xref: /vim-8.2.3635/src/mbyte.c (revision 044b68f4)
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_DECMCS	58
315     {"dec-mcs",		ENC_8BIT,		0},	/* DEC MCS */
316 #define IDX_HPROMAN8	59
317     {"hp-roman8",	ENC_8BIT,		0},	/* HP Roman8 */
318 #define IDX_COUNT	60
319 };
320 
321 /*
322  * Aliases for encoding names.
323  */
324 static struct
325 {   char *name;		int canon;}
326 enc_alias_table[] =
327 {
328     {"ansi",		IDX_LATIN_1},
329     {"iso-8859-1",	IDX_LATIN_1},
330     {"latin2",		IDX_ISO_2},
331     {"latin3",		IDX_ISO_3},
332     {"latin4",		IDX_ISO_4},
333     {"cyrillic",	IDX_ISO_5},
334     {"arabic",		IDX_ISO_6},
335     {"greek",		IDX_ISO_7},
336 #ifdef WIN3264
337     {"hebrew",		IDX_CP1255},
338 #else
339     {"hebrew",		IDX_ISO_8},
340 #endif
341     {"latin5",		IDX_ISO_9},
342     {"turkish",		IDX_ISO_9}, /* ? */
343     {"latin6",		IDX_ISO_10},
344     {"nordic",		IDX_ISO_10}, /* ? */
345     {"thai",		IDX_ISO_11}, /* ? */
346     {"latin7",		IDX_ISO_13},
347     {"latin8",		IDX_ISO_14},
348     {"latin9",		IDX_ISO_15},
349     {"utf8",		IDX_UTF8},
350     {"unicode",		IDX_UCS2},
351     {"ucs2",		IDX_UCS2},
352     {"ucs2be",		IDX_UCS2},
353     {"ucs-2be",		IDX_UCS2},
354     {"ucs2le",		IDX_UCS2LE},
355     {"utf16",		IDX_UTF16},
356     {"utf16be",		IDX_UTF16},
357     {"utf-16be",	IDX_UTF16},
358     {"utf16le",		IDX_UTF16LE},
359     {"ucs4",		IDX_UCS4},
360     {"ucs4be",		IDX_UCS4},
361     {"ucs-4be",		IDX_UCS4},
362     {"ucs4le",		IDX_UCS4LE},
363     {"932",		IDX_CP932},
364     {"949",		IDX_CP949},
365     {"936",		IDX_CP936},
366     {"gbk",		IDX_CP936},
367     {"950",		IDX_CP950},
368     {"eucjp",		IDX_EUC_JP},
369     {"unix-jis",	IDX_EUC_JP},
370     {"ujis",		IDX_EUC_JP},
371     {"shift-jis",	IDX_SJIS},
372     {"euckr",		IDX_EUC_KR},
373     {"5601",		IDX_EUC_KR},	/* Sun: KS C 5601 */
374     {"euccn",		IDX_EUC_CN},
375     {"gb2312",		IDX_EUC_CN},
376     {"euctw",		IDX_EUC_TW},
377 #if defined(WIN3264) || defined(WIN32UNIX) || defined(MACOS)
378     {"japan",		IDX_CP932},
379     {"korea",		IDX_CP949},
380     {"prc",		IDX_CP936},
381     {"chinese",		IDX_CP936},
382     {"taiwan",		IDX_CP950},
383     {"big5",		IDX_CP950},
384 #else
385     {"japan",		IDX_EUC_JP},
386     {"korea",		IDX_EUC_KR},
387     {"prc",		IDX_EUC_CN},
388     {"chinese",		IDX_EUC_CN},
389     {"taiwan",		IDX_EUC_TW},
390     {"cp950",		IDX_BIG5},
391     {"950",		IDX_BIG5},
392 #endif
393     {"mac",		IDX_MACROMAN},
394     {"mac-roman",	IDX_MACROMAN},
395     {NULL,		0}
396 };
397 
398 #ifndef CP_UTF8
399 # define CP_UTF8 65001	/* magic number from winnls.h */
400 #endif
401 
402 /*
403  * Find encoding "name" in the list of canonical encoding names.
404  * Returns -1 if not found.
405  */
406     static int
407 enc_canon_search(name)
408     char_u	*name;
409 {
410     int		i;
411 
412     for (i = 0; i < IDX_COUNT; ++i)
413 	if (STRCMP(name, enc_canon_table[i].name) == 0)
414 	    return i;
415     return -1;
416 }
417 
418 #endif
419 
420 #if defined(FEAT_MBYTE) || defined(PROTO)
421 
422 /*
423  * Find canonical encoding "name" in the list and return its properties.
424  * Returns 0 if not found.
425  */
426     int
427 enc_canon_props(name)
428     char_u	*name;
429 {
430     int		i;
431 
432     i = enc_canon_search(name);
433     if (i >= 0)
434 	return enc_canon_table[i].prop;
435 #ifdef WIN3264
436     if (name[0] == 'c' && name[1] == 'p' && VIM_ISDIGIT(name[2]))
437     {
438 	CPINFO	cpinfo;
439 
440 	/* Get info on this codepage to find out what it is. */
441 	if (GetCPInfo(atoi(name + 2), &cpinfo) != 0)
442 	{
443 	    if (cpinfo.MaxCharSize == 1) /* some single-byte encoding */
444 		return ENC_8BIT;
445 	    if (cpinfo.MaxCharSize == 2
446 		    && (cpinfo.LeadByte[0] != 0 || cpinfo.LeadByte[1] != 0))
447 		/* must be a DBCS encoding */
448 		return ENC_DBCS;
449 	}
450 	return 0;
451     }
452 #endif
453     if (STRNCMP(name, "2byte-", 6) == 0)
454 	return ENC_DBCS;
455     if (STRNCMP(name, "8bit-", 5) == 0 || STRNCMP(name, "iso-8859-", 9) == 0)
456 	return ENC_8BIT;
457     return 0;
458 }
459 
460 /*
461  * Set up for using multi-byte characters.
462  * Called in three cases:
463  * - by main() to initialize (p_enc == NULL)
464  * - by set_init_1() after 'encoding' was set to its default.
465  * - by do_set() when 'encoding' has been set.
466  * p_enc must have been passed through enc_canonize() already.
467  * Sets the "enc_unicode", "enc_utf8", "enc_dbcs" and "has_mbyte" flags.
468  * Fills mb_bytelen_tab[] and returns NULL when there are no problems.
469  * When there is something wrong: Returns an error message and doesn't change
470  * anything.
471  */
472     char_u *
473 mb_init()
474 {
475     int		i;
476     int		idx;
477     int		n;
478     int		enc_dbcs_new = 0;
479 #if defined(USE_ICONV) && !defined(WIN3264) && !defined(WIN32UNIX) \
480 	&& !defined(MACOS)
481 # define LEN_FROM_CONV
482     vimconv_T	vimconv;
483     char_u	*p;
484 #endif
485 
486     if (p_enc == NULL)
487     {
488 	/* Just starting up: set the whole table to one's. */
489 	for (i = 0; i < 256; ++i)
490 	    mb_bytelen_tab[i] = 1;
491 	input_conv.vc_type = CONV_NONE;
492 	input_conv.vc_factor = 1;
493 	output_conv.vc_type = CONV_NONE;
494 	return NULL;
495     }
496 
497 #ifdef WIN3264
498     if (p_enc[0] == 'c' && p_enc[1] == 'p' && VIM_ISDIGIT(p_enc[2]))
499     {
500 	CPINFO	cpinfo;
501 
502 	/* Get info on this codepage to find out what it is. */
503 	if (GetCPInfo(atoi(p_enc + 2), &cpinfo) != 0)
504 	{
505 	    if (cpinfo.MaxCharSize == 1)
506 	    {
507 		/* some single-byte encoding */
508 		enc_unicode = 0;
509 		enc_utf8 = FALSE;
510 	    }
511 	    else if (cpinfo.MaxCharSize == 2
512 		    && (cpinfo.LeadByte[0] != 0 || cpinfo.LeadByte[1] != 0))
513 	    {
514 		/* must be a DBCS encoding, check below */
515 		enc_dbcs_new = atoi(p_enc + 2);
516 	    }
517 	    else
518 		goto codepage_invalid;
519 	}
520 	else if (GetLastError() == ERROR_INVALID_PARAMETER)
521 	{
522 codepage_invalid:
523 	    return (char_u *)N_("E543: Not a valid codepage");
524 	}
525     }
526 #endif
527     else if (STRNCMP(p_enc, "8bit-", 5) == 0
528 	    || STRNCMP(p_enc, "iso-8859-", 9) == 0)
529     {
530 	/* Accept any "8bit-" or "iso-8859-" name. */
531 	enc_unicode = 0;
532 	enc_utf8 = FALSE;
533     }
534     else if (STRNCMP(p_enc, "2byte-", 6) == 0)
535     {
536 #ifdef WIN3264
537 	/* Windows: accept only valid codepage numbers, check below. */
538 	if (p_enc[6] != 'c' || p_enc[7] != 'p'
539 				      || (enc_dbcs_new = atoi(p_enc + 8)) == 0)
540 	    return e_invarg;
541 #else
542 	/* Unix: accept any "2byte-" name, assume current locale. */
543 	enc_dbcs_new = DBCS_2BYTE;
544 #endif
545     }
546     else if ((idx = enc_canon_search(p_enc)) >= 0)
547     {
548 	i = enc_canon_table[idx].prop;
549 	if (i & ENC_UNICODE)
550 	{
551 	    /* Unicode */
552 	    enc_utf8 = TRUE;
553 	    if (i & (ENC_2BYTE | ENC_2WORD))
554 		enc_unicode = 2;
555 	    else if (i & ENC_4BYTE)
556 		enc_unicode = 4;
557 	    else
558 		enc_unicode = 0;
559 	}
560 	else if (i & ENC_DBCS)
561 	{
562 	    /* 2byte, handle below */
563 	    enc_dbcs_new = enc_canon_table[idx].codepage;
564 	}
565 	else
566 	{
567 	    /* Must be 8-bit. */
568 	    enc_unicode = 0;
569 	    enc_utf8 = FALSE;
570 	}
571     }
572     else    /* Don't know what encoding this is, reject it. */
573 	return e_invarg;
574 
575     if (enc_dbcs_new != 0)
576     {
577 #ifdef WIN3264
578 	/* Check if the DBCS code page is OK. */
579 	if (!IsValidCodePage(enc_dbcs_new))
580 	    goto codepage_invalid;
581 #endif
582 	enc_unicode = 0;
583 	enc_utf8 = FALSE;
584     }
585     enc_dbcs = enc_dbcs_new;
586     has_mbyte = (enc_dbcs != 0 || enc_utf8);
587 
588 #ifdef WIN3264
589     enc_codepage = encname2codepage(p_enc);
590     enc_latin9 = (STRCMP(p_enc, "iso-8859-15") == 0);
591 #endif
592 
593     /* Detect an encoding that uses latin1 characters. */
594     enc_latin1like = (enc_utf8 || STRCMP(p_enc, "latin1") == 0
595 					|| STRCMP(p_enc, "iso-8859-15") == 0);
596 
597     /*
598      * Set the function pointers.
599      */
600     if (enc_utf8)
601     {
602 	mb_ptr2len = utfc_ptr2len;
603 	mb_char2len = utf_char2len;
604 	mb_char2bytes = utf_char2bytes;
605 	mb_ptr2cells = utf_ptr2cells;
606 	mb_char2cells = utf_char2cells;
607 	mb_off2cells = utf_off2cells;
608 	mb_ptr2char = utf_ptr2char;
609 	mb_head_off = utf_head_off;
610     }
611     else if (enc_dbcs != 0)
612     {
613 	mb_ptr2len = dbcs_ptr2len;
614 	mb_char2len = dbcs_char2len;
615 	mb_char2bytes = dbcs_char2bytes;
616 	mb_ptr2cells = dbcs_ptr2cells;
617 	mb_char2cells = dbcs_char2cells;
618 	mb_off2cells = dbcs_off2cells;
619 	mb_ptr2char = dbcs_ptr2char;
620 	mb_head_off = dbcs_head_off;
621     }
622     else
623     {
624 	mb_ptr2len = latin_ptr2len;
625 	mb_char2len = latin_char2len;
626 	mb_char2bytes = latin_char2bytes;
627 	mb_ptr2cells = latin_ptr2cells;
628 	mb_char2cells = latin_char2cells;
629 	mb_off2cells = latin_off2cells;
630 	mb_ptr2char = latin_ptr2char;
631 	mb_head_off = latin_head_off;
632     }
633 
634     /*
635      * Fill the mb_bytelen_tab[] for MB_BYTE2LEN().
636      */
637 #ifdef LEN_FROM_CONV
638     /* When 'encoding' is different from the current locale mblen() won't
639      * work.  Use conversion to "utf-8" instead. */
640     vimconv.vc_type = CONV_NONE;
641     if (enc_dbcs)
642     {
643 	p = enc_locale();
644 	if (p == NULL || STRCMP(p, p_enc) != 0)
645 	{
646 	    convert_setup(&vimconv, p_enc, (char_u *)"utf-8");
647 	    vimconv.vc_fail = TRUE;
648 	}
649 	vim_free(p);
650     }
651 #endif
652 
653     for (i = 0; i < 256; ++i)
654     {
655 	/* Our own function to reliably check the length of UTF-8 characters,
656 	 * independent of mblen(). */
657 	if (enc_utf8)
658 	    n = utf8len_tab[i];
659 	else if (enc_dbcs == 0)
660 	    n = 1;
661 	else
662 	{
663 #if defined(WIN3264) || defined(WIN32UNIX)
664 	    /* enc_dbcs is set by setting 'fileencoding'.  It becomes a Windows
665 	     * CodePage identifier, which we can pass directly in to Windows
666 	     * API */
667 	    n = IsDBCSLeadByteEx(enc_dbcs, (BYTE)i) ? 2 : 1;
668 #else
669 # if defined(MACOS) || defined(__amigaos4__)
670 	    /*
671 	     * if mblen() is not available, character which MSB is turned on
672 	     * are treated as leading byte character. (note : This assumption
673 	     * is not always true.)
674 	     */
675 	    n = (i & 0x80) ? 2 : 1;
676 # else
677 	    char buf[MB_MAXBYTES];
678 # ifdef X_LOCALE
679 #  ifndef mblen
680 #   define mblen _Xmblen
681 #  endif
682 # endif
683 	    if (i == NUL)	/* just in case mblen() can't handle "" */
684 		n = 1;
685 	    else
686 	    {
687 		buf[0] = i;
688 		buf[1] = 0;
689 #ifdef LEN_FROM_CONV
690 		if (vimconv.vc_type != CONV_NONE)
691 		{
692 		    /*
693 		     * string_convert() should fail when converting the first
694 		     * byte of a double-byte character.
695 		     */
696 		    p = string_convert(&vimconv, (char_u *)buf, NULL);
697 		    if (p != NULL)
698 		    {
699 			vim_free(p);
700 			n = 1;
701 		    }
702 		    else
703 			n = 2;
704 		}
705 		else
706 #endif
707 		{
708 		    /*
709 		     * mblen() should return -1 for invalid (means the leading
710 		     * multibyte) character.  However there are some platforms
711 		     * where mblen() returns 0 for invalid character.
712 		     * Therefore, following condition includes 0.
713 		     */
714 		    (void)mblen(NULL, 0);	/* First reset the state. */
715 		    if (mblen(buf, (size_t)1) <= 0)
716 			n = 2;
717 		    else
718 			n = 1;
719 		}
720 	    }
721 # endif
722 #endif
723 	}
724 
725 	mb_bytelen_tab[i] = n;
726     }
727 
728 #ifdef LEN_FROM_CONV
729     convert_setup(&vimconv, NULL, NULL);
730 #endif
731 
732     /* The cell width depends on the type of multi-byte characters. */
733     (void)init_chartab();
734 
735     /* When enc_utf8 is set or reset, (de)allocate ScreenLinesUC[] */
736     screenalloc(FALSE);
737 
738     /* When using Unicode, set default for 'fileencodings'. */
739     if (enc_utf8 && !option_was_set((char_u *)"fencs"))
740 	set_string_option_direct((char_u *)"fencs", -1,
741 		       (char_u *)"ucs-bom,utf-8,default,latin1", OPT_FREE, 0);
742 
743 #if defined(HAVE_BIND_TEXTDOMAIN_CODESET) && defined(FEAT_GETTEXT)
744     /* GNU gettext 0.10.37 supports this feature: set the codeset used for
745      * translated messages independently from the current locale. */
746     (void)bind_textdomain_codeset(VIMPACKAGE,
747 					  enc_utf8 ? "utf-8" : (char *)p_enc);
748 #endif
749 
750 #ifdef WIN32
751     /* When changing 'encoding' while starting up, then convert the command
752      * line arguments from the active codepage to 'encoding'. */
753     if (starting != 0)
754 	fix_arg_enc();
755 #endif
756 
757 #ifdef FEAT_AUTOCMD
758     /* Fire an autocommand to let people do custom font setup. This must be
759      * after Vim has been setup for the new encoding. */
760     apply_autocmds(EVENT_ENCODINGCHANGED, NULL, (char_u *)"", FALSE, curbuf);
761 #endif
762 
763 #ifdef FEAT_SPELL
764     /* Need to reload spell dictionaries */
765     spell_reload();
766 #endif
767 
768     return NULL;
769 }
770 
771 /*
772  * Return the size of the BOM for the current buffer:
773  * 0 - no BOM
774  * 2 - UCS-2 or UTF-16 BOM
775  * 4 - UCS-4 BOM
776  * 3 - UTF-8 BOM
777  */
778     int
779 bomb_size()
780 {
781     int n = 0;
782 
783     if (curbuf->b_p_bomb && !curbuf->b_p_bin)
784     {
785 	if (*curbuf->b_p_fenc == NUL)
786 	{
787 	    if (enc_utf8)
788 	    {
789 		if (enc_unicode != 0)
790 		    n = enc_unicode;
791 		else
792 		    n = 3;
793 	    }
794 	}
795 	else if (STRCMP(curbuf->b_p_fenc, "utf-8") == 0)
796 	    n = 3;
797 	else if (STRNCMP(curbuf->b_p_fenc, "ucs-2", 5) == 0
798 		|| STRNCMP(curbuf->b_p_fenc, "utf-16", 6) == 0)
799 	    n = 2;
800 	else if (STRNCMP(curbuf->b_p_fenc, "ucs-4", 5) == 0)
801 	    n = 4;
802     }
803     return n;
804 }
805 
806 /*
807  * Get class of pointer:
808  * 0 for blank or NUL
809  * 1 for punctuation
810  * 2 for an (ASCII) word character
811  * >2 for other word characters
812  */
813     int
814 mb_get_class(p)
815     char_u	*p;
816 {
817     if (MB_BYTE2LEN(p[0]) == 1)
818     {
819 	if (p[0] == NUL || vim_iswhite(p[0]))
820 	    return 0;
821 	if (vim_iswordc(p[0]))
822 	    return 2;
823 	return 1;
824     }
825     if (enc_dbcs != 0 && p[0] != NUL && p[1] != NUL)
826 	return dbcs_class(p[0], p[1]);
827     if (enc_utf8)
828 	return utf_class(utf_ptr2char(p));
829     return 0;
830 }
831 
832 /*
833  * Get class of a double-byte character.  This always returns 3 or bigger.
834  * TODO: Should return 1 for punctuation.
835  */
836     int
837 dbcs_class(lead, trail)
838     unsigned	lead;
839     unsigned	trail;
840 {
841     switch (enc_dbcs)
842     {
843 	/* please add classfy routine for your language in here */
844 
845 	case DBCS_JPNU:	/* ? */
846 	case DBCS_JPN:
847 	    {
848 		/* JIS code classification */
849 		unsigned char lb = lead;
850 		unsigned char tb = trail;
851 
852 		/* convert process code to JIS */
853 # if defined(WIN3264) || defined(WIN32UNIX) || defined(MACOS)
854 		/* process code is SJIS */
855 		if (lb <= 0x9f)
856 		    lb = (lb - 0x81) * 2 + 0x21;
857 		else
858 		    lb = (lb - 0xc1) * 2 + 0x21;
859 		if (tb <= 0x7e)
860 		    tb -= 0x1f;
861 		else if (tb <= 0x9e)
862 		    tb -= 0x20;
863 		else
864 		{
865 		    tb -= 0x7e;
866 		    lb += 1;
867 		}
868 # else
869 		/*
870 		 * XXX: Code page identification can not use with all
871 		 *	    system! So, some other encoding information
872 		 *	    will be needed.
873 		 *	    In japanese: SJIS,EUC,UNICODE,(JIS)
874 		 *	    Note that JIS-code system don't use as
875 		 *	    process code in most system because it uses
876 		 *	    escape sequences(JIS is context depend encoding).
877 		 */
878 		/* assume process code is JAPANESE-EUC */
879 		lb &= 0x7f;
880 		tb &= 0x7f;
881 # endif
882 		/* exceptions */
883 		switch (lb << 8 | tb)
884 		{
885 		    case 0x2121: /* ZENKAKU space */
886 			return 0;
887 		    case 0x2122: /* KU-TEN (Japanese comma) */
888 		    case 0x2123: /* TOU-TEN (Japanese period) */
889 		    case 0x2124: /* ZENKAKU comma */
890 		    case 0x2125: /* ZENKAKU period */
891 			return 1;
892 		    case 0x213c: /* prolongedsound handled as KATAKANA */
893 			return 13;
894 		}
895 		/* sieved by KU code */
896 		switch (lb)
897 		{
898 		    case 0x21:
899 		    case 0x22:
900 			/* special symbols */
901 			return 10;
902 		    case 0x23:
903 			/* alpha-numeric */
904 			return 11;
905 		    case 0x24:
906 			/* hiragana */
907 			return 12;
908 		    case 0x25:
909 			/* katakana */
910 			return 13;
911 		    case 0x26:
912 			/* greek */
913 			return 14;
914 		    case 0x27:
915 			/* russian */
916 			return 15;
917 		    case 0x28:
918 			/* lines */
919 			return 16;
920 		    default:
921 			/* kanji */
922 			return 17;
923 		}
924 	    }
925 
926 	case DBCS_KORU:	/* ? */
927 	case DBCS_KOR:
928 	    {
929 		/* KS code classification */
930 		unsigned char c1 = lead;
931 		unsigned char c2 = trail;
932 
933 		/*
934 		 * 20 : Hangul
935 		 * 21 : Hanja
936 		 * 22 : Symbols
937 		 * 23 : Alpha-numeric/Roman Letter (Full width)
938 		 * 24 : Hangul Letter(Alphabet)
939 		 * 25 : Roman Numeral/Greek Letter
940 		 * 26 : Box Drawings
941 		 * 27 : Unit Symbols
942 		 * 28 : Circled/Parenthesized Letter
943 		 * 29 : Hirigana/Katakana
944 		 * 30 : Cyrillic Letter
945 		 */
946 
947 		if (c1 >= 0xB0 && c1 <= 0xC8)
948 		    /* Hangul */
949 		    return 20;
950 #if defined(WIN3264) || defined(WIN32UNIX)
951 		else if (c1 <= 0xA0 || c2 <= 0xA0)
952 		    /* Extended Hangul Region : MS UHC(Unified Hangul Code) */
953 		    /* c1: 0x81-0xA0 with c2: 0x41-0x5A, 0x61-0x7A, 0x81-0xFE
954 		     * c1: 0xA1-0xC6 with c2: 0x41-0x5A, 0x61-0x7A, 0x81-0xA0
955 		     */
956 		    return 20;
957 #endif
958 
959 		else if (c1 >= 0xCA && c1 <= 0xFD)
960 		    /* Hanja */
961 		    return 21;
962 		else switch (c1)
963 		{
964 		    case 0xA1:
965 		    case 0xA2:
966 			/* Symbols */
967 			return 22;
968 		    case 0xA3:
969 			/* Alpha-numeric */
970 			return 23;
971 		    case 0xA4:
972 			/* Hangul Letter(Alphabet) */
973 			return 24;
974 		    case 0xA5:
975 			/* Roman Numeral/Greek Letter */
976 			return 25;
977 		    case 0xA6:
978 			/* Box Drawings */
979 			return 26;
980 		    case 0xA7:
981 			/* Unit Symbols */
982 			return 27;
983 		    case 0xA8:
984 		    case 0xA9:
985 			if (c2 <= 0xAF)
986 			    return 25;  /* Roman Letter */
987 			else if (c2 >= 0xF6)
988 			    return 22;  /* Symbols */
989 			else
990 			    /* Circled/Parenthesized Letter */
991 			    return 28;
992 		    case 0xAA:
993 		    case 0xAB:
994 			/* Hirigana/Katakana */
995 			return 29;
996 		    case 0xAC:
997 			/* Cyrillic Letter */
998 			return 30;
999 		}
1000 	    }
1001 	default:
1002 	    break;
1003     }
1004     return 3;
1005 }
1006 
1007 /*
1008  * mb_char2len() function pointer.
1009  * Return length in bytes of character "c".
1010  * Returns 1 for a single-byte character.
1011  */
1012 /* ARGSUSED */
1013     int
1014 latin_char2len(c)
1015     int		c;
1016 {
1017     return 1;
1018 }
1019 
1020     static int
1021 dbcs_char2len(c)
1022     int		c;
1023 {
1024     if (c >= 0x100)
1025 	return 2;
1026     return 1;
1027 }
1028 
1029 /*
1030  * mb_char2bytes() function pointer.
1031  * Convert a character to its bytes.
1032  * Returns the length in bytes.
1033  */
1034     int
1035 latin_char2bytes(c, buf)
1036     int		c;
1037     char_u	*buf;
1038 {
1039     buf[0] = c;
1040     return 1;
1041 }
1042 
1043     static int
1044 dbcs_char2bytes(c, buf)
1045     int		c;
1046     char_u	*buf;
1047 {
1048     if (c >= 0x100)
1049     {
1050 	buf[0] = (unsigned)c >> 8;
1051 	buf[1] = c;
1052 	/* Never use a NUL byte, it causes lots of trouble.  It's an invalid
1053 	 * character anyway. */
1054 	if (buf[1] == NUL)
1055 	    buf[1] = '\n';
1056 	return 2;
1057     }
1058     buf[0] = c;
1059     return 1;
1060 }
1061 
1062 /*
1063  * mb_ptr2len() function pointer.
1064  * Get byte length of character at "*p" but stop at a NUL.
1065  * For UTF-8 this includes following composing characters.
1066  * Returns 0 when *p is NUL.
1067  *
1068  */
1069     int
1070 latin_ptr2len(p)
1071     char_u	*p;
1072 {
1073     return MB_BYTE2LEN(*p);
1074 }
1075 
1076     static int
1077 dbcs_ptr2len(p)
1078     char_u	*p;
1079 {
1080     int		len;
1081 
1082     /* Check if second byte is not missing. */
1083     len = MB_BYTE2LEN(*p);
1084     if (len == 2 && p[1] == NUL)
1085 	len = 1;
1086     return len;
1087 }
1088 
1089 struct interval
1090 {
1091     unsigned short first;
1092     unsigned short last;
1093 };
1094 static int intable __ARGS((struct interval *table, size_t size, int c));
1095 
1096 /*
1097  * Return TRUE if "c" is in "table[size / sizeof(struct interval)]".
1098  */
1099     static int
1100 intable(table, size, c)
1101     struct interval	*table;
1102     size_t		size;
1103     int			c;
1104 {
1105     int mid, bot, top;
1106 
1107     /* first quick check for Latin1 etc. characters */
1108     if (c < table[0].first)
1109 	return FALSE;
1110 
1111     /* binary search in table */
1112     bot = 0;
1113     top = (int)(size / sizeof(struct interval) - 1);
1114     while (top >= bot)
1115     {
1116 	mid = (bot + top) / 2;
1117 	if (table[mid].last < c)
1118 	    bot = mid + 1;
1119 	else if (table[mid].first > c)
1120 	    top = mid - 1;
1121 	else
1122 	    return TRUE;
1123     }
1124     return FALSE;
1125 }
1126 
1127 /*
1128  * For UTF-8 character "c" return 2 for a double-width character, 1 for others.
1129  * Returns 4 or 6 for an unprintable character.
1130  * Is only correct for characters >= 0x80.
1131  * When p_ambw is "double", return 2 for a character with East Asian Width
1132  * class 'A'(mbiguous).
1133  */
1134     int
1135 utf_char2cells(c)
1136     int		c;
1137 {
1138     /* sorted list of non-overlapping intervals of East Asian Ambiguous
1139      * characters, generated with:
1140      * "uniset +WIDTH-A -cat=Me -cat=Mn -cat=Cf c" */
1141     static struct interval ambiguous[] = {
1142 	{0x00A1, 0x00A1}, {0x00A4, 0x00A4}, {0x00A7, 0x00A8},
1143 	{0x00AA, 0x00AA}, {0x00AE, 0x00AE}, {0x00B0, 0x00B4},
1144 	{0x00B6, 0x00BA}, {0x00BC, 0x00BF}, {0x00C6, 0x00C6},
1145 	{0x00D0, 0x00D0}, {0x00D7, 0x00D8}, {0x00DE, 0x00E1},
1146 	{0x00E6, 0x00E6}, {0x00E8, 0x00EA}, {0x00EC, 0x00ED},
1147 	{0x00F0, 0x00F0}, {0x00F2, 0x00F3}, {0x00F7, 0x00FA},
1148 	{0x00FC, 0x00FC}, {0x00FE, 0x00FE}, {0x0101, 0x0101},
1149 	{0x0111, 0x0111}, {0x0113, 0x0113}, {0x011B, 0x011B},
1150 	{0x0126, 0x0127}, {0x012B, 0x012B}, {0x0131, 0x0133},
1151 	{0x0138, 0x0138}, {0x013F, 0x0142}, {0x0144, 0x0144},
1152 	{0x0148, 0x014B}, {0x014D, 0x014D}, {0x0152, 0x0153},
1153 	{0x0166, 0x0167}, {0x016B, 0x016B}, {0x01CE, 0x01CE},
1154 	{0x01D0, 0x01D0}, {0x01D2, 0x01D2}, {0x01D4, 0x01D4},
1155 	{0x01D6, 0x01D6}, {0x01D8, 0x01D8}, {0x01DA, 0x01DA},
1156 	{0x01DC, 0x01DC}, {0x0251, 0x0251}, {0x0261, 0x0261},
1157 	{0x02C4, 0x02C4}, {0x02C7, 0x02C7}, {0x02C9, 0x02CB},
1158 	{0x02CD, 0x02CD}, {0x02D0, 0x02D0}, {0x02D8, 0x02DB},
1159 	{0x02DD, 0x02DD}, {0x02DF, 0x02DF}, {0x0391, 0x03A1},
1160 	{0x03A3, 0x03A9}, {0x03B1, 0x03C1}, {0x03C3, 0x03C9},
1161 	{0x0401, 0x0401}, {0x0410, 0x044F}, {0x0451, 0x0451},
1162 	{0x2010, 0x2010}, {0x2013, 0x2016}, {0x2018, 0x2019},
1163 	{0x201C, 0x201D}, {0x2020, 0x2022}, {0x2024, 0x2027},
1164 	{0x2030, 0x2030}, {0x2032, 0x2033}, {0x2035, 0x2035},
1165 	{0x203B, 0x203B}, {0x203E, 0x203E}, {0x2074, 0x2074},
1166 	{0x207F, 0x207F}, {0x2081, 0x2084}, {0x20AC, 0x20AC},
1167 	{0x2103, 0x2103}, {0x2105, 0x2105}, {0x2109, 0x2109},
1168 	{0x2113, 0x2113}, {0x2116, 0x2116}, {0x2121, 0x2122},
1169 	{0x2126, 0x2126}, {0x212B, 0x212B}, {0x2153, 0x2154},
1170 	{0x215B, 0x215E}, {0x2160, 0x216B}, {0x2170, 0x2179},
1171 	{0x2190, 0x2199}, {0x21B8, 0x21B9}, {0x21D2, 0x21D2},
1172 	{0x21D4, 0x21D4}, {0x21E7, 0x21E7}, {0x2200, 0x2200},
1173 	{0x2202, 0x2203}, {0x2207, 0x2208}, {0x220B, 0x220B},
1174 	{0x220F, 0x220F}, {0x2211, 0x2211}, {0x2215, 0x2215},
1175 	{0x221A, 0x221A}, {0x221D, 0x2220}, {0x2223, 0x2223},
1176 	{0x2225, 0x2225}, {0x2227, 0x222C}, {0x222E, 0x222E},
1177 	{0x2234, 0x2237}, {0x223C, 0x223D}, {0x2248, 0x2248},
1178 	{0x224C, 0x224C}, {0x2252, 0x2252}, {0x2260, 0x2261},
1179 	{0x2264, 0x2267}, {0x226A, 0x226B}, {0x226E, 0x226F},
1180 	{0x2282, 0x2283}, {0x2286, 0x2287}, {0x2295, 0x2295},
1181 	{0x2299, 0x2299}, {0x22A5, 0x22A5}, {0x22BF, 0x22BF},
1182 	{0x2312, 0x2312}, {0x2460, 0x24E9}, {0x24EB, 0x254B},
1183 	{0x2550, 0x2573}, {0x2580, 0x258F}, {0x2592, 0x2595},
1184 	{0x25A0, 0x25A1}, {0x25A3, 0x25A9}, {0x25B2, 0x25B3},
1185 	{0x25B6, 0x25B7}, {0x25BC, 0x25BD}, {0x25C0, 0x25C1},
1186 	{0x25C6, 0x25C8}, {0x25CB, 0x25CB}, {0x25CE, 0x25D1},
1187 	{0x25E2, 0x25E5}, {0x25EF, 0x25EF}, {0x2605, 0x2606},
1188 	{0x2609, 0x2609}, {0x260E, 0x260F}, {0x2614, 0x2615},
1189 	{0x261C, 0x261C}, {0x261E, 0x261E}, {0x2640, 0x2640},
1190 	{0x2642, 0x2642}, {0x2660, 0x2661}, {0x2663, 0x2665},
1191 	{0x2667, 0x266A}, {0x266C, 0x266D}, {0x266F, 0x266F},
1192 	{0x273D, 0x273D}, {0x2776, 0x277F}, {0xE000, 0xF8FF},
1193 	{0xFFFD, 0xFFFD}, /* {0xF0000, 0xFFFFD}, {0x100000, 0x10FFFD} */
1194     };
1195 
1196     if (c >= 0x100)
1197     {
1198 #ifdef USE_WCHAR_FUNCTIONS
1199 	/*
1200 	 * Assume the library function wcwidth() works better than our own
1201 	 * stuff.  It should return 1 for ambiguous width chars!
1202 	 */
1203 	int	n = wcwidth(c);
1204 
1205 	if (n < 0)
1206 	    return 6;		/* unprintable, displays <xxxx> */
1207 	if (n > 1)
1208 	    return n;
1209 #else
1210 	if (!utf_printable(c))
1211 	    return 6;		/* unprintable, displays <xxxx> */
1212 	if (c >= 0x1100
1213 	    && (c <= 0x115f			/* Hangul Jamo */
1214 		|| c == 0x2329
1215 		|| c == 0x232a
1216 		|| (c >= 0x2e80 && c <= 0xa4cf
1217 		    && c != 0x303f)		/* CJK ... Yi */
1218 		|| (c >= 0xac00 && c <= 0xd7a3)	/* Hangul Syllables */
1219 		|| (c >= 0xf900 && c <= 0xfaff)	/* CJK Compatibility
1220 						   Ideographs */
1221 		|| (c >= 0xfe30 && c <= 0xfe6f)	/* CJK Compatibility Forms */
1222 		|| (c >= 0xff00 && c <= 0xff60)	/* Fullwidth Forms */
1223 		|| (c >= 0xffe0 && c <= 0xffe6)
1224 		|| (c >= 0x20000 && c <= 0x2fffd)
1225 		|| (c >= 0x30000 && c <= 0x3fffd)))
1226 	    return 2;
1227 #endif
1228     }
1229 
1230     /* Characters below 0x100 are influenced by 'isprint' option */
1231     else if (c >= 0x80 && !vim_isprintc(c))
1232 	return 4;		/* unprintable, displays <xx> */
1233 
1234     if (c >= 0x80 && *p_ambw == 'd' && intable(ambiguous, sizeof(ambiguous), c))
1235 	return 2;
1236 
1237     return 1;
1238 }
1239 
1240 /*
1241  * mb_ptr2cells() function pointer.
1242  * Return the number of display cells character at "*p" occupies.
1243  * This doesn't take care of unprintable characters, use ptr2cells() for that.
1244  */
1245 /*ARGSUSED*/
1246     int
1247 latin_ptr2cells(p)
1248     char_u	*p;
1249 {
1250     return 1;
1251 }
1252 
1253     int
1254 utf_ptr2cells(p)
1255     char_u	*p;
1256 {
1257     int		c;
1258 
1259     /* Need to convert to a wide character. */
1260     if (*p >= 0x80)
1261     {
1262 	c = utf_ptr2char(p);
1263 	/* An illegal byte is displayed as <xx>. */
1264 	if (utf_ptr2len(p) == 1 || c == NUL)
1265 	    return 4;
1266 	/* If the char is ASCII it must be an overlong sequence. */
1267 	if (c < 0x80)
1268 	    return char2cells(c);
1269 	return utf_char2cells(c);
1270     }
1271     return 1;
1272 }
1273 
1274     int
1275 dbcs_ptr2cells(p)
1276     char_u	*p;
1277 {
1278     /* Number of cells is equal to number of bytes, except for euc-jp when
1279      * the first byte is 0x8e. */
1280     if (enc_dbcs == DBCS_JPNU && *p == 0x8e)
1281 	return 1;
1282     return MB_BYTE2LEN(*p);
1283 }
1284 
1285 /*
1286  * mb_char2cells() function pointer.
1287  * Return the number of display cells character "c" occupies.
1288  * Only takes care of multi-byte chars, not "^C" and such.
1289  */
1290 /*ARGSUSED*/
1291     int
1292 latin_char2cells(c)
1293     int		c;
1294 {
1295     return 1;
1296 }
1297 
1298     static int
1299 dbcs_char2cells(c)
1300     int		c;
1301 {
1302     /* Number of cells is equal to number of bytes, except for euc-jp when
1303      * the first byte is 0x8e. */
1304     if (enc_dbcs == DBCS_JPNU && ((unsigned)c >> 8) == 0x8e)
1305 	return 1;
1306     /* use the first byte */
1307     return MB_BYTE2LEN((unsigned)c >> 8);
1308 }
1309 
1310 /*
1311  * mb_off2cells() function pointer.
1312  * Return number of display cells for char at ScreenLines[off].
1313  * Caller must make sure "off" and "off + 1" are valid!
1314  */
1315 /*ARGSUSED*/
1316     int
1317 latin_off2cells(off)
1318     unsigned	off;
1319 {
1320     return 1;
1321 }
1322 
1323     int
1324 dbcs_off2cells(off)
1325     unsigned	off;
1326 {
1327     /* Number of cells is equal to number of bytes, except for euc-jp when
1328      * the first byte is 0x8e. */
1329     if (enc_dbcs == DBCS_JPNU && ScreenLines[off] == 0x8e)
1330 	return 1;
1331     return MB_BYTE2LEN(ScreenLines[off]);
1332 }
1333 
1334     int
1335 utf_off2cells(off)
1336     unsigned	off;
1337 {
1338     return ScreenLines[off + 1] == 0 ? 2 : 1;
1339 }
1340 
1341 /*
1342  * mb_ptr2char() function pointer.
1343  * Convert a byte sequence into a character.
1344  */
1345     int
1346 latin_ptr2char(p)
1347     char_u	*p;
1348 {
1349     return *p;
1350 }
1351 
1352     static int
1353 dbcs_ptr2char(p)
1354     char_u	*p;
1355 {
1356     if (MB_BYTE2LEN(*p) > 1 && p[1] != NUL)
1357 	return (p[0] << 8) + p[1];
1358     return *p;
1359 }
1360 
1361 /*
1362  * Convert a UTF-8 byte sequence to a wide character.
1363  * If the sequence is illegal or truncated by a NUL the first byte is
1364  * returned.
1365  * Does not include composing characters, of course.
1366  */
1367     int
1368 utf_ptr2char(p)
1369     char_u	*p;
1370 {
1371     int		len;
1372 
1373     if (p[0] < 0x80)	/* be quick for ASCII */
1374 	return p[0];
1375 
1376     len = utf8len_tab[p[0]];
1377     if ((p[1] & 0xc0) == 0x80)
1378     {
1379 	if (len == 2)
1380 	    return ((p[0] & 0x1f) << 6) + (p[1] & 0x3f);
1381 	if ((p[2] & 0xc0) == 0x80)
1382 	{
1383 	    if (len == 3)
1384 		return ((p[0] & 0x0f) << 12) + ((p[1] & 0x3f) << 6)
1385 		    + (p[2] & 0x3f);
1386 	    if ((p[3] & 0xc0) == 0x80)
1387 	    {
1388 		if (len == 4)
1389 		    return ((p[0] & 0x07) << 18) + ((p[1] & 0x3f) << 12)
1390 			+ ((p[2] & 0x3f) << 6) + (p[3] & 0x3f);
1391 		if ((p[4] & 0xc0) == 0x80)
1392 		{
1393 		    if (len == 5)
1394 			return ((p[0] & 0x03) << 24) + ((p[1] & 0x3f) << 18)
1395 			    + ((p[2] & 0x3f) << 12) + ((p[3] & 0x3f) << 6)
1396 			    + (p[4] & 0x3f);
1397 		    if ((p[5] & 0xc0) == 0x80 && len == 6)
1398 			return ((p[0] & 0x01) << 30) + ((p[1] & 0x3f) << 24)
1399 			    + ((p[2] & 0x3f) << 18) + ((p[3] & 0x3f) << 12)
1400 			    + ((p[4] & 0x3f) << 6) + (p[5] & 0x3f);
1401 		}
1402 	    }
1403 	}
1404     }
1405     /* Illegal value, just return the first byte */
1406     return p[0];
1407 }
1408 
1409 /*
1410  * Get character at **pp and advance *pp to the next character.
1411  * Note: composing characters are skipped!
1412  */
1413     int
1414 mb_ptr2char_adv(pp)
1415     char_u	**pp;
1416 {
1417     int		c;
1418 
1419     c = (*mb_ptr2char)(*pp);
1420     *pp += (*mb_ptr2len)(*pp);
1421     return c;
1422 }
1423 
1424 /*
1425  * Get character at **pp and advance *pp to the next character.
1426  * Note: composing characters are returned as separate characters.
1427  */
1428     int
1429 mb_cptr2char_adv(pp)
1430     char_u	**pp;
1431 {
1432     int		c;
1433 
1434     c = (*mb_ptr2char)(*pp);
1435     if (enc_utf8)
1436 	*pp += utf_ptr2len(*pp);
1437     else
1438 	*pp += (*mb_ptr2len)(*pp);
1439     return c;
1440 }
1441 
1442 #if defined(FEAT_ARABIC) || defined(PROTO)
1443 /*
1444  * Check whether we are dealing with Arabic combining characters.
1445  * Note: these are NOT really composing characters!
1446  */
1447     int
1448 arabic_combine(one, two)
1449     int		one;	    /* first character */
1450     int		two;	    /* character just after "one" */
1451 {
1452     if (one == a_LAM)
1453 	return arabic_maycombine(two);
1454     return FALSE;
1455 }
1456 
1457 /*
1458  * Check whether we are dealing with a character that could be regarded as an
1459  * Arabic combining character, need to check the character before this.
1460  */
1461     int
1462 arabic_maycombine(two)
1463     int		two;
1464 {
1465     if (p_arshape && !p_tbidi)
1466 	return (two == a_ALEF_MADDA
1467 		    || two == a_ALEF_HAMZA_ABOVE
1468 		    || two == a_ALEF_HAMZA_BELOW
1469 		    || two == a_ALEF);
1470     return FALSE;
1471 }
1472 
1473 /*
1474  * Check if the character pointed to by "p2" is a composing character when it
1475  * comes after "p1".  For Arabic sometimes "ab" is replaced with "c", which
1476  * behaves like a composing character.
1477  */
1478     int
1479 utf_composinglike(p1, p2)
1480     char_u	*p1;
1481     char_u	*p2;
1482 {
1483     int		c2;
1484 
1485     c2 = utf_ptr2char(p2);
1486     if (utf_iscomposing(c2))
1487 	return TRUE;
1488     if (!arabic_maycombine(c2))
1489 	return FALSE;
1490     return arabic_combine(utf_ptr2char(p1), c2);
1491 }
1492 #endif
1493 
1494 /*
1495  * Convert a UTF-8 byte string to a wide chararacter.  Also get up to MAX_MCO
1496  * composing characters.
1497  */
1498     int
1499 utfc_ptr2char(p, pcc)
1500     char_u	*p;
1501     int		*pcc;	/* return: composing chars, last one is 0 */
1502 {
1503     int		len;
1504     int		c;
1505     int		cc;
1506     int		i = 0;
1507 
1508     c = utf_ptr2char(p);
1509     len = utf_ptr2len(p);
1510 
1511     /* Only accept a composing char when the first char isn't illegal. */
1512     if ((len > 1 || *p < 0x80)
1513 	    && p[len] >= 0x80
1514 	    && UTF_COMPOSINGLIKE(p, p + len))
1515     {
1516 	cc = utf_ptr2char(p + len);
1517 	for (;;)
1518 	{
1519 	    pcc[i++] = cc;
1520 	    if (i == MAX_MCO)
1521 		break;
1522 	    len += utf_ptr2len(p + len);
1523 	    if (p[len] < 0x80 || !utf_iscomposing(cc = utf_ptr2char(p + len)))
1524 		break;
1525 	}
1526     }
1527 
1528     if (i < MAX_MCO)	/* last composing char must be 0 */
1529 	pcc[i] = 0;
1530 
1531     return c;
1532 }
1533 
1534 /*
1535  * Convert a UTF-8 byte string to a wide chararacter.  Also get up to MAX_MCO
1536  * composing characters.  Use no more than p[maxlen].
1537  */
1538     int
1539 utfc_ptr2char_len(p, pcc, maxlen)
1540     char_u	*p;
1541     int		*pcc;	/* return: composing chars, last one is 0 */
1542     int		maxlen;
1543 {
1544     int		len;
1545     int		c;
1546     int		cc;
1547     int		i = 0;
1548 
1549     c = utf_ptr2char(p);
1550     len = utf_ptr2len_len(p, maxlen);
1551     /* Only accept a composing char when the first char isn't illegal. */
1552     if ((len > 1 || *p < 0x80)
1553 	    && len < maxlen
1554 	    && p[len] >= 0x80
1555 	    && UTF_COMPOSINGLIKE(p, p + len))
1556     {
1557 	cc = utf_ptr2char(p + len);
1558 	for (;;)
1559 	{
1560 	    pcc[i++] = cc;
1561 	    if (i == MAX_MCO)
1562 		break;
1563 	    len += utf_ptr2len_len(p + len, maxlen - len);
1564 	    if (len >= maxlen
1565 		    || p[len] < 0x80
1566 		    || !utf_iscomposing(cc = utf_ptr2char(p + len)))
1567 		break;
1568 	}
1569     }
1570 
1571     if (i < MAX_MCO)	/* last composing char must be 0 */
1572 	pcc[i] = 0;
1573 
1574     return c;
1575 }
1576 
1577 /*
1578  * Convert the character at screen position "off" to a sequence of bytes.
1579  * Includes the composing characters.
1580  * "buf" must at least have the length MB_MAXBYTES.
1581  * Returns the produced number of bytes.
1582  */
1583     int
1584 utfc_char2bytes(off, buf)
1585     int		off;
1586     char_u	*buf;
1587 {
1588     int		len;
1589     int		i;
1590 
1591     len = utf_char2bytes(ScreenLinesUC[off], buf);
1592     for (i = 0; i < Screen_mco; ++i)
1593     {
1594 	if (ScreenLinesC[i][off] == 0)
1595 	    break;
1596 	len += utf_char2bytes(ScreenLinesC[i][off], buf + len);
1597     }
1598     return len;
1599 }
1600 
1601 /*
1602  * Get the length of a UTF-8 byte sequence, not including any following
1603  * composing characters.
1604  * Returns 0 for "".
1605  * Returns 1 for an illegal byte sequence.
1606  */
1607     int
1608 utf_ptr2len(p)
1609     char_u	*p;
1610 {
1611     int		len;
1612     int		i;
1613 
1614     if (*p == NUL)
1615 	return 0;
1616     len = utf8len_tab[*p];
1617     for (i = 1; i < len; ++i)
1618 	if ((p[i] & 0xc0) != 0x80)
1619 	    return 1;
1620     return len;
1621 }
1622 
1623 /*
1624  * Return length of UTF-8 character, obtained from the first byte.
1625  * "b" must be between 0 and 255!
1626  */
1627     int
1628 utf_byte2len(b)
1629     int		b;
1630 {
1631     return utf8len_tab[b];
1632 }
1633 
1634 /*
1635  * Get the length of UTF-8 byte sequence "p[size]".  Does not include any
1636  * following composing characters.
1637  * Returns 1 for "".
1638  * Returns 1 for an illegal byte sequence.
1639  * Returns number > "size" for an incomplete byte sequence.
1640  */
1641     int
1642 utf_ptr2len_len(p, size)
1643     char_u	*p;
1644     int		size;
1645 {
1646     int		len;
1647     int		i;
1648 
1649     if (*p == NUL)
1650 	return 1;
1651     len = utf8len_tab[*p];
1652     if (len > size)
1653 	return len;	/* incomplete byte sequence. */
1654     for (i = 1; i < len; ++i)
1655 	if ((p[i] & 0xc0) != 0x80)
1656 	    return 1;
1657     return len;
1658 }
1659 
1660 /*
1661  * Return the number of bytes the UTF-8 encoding of the character at "p" takes.
1662  * This includes following composing characters.
1663  */
1664     int
1665 utfc_ptr2len(p)
1666     char_u	*p;
1667 {
1668     int		len;
1669     int		b0 = *p;
1670 #ifdef FEAT_ARABIC
1671     int		prevlen;
1672 #endif
1673 
1674     if (b0 == NUL)
1675 	return 0;
1676     if (b0 < 0x80 && p[1] < 0x80)	/* be quick for ASCII */
1677 	return 1;
1678 
1679     /* Skip over first UTF-8 char, stopping at a NUL byte. */
1680     len = utf_ptr2len(p);
1681 
1682     /* Check for illegal byte. */
1683     if (len == 1 && b0 >= 0x80)
1684 	return 1;
1685 
1686     /*
1687      * Check for composing characters.  We can handle only the first two, but
1688      * skip all of them (otherwise the cursor would get stuck).
1689      */
1690 #ifdef FEAT_ARABIC
1691     prevlen = 0;
1692 #endif
1693     for (;;)
1694     {
1695 	if (p[len] < 0x80 || !UTF_COMPOSINGLIKE(p + prevlen, p + len))
1696 	    return len;
1697 
1698 	/* Skip over composing char */
1699 #ifdef FEAT_ARABIC
1700 	prevlen = len;
1701 #endif
1702 	len += utf_ptr2len(p + len);
1703     }
1704 }
1705 
1706 /*
1707  * Return the number of bytes the UTF-8 encoding of the character at "p[size]"
1708  * takes.  This includes following composing characters.
1709  * Returns 1 for an illegal char or an incomplete byte sequence.
1710  */
1711     int
1712 utfc_ptr2len_len(p, size)
1713     char_u	*p;
1714     int		size;
1715 {
1716     int		len;
1717 #ifdef FEAT_ARABIC
1718     int		prevlen;
1719 #endif
1720 
1721     if (*p == NUL)
1722 	return 0;
1723     if (p[0] < 0x80 && (size == 1 || p[1] < 0x80)) /* be quick for ASCII */
1724 	return 1;
1725 
1726     /* Skip over first UTF-8 char, stopping at a NUL byte. */
1727     len = utf_ptr2len_len(p, size);
1728 
1729     /* Check for illegal byte and incomplete byte sequence. */
1730     if ((len == 1 && p[0] >= 0x80) || len > size)
1731 	return 1;
1732 
1733     /*
1734      * Check for composing characters.  We can handle only the first two, but
1735      * skip all of them (otherwise the cursor would get stuck).
1736      */
1737 #ifdef FEAT_ARABIC
1738     prevlen = 0;
1739 #endif
1740     while (len < size)
1741     {
1742 	if (p[len] < 0x80 || !UTF_COMPOSINGLIKE(p + prevlen, p + len))
1743 	    break;
1744 
1745 	/* Skip over composing char */
1746 #ifdef FEAT_ARABIC
1747 	prevlen = len;
1748 #endif
1749 	len += utf_ptr2len_len(p + len, size - len);
1750     }
1751     return len;
1752 }
1753 
1754 /*
1755  * Return the number of bytes the UTF-8 encoding of character "c" takes.
1756  * This does not include composing characters.
1757  */
1758     int
1759 utf_char2len(c)
1760     int		c;
1761 {
1762     if (c < 0x80)
1763 	return 1;
1764     if (c < 0x800)
1765 	return 2;
1766     if (c < 0x10000)
1767 	return 3;
1768     if (c < 0x200000)
1769 	return 4;
1770     if (c < 0x4000000)
1771 	return 5;
1772     return 6;
1773 }
1774 
1775 /*
1776  * Convert Unicode character "c" to UTF-8 string in "buf[]".
1777  * Returns the number of bytes.
1778  * This does not include composing characters.
1779  */
1780     int
1781 utf_char2bytes(c, buf)
1782     int		c;
1783     char_u	*buf;
1784 {
1785     if (c < 0x80)		/* 7 bits */
1786     {
1787 	buf[0] = c;
1788 	return 1;
1789     }
1790     if (c < 0x800)		/* 11 bits */
1791     {
1792 	buf[0] = 0xc0 + ((unsigned)c >> 6);
1793 	buf[1] = 0x80 + (c & 0x3f);
1794 	return 2;
1795     }
1796     if (c < 0x10000)		/* 16 bits */
1797     {
1798 	buf[0] = 0xe0 + ((unsigned)c >> 12);
1799 	buf[1] = 0x80 + (((unsigned)c >> 6) & 0x3f);
1800 	buf[2] = 0x80 + (c & 0x3f);
1801 	return 3;
1802     }
1803     if (c < 0x200000)		/* 21 bits */
1804     {
1805 	buf[0] = 0xf0 + ((unsigned)c >> 18);
1806 	buf[1] = 0x80 + (((unsigned)c >> 12) & 0x3f);
1807 	buf[2] = 0x80 + (((unsigned)c >> 6) & 0x3f);
1808 	buf[3] = 0x80 + (c & 0x3f);
1809 	return 4;
1810     }
1811     if (c < 0x4000000)		/* 26 bits */
1812     {
1813 	buf[0] = 0xf8 + ((unsigned)c >> 24);
1814 	buf[1] = 0x80 + (((unsigned)c >> 18) & 0x3f);
1815 	buf[2] = 0x80 + (((unsigned)c >> 12) & 0x3f);
1816 	buf[3] = 0x80 + (((unsigned)c >> 6) & 0x3f);
1817 	buf[4] = 0x80 + (c & 0x3f);
1818 	return 5;
1819     }
1820 				/* 31 bits */
1821     buf[0] = 0xfc + ((unsigned)c >> 30);
1822     buf[1] = 0x80 + (((unsigned)c >> 24) & 0x3f);
1823     buf[2] = 0x80 + (((unsigned)c >> 18) & 0x3f);
1824     buf[3] = 0x80 + (((unsigned)c >> 12) & 0x3f);
1825     buf[4] = 0x80 + (((unsigned)c >> 6) & 0x3f);
1826     buf[5] = 0x80 + (c & 0x3f);
1827     return 6;
1828 }
1829 
1830 /*
1831  * Return TRUE if "c" is a composing UTF-8 character.  This means it will be
1832  * drawn on top of the preceding character.
1833  * Based on code from Markus Kuhn.
1834  */
1835     int
1836 utf_iscomposing(c)
1837     int		c;
1838 {
1839     /* sorted list of non-overlapping intervals */
1840     static struct interval combining[] =
1841     {
1842 	{0x0300, 0x034f}, {0x0360, 0x036f}, {0x0483, 0x0486}, {0x0488, 0x0489},
1843 	{0x0591, 0x05a1}, {0x05a3, 0x05b9}, {0x05bb, 0x05bd}, {0x05bf, 0x05bf},
1844 	{0x05c1, 0x05c2}, {0x05c4, 0x05c4}, {0x0610, 0x0615}, {0x064b, 0x0658},
1845 	{0x0670, 0x0670}, {0x06d6, 0x06dc}, {0x06de, 0x06e4}, {0x06e7, 0x06e8},
1846 	{0x06ea, 0x06ed}, {0x0711, 0x0711}, {0x0730, 0x074a}, {0x07a6, 0x07b0},
1847 	{0x0901, 0x0903}, {0x093c, 0x093c}, {0x093e, 0x094d}, {0x0951, 0x0954},
1848 	{0x0962, 0x0963}, {0x0981, 0x0983}, {0x09bc, 0x09bc}, {0x09be, 0x09c4},
1849 	{0x09c7, 0x09c8}, {0x09cb, 0x09cd}, {0x09d7, 0x09d7}, {0x09e2, 0x09e3},
1850 	{0x0a01, 0x0a03}, {0x0a3c, 0x0a3c}, {0x0a3e, 0x0a42}, {0x0a47, 0x0a48},
1851 	{0x0a4b, 0x0a4d}, {0x0a70, 0x0a71}, {0x0a81, 0x0a83}, {0x0abc, 0x0abc},
1852 	{0x0abe, 0x0ac5}, {0x0ac7, 0x0ac9}, {0x0acb, 0x0acd}, {0x0ae2, 0x0ae3},
1853 	{0x0b01, 0x0b03}, {0x0b3c, 0x0b3c}, {0x0b3e, 0x0b43}, {0x0b47, 0x0b48},
1854 	{0x0b4b, 0x0b4d}, {0x0b56, 0x0b57}, {0x0b82, 0x0b82}, {0x0bbe, 0x0bc2},
1855 	{0x0bc6, 0x0bc8}, {0x0bca, 0x0bcd}, {0x0bd7, 0x0bd7}, {0x0c01, 0x0c03},
1856 	{0x0c3e, 0x0c44}, {0x0c46, 0x0c48}, {0x0c4a, 0x0c4d}, {0x0c55, 0x0c56},
1857 	{0x0c82, 0x0c83}, {0x0cbc, 0x0cbc}, {0x0cbe, 0x0cc4}, {0x0cc6, 0x0cc8},
1858 	{0x0cca, 0x0ccd}, {0x0cd5, 0x0cd6}, {0x0d02, 0x0d03}, {0x0d3e, 0x0d43},
1859 	{0x0d46, 0x0d48}, {0x0d4a, 0x0d4d}, {0x0d57, 0x0d57}, {0x0d82, 0x0d83},
1860 	{0x0dca, 0x0dca}, {0x0dcf, 0x0dd4}, {0x0dd6, 0x0dd6}, {0x0dd8, 0x0ddf},
1861 	{0x0df2, 0x0df3}, {0x0e31, 0x0e31}, {0x0e34, 0x0e3a}, {0x0e47, 0x0e4e},
1862 	{0x0eb1, 0x0eb1}, {0x0eb4, 0x0eb9}, {0x0ebb, 0x0ebc}, {0x0ec8, 0x0ecd},
1863 	{0x0f18, 0x0f19}, {0x0f35, 0x0f35}, {0x0f37, 0x0f37}, {0x0f39, 0x0f39},
1864 	{0x0f3e, 0x0f3f}, {0x0f71, 0x0f84}, {0x0f86, 0x0f87}, {0x0f90, 0x0f97},
1865 	{0x0f99, 0x0fbc}, {0x0fc6, 0x0fc6}, {0x102c, 0x1032}, {0x1036, 0x1039},
1866 	{0x1056, 0x1059}, {0x1712, 0x1714}, {0x1732, 0x1734}, {0x1752, 0x1753},
1867 	{0x1772, 0x1773}, {0x17b6, 0x17d3}, {0x17dd, 0x17dd}, {0x180b, 0x180d},
1868 	{0x18a9, 0x18a9}, {0x1920, 0x192b}, {0x1930, 0x193b}, {0x20d0, 0x20ea},
1869 	{0x302a, 0x302f}, {0x3099, 0x309a}, {0xfb1e, 0xfb1e}, {0xfe00, 0xfe0f},
1870 	{0xfe20, 0xfe23},
1871     };
1872 
1873     return intable(combining, sizeof(combining), c);
1874 }
1875 
1876 /*
1877  * Return TRUE for characters that can be displayed in a normal way.
1878  * Only for characters of 0x100 and above!
1879  */
1880     int
1881 utf_printable(c)
1882     int		c;
1883 {
1884 #ifdef USE_WCHAR_FUNCTIONS
1885     /*
1886      * Assume the iswprint() library function works better than our own stuff.
1887      */
1888     return iswprint(c);
1889 #else
1890     /* Sorted list of non-overlapping intervals.
1891      * 0xd800-0xdfff is reserved for UTF-16, actually illegal. */
1892     static struct interval nonprint[] =
1893     {
1894 	{0x070f, 0x070f}, {0x180b, 0x180e}, {0x200b, 0x200f}, {0x202a, 0x202e},
1895 	{0x206a, 0x206f}, {0xd800, 0xdfff}, {0xfeff, 0xfeff}, {0xfff9, 0xfffb},
1896 	{0xfffe, 0xffff}
1897     };
1898 
1899     return !intable(nonprint, sizeof(nonprint), c);
1900 #endif
1901 }
1902 
1903 /*
1904  * Get class of a Unicode character.
1905  * 0: white space
1906  * 1: punctuation
1907  * 2 or bigger: some class of word character.
1908  */
1909     int
1910 utf_class(c)
1911     int		c;
1912 {
1913     /* sorted list of non-overlapping intervals */
1914     static struct clinterval
1915     {
1916 	unsigned short first;
1917 	unsigned short last;
1918 	unsigned short class;
1919     } classes[] =
1920     {
1921 	{0x037e, 0x037e, 1},		/* Greek question mark */
1922 	{0x0387, 0x0387, 1},		/* Greek ano teleia */
1923 	{0x055a, 0x055f, 1},		/* Armenian punctuation */
1924 	{0x0589, 0x0589, 1},		/* Armenian full stop */
1925 	{0x05be, 0x05be, 1},
1926 	{0x05c0, 0x05c0, 1},
1927 	{0x05c3, 0x05c3, 1},
1928 	{0x05f3, 0x05f4, 1},
1929 	{0x060c, 0x060c, 1},
1930 	{0x061b, 0x061b, 1},
1931 	{0x061f, 0x061f, 1},
1932 	{0x066a, 0x066d, 1},
1933 	{0x06d4, 0x06d4, 1},
1934 	{0x0700, 0x070d, 1},		/* Syriac punctuation */
1935 	{0x0964, 0x0965, 1},
1936 	{0x0970, 0x0970, 1},
1937 	{0x0df4, 0x0df4, 1},
1938 	{0x0e4f, 0x0e4f, 1},
1939 	{0x0e5a, 0x0e5b, 1},
1940 	{0x0f04, 0x0f12, 1},
1941 	{0x0f3a, 0x0f3d, 1},
1942 	{0x0f85, 0x0f85, 1},
1943 	{0x104a, 0x104f, 1},		/* Myanmar punctuation */
1944 	{0x10fb, 0x10fb, 1},		/* Georgian punctuation */
1945 	{0x1361, 0x1368, 1},		/* Ethiopic punctuation */
1946 	{0x166d, 0x166e, 1},		/* Canadian Syl. punctuation */
1947 	{0x1680, 0x1680, 0},
1948 	{0x169b, 0x169c, 1},
1949 	{0x16eb, 0x16ed, 1},
1950 	{0x1735, 0x1736, 1},
1951 	{0x17d4, 0x17dc, 1},		/* Khmer punctuation */
1952 	{0x1800, 0x180a, 1},		/* Mongolian punctuation */
1953 	{0x2000, 0x200b, 0},		/* spaces */
1954 	{0x200c, 0x2027, 1},		/* punctuation and symbols */
1955 	{0x2028, 0x2029, 0},
1956 	{0x202a, 0x202e, 1},		/* punctuation and symbols */
1957 	{0x202f, 0x202f, 0},
1958 	{0x2030, 0x205e, 1},		/* punctuation and symbols */
1959 	{0x205f, 0x205f, 0},
1960 	{0x2060, 0x27ff, 1},		/* punctuation and symbols */
1961 	{0x2070, 0x207f, 0x2070},	/* superscript */
1962 	{0x2080, 0x208f, 0x2080},	/* subscript */
1963 	{0x2983, 0x2998, 1},
1964 	{0x29d8, 0x29db, 1},
1965 	{0x29fc, 0x29fd, 1},
1966 	{0x3000, 0x3000, 0},		/* ideographic space */
1967 	{0x3001, 0x3020, 1},		/* ideographic punctuation */
1968 	{0x3030, 0x3030, 1},
1969 	{0x303d, 0x303d, 1},
1970 	{0x3040, 0x309f, 0x3040},	/* Hiragana */
1971 	{0x30a0, 0x30ff, 0x30a0},	/* Katakana */
1972 	{0x3300, 0x9fff, 0x4e00},	/* CJK Ideographs */
1973 	{0xac00, 0xd7a3, 0xac00},	/* Hangul Syllables */
1974 	{0xf900, 0xfaff, 0x4e00},	/* CJK Ideographs */
1975 	{0xfd3e, 0xfd3f, 1},
1976 	{0xfe30, 0xfe6b, 1},		/* punctuation forms */
1977 	{0xff00, 0xff0f, 1},		/* half/fullwidth ASCII */
1978 	{0xff1a, 0xff20, 1},		/* half/fullwidth ASCII */
1979 	{0xff3b, 0xff40, 1},		/* half/fullwidth ASCII */
1980 	{0xff5b, 0xff65, 1},		/* half/fullwidth ASCII */
1981     };
1982     int bot = 0;
1983     int top = sizeof(classes) / sizeof(struct clinterval) - 1;
1984     int mid;
1985 
1986     /* First quick check for Latin1 characters, use 'iskeyword'. */
1987     if (c < 0x100)
1988     {
1989 	if (c == ' ' || c == '\t' || c == NUL || c == 0xa0)
1990 	    return 0;	    /* blank */
1991 	if (vim_iswordc(c))
1992 	    return 2;	    /* word character */
1993 	return 1;	    /* punctuation */
1994     }
1995 
1996     /* binary search in table */
1997     while (top >= bot)
1998     {
1999 	mid = (bot + top) / 2;
2000 	if (classes[mid].last < c)
2001 	    bot = mid + 1;
2002 	else if (classes[mid].first > c)
2003 	    top = mid - 1;
2004 	else
2005 	    return (int)classes[mid].class;
2006     }
2007 
2008     /* most other characters are "word" characters */
2009     return 2;
2010 }
2011 
2012 /*
2013  * Code for Unicode case-dependent operations.  Based on notes in
2014  * http://www.unicode.org/Public/UNIDATA/CaseFolding.txt
2015  * This code uses simple case folding, not full case folding.
2016  */
2017 
2018 /*
2019  * The following table is built by foldExtract.pl < CaseFolding.txt .
2020  * It must be in numeric order, because we use binary search on it.
2021  * An entry such as {0x41,0x5a,1,32} means that UCS-4 characters in the range
2022  * from 0x41 to 0x5a inclusive, stepping by 1, are folded by adding 32.
2023  */
2024 
2025 typedef struct
2026 {
2027     int rangeStart;
2028     int rangeEnd;
2029     int step;
2030     int offset;
2031 } convertStruct;
2032 
2033 static convertStruct foldCase[] =
2034 {
2035 	{0x41,0x5a,1,32}, {0xc0,0xd6,1,32}, {0xd8,0xde,1,32},
2036 	{0x100,0x12e,2,1}, {0x130,0x130,-1,-199}, {0x132,0x136,2,1},
2037 	{0x139,0x147,2,1}, {0x14a,0x176,2,1}, {0x178,0x178,-1,-121},
2038 	{0x179,0x17d,2,1}, {0x181,0x181,-1,210}, {0x182,0x184,2,1},
2039 	{0x186,0x186,-1,206}, {0x187,0x187,-1,1}, {0x189,0x18a,1,205},
2040 	{0x18b,0x18b,-1,1}, {0x18e,0x18e,-1,79}, {0x18f,0x18f,-1,202},
2041 	{0x190,0x190,-1,203}, {0x191,0x191,-1,1}, {0x193,0x193,-1,205},
2042 	{0x194,0x194,-1,207}, {0x196,0x196,-1,211}, {0x197,0x197,-1,209},
2043 	{0x198,0x198,-1,1}, {0x19c,0x19c,-1,211}, {0x19d,0x19d,-1,213},
2044 	{0x19f,0x19f,-1,214}, {0x1a0,0x1a4,2,1}, {0x1a6,0x1a6,-1,218},
2045 	{0x1a7,0x1a7,-1,1}, {0x1a9,0x1a9,-1,218}, {0x1ac,0x1ac,-1,1},
2046 	{0x1ae,0x1ae,-1,218}, {0x1af,0x1af,-1,1}, {0x1b1,0x1b2,1,217},
2047 	{0x1b3,0x1b5,2,1}, {0x1b7,0x1b7,-1,219}, {0x1b8,0x1bc,4,1},
2048 	{0x1c4,0x1c4,-1,2}, {0x1c5,0x1c5,-1,1}, {0x1c7,0x1c7,-1,2},
2049 	{0x1c8,0x1c8,-1,1}, {0x1ca,0x1ca,-1,2}, {0x1cb,0x1db,2,1},
2050 	{0x1de,0x1ee,2,1}, {0x1f1,0x1f1,-1,2}, {0x1f2,0x1f4,2,1},
2051 	{0x1f6,0x1f6,-1,-97}, {0x1f7,0x1f7,-1,-56}, {0x1f8,0x21e,2,1},
2052 	{0x220,0x220,-1,-130}, {0x222,0x232,2,1}, {0x386,0x386,-1,38},
2053 	{0x388,0x38a,1,37}, {0x38c,0x38c,-1,64}, {0x38e,0x38f,1,63},
2054 	{0x391,0x3a1,1,32}, {0x3a3,0x3ab,1,32}, {0x3d8,0x3ee,2,1},
2055 	{0x3f4,0x3f4,-1,-60}, {0x3f7,0x3f7,-1,1}, {0x3f9,0x3f9,-1,-7},
2056 	{0x3fa,0x3fa,-1,1}, {0x400,0x40f,1,80}, {0x410,0x42f,1,32},
2057 	{0x460,0x480,2,1}, {0x48a,0x4be,2,1}, {0x4c1,0x4cd,2,1},
2058 	{0x4d0,0x4f4,2,1}, {0x4f8,0x500,8,1}, {0x502,0x50e,2,1},
2059 	{0x531,0x556,1,48}, {0x1e00,0x1e94,2,1}, {0x1ea0,0x1ef8,2,1},
2060 	{0x1f08,0x1f0f,1,-8}, {0x1f18,0x1f1d,1,-8}, {0x1f28,0x1f2f,1,-8},
2061 	{0x1f38,0x1f3f,1,-8}, {0x1f48,0x1f4d,1,-8}, {0x1f59,0x1f5f,2,-8},
2062 	{0x1f68,0x1f6f,1,-8}, {0x1f88,0x1f8f,1,-8}, {0x1f98,0x1f9f,1,-8},
2063 	{0x1fa8,0x1faf,1,-8}, {0x1fb8,0x1fb9,1,-8}, {0x1fba,0x1fbb,1,-74},
2064 	{0x1fbc,0x1fbc,-1,-9}, {0x1fc8,0x1fcb,1,-86}, {0x1fcc,0x1fcc,-1,-9},
2065 	{0x1fd8,0x1fd9,1,-8}, {0x1fda,0x1fdb,1,-100}, {0x1fe8,0x1fe9,1,-8},
2066 	{0x1fea,0x1feb,1,-112}, {0x1fec,0x1fec,-1,-7}, {0x1ff8,0x1ff9,1,-128},
2067 	{0x1ffa,0x1ffb,1,-126}, {0x1ffc,0x1ffc,-1,-9}, {0x2126,0x2126,-1,-7517},
2068 	{0x212a,0x212a,-1,-8383}, {0x212b,0x212b,-1,-8262},
2069 	{0x2160,0x216f,1,16}, {0x24b6,0x24cf,1,26}, {0xff21,0xff3a,1,32},
2070 	{0x10400,0x10427,1,40}
2071 };
2072 
2073 static int utf_convert(int a, convertStruct table[], int tableSize);
2074 
2075 /*
2076  * Generic conversion function for case operations.
2077  * Return the converted equivalent of "a", which is a UCS-4 character.  Use
2078  * the given conversion "table".  Uses binary search on "table".
2079  */
2080     static int
2081 utf_convert(a, table, tableSize)
2082     int			a;
2083     convertStruct	table[];
2084     int			tableSize;
2085 {
2086     int start, mid, end; /* indices into table */
2087 
2088     start = 0;
2089     end = tableSize / sizeof(convertStruct);
2090     while (start < end)
2091     {
2092 	/* need to search further */
2093 	mid = (end + start) /2;
2094 	if (table[mid].rangeEnd < a)
2095 	    start = mid + 1;
2096 	else
2097 	    end = mid;
2098     }
2099     if (table[start].rangeStart <= a && a <= table[start].rangeEnd
2100 	    && (a - table[start].rangeStart) % table[start].step == 0)
2101 	return (a + table[start].offset);
2102     else
2103 	return a;
2104 }
2105 
2106 /*
2107  * Return the folded-case equivalent of "a", which is a UCS-4 character.  Uses
2108  * simple case folding.
2109  */
2110     int
2111 utf_fold(a)
2112     int		a;
2113 {
2114     return utf_convert(a, foldCase, sizeof(foldCase));
2115 }
2116 
2117 /*
2118  * The following tables are built by upperLowerExtract.pl < UnicodeData.txt .
2119  * They must be in numeric order, because we use binary search on them.
2120  * An entry such as {0x41,0x5a,1,32} means that UCS-4 characters in the range
2121  * from 0x41 to 0x5a inclusive, stepping by 1, are switched to lower (for
2122  * example) by adding 32.
2123  */
2124 static convertStruct toLower[] =
2125 {
2126 	{0x41,0x5a,1,32}, {0xc0,0xd6,1,32}, {0xd8,0xde,1,32},
2127 	{0x100,0x12e,2,1}, {0x130,0x130,-1,-199}, {0x132,0x136,2,1},
2128 	{0x139,0x147,2,1}, {0x14a,0x176,2,1}, {0x178,0x178,-1,-121},
2129 	{0x179,0x17d,2,1}, {0x181,0x181,-1,210}, {0x182,0x184,2,1},
2130 	{0x186,0x186,-1,206}, {0x187,0x187,-1,1}, {0x189,0x18a,1,205},
2131 	{0x18b,0x18b,-1,1}, {0x18e,0x18e,-1,79}, {0x18f,0x18f,-1,202},
2132 	{0x190,0x190,-1,203}, {0x191,0x191,-1,1}, {0x193,0x193,-1,205},
2133 	{0x194,0x194,-1,207}, {0x196,0x196,-1,211}, {0x197,0x197,-1,209},
2134 	{0x198,0x198,-1,1}, {0x19c,0x19c,-1,211}, {0x19d,0x19d,-1,213},
2135 	{0x19f,0x19f,-1,214}, {0x1a0,0x1a4,2,1}, {0x1a6,0x1a6,-1,218},
2136 	{0x1a7,0x1a7,-1,1}, {0x1a9,0x1a9,-1,218}, {0x1ac,0x1ac,-1,1},
2137 	{0x1ae,0x1ae,-1,218}, {0x1af,0x1af,-1,1}, {0x1b1,0x1b2,1,217},
2138 	{0x1b3,0x1b5,2,1}, {0x1b7,0x1b7,-1,219}, {0x1b8,0x1bc,4,1},
2139 	{0x1c4,0x1ca,3,2}, {0x1cd,0x1db,2,1}, {0x1de,0x1ee,2,1},
2140 	{0x1f1,0x1f1,-1,2}, {0x1f4,0x1f4,-1,1}, {0x1f6,0x1f6,-1,-97},
2141 	{0x1f7,0x1f7,-1,-56}, {0x1f8,0x21e,2,1}, {0x220,0x220,-1,-130},
2142 	{0x222,0x232,2,1}, {0x386,0x386,-1,38}, {0x388,0x38a,1,37},
2143 	{0x38c,0x38c,-1,64}, {0x38e,0x38f,1,63}, {0x391,0x3a1,1,32},
2144 	{0x3a3,0x3ab,1,32}, {0x3d8,0x3ee,2,1}, {0x3f4,0x3f4,-1,-60},
2145 	{0x3f7,0x3f7,-1,1}, {0x3f9,0x3f9,-1,-7}, {0x3fa,0x3fa,-1,1},
2146 	{0x400,0x40f,1,80}, {0x410,0x42f,1,32}, {0x460,0x480,2,1},
2147 	{0x48a,0x4be,2,1}, {0x4c1,0x4cd,2,1}, {0x4d0,0x4f4,2,1},
2148 	{0x4f8,0x500,8,1}, {0x502,0x50e,2,1}, {0x531,0x556,1,48},
2149 	{0x1e00,0x1e94,2,1}, {0x1ea0,0x1ef8,2,1}, {0x1f08,0x1f0f,1,-8},
2150 	{0x1f18,0x1f1d,1,-8}, {0x1f28,0x1f2f,1,-8}, {0x1f38,0x1f3f,1,-8},
2151 	{0x1f48,0x1f4d,1,-8}, {0x1f59,0x1f5f,2,-8}, {0x1f68,0x1f6f,1,-8},
2152 	{0x1fb8,0x1fb9,1,-8}, {0x1fba,0x1fbb,1,-74}, {0x1fc8,0x1fcb,1,-86},
2153 	{0x1fd8,0x1fd9,1,-8}, {0x1fda,0x1fdb,1,-100}, {0x1fe8,0x1fe9,1,-8},
2154 	{0x1fea,0x1feb,1,-112}, {0x1fec,0x1fec,-1,-7}, {0x1ff8,0x1ff9,1,-128},
2155 	{0x1ffa,0x1ffb,1,-126}, {0x2126,0x2126,-1,-7517}, {0x212a,0x212a,-1,-8383},
2156 	{0x212b,0x212b,-1,-8262}, {0xff21,0xff3a,1,32}, {0x10400,0x10427,1,40}
2157 };
2158 
2159 static convertStruct toUpper[] =
2160 {
2161 	{0x61,0x7a,1,-32}, {0xb5,0xb5,-1,743}, {0xe0,0xf6,1,-32},
2162 	{0xf8,0xfe,1,-32}, {0xff,0xff,-1,121}, {0x101,0x12f,2,-1},
2163 	{0x131,0x131,-1,-232}, {0x133,0x137,2,-1}, {0x13a,0x148,2,-1},
2164 	{0x14b,0x177,2,-1}, {0x17a,0x17e,2,-1}, {0x17f,0x17f,-1,-300},
2165 	{0x183,0x185,2,-1}, {0x188,0x18c,4,-1}, {0x192,0x192,-1,-1},
2166 	{0x195,0x195,-1,97}, {0x199,0x199,-1,-1}, {0x19e,0x19e,-1,130},
2167 	{0x1a1,0x1a5,2,-1}, {0x1a8,0x1ad,5,-1}, {0x1b0,0x1b4,4,-1},
2168 	{0x1b6,0x1b9,3,-1}, {0x1bd,0x1bd,-1,-1}, {0x1bf,0x1bf,-1,56},
2169 	{0x1c5,0x1c6,1,-1}, {0x1c8,0x1c9,1,-1}, {0x1cb,0x1cc,1,-1},
2170 	{0x1ce,0x1dc,2,-1}, {0x1dd,0x1dd,-1,-79}, {0x1df,0x1ef,2,-1},
2171 	{0x1f2,0x1f3,1,-1}, {0x1f5,0x1f9,4,-1}, {0x1fb,0x21f,2,-1},
2172 	{0x223,0x233,2,-1}, {0x253,0x253,-1,-210}, {0x254,0x254,-1,-206},
2173 	{0x256,0x257,1,-205}, {0x259,0x259,-1,-202}, {0x25b,0x25b,-1,-203},
2174 	{0x260,0x260,-1,-205}, {0x263,0x263,-1,-207}, {0x268,0x268,-1,-209},
2175 	{0x269,0x26f,6,-211}, {0x272,0x272,-1,-213}, {0x275,0x275,-1,-214},
2176 	{0x280,0x283,3,-218}, {0x288,0x288,-1,-218}, {0x28a,0x28b,1,-217},
2177 	{0x292,0x292,-1,-219}, {0x3ac,0x3ac,-1,-38}, {0x3ad,0x3af,1,-37},
2178 	{0x3b1,0x3c1,1,-32}, {0x3c2,0x3c2,-1,-31}, {0x3c3,0x3cb,1,-32},
2179 	{0x3cc,0x3cc,-1,-64}, {0x3cd,0x3ce,1,-63}, {0x3d0,0x3d0,-1,-62},
2180 	{0x3d1,0x3d1,-1,-57}, {0x3d5,0x3d5,-1,-47}, {0x3d6,0x3d6,-1,-54},
2181 	{0x3d9,0x3ef,2,-1}, {0x3f0,0x3f0,-1,-86}, {0x3f1,0x3f1,-1,-80},
2182 	{0x3f2,0x3f2,-1,7}, {0x3f5,0x3f5,-1,-96}, {0x3f8,0x3fb,3,-1},
2183 	{0x430,0x44f,1,-32}, {0x450,0x45f,1,-80}, {0x461,0x481,2,-1},
2184 	{0x48b,0x4bf,2,-1}, {0x4c2,0x4ce,2,-1}, {0x4d1,0x4f5,2,-1},
2185 	{0x4f9,0x501,8,-1}, {0x503,0x50f,2,-1}, {0x561,0x586,1,-48},
2186 	{0x1e01,0x1e95,2,-1}, {0x1e9b,0x1e9b,-1,-59}, {0x1ea1,0x1ef9,2,-1},
2187 	{0x1f00,0x1f07,1,8}, {0x1f10,0x1f15,1,8}, {0x1f20,0x1f27,1,8},
2188 	{0x1f30,0x1f37,1,8}, {0x1f40,0x1f45,1,8}, {0x1f51,0x1f57,2,8},
2189 	{0x1f60,0x1f67,1,8}, {0x1f70,0x1f71,1,74}, {0x1f72,0x1f75,1,86},
2190 	{0x1f76,0x1f77,1,100}, {0x1f78,0x1f79,1,128}, {0x1f7a,0x1f7b,1,112},
2191 	{0x1f7c,0x1f7d,1,126}, {0x1f80,0x1f87,1,8}, {0x1f90,0x1f97,1,8},
2192 	{0x1fa0,0x1fa7,1,8}, {0x1fb0,0x1fb1,1,8}, {0x1fb3,0x1fb3,-1,9},
2193 	{0x1fbe,0x1fbe,-1,-7205}, {0x1fc3,0x1fc3,-1,9}, {0x1fd0,0x1fd1,1,8},
2194 	{0x1fe0,0x1fe1,1,8}, {0x1fe5,0x1fe5,-1,7}, {0x1ff3,0x1ff3,-1,9},
2195 	{0xff41,0xff5a,1,-32}, {0x10428,0x1044f,1,-40}
2196 };
2197 
2198 /*
2199  * Return the upper-case equivalent of "a", which is a UCS-4 character.  Use
2200  * simple case folding.
2201  */
2202     int
2203 utf_toupper(a)
2204     int		a;
2205 {
2206     /* If 'casemap' contains "keepascii" use ASCII style toupper(). */
2207     if (a < 128 && (cmp_flags & CMP_KEEPASCII))
2208 	return TOUPPER_ASC(a);
2209 
2210 #if defined(HAVE_TOWUPPER) && defined(__STDC_ISO_10646__)
2211     /* If towupper() is available and handles Unicode, use it. */
2212     if (!(cmp_flags & CMP_INTERNAL))
2213 	return towupper(a);
2214 #endif
2215 
2216     /* For characters below 128 use locale sensitive toupper(). */
2217     if (a < 128)
2218 	return TOUPPER_LOC(a);
2219 
2220     /* For any other characters use the above mapping table. */
2221     return utf_convert(a, toUpper, sizeof(toUpper));
2222 }
2223 
2224     int
2225 utf_islower(a)
2226     int		a;
2227 {
2228     return (utf_toupper(a) != a);
2229 }
2230 
2231 /*
2232  * Return the lower-case equivalent of "a", which is a UCS-4 character.  Use
2233  * simple case folding.
2234  */
2235     int
2236 utf_tolower(a)
2237     int		a;
2238 {
2239     /* If 'casemap' contains "keepascii" use ASCII style tolower(). */
2240     if (a < 128 && (cmp_flags & CMP_KEEPASCII))
2241 	return TOLOWER_ASC(a);
2242 
2243 #if defined(HAVE_TOWLOWER) && defined(__STDC_ISO_10646__)
2244     /* If towlower() is available and handles Unicode, use it. */
2245     if (!(cmp_flags & CMP_INTERNAL))
2246 	return towlower(a);
2247 #endif
2248 
2249     /* For characters below 128 use locale sensitive tolower(). */
2250     if (a < 128)
2251 	return TOLOWER_LOC(a);
2252 
2253     /* For any other characters use the above mapping table. */
2254     return utf_convert(a, toLower, sizeof(toLower));
2255 }
2256 
2257     int
2258 utf_isupper(a)
2259     int		a;
2260 {
2261     return (utf_tolower(a) != a);
2262 }
2263 
2264 /*
2265  * Version of strnicmp() that handles multi-byte characters.
2266  * Needed for Big5, Sjift-JIS and UTF-8 encoding.  Other DBCS encodings can
2267  * probably use strnicmp(), because there are no ASCII characters in the
2268  * second byte.
2269  * Returns zero if s1 and s2 are equal (ignoring case), the difference between
2270  * two characters otherwise.
2271  */
2272     int
2273 mb_strnicmp(s1, s2, nn)
2274     char_u	*s1, *s2;
2275     size_t	nn;
2276 {
2277     int		i, j, l;
2278     int		cdiff;
2279     int		incomplete = FALSE;
2280     int		n = (int)nn;
2281 
2282     for (i = 0; i < n; i += l)
2283     {
2284 	if (s1[i] == NUL && s2[i] == NUL)   /* both strings end */
2285 	    return 0;
2286 	if (enc_utf8)
2287 	{
2288 	    l = utf_byte2len(s1[i]);
2289 	    if (l > n - i)
2290 	    {
2291 		l = n - i;		    /* incomplete character */
2292 		incomplete = TRUE;
2293 	    }
2294 	    /* Check directly first, it's faster. */
2295 	    for (j = 0; j < l; ++j)
2296 	    {
2297 		if (s1[i + j] != s2[i + j])
2298 		    break;
2299 		if (s1[i + j] == 0)
2300 		    /* Both stings have the same bytes but are incomplete or
2301 		     * have illegal bytes, accept them as equal. */
2302 		    l = j;
2303 	    }
2304 	    if (j < l)
2305 	    {
2306 		/* If one of the two characters is incomplete return -1. */
2307 		if (incomplete || i + utf_byte2len(s2[i]) > n)
2308 		    return -1;
2309 		cdiff = utf_fold(utf_ptr2char(s1 + i))
2310 					     - utf_fold(utf_ptr2char(s2 + i));
2311 		if (cdiff != 0)
2312 		    return cdiff;
2313 	    }
2314 	}
2315 	else
2316 	{
2317 	    l = (*mb_ptr2len)(s1 + i);
2318 	    if (l <= 1)
2319 	    {
2320 		/* Single byte: first check normally, then with ignore case. */
2321 		if (s1[i] != s2[i])
2322 		{
2323 		    cdiff = TOLOWER_LOC(s1[i]) - TOLOWER_LOC(s2[i]);
2324 		    if (cdiff != 0)
2325 			return cdiff;
2326 		}
2327 	    }
2328 	    else
2329 	    {
2330 		/* For non-Unicode multi-byte don't ignore case. */
2331 		if (l > n - i)
2332 		    l = n - i;
2333 		cdiff = STRNCMP(s1 + i, s2 + i, l);
2334 		if (cdiff != 0)
2335 		    return cdiff;
2336 	    }
2337 	}
2338     }
2339     return 0;
2340 }
2341 
2342 /*
2343  * "g8": show bytes of the UTF-8 char under the cursor.  Doesn't matter what
2344  * 'encoding' has been set to.
2345  */
2346     void
2347 show_utf8()
2348 {
2349     int		len;
2350     int		rlen = 0;
2351     char_u	*line;
2352     int		clen;
2353     int		i;
2354 
2355     /* Get the byte length of the char under the cursor, including composing
2356      * characters. */
2357     line = ml_get_cursor();
2358     len = utfc_ptr2len(line);
2359     if (len == 0)
2360     {
2361 	MSG("NUL");
2362 	return;
2363     }
2364 
2365     clen = 0;
2366     for (i = 0; i < len; ++i)
2367     {
2368 	if (clen == 0)
2369 	{
2370 	    /* start of (composing) character, get its length */
2371 	    if (i > 0)
2372 	    {
2373 		STRCPY(IObuff + rlen, "+ ");
2374 		rlen += 2;
2375 	    }
2376 	    clen = utf_ptr2len(line + i);
2377 	}
2378 	sprintf((char *)IObuff + rlen, "%02x ", line[i]);
2379 	--clen;
2380 	rlen += (int)STRLEN(IObuff + rlen);
2381 	if (rlen > IOSIZE - 20)
2382 	    break;
2383     }
2384 
2385     msg(IObuff);
2386 }
2387 
2388 /*
2389  * mb_head_off() function pointer.
2390  * Return offset from "p" to the first byte of the character it points into.
2391  * Returns 0 when already at the first byte of a character.
2392  */
2393 /*ARGSUSED*/
2394     int
2395 latin_head_off(base, p)
2396     char_u	*base;
2397     char_u	*p;
2398 {
2399     return 0;
2400 }
2401 
2402     int
2403 dbcs_head_off(base, p)
2404     char_u	*base;
2405     char_u	*p;
2406 {
2407     char_u	*q;
2408 
2409     /* It can't be a trailing byte when not using DBCS, at the start of the
2410      * string or the previous byte can't start a double-byte. */
2411     if (p <= base || MB_BYTE2LEN(p[-1]) == 1)
2412 	return 0;
2413 
2414     /* This is slow: need to start at the base and go forward until the
2415      * byte we are looking for.  Return 1 when we went past it, 0 otherwise. */
2416     q = base;
2417     while (q < p)
2418 	q += dbcs_ptr2len(q);
2419     return (q == p) ? 0 : 1;
2420 }
2421 
2422 #if defined(FEAT_CLIPBOARD) || defined(FEAT_GUI) || defined(FEAT_RIGHTLEFT) \
2423 	|| defined(PROTO)
2424 /*
2425  * Special version of dbcs_head_off() that works for ScreenLines[], where
2426  * single-width DBCS_JPNU characters are stored separately.
2427  */
2428     int
2429 dbcs_screen_head_off(base, p)
2430     char_u	*base;
2431     char_u	*p;
2432 {
2433     char_u	*q;
2434 
2435     /* It can't be a trailing byte when not using DBCS, at the start of the
2436      * string or the previous byte can't start a double-byte.
2437      * For euc-jp an 0x8e byte in the previous cell always means we have a
2438      * lead byte in the current cell. */
2439     if (p <= base
2440 	    || (enc_dbcs == DBCS_JPNU && p[-1] == 0x8e)
2441 	    || MB_BYTE2LEN(p[-1]) == 1)
2442 	return 0;
2443 
2444     /* This is slow: need to start at the base and go forward until the
2445      * byte we are looking for.  Return 1 when we went past it, 0 otherwise.
2446      * For DBCS_JPNU look out for 0x8e, which means the second byte is not
2447      * stored as the next byte. */
2448     q = base;
2449     while (q < p)
2450     {
2451 	if (enc_dbcs == DBCS_JPNU && *q == 0x8e)
2452 	    ++q;
2453 	else
2454 	    q += dbcs_ptr2len(q);
2455     }
2456     return (q == p) ? 0 : 1;
2457 }
2458 #endif
2459 
2460     int
2461 utf_head_off(base, p)
2462     char_u	*base;
2463     char_u	*p;
2464 {
2465     char_u	*q;
2466     char_u	*s;
2467     int		c;
2468 #ifdef FEAT_ARABIC
2469     char_u	*j;
2470 #endif
2471 
2472     if (*p < 0x80)		/* be quick for ASCII */
2473 	return 0;
2474 
2475     /* Skip backwards over trailing bytes: 10xx.xxxx
2476      * Skip backwards again if on a composing char. */
2477     for (q = p; ; --q)
2478     {
2479 	/* Move s to the last byte of this char. */
2480 	for (s = q; (s[1] & 0xc0) == 0x80; ++s)
2481 	    ;
2482 	/* Move q to the first byte of this char. */
2483 	while (q > base && (*q & 0xc0) == 0x80)
2484 	    --q;
2485 	/* Check for illegal sequence. Do allow an illegal byte after where we
2486 	 * started. */
2487 	if (utf8len_tab[*q] != (int)(s - q + 1)
2488 				       && utf8len_tab[*q] != (int)(p - q + 1))
2489 	    return 0;
2490 
2491 	if (q <= base)
2492 	    break;
2493 
2494 	c = utf_ptr2char(q);
2495 	if (utf_iscomposing(c))
2496 	    continue;
2497 
2498 #ifdef FEAT_ARABIC
2499 	if (arabic_maycombine(c))
2500 	{
2501 	    /* Advance to get a sneak-peak at the next char */
2502 	    j = q;
2503 	    --j;
2504 	    /* Move j to the first byte of this char. */
2505 	    while (j > base && (*j & 0xc0) == 0x80)
2506 		--j;
2507 	    if (arabic_combine(utf_ptr2char(j), c))
2508 		continue;
2509 	}
2510 #endif
2511 	break;
2512     }
2513 
2514     return (int)(p - q);
2515 }
2516 
2517 #if defined(FEAT_EVAL) || defined(PROTO)
2518 /*
2519  * Copy a character from "*fp" to "*tp" and advance the pointers.
2520  */
2521     void
2522 mb_copy_char(fp, tp)
2523     char_u	**fp;
2524     char_u	**tp;
2525 {
2526     int	    l = (*mb_ptr2len)(*fp);
2527 
2528     mch_memmove(*tp, *fp, (size_t)l);
2529     *tp += l;
2530     *fp += l;
2531 }
2532 #endif
2533 
2534 /*
2535  * Return the offset from "p" to the first byte of a character.  When "p" is
2536  * at the start of a character 0 is returned, otherwise the offset to the next
2537  * character.  Can start anywhere in a stream of bytes.
2538  */
2539     int
2540 mb_off_next(base, p)
2541     char_u	*base;
2542     char_u	*p;
2543 {
2544     int		i;
2545     int		j;
2546 
2547     if (enc_utf8)
2548     {
2549 	if (*p < 0x80)		/* be quick for ASCII */
2550 	    return 0;
2551 
2552 	/* Find the next character that isn't 10xx.xxxx */
2553 	for (i = 0; (p[i] & 0xc0) == 0x80; ++i)
2554 	    ;
2555 	if (i > 0)
2556 	{
2557 	    /* Check for illegal sequence. */
2558 	    for (j = 0; p - j > base; ++j)
2559 		if ((p[-j] & 0xc0) != 0x80)
2560 		    break;
2561 	    if (utf8len_tab[p[-j]] != i + j)
2562 		return 0;
2563 	}
2564 	return i;
2565     }
2566 
2567     /* Only need to check if we're on a trail byte, it doesn't matter if we
2568      * want the offset to the next or current character. */
2569     return (*mb_head_off)(base, p);
2570 }
2571 
2572 /*
2573  * Return the offset from "p" to the last byte of the character it points
2574  * into.  Can start anywhere in a stream of bytes.
2575  */
2576     int
2577 mb_tail_off(base, p)
2578     char_u	*base;
2579     char_u	*p;
2580 {
2581     int		i;
2582     int		j;
2583 
2584     if (*p == NUL)
2585 	return 0;
2586 
2587     if (enc_utf8)
2588     {
2589 	/* Find the last character that is 10xx.xxxx */
2590 	for (i = 0; (p[i + 1] & 0xc0) == 0x80; ++i)
2591 	    ;
2592 	/* Check for illegal sequence. */
2593 	for (j = 0; p - j > base; ++j)
2594 	    if ((p[-j] & 0xc0) != 0x80)
2595 		break;
2596 	if (utf8len_tab[p[-j]] != i + j + 1)
2597 	    return 0;
2598 	return i;
2599     }
2600 
2601     /* It can't be the first byte if a double-byte when not using DBCS, at the
2602      * end of the string or the byte can't start a double-byte. */
2603     if (enc_dbcs == 0 || p[1] == NUL || MB_BYTE2LEN(*p) == 1)
2604 	return 0;
2605 
2606     /* Return 1 when on the lead byte, 0 when on the tail byte. */
2607     return 1 - dbcs_head_off(base, p);
2608 }
2609 
2610 /*
2611  * Find the next illegal byte sequence.
2612  */
2613     void
2614 utf_find_illegal()
2615 {
2616     pos_T	pos = curwin->w_cursor;
2617     char_u	*p;
2618     int		len;
2619     vimconv_T	vimconv;
2620     char_u	*tofree = NULL;
2621 
2622     vimconv.vc_type = CONV_NONE;
2623     if (enc_utf8 && (enc_canon_props(curbuf->b_p_fenc) & ENC_8BIT))
2624     {
2625 	/* 'encoding' is "utf-8" but we are editing a 8-bit encoded file,
2626 	 * possibly a utf-8 file with illegal bytes.  Setup for conversion
2627 	 * from utf-8 to 'fileencoding'. */
2628 	convert_setup(&vimconv, p_enc, curbuf->b_p_fenc);
2629     }
2630 
2631 #ifdef FEAT_VIRTUALEDIT
2632     curwin->w_cursor.coladd = 0;
2633 #endif
2634     for (;;)
2635     {
2636 	p = ml_get_cursor();
2637 	if (vimconv.vc_type != CONV_NONE)
2638 	{
2639 	    vim_free(tofree);
2640 	    tofree = string_convert(&vimconv, p, NULL);
2641 	    if (tofree == NULL)
2642 		break;
2643 	    p = tofree;
2644 	}
2645 
2646 	while (*p != NUL)
2647 	{
2648 	    /* Illegal means that there are not enough trail bytes (checked by
2649 	     * utf_ptr2len()) or too many of them (overlong sequence). */
2650 	    len = utf_ptr2len(p);
2651 	    if (*p >= 0x80 && (len == 1
2652 				     || utf_char2len(utf_ptr2char(p)) != len))
2653 	    {
2654 		if (vimconv.vc_type == CONV_NONE)
2655 		    curwin->w_cursor.col += (colnr_T)(p - ml_get_cursor());
2656 		else
2657 		{
2658 		    int	    l;
2659 
2660 		    len = (int)(p - tofree);
2661 		    for (p = ml_get_cursor(); *p != NUL && len-- > 0; p += l)
2662 		    {
2663 			l = utf_ptr2len(p);
2664 			curwin->w_cursor.col += l;
2665 		    }
2666 		}
2667 		goto theend;
2668 	    }
2669 	    p += len;
2670 	}
2671 	if (curwin->w_cursor.lnum == curbuf->b_ml.ml_line_count)
2672 	    break;
2673 	++curwin->w_cursor.lnum;
2674 	curwin->w_cursor.col = 0;
2675     }
2676 
2677     /* didn't find it: don't move and beep */
2678     curwin->w_cursor = pos;
2679     beep_flush();
2680 
2681 theend:
2682     vim_free(tofree);
2683     convert_setup(&vimconv, NULL, NULL);
2684 }
2685 
2686 #if defined(HAVE_GTK2) || defined(PROTO)
2687 /*
2688  * Return TRUE if string "s" is a valid utf-8 string.
2689  * When "end" is NULL stop at the first NUL.
2690  * When "end" is positive stop there.
2691  */
2692     int
2693 utf_valid_string(s, end)
2694     char_u	*s;
2695     char_u	*end;
2696 {
2697     int		l;
2698     char_u	*p = s;
2699 
2700     while (end == NULL ? *p != NUL : p < end)
2701     {
2702 	if ((*p & 0xc0) == 0x80)
2703 	    return FALSE;	/* invalid lead byte */
2704 	l = utf8len_tab[*p];
2705 	if (end != NULL && p + l > end)
2706 	    return FALSE;	/* incomplete byte sequence */
2707 	++p;
2708 	while (--l > 0)
2709 	    if ((*p++ & 0xc0) != 0x80)
2710 		return FALSE;	/* invalid trail byte */
2711     }
2712     return TRUE;
2713 }
2714 #endif
2715 
2716 #if defined(FEAT_GUI) || defined(PROTO)
2717 /*
2718  * Special version of mb_tail_off() for use in ScreenLines[].
2719  */
2720     int
2721 dbcs_screen_tail_off(base, p)
2722     char_u	*base;
2723     char_u	*p;
2724 {
2725     /* It can't be the first byte if a double-byte when not using DBCS, at the
2726      * end of the string or the byte can't start a double-byte.
2727      * For euc-jp an 0x8e byte always means we have a lead byte in the current
2728      * cell. */
2729     if (*p == NUL || p[1] == NUL
2730 	    || (enc_dbcs == DBCS_JPNU && *p == 0x8e)
2731 	    || MB_BYTE2LEN(*p) == 1)
2732 	return 0;
2733 
2734     /* Return 1 when on the lead byte, 0 when on the tail byte. */
2735     return 1 - dbcs_screen_head_off(base, p);
2736 }
2737 #endif
2738 
2739 /*
2740  * If the cursor moves on an trail byte, set the cursor on the lead byte.
2741  * Thus it moves left if necessary.
2742  * Return TRUE when the cursor was adjusted.
2743  */
2744     void
2745 mb_adjust_cursor()
2746 {
2747     mb_adjustpos(&curwin->w_cursor);
2748 }
2749 
2750 /*
2751  * Adjust position "*lp" to point to the first byte of a multi-byte character.
2752  * If it points to a tail byte it's moved backwards to the head byte.
2753  */
2754     void
2755 mb_adjustpos(lp)
2756     pos_T	*lp;
2757 {
2758     char_u	*p;
2759 
2760     if (lp->col > 0
2761 #ifdef FEAT_VIRTUALEDIT
2762 	    || lp->coladd > 1
2763 #endif
2764 	    )
2765     {
2766 	p = ml_get(lp->lnum);
2767 	lp->col -= (*mb_head_off)(p, p + lp->col);
2768 #ifdef FEAT_VIRTUALEDIT
2769 	/* Reset "coladd" when the cursor would be on the right half of a
2770 	 * double-wide character. */
2771 	if (lp->coladd == 1
2772 		&& p[lp->col] != TAB
2773 		&& vim_isprintc((*mb_ptr2char)(p + lp->col))
2774 		&& ptr2cells(p + lp->col) > 1)
2775 	    lp->coladd = 0;
2776 #endif
2777     }
2778 }
2779 
2780 /*
2781  * Return a pointer to the character before "*p", if there is one.
2782  */
2783     char_u *
2784 mb_prevptr(line, p)
2785     char_u *line;	/* start of the string */
2786     char_u *p;
2787 {
2788     if (p > line)
2789 	mb_ptr_back(line, p);
2790     return p;
2791 }
2792 
2793 /*
2794  * Return the character length of "str".  Each multi-byte character (with
2795  * following composing characters) counts as one.
2796  */
2797     int
2798 mb_charlen(str)
2799     char_u	*str;
2800 {
2801     char_u	*p = str;
2802     int		count;
2803 
2804     if (p == NULL)
2805 	return 0;
2806 
2807     for (count = 0; *p != NUL; count++)
2808 	p += (*mb_ptr2len)(p);
2809 
2810     return count;
2811 }
2812 
2813 #if defined(FEAT_SPELL) || defined(PROTO)
2814 /*
2815  * Like mb_charlen() but for a string with specified length.
2816  */
2817     int
2818 mb_charlen_len(str, len)
2819     char_u	*str;
2820     int		len;
2821 {
2822     char_u	*p = str;
2823     int		count;
2824 
2825     for (count = 0; *p != NUL && p < str + len; count++)
2826 	p += (*mb_ptr2len)(p);
2827 
2828     return count;
2829 }
2830 #endif
2831 
2832 /*
2833  * Try to un-escape a multi-byte character.
2834  * Used for the "to" and "from" part of a mapping.
2835  * Return the un-escaped string if it is a multi-byte character, and advance
2836  * "pp" to just after the bytes that formed it.
2837  * Return NULL if no multi-byte char was found.
2838  */
2839     char_u *
2840 mb_unescape(pp)
2841     char_u **pp;
2842 {
2843     static char_u	buf[MB_MAXBYTES + 1];
2844     int			n, m = 0;
2845     char_u		*str = *pp;
2846 
2847     /* Must translate K_SPECIAL KS_SPECIAL KE_FILLER to K_SPECIAL and CSI
2848      * KS_EXTRA KE_CSI to CSI. */
2849     for (n = 0; str[n] != NUL && m <= MB_MAXBYTES; ++n)
2850     {
2851 	if (str[n] == K_SPECIAL
2852 		&& str[n + 1] == KS_SPECIAL
2853 		&& str[n + 2] == KE_FILLER)
2854 	{
2855 	    buf[m++] = K_SPECIAL;
2856 	    n += 2;
2857 	}
2858 # ifdef FEAT_GUI
2859 	else if (str[n] == CSI
2860 		&& str[n + 1] == KS_EXTRA
2861 		&& str[n + 2] == (int)KE_CSI)
2862 	{
2863 	    buf[m++] = CSI;
2864 	    n += 2;
2865 	}
2866 # endif
2867 	else if (str[n] == K_SPECIAL
2868 # ifdef FEAT_GUI
2869 		|| str[n] == CSI
2870 # endif
2871 		)
2872 	    break;		/* a special key can't be a multibyte char */
2873 	else
2874 	    buf[m++] = str[n];
2875 	buf[m] = NUL;
2876 
2877 	/* Return a multi-byte character if it's found.  An illegal sequence
2878 	 * will result in a 1 here. */
2879 	if ((*mb_ptr2len)(buf) > 1)
2880 	{
2881 	    *pp = str + n + 1;
2882 	    return buf;
2883 	}
2884     }
2885     return NULL;
2886 }
2887 
2888 /*
2889  * Return TRUE if the character at "row"/"col" on the screen is the left side
2890  * of a double-width character.
2891  * Caller must make sure "row" and "col" are not invalid!
2892  */
2893     int
2894 mb_lefthalve(row, col)
2895     int	    row;
2896     int	    col;
2897 {
2898 #ifdef FEAT_HANGULIN
2899     if (composing_hangul)
2900 	return TRUE;
2901 #endif
2902     if (enc_dbcs != 0)
2903 	return dbcs_off2cells(LineOffset[row] + col) > 1;
2904     if (enc_utf8)
2905 	return (col + 1 < Columns
2906 		&& ScreenLines[LineOffset[row] + col + 1] == 0);
2907     return FALSE;
2908 }
2909 
2910 # if defined(FEAT_CLIPBOARD) || defined(FEAT_GUI) || defined(FEAT_RIGHTLEFT) \
2911 	|| defined(PROTO)
2912 /*
2913  * Correct a position on the screen, if it's the right halve of a double-wide
2914  * char move it to the left halve.  Returns the corrected column.
2915  */
2916     int
2917 mb_fix_col(col, row)
2918     int		col;
2919     int		row;
2920 {
2921     col = check_col(col);
2922     row = check_row(row);
2923     if (has_mbyte && ScreenLines != NULL && col > 0
2924 	    && ((enc_dbcs
2925 		    && ScreenLines[LineOffset[row] + col] != NUL
2926 		    && dbcs_screen_head_off(ScreenLines + LineOffset[row],
2927 					 ScreenLines + LineOffset[row] + col))
2928 		|| (enc_utf8 && ScreenLines[LineOffset[row] + col] == 0)))
2929 	--col;
2930     return col;
2931 }
2932 # endif
2933 #endif
2934 
2935 #if defined(FEAT_MBYTE) || defined(FEAT_POSTSCRIPT) || defined(PROTO)
2936 static int enc_alias_search __ARGS((char_u *name));
2937 
2938 /*
2939  * Skip the Vim specific head of a 'encoding' name.
2940  */
2941     char_u *
2942 enc_skip(p)
2943     char_u	*p;
2944 {
2945     if (STRNCMP(p, "2byte-", 6) == 0)
2946 	return p + 6;
2947     if (STRNCMP(p, "8bit-", 5) == 0)
2948 	return p + 5;
2949     return p;
2950 }
2951 
2952 /*
2953  * Find the canonical name for encoding "enc".
2954  * When the name isn't recognized, returns "enc" itself, but with all lower
2955  * case characters and '_' replaced with '-'.
2956  * Returns an allocated string.  NULL for out-of-memory.
2957  */
2958     char_u *
2959 enc_canonize(enc)
2960     char_u	*enc;
2961 {
2962     char_u	*r;
2963     char_u	*p, *s;
2964     int		i;
2965 
2966 # ifdef FEAT_MBYTE
2967     if (STRCMP(enc, "default") == 0)
2968     {
2969 	/* Use the default encoding as it's found by set_init_1(). */
2970 	r = get_encoding_default();
2971 	if (r == NULL)
2972 	    r = (char_u *)"latin1";
2973 	return vim_strsave(r);
2974     }
2975 # endif
2976 
2977     /* copy "enc" to allocted memory, with room for two '-' */
2978     r = alloc((unsigned)(STRLEN(enc) + 3));
2979     if (r != NULL)
2980     {
2981 	/* Make it all lower case and replace '_' with '-'. */
2982 	p = r;
2983 	for (s = enc; *s != NUL; ++s)
2984 	{
2985 	    if (*s == '_')
2986 		*p++ = '-';
2987 	    else
2988 		*p++ = TOLOWER_ASC(*s);
2989 	}
2990 	*p = NUL;
2991 
2992 	/* Skip "2byte-" and "8bit-". */
2993 	p = enc_skip(r);
2994 
2995 	/* Change "microsoft-cp" to "cp".  Used in some spell files. */
2996 	if (STRNCMP(p, "microsoft-cp", 12) == 0)
2997 	    mch_memmove(p, p + 10, STRLEN(p + 10) + 1);
2998 
2999 	/* "iso8859" -> "iso-8859" */
3000 	if (STRNCMP(p, "iso8859", 7) == 0)
3001 	{
3002 	    mch_memmove(p + 4, p + 3, STRLEN(p + 2));
3003 	    p[3] = '-';
3004 	}
3005 
3006 	/* "iso-8859n" -> "iso-8859-n" */
3007 	if (STRNCMP(p, "iso-8859", 8) == 0 && p[8] != '-')
3008 	{
3009 	    mch_memmove(p + 9, p + 8, STRLEN(p + 7));
3010 	    p[8] = '-';
3011 	}
3012 
3013 	/* "latin-N" -> "latinN" */
3014 	if (STRNCMP(p, "latin-", 6) == 0)
3015 	    mch_memmove(p + 5, p + 6, STRLEN(p + 5));
3016 
3017 	if (enc_canon_search(p) >= 0)
3018 	{
3019 	    /* canonical name can be used unmodified */
3020 	    if (p != r)
3021 		mch_memmove(r, p, STRLEN(p) + 1);
3022 	}
3023 	else if ((i = enc_alias_search(p)) >= 0)
3024 	{
3025 	    /* alias recognized, get canonical name */
3026 	    vim_free(r);
3027 	    r = vim_strsave((char_u *)enc_canon_table[i].name);
3028 	}
3029     }
3030     return r;
3031 }
3032 
3033 /*
3034  * Search for an encoding alias of "name".
3035  * Returns -1 when not found.
3036  */
3037     static int
3038 enc_alias_search(name)
3039     char_u	*name;
3040 {
3041     int		i;
3042 
3043     for (i = 0; enc_alias_table[i].name != NULL; ++i)
3044 	if (STRCMP(name, enc_alias_table[i].name) == 0)
3045 	    return enc_alias_table[i].canon;
3046     return -1;
3047 }
3048 #endif
3049 
3050 #if defined(FEAT_MBYTE) || defined(PROTO)
3051 
3052 #ifdef HAVE_LANGINFO_H
3053 # include <langinfo.h>
3054 #endif
3055 
3056 /*
3057  * Get the canonicalized encoding of the current locale.
3058  * Returns an allocated string when successful, NULL when not.
3059  */
3060     char_u *
3061 enc_locale()
3062 {
3063 #ifndef WIN3264
3064     char	*s;
3065     char	*p;
3066     int		i;
3067 #endif
3068     char	buf[50];
3069 #ifdef WIN3264
3070     long	acp = GetACP();
3071 
3072     if (acp == 1200)
3073 	STRCPY(buf, "ucs-2le");
3074     else if (acp == 1252)	    /* cp1252 is used as latin1 */
3075 	STRCPY(buf, "latin1");
3076     else
3077 	sprintf(buf, "cp%ld", acp);
3078 #else
3079 # ifdef HAVE_NL_LANGINFO_CODESET
3080     if ((s = nl_langinfo(CODESET)) == NULL || *s == NUL)
3081 # endif
3082 #  if defined(HAVE_LOCALE_H) || defined(X_LOCALE)
3083 	if ((s = setlocale(LC_CTYPE, NULL)) == NULL || *s == NUL)
3084 #  endif
3085 	    if ((s = getenv("LC_ALL")) == NULL || *s == NUL)
3086 		if ((s = getenv("LC_CTYPE")) == NULL || *s == NUL)
3087 		    s = getenv("LANG");
3088 
3089     if (s == NULL || *s == NUL)
3090 	return FAIL;
3091 
3092     /* The most generic locale format is:
3093      * language[_territory][.codeset][@modifier][+special][,[sponsor][_revision]]
3094      * If there is a '.' remove the part before it.
3095      * if there is something after the codeset, remove it.
3096      * Make the name lowercase and replace '_' with '-'.
3097      * Exception: "ja_JP.EUC" == "euc-jp", "zh_CN.EUC" = "euc-cn",
3098      * "ko_KR.EUC" == "euc-kr"
3099      */
3100     if ((p = (char *)vim_strchr((char_u *)s, '.')) != NULL)
3101     {
3102 	if (p > s + 2 && STRNICMP(p + 1, "EUC", 3) == 0
3103 			&& !isalnum((int)p[4]) && p[4] != '-' && p[-3] == '_')
3104 	{
3105 	    /* copy "XY.EUC" to "euc-XY" to buf[10] */
3106 	    STRCPY(buf + 10, "euc-");
3107 	    buf[14] = p[-2];
3108 	    buf[15] = p[-1];
3109 	    buf[16] = 0;
3110 	    s = buf + 10;
3111 	}
3112 	else
3113 	    s = p + 1;
3114     }
3115     for (i = 0; s[i] != NUL && s + i < buf + sizeof(buf) - 1; ++i)
3116     {
3117 	if (s[i] == '_' || s[i] == '-')
3118 	    buf[i] = '-';
3119 	else if (isalnum((int)s[i]))
3120 	    buf[i] = TOLOWER_ASC(s[i]);
3121 	else
3122 	    break;
3123     }
3124     buf[i] = NUL;
3125 #endif
3126 
3127     return enc_canonize((char_u *)buf);
3128 }
3129 
3130 #if defined(WIN3264) || defined(PROTO)
3131 /*
3132  * Convert an encoding name to an MS-Windows codepage.
3133  * Returns zero if no codepage can be figured out.
3134  */
3135     int
3136 encname2codepage(name)
3137     char_u	*name;
3138 {
3139     int		cp;
3140     char_u	*p = name;
3141     int		idx;
3142 
3143     if (STRNCMP(p, "8bit-", 5) == 0)
3144 	p += 5;
3145     else if (STRNCMP(p_enc, "2byte-", 6) == 0)
3146 	p += 6;
3147 
3148     if (p[0] == 'c' && p[1] == 'p')
3149 	cp = atoi(p + 2);
3150     else if ((idx = enc_canon_search(p)) >= 0)
3151 	cp = enc_canon_table[idx].codepage;
3152     else
3153 	return 0;
3154     if (IsValidCodePage(cp))
3155 	return cp;
3156     return 0;
3157 }
3158 #endif
3159 
3160 # if defined(USE_ICONV) || defined(PROTO)
3161 
3162 static char_u *iconv_string __ARGS((vimconv_T *vcp, char_u *str, int slen, int *unconvlenp));
3163 
3164 /*
3165  * Call iconv_open() with a check if iconv() works properly (there are broken
3166  * versions).
3167  * Returns (void *)-1 if failed.
3168  * (should return iconv_t, but that causes problems with prototypes).
3169  */
3170     void *
3171 my_iconv_open(to, from)
3172     char_u	*to;
3173     char_u	*from;
3174 {
3175     iconv_t	fd;
3176 #define ICONV_TESTLEN 400
3177     char_u	tobuf[ICONV_TESTLEN];
3178     char	*p;
3179     size_t	tolen;
3180     static int	iconv_ok = -1;
3181 
3182     if (iconv_ok == FALSE)
3183 	return (void *)-1;	/* detected a broken iconv() previously */
3184 
3185 #ifdef DYNAMIC_ICONV
3186     /* Check if the iconv.dll can be found. */
3187     if (!iconv_enabled(TRUE))
3188 	return (void *)-1;
3189 #endif
3190 
3191     fd = iconv_open((char *)enc_skip(to), (char *)enc_skip(from));
3192 
3193     if (fd != (iconv_t)-1 && iconv_ok == -1)
3194     {
3195 	/*
3196 	 * Do a dummy iconv() call to check if it actually works.  There is a
3197 	 * version of iconv() on Linux that is broken.  We can't ignore it,
3198 	 * because it's wide-spread.  The symptoms are that after outputting
3199 	 * the initial shift state the "to" pointer is NULL and conversion
3200 	 * stops for no apparent reason after about 8160 characters.
3201 	 */
3202 	p = (char *)tobuf;
3203 	tolen = ICONV_TESTLEN;
3204 	(void)iconv(fd, NULL, NULL, &p, &tolen);
3205 	if (p == NULL)
3206 	{
3207 	    iconv_ok = FALSE;
3208 	    iconv_close(fd);
3209 	    fd = (iconv_t)-1;
3210 	}
3211 	else
3212 	    iconv_ok = TRUE;
3213     }
3214 
3215     return (void *)fd;
3216 }
3217 
3218 /*
3219  * Convert the string "str[slen]" with iconv().
3220  * If "unconvlenp" is not NULL handle the string ending in an incomplete
3221  * sequence and set "*unconvlenp" to the length of it.
3222  * Returns the converted string in allocated memory.  NULL for an error.
3223  */
3224     static char_u *
3225 iconv_string(vcp, str, slen, unconvlenp)
3226     vimconv_T	*vcp;
3227     char_u	*str;
3228     int		slen;
3229     int		*unconvlenp;
3230 {
3231     const char	*from;
3232     size_t	fromlen;
3233     char	*to;
3234     size_t	tolen;
3235     size_t	len = 0;
3236     size_t	done = 0;
3237     char_u	*result = NULL;
3238     char_u	*p;
3239     int		l;
3240 
3241     from = (char *)str;
3242     fromlen = slen;
3243     for (;;)
3244     {
3245 	if (len == 0 || ICONV_ERRNO == ICONV_E2BIG)
3246 	{
3247 	    /* Allocate enough room for most conversions.  When re-allocating
3248 	     * increase the buffer size. */
3249 	    len = len + fromlen * 2 + 40;
3250 	    p = alloc((unsigned)len);
3251 	    if (p != NULL && done > 0)
3252 		mch_memmove(p, result, done);
3253 	    vim_free(result);
3254 	    result = p;
3255 	    if (result == NULL)	/* out of memory */
3256 		break;
3257 	}
3258 
3259 	to = (char *)result + done;
3260 	tolen = len - done - 2;
3261 	/* Avoid a warning for systems with a wrong iconv() prototype by
3262 	 * casting the second argument to void *. */
3263 	if (iconv(vcp->vc_fd, (void *)&from, &fromlen, &to, &tolen)
3264 								!= (size_t)-1)
3265 	{
3266 	    /* Finished, append a NUL. */
3267 	    *to = NUL;
3268 	    break;
3269 	}
3270 
3271 	/* Check both ICONV_EINVAL and EINVAL, because the dynamically loaded
3272 	 * iconv library may use one of them. */
3273 	if (!vcp->vc_fail && unconvlenp != NULL
3274 		&& (ICONV_ERRNO == ICONV_EINVAL || ICONV_ERRNO == EINVAL))
3275 	{
3276 	    /* Handle an incomplete sequence at the end. */
3277 	    *to = NUL;
3278 	    *unconvlenp = (int)fromlen;
3279 	    break;
3280 	}
3281 
3282 	/* Check both ICONV_EILSEQ and EILSEQ, because the dynamically loaded
3283 	 * iconv library may use one of them. */
3284 	else if (!vcp->vc_fail
3285 		&& (ICONV_ERRNO == ICONV_EILSEQ || ICONV_ERRNO == EILSEQ
3286 		    || ICONV_ERRNO == ICONV_EINVAL || ICONV_ERRNO == EINVAL))
3287 	{
3288 	    /* Can't convert: insert a '?' and skip a character.  This assumes
3289 	     * conversion from 'encoding' to something else.  In other
3290 	     * situations we don't know what to skip anyway. */
3291 	    *to++ = '?';
3292 	    if ((*mb_ptr2cells)((char_u *)from) > 1)
3293 		*to++ = '?';
3294 	    if (enc_utf8)
3295 		l = utfc_ptr2len_len((char_u *)from, (int)fromlen);
3296 	    else
3297 	    {
3298 		l = (*mb_ptr2len)((char_u *)from);
3299 		if (l > (int)fromlen)
3300 		    l = (int)fromlen;
3301 	    }
3302 	    from += l;
3303 	    fromlen -= l;
3304 	}
3305 	else if (ICONV_ERRNO != ICONV_E2BIG)
3306 	{
3307 	    /* conversion failed */
3308 	    vim_free(result);
3309 	    result = NULL;
3310 	    break;
3311 	}
3312 	/* Not enough room or skipping illegal sequence. */
3313 	done = to - (char *)result;
3314     }
3315     return result;
3316 }
3317 
3318 #  if defined(DYNAMIC_ICONV) || defined(PROTO)
3319 /*
3320  * Dynamically load the "iconv.dll" on Win32.
3321  */
3322 
3323 #ifndef DYNAMIC_ICONV	    /* just generating prototypes */
3324 # define HINSTANCE int
3325 #endif
3326 static HINSTANCE hIconvDLL = 0;
3327 static HINSTANCE hMsvcrtDLL = 0;
3328 
3329 #  ifndef DYNAMIC_ICONV_DLL
3330 #   define DYNAMIC_ICONV_DLL "iconv.dll"
3331 #   define DYNAMIC_ICONV_DLL_ALT "libiconv.dll"
3332 #  endif
3333 #  ifndef DYNAMIC_MSVCRT_DLL
3334 #   define DYNAMIC_MSVCRT_DLL "msvcrt.dll"
3335 #  endif
3336 
3337 /*
3338  * Try opening the iconv.dll and return TRUE if iconv() can be used.
3339  */
3340     int
3341 iconv_enabled(verbose)
3342     int		verbose;
3343 {
3344     if (hIconvDLL != 0 && hMsvcrtDLL != 0)
3345 	return TRUE;
3346     hIconvDLL = LoadLibrary(DYNAMIC_ICONV_DLL);
3347     if (hIconvDLL == 0)		/* sometimes it's called libiconv.dll */
3348 	hIconvDLL = LoadLibrary(DYNAMIC_ICONV_DLL_ALT);
3349     if (hIconvDLL != 0)
3350 	hMsvcrtDLL = LoadLibrary(DYNAMIC_MSVCRT_DLL);
3351     if (hIconvDLL == 0 || hMsvcrtDLL == 0)
3352     {
3353 	/* Only give the message when 'verbose' is set, otherwise it might be
3354 	 * done whenever a conversion is attempted. */
3355 	if (verbose && p_verbose > 0)
3356 	{
3357 	    verbose_enter();
3358 	    EMSG2(_(e_loadlib),
3359 		    hIconvDLL == 0 ? DYNAMIC_ICONV_DLL : DYNAMIC_MSVCRT_DLL);
3360 	    verbose_leave();
3361 	}
3362 	iconv_end();
3363 	return FALSE;
3364     }
3365 
3366     iconv	= (void *)GetProcAddress(hIconvDLL, "libiconv");
3367     iconv_open	= (void *)GetProcAddress(hIconvDLL, "libiconv_open");
3368     iconv_close	= (void *)GetProcAddress(hIconvDLL, "libiconv_close");
3369     iconvctl	= (void *)GetProcAddress(hIconvDLL, "libiconvctl");
3370     iconv_errno	= (void *)GetProcAddress(hMsvcrtDLL, "_errno");
3371     if (iconv == NULL || iconv_open == NULL || iconv_close == NULL
3372 	    || iconvctl == NULL || iconv_errno == NULL)
3373     {
3374 	iconv_end();
3375 	if (verbose && p_verbose > 0)
3376 	{
3377 	    verbose_enter();
3378 	    EMSG2(_(e_loadfunc), "for libiconv");
3379 	    verbose_leave();
3380 	}
3381 	return FALSE;
3382     }
3383     return TRUE;
3384 }
3385 
3386     void
3387 iconv_end()
3388 {
3389     /* Don't use iconv() when inputting or outputting characters. */
3390     if (input_conv.vc_type == CONV_ICONV)
3391 	convert_setup(&input_conv, NULL, NULL);
3392     if (output_conv.vc_type == CONV_ICONV)
3393 	convert_setup(&output_conv, NULL, NULL);
3394 
3395     if (hIconvDLL != 0)
3396 	FreeLibrary(hIconvDLL);
3397     if (hMsvcrtDLL != 0)
3398 	FreeLibrary(hMsvcrtDLL);
3399     hIconvDLL = 0;
3400     hMsvcrtDLL = 0;
3401 }
3402 #  endif /* DYNAMIC_ICONV */
3403 # endif /* USE_ICONV */
3404 
3405 #endif /* FEAT_MBYTE */
3406 
3407 #if defined(FEAT_XIM) || defined(PROTO)
3408 
3409 # ifdef FEAT_GUI_GTK
3410 static int xim_has_preediting INIT(= FALSE);  /* IM current status */
3411 
3412 /*
3413  * Set preedit_start_col to the current cursor position.
3414  */
3415     static void
3416 init_preedit_start_col(void)
3417 {
3418     if (State & CMDLINE)
3419 	preedit_start_col = cmdline_getvcol_cursor();
3420     else if (curwin != NULL)
3421 	getvcol(curwin, &curwin->w_cursor, &preedit_start_col, NULL, NULL);
3422     /* Prevent that preediting marks the buffer as changed. */
3423     xim_changed_while_preediting = curbuf->b_changed;
3424 }
3425 # endif
3426 
3427 # if defined(HAVE_GTK2) && !defined(PROTO)
3428 
3429 static int im_is_active	       = FALSE;	/* IM is enabled for current mode    */
3430 static int im_preedit_cursor   = 0;	/* cursor offset in characters       */
3431 static int im_preedit_trailing = 0;	/* number of characters after cursor */
3432 
3433 static unsigned long im_commit_handler_id  = 0;
3434 static unsigned int  im_activatekey_keyval = GDK_VoidSymbol;
3435 static unsigned int  im_activatekey_state  = 0;
3436 
3437     void
3438 im_set_active(int active)
3439 {
3440     int was_active;
3441 
3442     was_active = !!im_is_active;
3443     im_is_active = (active && !p_imdisable);
3444 
3445     if (im_is_active != was_active)
3446 	xim_reset();
3447 }
3448 
3449     void
3450 xim_set_focus(int focus)
3451 {
3452     if (xic != NULL)
3453     {
3454 	if (focus)
3455 	    gtk_im_context_focus_in(xic);
3456 	else
3457 	    gtk_im_context_focus_out(xic);
3458     }
3459 }
3460 
3461     void
3462 im_set_position(int row, int col)
3463 {
3464     if (xic != NULL)
3465     {
3466 	GdkRectangle area;
3467 
3468 	area.x = FILL_X(col);
3469 	area.y = FILL_Y(row);
3470 	area.width  = gui.char_width * (mb_lefthalve(row, col) ? 2 : 1);
3471 	area.height = gui.char_height;
3472 
3473 	gtk_im_context_set_cursor_location(xic, &area);
3474     }
3475 }
3476 
3477 #  if 0 || defined(PROTO) /* apparently only used in gui_x11.c */
3478     void
3479 xim_set_preedit(void)
3480 {
3481     im_set_position(gui.row, gui.col);
3482 }
3483 #  endif
3484 
3485     static void
3486 im_add_to_input(char_u *str, int len)
3487 {
3488     /* Convert from 'termencoding' (always "utf-8") to 'encoding' */
3489     if (input_conv.vc_type != CONV_NONE)
3490     {
3491 	str = string_convert(&input_conv, str, &len);
3492 	g_return_if_fail(str != NULL);
3493     }
3494 
3495     add_to_input_buf_csi(str, len);
3496 
3497     if (input_conv.vc_type != CONV_NONE)
3498 	vim_free(str);
3499 
3500     if (p_mh) /* blank out the pointer if necessary */
3501 	gui_mch_mousehide(TRUE);
3502 }
3503 
3504     static void
3505 im_delete_preedit(void)
3506 {
3507     char_u bskey[]  = {CSI, 'k', 'b'};
3508     char_u delkey[] = {CSI, 'k', 'D'};
3509 
3510     if (State & NORMAL)
3511     {
3512 	im_preedit_cursor = 0;
3513 	return;
3514     }
3515     for (; im_preedit_cursor > 0; --im_preedit_cursor)
3516 	add_to_input_buf(bskey, (int)sizeof(bskey));
3517 
3518     for (; im_preedit_trailing > 0; --im_preedit_trailing)
3519 	add_to_input_buf(delkey, (int)sizeof(delkey));
3520 }
3521 
3522 /*
3523  * Move the cursor left by "num_move_back" characters.
3524  * Note that ins_left() checks im_is_preediting() to avoid breaking undo for
3525  * these K_LEFT keys.
3526  */
3527     static void
3528 im_correct_cursor(int num_move_back)
3529 {
3530     char_u backkey[] = {CSI, 'k', 'l'};
3531 
3532     if (State & NORMAL)
3533 	return;
3534 #  ifdef FEAT_RIGHTLEFT
3535     if ((State & CMDLINE) == 0 && curwin != NULL && curwin->w_p_rl)
3536 	backkey[2] = 'r';
3537 #  endif
3538     for (; num_move_back > 0; --num_move_back)
3539 	add_to_input_buf(backkey, (int)sizeof(backkey));
3540 }
3541 
3542 static int xim_expected_char = NUL;
3543 static int xim_ignored_char = FALSE;
3544 
3545 /*
3546  * Update the mode and cursor while in an IM callback.
3547  */
3548     static void
3549 im_show_info(void)
3550 {
3551     int	    old_vgetc_busy;
3552 
3553     old_vgetc_busy = vgetc_busy;
3554     vgetc_busy = TRUE;
3555     showmode();
3556     vgetc_busy = old_vgetc_busy;
3557     setcursor();
3558     out_flush();
3559 }
3560 
3561 /*
3562  * Callback invoked when the user finished preediting.
3563  * Put the final string into the input buffer.
3564  */
3565 /*ARGSUSED0*/
3566     static void
3567 im_commit_cb(GtkIMContext *context, const gchar *str, gpointer data)
3568 {
3569     int	slen = (int)STRLEN(str);
3570     int	add_to_input = TRUE;
3571     int	clen;
3572     int	len = slen;
3573     int	commit_with_preedit = TRUE;
3574     char_u	*im_str, *p;
3575 
3576 #ifdef XIM_DEBUG
3577     xim_log("im_commit_cb(): %s\n", str);
3578 #endif
3579 
3580     /* The imhangul module doesn't reset the preedit string before
3581      * committing.  Call im_delete_preedit() to work around that. */
3582     im_delete_preedit();
3583 
3584     /* Indicate that preediting has finished. */
3585     if (preedit_start_col == MAXCOL)
3586     {
3587 	init_preedit_start_col();
3588 	commit_with_preedit = FALSE;
3589     }
3590 
3591     /* The thing which setting "preedit_start_col" to MAXCOL means that
3592      * "preedit_start_col" will be set forcely when calling
3593      * preedit_changed_cb() next time.
3594      * "preedit_start_col" should not reset with MAXCOL on this part. Vim
3595      * is simulating the preediting by using add_to_input_str(). when
3596      * preedit begin immediately before committed, the typebuf is not
3597      * flushed to screen, then it can't get correct "preedit_start_col".
3598      * Thus, it should calculate the cells by adding cells of the committed
3599      * string. */
3600     if (input_conv.vc_type != CONV_NONE)
3601     {
3602 	im_str = string_convert(&input_conv, (char_u *)str, &len);
3603 	g_return_if_fail(im_str != NULL);
3604     }
3605     else
3606 	im_str = (char_u *)str;
3607     clen = 0;
3608     for (p = im_str; p < im_str + len; p += (*mb_ptr2len)(p))
3609 	clen += (*mb_ptr2cells)(p);
3610     if (input_conv.vc_type != CONV_NONE)
3611 	vim_free(im_str);
3612     preedit_start_col += clen;
3613 
3614     /* Is this a single character that matches a keypad key that's just
3615      * been pressed?  If so, we don't want it to be entered as such - let
3616      * us carry on processing the raw keycode so that it may be used in
3617      * mappings as <kSomething>. */
3618     if (xim_expected_char != NUL)
3619     {
3620 	/* We're currently processing a keypad or other special key */
3621 	if (slen == 1 && str[0] == xim_expected_char)
3622 	{
3623 	    /* It's a match - don't do it here */
3624 	    xim_ignored_char = TRUE;
3625 	    add_to_input = FALSE;
3626 	}
3627 	else
3628 	{
3629 	    /* Not a match */
3630 	    xim_ignored_char = FALSE;
3631 	}
3632     }
3633 
3634     if (add_to_input)
3635 	im_add_to_input((char_u *)str, slen);
3636 
3637     /* Inserting chars while "im_is_active" is set does not cause a change of
3638      * buffer.  When the chars are committed the buffer must be marked as
3639      * changed. */
3640     if (!commit_with_preedit)
3641 	preedit_start_col = MAXCOL;
3642 
3643     /* This flag is used in changed() at next call. */
3644     xim_changed_while_preediting = TRUE;
3645 
3646     if (gtk_main_level() > 0)
3647 	gtk_main_quit();
3648 }
3649 
3650 /*
3651  * Callback invoked after start to the preedit.
3652  */
3653 /*ARGSUSED*/
3654     static void
3655 im_preedit_start_cb(GtkIMContext *context, gpointer data)
3656 {
3657 #ifdef XIM_DEBUG
3658     xim_log("im_preedit_start_cb()\n");
3659 #endif
3660 
3661     im_is_active = TRUE;
3662     gui_update_cursor(TRUE, FALSE);
3663 }
3664 
3665 /*
3666  * Callback invoked after end to the preedit.
3667  */
3668 /*ARGSUSED*/
3669     static void
3670 im_preedit_end_cb(GtkIMContext *context, gpointer data)
3671 {
3672 #ifdef XIM_DEBUG
3673     xim_log("im_preedit_end_cb()\n");
3674 #endif
3675     im_delete_preedit();
3676 
3677     /* Indicate that preediting has finished */
3678     preedit_start_col = MAXCOL;
3679     xim_has_preediting = FALSE;
3680 
3681     im_is_active = FALSE;
3682     gui_update_cursor(TRUE, FALSE);
3683     im_show_info();
3684 }
3685 
3686 /*
3687  * Callback invoked after changes to the preedit string.  If the preedit
3688  * string was empty before, remember the preedit start column so we know
3689  * where to apply feedback attributes.  Delete the previous preedit string
3690  * if there was one, save the new preedit cursor offset, and put the new
3691  * string into the input buffer.
3692  *
3693  * TODO: The pragmatic "put into input buffer" approach used here has
3694  *       several fundamental problems:
3695  *
3696  * - The characters in the preedit string are subject to remapping.
3697  *   That's broken, only the finally committed string should be remapped.
3698  *
3699  * - There is a race condition involved:  The retrieved value for the
3700  *   current cursor position will be wrong if any unprocessed characters
3701  *   are still queued in the input buffer.
3702  *
3703  * - Due to the lack of synchronization between the file buffer in memory
3704  *   and any typed characters, it's practically impossible to implement the
3705  *   "retrieve_surrounding" and "delete_surrounding" signals reliably.  IM
3706  *   modules for languages such as Thai are likely to rely on this feature
3707  *   for proper operation.
3708  *
3709  * Conclusions:  I think support for preediting needs to be moved to the
3710  * core parts of Vim.  Ideally, until it has been committed, the preediting
3711  * string should only be displayed and not affect the buffer content at all.
3712  * The question how to deal with the synchronization issue still remains.
3713  * Circumventing the input buffer is probably not desirable.  Anyway, I think
3714  * implementing "retrieve_surrounding" is the only hard problem.
3715  *
3716  * One way to solve all of this in a clean manner would be to queue all key
3717  * press/release events "as is" in the input buffer, and apply the IM filtering
3718  * at the receiving end of the queue.  This, however, would have a rather large
3719  * impact on the code base.  If there is an easy way to force processing of all
3720  * remaining input from within the "retrieve_surrounding" signal handler, this
3721  * might not be necessary.  Gotta ask on vim-dev for opinions.
3722  */
3723 /*ARGSUSED1*/
3724     static void
3725 im_preedit_changed_cb(GtkIMContext *context, gpointer data)
3726 {
3727     char    *preedit_string = NULL;
3728     int	    cursor_index    = 0;
3729     int	    num_move_back   = 0;
3730     char_u  *str;
3731     char_u  *p;
3732     int	    i;
3733 
3734     gtk_im_context_get_preedit_string(context,
3735 				      &preedit_string, NULL,
3736 				      &cursor_index);
3737 
3738 #ifdef XIM_DEBUG
3739     xim_log("im_preedit_changed_cb(): %s\n", preedit_string);
3740 #endif
3741 
3742     g_return_if_fail(preedit_string != NULL); /* just in case */
3743 
3744     /* If preedit_start_col is MAXCOL set it to the current cursor position. */
3745     if (preedit_start_col == MAXCOL && preedit_string[0] != '\0')
3746     {
3747 	xim_has_preediting = TRUE;
3748 
3749 	/* Urgh, this breaks if the input buffer isn't empty now */
3750 	init_preedit_start_col();
3751     }
3752     else if (cursor_index == 0 && preedit_string[0] == '\0')
3753     {
3754 	xim_has_preediting = FALSE;
3755 
3756 	/* If at the start position (after typing backspace)
3757 	 * preedit_start_col must be reset. */
3758 	preedit_start_col = MAXCOL;
3759     }
3760 
3761     im_delete_preedit();
3762 
3763     /*
3764      * Compute the end of the preediting area: "preedit_end_col".
3765      * According to the documentation of gtk_im_context_get_preedit_string(),
3766      * the cursor_pos output argument returns the offset in bytes.  This is
3767      * unfortunately not true -- real life shows the offset is in characters,
3768      * and the GTK+ source code agrees with me.  Will file a bug later.
3769      */
3770     if (preedit_start_col != MAXCOL)
3771 	preedit_end_col = preedit_start_col;
3772     str = (char_u *)preedit_string;
3773     for (p = str, i = 0; *p != NUL; p += utf_byte2len(*p), ++i)
3774     {
3775 	int is_composing;
3776 
3777 	is_composing = ((*p & 0x80) != 0 && utf_iscomposing(utf_ptr2char(p)));
3778 	/*
3779 	 * These offsets are used as counters when generating <BS> and <Del>
3780 	 * to delete the preedit string.  So don't count composing characters
3781 	 * unless 'delcombine' is enabled.
3782 	 */
3783 	if (!is_composing || p_deco)
3784 	{
3785 	    if (i < cursor_index)
3786 		++im_preedit_cursor;
3787 	    else
3788 		++im_preedit_trailing;
3789 	}
3790 	if (!is_composing && i >= cursor_index)
3791 	{
3792 	    /* This is essentially the same as im_preedit_trailing, except
3793 	     * composing characters are not counted even if p_deco is set. */
3794 	    ++num_move_back;
3795 	}
3796 	if (preedit_start_col != MAXCOL)
3797 	    preedit_end_col += utf_ptr2cells(p);
3798     }
3799 
3800     if (p > str)
3801     {
3802 	im_add_to_input(str, (int)(p - str));
3803 	im_correct_cursor(num_move_back);
3804     }
3805 
3806     g_free(preedit_string);
3807 
3808     if (gtk_main_level() > 0)
3809 	gtk_main_quit();
3810 }
3811 
3812 /*
3813  * Translate the Pango attributes at iter to Vim highlighting attributes.
3814  * Ignore attributes not supported by Vim highlighting.  This shouldn't have
3815  * too much impact -- right now we handle even more attributes than necessary
3816  * for the IM modules I tested with.
3817  */
3818     static int
3819 translate_pango_attributes(PangoAttrIterator *iter)
3820 {
3821     PangoAttribute  *attr;
3822     int		    char_attr = HL_NORMAL;
3823 
3824     attr = pango_attr_iterator_get(iter, PANGO_ATTR_UNDERLINE);
3825     if (attr != NULL && ((PangoAttrInt *)attr)->value
3826 						 != (int)PANGO_UNDERLINE_NONE)
3827 	char_attr |= HL_UNDERLINE;
3828 
3829     attr = pango_attr_iterator_get(iter, PANGO_ATTR_WEIGHT);
3830     if (attr != NULL && ((PangoAttrInt *)attr)->value >= (int)PANGO_WEIGHT_BOLD)
3831 	char_attr |= HL_BOLD;
3832 
3833     attr = pango_attr_iterator_get(iter, PANGO_ATTR_STYLE);
3834     if (attr != NULL && ((PangoAttrInt *)attr)->value
3835 						   != (int)PANGO_STYLE_NORMAL)
3836 	char_attr |= HL_ITALIC;
3837 
3838     attr = pango_attr_iterator_get(iter, PANGO_ATTR_BACKGROUND);
3839     if (attr != NULL)
3840     {
3841 	const PangoColor *color = &((PangoAttrColor *)attr)->color;
3842 
3843 	/* Assume inverse if black background is requested */
3844 	if ((color->red | color->green | color->blue) == 0)
3845 	    char_attr |= HL_INVERSE;
3846     }
3847 
3848     return char_attr;
3849 }
3850 
3851 /*
3852  * Retrieve the highlighting attributes at column col in the preedit string.
3853  * Return -1 if not in preediting mode or if col is out of range.
3854  */
3855     int
3856 im_get_feedback_attr(int col)
3857 {
3858     char	    *preedit_string = NULL;
3859     PangoAttrList   *attr_list	    = NULL;
3860     int		    char_attr	    = -1;
3861 
3862     if (xic == NULL)
3863 	return char_attr;
3864 
3865     gtk_im_context_get_preedit_string(xic, &preedit_string, &attr_list, NULL);
3866 
3867     if (preedit_string != NULL && attr_list != NULL)
3868     {
3869 	int idx;
3870 
3871 	/* Get the byte index as used by PangoAttrIterator */
3872 	for (idx = 0; col > 0 && preedit_string[idx] != '\0'; --col)
3873 	    idx += utfc_ptr2len((char_u *)preedit_string + idx);
3874 
3875 	if (preedit_string[idx] != '\0')
3876 	{
3877 	    PangoAttrIterator	*iter;
3878 	    int			start, end;
3879 
3880 	    char_attr = HL_NORMAL;
3881 	    iter = pango_attr_list_get_iterator(attr_list);
3882 
3883 	    /* Extract all relevant attributes from the list. */
3884 	    do
3885 	    {
3886 		pango_attr_iterator_range(iter, &start, &end);
3887 
3888 		if (idx >= start && idx < end)
3889 		    char_attr |= translate_pango_attributes(iter);
3890 	    }
3891 	    while (pango_attr_iterator_next(iter));
3892 
3893 	    pango_attr_iterator_destroy(iter);
3894 	}
3895     }
3896 
3897     if (attr_list != NULL)
3898 	pango_attr_list_unref(attr_list);
3899     g_free(preedit_string);
3900 
3901     return char_attr;
3902 }
3903 
3904     void
3905 xim_init(void)
3906 {
3907 #ifdef XIM_DEBUG
3908     xim_log("xim_init()\n");
3909 #endif
3910 
3911     g_return_if_fail(gui.drawarea != NULL);
3912     g_return_if_fail(gui.drawarea->window != NULL);
3913 
3914     xic = gtk_im_multicontext_new();
3915     g_object_ref(xic);
3916 
3917     im_commit_handler_id = g_signal_connect(G_OBJECT(xic), "commit",
3918 					    G_CALLBACK(&im_commit_cb), NULL);
3919     g_signal_connect(G_OBJECT(xic), "preedit_changed",
3920 		     G_CALLBACK(&im_preedit_changed_cb), NULL);
3921     g_signal_connect(G_OBJECT(xic), "preedit_start",
3922 		     G_CALLBACK(&im_preedit_start_cb), NULL);
3923     g_signal_connect(G_OBJECT(xic), "preedit_end",
3924 		     G_CALLBACK(&im_preedit_end_cb), NULL);
3925 
3926     gtk_im_context_set_client_window(xic, gui.drawarea->window);
3927 }
3928 
3929     void
3930 im_shutdown(void)
3931 {
3932 #ifdef XIM_DEBUG
3933     xim_log("im_shutdown()\n");
3934 #endif
3935 
3936     if (xic != NULL)
3937     {
3938 	gtk_im_context_focus_out(xic);
3939 	g_object_unref(xic);
3940 	xic = NULL;
3941     }
3942     im_is_active = FALSE;
3943     im_commit_handler_id = 0;
3944     preedit_start_col = MAXCOL;
3945     xim_has_preediting = FALSE;
3946 }
3947 
3948 /*
3949  * Convert the string argument to keyval and state for GdkEventKey.
3950  * If str is valid return TRUE, otherwise FALSE.
3951  *
3952  * See 'imactivatekey' for documentation of the format.
3953  */
3954     static int
3955 im_string_to_keyval(const char *str, unsigned int *keyval, unsigned int *state)
3956 {
3957     const char	    *mods_end;
3958     unsigned	    tmp_keyval;
3959     unsigned	    tmp_state = 0;
3960 
3961     mods_end = strrchr(str, '-');
3962     mods_end = (mods_end != NULL) ? mods_end + 1 : str;
3963 
3964     /* Parse modifier keys */
3965     while (str < mods_end)
3966 	switch (*str++)
3967 	{
3968 	    case '-':							break;
3969 	    case 'S': case 's': tmp_state |= (unsigned)GDK_SHIFT_MASK;	break;
3970 	    case 'L': case 'l': tmp_state |= (unsigned)GDK_LOCK_MASK;	break;
3971 	    case 'C': case 'c': tmp_state |= (unsigned)GDK_CONTROL_MASK;break;
3972 	    case '1':		tmp_state |= (unsigned)GDK_MOD1_MASK;	break;
3973 	    case '2':		tmp_state |= (unsigned)GDK_MOD2_MASK;	break;
3974 	    case '3':		tmp_state |= (unsigned)GDK_MOD3_MASK;	break;
3975 	    case '4':		tmp_state |= (unsigned)GDK_MOD4_MASK;	break;
3976 	    case '5':		tmp_state |= (unsigned)GDK_MOD5_MASK;	break;
3977 	    default:
3978 		return FALSE;
3979 	}
3980 
3981     tmp_keyval = gdk_keyval_from_name(str);
3982 
3983     if (tmp_keyval == 0 || tmp_keyval == GDK_VoidSymbol)
3984 	return FALSE;
3985 
3986     if (keyval != NULL)
3987 	*keyval = tmp_keyval;
3988     if (state != NULL)
3989 	*state = tmp_state;
3990 
3991     return TRUE;
3992 }
3993 
3994 /*
3995  * Return TRUE if p_imak is valid, otherwise FALSE.  As a special case, an
3996  * empty string is also regarded as valid.
3997  *
3998  * Note: The numerical key value of p_imak is cached if it was valid; thus
3999  * boldly assuming im_xim_isvalid_imactivate() will always be called whenever
4000  * 'imak' changes.  This is currently the case but not obvious -- should
4001  * probably rename the function for clarity.
4002  */
4003     int
4004 im_xim_isvalid_imactivate(void)
4005 {
4006     if (p_imak[0] == NUL)
4007     {
4008 	im_activatekey_keyval = GDK_VoidSymbol;
4009 	im_activatekey_state  = 0;
4010 	return TRUE;
4011     }
4012 
4013     return im_string_to_keyval((const char *)p_imak,
4014 			       &im_activatekey_keyval,
4015 			       &im_activatekey_state);
4016 }
4017 
4018     static void
4019 im_synthesize_keypress(unsigned int keyval, unsigned int state)
4020 {
4021     GdkEventKey *event;
4022 
4023 #  ifdef HAVE_GTK_MULTIHEAD
4024     event = (GdkEventKey *)gdk_event_new(GDK_KEY_PRESS);
4025     g_object_ref(gui.drawarea->window); /* unreffed by gdk_event_free() */
4026 #  else
4027     event = (GdkEventKey *)g_malloc0((gulong)sizeof(GdkEvent));
4028     event->type = GDK_KEY_PRESS;
4029 #  endif
4030     event->window = gui.drawarea->window;
4031     event->send_event = TRUE;
4032     event->time = GDK_CURRENT_TIME;
4033     event->state  = state;
4034     event->keyval = keyval;
4035     event->hardware_keycode = /* needed for XIM */
4036 	XKeysymToKeycode(GDK_WINDOW_XDISPLAY(event->window), (KeySym)keyval);
4037     event->length = 0;
4038     event->string = NULL;
4039 
4040     gtk_im_context_filter_keypress(xic, event);
4041 
4042     /* For consistency, also send the corresponding release event. */
4043     event->type = GDK_KEY_RELEASE;
4044     event->send_event = FALSE;
4045     gtk_im_context_filter_keypress(xic, event);
4046 
4047 #  ifdef HAVE_GTK_MULTIHEAD
4048     gdk_event_free((GdkEvent *)event);
4049 #  else
4050     g_free(event);
4051 #  endif
4052 }
4053 
4054     void
4055 xim_reset(void)
4056 {
4057     if (xic != NULL)
4058     {
4059 	/*
4060 	 * The third-party imhangul module (and maybe others too) ignores
4061 	 * gtk_im_context_reset() or at least doesn't reset the active state.
4062 	 * Thus sending imactivatekey would turn it off if it was on before,
4063 	 * which is clearly not what we want.  Fortunately we can work around
4064 	 * that for imhangul by sending GDK_Escape, but I don't know if it
4065 	 * works with all IM modules that support an activation key :/
4066 	 *
4067 	 * An alternative approach would be to destroy the IM context and
4068 	 * recreate it.  But that means loading/unloading the IM module on
4069 	 * every mode switch, which causes a quite noticable delay even on
4070 	 * my rather fast box...
4071 	 * *
4072 	 * Moreover, there are some XIM which cannot respond to
4073 	 * im_synthesize_keypress(). we hope that they reset by
4074 	 * xim_shutdown().
4075 	 */
4076 	if (im_activatekey_keyval != GDK_VoidSymbol && im_is_active)
4077 	    im_synthesize_keypress(GDK_Escape, 0U);
4078 
4079 	gtk_im_context_reset(xic);
4080 
4081 	/*
4082 	 * HACK for Ami: This sequence of function calls makes Ami handle
4083 	 * the IM reset gratiously, without breaking loads of other stuff.
4084 	 * It seems to force English mode as well, which is exactly what we
4085 	 * want because it makes the Ami status display work reliably.
4086 	 */
4087 	gtk_im_context_set_use_preedit(xic, FALSE);
4088 
4089 	if (p_imdisable)
4090 	    im_shutdown();
4091 	else
4092 	{
4093 	    gtk_im_context_set_use_preedit(xic, TRUE);
4094 	    xim_set_focus(gui.in_focus);
4095 
4096 	    if (im_activatekey_keyval != GDK_VoidSymbol)
4097 	    {
4098 		if (im_is_active)
4099 		{
4100 		    g_signal_handler_block(xic, im_commit_handler_id);
4101 		    im_synthesize_keypress(im_activatekey_keyval,
4102 						    im_activatekey_state);
4103 		    g_signal_handler_unblock(xic, im_commit_handler_id);
4104 		}
4105 	    }
4106 	    else
4107 	    {
4108 		im_shutdown();
4109 		xim_init();
4110 		xim_set_focus(gui.in_focus);
4111 	    }
4112 	}
4113     }
4114 
4115     preedit_start_col = MAXCOL;
4116     xim_has_preediting = FALSE;
4117 }
4118 
4119     int
4120 xim_queue_key_press_event(GdkEventKey *event, int down)
4121 {
4122     if (down)
4123     {
4124 	/*
4125 	 * Workaround GTK2 XIM 'feature' that always converts keypad keys to
4126 	 * chars., even when not part of an IM sequence (ref. feature of
4127 	 * gdk/gdkkeyuni.c).
4128 	 * Flag any keypad keys that might represent a single char.
4129 	 * If this (on its own - i.e., not part of an IM sequence) is
4130 	 * committed while we're processing one of these keys, we can ignore
4131 	 * that commit and go ahead & process it ourselves.  That way we can
4132 	 * still distinguish keypad keys for use in mappings.
4133 	 */
4134 	switch (event->keyval)
4135 	{
4136 	    case GDK_KP_Add:      xim_expected_char = '+';  break;
4137 	    case GDK_KP_Subtract: xim_expected_char = '-';  break;
4138 	    case GDK_KP_Divide:   xim_expected_char = '/';  break;
4139 	    case GDK_KP_Multiply: xim_expected_char = '*';  break;
4140 	    case GDK_KP_Decimal:  xim_expected_char = '.';  break;
4141 	    case GDK_KP_Equal:    xim_expected_char = '=';  break;
4142 	    case GDK_KP_0:	  xim_expected_char = '0';  break;
4143 	    case GDK_KP_1:	  xim_expected_char = '1';  break;
4144 	    case GDK_KP_2:	  xim_expected_char = '2';  break;
4145 	    case GDK_KP_3:	  xim_expected_char = '3';  break;
4146 	    case GDK_KP_4:	  xim_expected_char = '4';  break;
4147 	    case GDK_KP_5:	  xim_expected_char = '5';  break;
4148 	    case GDK_KP_6:	  xim_expected_char = '6';  break;
4149 	    case GDK_KP_7:	  xim_expected_char = '7';  break;
4150 	    case GDK_KP_8:	  xim_expected_char = '8';  break;
4151 	    case GDK_KP_9:	  xim_expected_char = '9';  break;
4152 	    default:		  xim_expected_char = NUL;
4153 	}
4154 	xim_ignored_char = FALSE;
4155     }
4156 
4157     /*
4158      * When typing fFtT, XIM may be activated. Thus it must pass
4159      * gtk_im_context_filter_keypress() in Normal mode.
4160      * And while doing :sh too.
4161      */
4162     if (xic != NULL && !p_imdisable
4163 		    && (State & (INSERT | CMDLINE | NORMAL | EXTERNCMD)) != 0)
4164     {
4165 	/*
4166 	 * Filter 'imactivatekey' and map it to CTRL-^.  This way, Vim is
4167 	 * always aware of the current status of IM, and can even emulate
4168 	 * the activation key for modules that don't support one.
4169 	 */
4170 	if (event->keyval == im_activatekey_keyval
4171 	     && (event->state & im_activatekey_state) == im_activatekey_state)
4172 	{
4173 	    unsigned int state_mask;
4174 
4175 	    /* Require the state of the 3 most used modifiers to match exactly.
4176 	     * Otherwise e.g. <S-C-space> would be unusable for other purposes
4177 	     * if the IM activate key is <S-space>. */
4178 	    state_mask  = im_activatekey_state;
4179 	    state_mask |= ((int)GDK_SHIFT_MASK | (int)GDK_CONTROL_MASK
4180 							| (int)GDK_MOD1_MASK);
4181 
4182 	    if ((event->state & state_mask) != im_activatekey_state)
4183 		return FALSE;
4184 
4185 	    /* Don't send it a second time on GDK_KEY_RELEASE. */
4186 	    if (event->type != GDK_KEY_PRESS)
4187 		return TRUE;
4188 
4189 	    if (map_to_exists_mode((char_u *)"", LANGMAP, FALSE))
4190 	    {
4191 		im_set_active(FALSE);
4192 
4193 		/* ":lmap" mappings exists, toggle use of mappings. */
4194 		State ^= LANGMAP;
4195 		if (State & LANGMAP)
4196 		{
4197 		    curbuf->b_p_iminsert = B_IMODE_NONE;
4198 		    State &= ~LANGMAP;
4199 		}
4200 		else
4201 		{
4202 		    curbuf->b_p_iminsert = B_IMODE_LMAP;
4203 		    State |= LANGMAP;
4204 		}
4205 		return TRUE;
4206 	    }
4207 
4208 	    return gtk_im_context_filter_keypress(xic, event);
4209 	}
4210 
4211 	/* Don't filter events through the IM context if IM isn't active
4212 	 * right now.  Unlike with GTK+ 1.2 we cannot rely on the IM module
4213 	 * not doing anything before the activation key was sent. */
4214 	if (im_activatekey_keyval == GDK_VoidSymbol || im_is_active)
4215 	{
4216 	    int imresult = gtk_im_context_filter_keypress(xic, event);
4217 
4218 	    /* Some XIM send following sequence:
4219 	     * 1. preedited string.
4220 	     * 2. committed string.
4221 	     * 3. line changed key.
4222 	     * 4. preedited string.
4223 	     * 5. remove preedited string.
4224 	     * if 3, Vim can't move back the above line for 5.
4225 	     * thus, this part should not parse the key. */
4226 	    if (!imresult && preedit_start_col != MAXCOL
4227 					       && event->keyval == GDK_Return)
4228 	    {
4229 		im_synthesize_keypress(GDK_Return, 0U);
4230 		return FALSE;
4231 	    }
4232 
4233 	    /* If XIM tried to commit a keypad key as a single char.,
4234 	     * ignore it so we can use the keypad key 'raw', for mappings. */
4235 	    if (xim_expected_char != NUL && xim_ignored_char)
4236 		/* We had a keypad key, and XIM tried to thieve it */
4237 		return FALSE;
4238 
4239 	    /* Normal processing */
4240 	    return imresult;
4241 	}
4242     }
4243 
4244     return FALSE;
4245 }
4246 
4247     int
4248 im_get_status(void)
4249 {
4250     return im_is_active;
4251 }
4252 
4253 # else /* !HAVE_GTK2 */
4254 
4255 static int	xim_is_active = FALSE;  /* XIM should be active in the current
4256 					   mode */
4257 static int	xim_has_focus = FALSE;	/* XIM is really being used for Vim */
4258 #ifdef FEAT_GUI_X11
4259 static XIMStyle	input_style;
4260 static int	status_area_enabled = TRUE;
4261 #endif
4262 
4263 #ifdef FEAT_GUI_GTK
4264 # ifdef WIN3264
4265 #  include <gdk/gdkwin32.h>
4266 # else
4267 #  include <gdk/gdkx.h>
4268 # endif
4269 #else
4270 # ifdef PROTO
4271 /* Define a few things to be able to generate prototypes while not configured
4272  * for GTK. */
4273 #  define GSList int
4274 #  define gboolean int
4275    typedef int GdkEvent;
4276    typedef int GdkEventKey;
4277 #  define GdkIC int
4278 # endif
4279 #endif
4280 
4281 #if defined(FEAT_GUI_GTK) || defined(PROTO)
4282 static int	preedit_buf_len = 0;
4283 static int	xim_can_preediting INIT(= FALSE);	/* XIM in showmode() */
4284 static int	xim_input_style;
4285 #ifndef FEAT_GUI_GTK
4286 # define gboolean int
4287 #endif
4288 static gboolean	use_status_area = 0;
4289 
4290 static int im_xim_str2keycode __ARGS((unsigned int *code, unsigned int *state));
4291 static void im_xim_send_event_imactivate __ARGS((void));
4292 
4293 /*
4294  * Convert string to keycode and state for XKeyEvent.
4295  * When string is valid return OK, when invalid return FAIL.
4296  *
4297  * See 'imactivatekey' documentation for the format.
4298  */
4299     static int
4300 im_xim_str2keycode(code, state)
4301     unsigned int *code;
4302     unsigned int *state;
4303 {
4304     int		retval = OK;
4305     int		len;
4306     unsigned	keycode = 0, keystate = 0;
4307     Window	window;
4308     Display	*display;
4309     char_u	*flag_end;
4310     char_u	*str;
4311 
4312     if (*p_imak != NUL)
4313     {
4314 	len = STRLEN(p_imak);
4315 	for (flag_end = p_imak + len - 1;
4316 			    flag_end > p_imak && *flag_end != '-'; --flag_end)
4317 	    ;
4318 
4319 	/* Parse modifier keys */
4320 	for (str = p_imak; str < flag_end; ++str)
4321 	{
4322 	    switch (*str)
4323 	    {
4324 		case 's': case 'S':
4325 		    keystate |= ShiftMask;
4326 		    break;
4327 		case 'l': case 'L':
4328 		    keystate |= LockMask;
4329 		    break;
4330 		case 'c': case 'C':
4331 		    keystate |= ControlMask;
4332 		    break;
4333 		case '1':
4334 		    keystate |= Mod1Mask;
4335 		    break;
4336 		case '2':
4337 		    keystate |= Mod2Mask;
4338 		    break;
4339 		case '3':
4340 		    keystate |= Mod3Mask;
4341 		    break;
4342 		case '4':
4343 		    keystate |= Mod4Mask;
4344 		    break;
4345 		case '5':
4346 		    keystate |= Mod5Mask;
4347 		    break;
4348 		case '-':
4349 		    break;
4350 		default:
4351 		    retval = FAIL;
4352 	    }
4353 	}
4354 	if (*str == '-')
4355 	    ++str;
4356 
4357 	/* Get keycode from string. */
4358 	gui_get_x11_windis(&window, &display);
4359 	if (display)
4360 	    keycode = XKeysymToKeycode(display, XStringToKeysym((char *)str));
4361 	if (keycode == 0)
4362 	    retval = FAIL;
4363 
4364 	if (code != NULL)
4365 	    *code = keycode;
4366 	if (state != NULL)
4367 	    *state = keystate;
4368     }
4369     return retval;
4370 }
4371 
4372     static void
4373 im_xim_send_event_imactivate()
4374 {
4375     /* Force turn on preedit state by symulate keypress event.
4376      * Keycode and state is specified by 'imactivatekey'.
4377      */
4378     XKeyEvent ev;
4379 
4380     gui_get_x11_windis(&ev.window, &ev.display);
4381     ev.root = RootWindow(ev.display, DefaultScreen(ev.display));
4382     ev.subwindow = None;
4383     ev.time = CurrentTime;
4384     ev.x = 1;
4385     ev.y = 1;
4386     ev.x_root = 1;
4387     ev.y_root = 1;
4388     ev.same_screen = 1;
4389     ev.type = KeyPress;
4390     if (im_xim_str2keycode(&ev.keycode, &ev.state) == OK)
4391 	XSendEvent(ev.display, ev.window, 1, KeyPressMask, (XEvent*)&ev);
4392 }
4393 
4394 /*
4395  * Return TRUE if 'imactivatekey' has a valid value.
4396  */
4397     int
4398 im_xim_isvalid_imactivate()
4399 {
4400     return im_xim_str2keycode(NULL, NULL) == OK;
4401 }
4402 #endif /* FEAT_GUI_GTK */
4403 
4404 /*
4405  * Switch using XIM on/off.  This is used by the code that changes "State".
4406  */
4407     void
4408 im_set_active(active)
4409     int		active;
4410 {
4411     if (xic == NULL)
4412 	return;
4413 
4414     /* If 'imdisable' is set, XIM is never active. */
4415     if (p_imdisable)
4416 	active = FALSE;
4417 #if !defined (FEAT_GUI_GTK)
4418     else if (input_style & XIMPreeditPosition)
4419 	/* There is a problem in switching XIM off when preediting is used,
4420 	 * and it is not clear how this can be solved.  For now, keep XIM on
4421 	 * all the time, like it was done in Vim 5.8. */
4422 	active = TRUE;
4423 #endif
4424 
4425     /* Remember the active state, it is needed when Vim gets keyboard focus. */
4426     xim_is_active = active;
4427 
4428 #ifdef FEAT_GUI_GTK
4429     /* When 'imactivatekey' has valid key-string, try to control XIM preedit
4430      * state.  When 'imactivatekey' has no or invalid string, try old XIM
4431      * focus control.
4432      */
4433     if (*p_imak != NUL)
4434     {
4435 	/* BASIC STRATEGY:
4436 	 * Destroy old Input Context (XIC), and create new one.  New XIC
4437 	 * would have a state of preedit that is off.  When argument:active
4438 	 * is false, that's all.  Else argument:active is true, send a key
4439 	 * event specified by 'imactivatekey' to activate XIM preedit state.
4440 	 */
4441 
4442 	xim_is_active = TRUE; /* Disable old XIM focus control */
4443 	/* If we can monitor preedit state with preedit callback functions,
4444 	 * try least creation of new XIC.
4445 	 */
4446 	if (xim_input_style & (int)GDK_IM_PREEDIT_CALLBACKS)
4447 	{
4448 	    if (xim_can_preediting && !active)
4449 	    {
4450 		/* Force turn off preedit state.  With some IM
4451 		 * implementations, we cannot turn off preedit state by
4452 		 * symulate keypress event.  It is why using such a method
4453 		 * that destroy old IC (input context), and create new one.
4454 		 * When create new IC, its preedit state is usually off.
4455 		 */
4456 		xim_reset();
4457 		xim_set_focus(FALSE);
4458 		gdk_ic_destroy(xic);
4459 		xim_init();
4460 		xim_can_preediting = FALSE;
4461 	    }
4462 	    else if (!xim_can_preediting && active)
4463 		im_xim_send_event_imactivate();
4464 	}
4465 	else
4466 	{
4467 	    /* First, force destroy old IC, and create new one.  It
4468 	     * symulates "turning off preedit state".
4469 	     */
4470 	    xim_set_focus(FALSE);
4471 	    gdk_ic_destroy(xic);
4472 	    xim_init();
4473 	    xim_can_preediting = FALSE;
4474 
4475 	    /* 2nd, when requested to activate IM, symulate this by sending
4476 	     * the event.
4477 	     */
4478 	    if (active)
4479 	    {
4480 		im_xim_send_event_imactivate();
4481 		xim_can_preediting = TRUE;
4482 	    }
4483 	}
4484     }
4485     else
4486     {
4487 # ifndef XIMPreeditUnKnown
4488 	/* X11R5 doesn't have these, it looks safe enough to define here. */
4489 	typedef unsigned long XIMPreeditState;
4490 #  define XIMPreeditUnKnown	0L
4491 #  define XIMPreeditEnable	1L
4492 #  define XIMPreeditDisable	(1L<<1)
4493 #  define XNPreeditState	"preeditState"
4494 # endif
4495 	XIMPreeditState preedit_state = XIMPreeditUnKnown;
4496 	XVaNestedList preedit_attr;
4497 	XIC pxic;
4498 
4499 	preedit_attr = XVaCreateNestedList(0,
4500 				XNPreeditState, &preedit_state,
4501 				NULL);
4502 	pxic = ((GdkICPrivate *)xic)->xic;
4503 
4504 	if (!XGetICValues(pxic, XNPreeditAttributes, preedit_attr, NULL))
4505 	{
4506 	    XFree(preedit_attr);
4507 	    preedit_attr = XVaCreateNestedList(0,
4508 				XNPreeditState,
4509 				active ? XIMPreeditEnable : XIMPreeditDisable,
4510 				NULL);
4511 	    XSetICValues(pxic, XNPreeditAttributes, preedit_attr, NULL);
4512 	    xim_can_preediting = active;
4513 	    xim_is_active = active;
4514 	}
4515 	XFree(preedit_attr);
4516     }
4517     if (xim_input_style & XIMPreeditCallbacks)
4518     {
4519 	preedit_buf_len = 0;
4520 	init_preedit_start_col();
4521     }
4522 #else
4523 # if 0
4524 	/* When had tested kinput2 + canna + Athena GUI version with
4525 	 * 'imactivatekey' is "s-space", im_xim_send_event_imactivate() did not
4526 	 * work correctly.  It just inserted one space.  I don't know why we
4527 	 * couldn't switch state of XIM preediting.  This is reason why these
4528 	 * codes are commented out.
4529 	 */
4530 	/* First, force destroy old IC, and create new one.  It symulates
4531 	 * "turning off preedit state".
4532 	 */
4533 	xim_set_focus(FALSE);
4534 	XDestroyIC(xic);
4535 	xic = NULL;
4536 	xim_init();
4537 
4538 	/* 2nd, when requested to activate IM, symulate this by sending the
4539 	 * event.
4540 	 */
4541 	if (active)
4542 	    im_xim_send_event_imactivate();
4543 # endif
4544 #endif
4545     xim_set_preedit();
4546 }
4547 
4548 /*
4549  * Adjust using XIM for gaining or losing keyboard focus.  Also called when
4550  * "xim_is_active" changes.
4551  */
4552     void
4553 xim_set_focus(focus)
4554     int		focus;
4555 {
4556     if (xic == NULL)
4557 	return;
4558 
4559     /*
4560      * XIM only gets focus when the Vim window has keyboard focus and XIM has
4561      * been set active for the current mode.
4562      */
4563     if (focus && xim_is_active)
4564     {
4565 	if (!xim_has_focus)
4566 	{
4567 	    xim_has_focus = TRUE;
4568 #ifdef FEAT_GUI_GTK
4569 	    gdk_im_begin(xic, gui.drawarea->window);
4570 #else
4571 	    XSetICFocus(xic);
4572 #endif
4573 	}
4574     }
4575     else
4576     {
4577 	if (xim_has_focus)
4578 	{
4579 	    xim_has_focus = FALSE;
4580 #ifdef FEAT_GUI_GTK
4581 	    gdk_im_end();
4582 #else
4583 	    XUnsetICFocus(xic);
4584 #endif
4585 	}
4586     }
4587 }
4588 
4589 /*ARGSUSED*/
4590     void
4591 im_set_position(row, col)
4592     int		row;
4593     int		col;
4594 {
4595     xim_set_preedit();
4596 }
4597 
4598 /*
4599  * Set the XIM to the current cursor position.
4600  */
4601     void
4602 xim_set_preedit()
4603 {
4604     if (xic == NULL)
4605 	return;
4606 
4607     xim_set_focus(TRUE);
4608 
4609 #ifdef FEAT_GUI_GTK
4610     if (gdk_im_ready())
4611     {
4612 	int		attrmask;
4613 	GdkICAttr	*attr;
4614 
4615 	if (!xic_attr)
4616 	    return;
4617 
4618 	attr = xic_attr;
4619 	attrmask = 0;
4620 
4621 # ifdef FEAT_XFONTSET
4622 	if ((xim_input_style & (int)GDK_IM_PREEDIT_POSITION)
4623 		&& gui.fontset != NOFONTSET
4624 		&& gui.fontset->type == GDK_FONT_FONTSET)
4625 	{
4626 	    if (!xim_has_focus)
4627 	    {
4628 		if (attr->spot_location.y >= 0)
4629 		{
4630 		    attr->spot_location.x = 0;
4631 		    attr->spot_location.y = -100;
4632 		    attrmask |= (int)GDK_IC_SPOT_LOCATION;
4633 		}
4634 	    }
4635 	    else
4636 	    {
4637 		gint	width, height;
4638 
4639 		if (attr->spot_location.x != TEXT_X(gui.col)
4640 		    || attr->spot_location.y != TEXT_Y(gui.row))
4641 		{
4642 		    attr->spot_location.x = TEXT_X(gui.col);
4643 		    attr->spot_location.y = TEXT_Y(gui.row);
4644 		    attrmask |= (int)GDK_IC_SPOT_LOCATION;
4645 		}
4646 
4647 		gdk_window_get_size(gui.drawarea->window, &width, &height);
4648 		width -= 2 * gui.border_offset;
4649 		height -= 2 * gui.border_offset;
4650 		if (xim_input_style & (int)GDK_IM_STATUS_AREA)
4651 		    height -= gui.char_height;
4652 		if (attr->preedit_area.width != width
4653 		    || attr->preedit_area.height != height)
4654 		{
4655 		    attr->preedit_area.x = gui.border_offset;
4656 		    attr->preedit_area.y = gui.border_offset;
4657 		    attr->preedit_area.width = width;
4658 		    attr->preedit_area.height = height;
4659 		    attrmask |= (int)GDK_IC_PREEDIT_AREA;
4660 		}
4661 
4662 		if (attr->preedit_fontset != gui.current_font)
4663 		{
4664 		    attr->preedit_fontset = gui.current_font;
4665 		    attrmask |= (int)GDK_IC_PREEDIT_FONTSET;
4666 		}
4667 	    }
4668 	}
4669 # endif /* FEAT_XFONTSET */
4670 
4671 	if (xim_fg_color == INVALCOLOR)
4672 	{
4673 	    xim_fg_color = gui.def_norm_pixel;
4674 	    xim_bg_color = gui.def_back_pixel;
4675 	}
4676 	if (attr->preedit_foreground.pixel != xim_fg_color)
4677 	{
4678 	    attr->preedit_foreground.pixel = xim_fg_color;
4679 	    attrmask |= (int)GDK_IC_PREEDIT_FOREGROUND;
4680 	}
4681 	if (attr->preedit_background.pixel != xim_bg_color)
4682 	{
4683 	    attr->preedit_background.pixel = xim_bg_color;
4684 	    attrmask |= (int)GDK_IC_PREEDIT_BACKGROUND;
4685 	}
4686 
4687 	if (attrmask != 0)
4688 	    gdk_ic_set_attr(xic, attr, (GdkICAttributesType)attrmask);
4689     }
4690 #else /* FEAT_GUI_GTK */
4691     {
4692 	XVaNestedList attr_list;
4693 	XRectangle spot_area;
4694 	XPoint over_spot;
4695 	int line_space;
4696 
4697 	if (!xim_has_focus)
4698 	{
4699 	    /* hide XIM cursor */
4700 	    over_spot.x = 0;
4701 	    over_spot.y = -100; /* arbitrary invisible position */
4702 	    attr_list = (XVaNestedList) XVaCreateNestedList(0,
4703 							    XNSpotLocation,
4704 							    &over_spot,
4705 							    NULL);
4706 	    XSetICValues(xic, XNPreeditAttributes, attr_list, NULL);
4707 	    XFree(attr_list);
4708 	    return;
4709 	}
4710 
4711 	if (input_style & XIMPreeditPosition)
4712 	{
4713 	    if (xim_fg_color == INVALCOLOR)
4714 	    {
4715 		xim_fg_color = gui.def_norm_pixel;
4716 		xim_bg_color = gui.def_back_pixel;
4717 	    }
4718 	    over_spot.x = TEXT_X(gui.col);
4719 	    over_spot.y = TEXT_Y(gui.row);
4720 	    spot_area.x = 0;
4721 	    spot_area.y = 0;
4722 	    spot_area.height = gui.char_height * Rows;
4723 	    spot_area.width  = gui.char_width * Columns;
4724 	    line_space = gui.char_height;
4725 	    attr_list = (XVaNestedList) XVaCreateNestedList(0,
4726 					    XNSpotLocation, &over_spot,
4727 					    XNForeground, (Pixel) xim_fg_color,
4728 					    XNBackground, (Pixel) xim_bg_color,
4729 					    XNArea, &spot_area,
4730 					    XNLineSpace, line_space,
4731 					    NULL);
4732 	    if (XSetICValues(xic, XNPreeditAttributes, attr_list, NULL))
4733 		EMSG(_("E284: Cannot set IC values"));
4734 	    XFree(attr_list);
4735 	}
4736     }
4737 #endif /* FEAT_GUI_GTK */
4738 }
4739 
4740 /*
4741  * Set up the status area.
4742  *
4743  * This should use a separate Widget, but that seems not possible, because
4744  * preedit_area and status_area should be set to the same window as for the
4745  * text input.  Unfortunately this means the status area pollutes the text
4746  * window...
4747  */
4748     void
4749 xim_set_status_area()
4750 {
4751     if (xic == NULL)
4752 	return;
4753 
4754 #ifdef FEAT_GUI_GTK
4755 # if defined(FEAT_XFONTSET)
4756     if (use_status_area)
4757     {
4758 	GdkICAttr	*attr;
4759 	int		style;
4760 	gint		width, height;
4761 	GtkWidget	*widget;
4762 	int		attrmask;
4763 
4764 	if (!xic_attr)
4765 	    return;
4766 
4767 	attr = xic_attr;
4768 	attrmask = 0;
4769 	style = (int)gdk_ic_get_style(xic);
4770 	if ((style & (int)GDK_IM_STATUS_MASK) == (int)GDK_IM_STATUS_AREA)
4771 	{
4772 	    if (gui.fontset != NOFONTSET
4773 		    && gui.fontset->type == GDK_FONT_FONTSET)
4774 	    {
4775 		widget = gui.mainwin;
4776 		gdk_window_get_size(widget->window, &width, &height);
4777 
4778 		attrmask |= (int)GDK_IC_STATUS_AREA;
4779 		attr->status_area.x = 0;
4780 		attr->status_area.y = height - gui.char_height - 1;
4781 		attr->status_area.width = width;
4782 		attr->status_area.height = gui.char_height;
4783 	    }
4784 	}
4785 	if (attrmask != 0)
4786 	    gdk_ic_set_attr(xic, attr, (GdkICAttributesType)attrmask);
4787     }
4788 # endif
4789 #else
4790     {
4791 	XVaNestedList preedit_list = 0, status_list = 0, list = 0;
4792 	XRectangle pre_area, status_area;
4793 
4794 	if (input_style & XIMStatusArea)
4795 	{
4796 	    if (input_style & XIMPreeditArea)
4797 	    {
4798 		XRectangle *needed_rect;
4799 
4800 		/* to get status_area width */
4801 		status_list = XVaCreateNestedList(0, XNAreaNeeded,
4802 						  &needed_rect, NULL);
4803 		XGetICValues(xic, XNStatusAttributes, status_list, NULL);
4804 		XFree(status_list);
4805 
4806 		status_area.width = needed_rect->width;
4807 	    }
4808 	    else
4809 		status_area.width = gui.char_width * Columns;
4810 
4811 	    status_area.x = 0;
4812 	    status_area.y = gui.char_height * Rows + gui.border_offset;
4813 	    if (gui.which_scrollbars[SBAR_BOTTOM])
4814 		status_area.y += gui.scrollbar_height;
4815 #ifdef FEAT_MENU
4816 	    if (gui.menu_is_active)
4817 		status_area.y += gui.menu_height;
4818 #endif
4819 	    status_area.height = gui.char_height;
4820 	    status_list = XVaCreateNestedList(0, XNArea, &status_area, NULL);
4821 	}
4822 	else
4823 	{
4824 	    status_area.x = 0;
4825 	    status_area.y = gui.char_height * Rows + gui.border_offset;
4826 	    if (gui.which_scrollbars[SBAR_BOTTOM])
4827 		status_area.y += gui.scrollbar_height;
4828 #ifdef FEAT_MENU
4829 	    if (gui.menu_is_active)
4830 		status_area.y += gui.menu_height;
4831 #endif
4832 	    status_area.width = 0;
4833 	    status_area.height = gui.char_height;
4834 	}
4835 
4836 	if (input_style & XIMPreeditArea)   /* off-the-spot */
4837 	{
4838 	    pre_area.x = status_area.x + status_area.width;
4839 	    pre_area.y = gui.char_height * Rows + gui.border_offset;
4840 	    pre_area.width = gui.char_width * Columns - pre_area.x;
4841 	    if (gui.which_scrollbars[SBAR_BOTTOM])
4842 		pre_area.y += gui.scrollbar_height;
4843 #ifdef FEAT_MENU
4844 	    if (gui.menu_is_active)
4845 		pre_area.y += gui.menu_height;
4846 #endif
4847 	    pre_area.height = gui.char_height;
4848 	    preedit_list = XVaCreateNestedList(0, XNArea, &pre_area, NULL);
4849 	}
4850 	else if (input_style & XIMPreeditPosition)   /* over-the-spot */
4851 	{
4852 	    pre_area.x = 0;
4853 	    pre_area.y = 0;
4854 	    pre_area.height = gui.char_height * Rows;
4855 	    pre_area.width = gui.char_width * Columns;
4856 	    preedit_list = XVaCreateNestedList(0, XNArea, &pre_area, NULL);
4857 	}
4858 
4859 	if (preedit_list && status_list)
4860 	    list = XVaCreateNestedList(0, XNPreeditAttributes, preedit_list,
4861 				       XNStatusAttributes, status_list, NULL);
4862 	else if (preedit_list)
4863 	    list = XVaCreateNestedList(0, XNPreeditAttributes, preedit_list,
4864 				       NULL);
4865 	else if (status_list)
4866 	    list = XVaCreateNestedList(0, XNStatusAttributes, status_list,
4867 				       NULL);
4868 	else
4869 	    list = NULL;
4870 
4871 	if (list)
4872 	{
4873 	    XSetICValues(xic, XNVaNestedList, list, NULL);
4874 	    XFree(list);
4875 	}
4876 	if (status_list)
4877 	    XFree(status_list);
4878 	if (preedit_list)
4879 	    XFree(preedit_list);
4880     }
4881 #endif
4882 }
4883 
4884 #if defined(FEAT_GUI_X11) || defined(FEAT_GUI_GTK)
4885 static char e_xim[] = N_("E285: Failed to create input context");
4886 #endif
4887 
4888 #if defined(FEAT_GUI_X11) || defined(PROTO)
4889 # if defined(XtSpecificationRelease) && XtSpecificationRelease >= 6 && !defined(sun)
4890 #  define USE_X11R6_XIM
4891 # endif
4892 
4893 static int xim_real_init __ARGS((Window x11_window, Display *x11_display));
4894 
4895 
4896 #ifdef USE_X11R6_XIM
4897 static void xim_instantiate_cb __ARGS((Display *display, XPointer client_data, XPointer	call_data));
4898 static void xim_destroy_cb __ARGS((XIM im, XPointer client_data, XPointer call_data));
4899 
4900 /*ARGSUSED*/
4901     static void
4902 xim_instantiate_cb(display, client_data, call_data)
4903     Display	*display;
4904     XPointer	client_data;
4905     XPointer	call_data;
4906 {
4907     Window	x11_window;
4908     Display	*x11_display;
4909 
4910 #ifdef XIM_DEBUG
4911     xim_log("xim_instantiate_cb()\n");
4912 #endif
4913 
4914     gui_get_x11_windis(&x11_window, &x11_display);
4915     if (display != x11_display)
4916 	return;
4917 
4918     xim_real_init(x11_window, x11_display);
4919     gui_set_shellsize(FALSE, FALSE, RESIZE_BOTH);
4920     if (xic != NULL)
4921 	XUnregisterIMInstantiateCallback(x11_display, NULL, NULL, NULL,
4922 					 xim_instantiate_cb, NULL);
4923 }
4924 
4925 /*ARGSUSED*/
4926     static void
4927 xim_destroy_cb(im, client_data, call_data)
4928     XIM		im;
4929     XPointer	client_data;
4930     XPointer	call_data;
4931 {
4932     Window	x11_window;
4933     Display	*x11_display;
4934 
4935 #ifdef XIM_DEBUG
4936     xim_log("xim_destroy_cb()\n");
4937 #endif
4938     gui_get_x11_windis(&x11_window, &x11_display);
4939 
4940     xic = NULL;
4941     status_area_enabled = FALSE;
4942 
4943     gui_set_shellsize(FALSE, FALSE, RESIZE_BOTH);
4944 
4945     XRegisterIMInstantiateCallback(x11_display, NULL, NULL, NULL,
4946 				   xim_instantiate_cb, NULL);
4947 }
4948 #endif
4949 
4950     void
4951 xim_init()
4952 {
4953     Window	x11_window;
4954     Display	*x11_display;
4955 
4956 #ifdef XIM_DEBUG
4957     xim_log("xim_init()\n");
4958 #endif
4959 
4960     gui_get_x11_windis(&x11_window, &x11_display);
4961 
4962     xic = NULL;
4963 
4964     if (xim_real_init(x11_window, x11_display))
4965 	return;
4966 
4967     gui_set_shellsize(FALSE, FALSE, RESIZE_BOTH);
4968 
4969 #ifdef USE_X11R6_XIM
4970     XRegisterIMInstantiateCallback(x11_display, NULL, NULL, NULL,
4971 				   xim_instantiate_cb, NULL);
4972 #endif
4973 }
4974 
4975     static int
4976 xim_real_init(x11_window, x11_display)
4977     Window	x11_window;
4978     Display	*x11_display;
4979 {
4980     int		i;
4981     char	*p,
4982 		*s,
4983 		*ns,
4984 		*end,
4985 		tmp[1024];
4986 #define IMLEN_MAX 40
4987     char	buf[IMLEN_MAX + 7];
4988     XIM		xim = NULL;
4989     XIMStyles	*xim_styles;
4990     XIMStyle	this_input_style = 0;
4991     Boolean	found;
4992     XPoint	over_spot;
4993     XVaNestedList preedit_list, status_list;
4994 
4995     input_style = 0;
4996     status_area_enabled = FALSE;
4997 
4998     if (xic != NULL)
4999 	return FALSE;
5000 
5001     if (gui.rsrc_input_method != NULL && *gui.rsrc_input_method != NUL)
5002     {
5003 	strcpy(tmp, gui.rsrc_input_method);
5004 	for (ns = s = tmp; ns != NULL && *s != NUL;)
5005 	{
5006 	    s = (char *)skipwhite((char_u *)s);
5007 	    if (*s == NUL)
5008 		break;
5009 	    if ((ns = end = strchr(s, ',')) == NULL)
5010 		end = s + strlen(s);
5011 	    while (isspace(((char_u *)end)[-1]))
5012 		end--;
5013 	    *end = NUL;
5014 
5015 	    if (strlen(s) <= IMLEN_MAX)
5016 	    {
5017 		strcpy(buf, "@im=");
5018 		strcat(buf, s);
5019 		if ((p = XSetLocaleModifiers(buf)) != NULL && *p != NUL
5020 			&& (xim = XOpenIM(x11_display, NULL, NULL, NULL))
5021 								      != NULL)
5022 		    break;
5023 	    }
5024 
5025 	    s = ns + 1;
5026 	}
5027     }
5028 
5029     if (xim == NULL && (p = XSetLocaleModifiers("")) != NULL && *p != NUL)
5030 	xim = XOpenIM(x11_display, NULL, NULL, NULL);
5031 
5032     /* This is supposed to be useful to obtain characters through
5033      * XmbLookupString() without really using a XIM. */
5034     if (xim == NULL && (p = XSetLocaleModifiers("@im=none")) != NULL
5035 								 && *p != NUL)
5036 	xim = XOpenIM(x11_display, NULL, NULL, NULL);
5037 
5038     if (xim == NULL)
5039     {
5040 	/* Only give this message when verbose is set, because too many people
5041 	 * got this message when they didn't want to use a XIM. */
5042 	if (p_verbose > 0)
5043 	{
5044 	    verbose_enter();
5045 	    EMSG(_("E286: Failed to open input method"));
5046 	    verbose_leave();
5047 	}
5048 	return FALSE;
5049     }
5050 
5051 #ifdef USE_X11R6_XIM
5052     {
5053 	XIMCallback destroy_cb;
5054 
5055 	destroy_cb.callback = xim_destroy_cb;
5056 	destroy_cb.client_data = NULL;
5057 	if (XSetIMValues(xim, XNDestroyCallback, &destroy_cb, NULL))
5058 	    EMSG(_("E287: Warning: Could not set destroy callback to IM"));
5059     }
5060 #endif
5061 
5062     if (XGetIMValues(xim, XNQueryInputStyle, &xim_styles, NULL) || !xim_styles)
5063     {
5064 	EMSG(_("E288: input method doesn't support any style"));
5065 	XCloseIM(xim);
5066 	return FALSE;
5067     }
5068 
5069     found = False;
5070     strcpy(tmp, gui.rsrc_preedit_type_name);
5071     for (s = tmp; s && !found; )
5072     {
5073 	while (*s && isspace((unsigned char)*s))
5074 	    s++;
5075 	if (!*s)
5076 	    break;
5077 	if ((ns = end = strchr(s, ',')) != 0)
5078 	    ns++;
5079 	else
5080 	    end = s + strlen(s);
5081 	while (isspace((unsigned char)*end))
5082 	    end--;
5083 	*end = '\0';
5084 
5085 	if (!strcmp(s, "OverTheSpot"))
5086 	    this_input_style = (XIMPreeditPosition | XIMStatusArea);
5087 	else if (!strcmp(s, "OffTheSpot"))
5088 	    this_input_style = (XIMPreeditArea | XIMStatusArea);
5089 	else if (!strcmp(s, "Root"))
5090 	    this_input_style = (XIMPreeditNothing | XIMStatusNothing);
5091 
5092 	for (i = 0; (unsigned short)i < xim_styles->count_styles; i++)
5093 	{
5094 	    if (this_input_style == xim_styles->supported_styles[i])
5095 	    {
5096 		found = True;
5097 		break;
5098 	    }
5099 	}
5100 	if (!found)
5101 	    for (i = 0; (unsigned short)i < xim_styles->count_styles; i++)
5102 	    {
5103 		if ((xim_styles->supported_styles[i] & this_input_style)
5104 			== (this_input_style & ~XIMStatusArea))
5105 		{
5106 		    this_input_style &= ~XIMStatusArea;
5107 		    found = True;
5108 		    break;
5109 		}
5110 	    }
5111 
5112 	s = ns;
5113     }
5114     XFree(xim_styles);
5115 
5116     if (!found)
5117     {
5118 	/* Only give this message when verbose is set, because too many people
5119 	 * got this message when they didn't want to use a XIM. */
5120 	if (p_verbose > 0)
5121 	{
5122 	    verbose_enter();
5123 	    EMSG(_("E289: input method doesn't support my preedit type"));
5124 	    verbose_leave();
5125 	}
5126 	XCloseIM(xim);
5127 	return FALSE;
5128     }
5129 
5130     over_spot.x = TEXT_X(gui.col);
5131     over_spot.y = TEXT_Y(gui.row);
5132     input_style = this_input_style;
5133 
5134     /* A crash was reported when trying to pass gui.norm_font as XNFontSet,
5135      * thus that has been removed.  Hopefully the default works... */
5136 #ifdef FEAT_XFONTSET
5137     if (gui.fontset != NOFONTSET)
5138     {
5139 	preedit_list = XVaCreateNestedList(0,
5140 				XNSpotLocation, &over_spot,
5141 				XNForeground, (Pixel)gui.def_norm_pixel,
5142 				XNBackground, (Pixel)gui.def_back_pixel,
5143 				XNFontSet, (XFontSet)gui.fontset,
5144 				NULL);
5145 	status_list = XVaCreateNestedList(0,
5146 				XNForeground, (Pixel)gui.def_norm_pixel,
5147 				XNBackground, (Pixel)gui.def_back_pixel,
5148 				XNFontSet, (XFontSet)gui.fontset,
5149 				NULL);
5150     }
5151     else
5152 #endif
5153     {
5154 	preedit_list = XVaCreateNestedList(0,
5155 				XNSpotLocation, &over_spot,
5156 				XNForeground, (Pixel)gui.def_norm_pixel,
5157 				XNBackground, (Pixel)gui.def_back_pixel,
5158 				NULL);
5159 	status_list = XVaCreateNestedList(0,
5160 				XNForeground, (Pixel)gui.def_norm_pixel,
5161 				XNBackground, (Pixel)gui.def_back_pixel,
5162 				NULL);
5163     }
5164 
5165     xic = XCreateIC(xim,
5166 		    XNInputStyle, input_style,
5167 		    XNClientWindow, x11_window,
5168 		    XNFocusWindow, gui.wid,
5169 		    XNPreeditAttributes, preedit_list,
5170 		    XNStatusAttributes, status_list,
5171 		    NULL);
5172     XFree(status_list);
5173     XFree(preedit_list);
5174     if (xic != NULL)
5175     {
5176 	if (input_style & XIMStatusArea)
5177 	{
5178 	    xim_set_status_area();
5179 	    status_area_enabled = TRUE;
5180 	}
5181 	else
5182 	    gui_set_shellsize(FALSE, FALSE, RESIZE_BOTH);
5183     }
5184     else
5185     {
5186 	EMSG(_(e_xim));
5187 	XCloseIM(xim);
5188 	return FALSE;
5189     }
5190 
5191     return TRUE;
5192 }
5193 
5194 #endif /* FEAT_GUI_X11 */
5195 
5196 #if defined(FEAT_GUI_GTK) || defined(PROTO)
5197 
5198 # ifdef FEAT_XFONTSET
5199 static char e_overthespot[] = N_("E290: over-the-spot style requires fontset");
5200 # endif
5201 
5202 # ifdef PROTO
5203 typedef int GdkIC;
5204 # endif
5205 
5206     void
5207 xim_decide_input_style()
5208 {
5209     /* GDK_IM_STATUS_CALLBACKS was disabled, enabled it to allow Japanese
5210      * OverTheSpot. */
5211     int supported_style = (int)GDK_IM_PREEDIT_NONE |
5212 				 (int)GDK_IM_PREEDIT_NOTHING |
5213 				 (int)GDK_IM_PREEDIT_POSITION |
5214 				 (int)GDK_IM_PREEDIT_CALLBACKS |
5215 				 (int)GDK_IM_STATUS_CALLBACKS |
5216 				 (int)GDK_IM_STATUS_AREA |
5217 				 (int)GDK_IM_STATUS_NONE |
5218 				 (int)GDK_IM_STATUS_NOTHING;
5219 
5220 #ifdef XIM_DEBUG
5221     xim_log("xim_decide_input_style()\n");
5222 #endif
5223 
5224     if (!gdk_im_ready())
5225 	xim_input_style = 0;
5226     else
5227     {
5228 	if (gtk_major_version > 1
5229 		|| (gtk_major_version == 1
5230 		    && (gtk_minor_version > 2
5231 			|| (gtk_minor_version == 2 && gtk_micro_version >= 3))))
5232 	    use_status_area = TRUE;
5233 	else
5234 	{
5235 	    EMSG(_("E291: Your GTK+ is older than 1.2.3. Status area disabled"));
5236 	    use_status_area = FALSE;
5237 	}
5238 #ifdef FEAT_XFONTSET
5239 	if (gui.fontset == NOFONTSET || gui.fontset->type != GDK_FONT_FONTSET)
5240 #endif
5241 	    supported_style &= ~((int)GDK_IM_PREEDIT_POSITION
5242 						   | (int)GDK_IM_STATUS_AREA);
5243 	if (!use_status_area)
5244 	    supported_style &= ~(int)GDK_IM_STATUS_AREA;
5245 	xim_input_style = (int)gdk_im_decide_style((GdkIMStyle)supported_style);
5246     }
5247 }
5248 
5249 /*ARGSUSED*/
5250     static void
5251 preedit_start_cbproc(XIC xic, XPointer client_data, XPointer call_data)
5252 {
5253 #ifdef XIM_DEBUG
5254     xim_log("xim_decide_input_style()\n");
5255 #endif
5256 
5257     draw_feedback = NULL;
5258     xim_can_preediting = TRUE;
5259     xim_has_preediting = TRUE;
5260     gui_update_cursor(TRUE, FALSE);
5261     if (showmode() > 0)
5262     {
5263 	setcursor();
5264 	out_flush();
5265     }
5266 }
5267 
5268     static void
5269 xim_back_delete(int n)
5270 {
5271     char_u str[3];
5272 
5273     str[0] = CSI;
5274     str[1] = 'k';
5275     str[2] = 'b';
5276     while (n-- > 0)
5277 	add_to_input_buf(str, 3);
5278 }
5279 
5280 static GSList *key_press_event_queue = NULL;
5281 static gboolean processing_queued_event = FALSE;
5282 
5283 /*ARGSUSED*/
5284     static void
5285 preedit_draw_cbproc(XIC xic, XPointer client_data, XPointer call_data)
5286 {
5287     XIMPreeditDrawCallbackStruct *draw_data;
5288     XIMText	*text;
5289     char	*src;
5290     GSList	*event_queue;
5291 
5292 #ifdef XIM_DEBUG
5293     xim_log("preedit_draw_cbproc()\n");
5294 #endif
5295 
5296     draw_data = (XIMPreeditDrawCallbackStruct *) call_data;
5297     text = (XIMText *) draw_data->text;
5298 
5299     if ((text == NULL && draw_data->chg_length == preedit_buf_len)
5300 						      || preedit_buf_len == 0)
5301     {
5302 	init_preedit_start_col();
5303 	vim_free(draw_feedback);
5304 	draw_feedback = NULL;
5305     }
5306     if (draw_data->chg_length > 0)
5307     {
5308 	int bs_cnt;
5309 
5310 	if (draw_data->chg_length > preedit_buf_len)
5311 	    bs_cnt = preedit_buf_len;
5312 	else
5313 	    bs_cnt = draw_data->chg_length;
5314 	xim_back_delete(bs_cnt);
5315 	preedit_buf_len -= bs_cnt;
5316     }
5317     if (text != NULL)
5318     {
5319 	int		len;
5320 #ifdef FEAT_MBYTE
5321 	char_u		*buf = NULL;
5322 	unsigned int	nfeedback = 0;
5323 #endif
5324 	char_u		*ptr;
5325 
5326 	src = text->string.multi_byte;
5327 	if (src != NULL && !text->encoding_is_wchar)
5328 	{
5329 	    len = strlen(src);
5330 	    ptr = (char_u *)src;
5331 	    /* Avoid the enter for decision */
5332 	    if (*ptr == '\n')
5333 		return;
5334 
5335 #ifdef FEAT_MBYTE
5336 	    if (input_conv.vc_type != CONV_NONE
5337 		    && (buf = string_convert(&input_conv,
5338 						 (char_u *)src, &len)) != NULL)
5339 	    {
5340 		/* Converted from 'termencoding' to 'encoding'. */
5341 		add_to_input_buf_csi(buf, len);
5342 		ptr = buf;
5343 	    }
5344 	    else
5345 #endif
5346 		add_to_input_buf_csi((char_u *)src, len);
5347 	    /* Add count of character to preedit_buf_len  */
5348 	    while (*ptr != NUL)
5349 	    {
5350 #ifdef FEAT_MBYTE
5351 		if (draw_data->text->feedback != NULL)
5352 		{
5353 		    if (draw_feedback == NULL)
5354 			draw_feedback = (char *)alloc(draw_data->chg_first
5355 							      + text->length);
5356 		    else
5357 			draw_feedback = realloc(draw_feedback,
5358 					 draw_data->chg_first + text->length);
5359 		    if (draw_feedback != NULL)
5360 		    {
5361 			draw_feedback[nfeedback + draw_data->chg_first]
5362 				       = draw_data->text->feedback[nfeedback];
5363 			nfeedback++;
5364 		    }
5365 		}
5366 		if (has_mbyte)
5367 		    ptr += (*mb_ptr2len)(ptr);
5368 		else
5369 #endif
5370 		    ptr++;
5371 		preedit_buf_len++;
5372 	    }
5373 #ifdef FEAT_MBYTE
5374 	    vim_free(buf);
5375 #endif
5376 	    preedit_end_col = MAXCOL;
5377 	}
5378     }
5379     if (text != NULL || draw_data->chg_length > 0)
5380     {
5381 	event_queue = key_press_event_queue;
5382 	processing_queued_event = TRUE;
5383 	while (event_queue != NULL && processing_queued_event)
5384 	{
5385 	    GdkEvent *ev = event_queue->data;
5386 
5387 	    gboolean *ret;
5388 	    gtk_signal_emit_by_name((GtkObject*)gui.mainwin, "key_press_event",
5389 								    ev, &ret);
5390 	    gdk_event_free(ev);
5391 	    event_queue = event_queue->next;
5392 	}
5393 	processing_queued_event = FALSE;
5394 	if (key_press_event_queue)
5395 	{
5396 	    g_slist_free(key_press_event_queue);
5397 	    key_press_event_queue = NULL;
5398 	}
5399     }
5400     if (gtk_main_level() > 0)
5401 	gtk_main_quit();
5402 }
5403 
5404 /*
5405  * Retrieve the highlighting attributes at column col in the preedit string.
5406  * Return -1 if not in preediting mode or if col is out of range.
5407  */
5408     int
5409 im_get_feedback_attr(int col)
5410 {
5411     if (draw_feedback != NULL && col < preedit_buf_len)
5412     {
5413 	if (draw_feedback[col] & XIMReverse)
5414 	    return HL_INVERSE;
5415 	else if (draw_feedback[col] & XIMUnderline)
5416 	    return HL_UNDERLINE;
5417 	else
5418 	    return hl_attr(HLF_V);
5419     }
5420 
5421     return -1;
5422 }
5423 
5424 /*ARGSUSED*/
5425     static void
5426 preedit_caret_cbproc(XIC xic, XPointer client_data, XPointer call_data)
5427 {
5428 #ifdef XIM_DEBUG
5429     xim_log("preedit_caret_cbproc()\n");
5430 #endif
5431 }
5432 
5433 /*ARGSUSED*/
5434     static void
5435 preedit_done_cbproc(XIC xic, XPointer client_data, XPointer call_data)
5436 {
5437 #ifdef XIM_DEBUG
5438     xim_log("preedit_done_cbproc()\n");
5439 #endif
5440 
5441     vim_free(draw_feedback);
5442     draw_feedback = NULL;
5443     xim_can_preediting = FALSE;
5444     xim_has_preediting = FALSE;
5445     gui_update_cursor(TRUE, FALSE);
5446     if (showmode() > 0)
5447     {
5448 	setcursor();
5449 	out_flush();
5450     }
5451 }
5452 
5453     void
5454 xim_reset(void)
5455 {
5456     char *text;
5457 
5458 #ifdef XIM_DEBUG
5459     xim_log("xim_reset()\n");
5460 #endif
5461 
5462     if (xic != NULL)
5463     {
5464 	text = XmbResetIC(((GdkICPrivate *)xic)->xic);
5465 	if (text != NULL && !(xim_input_style & (int)GDK_IM_PREEDIT_CALLBACKS))
5466 	    add_to_input_buf_csi((char_u *)text, strlen(text));
5467 	else
5468 	    preedit_buf_len = 0;
5469 	if (text != NULL)
5470 	    XFree(text);
5471     }
5472 }
5473 
5474 /*ARGSUSED*/
5475     int
5476 xim_queue_key_press_event(GdkEventKey *event, int down)
5477 {
5478 #ifdef XIM_DEBUG
5479     xim_log("xim_queue_key_press_event()\n");
5480 #endif
5481 
5482     if (preedit_buf_len <= 0)
5483 	return FALSE;
5484     if (processing_queued_event)
5485 	processing_queued_event = FALSE;
5486 
5487     key_press_event_queue = g_slist_append(key_press_event_queue,
5488 					   gdk_event_copy((GdkEvent *)event));
5489     return TRUE;
5490 }
5491 
5492 /*ARGSUSED*/
5493     static void
5494 preedit_callback_setup(GdkIC *ic)
5495 {
5496     XIC xxic;
5497     XVaNestedList preedit_attr;
5498     XIMCallback preedit_start_cb;
5499     XIMCallback preedit_draw_cb;
5500     XIMCallback preedit_caret_cb;
5501     XIMCallback preedit_done_cb;
5502 
5503     xxic = ((GdkICPrivate*)xic)->xic;
5504     preedit_start_cb.callback = (XIMProc)preedit_start_cbproc;
5505     preedit_draw_cb.callback = (XIMProc)preedit_draw_cbproc;
5506     preedit_caret_cb.callback = (XIMProc)preedit_caret_cbproc;
5507     preedit_done_cb.callback = (XIMProc)preedit_done_cbproc;
5508     preedit_attr
5509 	= XVaCreateNestedList (0,
5510 			       XNPreeditStartCallback, &preedit_start_cb,
5511 			       XNPreeditDrawCallback, &preedit_draw_cb,
5512 			       XNPreeditCaretCallback, &preedit_caret_cb,
5513 			       XNPreeditDoneCallback, &preedit_done_cb,
5514 			       0);
5515     XSetICValues (xxic, XNPreeditAttributes, preedit_attr, 0);
5516     XFree(preedit_attr);
5517 }
5518 
5519 /*ARGSUSED*/
5520     static void
5521 reset_state_setup(GdkIC *ic)
5522 {
5523 #ifdef USE_X11R6_XIM
5524     /* don't change the input context when we call reset */
5525     XSetICValues(((GdkICPrivate*)ic)->xic, XNResetState, XIMPreserveState, 0);
5526 #endif
5527 }
5528 
5529     void
5530 xim_init(void)
5531 {
5532 #ifdef XIM_DEBUG
5533     xim_log("xim_init()\n");
5534 #endif
5535 
5536     xic = NULL;
5537     xic_attr = NULL;
5538 
5539     if (!gdk_im_ready())
5540     {
5541 	if (p_verbose > 0)
5542 	{
5543 	    verbose_enter();
5544 	    EMSG(_("E292: Input Method Server is not running"));
5545 	    verbose_leave();
5546 	}
5547 	return;
5548     }
5549     if ((xic_attr = gdk_ic_attr_new()) != NULL)
5550     {
5551 #ifdef FEAT_XFONTSET
5552 	gint width, height;
5553 #endif
5554 	int		mask;
5555 	GdkColormap	*colormap;
5556 	GdkICAttr	*attr = xic_attr;
5557 	int		attrmask = (int)GDK_IC_ALL_REQ;
5558 	GtkWidget	*widget = gui.drawarea;
5559 
5560 	attr->style = (GdkIMStyle)xim_input_style;
5561 	attr->client_window = gui.mainwin->window;
5562 
5563 	if ((colormap = gtk_widget_get_colormap(widget)) !=
5564 	    gtk_widget_get_default_colormap())
5565 	{
5566 	    attrmask |= (int)GDK_IC_PREEDIT_COLORMAP;
5567 	    attr->preedit_colormap = colormap;
5568 	}
5569 	attrmask |= (int)GDK_IC_PREEDIT_FOREGROUND;
5570 	attrmask |= (int)GDK_IC_PREEDIT_BACKGROUND;
5571 	attr->preedit_foreground = widget->style->fg[GTK_STATE_NORMAL];
5572 	attr->preedit_background = widget->style->base[GTK_STATE_NORMAL];
5573 
5574 #ifdef FEAT_XFONTSET
5575 	if ((xim_input_style & (int)GDK_IM_PREEDIT_MASK)
5576 					      == (int)GDK_IM_PREEDIT_POSITION)
5577 	{
5578 	    if (gui.fontset == NOFONTSET
5579 		    || gui.fontset->type != GDK_FONT_FONTSET)
5580 	    {
5581 		EMSG(_(e_overthespot));
5582 	    }
5583 	    else
5584 	    {
5585 		gdk_window_get_size(widget->window, &width, &height);
5586 
5587 		attrmask |= (int)GDK_IC_PREEDIT_POSITION_REQ;
5588 		attr->spot_location.x = TEXT_X(0);
5589 		attr->spot_location.y = TEXT_Y(0);
5590 		attr->preedit_area.x = gui.border_offset;
5591 		attr->preedit_area.y = gui.border_offset;
5592 		attr->preedit_area.width = width - 2*gui.border_offset;
5593 		attr->preedit_area.height = height - 2*gui.border_offset;
5594 		attr->preedit_fontset = gui.fontset;
5595 	    }
5596 	}
5597 
5598 	if ((xim_input_style & (int)GDK_IM_STATUS_MASK)
5599 						   == (int)GDK_IM_STATUS_AREA)
5600 	{
5601 	    if (gui.fontset == NOFONTSET
5602 		    || gui.fontset->type != GDK_FONT_FONTSET)
5603 	    {
5604 		EMSG(_(e_overthespot));
5605 	    }
5606 	    else
5607 	    {
5608 		gdk_window_get_size(gui.mainwin->window, &width, &height);
5609 		attrmask |= (int)GDK_IC_STATUS_AREA_REQ;
5610 		attr->status_area.x = 0;
5611 		attr->status_area.y = height - gui.char_height - 1;
5612 		attr->status_area.width = width;
5613 		attr->status_area.height = gui.char_height;
5614 		attr->status_fontset = gui.fontset;
5615 	    }
5616 	}
5617 	else if ((xim_input_style & (int)GDK_IM_STATUS_MASK)
5618 					      == (int)GDK_IM_STATUS_CALLBACKS)
5619 	{
5620 	    /* FIXME */
5621 	}
5622 #endif
5623 
5624 	xic = gdk_ic_new(attr, (GdkICAttributesType)attrmask);
5625 
5626 	if (xic == NULL)
5627 	    EMSG(_(e_xim));
5628 	else
5629 	{
5630 	    mask = (int)gdk_window_get_events(widget->window);
5631 	    mask |= (int)gdk_ic_get_events(xic);
5632 	    gdk_window_set_events(widget->window, (GdkEventMask)mask);
5633 	    if (xim_input_style & (int)GDK_IM_PREEDIT_CALLBACKS)
5634 		preedit_callback_setup(xic);
5635 	    reset_state_setup(xic);
5636 	}
5637     }
5638 }
5639 
5640     void
5641 im_shutdown(void)
5642 {
5643 #ifdef XIM_DEBUG
5644     xim_log("im_shutdown()\n");
5645 #endif
5646 
5647     if (xic != NULL)
5648     {
5649 	gdk_im_end();
5650 	gdk_ic_destroy(xic);
5651 	xic = NULL;
5652     }
5653     xim_is_active = FALSE;
5654     xim_can_preediting = FALSE;
5655     preedit_start_col = MAXCOL;
5656     xim_has_preediting = FALSE;
5657 }
5658 
5659 #endif /* FEAT_GUI_GTK */
5660 
5661     int
5662 xim_get_status_area_height()
5663 {
5664 #ifdef FEAT_GUI_GTK
5665     if (xim_input_style & (int)GDK_IM_STATUS_AREA)
5666 	return gui.char_height;
5667 #else
5668     if (status_area_enabled)
5669 	return gui.char_height;
5670 #endif
5671     return 0;
5672 }
5673 
5674 /*
5675  * Get IM status.  When IM is on, return TRUE.  Else return FALSE.
5676  * FIXME: This doesn't work correctly: Having focus doesn't always mean XIM is
5677  * active, when not having focus XIM may still be active (e.g., when using a
5678  * tear-off menu item).
5679  */
5680     int
5681 im_get_status()
5682 {
5683 #  ifdef FEAT_GUI_GTK
5684     if (xim_input_style & (int)GDK_IM_PREEDIT_CALLBACKS)
5685 	return xim_can_preediting;
5686 #  endif
5687     return xim_has_focus;
5688 }
5689 
5690 # endif /* !HAVE_GTK2 */
5691 
5692 # if defined(FEAT_GUI_GTK) || defined(PROTO)
5693     int
5694 im_is_preediting()
5695 {
5696     return xim_has_preediting;
5697 }
5698 # endif
5699 #endif /* FEAT_XIM */
5700 
5701 #if defined(FEAT_MBYTE) || defined(PROTO)
5702 
5703 /*
5704  * Setup "vcp" for conversion from "from" to "to".
5705  * The names must have been made canonical with enc_canonize().
5706  * vcp->vc_type must have been initialized to CONV_NONE.
5707  * Note: cannot be used for conversion from/to ucs-2 and ucs-4 (will use utf-8
5708  * instead).
5709  * Afterwards invoke with "from" and "to" equal to NULL to cleanup.
5710  * Return FAIL when conversion is not supported, OK otherwise.
5711  */
5712     int
5713 convert_setup(vcp, from, to)
5714     vimconv_T	*vcp;
5715     char_u	*from;
5716     char_u	*to;
5717 {
5718     int		from_prop;
5719     int		to_prop;
5720 
5721     /* Reset to no conversion. */
5722 # ifdef USE_ICONV
5723     if (vcp->vc_type == CONV_ICONV && vcp->vc_fd != (iconv_t)-1)
5724 	iconv_close(vcp->vc_fd);
5725 # endif
5726     vcp->vc_type = CONV_NONE;
5727     vcp->vc_factor = 1;
5728     vcp->vc_fail = FALSE;
5729 
5730     /* No conversion when one of the names is empty or they are equal. */
5731     if (from == NULL || *from == NUL || to == NULL || *to == NUL
5732 						     || STRCMP(from, to) == 0)
5733 	return OK;
5734 
5735     from_prop = enc_canon_props(from);
5736     to_prop = enc_canon_props(to);
5737     if ((from_prop & ENC_LATIN1) && (to_prop & ENC_UNICODE))
5738     {
5739 	/* Internal latin1 -> utf-8 conversion. */
5740 	vcp->vc_type = CONV_TO_UTF8;
5741 	vcp->vc_factor = 2;	/* up to twice as long */
5742     }
5743     else if ((from_prop & ENC_LATIN9) && (to_prop & ENC_UNICODE))
5744     {
5745 	/* Internal latin9 -> utf-8 conversion. */
5746 	vcp->vc_type = CONV_9_TO_UTF8;
5747 	vcp->vc_factor = 3;	/* up to three as long (euro sign) */
5748     }
5749     else if ((from_prop & ENC_UNICODE) && (to_prop & ENC_LATIN1))
5750     {
5751 	/* Internal utf-8 -> latin1 conversion. */
5752 	vcp->vc_type = CONV_TO_LATIN1;
5753     }
5754     else if ((from_prop & ENC_UNICODE) && (to_prop & ENC_LATIN9))
5755     {
5756 	/* Internal utf-8 -> latin9 conversion. */
5757 	vcp->vc_type = CONV_TO_LATIN9;
5758     }
5759 #ifdef WIN3264
5760     /* Win32-specific codepage <-> codepage conversion without iconv. */
5761     else if (((from_prop & ENC_UNICODE) || encname2codepage(from) > 0)
5762 	    && ((to_prop & ENC_UNICODE) || encname2codepage(to) > 0))
5763     {
5764 	vcp->vc_type = CONV_CODEPAGE;
5765 	vcp->vc_factor = 2;	/* up to twice as long */
5766 	vcp->vc_cpfrom = (from_prop & ENC_UNICODE) ? 0 : encname2codepage(from);
5767 	vcp->vc_cpto = (to_prop & ENC_UNICODE) ? 0 : encname2codepage(to);
5768     }
5769 #endif
5770 #ifdef MACOS_X
5771     else if ((from_prop & ENC_MACROMAN) && (to_prop & ENC_LATIN1))
5772     {
5773 	vcp->vc_type = CONV_MAC_LATIN1;
5774     }
5775     else if ((from_prop & ENC_MACROMAN) && (to_prop & ENC_UNICODE))
5776     {
5777 	vcp->vc_type = CONV_MAC_UTF8;
5778 	vcp->vc_factor = 2;	/* up to twice as long */
5779     }
5780     else if ((from_prop & ENC_LATIN1) && (to_prop & ENC_MACROMAN))
5781     {
5782 	vcp->vc_type = CONV_LATIN1_MAC;
5783     }
5784     else if ((from_prop & ENC_UNICODE) && (to_prop & ENC_MACROMAN))
5785     {
5786 	vcp->vc_type = CONV_UTF8_MAC;
5787     }
5788 #endif
5789 # ifdef USE_ICONV
5790     else
5791     {
5792 	/* Use iconv() for conversion. */
5793 	vcp->vc_fd = (iconv_t)my_iconv_open(
5794 		(to_prop & ENC_UNICODE) ? (char_u *)"utf-8" : to,
5795 		(from_prop & ENC_UNICODE) ? (char_u *)"utf-8" : from);
5796 	if (vcp->vc_fd != (iconv_t)-1)
5797 	{
5798 	    vcp->vc_type = CONV_ICONV;
5799 	    vcp->vc_factor = 4;	/* could be longer too... */
5800 	}
5801     }
5802 # endif
5803     if (vcp->vc_type == CONV_NONE)
5804 	return FAIL;
5805 
5806     return OK;
5807 }
5808 
5809 #if defined(FEAT_GUI) || defined(AMIGA) || defined(WIN3264) \
5810 	|| defined(MSDOS) || defined(PROTO)
5811 /*
5812  * Do conversion on typed input characters in-place.
5813  * The input and output are not NUL terminated!
5814  * Returns the length after conversion.
5815  */
5816     int
5817 convert_input(ptr, len, maxlen)
5818     char_u	*ptr;
5819     int		len;
5820     int		maxlen;
5821 {
5822     return convert_input_safe(ptr, len, maxlen, NULL, NULL);
5823 }
5824 #endif
5825 
5826 /*
5827  * Like convert_input(), but when there is an incomplete byte sequence at the
5828  * end return that as an allocated string in "restp" and set "*restlenp" to
5829  * the length.  If "restp" is NULL it is not used.
5830  */
5831     int
5832 convert_input_safe(ptr, len, maxlen, restp, restlenp)
5833     char_u	*ptr;
5834     int		len;
5835     int		maxlen;
5836     char_u	**restp;
5837     int		*restlenp;
5838 {
5839     char_u	*d;
5840     int		dlen = len;
5841     int		unconvertlen = 0;
5842 
5843     d = string_convert_ext(&input_conv, ptr, &dlen,
5844 					restp == NULL ? NULL : &unconvertlen);
5845     if (d != NULL)
5846     {
5847 	if (dlen <= maxlen)
5848 	{
5849 	    if (unconvertlen > 0)
5850 	    {
5851 		/* Move the unconverted characters to allocated memory. */
5852 		*restp = alloc(unconvertlen);
5853 		if (*restp != NULL)
5854 		    mch_memmove(*restp, ptr + len - unconvertlen, unconvertlen);
5855 		*restlenp = unconvertlen;
5856 	    }
5857 	    mch_memmove(ptr, d, dlen);
5858 	}
5859 	else
5860 	    /* result is too long, keep the unconverted text (the caller must
5861 	     * have done something wrong!) */
5862 	    dlen = len;
5863 	vim_free(d);
5864     }
5865     return dlen;
5866 }
5867 
5868 /*
5869  * Convert text "ptr[*lenp]" according to "vcp".
5870  * Returns the result in allocated memory and sets "*lenp".
5871  * When "lenp" is NULL, use NUL terminated strings.
5872  * Illegal chars are often changed to "?", unless vcp->vc_fail is set.
5873  * When something goes wrong, NULL is returned and "*lenp" is unchanged.
5874  */
5875     char_u *
5876 string_convert(vcp, ptr, lenp)
5877     vimconv_T	*vcp;
5878     char_u	*ptr;
5879     int		*lenp;
5880 {
5881     return string_convert_ext(vcp, ptr, lenp, NULL);
5882 }
5883 
5884 /*
5885  * Like string_convert(), but when "unconvlenp" is not NULL and there are is
5886  * an incomplete sequence at the end it is not converted and "*unconvlenp" is
5887  * set to the number of remaining bytes.
5888  */
5889     char_u *
5890 string_convert_ext(vcp, ptr, lenp, unconvlenp)
5891     vimconv_T	*vcp;
5892     char_u	*ptr;
5893     int		*lenp;
5894     int		*unconvlenp;
5895 {
5896     char_u	*retval = NULL;
5897     char_u	*d;
5898     int		len;
5899     int		i;
5900     int		l;
5901     int		c;
5902 
5903     if (lenp == NULL)
5904 	len = (int)STRLEN(ptr);
5905     else
5906 	len = *lenp;
5907     if (len == 0)
5908 	return vim_strsave((char_u *)"");
5909 
5910     switch (vcp->vc_type)
5911     {
5912 	case CONV_TO_UTF8:	/* latin1 to utf-8 conversion */
5913 	    retval = alloc(len * 2 + 1);
5914 	    if (retval == NULL)
5915 		break;
5916 	    d = retval;
5917 	    for (i = 0; i < len; ++i)
5918 	    {
5919 		c = ptr[i];
5920 		if (c < 0x80)
5921 		    *d++ = c;
5922 		else
5923 		{
5924 		    *d++ = 0xc0 + ((unsigned)c >> 6);
5925 		    *d++ = 0x80 + (c & 0x3f);
5926 		}
5927 	    }
5928 	    *d = NUL;
5929 	    if (lenp != NULL)
5930 		*lenp = (int)(d - retval);
5931 	    break;
5932 
5933 	case CONV_9_TO_UTF8:	/* latin9 to utf-8 conversion */
5934 	    retval = alloc(len * 3 + 1);
5935 	    if (retval == NULL)
5936 		break;
5937 	    d = retval;
5938 	    for (i = 0; i < len; ++i)
5939 	    {
5940 		c = ptr[i];
5941 		switch (c)
5942 		{
5943 		    case 0xa4: c = 0x20ac; break;   /* euro */
5944 		    case 0xa6: c = 0x0160; break;   /* S hat */
5945 		    case 0xa8: c = 0x0161; break;   /* S -hat */
5946 		    case 0xb4: c = 0x017d; break;   /* Z hat */
5947 		    case 0xb8: c = 0x017e; break;   /* Z -hat */
5948 		    case 0xbc: c = 0x0152; break;   /* OE */
5949 		    case 0xbd: c = 0x0153; break;   /* oe */
5950 		    case 0xbe: c = 0x0178; break;   /* Y */
5951 		}
5952 		d += utf_char2bytes(c, d);
5953 	    }
5954 	    *d = NUL;
5955 	    if (lenp != NULL)
5956 		*lenp = (int)(d - retval);
5957 	    break;
5958 
5959 	case CONV_TO_LATIN1:	/* utf-8 to latin1 conversion */
5960 	case CONV_TO_LATIN9:	/* utf-8 to latin9 conversion */
5961 	    retval = alloc(len + 1);
5962 	    if (retval == NULL)
5963 		break;
5964 	    d = retval;
5965 	    for (i = 0; i < len; ++i)
5966 	    {
5967 		l = utf_ptr2len(ptr + i);
5968 		if (l == 0)
5969 		    *d++ = NUL;
5970 		else if (l == 1)
5971 		{
5972 		    if (unconvlenp != NULL && utf8len_tab[ptr[i]] > len - i)
5973 		    {
5974 			/* Incomplete sequence at the end. */
5975 			*unconvlenp = len - i;
5976 			break;
5977 		    }
5978 		    *d++ = ptr[i];
5979 		}
5980 		else
5981 		{
5982 		    c = utf_ptr2char(ptr + i);
5983 		    if (vcp->vc_type == CONV_TO_LATIN9)
5984 			switch (c)
5985 			{
5986 			    case 0x20ac: c = 0xa4; break;   /* euro */
5987 			    case 0x0160: c = 0xa6; break;   /* S hat */
5988 			    case 0x0161: c = 0xa8; break;   /* S -hat */
5989 			    case 0x017d: c = 0xb4; break;   /* Z hat */
5990 			    case 0x017e: c = 0xb8; break;   /* Z -hat */
5991 			    case 0x0152: c = 0xbc; break;   /* OE */
5992 			    case 0x0153: c = 0xbd; break;   /* oe */
5993 			    case 0x0178: c = 0xbe; break;   /* Y */
5994 			    case 0xa4:
5995 			    case 0xa6:
5996 			    case 0xa8:
5997 			    case 0xb4:
5998 			    case 0xb8:
5999 			    case 0xbc:
6000 			    case 0xbd:
6001 			    case 0xbe: c = 0x100; break; /* not in latin9 */
6002 			}
6003 		    if (!utf_iscomposing(c))	/* skip composing chars */
6004 		    {
6005 			if (c < 0x100)
6006 			    *d++ = c;
6007 			else if (vcp->vc_fail)
6008 			{
6009 			    vim_free(retval);
6010 			    return NULL;
6011 			}
6012 			else
6013 			{
6014 			    *d++ = 0xbf;
6015 			    if (utf_char2cells(c) > 1)
6016 				*d++ = '?';
6017 			}
6018 		    }
6019 		    i += l - 1;
6020 		}
6021 	    }
6022 	    *d = NUL;
6023 	    if (lenp != NULL)
6024 		*lenp = (int)(d - retval);
6025 	    break;
6026 
6027 # ifdef MACOS_CONVERT
6028 	case CONV_MAC_LATIN1:
6029 	    retval = mac_string_convert(ptr, len, lenp, vcp->vc_fail,
6030 					'm', 'l', unconvlenp);
6031 	    break;
6032 
6033 	case CONV_LATIN1_MAC:
6034 	    retval = mac_string_convert(ptr, len, lenp, vcp->vc_fail,
6035 					'l', 'm', unconvlenp);
6036 	    break;
6037 
6038 	case CONV_MAC_UTF8:
6039 	    retval = mac_string_convert(ptr, len, lenp, vcp->vc_fail,
6040 					'm', 'u', unconvlenp);
6041 	    break;
6042 
6043 	case CONV_UTF8_MAC:
6044 	    retval = mac_string_convert(ptr, len, lenp, vcp->vc_fail,
6045 					'u', 'm', unconvlenp);
6046 	    break;
6047 # endif
6048 
6049 # ifdef USE_ICONV
6050 	case CONV_ICONV:	/* conversion with output_conv.vc_fd */
6051 	    retval = iconv_string(vcp, ptr, len, unconvlenp);
6052 	    if (retval != NULL && lenp != NULL)
6053 		*lenp = (int)STRLEN(retval);
6054 	    break;
6055 # endif
6056 # ifdef WIN3264
6057 	case CONV_CODEPAGE:		/* codepage -> codepage */
6058 	{
6059 	    int		retlen;
6060 	    int		tmp_len;
6061 	    short_u	*tmp;
6062 
6063 	    /* 1. codepage/UTF-8  ->  ucs-2. */
6064 	    if (vcp->vc_cpfrom == 0)
6065 		tmp_len = utf8_to_ucs2(ptr, len, NULL, NULL);
6066 	    else
6067 		tmp_len = MultiByteToWideChar(vcp->vc_cpfrom, 0,
6068 							      ptr, len, 0, 0);
6069 	    tmp = (short_u *)alloc(sizeof(short_u) * tmp_len);
6070 	    if (tmp == NULL)
6071 		break;
6072 	    if (vcp->vc_cpfrom == 0)
6073 		utf8_to_ucs2(ptr, len, tmp, unconvlenp);
6074 	    else
6075 		MultiByteToWideChar(vcp->vc_cpfrom, 0, ptr, len, tmp, tmp_len);
6076 
6077 	    /* 2. ucs-2  ->  codepage/UTF-8. */
6078 	    if (vcp->vc_cpto == 0)
6079 		retlen = ucs2_to_utf8(tmp, tmp_len, NULL);
6080 	    else
6081 		retlen = WideCharToMultiByte(vcp->vc_cpto, 0,
6082 						    tmp, tmp_len, 0, 0, 0, 0);
6083 	    retval = alloc(retlen + 1);
6084 	    if (retval != NULL)
6085 	    {
6086 		if (vcp->vc_cpto == 0)
6087 		    ucs2_to_utf8(tmp, tmp_len, retval);
6088 		else
6089 		    WideCharToMultiByte(vcp->vc_cpto, 0,
6090 					  tmp, tmp_len, retval, retlen, 0, 0);
6091 		retval[retlen] = NUL;
6092 		if (lenp != NULL)
6093 		    *lenp = retlen;
6094 	    }
6095 	    vim_free(tmp);
6096 	    break;
6097 	}
6098 # endif
6099     }
6100 
6101     return retval;
6102 }
6103 #endif
6104