(vcp->vc_cpto, 0,
6065 						    tmp, tmp_len, 0, 0, 0, 0);
6066 	    retval = alloc(retlen + 1);
6067 	    if (retval != NULL)
6068 	    {
6069 		if (vcp->vc_cpto == 0)
6070 		    ucs2_to_utf8(tmp, tmp_len, retval);
6071 		else
6072 		    WideCharToMultiByte(vcp->vc_cpto, 0,
6073 					  tmp, tmp_len, retval, retlen, 0, 0);
6074 		retval[retlen] = NUL;
6075 		if (lenp != NULL)
6076 		    *lenp = retlen;
6077 	    }
6078 	    vim_free(tmp);
6079 	    break;
6080 	}
6081 # endif
6082     }
6083 
6084     return retval;
6085 }
6086 #endif
6087