xref: /vim-8.2.3635/src/mbyte.c (revision dee2e315)
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 six 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)
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_ptr2len_len __ARGS((char_u *p, int size));
131 static int utf_ptr2cells_len __ARGS((char_u *p, int size));
132 static int dbcs_char2cells __ARGS((int c));
133 static int dbcs_ptr2cells_len __ARGS((char_u *p, int size));
134 static int dbcs_ptr2char __ARGS((char_u *p));
135 static int utf_safe_read_char_adv __ARGS((char_u **s, size_t *n));
136 
137 /*
138  * Lookup table to quickly get the length in bytes of a UTF-8 character from
139  * the first byte of a UTF-8 string.
140  * Bytes which are illegal when used as the first byte have a 1.
141  * The NUL byte has length 1.
142  */
143 static char utf8len_tab[256] =
144 {
145     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,
146     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,
147     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,
148     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,
149     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,
150     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,
151     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,
152     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,
153 };
154 
155 /*
156  * Like utf8len_tab above, but using a zero for illegal lead bytes.
157  */
158 static char utf8len_tab_zero[256] =
159 {
160     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,
161     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,
162     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,
163     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,
164     0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
165     0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
166     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,
167     3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,4,4,4,4,4,4,4,4,5,5,5,5,6,6,0,0,
168 };
169 
170 /*
171  * XIM often causes trouble.  Define XIM_DEBUG to get a log of XIM callbacks
172  * in the "xim.log" file.
173  */
174 /* #define XIM_DEBUG */
175 #ifdef XIM_DEBUG
176     static void
177 xim_log(char *s, ...)
178 {
179     va_list arglist;
180     static FILE *fd = NULL;
181 
182     if (fd == (FILE *)-1)
183 	return;
184     if (fd == NULL)
185     {
186 	fd = mch_fopen("xim.log", "w");
187 	if (fd == NULL)
188 	{
189 	    EMSG("Cannot open xim.log");
190 	    fd = (FILE *)-1;
191 	    return;
192 	}
193     }
194 
195     va_start(arglist, s);
196     vfprintf(fd, s, arglist);
197     va_end(arglist);
198 }
199 #endif
200 
201 #endif
202 
203 #if defined(FEAT_MBYTE) || defined(FEAT_POSTSCRIPT) || defined(PROTO)
204 /*
205  * Canonical encoding names and their properties.
206  * "iso-8859-n" is handled by enc_canonize() directly.
207  */
208 static struct
209 {   char *name;		int prop;		int codepage;}
210 enc_canon_table[] =
211 {
212 #define IDX_LATIN_1	0
213     {"latin1",		ENC_8BIT + ENC_LATIN1,	1252},
214 #define IDX_ISO_2	1
215     {"iso-8859-2",	ENC_8BIT,		0},
216 #define IDX_ISO_3	2
217     {"iso-8859-3",	ENC_8BIT,		0},
218 #define IDX_ISO_4	3
219     {"iso-8859-4",	ENC_8BIT,		0},
220 #define IDX_ISO_5	4
221     {"iso-8859-5",	ENC_8BIT,		0},
222 #define IDX_ISO_6	5
223     {"iso-8859-6",	ENC_8BIT,		0},
224 #define IDX_ISO_7	6
225     {"iso-8859-7",	ENC_8BIT,		0},
226 #define IDX_ISO_8	7
227     {"iso-8859-8",	ENC_8BIT,		0},
228 #define IDX_ISO_9	8
229     {"iso-8859-9",	ENC_8BIT,		0},
230 #define IDX_ISO_10	9
231     {"iso-8859-10",	ENC_8BIT,		0},
232 #define IDX_ISO_11	10
233     {"iso-8859-11",	ENC_8BIT,		0},
234 #define IDX_ISO_13	11
235     {"iso-8859-13",	ENC_8BIT,		0},
236 #define IDX_ISO_14	12
237     {"iso-8859-14",	ENC_8BIT,		0},
238 #define IDX_ISO_15	13
239     {"iso-8859-15",	ENC_8BIT + ENC_LATIN9,	0},
240 #define IDX_KOI8_R	14
241     {"koi8-r",		ENC_8BIT,		0},
242 #define IDX_KOI8_U	15
243     {"koi8-u",		ENC_8BIT,		0},
244 #define IDX_UTF8	16
245     {"utf-8",		ENC_UNICODE,		0},
246 #define IDX_UCS2	17
247     {"ucs-2",		ENC_UNICODE + ENC_ENDIAN_B + ENC_2BYTE, 0},
248 #define IDX_UCS2LE	18
249     {"ucs-2le",		ENC_UNICODE + ENC_ENDIAN_L + ENC_2BYTE, 0},
250 #define IDX_UTF16	19
251     {"utf-16",		ENC_UNICODE + ENC_ENDIAN_B + ENC_2WORD, 0},
252 #define IDX_UTF16LE	20
253     {"utf-16le",	ENC_UNICODE + ENC_ENDIAN_L + ENC_2WORD, 0},
254 #define IDX_UCS4	21
255     {"ucs-4",		ENC_UNICODE + ENC_ENDIAN_B + ENC_4BYTE, 0},
256 #define IDX_UCS4LE	22
257     {"ucs-4le",		ENC_UNICODE + ENC_ENDIAN_L + ENC_4BYTE, 0},
258 
259     /* For debugging DBCS encoding on Unix. */
260 #define IDX_DEBUG	23
261     {"debug",		ENC_DBCS,		DBCS_DEBUG},
262 #define IDX_EUC_JP	24
263     {"euc-jp",		ENC_DBCS,		DBCS_JPNU},
264 #define IDX_SJIS	25
265     {"sjis",		ENC_DBCS,		DBCS_JPN},
266 #define IDX_EUC_KR	26
267     {"euc-kr",		ENC_DBCS,		DBCS_KORU},
268 #define IDX_EUC_CN	27
269     {"euc-cn",		ENC_DBCS,		DBCS_CHSU},
270 #define IDX_EUC_TW	28
271     {"euc-tw",		ENC_DBCS,		DBCS_CHTU},
272 #define IDX_BIG5	29
273     {"big5",		ENC_DBCS,		DBCS_CHT},
274 
275     /* MS-DOS and MS-Windows codepages are included here, so that they can be
276      * used on Unix too.  Most of them are similar to ISO-8859 encodings, but
277      * not exactly the same. */
278 #define IDX_CP437	30
279     {"cp437",		ENC_8BIT,		437}, /* like iso-8859-1 */
280 #define IDX_CP737	31
281     {"cp737",		ENC_8BIT,		737}, /* like iso-8859-7 */
282 #define IDX_CP775	32
283     {"cp775",		ENC_8BIT,		775}, /* Baltic */
284 #define IDX_CP850	33
285     {"cp850",		ENC_8BIT,		850}, /* like iso-8859-4 */
286 #define IDX_CP852	34
287     {"cp852",		ENC_8BIT,		852}, /* like iso-8859-1 */
288 #define IDX_CP855	35
289     {"cp855",		ENC_8BIT,		855}, /* like iso-8859-2 */
290 #define IDX_CP857	36
291     {"cp857",		ENC_8BIT,		857}, /* like iso-8859-5 */
292 #define IDX_CP860	37
293     {"cp860",		ENC_8BIT,		860}, /* like iso-8859-9 */
294 #define IDX_CP861	38
295     {"cp861",		ENC_8BIT,		861}, /* like iso-8859-1 */
296 #define IDX_CP862	39
297     {"cp862",		ENC_8BIT,		862}, /* like iso-8859-1 */
298 #define IDX_CP863	40
299     {"cp863",		ENC_8BIT,		863}, /* like iso-8859-8 */
300 #define IDX_CP865	41
301     {"cp865",		ENC_8BIT,		865}, /* like iso-8859-1 */
302 #define IDX_CP866	42
303     {"cp866",		ENC_8BIT,		866}, /* like iso-8859-5 */
304 #define IDX_CP869	43
305     {"cp869",		ENC_8BIT,		869}, /* like iso-8859-7 */
306 #define IDX_CP874	44
307     {"cp874",		ENC_8BIT,		874}, /* Thai */
308 #define IDX_CP932	45
309     {"cp932",		ENC_DBCS,		DBCS_JPN},
310 #define IDX_CP936	46
311     {"cp936",		ENC_DBCS,		DBCS_CHS},
312 #define IDX_CP949	47
313     {"cp949",		ENC_DBCS,		DBCS_KOR},
314 #define IDX_CP950	48
315     {"cp950",		ENC_DBCS,		DBCS_CHT},
316 #define IDX_CP1250	49
317     {"cp1250",		ENC_8BIT,		1250}, /* Czech, Polish, etc. */
318 #define IDX_CP1251	50
319     {"cp1251",		ENC_8BIT,		1251}, /* Cyrillic */
320     /* cp1252 is considered to be equal to latin1 */
321 #define IDX_CP1253	51
322     {"cp1253",		ENC_8BIT,		1253}, /* Greek */
323 #define IDX_CP1254	52
324     {"cp1254",		ENC_8BIT,		1254}, /* Turkish */
325 #define IDX_CP1255	53
326     {"cp1255",		ENC_8BIT,		1255}, /* Hebrew */
327 #define IDX_CP1256	54
328     {"cp1256",		ENC_8BIT,		1256}, /* Arabic */
329 #define IDX_CP1257	55
330     {"cp1257",		ENC_8BIT,		1257}, /* Baltic */
331 #define IDX_CP1258	56
332     {"cp1258",		ENC_8BIT,		1258}, /* Vietnamese */
333 
334 #define IDX_MACROMAN	57
335     {"macroman",	ENC_8BIT + ENC_MACROMAN, 0},	/* Mac OS */
336 #define IDX_DECMCS	58
337     {"dec-mcs",		ENC_8BIT,		0},	/* DEC MCS */
338 #define IDX_HPROMAN8	59
339     {"hp-roman8",	ENC_8BIT,		0},	/* HP Roman8 */
340 #define IDX_COUNT	60
341 };
342 
343 /*
344  * Aliases for encoding names.
345  */
346 static struct
347 {   char *name;		int canon;}
348 enc_alias_table[] =
349 {
350     {"ansi",		IDX_LATIN_1},
351     {"iso-8859-1",	IDX_LATIN_1},
352     {"latin2",		IDX_ISO_2},
353     {"latin3",		IDX_ISO_3},
354     {"latin4",		IDX_ISO_4},
355     {"cyrillic",	IDX_ISO_5},
356     {"arabic",		IDX_ISO_6},
357     {"greek",		IDX_ISO_7},
358 #ifdef WIN3264
359     {"hebrew",		IDX_CP1255},
360 #else
361     {"hebrew",		IDX_ISO_8},
362 #endif
363     {"latin5",		IDX_ISO_9},
364     {"turkish",		IDX_ISO_9}, /* ? */
365     {"latin6",		IDX_ISO_10},
366     {"nordic",		IDX_ISO_10}, /* ? */
367     {"thai",		IDX_ISO_11}, /* ? */
368     {"latin7",		IDX_ISO_13},
369     {"latin8",		IDX_ISO_14},
370     {"latin9",		IDX_ISO_15},
371     {"utf8",		IDX_UTF8},
372     {"unicode",		IDX_UCS2},
373     {"ucs2",		IDX_UCS2},
374     {"ucs2be",		IDX_UCS2},
375     {"ucs-2be",		IDX_UCS2},
376     {"ucs2le",		IDX_UCS2LE},
377     {"utf16",		IDX_UTF16},
378     {"utf16be",		IDX_UTF16},
379     {"utf-16be",	IDX_UTF16},
380     {"utf16le",		IDX_UTF16LE},
381     {"ucs4",		IDX_UCS4},
382     {"ucs4be",		IDX_UCS4},
383     {"ucs-4be",		IDX_UCS4},
384     {"ucs4le",		IDX_UCS4LE},
385     {"utf32",		IDX_UCS4},
386     {"utf-32",		IDX_UCS4},
387     {"utf32be",		IDX_UCS4},
388     {"utf-32be",	IDX_UCS4},
389     {"utf32le",		IDX_UCS4LE},
390     {"utf-32le",	IDX_UCS4LE},
391     {"932",		IDX_CP932},
392     {"949",		IDX_CP949},
393     {"936",		IDX_CP936},
394     {"gbk",		IDX_CP936},
395     {"950",		IDX_CP950},
396     {"eucjp",		IDX_EUC_JP},
397     {"unix-jis",	IDX_EUC_JP},
398     {"ujis",		IDX_EUC_JP},
399     {"shift-jis",	IDX_SJIS},
400     {"euckr",		IDX_EUC_KR},
401     {"5601",		IDX_EUC_KR},	/* Sun: KS C 5601 */
402     {"euccn",		IDX_EUC_CN},
403     {"gb2312",		IDX_EUC_CN},
404     {"euctw",		IDX_EUC_TW},
405 #if defined(WIN3264) || defined(WIN32UNIX) || defined(MACOS)
406     {"japan",		IDX_CP932},
407     {"korea",		IDX_CP949},
408     {"prc",		IDX_CP936},
409     {"chinese",		IDX_CP936},
410     {"taiwan",		IDX_CP950},
411     {"big5",		IDX_CP950},
412 #else
413     {"japan",		IDX_EUC_JP},
414     {"korea",		IDX_EUC_KR},
415     {"prc",		IDX_EUC_CN},
416     {"chinese",		IDX_EUC_CN},
417     {"taiwan",		IDX_EUC_TW},
418     {"cp950",		IDX_BIG5},
419     {"950",		IDX_BIG5},
420 #endif
421     {"mac",		IDX_MACROMAN},
422     {"mac-roman",	IDX_MACROMAN},
423     {NULL,		0}
424 };
425 
426 #ifndef CP_UTF8
427 # define CP_UTF8 65001	/* magic number from winnls.h */
428 #endif
429 
430 /*
431  * Find encoding "name" in the list of canonical encoding names.
432  * Returns -1 if not found.
433  */
434     static int
435 enc_canon_search(name)
436     char_u	*name;
437 {
438     int		i;
439 
440     for (i = 0; i < IDX_COUNT; ++i)
441 	if (STRCMP(name, enc_canon_table[i].name) == 0)
442 	    return i;
443     return -1;
444 }
445 
446 #endif
447 
448 #if defined(FEAT_MBYTE) || defined(PROTO)
449 
450 /*
451  * Find canonical encoding "name" in the list and return its properties.
452  * Returns 0 if not found.
453  */
454     int
455 enc_canon_props(name)
456     char_u	*name;
457 {
458     int		i;
459 
460     i = enc_canon_search(name);
461     if (i >= 0)
462 	return enc_canon_table[i].prop;
463 #ifdef WIN3264
464     if (name[0] == 'c' && name[1] == 'p' && VIM_ISDIGIT(name[2]))
465     {
466 	CPINFO	cpinfo;
467 
468 	/* Get info on this codepage to find out what it is. */
469 	if (GetCPInfo(atoi(name + 2), &cpinfo) != 0)
470 	{
471 	    if (cpinfo.MaxCharSize == 1) /* some single-byte encoding */
472 		return ENC_8BIT;
473 	    if (cpinfo.MaxCharSize == 2
474 		    && (cpinfo.LeadByte[0] != 0 || cpinfo.LeadByte[1] != 0))
475 		/* must be a DBCS encoding */
476 		return ENC_DBCS;
477 	}
478 	return 0;
479     }
480 #endif
481     if (STRNCMP(name, "2byte-", 6) == 0)
482 	return ENC_DBCS;
483     if (STRNCMP(name, "8bit-", 5) == 0 || STRNCMP(name, "iso-8859-", 9) == 0)
484 	return ENC_8BIT;
485     return 0;
486 }
487 
488 /*
489  * Set up for using multi-byte characters.
490  * Called in three cases:
491  * - by main() to initialize (p_enc == NULL)
492  * - by set_init_1() after 'encoding' was set to its default.
493  * - by do_set() when 'encoding' has been set.
494  * p_enc must have been passed through enc_canonize() already.
495  * Sets the "enc_unicode", "enc_utf8", "enc_dbcs" and "has_mbyte" flags.
496  * Fills mb_bytelen_tab[] and returns NULL when there are no problems.
497  * When there is something wrong: Returns an error message and doesn't change
498  * anything.
499  */
500     char_u *
501 mb_init()
502 {
503     int		i;
504     int		idx;
505     int		n;
506     int		enc_dbcs_new = 0;
507 #if defined(USE_ICONV) && !defined(WIN3264) && !defined(WIN32UNIX) \
508 	&& !defined(MACOS)
509 # define LEN_FROM_CONV
510     vimconv_T	vimconv;
511     char_u	*p;
512 #endif
513 
514     if (p_enc == NULL)
515     {
516 	/* Just starting up: set the whole table to one's. */
517 	for (i = 0; i < 256; ++i)
518 	    mb_bytelen_tab[i] = 1;
519 	input_conv.vc_type = CONV_NONE;
520 	input_conv.vc_factor = 1;
521 	output_conv.vc_type = CONV_NONE;
522 	return NULL;
523     }
524 
525 #ifdef WIN3264
526     if (p_enc[0] == 'c' && p_enc[1] == 'p' && VIM_ISDIGIT(p_enc[2]))
527     {
528 	CPINFO	cpinfo;
529 
530 	/* Get info on this codepage to find out what it is. */
531 	if (GetCPInfo(atoi(p_enc + 2), &cpinfo) != 0)
532 	{
533 	    if (cpinfo.MaxCharSize == 1)
534 	    {
535 		/* some single-byte encoding */
536 		enc_unicode = 0;
537 		enc_utf8 = FALSE;
538 	    }
539 	    else if (cpinfo.MaxCharSize == 2
540 		    && (cpinfo.LeadByte[0] != 0 || cpinfo.LeadByte[1] != 0))
541 	    {
542 		/* must be a DBCS encoding, check below */
543 		enc_dbcs_new = atoi(p_enc + 2);
544 	    }
545 	    else
546 		goto codepage_invalid;
547 	}
548 	else if (GetLastError() == ERROR_INVALID_PARAMETER)
549 	{
550 codepage_invalid:
551 	    return (char_u *)N_("E543: Not a valid codepage");
552 	}
553     }
554 #endif
555     else if (STRNCMP(p_enc, "8bit-", 5) == 0
556 	    || STRNCMP(p_enc, "iso-8859-", 9) == 0)
557     {
558 	/* Accept any "8bit-" or "iso-8859-" name. */
559 	enc_unicode = 0;
560 	enc_utf8 = FALSE;
561     }
562     else if (STRNCMP(p_enc, "2byte-", 6) == 0)
563     {
564 #ifdef WIN3264
565 	/* Windows: accept only valid codepage numbers, check below. */
566 	if (p_enc[6] != 'c' || p_enc[7] != 'p'
567 				      || (enc_dbcs_new = atoi(p_enc + 8)) == 0)
568 	    return e_invarg;
569 #else
570 	/* Unix: accept any "2byte-" name, assume current locale. */
571 	enc_dbcs_new = DBCS_2BYTE;
572 #endif
573     }
574     else if ((idx = enc_canon_search(p_enc)) >= 0)
575     {
576 	i = enc_canon_table[idx].prop;
577 	if (i & ENC_UNICODE)
578 	{
579 	    /* Unicode */
580 	    enc_utf8 = TRUE;
581 	    if (i & (ENC_2BYTE | ENC_2WORD))
582 		enc_unicode = 2;
583 	    else if (i & ENC_4BYTE)
584 		enc_unicode = 4;
585 	    else
586 		enc_unicode = 0;
587 	}
588 	else if (i & ENC_DBCS)
589 	{
590 	    /* 2byte, handle below */
591 	    enc_dbcs_new = enc_canon_table[idx].codepage;
592 	}
593 	else
594 	{
595 	    /* Must be 8-bit. */
596 	    enc_unicode = 0;
597 	    enc_utf8 = FALSE;
598 	}
599     }
600     else    /* Don't know what encoding this is, reject it. */
601 	return e_invarg;
602 
603     if (enc_dbcs_new != 0)
604     {
605 #ifdef WIN3264
606 	/* Check if the DBCS code page is OK. */
607 	if (!IsValidCodePage(enc_dbcs_new))
608 	    goto codepage_invalid;
609 #endif
610 	enc_unicode = 0;
611 	enc_utf8 = FALSE;
612     }
613     enc_dbcs = enc_dbcs_new;
614     has_mbyte = (enc_dbcs != 0 || enc_utf8);
615 
616 #if defined(WIN3264) || defined(FEAT_CYGWIN_WIN32_CLIPBOARD)
617     enc_codepage = encname2codepage(p_enc);
618     enc_latin9 = (STRCMP(p_enc, "iso-8859-15") == 0);
619 #endif
620 
621     /* Detect an encoding that uses latin1 characters. */
622     enc_latin1like = (enc_utf8 || STRCMP(p_enc, "latin1") == 0
623 					|| STRCMP(p_enc, "iso-8859-15") == 0);
624 
625     /*
626      * Set the function pointers.
627      */
628     if (enc_utf8)
629     {
630 	mb_ptr2len = utfc_ptr2len;
631 	mb_ptr2len_len = utfc_ptr2len_len;
632 	mb_char2len = utf_char2len;
633 	mb_char2bytes = utf_char2bytes;
634 	mb_ptr2cells = utf_ptr2cells;
635 	mb_ptr2cells_len = utf_ptr2cells_len;
636 	mb_char2cells = utf_char2cells;
637 	mb_off2cells = utf_off2cells;
638 	mb_ptr2char = utf_ptr2char;
639 	mb_head_off = utf_head_off;
640     }
641     else if (enc_dbcs != 0)
642     {
643 	mb_ptr2len = dbcs_ptr2len;
644 	mb_ptr2len_len = dbcs_ptr2len_len;
645 	mb_char2len = dbcs_char2len;
646 	mb_char2bytes = dbcs_char2bytes;
647 	mb_ptr2cells = dbcs_ptr2cells;
648 	mb_ptr2cells_len = dbcs_ptr2cells_len;
649 	mb_char2cells = dbcs_char2cells;
650 	mb_off2cells = dbcs_off2cells;
651 	mb_ptr2char = dbcs_ptr2char;
652 	mb_head_off = dbcs_head_off;
653     }
654     else
655     {
656 	mb_ptr2len = latin_ptr2len;
657 	mb_ptr2len_len = latin_ptr2len_len;
658 	mb_char2len = latin_char2len;
659 	mb_char2bytes = latin_char2bytes;
660 	mb_ptr2cells = latin_ptr2cells;
661 	mb_ptr2cells_len = latin_ptr2cells_len;
662 	mb_char2cells = latin_char2cells;
663 	mb_off2cells = latin_off2cells;
664 	mb_ptr2char = latin_ptr2char;
665 	mb_head_off = latin_head_off;
666     }
667 
668     /*
669      * Fill the mb_bytelen_tab[] for MB_BYTE2LEN().
670      */
671 #ifdef LEN_FROM_CONV
672     /* When 'encoding' is different from the current locale mblen() won't
673      * work.  Use conversion to "utf-8" instead. */
674     vimconv.vc_type = CONV_NONE;
675     if (enc_dbcs)
676     {
677 	p = enc_locale();
678 	if (p == NULL || STRCMP(p, p_enc) != 0)
679 	{
680 	    convert_setup(&vimconv, p_enc, (char_u *)"utf-8");
681 	    vimconv.vc_fail = TRUE;
682 	}
683 	vim_free(p);
684     }
685 #endif
686 
687     for (i = 0; i < 256; ++i)
688     {
689 	/* Our own function to reliably check the length of UTF-8 characters,
690 	 * independent of mblen(). */
691 	if (enc_utf8)
692 	    n = utf8len_tab[i];
693 	else if (enc_dbcs == 0)
694 	    n = 1;
695 	else
696 	{
697 #if defined(WIN3264) || defined(WIN32UNIX)
698 	    /* enc_dbcs is set by setting 'fileencoding'.  It becomes a Windows
699 	     * CodePage identifier, which we can pass directly in to Windows
700 	     * API */
701 	    n = IsDBCSLeadByteEx(enc_dbcs, (BYTE)i) ? 2 : 1;
702 #else
703 # if defined(MACOS) || defined(__amigaos4__)
704 	    /*
705 	     * if mblen() is not available, character which MSB is turned on
706 	     * are treated as leading byte character. (note : This assumption
707 	     * is not always true.)
708 	     */
709 	    n = (i & 0x80) ? 2 : 1;
710 # else
711 	    char buf[MB_MAXBYTES + 1];
712 # ifdef X_LOCALE
713 #  ifndef mblen
714 #   define mblen _Xmblen
715 #  endif
716 # endif
717 	    if (i == NUL)	/* just in case mblen() can't handle "" */
718 		n = 1;
719 	    else
720 	    {
721 		buf[0] = i;
722 		buf[1] = 0;
723 #ifdef LEN_FROM_CONV
724 		if (vimconv.vc_type != CONV_NONE)
725 		{
726 		    /*
727 		     * string_convert() should fail when converting the first
728 		     * byte of a double-byte character.
729 		     */
730 		    p = string_convert(&vimconv, (char_u *)buf, NULL);
731 		    if (p != NULL)
732 		    {
733 			vim_free(p);
734 			n = 1;
735 		    }
736 		    else
737 			n = 2;
738 		}
739 		else
740 #endif
741 		{
742 		    /*
743 		     * mblen() should return -1 for invalid (means the leading
744 		     * multibyte) character.  However there are some platforms
745 		     * where mblen() returns 0 for invalid character.
746 		     * Therefore, following condition includes 0.
747 		     */
748 		    ignored = mblen(NULL, 0);	/* First reset the state. */
749 		    if (mblen(buf, (size_t)1) <= 0)
750 			n = 2;
751 		    else
752 			n = 1;
753 		}
754 	    }
755 # endif
756 #endif
757 	}
758 
759 	mb_bytelen_tab[i] = n;
760     }
761 
762 #ifdef LEN_FROM_CONV
763     convert_setup(&vimconv, NULL, NULL);
764 #endif
765 
766     /* The cell width depends on the type of multi-byte characters. */
767     (void)init_chartab();
768 
769     /* When enc_utf8 is set or reset, (de)allocate ScreenLinesUC[] */
770     screenalloc(FALSE);
771 
772     /* When using Unicode, set default for 'fileencodings'. */
773     if (enc_utf8 && !option_was_set((char_u *)"fencs"))
774 	set_string_option_direct((char_u *)"fencs", -1,
775 		       (char_u *)"ucs-bom,utf-8,default,latin1", OPT_FREE, 0);
776 
777 #if defined(HAVE_BIND_TEXTDOMAIN_CODESET) && defined(FEAT_GETTEXT)
778     /* GNU gettext 0.10.37 supports this feature: set the codeset used for
779      * translated messages independently from the current locale. */
780     (void)bind_textdomain_codeset(VIMPACKAGE,
781 					  enc_utf8 ? "utf-8" : (char *)p_enc);
782 #endif
783 
784 #ifdef WIN32
785     /* When changing 'encoding' while starting up, then convert the command
786      * line arguments from the active codepage to 'encoding'. */
787     if (starting != 0)
788 	fix_arg_enc();
789 #endif
790 
791 #ifdef FEAT_AUTOCMD
792     /* Fire an autocommand to let people do custom font setup. This must be
793      * after Vim has been setup for the new encoding. */
794     apply_autocmds(EVENT_ENCODINGCHANGED, NULL, (char_u *)"", FALSE, curbuf);
795 #endif
796 
797 #ifdef FEAT_SPELL
798     /* Need to reload spell dictionaries */
799     spell_reload();
800 #endif
801 
802     return NULL;
803 }
804 
805 /*
806  * Return the size of the BOM for the current buffer:
807  * 0 - no BOM
808  * 2 - UCS-2 or UTF-16 BOM
809  * 4 - UCS-4 BOM
810  * 3 - UTF-8 BOM
811  */
812     int
813 bomb_size()
814 {
815     int n = 0;
816 
817     if (curbuf->b_p_bomb && !curbuf->b_p_bin)
818     {
819 	if (*curbuf->b_p_fenc == NUL)
820 	{
821 	    if (enc_utf8)
822 	    {
823 		if (enc_unicode != 0)
824 		    n = enc_unicode;
825 		else
826 		    n = 3;
827 	    }
828 	}
829 	else if (STRCMP(curbuf->b_p_fenc, "utf-8") == 0)
830 	    n = 3;
831 	else if (STRNCMP(curbuf->b_p_fenc, "ucs-2", 5) == 0
832 		|| STRNCMP(curbuf->b_p_fenc, "utf-16", 6) == 0)
833 	    n = 2;
834 	else if (STRNCMP(curbuf->b_p_fenc, "ucs-4", 5) == 0)
835 	    n = 4;
836     }
837     return n;
838 }
839 
840 /*
841  * Remove all BOM from "s" by moving remaining text.
842  */
843     void
844 remove_bom(s)
845     char_u *s;
846 {
847     if (enc_utf8)
848     {
849 	char_u *p = s;
850 
851 	while ((p = vim_strbyte(p, 0xef)) != NULL)
852 	{
853 	    if (p[1] == 0xbb && p[2] == 0xbf)
854 		STRMOVE(p, p + 3);
855 	    else
856 		++p;
857 	}
858     }
859 }
860 
861 /*
862  * Get class of pointer:
863  * 0 for blank or NUL
864  * 1 for punctuation
865  * 2 for an (ASCII) word character
866  * >2 for other word characters
867  */
868     int
869 mb_get_class(p)
870     char_u	*p;
871 {
872     return mb_get_class_buf(p, curbuf);
873 }
874 
875     int
876 mb_get_class_buf(p, buf)
877     char_u	*p;
878     buf_T	*buf;
879 {
880     if (MB_BYTE2LEN(p[0]) == 1)
881     {
882 	if (p[0] == NUL || vim_iswhite(p[0]))
883 	    return 0;
884 	if (vim_iswordc_buf(p[0], buf))
885 	    return 2;
886 	return 1;
887     }
888     if (enc_dbcs != 0 && p[0] != NUL && p[1] != NUL)
889 	return dbcs_class(p[0], p[1]);
890     if (enc_utf8)
891 	return utf_class(utf_ptr2char(p));
892     return 0;
893 }
894 
895 /*
896  * Get class of a double-byte character.  This always returns 3 or bigger.
897  * TODO: Should return 1 for punctuation.
898  */
899     int
900 dbcs_class(lead, trail)
901     unsigned	lead;
902     unsigned	trail;
903 {
904     switch (enc_dbcs)
905     {
906 	/* please add classify routine for your language in here */
907 
908 	case DBCS_JPNU:	/* ? */
909 	case DBCS_JPN:
910 	    {
911 		/* JIS code classification */
912 		unsigned char lb = lead;
913 		unsigned char tb = trail;
914 
915 		/* convert process code to JIS */
916 # if defined(WIN3264) || defined(WIN32UNIX) || defined(MACOS)
917 		/* process code is SJIS */
918 		if (lb <= 0x9f)
919 		    lb = (lb - 0x81) * 2 + 0x21;
920 		else
921 		    lb = (lb - 0xc1) * 2 + 0x21;
922 		if (tb <= 0x7e)
923 		    tb -= 0x1f;
924 		else if (tb <= 0x9e)
925 		    tb -= 0x20;
926 		else
927 		{
928 		    tb -= 0x7e;
929 		    lb += 1;
930 		}
931 # else
932 		/*
933 		 * XXX: Code page identification can not use with all
934 		 *	    system! So, some other encoding information
935 		 *	    will be needed.
936 		 *	    In japanese: SJIS,EUC,UNICODE,(JIS)
937 		 *	    Note that JIS-code system don't use as
938 		 *	    process code in most system because it uses
939 		 *	    escape sequences(JIS is context depend encoding).
940 		 */
941 		/* assume process code is JAPANESE-EUC */
942 		lb &= 0x7f;
943 		tb &= 0x7f;
944 # endif
945 		/* exceptions */
946 		switch (lb << 8 | tb)
947 		{
948 		    case 0x2121: /* ZENKAKU space */
949 			return 0;
950 		    case 0x2122: /* KU-TEN (Japanese comma) */
951 		    case 0x2123: /* TOU-TEN (Japanese period) */
952 		    case 0x2124: /* ZENKAKU comma */
953 		    case 0x2125: /* ZENKAKU period */
954 			return 1;
955 		    case 0x213c: /* prolongedsound handled as KATAKANA */
956 			return 13;
957 		}
958 		/* sieved by KU code */
959 		switch (lb)
960 		{
961 		    case 0x21:
962 		    case 0x22:
963 			/* special symbols */
964 			return 10;
965 		    case 0x23:
966 			/* alpha-numeric */
967 			return 11;
968 		    case 0x24:
969 			/* hiragana */
970 			return 12;
971 		    case 0x25:
972 			/* katakana */
973 			return 13;
974 		    case 0x26:
975 			/* greek */
976 			return 14;
977 		    case 0x27:
978 			/* russian */
979 			return 15;
980 		    case 0x28:
981 			/* lines */
982 			return 16;
983 		    default:
984 			/* kanji */
985 			return 17;
986 		}
987 	    }
988 
989 	case DBCS_KORU:	/* ? */
990 	case DBCS_KOR:
991 	    {
992 		/* KS code classification */
993 		unsigned char c1 = lead;
994 		unsigned char c2 = trail;
995 
996 		/*
997 		 * 20 : Hangul
998 		 * 21 : Hanja
999 		 * 22 : Symbols
1000 		 * 23 : Alpha-numeric/Roman Letter (Full width)
1001 		 * 24 : Hangul Letter(Alphabet)
1002 		 * 25 : Roman Numeral/Greek Letter
1003 		 * 26 : Box Drawings
1004 		 * 27 : Unit Symbols
1005 		 * 28 : Circled/Parenthesized Letter
1006 		 * 29 : Hiragana/Katakana
1007 		 * 30 : Cyrillic Letter
1008 		 */
1009 
1010 		if (c1 >= 0xB0 && c1 <= 0xC8)
1011 		    /* Hangul */
1012 		    return 20;
1013 #if defined(WIN3264) || defined(WIN32UNIX)
1014 		else if (c1 <= 0xA0 || c2 <= 0xA0)
1015 		    /* Extended Hangul Region : MS UHC(Unified Hangul Code) */
1016 		    /* c1: 0x81-0xA0 with c2: 0x41-0x5A, 0x61-0x7A, 0x81-0xFE
1017 		     * c1: 0xA1-0xC6 with c2: 0x41-0x5A, 0x61-0x7A, 0x81-0xA0
1018 		     */
1019 		    return 20;
1020 #endif
1021 
1022 		else if (c1 >= 0xCA && c1 <= 0xFD)
1023 		    /* Hanja */
1024 		    return 21;
1025 		else switch (c1)
1026 		{
1027 		    case 0xA1:
1028 		    case 0xA2:
1029 			/* Symbols */
1030 			return 22;
1031 		    case 0xA3:
1032 			/* Alpha-numeric */
1033 			return 23;
1034 		    case 0xA4:
1035 			/* Hangul Letter(Alphabet) */
1036 			return 24;
1037 		    case 0xA5:
1038 			/* Roman Numeral/Greek Letter */
1039 			return 25;
1040 		    case 0xA6:
1041 			/* Box Drawings */
1042 			return 26;
1043 		    case 0xA7:
1044 			/* Unit Symbols */
1045 			return 27;
1046 		    case 0xA8:
1047 		    case 0xA9:
1048 			if (c2 <= 0xAF)
1049 			    return 25;  /* Roman Letter */
1050 			else if (c2 >= 0xF6)
1051 			    return 22;  /* Symbols */
1052 			else
1053 			    /* Circled/Parenthesized Letter */
1054 			    return 28;
1055 		    case 0xAA:
1056 		    case 0xAB:
1057 			/* Hiragana/Katakana */
1058 			return 29;
1059 		    case 0xAC:
1060 			/* Cyrillic Letter */
1061 			return 30;
1062 		}
1063 	    }
1064 	default:
1065 	    break;
1066     }
1067     return 3;
1068 }
1069 
1070 /*
1071  * mb_char2len() function pointer.
1072  * Return length in bytes of character "c".
1073  * Returns 1 for a single-byte character.
1074  */
1075     int
1076 latin_char2len(c)
1077     int		c UNUSED;
1078 {
1079     return 1;
1080 }
1081 
1082     static int
1083 dbcs_char2len(c)
1084     int		c;
1085 {
1086     if (c >= 0x100)
1087 	return 2;
1088     return 1;
1089 }
1090 
1091 /*
1092  * mb_char2bytes() function pointer.
1093  * Convert a character to its bytes.
1094  * Returns the length in bytes.
1095  */
1096     int
1097 latin_char2bytes(c, buf)
1098     int		c;
1099     char_u	*buf;
1100 {
1101     buf[0] = c;
1102     return 1;
1103 }
1104 
1105     static int
1106 dbcs_char2bytes(c, buf)
1107     int		c;
1108     char_u	*buf;
1109 {
1110     if (c >= 0x100)
1111     {
1112 	buf[0] = (unsigned)c >> 8;
1113 	buf[1] = c;
1114 	/* Never use a NUL byte, it causes lots of trouble.  It's an invalid
1115 	 * character anyway. */
1116 	if (buf[1] == NUL)
1117 	    buf[1] = '\n';
1118 	return 2;
1119     }
1120     buf[0] = c;
1121     return 1;
1122 }
1123 
1124 /*
1125  * mb_ptr2len() function pointer.
1126  * Get byte length of character at "*p" but stop at a NUL.
1127  * For UTF-8 this includes following composing characters.
1128  * Returns 0 when *p is NUL.
1129  */
1130     int
1131 latin_ptr2len(p)
1132     char_u	*p;
1133 {
1134     return MB_BYTE2LEN(*p);
1135 }
1136 
1137     static int
1138 dbcs_ptr2len(p)
1139     char_u	*p;
1140 {
1141     int		len;
1142 
1143     /* Check if second byte is not missing. */
1144     len = MB_BYTE2LEN(*p);
1145     if (len == 2 && p[1] == NUL)
1146 	len = 1;
1147     return len;
1148 }
1149 
1150 /*
1151  * mb_ptr2len_len() function pointer.
1152  * Like mb_ptr2len(), but limit to read "size" bytes.
1153  * Returns 0 for an empty string.
1154  * Returns 1 for an illegal char or an incomplete byte sequence.
1155  */
1156     int
1157 latin_ptr2len_len(p, size)
1158     char_u	*p;
1159     int		size;
1160 {
1161     if (size < 1 || *p == NUL)
1162 	return 0;
1163     return 1;
1164 }
1165 
1166     static int
1167 dbcs_ptr2len_len(p, size)
1168     char_u	*p;
1169     int		size;
1170 {
1171     int		len;
1172 
1173     if (size < 1 || *p == NUL)
1174 	return 0;
1175     if (size == 1)
1176 	return 1;
1177     /* Check that second byte is not missing. */
1178     len = MB_BYTE2LEN(*p);
1179     if (len == 2 && p[1] == NUL)
1180 	len = 1;
1181     return len;
1182 }
1183 
1184 struct interval
1185 {
1186     long first;
1187     long last;
1188 };
1189 static int intable __ARGS((struct interval *table, size_t size, int c));
1190 
1191 /*
1192  * Return TRUE if "c" is in "table[size / sizeof(struct interval)]".
1193  */
1194     static int
1195 intable(table, size, c)
1196     struct interval	*table;
1197     size_t		size;
1198     int			c;
1199 {
1200     int mid, bot, top;
1201 
1202     /* first quick check for Latin1 etc. characters */
1203     if (c < table[0].first)
1204 	return FALSE;
1205 
1206     /* binary search in table */
1207     bot = 0;
1208     top = (int)(size / sizeof(struct interval) - 1);
1209     while (top >= bot)
1210     {
1211 	mid = (bot + top) / 2;
1212 	if (table[mid].last < c)
1213 	    bot = mid + 1;
1214 	else if (table[mid].first > c)
1215 	    top = mid - 1;
1216 	else
1217 	    return TRUE;
1218     }
1219     return FALSE;
1220 }
1221 
1222 /*
1223  * For UTF-8 character "c" return 2 for a double-width character, 1 for others.
1224  * Returns 4 or 6 for an unprintable character.
1225  * Is only correct for characters >= 0x80.
1226  * When p_ambw is "double", return 2 for a character with East Asian Width
1227  * class 'A'(mbiguous).
1228  */
1229     int
1230 utf_char2cells(c)
1231     int		c;
1232 {
1233     /* Sorted list of non-overlapping intervals of East Asian double width
1234      * characters, generated with ../runtime/tools/unicode.vim. */
1235     static struct interval doublewidth[] =
1236     {
1237 	{0x1100, 0x115f},
1238 	{0x11a3, 0x11a7},
1239 	{0x11fa, 0x11ff},
1240 	{0x2329, 0x232a},
1241 	{0x2e80, 0x2e99},
1242 	{0x2e9b, 0x2ef3},
1243 	{0x2f00, 0x2fd5},
1244 	{0x2ff0, 0x2ffb},
1245 	{0x3000, 0x3029},
1246 	{0x3030, 0x303e},
1247 	{0x3041, 0x3096},
1248 	{0x309b, 0x30ff},
1249 	{0x3105, 0x312d},
1250 	{0x3131, 0x318e},
1251 	{0x3190, 0x31b7},
1252 	{0x31c0, 0x31e3},
1253 	{0x31f0, 0x321e},
1254 	{0x3220, 0x3247},
1255 	{0x3250, 0x32fe},
1256 	{0x3300, 0x4dbf},
1257 	{0x4e00, 0xa48c},
1258 	{0xa490, 0xa4c6},
1259 	{0xa960, 0xa97c},
1260 	{0xac00, 0xd7a3},
1261 	{0xd7b0, 0xd7c6},
1262 	{0xd7cb, 0xd7fb},
1263 	{0xf900, 0xfaff},
1264 	{0xfe10, 0xfe19},
1265 	{0xfe30, 0xfe52},
1266 	{0xfe54, 0xfe66},
1267 	{0xfe68, 0xfe6b},
1268 	{0xff01, 0xff60},
1269 	{0xffe0, 0xffe6},
1270 	{0x1f200, 0x1f200},
1271 	{0x1f210, 0x1f231},
1272 	{0x1f240, 0x1f248},
1273 	{0x20000, 0x2fffd},
1274 	{0x30000, 0x3fffd}
1275     };
1276     /* Sorted list of non-overlapping intervals of East Asian Ambiguous
1277      * characters, generated with ../runtime/tools/unicode.vim. */
1278     static struct interval ambiguous[] =
1279     {
1280 	{0x00a1, 0x00a1},
1281 	{0x00a4, 0x00a4},
1282 	{0x00a7, 0x00a8},
1283 	{0x00aa, 0x00aa},
1284 	{0x00ad, 0x00ae},
1285 	{0x00b0, 0x00b4},
1286 	{0x00b6, 0x00ba},
1287 	{0x00bc, 0x00bf},
1288 	{0x00c6, 0x00c6},
1289 	{0x00d0, 0x00d0},
1290 	{0x00d7, 0x00d8},
1291 	{0x00de, 0x00e1},
1292 	{0x00e6, 0x00e6},
1293 	{0x00e8, 0x00ea},
1294 	{0x00ec, 0x00ed},
1295 	{0x00f0, 0x00f0},
1296 	{0x00f2, 0x00f3},
1297 	{0x00f7, 0x00fa},
1298 	{0x00fc, 0x00fc},
1299 	{0x00fe, 0x00fe},
1300 	{0x0101, 0x0101},
1301 	{0x0111, 0x0111},
1302 	{0x0113, 0x0113},
1303 	{0x011b, 0x011b},
1304 	{0x0126, 0x0127},
1305 	{0x012b, 0x012b},
1306 	{0x0131, 0x0133},
1307 	{0x0138, 0x0138},
1308 	{0x013f, 0x0142},
1309 	{0x0144, 0x0144},
1310 	{0x0148, 0x014b},
1311 	{0x014d, 0x014d},
1312 	{0x0152, 0x0153},
1313 	{0x0166, 0x0167},
1314 	{0x016b, 0x016b},
1315 	{0x01ce, 0x01ce},
1316 	{0x01d0, 0x01d0},
1317 	{0x01d2, 0x01d2},
1318 	{0x01d4, 0x01d4},
1319 	{0x01d6, 0x01d6},
1320 	{0x01d8, 0x01d8},
1321 	{0x01da, 0x01da},
1322 	{0x01dc, 0x01dc},
1323 	{0x0251, 0x0251},
1324 	{0x0261, 0x0261},
1325 	{0x02c4, 0x02c4},
1326 	{0x02c7, 0x02c7},
1327 	{0x02c9, 0x02cb},
1328 	{0x02cd, 0x02cd},
1329 	{0x02d0, 0x02d0},
1330 	{0x02d8, 0x02db},
1331 	{0x02dd, 0x02dd},
1332 	{0x02df, 0x02df},
1333 	{0x0391, 0x03a1},
1334 	{0x03a3, 0x03a9},
1335 	{0x03b1, 0x03c1},
1336 	{0x03c3, 0x03c9},
1337 	{0x0401, 0x0401},
1338 	{0x0410, 0x044f},
1339 	{0x0451, 0x0451},
1340 	{0x2010, 0x2010},
1341 	{0x2013, 0x2016},
1342 	{0x2018, 0x2019},
1343 	{0x201c, 0x201d},
1344 	{0x2020, 0x2022},
1345 	{0x2024, 0x2027},
1346 	{0x2030, 0x2030},
1347 	{0x2032, 0x2033},
1348 	{0x2035, 0x2035},
1349 	{0x203b, 0x203b},
1350 	{0x203e, 0x203e},
1351 	{0x2074, 0x2074},
1352 	{0x207f, 0x207f},
1353 	{0x2081, 0x2084},
1354 	{0x20ac, 0x20ac},
1355 	{0x2103, 0x2103},
1356 	{0x2105, 0x2105},
1357 	{0x2109, 0x2109},
1358 	{0x2113, 0x2113},
1359 	{0x2116, 0x2116},
1360 	{0x2121, 0x2122},
1361 	{0x2126, 0x2126},
1362 	{0x212b, 0x212b},
1363 	{0x2153, 0x2154},
1364 	{0x215b, 0x215e},
1365 	{0x2160, 0x216b},
1366 	{0x2170, 0x2179},
1367 	{0x2189, 0x2189},
1368 	{0x2190, 0x2199},
1369 	{0x21b8, 0x21b9},
1370 	{0x21d2, 0x21d2},
1371 	{0x21d4, 0x21d4},
1372 	{0x21e7, 0x21e7},
1373 	{0x2200, 0x2200},
1374 	{0x2202, 0x2203},
1375 	{0x2207, 0x2208},
1376 	{0x220b, 0x220b},
1377 	{0x220f, 0x220f},
1378 	{0x2211, 0x2211},
1379 	{0x2215, 0x2215},
1380 	{0x221a, 0x221a},
1381 	{0x221d, 0x2220},
1382 	{0x2223, 0x2223},
1383 	{0x2225, 0x2225},
1384 	{0x2227, 0x222c},
1385 	{0x222e, 0x222e},
1386 	{0x2234, 0x2237},
1387 	{0x223c, 0x223d},
1388 	{0x2248, 0x2248},
1389 	{0x224c, 0x224c},
1390 	{0x2252, 0x2252},
1391 	{0x2260, 0x2261},
1392 	{0x2264, 0x2267},
1393 	{0x226a, 0x226b},
1394 	{0x226e, 0x226f},
1395 	{0x2282, 0x2283},
1396 	{0x2286, 0x2287},
1397 	{0x2295, 0x2295},
1398 	{0x2299, 0x2299},
1399 	{0x22a5, 0x22a5},
1400 	{0x22bf, 0x22bf},
1401 	{0x2312, 0x2312},
1402 	{0x2460, 0x24e9},
1403 	{0x24eb, 0x254b},
1404 	{0x2550, 0x2573},
1405 	{0x2580, 0x258f},
1406 	{0x2592, 0x2595},
1407 	{0x25a0, 0x25a1},
1408 	{0x25a3, 0x25a9},
1409 	{0x25b2, 0x25b3},
1410 	{0x25b6, 0x25b7},
1411 	{0x25bc, 0x25bd},
1412 	{0x25c0, 0x25c1},
1413 	{0x25c6, 0x25c8},
1414 	{0x25cb, 0x25cb},
1415 	{0x25ce, 0x25d1},
1416 	{0x25e2, 0x25e5},
1417 	{0x25ef, 0x25ef},
1418 	{0x2605, 0x2606},
1419 	{0x2609, 0x2609},
1420 	{0x260e, 0x260f},
1421 	{0x2614, 0x2615},
1422 	{0x261c, 0x261c},
1423 	{0x261e, 0x261e},
1424 	{0x2640, 0x2640},
1425 	{0x2642, 0x2642},
1426 	{0x2660, 0x2661},
1427 	{0x2663, 0x2665},
1428 	{0x2667, 0x266a},
1429 	{0x266c, 0x266d},
1430 	{0x266f, 0x266f},
1431 	{0x269e, 0x269f},
1432 	{0x26be, 0x26bf},
1433 	{0x26c4, 0x26cd},
1434 	{0x26cf, 0x26e1},
1435 	{0x26e3, 0x26e3},
1436 	{0x26e8, 0x26ff},
1437 	{0x273d, 0x273d},
1438 	{0x2757, 0x2757},
1439 	{0x2776, 0x277f},
1440 	{0x2b55, 0x2b59},
1441 	{0x3248, 0x324f},
1442 	{0xe000, 0xf8ff},
1443 	{0xfffd, 0xfffd},
1444 	{0x1f100, 0x1f10a},
1445 	{0x1f110, 0x1f12d},
1446 	{0x1f131, 0x1f131},
1447 	{0x1f13d, 0x1f13d},
1448 	{0x1f13f, 0x1f13f},
1449 	{0x1f142, 0x1f142},
1450 	{0x1f146, 0x1f146},
1451 	{0x1f14a, 0x1f14e},
1452 	{0x1f157, 0x1f157},
1453 	{0x1f15f, 0x1f15f},
1454 	{0x1f179, 0x1f179},
1455 	{0x1f17b, 0x1f17c},
1456 	{0x1f17f, 0x1f17f},
1457 	{0x1f18a, 0x1f18d},
1458 	{0x1f190, 0x1f190},
1459 	{0xf0000, 0xffffd},
1460 	{0x100000, 0x10fffd}
1461     };
1462 
1463     if (c >= 0x100)
1464     {
1465 #ifdef USE_WCHAR_FUNCTIONS
1466 	/*
1467 	 * Assume the library function wcwidth() works better than our own
1468 	 * stuff.  It should return 1 for ambiguous width chars!
1469 	 */
1470 	int	n = wcwidth(c);
1471 
1472 	if (n < 0)
1473 	    return 6;		/* unprintable, displays <xxxx> */
1474 	if (n > 1)
1475 	    return n;
1476 #else
1477 	if (!utf_printable(c))
1478 	    return 6;		/* unprintable, displays <xxxx> */
1479 	if (intable(doublewidth, sizeof(doublewidth), c))
1480 	    return 2;
1481 #endif
1482     }
1483 
1484     /* Characters below 0x100 are influenced by 'isprint' option */
1485     else if (c >= 0x80 && !vim_isprintc(c))
1486 	return 4;		/* unprintable, displays <xx> */
1487 
1488     if (c >= 0x80 && *p_ambw == 'd' && intable(ambiguous, sizeof(ambiguous), c))
1489 	return 2;
1490 
1491     return 1;
1492 }
1493 
1494 /*
1495  * mb_ptr2cells() function pointer.
1496  * Return the number of display cells character at "*p" occupies.
1497  * This doesn't take care of unprintable characters, use ptr2cells() for that.
1498  */
1499     int
1500 latin_ptr2cells(p)
1501     char_u	*p UNUSED;
1502 {
1503     return 1;
1504 }
1505 
1506     int
1507 utf_ptr2cells(p)
1508     char_u	*p;
1509 {
1510     int		c;
1511 
1512     /* Need to convert to a wide character. */
1513     if (*p >= 0x80)
1514     {
1515 	c = utf_ptr2char(p);
1516 	/* An illegal byte is displayed as <xx>. */
1517 	if (utf_ptr2len(p) == 1 || c == NUL)
1518 	    return 4;
1519 	/* If the char is ASCII it must be an overlong sequence. */
1520 	if (c < 0x80)
1521 	    return char2cells(c);
1522 	return utf_char2cells(c);
1523     }
1524     return 1;
1525 }
1526 
1527     int
1528 dbcs_ptr2cells(p)
1529     char_u	*p;
1530 {
1531     /* Number of cells is equal to number of bytes, except for euc-jp when
1532      * the first byte is 0x8e. */
1533     if (enc_dbcs == DBCS_JPNU && *p == 0x8e)
1534 	return 1;
1535     return MB_BYTE2LEN(*p);
1536 }
1537 
1538 /*
1539  * mb_ptr2cells_len() function pointer.
1540  * Like mb_ptr2cells(), but limit string length to "size".
1541  * For an empty string or truncated character returns 1.
1542  */
1543     int
1544 latin_ptr2cells_len(p, size)
1545     char_u	*p UNUSED;
1546     int		size UNUSED;
1547 {
1548     return 1;
1549 }
1550 
1551     static int
1552 utf_ptr2cells_len(p, size)
1553     char_u	*p;
1554     int		size;
1555 {
1556     int		c;
1557 
1558     /* Need to convert to a wide character. */
1559     if (size > 0 && *p >= 0x80)
1560     {
1561 	if (utf_ptr2len_len(p, size) < utf8len_tab[*p])
1562 	    return 1;  /* truncated */
1563 	c = utf_ptr2char(p);
1564 	/* An illegal byte is displayed as <xx>. */
1565 	if (utf_ptr2len(p) == 1 || c == NUL)
1566 	    return 4;
1567 	/* If the char is ASCII it must be an overlong sequence. */
1568 	if (c < 0x80)
1569 	    return char2cells(c);
1570 	return utf_char2cells(c);
1571     }
1572     return 1;
1573 }
1574 
1575     static int
1576 dbcs_ptr2cells_len(p, size)
1577     char_u	*p;
1578     int		size;
1579 {
1580     /* Number of cells is equal to number of bytes, except for euc-jp when
1581      * the first byte is 0x8e. */
1582     if (size <= 1 || (enc_dbcs == DBCS_JPNU && *p == 0x8e))
1583 	return 1;
1584     return MB_BYTE2LEN(*p);
1585 }
1586 
1587 /*
1588  * mb_char2cells() function pointer.
1589  * Return the number of display cells character "c" occupies.
1590  * Only takes care of multi-byte chars, not "^C" and such.
1591  */
1592     int
1593 latin_char2cells(c)
1594     int		c UNUSED;
1595 {
1596     return 1;
1597 }
1598 
1599     static int
1600 dbcs_char2cells(c)
1601     int		c;
1602 {
1603     /* Number of cells is equal to number of bytes, except for euc-jp when
1604      * the first byte is 0x8e. */
1605     if (enc_dbcs == DBCS_JPNU && ((unsigned)c >> 8) == 0x8e)
1606 	return 1;
1607     /* use the first byte */
1608     return MB_BYTE2LEN((unsigned)c >> 8);
1609 }
1610 
1611 /*
1612  * Return the number of cells occupied by string "p".
1613  * Stop at a NUL character.  When "len" >= 0 stop at character "p[len]".
1614  */
1615     int
1616 mb_string2cells(p, len)
1617     char_u  *p;
1618     int	    len;
1619 {
1620     int i;
1621     int clen = 0;
1622 
1623     for (i = 0; (len < 0 || i < len) && p[i] != NUL; i += (*mb_ptr2len)(p + i))
1624 	clen += (*mb_ptr2cells)(p + i);
1625     return clen;
1626 }
1627 
1628 /*
1629  * mb_off2cells() function pointer.
1630  * Return number of display cells for char at ScreenLines[off].
1631  * We make sure that the offset used is less than "max_off".
1632  */
1633     int
1634 latin_off2cells(off, max_off)
1635     unsigned	off UNUSED;
1636     unsigned	max_off UNUSED;
1637 {
1638     return 1;
1639 }
1640 
1641     int
1642 dbcs_off2cells(off, max_off)
1643     unsigned	off;
1644     unsigned	max_off;
1645 {
1646     /* never check beyond end of the line */
1647     if (off >= max_off)
1648 	return 1;
1649 
1650     /* Number of cells is equal to number of bytes, except for euc-jp when
1651      * the first byte is 0x8e. */
1652     if (enc_dbcs == DBCS_JPNU && ScreenLines[off] == 0x8e)
1653 	return 1;
1654     return MB_BYTE2LEN(ScreenLines[off]);
1655 }
1656 
1657     int
1658 utf_off2cells(off, max_off)
1659     unsigned	off;
1660     unsigned	max_off;
1661 {
1662     return (off + 1 < max_off && ScreenLines[off + 1] == 0) ? 2 : 1;
1663 }
1664 
1665 /*
1666  * mb_ptr2char() function pointer.
1667  * Convert a byte sequence into a character.
1668  */
1669     int
1670 latin_ptr2char(p)
1671     char_u	*p;
1672 {
1673     return *p;
1674 }
1675 
1676     static int
1677 dbcs_ptr2char(p)
1678     char_u	*p;
1679 {
1680     if (MB_BYTE2LEN(*p) > 1 && p[1] != NUL)
1681 	return (p[0] << 8) + p[1];
1682     return *p;
1683 }
1684 
1685 /*
1686  * Convert a UTF-8 byte sequence to a wide character.
1687  * If the sequence is illegal or truncated by a NUL the first byte is
1688  * returned.
1689  * Does not include composing characters, of course.
1690  */
1691     int
1692 utf_ptr2char(p)
1693     char_u	*p;
1694 {
1695     int		len;
1696 
1697     if (p[0] < 0x80)	/* be quick for ASCII */
1698 	return p[0];
1699 
1700     len = utf8len_tab_zero[p[0]];
1701     if (len > 1 && (p[1] & 0xc0) == 0x80)
1702     {
1703 	if (len == 2)
1704 	    return ((p[0] & 0x1f) << 6) + (p[1] & 0x3f);
1705 	if ((p[2] & 0xc0) == 0x80)
1706 	{
1707 	    if (len == 3)
1708 		return ((p[0] & 0x0f) << 12) + ((p[1] & 0x3f) << 6)
1709 		    + (p[2] & 0x3f);
1710 	    if ((p[3] & 0xc0) == 0x80)
1711 	    {
1712 		if (len == 4)
1713 		    return ((p[0] & 0x07) << 18) + ((p[1] & 0x3f) << 12)
1714 			+ ((p[2] & 0x3f) << 6) + (p[3] & 0x3f);
1715 		if ((p[4] & 0xc0) == 0x80)
1716 		{
1717 		    if (len == 5)
1718 			return ((p[0] & 0x03) << 24) + ((p[1] & 0x3f) << 18)
1719 			    + ((p[2] & 0x3f) << 12) + ((p[3] & 0x3f) << 6)
1720 			    + (p[4] & 0x3f);
1721 		    if ((p[5] & 0xc0) == 0x80 && len == 6)
1722 			return ((p[0] & 0x01) << 30) + ((p[1] & 0x3f) << 24)
1723 			    + ((p[2] & 0x3f) << 18) + ((p[3] & 0x3f) << 12)
1724 			    + ((p[4] & 0x3f) << 6) + (p[5] & 0x3f);
1725 		}
1726 	    }
1727 	}
1728     }
1729     /* Illegal value, just return the first byte */
1730     return p[0];
1731 }
1732 
1733 /*
1734  * Convert a UTF-8 byte sequence to a wide character.
1735  * String is assumed to be terminated by NUL or after "n" bytes, whichever
1736  * comes first.
1737  * The function is safe in the sense that it never accesses memory beyond the
1738  * first "n" bytes of "s".
1739  *
1740  * On success, returns decoded codepoint, advances "s" to the beginning of
1741  * next character and decreases "n" accordingly.
1742  *
1743  * If end of string was reached, returns 0 and, if "n" > 0, advances "s" past
1744  * NUL byte.
1745  *
1746  * If byte sequence is illegal or incomplete, returns -1 and does not advance
1747  * "s".
1748  */
1749     static int
1750 utf_safe_read_char_adv(s, n)
1751     char_u      **s;
1752     size_t      *n;
1753 {
1754     int		c, k;
1755 
1756     if (*n == 0) /* end of buffer */
1757 	return 0;
1758 
1759     k = utf8len_tab_zero[**s];
1760 
1761     if (k == 1)
1762     {
1763 	/* ASCII character or NUL */
1764 	(*n)--;
1765 	return *(*s)++;
1766     }
1767 
1768     if ((size_t)k <= *n)
1769     {
1770 	/* We have a multibyte sequence and it isn't truncated by buffer
1771 	 * limits so utf_ptr2char() is safe to use. Or the first byte is
1772 	 * illegal (k=0), and it's also safe to use utf_ptr2char(). */
1773 	c = utf_ptr2char(*s);
1774 
1775 	/* On failure, utf_ptr2char() returns the first byte, so here we
1776 	 * check equality with the first byte. The only non-ASCII character
1777 	 * which equals the first byte of its own UTF-8 representation is
1778 	 * U+00C3 (UTF-8: 0xC3 0x83), so need to check that special case too.
1779 	 * It's safe even if n=1, else we would have k=2 > n. */
1780 	if (c != (int)(**s) || (c == 0xC3 && (*s)[1] == 0x83))
1781 	{
1782 	    /* byte sequence was successfully decoded */
1783 	    *s += k;
1784 	    *n -= k;
1785 	    return c;
1786 	}
1787     }
1788 
1789     /* byte sequence is incomplete or illegal */
1790     return -1;
1791 }
1792 
1793 /*
1794  * Get character at **pp and advance *pp to the next character.
1795  * Note: composing characters are skipped!
1796  */
1797     int
1798 mb_ptr2char_adv(pp)
1799     char_u	**pp;
1800 {
1801     int		c;
1802 
1803     c = (*mb_ptr2char)(*pp);
1804     *pp += (*mb_ptr2len)(*pp);
1805     return c;
1806 }
1807 
1808 /*
1809  * Get character at **pp and advance *pp to the next character.
1810  * Note: composing characters are returned as separate characters.
1811  */
1812     int
1813 mb_cptr2char_adv(pp)
1814     char_u	**pp;
1815 {
1816     int		c;
1817 
1818     c = (*mb_ptr2char)(*pp);
1819     if (enc_utf8)
1820 	*pp += utf_ptr2len(*pp);
1821     else
1822 	*pp += (*mb_ptr2len)(*pp);
1823     return c;
1824 }
1825 
1826 #if defined(FEAT_ARABIC) || defined(PROTO)
1827 /*
1828  * Check whether we are dealing with Arabic combining characters.
1829  * Note: these are NOT really composing characters!
1830  */
1831     int
1832 arabic_combine(one, two)
1833     int		one;	    /* first character */
1834     int		two;	    /* character just after "one" */
1835 {
1836     if (one == a_LAM)
1837 	return arabic_maycombine(two);
1838     return FALSE;
1839 }
1840 
1841 /*
1842  * Check whether we are dealing with a character that could be regarded as an
1843  * Arabic combining character, need to check the character before this.
1844  */
1845     int
1846 arabic_maycombine(two)
1847     int		two;
1848 {
1849     if (p_arshape && !p_tbidi)
1850 	return (two == a_ALEF_MADDA
1851 		    || two == a_ALEF_HAMZA_ABOVE
1852 		    || two == a_ALEF_HAMZA_BELOW
1853 		    || two == a_ALEF);
1854     return FALSE;
1855 }
1856 
1857 /*
1858  * Check if the character pointed to by "p2" is a composing character when it
1859  * comes after "p1".  For Arabic sometimes "ab" is replaced with "c", which
1860  * behaves like a composing character.
1861  */
1862     int
1863 utf_composinglike(p1, p2)
1864     char_u	*p1;
1865     char_u	*p2;
1866 {
1867     int		c2;
1868 
1869     c2 = utf_ptr2char(p2);
1870     if (utf_iscomposing(c2))
1871 	return TRUE;
1872     if (!arabic_maycombine(c2))
1873 	return FALSE;
1874     return arabic_combine(utf_ptr2char(p1), c2);
1875 }
1876 #endif
1877 
1878 /*
1879  * Convert a UTF-8 byte string to a wide character.  Also get up to MAX_MCO
1880  * composing characters.
1881  */
1882     int
1883 utfc_ptr2char(p, pcc)
1884     char_u	*p;
1885     int		*pcc;	/* return: composing chars, last one is 0 */
1886 {
1887     int		len;
1888     int		c;
1889     int		cc;
1890     int		i = 0;
1891 
1892     c = utf_ptr2char(p);
1893     len = utf_ptr2len(p);
1894 
1895     /* Only accept a composing char when the first char isn't illegal. */
1896     if ((len > 1 || *p < 0x80)
1897 	    && p[len] >= 0x80
1898 	    && UTF_COMPOSINGLIKE(p, p + len))
1899     {
1900 	cc = utf_ptr2char(p + len);
1901 	for (;;)
1902 	{
1903 	    pcc[i++] = cc;
1904 	    if (i == MAX_MCO)
1905 		break;
1906 	    len += utf_ptr2len(p + len);
1907 	    if (p[len] < 0x80 || !utf_iscomposing(cc = utf_ptr2char(p + len)))
1908 		break;
1909 	}
1910     }
1911 
1912     if (i < MAX_MCO)	/* last composing char must be 0 */
1913 	pcc[i] = 0;
1914 
1915     return c;
1916 }
1917 
1918 /*
1919  * Convert a UTF-8 byte string to a wide character.  Also get up to MAX_MCO
1920  * composing characters.  Use no more than p[maxlen].
1921  */
1922     int
1923 utfc_ptr2char_len(p, pcc, maxlen)
1924     char_u	*p;
1925     int		*pcc;	/* return: composing chars, last one is 0 */
1926     int		maxlen;
1927 {
1928     int		len;
1929     int		c;
1930     int		cc;
1931     int		i = 0;
1932 
1933     c = utf_ptr2char(p);
1934     len = utf_ptr2len_len(p, maxlen);
1935     /* Only accept a composing char when the first char isn't illegal. */
1936     if ((len > 1 || *p < 0x80)
1937 	    && len < maxlen
1938 	    && p[len] >= 0x80
1939 	    && UTF_COMPOSINGLIKE(p, p + len))
1940     {
1941 	cc = utf_ptr2char(p + len);
1942 	for (;;)
1943 	{
1944 	    pcc[i++] = cc;
1945 	    if (i == MAX_MCO)
1946 		break;
1947 	    len += utf_ptr2len_len(p + len, maxlen - len);
1948 	    if (len >= maxlen
1949 		    || p[len] < 0x80
1950 		    || !utf_iscomposing(cc = utf_ptr2char(p + len)))
1951 		break;
1952 	}
1953     }
1954 
1955     if (i < MAX_MCO)	/* last composing char must be 0 */
1956 	pcc[i] = 0;
1957 
1958     return c;
1959 }
1960 
1961 /*
1962  * Convert the character at screen position "off" to a sequence of bytes.
1963  * Includes the composing characters.
1964  * "buf" must at least have the length MB_MAXBYTES + 1.
1965  * Only to be used when ScreenLinesUC[off] != 0.
1966  * Returns the produced number of bytes.
1967  */
1968     int
1969 utfc_char2bytes(off, buf)
1970     int		off;
1971     char_u	*buf;
1972 {
1973     int		len;
1974     int		i;
1975 
1976     len = utf_char2bytes(ScreenLinesUC[off], buf);
1977     for (i = 0; i < Screen_mco; ++i)
1978     {
1979 	if (ScreenLinesC[i][off] == 0)
1980 	    break;
1981 	len += utf_char2bytes(ScreenLinesC[i][off], buf + len);
1982     }
1983     return len;
1984 }
1985 
1986 /*
1987  * Get the length of a UTF-8 byte sequence, not including any following
1988  * composing characters.
1989  * Returns 0 for "".
1990  * Returns 1 for an illegal byte sequence.
1991  */
1992     int
1993 utf_ptr2len(p)
1994     char_u	*p;
1995 {
1996     int		len;
1997     int		i;
1998 
1999     if (*p == NUL)
2000 	return 0;
2001     len = utf8len_tab[*p];
2002     for (i = 1; i < len; ++i)
2003 	if ((p[i] & 0xc0) != 0x80)
2004 	    return 1;
2005     return len;
2006 }
2007 
2008 /*
2009  * Return length of UTF-8 character, obtained from the first byte.
2010  * "b" must be between 0 and 255!
2011  * Returns 1 for an invalid first byte value.
2012  */
2013     int
2014 utf_byte2len(b)
2015     int		b;
2016 {
2017     return utf8len_tab[b];
2018 }
2019 
2020 /*
2021  * Get the length of UTF-8 byte sequence "p[size]".  Does not include any
2022  * following composing characters.
2023  * Returns 1 for "".
2024  * Returns 1 for an illegal byte sequence (also in incomplete byte seq.).
2025  * Returns number > "size" for an incomplete byte sequence.
2026  * Never returns zero.
2027  */
2028     int
2029 utf_ptr2len_len(p, size)
2030     char_u	*p;
2031     int		size;
2032 {
2033     int		len;
2034     int		i;
2035     int		m;
2036 
2037     len = utf8len_tab[*p];
2038     if (len == 1)
2039 	return 1;	/* NUL, ascii or illegal lead byte */
2040     if (len > size)
2041 	m = size;	/* incomplete byte sequence. */
2042     else
2043 	m = len;
2044     for (i = 1; i < m; ++i)
2045 	if ((p[i] & 0xc0) != 0x80)
2046 	    return 1;
2047     return len;
2048 }
2049 
2050 /*
2051  * Return the number of bytes the UTF-8 encoding of the character at "p" takes.
2052  * This includes following composing characters.
2053  */
2054     int
2055 utfc_ptr2len(p)
2056     char_u	*p;
2057 {
2058     int		len;
2059     int		b0 = *p;
2060 #ifdef FEAT_ARABIC
2061     int		prevlen;
2062 #endif
2063 
2064     if (b0 == NUL)
2065 	return 0;
2066     if (b0 < 0x80 && p[1] < 0x80)	/* be quick for ASCII */
2067 	return 1;
2068 
2069     /* Skip over first UTF-8 char, stopping at a NUL byte. */
2070     len = utf_ptr2len(p);
2071 
2072     /* Check for illegal byte. */
2073     if (len == 1 && b0 >= 0x80)
2074 	return 1;
2075 
2076     /*
2077      * Check for composing characters.  We can handle only the first six, but
2078      * skip all of them (otherwise the cursor would get stuck).
2079      */
2080 #ifdef FEAT_ARABIC
2081     prevlen = 0;
2082 #endif
2083     for (;;)
2084     {
2085 	if (p[len] < 0x80 || !UTF_COMPOSINGLIKE(p + prevlen, p + len))
2086 	    return len;
2087 
2088 	/* Skip over composing char */
2089 #ifdef FEAT_ARABIC
2090 	prevlen = len;
2091 #endif
2092 	len += utf_ptr2len(p + len);
2093     }
2094 }
2095 
2096 /*
2097  * Return the number of bytes the UTF-8 encoding of the character at "p[size]"
2098  * takes.  This includes following composing characters.
2099  * Returns 0 for an empty string.
2100  * Returns 1 for an illegal char or an incomplete byte sequence.
2101  */
2102     int
2103 utfc_ptr2len_len(p, size)
2104     char_u	*p;
2105     int		size;
2106 {
2107     int		len;
2108 #ifdef FEAT_ARABIC
2109     int		prevlen;
2110 #endif
2111 
2112     if (size < 1 || *p == NUL)
2113 	return 0;
2114     if (p[0] < 0x80 && (size == 1 || p[1] < 0x80)) /* be quick for ASCII */
2115 	return 1;
2116 
2117     /* Skip over first UTF-8 char, stopping at a NUL byte. */
2118     len = utf_ptr2len_len(p, size);
2119 
2120     /* Check for illegal byte and incomplete byte sequence. */
2121     if ((len == 1 && p[0] >= 0x80) || len > size)
2122 	return 1;
2123 
2124     /*
2125      * Check for composing characters.  We can handle only the first six, but
2126      * skip all of them (otherwise the cursor would get stuck).
2127      */
2128 #ifdef FEAT_ARABIC
2129     prevlen = 0;
2130 #endif
2131     while (len < size)
2132     {
2133 	int	len_next_char;
2134 
2135 	if (p[len] < 0x80)
2136 	    break;
2137 
2138 	/*
2139 	 * Next character length should not go beyond size to ensure that
2140 	 * UTF_COMPOSINGLIKE(...) does not read beyond size.
2141 	 */
2142 	len_next_char = utf_ptr2len_len(p + len, size - len);
2143 	if (len_next_char > size - len)
2144 	    break;
2145 
2146 	if (!UTF_COMPOSINGLIKE(p + prevlen, p + len))
2147 	    break;
2148 
2149 	/* Skip over composing char */
2150 #ifdef FEAT_ARABIC
2151 	prevlen = len;
2152 #endif
2153 	len += len_next_char;
2154     }
2155     return len;
2156 }
2157 
2158 /*
2159  * Return the number of bytes the UTF-8 encoding of character "c" takes.
2160  * This does not include composing characters.
2161  */
2162     int
2163 utf_char2len(c)
2164     int		c;
2165 {
2166     if (c < 0x80)
2167 	return 1;
2168     if (c < 0x800)
2169 	return 2;
2170     if (c < 0x10000)
2171 	return 3;
2172     if (c < 0x200000)
2173 	return 4;
2174     if (c < 0x4000000)
2175 	return 5;
2176     return 6;
2177 }
2178 
2179 /*
2180  * Convert Unicode character "c" to UTF-8 string in "buf[]".
2181  * Returns the number of bytes.
2182  * This does not include composing characters.
2183  */
2184     int
2185 utf_char2bytes(c, buf)
2186     int		c;
2187     char_u	*buf;
2188 {
2189     if (c < 0x80)		/* 7 bits */
2190     {
2191 	buf[0] = c;
2192 	return 1;
2193     }
2194     if (c < 0x800)		/* 11 bits */
2195     {
2196 	buf[0] = 0xc0 + ((unsigned)c >> 6);
2197 	buf[1] = 0x80 + (c & 0x3f);
2198 	return 2;
2199     }
2200     if (c < 0x10000)		/* 16 bits */
2201     {
2202 	buf[0] = 0xe0 + ((unsigned)c >> 12);
2203 	buf[1] = 0x80 + (((unsigned)c >> 6) & 0x3f);
2204 	buf[2] = 0x80 + (c & 0x3f);
2205 	return 3;
2206     }
2207     if (c < 0x200000)		/* 21 bits */
2208     {
2209 	buf[0] = 0xf0 + ((unsigned)c >> 18);
2210 	buf[1] = 0x80 + (((unsigned)c >> 12) & 0x3f);
2211 	buf[2] = 0x80 + (((unsigned)c >> 6) & 0x3f);
2212 	buf[3] = 0x80 + (c & 0x3f);
2213 	return 4;
2214     }
2215     if (c < 0x4000000)		/* 26 bits */
2216     {
2217 	buf[0] = 0xf8 + ((unsigned)c >> 24);
2218 	buf[1] = 0x80 + (((unsigned)c >> 18) & 0x3f);
2219 	buf[2] = 0x80 + (((unsigned)c >> 12) & 0x3f);
2220 	buf[3] = 0x80 + (((unsigned)c >> 6) & 0x3f);
2221 	buf[4] = 0x80 + (c & 0x3f);
2222 	return 5;
2223     }
2224 				/* 31 bits */
2225     buf[0] = 0xfc + ((unsigned)c >> 30);
2226     buf[1] = 0x80 + (((unsigned)c >> 24) & 0x3f);
2227     buf[2] = 0x80 + (((unsigned)c >> 18) & 0x3f);
2228     buf[3] = 0x80 + (((unsigned)c >> 12) & 0x3f);
2229     buf[4] = 0x80 + (((unsigned)c >> 6) & 0x3f);
2230     buf[5] = 0x80 + (c & 0x3f);
2231     return 6;
2232 }
2233 
2234 /*
2235  * Return TRUE if "c" is a composing UTF-8 character.  This means it will be
2236  * drawn on top of the preceding character.
2237  * Based on code from Markus Kuhn.
2238  */
2239     int
2240 utf_iscomposing(c)
2241     int		c;
2242 {
2243     /* Sorted list of non-overlapping intervals.
2244      * Generated by ../runtime/tools/unicode.vim. */
2245     static struct interval combining[] =
2246     {
2247 	{0x0300, 0x036f},
2248 	{0x0483, 0x0489},
2249 	{0x0591, 0x05bd},
2250 	{0x05bf, 0x05bf},
2251 	{0x05c1, 0x05c2},
2252 	{0x05c4, 0x05c5},
2253 	{0x05c7, 0x05c7},
2254 	{0x0610, 0x061a},
2255 	{0x064b, 0x065e},
2256 	{0x0670, 0x0670},
2257 	{0x06d6, 0x06dc},
2258 	{0x06de, 0x06e4},
2259 	{0x06e7, 0x06e8},
2260 	{0x06ea, 0x06ed},
2261 	{0x0711, 0x0711},
2262 	{0x0730, 0x074a},
2263 	{0x07a6, 0x07b0},
2264 	{0x07eb, 0x07f3},
2265 	{0x0816, 0x0819},
2266 	{0x081b, 0x0823},
2267 	{0x0825, 0x0827},
2268 	{0x0829, 0x082d},
2269 	{0x0900, 0x0903},
2270 	{0x093c, 0x093c},
2271 	{0x093e, 0x094e},
2272 	{0x0951, 0x0955},
2273 	{0x0962, 0x0963},
2274 	{0x0981, 0x0983},
2275 	{0x09bc, 0x09bc},
2276 	{0x09be, 0x09c4},
2277 	{0x09c7, 0x09c8},
2278 	{0x09cb, 0x09cd},
2279 	{0x09d7, 0x09d7},
2280 	{0x09e2, 0x09e3},
2281 	{0x0a01, 0x0a03},
2282 	{0x0a3c, 0x0a3c},
2283 	{0x0a3e, 0x0a42},
2284 	{0x0a47, 0x0a48},
2285 	{0x0a4b, 0x0a4d},
2286 	{0x0a51, 0x0a51},
2287 	{0x0a70, 0x0a71},
2288 	{0x0a75, 0x0a75},
2289 	{0x0a81, 0x0a83},
2290 	{0x0abc, 0x0abc},
2291 	{0x0abe, 0x0ac5},
2292 	{0x0ac7, 0x0ac9},
2293 	{0x0acb, 0x0acd},
2294 	{0x0ae2, 0x0ae3},
2295 	{0x0b01, 0x0b03},
2296 	{0x0b3c, 0x0b3c},
2297 	{0x0b3e, 0x0b44},
2298 	{0x0b47, 0x0b48},
2299 	{0x0b4b, 0x0b4d},
2300 	{0x0b56, 0x0b57},
2301 	{0x0b62, 0x0b63},
2302 	{0x0b82, 0x0b82},
2303 	{0x0bbe, 0x0bc2},
2304 	{0x0bc6, 0x0bc8},
2305 	{0x0bca, 0x0bcd},
2306 	{0x0bd7, 0x0bd7},
2307 	{0x0c01, 0x0c03},
2308 	{0x0c3e, 0x0c44},
2309 	{0x0c46, 0x0c48},
2310 	{0x0c4a, 0x0c4d},
2311 	{0x0c55, 0x0c56},
2312 	{0x0c62, 0x0c63},
2313 	{0x0c82, 0x0c83},
2314 	{0x0cbc, 0x0cbc},
2315 	{0x0cbe, 0x0cc4},
2316 	{0x0cc6, 0x0cc8},
2317 	{0x0cca, 0x0ccd},
2318 	{0x0cd5, 0x0cd6},
2319 	{0x0ce2, 0x0ce3},
2320 	{0x0d02, 0x0d03},
2321 	{0x0d3e, 0x0d44},
2322 	{0x0d46, 0x0d48},
2323 	{0x0d4a, 0x0d4d},
2324 	{0x0d57, 0x0d57},
2325 	{0x0d62, 0x0d63},
2326 	{0x0d82, 0x0d83},
2327 	{0x0dca, 0x0dca},
2328 	{0x0dcf, 0x0dd4},
2329 	{0x0dd6, 0x0dd6},
2330 	{0x0dd8, 0x0ddf},
2331 	{0x0df2, 0x0df3},
2332 	{0x0e31, 0x0e31},
2333 	{0x0e34, 0x0e3a},
2334 	{0x0e47, 0x0e4e},
2335 	{0x0eb1, 0x0eb1},
2336 	{0x0eb4, 0x0eb9},
2337 	{0x0ebb, 0x0ebc},
2338 	{0x0ec8, 0x0ecd},
2339 	{0x0f18, 0x0f19},
2340 	{0x0f35, 0x0f35},
2341 	{0x0f37, 0x0f37},
2342 	{0x0f39, 0x0f39},
2343 	{0x0f3e, 0x0f3f},
2344 	{0x0f71, 0x0f84},
2345 	{0x0f86, 0x0f87},
2346 	{0x0f90, 0x0f97},
2347 	{0x0f99, 0x0fbc},
2348 	{0x0fc6, 0x0fc6},
2349 	{0x102b, 0x103e},
2350 	{0x1056, 0x1059},
2351 	{0x105e, 0x1060},
2352 	{0x1062, 0x1064},
2353 	{0x1067, 0x106d},
2354 	{0x1071, 0x1074},
2355 	{0x1082, 0x108d},
2356 	{0x108f, 0x108f},
2357 	{0x109a, 0x109d},
2358 	{0x135f, 0x135f},
2359 	{0x1712, 0x1714},
2360 	{0x1732, 0x1734},
2361 	{0x1752, 0x1753},
2362 	{0x1772, 0x1773},
2363 	{0x17b6, 0x17d3},
2364 	{0x17dd, 0x17dd},
2365 	{0x180b, 0x180d},
2366 	{0x18a9, 0x18a9},
2367 	{0x1920, 0x192b},
2368 	{0x1930, 0x193b},
2369 	{0x19b0, 0x19c0},
2370 	{0x19c8, 0x19c9},
2371 	{0x1a17, 0x1a1b},
2372 	{0x1a55, 0x1a5e},
2373 	{0x1a60, 0x1a7c},
2374 	{0x1a7f, 0x1a7f},
2375 	{0x1b00, 0x1b04},
2376 	{0x1b34, 0x1b44},
2377 	{0x1b6b, 0x1b73},
2378 	{0x1b80, 0x1b82},
2379 	{0x1ba1, 0x1baa},
2380 	{0x1c24, 0x1c37},
2381 	{0x1cd0, 0x1cd2},
2382 	{0x1cd4, 0x1ce8},
2383 	{0x1ced, 0x1ced},
2384 	{0x1cf2, 0x1cf2},
2385 	{0x1dc0, 0x1de6},
2386 	{0x1dfd, 0x1dff},
2387 	{0x20d0, 0x20f0},
2388 	{0x2cef, 0x2cf1},
2389 	{0x2de0, 0x2dff},
2390 	{0x302a, 0x302f},
2391 	{0x3099, 0x309a},
2392 	{0xa66f, 0xa672},
2393 	{0xa67c, 0xa67d},
2394 	{0xa6f0, 0xa6f1},
2395 	{0xa802, 0xa802},
2396 	{0xa806, 0xa806},
2397 	{0xa80b, 0xa80b},
2398 	{0xa823, 0xa827},
2399 	{0xa880, 0xa881},
2400 	{0xa8b4, 0xa8c4},
2401 	{0xa8e0, 0xa8f1},
2402 	{0xa926, 0xa92d},
2403 	{0xa947, 0xa953},
2404 	{0xa980, 0xa983},
2405 	{0xa9b3, 0xa9c0},
2406 	{0xaa29, 0xaa36},
2407 	{0xaa43, 0xaa43},
2408 	{0xaa4c, 0xaa4d},
2409 	{0xaa7b, 0xaa7b},
2410 	{0xaab0, 0xaab0},
2411 	{0xaab2, 0xaab4},
2412 	{0xaab7, 0xaab8},
2413 	{0xaabe, 0xaabf},
2414 	{0xaac1, 0xaac1},
2415 	{0xabe3, 0xabea},
2416 	{0xabec, 0xabed},
2417 	{0xfb1e, 0xfb1e},
2418 	{0xfe00, 0xfe0f},
2419 	{0xfe20, 0xfe26},
2420 	{0x101fd, 0x101fd},
2421 	{0x10a01, 0x10a03},
2422 	{0x10a05, 0x10a06},
2423 	{0x10a0c, 0x10a0f},
2424 	{0x10a38, 0x10a3a},
2425 	{0x10a3f, 0x10a3f},
2426 	{0x11080, 0x11082},
2427 	{0x110b0, 0x110ba},
2428 	{0x1d165, 0x1d169},
2429 	{0x1d16d, 0x1d172},
2430 	{0x1d17b, 0x1d182},
2431 	{0x1d185, 0x1d18b},
2432 	{0x1d1aa, 0x1d1ad},
2433 	{0x1d242, 0x1d244},
2434 	{0xe0100, 0xe01ef}
2435     };
2436 
2437     return intable(combining, sizeof(combining), c);
2438 }
2439 
2440 /*
2441  * Return TRUE for characters that can be displayed in a normal way.
2442  * Only for characters of 0x100 and above!
2443  */
2444     int
2445 utf_printable(c)
2446     int		c;
2447 {
2448 #ifdef USE_WCHAR_FUNCTIONS
2449     /*
2450      * Assume the iswprint() library function works better than our own stuff.
2451      */
2452     return iswprint(c);
2453 #else
2454     /* Sorted list of non-overlapping intervals.
2455      * 0xd800-0xdfff is reserved for UTF-16, actually illegal. */
2456     static struct interval nonprint[] =
2457     {
2458 	{0x070f, 0x070f}, {0x180b, 0x180e}, {0x200b, 0x200f}, {0x202a, 0x202e},
2459 	{0x206a, 0x206f}, {0xd800, 0xdfff}, {0xfeff, 0xfeff}, {0xfff9, 0xfffb},
2460 	{0xfffe, 0xffff}
2461     };
2462 
2463     return !intable(nonprint, sizeof(nonprint), c);
2464 #endif
2465 }
2466 
2467 /*
2468  * Get class of a Unicode character.
2469  * 0: white space
2470  * 1: punctuation
2471  * 2 or bigger: some class of word character.
2472  */
2473     int
2474 utf_class(c)
2475     int		c;
2476 {
2477     /* sorted list of non-overlapping intervals */
2478     static struct clinterval
2479     {
2480 	unsigned short first;
2481 	unsigned short last;
2482 	unsigned short class;
2483     } classes[] =
2484     {
2485 	{0x037e, 0x037e, 1},		/* Greek question mark */
2486 	{0x0387, 0x0387, 1},		/* Greek ano teleia */
2487 	{0x055a, 0x055f, 1},		/* Armenian punctuation */
2488 	{0x0589, 0x0589, 1},		/* Armenian full stop */
2489 	{0x05be, 0x05be, 1},
2490 	{0x05c0, 0x05c0, 1},
2491 	{0x05c3, 0x05c3, 1},
2492 	{0x05f3, 0x05f4, 1},
2493 	{0x060c, 0x060c, 1},
2494 	{0x061b, 0x061b, 1},
2495 	{0x061f, 0x061f, 1},
2496 	{0x066a, 0x066d, 1},
2497 	{0x06d4, 0x06d4, 1},
2498 	{0x0700, 0x070d, 1},		/* Syriac punctuation */
2499 	{0x0964, 0x0965, 1},
2500 	{0x0970, 0x0970, 1},
2501 	{0x0df4, 0x0df4, 1},
2502 	{0x0e4f, 0x0e4f, 1},
2503 	{0x0e5a, 0x0e5b, 1},
2504 	{0x0f04, 0x0f12, 1},
2505 	{0x0f3a, 0x0f3d, 1},
2506 	{0x0f85, 0x0f85, 1},
2507 	{0x104a, 0x104f, 1},		/* Myanmar punctuation */
2508 	{0x10fb, 0x10fb, 1},		/* Georgian punctuation */
2509 	{0x1361, 0x1368, 1},		/* Ethiopic punctuation */
2510 	{0x166d, 0x166e, 1},		/* Canadian Syl. punctuation */
2511 	{0x1680, 0x1680, 0},
2512 	{0x169b, 0x169c, 1},
2513 	{0x16eb, 0x16ed, 1},
2514 	{0x1735, 0x1736, 1},
2515 	{0x17d4, 0x17dc, 1},		/* Khmer punctuation */
2516 	{0x1800, 0x180a, 1},		/* Mongolian punctuation */
2517 	{0x2000, 0x200b, 0},		/* spaces */
2518 	{0x200c, 0x2027, 1},		/* punctuation and symbols */
2519 	{0x2028, 0x2029, 0},
2520 	{0x202a, 0x202e, 1},		/* punctuation and symbols */
2521 	{0x202f, 0x202f, 0},
2522 	{0x2030, 0x205e, 1},		/* punctuation and symbols */
2523 	{0x205f, 0x205f, 0},
2524 	{0x2060, 0x27ff, 1},		/* punctuation and symbols */
2525 	{0x2070, 0x207f, 0x2070},	/* superscript */
2526 	{0x2080, 0x2094, 0x2080},	/* subscript */
2527 	{0x20a0, 0x27ff, 1},		/* all kinds of symbols */
2528 	{0x2800, 0x28ff, 0x2800},	/* braille */
2529 	{0x2900, 0x2998, 1},		/* arrows, brackets, etc. */
2530 	{0x29d8, 0x29db, 1},
2531 	{0x29fc, 0x29fd, 1},
2532 	{0x3000, 0x3000, 0},		/* ideographic space */
2533 	{0x3001, 0x3020, 1},		/* ideographic punctuation */
2534 	{0x3030, 0x3030, 1},
2535 	{0x303d, 0x303d, 1},
2536 	{0x3040, 0x309f, 0x3040},	/* Hiragana */
2537 	{0x30a0, 0x30ff, 0x30a0},	/* Katakana */
2538 	{0x3300, 0x9fff, 0x4e00},	/* CJK Ideographs */
2539 	{0xac00, 0xd7a3, 0xac00},	/* Hangul Syllables */
2540 	{0xf900, 0xfaff, 0x4e00},	/* CJK Ideographs */
2541 	{0xfd3e, 0xfd3f, 1},
2542 	{0xfe30, 0xfe6b, 1},		/* punctuation forms */
2543 	{0xff00, 0xff0f, 1},		/* half/fullwidth ASCII */
2544 	{0xff1a, 0xff20, 1},		/* half/fullwidth ASCII */
2545 	{0xff3b, 0xff40, 1},		/* half/fullwidth ASCII */
2546 	{0xff5b, 0xff65, 1},		/* half/fullwidth ASCII */
2547     };
2548     int bot = 0;
2549     int top = sizeof(classes) / sizeof(struct clinterval) - 1;
2550     int mid;
2551 
2552     /* First quick check for Latin1 characters, use 'iskeyword'. */
2553     if (c < 0x100)
2554     {
2555 	if (c == ' ' || c == '\t' || c == NUL || c == 0xa0)
2556 	    return 0;	    /* blank */
2557 	if (vim_iswordc(c))
2558 	    return 2;	    /* word character */
2559 	return 1;	    /* punctuation */
2560     }
2561 
2562     /* binary search in table */
2563     while (top >= bot)
2564     {
2565 	mid = (bot + top) / 2;
2566 	if (classes[mid].last < c)
2567 	    bot = mid + 1;
2568 	else if (classes[mid].first > c)
2569 	    top = mid - 1;
2570 	else
2571 	    return (int)classes[mid].class;
2572     }
2573 
2574     /* most other characters are "word" characters */
2575     return 2;
2576 }
2577 
2578 /*
2579  * Code for Unicode case-dependent operations.  Based on notes in
2580  * http://www.unicode.org/Public/UNIDATA/CaseFolding.txt
2581  * This code uses simple case folding, not full case folding.
2582  * Last updated for Unicode 5.2.
2583  */
2584 
2585 /*
2586  * The following tables are built by ../runtime/tools/unicode.vim.
2587  * They must be in numeric order, because we use binary search.
2588  * An entry such as {0x41,0x5a,1,32} means that Unicode characters in the
2589  * range from 0x41 to 0x5a inclusive, stepping by 1, are changed to
2590  * folded/upper/lower by adding 32.
2591  */
2592 typedef struct
2593 {
2594     int rangeStart;
2595     int rangeEnd;
2596     int step;
2597     int offset;
2598 } convertStruct;
2599 
2600 static convertStruct foldCase[] =
2601 {
2602 	{0x41,0x5a,1,32},
2603 	{0xb5,0xb5,-1,775},
2604 	{0xc0,0xd6,1,32},
2605 	{0xd8,0xde,1,32},
2606 	{0x100,0x12e,2,1},
2607 	{0x132,0x136,2,1},
2608 	{0x139,0x147,2,1},
2609 	{0x14a,0x176,2,1},
2610 	{0x178,0x178,-1,-121},
2611 	{0x179,0x17d,2,1},
2612 	{0x17f,0x17f,-1,-268},
2613 	{0x181,0x181,-1,210},
2614 	{0x182,0x184,2,1},
2615 	{0x186,0x186,-1,206},
2616 	{0x187,0x187,-1,1},
2617 	{0x189,0x18a,1,205},
2618 	{0x18b,0x18b,-1,1},
2619 	{0x18e,0x18e,-1,79},
2620 	{0x18f,0x18f,-1,202},
2621 	{0x190,0x190,-1,203},
2622 	{0x191,0x191,-1,1},
2623 	{0x193,0x193,-1,205},
2624 	{0x194,0x194,-1,207},
2625 	{0x196,0x196,-1,211},
2626 	{0x197,0x197,-1,209},
2627 	{0x198,0x198,-1,1},
2628 	{0x19c,0x19c,-1,211},
2629 	{0x19d,0x19d,-1,213},
2630 	{0x19f,0x19f,-1,214},
2631 	{0x1a0,0x1a4,2,1},
2632 	{0x1a6,0x1a6,-1,218},
2633 	{0x1a7,0x1a7,-1,1},
2634 	{0x1a9,0x1a9,-1,218},
2635 	{0x1ac,0x1ac,-1,1},
2636 	{0x1ae,0x1ae,-1,218},
2637 	{0x1af,0x1af,-1,1},
2638 	{0x1b1,0x1b2,1,217},
2639 	{0x1b3,0x1b5,2,1},
2640 	{0x1b7,0x1b7,-1,219},
2641 	{0x1b8,0x1bc,4,1},
2642 	{0x1c4,0x1c4,-1,2},
2643 	{0x1c5,0x1c5,-1,1},
2644 	{0x1c7,0x1c7,-1,2},
2645 	{0x1c8,0x1c8,-1,1},
2646 	{0x1ca,0x1ca,-1,2},
2647 	{0x1cb,0x1db,2,1},
2648 	{0x1de,0x1ee,2,1},
2649 	{0x1f1,0x1f1,-1,2},
2650 	{0x1f2,0x1f4,2,1},
2651 	{0x1f6,0x1f6,-1,-97},
2652 	{0x1f7,0x1f7,-1,-56},
2653 	{0x1f8,0x21e,2,1},
2654 	{0x220,0x220,-1,-130},
2655 	{0x222,0x232,2,1},
2656 	{0x23a,0x23a,-1,10795},
2657 	{0x23b,0x23b,-1,1},
2658 	{0x23d,0x23d,-1,-163},
2659 	{0x23e,0x23e,-1,10792},
2660 	{0x241,0x241,-1,1},
2661 	{0x243,0x243,-1,-195},
2662 	{0x244,0x244,-1,69},
2663 	{0x245,0x245,-1,71},
2664 	{0x246,0x24e,2,1},
2665 	{0x345,0x345,-1,116},
2666 	{0x370,0x372,2,1},
2667 	{0x376,0x376,-1,1},
2668 	{0x386,0x386,-1,38},
2669 	{0x388,0x38a,1,37},
2670 	{0x38c,0x38c,-1,64},
2671 	{0x38e,0x38f,1,63},
2672 	{0x391,0x3a1,1,32},
2673 	{0x3a3,0x3ab,1,32},
2674 	{0x3c2,0x3c2,-1,1},
2675 	{0x3cf,0x3cf,-1,8},
2676 	{0x3d0,0x3d0,-1,-30},
2677 	{0x3d1,0x3d1,-1,-25},
2678 	{0x3d5,0x3d5,-1,-15},
2679 	{0x3d6,0x3d6,-1,-22},
2680 	{0x3d8,0x3ee,2,1},
2681 	{0x3f0,0x3f0,-1,-54},
2682 	{0x3f1,0x3f1,-1,-48},
2683 	{0x3f4,0x3f4,-1,-60},
2684 	{0x3f5,0x3f5,-1,-64},
2685 	{0x3f7,0x3f7,-1,1},
2686 	{0x3f9,0x3f9,-1,-7},
2687 	{0x3fa,0x3fa,-1,1},
2688 	{0x3fd,0x3ff,1,-130},
2689 	{0x400,0x40f,1,80},
2690 	{0x410,0x42f,1,32},
2691 	{0x460,0x480,2,1},
2692 	{0x48a,0x4be,2,1},
2693 	{0x4c0,0x4c0,-1,15},
2694 	{0x4c1,0x4cd,2,1},
2695 	{0x4d0,0x524,2,1},
2696 	{0x531,0x556,1,48},
2697 	{0x10a0,0x10c5,1,7264},
2698 	{0x1e00,0x1e94,2,1},
2699 	{0x1e9b,0x1e9b,-1,-58},
2700 	{0x1e9e,0x1e9e,-1,-7615},
2701 	{0x1ea0,0x1efe,2,1},
2702 	{0x1f08,0x1f0f,1,-8},
2703 	{0x1f18,0x1f1d,1,-8},
2704 	{0x1f28,0x1f2f,1,-8},
2705 	{0x1f38,0x1f3f,1,-8},
2706 	{0x1f48,0x1f4d,1,-8},
2707 	{0x1f59,0x1f5f,2,-8},
2708 	{0x1f68,0x1f6f,1,-8},
2709 	{0x1f88,0x1f8f,1,-8},
2710 	{0x1f98,0x1f9f,1,-8},
2711 	{0x1fa8,0x1faf,1,-8},
2712 	{0x1fb8,0x1fb9,1,-8},
2713 	{0x1fba,0x1fbb,1,-74},
2714 	{0x1fbc,0x1fbc,-1,-9},
2715 	{0x1fbe,0x1fbe,-1,-7173},
2716 	{0x1fc8,0x1fcb,1,-86},
2717 	{0x1fcc,0x1fcc,-1,-9},
2718 	{0x1fd8,0x1fd9,1,-8},
2719 	{0x1fda,0x1fdb,1,-100},
2720 	{0x1fe8,0x1fe9,1,-8},
2721 	{0x1fea,0x1feb,1,-112},
2722 	{0x1fec,0x1fec,-1,-7},
2723 	{0x1ff8,0x1ff9,1,-128},
2724 	{0x1ffa,0x1ffb,1,-126},
2725 	{0x1ffc,0x1ffc,-1,-9},
2726 	{0x2126,0x2126,-1,-7517},
2727 	{0x212a,0x212a,-1,-8383},
2728 	{0x212b,0x212b,-1,-8262},
2729 	{0x2132,0x2132,-1,28},
2730 	{0x2160,0x216f,1,16},
2731 	{0x2183,0x2183,-1,1},
2732 	{0x24b6,0x24cf,1,26},
2733 	{0x2c00,0x2c2e,1,48},
2734 	{0x2c60,0x2c60,-1,1},
2735 	{0x2c62,0x2c62,-1,-10743},
2736 	{0x2c63,0x2c63,-1,-3814},
2737 	{0x2c64,0x2c64,-1,-10727},
2738 	{0x2c67,0x2c6b,2,1},
2739 	{0x2c6d,0x2c6d,-1,-10780},
2740 	{0x2c6e,0x2c6e,-1,-10749},
2741 	{0x2c6f,0x2c6f,-1,-10783},
2742 	{0x2c70,0x2c70,-1,-10782},
2743 	{0x2c72,0x2c75,3,1},
2744 	{0x2c7e,0x2c7f,1,-10815},
2745 	{0x2c80,0x2ce2,2,1},
2746 	{0x2ceb,0x2ced,2,1},
2747 	{0xa640,0xa65e,2,1},
2748 	{0xa662,0xa66c,2,1},
2749 	{0xa680,0xa696,2,1},
2750 	{0xa722,0xa72e,2,1},
2751 	{0xa732,0xa76e,2,1},
2752 	{0xa779,0xa77b,2,1},
2753 	{0xa77d,0xa77d,-1,-35332},
2754 	{0xa77e,0xa786,2,1},
2755 	{0xa78b,0xa78b,-1,1},
2756 	{0xff21,0xff3a,1,32},
2757 	{0x10400,0x10427,1,40}
2758 };
2759 
2760 static int utf_convert __ARGS((int a, convertStruct table[], int tableSize));
2761 static int utf_strnicmp __ARGS((char_u *s1, char_u *s2, size_t n1, size_t n2));
2762 
2763 /*
2764  * Generic conversion function for case operations.
2765  * Return the converted equivalent of "a", which is a UCS-4 character.  Use
2766  * the given conversion "table".  Uses binary search on "table".
2767  */
2768     static int
2769 utf_convert(a, table, tableSize)
2770     int			a;
2771     convertStruct	table[];
2772     int			tableSize;
2773 {
2774     int start, mid, end; /* indices into table */
2775     int entries = tableSize / sizeof(convertStruct);
2776 
2777     start = 0;
2778     end = entries;
2779     while (start < end)
2780     {
2781 	/* need to search further */
2782 	mid = (end + start) / 2;
2783 	if (table[mid].rangeEnd < a)
2784 	    start = mid + 1;
2785 	else
2786 	    end = mid;
2787     }
2788     if (start < entries
2789 	    && table[start].rangeStart <= a
2790 	    && a <= table[start].rangeEnd
2791 	    && (a - table[start].rangeStart) % table[start].step == 0)
2792 	return (a + table[start].offset);
2793     else
2794 	return a;
2795 }
2796 
2797 /*
2798  * Return the folded-case equivalent of "a", which is a UCS-4 character.  Uses
2799  * simple case folding.
2800  */
2801     int
2802 utf_fold(a)
2803     int		a;
2804 {
2805     return utf_convert(a, foldCase, (int)sizeof(foldCase));
2806 }
2807 
2808 static convertStruct toLower[] =
2809 {
2810 	{0x41,0x5a,1,32},
2811 	{0xc0,0xd6,1,32},
2812 	{0xd8,0xde,1,32},
2813 	{0x100,0x12e,2,1},
2814 	{0x130,0x130,-1,-199},
2815 	{0x132,0x136,2,1},
2816 	{0x139,0x147,2,1},
2817 	{0x14a,0x176,2,1},
2818 	{0x178,0x178,-1,-121},
2819 	{0x179,0x17d,2,1},
2820 	{0x181,0x181,-1,210},
2821 	{0x182,0x184,2,1},
2822 	{0x186,0x186,-1,206},
2823 	{0x187,0x187,-1,1},
2824 	{0x189,0x18a,1,205},
2825 	{0x18b,0x18b,-1,1},
2826 	{0x18e,0x18e,-1,79},
2827 	{0x18f,0x18f,-1,202},
2828 	{0x190,0x190,-1,203},
2829 	{0x191,0x191,-1,1},
2830 	{0x193,0x193,-1,205},
2831 	{0x194,0x194,-1,207},
2832 	{0x196,0x196,-1,211},
2833 	{0x197,0x197,-1,209},
2834 	{0x198,0x198,-1,1},
2835 	{0x19c,0x19c,-1,211},
2836 	{0x19d,0x19d,-1,213},
2837 	{0x19f,0x19f,-1,214},
2838 	{0x1a0,0x1a4,2,1},
2839 	{0x1a6,0x1a6,-1,218},
2840 	{0x1a7,0x1a7,-1,1},
2841 	{0x1a9,0x1a9,-1,218},
2842 	{0x1ac,0x1ac,-1,1},
2843 	{0x1ae,0x1ae,-1,218},
2844 	{0x1af,0x1af,-1,1},
2845 	{0x1b1,0x1b2,1,217},
2846 	{0x1b3,0x1b5,2,1},
2847 	{0x1b7,0x1b7,-1,219},
2848 	{0x1b8,0x1bc,4,1},
2849 	{0x1c4,0x1c4,-1,2},
2850 	{0x1c5,0x1c5,-1,1},
2851 	{0x1c7,0x1c7,-1,2},
2852 	{0x1c8,0x1c8,-1,1},
2853 	{0x1ca,0x1ca,-1,2},
2854 	{0x1cb,0x1db,2,1},
2855 	{0x1de,0x1ee,2,1},
2856 	{0x1f1,0x1f1,-1,2},
2857 	{0x1f2,0x1f4,2,1},
2858 	{0x1f6,0x1f6,-1,-97},
2859 	{0x1f7,0x1f7,-1,-56},
2860 	{0x1f8,0x21e,2,1},
2861 	{0x220,0x220,-1,-130},
2862 	{0x222,0x232,2,1},
2863 	{0x23a,0x23a,-1,10795},
2864 	{0x23b,0x23b,-1,1},
2865 	{0x23d,0x23d,-1,-163},
2866 	{0x23e,0x23e,-1,10792},
2867 	{0x241,0x241,-1,1},
2868 	{0x243,0x243,-1,-195},
2869 	{0x244,0x244,-1,69},
2870 	{0x245,0x245,-1,71},
2871 	{0x246,0x24e,2,1},
2872 	{0x370,0x372,2,1},
2873 	{0x376,0x376,-1,1},
2874 	{0x386,0x386,-1,38},
2875 	{0x388,0x38a,1,37},
2876 	{0x38c,0x38c,-1,64},
2877 	{0x38e,0x38f,1,63},
2878 	{0x391,0x3a1,1,32},
2879 	{0x3a3,0x3ab,1,32},
2880 	{0x3cf,0x3cf,-1,8},
2881 	{0x3d8,0x3ee,2,1},
2882 	{0x3f4,0x3f4,-1,-60},
2883 	{0x3f7,0x3f7,-1,1},
2884 	{0x3f9,0x3f9,-1,-7},
2885 	{0x3fa,0x3fa,-1,1},
2886 	{0x3fd,0x3ff,1,-130},
2887 	{0x400,0x40f,1,80},
2888 	{0x410,0x42f,1,32},
2889 	{0x460,0x480,2,1},
2890 	{0x48a,0x4be,2,1},
2891 	{0x4c0,0x4c0,-1,15},
2892 	{0x4c1,0x4cd,2,1},
2893 	{0x4d0,0x524,2,1},
2894 	{0x531,0x556,1,48},
2895 	{0x10a0,0x10c5,1,7264},
2896 	{0x1e00,0x1e94,2,1},
2897 	{0x1e9e,0x1e9e,-1,-7615},
2898 	{0x1ea0,0x1efe,2,1},
2899 	{0x1f08,0x1f0f,1,-8},
2900 	{0x1f18,0x1f1d,1,-8},
2901 	{0x1f28,0x1f2f,1,-8},
2902 	{0x1f38,0x1f3f,1,-8},
2903 	{0x1f48,0x1f4d,1,-8},
2904 	{0x1f59,0x1f5f,2,-8},
2905 	{0x1f68,0x1f6f,1,-8},
2906 	{0x1f88,0x1f8f,1,-8},
2907 	{0x1f98,0x1f9f,1,-8},
2908 	{0x1fa8,0x1faf,1,-8},
2909 	{0x1fb8,0x1fb9,1,-8},
2910 	{0x1fba,0x1fbb,1,-74},
2911 	{0x1fbc,0x1fbc,-1,-9},
2912 	{0x1fc8,0x1fcb,1,-86},
2913 	{0x1fcc,0x1fcc,-1,-9},
2914 	{0x1fd8,0x1fd9,1,-8},
2915 	{0x1fda,0x1fdb,1,-100},
2916 	{0x1fe8,0x1fe9,1,-8},
2917 	{0x1fea,0x1feb,1,-112},
2918 	{0x1fec,0x1fec,-1,-7},
2919 	{0x1ff8,0x1ff9,1,-128},
2920 	{0x1ffa,0x1ffb,1,-126},
2921 	{0x1ffc,0x1ffc,-1,-9},
2922 	{0x2126,0x2126,-1,-7517},
2923 	{0x212a,0x212a,-1,-8383},
2924 	{0x212b,0x212b,-1,-8262},
2925 	{0x2132,0x2132,-1,28},
2926 	{0x2160,0x216f,1,16},
2927 	{0x2183,0x2183,-1,1},
2928 	{0x24b6,0x24cf,1,26},
2929 	{0x2c00,0x2c2e,1,48},
2930 	{0x2c60,0x2c60,-1,1},
2931 	{0x2c62,0x2c62,-1,-10743},
2932 	{0x2c63,0x2c63,-1,-3814},
2933 	{0x2c64,0x2c64,-1,-10727},
2934 	{0x2c67,0x2c6b,2,1},
2935 	{0x2c6d,0x2c6d,-1,-10780},
2936 	{0x2c6e,0x2c6e,-1,-10749},
2937 	{0x2c6f,0x2c6f,-1,-10783},
2938 	{0x2c70,0x2c70,-1,-10782},
2939 	{0x2c72,0x2c75,3,1},
2940 	{0x2c7e,0x2c7f,1,-10815},
2941 	{0x2c80,0x2ce2,2,1},
2942 	{0x2ceb,0x2ced,2,1},
2943 	{0xa640,0xa65e,2,1},
2944 	{0xa662,0xa66c,2,1},
2945 	{0xa680,0xa696,2,1},
2946 	{0xa722,0xa72e,2,1},
2947 	{0xa732,0xa76e,2,1},
2948 	{0xa779,0xa77b,2,1},
2949 	{0xa77d,0xa77d,-1,-35332},
2950 	{0xa77e,0xa786,2,1},
2951 	{0xa78b,0xa78b,-1,1},
2952 	{0xff21,0xff3a,1,32},
2953 	{0x10400,0x10427,1,40}
2954 };
2955 
2956 static convertStruct toUpper[] =
2957 {
2958 	{0x61,0x7a,1,-32},
2959 	{0xb5,0xb5,-1,743},
2960 	{0xe0,0xf6,1,-32},  /* 0xdf (German sharp s) is not upper-cased */
2961 	{0xf8,0xfe,1,-32},
2962 	{0xff,0xff,-1,121},
2963 	{0x101,0x12f,2,-1},
2964 	{0x131,0x131,-1,-232},
2965 	{0x133,0x137,2,-1},
2966 	{0x13a,0x148,2,-1},
2967 	{0x14b,0x177,2,-1},
2968 	{0x17a,0x17e,2,-1},
2969 	{0x17f,0x17f,-1,-300},
2970 	{0x180,0x180,-1,195},
2971 	{0x183,0x185,2,-1},
2972 	{0x188,0x18c,4,-1},
2973 	{0x192,0x192,-1,-1},
2974 	{0x195,0x195,-1,97},
2975 	{0x199,0x199,-1,-1},
2976 	{0x19a,0x19a,-1,163},
2977 	{0x19e,0x19e,-1,130},
2978 	{0x1a1,0x1a5,2,-1},
2979 	{0x1a8,0x1ad,5,-1},
2980 	{0x1b0,0x1b4,4,-1},
2981 	{0x1b6,0x1b9,3,-1},
2982 	{0x1bd,0x1bd,-1,-1},
2983 	{0x1bf,0x1bf,-1,56},
2984 	{0x1c5,0x1c5,-1,-1},
2985 	{0x1c6,0x1c6,-1,-2},
2986 	{0x1c8,0x1c8,-1,-1},
2987 	{0x1c9,0x1c9,-1,-2},
2988 	{0x1cb,0x1cb,-1,-1},
2989 	{0x1cc,0x1cc,-1,-2},
2990 	{0x1ce,0x1dc,2,-1},
2991 	{0x1dd,0x1dd,-1,-79},
2992 	{0x1df,0x1ef,2,-1},
2993 	{0x1f2,0x1f2,-1,-1},
2994 	{0x1f3,0x1f3,-1,-2},
2995 	{0x1f5,0x1f9,4,-1},
2996 	{0x1fb,0x21f,2,-1},
2997 	{0x223,0x233,2,-1},
2998 	{0x23c,0x23c,-1,-1},
2999 	{0x23f,0x240,1,10815},
3000 	{0x242,0x247,5,-1},
3001 	{0x249,0x24f,2,-1},
3002 	{0x250,0x250,-1,10783},
3003 	{0x251,0x251,-1,10780},
3004 	{0x252,0x252,-1,10782},
3005 	{0x253,0x253,-1,-210},
3006 	{0x254,0x254,-1,-206},
3007 	{0x256,0x257,1,-205},
3008 	{0x259,0x259,-1,-202},
3009 	{0x25b,0x25b,-1,-203},
3010 	{0x260,0x260,-1,-205},
3011 	{0x263,0x263,-1,-207},
3012 	{0x268,0x268,-1,-209},
3013 	{0x269,0x269,-1,-211},
3014 	{0x26b,0x26b,-1,10743},
3015 	{0x26f,0x26f,-1,-211},
3016 	{0x271,0x271,-1,10749},
3017 	{0x272,0x272,-1,-213},
3018 	{0x275,0x275,-1,-214},
3019 	{0x27d,0x27d,-1,10727},
3020 	{0x280,0x283,3,-218},
3021 	{0x288,0x288,-1,-218},
3022 	{0x289,0x289,-1,-69},
3023 	{0x28a,0x28b,1,-217},
3024 	{0x28c,0x28c,-1,-71},
3025 	{0x292,0x292,-1,-219},
3026 	{0x345,0x345,-1,84},
3027 	{0x371,0x373,2,-1},
3028 	{0x377,0x377,-1,-1},
3029 	{0x37b,0x37d,1,130},
3030 	{0x3ac,0x3ac,-1,-38},
3031 	{0x3ad,0x3af,1,-37},
3032 	{0x3b1,0x3c1,1,-32},
3033 	{0x3c2,0x3c2,-1,-31},
3034 	{0x3c3,0x3cb,1,-32},
3035 	{0x3cc,0x3cc,-1,-64},
3036 	{0x3cd,0x3ce,1,-63},
3037 	{0x3d0,0x3d0,-1,-62},
3038 	{0x3d1,0x3d1,-1,-57},
3039 	{0x3d5,0x3d5,-1,-47},
3040 	{0x3d6,0x3d6,-1,-54},
3041 	{0x3d7,0x3d7,-1,-8},
3042 	{0x3d9,0x3ef,2,-1},
3043 	{0x3f0,0x3f0,-1,-86},
3044 	{0x3f1,0x3f1,-1,-80},
3045 	{0x3f2,0x3f2,-1,7},
3046 	{0x3f5,0x3f5,-1,-96},
3047 	{0x3f8,0x3fb,3,-1},
3048 	{0x430,0x44f,1,-32},
3049 	{0x450,0x45f,1,-80},
3050 	{0x461,0x481,2,-1},
3051 	{0x48b,0x4bf,2,-1},
3052 	{0x4c2,0x4ce,2,-1},
3053 	{0x4cf,0x4cf,-1,-15},
3054 	{0x4d1,0x525,2,-1},
3055 	{0x561,0x586,1,-48},
3056 	{0x1d79,0x1d79,-1,35332},
3057 	{0x1d7d,0x1d7d,-1,3814},
3058 	{0x1e01,0x1e95,2,-1},
3059 	{0x1e9b,0x1e9b,-1,-59},
3060 	{0x1ea1,0x1eff,2,-1},
3061 	{0x1f00,0x1f07,1,8},
3062 	{0x1f10,0x1f15,1,8},
3063 	{0x1f20,0x1f27,1,8},
3064 	{0x1f30,0x1f37,1,8},
3065 	{0x1f40,0x1f45,1,8},
3066 	{0x1f51,0x1f57,2,8},
3067 	{0x1f60,0x1f67,1,8},
3068 	{0x1f70,0x1f71,1,74},
3069 	{0x1f72,0x1f75,1,86},
3070 	{0x1f76,0x1f77,1,100},
3071 	{0x1f78,0x1f79,1,128},
3072 	{0x1f7a,0x1f7b,1,112},
3073 	{0x1f7c,0x1f7d,1,126},
3074 	{0x1f80,0x1f87,1,8},
3075 	{0x1f90,0x1f97,1,8},
3076 	{0x1fa0,0x1fa7,1,8},
3077 	{0x1fb0,0x1fb1,1,8},
3078 	{0x1fb3,0x1fb3,-1,9},
3079 	{0x1fbe,0x1fbe,-1,-7205},
3080 	{0x1fc3,0x1fc3,-1,9},
3081 	{0x1fd0,0x1fd1,1,8},
3082 	{0x1fe0,0x1fe1,1,8},
3083 	{0x1fe5,0x1fe5,-1,7},
3084 	{0x1ff3,0x1ff3,-1,9},
3085 	{0x214e,0x214e,-1,-28},
3086 	{0x2170,0x217f,1,-16},
3087 	{0x2184,0x2184,-1,-1},
3088 	{0x24d0,0x24e9,1,-26},
3089 	{0x2c30,0x2c5e,1,-48},
3090 	{0x2c61,0x2c61,-1,-1},
3091 	{0x2c65,0x2c65,-1,-10795},
3092 	{0x2c66,0x2c66,-1,-10792},
3093 	{0x2c68,0x2c6c,2,-1},
3094 	{0x2c73,0x2c76,3,-1},
3095 	{0x2c81,0x2ce3,2,-1},
3096 	{0x2cec,0x2cee,2,-1},
3097 	{0x2d00,0x2d25,1,-7264},
3098 	{0xa641,0xa65f,2,-1},
3099 	{0xa663,0xa66d,2,-1},
3100 	{0xa681,0xa697,2,-1},
3101 	{0xa723,0xa72f,2,-1},
3102 	{0xa733,0xa76f,2,-1},
3103 	{0xa77a,0xa77c,2,-1},
3104 	{0xa77f,0xa787,2,-1},
3105 	{0xa78c,0xa78c,-1,-1},
3106 	{0xff41,0xff5a,1,-32},
3107 	{0x10428,0x1044f,1,-40}
3108 };
3109 
3110 /*
3111  * Return the upper-case equivalent of "a", which is a UCS-4 character.  Use
3112  * simple case folding.
3113  */
3114     int
3115 utf_toupper(a)
3116     int		a;
3117 {
3118     /* If 'casemap' contains "keepascii" use ASCII style toupper(). */
3119     if (a < 128 && (cmp_flags & CMP_KEEPASCII))
3120 	return TOUPPER_ASC(a);
3121 
3122 #if defined(HAVE_TOWUPPER) && defined(__STDC_ISO_10646__)
3123     /* If towupper() is available and handles Unicode, use it. */
3124     if (!(cmp_flags & CMP_INTERNAL))
3125 	return towupper(a);
3126 #endif
3127 
3128     /* For characters below 128 use locale sensitive toupper(). */
3129     if (a < 128)
3130 	return TOUPPER_LOC(a);
3131 
3132     /* For any other characters use the above mapping table. */
3133     return utf_convert(a, toUpper, (int)sizeof(toUpper));
3134 }
3135 
3136     int
3137 utf_islower(a)
3138     int		a;
3139 {
3140     /* German sharp s is lower case but has no upper case equivalent. */
3141     return (utf_toupper(a) != a) || a == 0xdf;
3142 }
3143 
3144 /*
3145  * Return the lower-case equivalent of "a", which is a UCS-4 character.  Use
3146  * simple case folding.
3147  */
3148     int
3149 utf_tolower(a)
3150     int		a;
3151 {
3152     /* If 'casemap' contains "keepascii" use ASCII style tolower(). */
3153     if (a < 128 && (cmp_flags & CMP_KEEPASCII))
3154 	return TOLOWER_ASC(a);
3155 
3156 #if defined(HAVE_TOWLOWER) && defined(__STDC_ISO_10646__)
3157     /* If towlower() is available and handles Unicode, use it. */
3158     if (!(cmp_flags & CMP_INTERNAL))
3159 	return towlower(a);
3160 #endif
3161 
3162     /* For characters below 128 use locale sensitive tolower(). */
3163     if (a < 128)
3164 	return TOLOWER_LOC(a);
3165 
3166     /* For any other characters use the above mapping table. */
3167     return utf_convert(a, toLower, (int)sizeof(toLower));
3168 }
3169 
3170     int
3171 utf_isupper(a)
3172     int		a;
3173 {
3174     return (utf_tolower(a) != a);
3175 }
3176 
3177     static int
3178 utf_strnicmp(s1, s2, n1, n2)
3179     char_u      *s1, *s2;
3180     size_t      n1, n2;
3181 {
3182     int		c1, c2, cdiff;
3183     char_u	buffer[6];
3184 
3185     for (;;)
3186     {
3187 	c1 = utf_safe_read_char_adv(&s1, &n1);
3188 	c2 = utf_safe_read_char_adv(&s2, &n2);
3189 
3190 	if (c1 <= 0 || c2 <= 0)
3191 	    break;
3192 
3193 	if (c1 == c2)
3194 	    continue;
3195 
3196 	cdiff = utf_fold(c1) - utf_fold(c2);
3197 	if (cdiff != 0)
3198 	    return cdiff;
3199     }
3200 
3201     /* some string ended or has an incomplete/illegal character sequence */
3202 
3203     if (c1 == 0 || c2 == 0)
3204     {
3205 	/* some string ended. shorter string is smaller */
3206 	if (c1 == 0 && c2 == 0)
3207 	    return 0;
3208 	return c1 == 0 ? -1 : 1;
3209     }
3210 
3211     /* Continue with bytewise comparison to produce some result that
3212      * would make comparison operations involving this function transitive.
3213      *
3214      * If only one string had an error, comparison should be made with
3215      * folded version of the other string. In this case it is enough
3216      * to fold just one character to determine the result of comparison. */
3217 
3218     if (c1 != -1 && c2 == -1)
3219     {
3220 	n1 = utf_char2bytes(utf_fold(c1), buffer);
3221 	s1 = buffer;
3222     }
3223     else if (c2 != -1 && c1 == -1)
3224     {
3225 	n2 = utf_char2bytes(utf_fold(c2), buffer);
3226 	s2 = buffer;
3227     }
3228 
3229     while (n1 > 0 && n2 > 0 && *s1 != NUL && *s2 != NUL)
3230     {
3231 	cdiff = (int)(*s1) - (int)(*s2);
3232 	if (cdiff != 0)
3233 	    return cdiff;
3234 
3235 	s1++;
3236 	s2++;
3237 	n1--;
3238 	n2--;
3239     }
3240 
3241     if (n1 > 0 && *s1 == NUL)
3242 	n1 = 0;
3243     if (n2 > 0 && *s2 == NUL)
3244 	n2 = 0;
3245 
3246     if (n1 == 0 && n2 == 0)
3247 	return 0;
3248     return n1 == 0 ? -1 : 1;
3249 }
3250 
3251 /*
3252  * Version of strnicmp() that handles multi-byte characters.
3253  * Needed for Big5, Shift-JIS and UTF-8 encoding.  Other DBCS encodings can
3254  * probably use strnicmp(), because there are no ASCII characters in the
3255  * second byte.
3256  * Returns zero if s1 and s2 are equal (ignoring case), the difference between
3257  * two characters otherwise.
3258  */
3259     int
3260 mb_strnicmp(s1, s2, nn)
3261     char_u	*s1, *s2;
3262     size_t	nn;
3263 {
3264     int		i, l;
3265     int		cdiff;
3266     int		n = (int)nn;
3267 
3268     if (enc_utf8)
3269     {
3270 	return utf_strnicmp(s1, s2, nn, nn);
3271     }
3272     else
3273     {
3274 	for (i = 0; i < n; i += l)
3275 	{
3276 	    if (s1[i] == NUL && s2[i] == NUL)	/* both strings end */
3277 		return 0;
3278 
3279 	    l = (*mb_ptr2len)(s1 + i);
3280 	    if (l <= 1)
3281 	    {
3282 		/* Single byte: first check normally, then with ignore case. */
3283 		if (s1[i] != s2[i])
3284 		{
3285 		    cdiff = MB_TOLOWER(s1[i]) - MB_TOLOWER(s2[i]);
3286 		    if (cdiff != 0)
3287 			return cdiff;
3288 		}
3289 	    }
3290 	    else
3291 	    {
3292 		/* For non-Unicode multi-byte don't ignore case. */
3293 		if (l > n - i)
3294 		    l = n - i;
3295 		cdiff = STRNCMP(s1 + i, s2 + i, l);
3296 		if (cdiff != 0)
3297 		    return cdiff;
3298 	    }
3299 	}
3300     }
3301     return 0;
3302 }
3303 
3304 /*
3305  * "g8": show bytes of the UTF-8 char under the cursor.  Doesn't matter what
3306  * 'encoding' has been set to.
3307  */
3308     void
3309 show_utf8()
3310 {
3311     int		len;
3312     int		rlen = 0;
3313     char_u	*line;
3314     int		clen;
3315     int		i;
3316 
3317     /* Get the byte length of the char under the cursor, including composing
3318      * characters. */
3319     line = ml_get_cursor();
3320     len = utfc_ptr2len(line);
3321     if (len == 0)
3322     {
3323 	MSG("NUL");
3324 	return;
3325     }
3326 
3327     clen = 0;
3328     for (i = 0; i < len; ++i)
3329     {
3330 	if (clen == 0)
3331 	{
3332 	    /* start of (composing) character, get its length */
3333 	    if (i > 0)
3334 	    {
3335 		STRCPY(IObuff + rlen, "+ ");
3336 		rlen += 2;
3337 	    }
3338 	    clen = utf_ptr2len(line + i);
3339 	}
3340 	sprintf((char *)IObuff + rlen, "%02x ",
3341 		(line[i] == NL) ? NUL : line[i]);  /* NUL is stored as NL */
3342 	--clen;
3343 	rlen += (int)STRLEN(IObuff + rlen);
3344 	if (rlen > IOSIZE - 20)
3345 	    break;
3346     }
3347 
3348     msg(IObuff);
3349 }
3350 
3351 /*
3352  * mb_head_off() function pointer.
3353  * Return offset from "p" to the first byte of the character it points into.
3354  * If "p" points to the NUL at the end of the string return 0.
3355  * Returns 0 when already at the first byte of a character.
3356  */
3357     int
3358 latin_head_off(base, p)
3359     char_u	*base UNUSED;
3360     char_u	*p UNUSED;
3361 {
3362     return 0;
3363 }
3364 
3365     int
3366 dbcs_head_off(base, p)
3367     char_u	*base;
3368     char_u	*p;
3369 {
3370     char_u	*q;
3371 
3372     /* It can't be a trailing byte when not using DBCS, at the start of the
3373      * string or the previous byte can't start a double-byte. */
3374     if (p <= base || MB_BYTE2LEN(p[-1]) == 1 || *p == NUL)
3375 	return 0;
3376 
3377     /* This is slow: need to start at the base and go forward until the
3378      * byte we are looking for.  Return 1 when we went past it, 0 otherwise. */
3379     q = base;
3380     while (q < p)
3381 	q += dbcs_ptr2len(q);
3382     return (q == p) ? 0 : 1;
3383 }
3384 
3385 /*
3386  * Special version of dbcs_head_off() that works for ScreenLines[], where
3387  * single-width DBCS_JPNU characters are stored separately.
3388  */
3389     int
3390 dbcs_screen_head_off(base, p)
3391     char_u	*base;
3392     char_u	*p;
3393 {
3394     char_u	*q;
3395 
3396     /* It can't be a trailing byte when not using DBCS, at the start of the
3397      * string or the previous byte can't start a double-byte.
3398      * For euc-jp an 0x8e byte in the previous cell always means we have a
3399      * lead byte in the current cell. */
3400     if (p <= base
3401 	    || (enc_dbcs == DBCS_JPNU && p[-1] == 0x8e)
3402 	    || MB_BYTE2LEN(p[-1]) == 1
3403 	    || *p == NUL)
3404 	return 0;
3405 
3406     /* This is slow: need to start at the base and go forward until the
3407      * byte we are looking for.  Return 1 when we went past it, 0 otherwise.
3408      * For DBCS_JPNU look out for 0x8e, which means the second byte is not
3409      * stored as the next byte. */
3410     q = base;
3411     while (q < p)
3412     {
3413 	if (enc_dbcs == DBCS_JPNU && *q == 0x8e)
3414 	    ++q;
3415 	else
3416 	    q += dbcs_ptr2len(q);
3417     }
3418     return (q == p) ? 0 : 1;
3419 }
3420 
3421     int
3422 utf_head_off(base, p)
3423     char_u	*base;
3424     char_u	*p;
3425 {
3426     char_u	*q;
3427     char_u	*s;
3428     int		c;
3429     int		len;
3430 #ifdef FEAT_ARABIC
3431     char_u	*j;
3432 #endif
3433 
3434     if (*p < 0x80)		/* be quick for ASCII */
3435 	return 0;
3436 
3437     /* Skip backwards over trailing bytes: 10xx.xxxx
3438      * Skip backwards again if on a composing char. */
3439     for (q = p; ; --q)
3440     {
3441 	/* Move s to the last byte of this char. */
3442 	for (s = q; (s[1] & 0xc0) == 0x80; ++s)
3443 	    ;
3444 	/* Move q to the first byte of this char. */
3445 	while (q > base && (*q & 0xc0) == 0x80)
3446 	    --q;
3447 	/* Check for illegal sequence. Do allow an illegal byte after where we
3448 	 * started. */
3449 	len = utf8len_tab[*q];
3450 	if (len != (int)(s - q + 1) && len != (int)(p - q + 1))
3451 	    return 0;
3452 
3453 	if (q <= base)
3454 	    break;
3455 
3456 	c = utf_ptr2char(q);
3457 	if (utf_iscomposing(c))
3458 	    continue;
3459 
3460 #ifdef FEAT_ARABIC
3461 	if (arabic_maycombine(c))
3462 	{
3463 	    /* Advance to get a sneak-peak at the next char */
3464 	    j = q;
3465 	    --j;
3466 	    /* Move j to the first byte of this char. */
3467 	    while (j > base && (*j & 0xc0) == 0x80)
3468 		--j;
3469 	    if (arabic_combine(utf_ptr2char(j), c))
3470 		continue;
3471 	}
3472 #endif
3473 	break;
3474     }
3475 
3476     return (int)(p - q);
3477 }
3478 
3479 /*
3480  * Copy a character from "*fp" to "*tp" and advance the pointers.
3481  */
3482     void
3483 mb_copy_char(fp, tp)
3484     char_u	**fp;
3485     char_u	**tp;
3486 {
3487     int	    l = (*mb_ptr2len)(*fp);
3488 
3489     mch_memmove(*tp, *fp, (size_t)l);
3490     *tp += l;
3491     *fp += l;
3492 }
3493 
3494 /*
3495  * Return the offset from "p" to the first byte of a character.  When "p" is
3496  * at the start of a character 0 is returned, otherwise the offset to the next
3497  * character.  Can start anywhere in a stream of bytes.
3498  */
3499     int
3500 mb_off_next(base, p)
3501     char_u	*base;
3502     char_u	*p;
3503 {
3504     int		i;
3505     int		j;
3506 
3507     if (enc_utf8)
3508     {
3509 	if (*p < 0x80)		/* be quick for ASCII */
3510 	    return 0;
3511 
3512 	/* Find the next character that isn't 10xx.xxxx */
3513 	for (i = 0; (p[i] & 0xc0) == 0x80; ++i)
3514 	    ;
3515 	if (i > 0)
3516 	{
3517 	    /* Check for illegal sequence. */
3518 	    for (j = 0; p - j > base; ++j)
3519 		if ((p[-j] & 0xc0) != 0x80)
3520 		    break;
3521 	    if (utf8len_tab[p[-j]] != i + j)
3522 		return 0;
3523 	}
3524 	return i;
3525     }
3526 
3527     /* Only need to check if we're on a trail byte, it doesn't matter if we
3528      * want the offset to the next or current character. */
3529     return (*mb_head_off)(base, p);
3530 }
3531 
3532 /*
3533  * Return the offset from "p" to the last byte of the character it points
3534  * into.  Can start anywhere in a stream of bytes.
3535  */
3536     int
3537 mb_tail_off(base, p)
3538     char_u	*base;
3539     char_u	*p;
3540 {
3541     int		i;
3542     int		j;
3543 
3544     if (*p == NUL)
3545 	return 0;
3546 
3547     if (enc_utf8)
3548     {
3549 	/* Find the last character that is 10xx.xxxx */
3550 	for (i = 0; (p[i + 1] & 0xc0) == 0x80; ++i)
3551 	    ;
3552 	/* Check for illegal sequence. */
3553 	for (j = 0; p - j > base; ++j)
3554 	    if ((p[-j] & 0xc0) != 0x80)
3555 		break;
3556 	if (utf8len_tab[p[-j]] != i + j + 1)
3557 	    return 0;
3558 	return i;
3559     }
3560 
3561     /* It can't be the first byte if a double-byte when not using DBCS, at the
3562      * end of the string or the byte can't start a double-byte. */
3563     if (enc_dbcs == 0 || p[1] == NUL || MB_BYTE2LEN(*p) == 1)
3564 	return 0;
3565 
3566     /* Return 1 when on the lead byte, 0 when on the tail byte. */
3567     return 1 - dbcs_head_off(base, p);
3568 }
3569 
3570 /*
3571  * Find the next illegal byte sequence.
3572  */
3573     void
3574 utf_find_illegal()
3575 {
3576     pos_T	pos = curwin->w_cursor;
3577     char_u	*p;
3578     int		len;
3579     vimconv_T	vimconv;
3580     char_u	*tofree = NULL;
3581 
3582     vimconv.vc_type = CONV_NONE;
3583     if (enc_utf8 && (enc_canon_props(curbuf->b_p_fenc) & ENC_8BIT))
3584     {
3585 	/* 'encoding' is "utf-8" but we are editing a 8-bit encoded file,
3586 	 * possibly a utf-8 file with illegal bytes.  Setup for conversion
3587 	 * from utf-8 to 'fileencoding'. */
3588 	convert_setup(&vimconv, p_enc, curbuf->b_p_fenc);
3589     }
3590 
3591 #ifdef FEAT_VIRTUALEDIT
3592     curwin->w_cursor.coladd = 0;
3593 #endif
3594     for (;;)
3595     {
3596 	p = ml_get_cursor();
3597 	if (vimconv.vc_type != CONV_NONE)
3598 	{
3599 	    vim_free(tofree);
3600 	    tofree = string_convert(&vimconv, p, NULL);
3601 	    if (tofree == NULL)
3602 		break;
3603 	    p = tofree;
3604 	}
3605 
3606 	while (*p != NUL)
3607 	{
3608 	    /* Illegal means that there are not enough trail bytes (checked by
3609 	     * utf_ptr2len()) or too many of them (overlong sequence). */
3610 	    len = utf_ptr2len(p);
3611 	    if (*p >= 0x80 && (len == 1
3612 				     || utf_char2len(utf_ptr2char(p)) != len))
3613 	    {
3614 		if (vimconv.vc_type == CONV_NONE)
3615 		    curwin->w_cursor.col += (colnr_T)(p - ml_get_cursor());
3616 		else
3617 		{
3618 		    int	    l;
3619 
3620 		    len = (int)(p - tofree);
3621 		    for (p = ml_get_cursor(); *p != NUL && len-- > 0; p += l)
3622 		    {
3623 			l = utf_ptr2len(p);
3624 			curwin->w_cursor.col += l;
3625 		    }
3626 		}
3627 		goto theend;
3628 	    }
3629 	    p += len;
3630 	}
3631 	if (curwin->w_cursor.lnum == curbuf->b_ml.ml_line_count)
3632 	    break;
3633 	++curwin->w_cursor.lnum;
3634 	curwin->w_cursor.col = 0;
3635     }
3636 
3637     /* didn't find it: don't move and beep */
3638     curwin->w_cursor = pos;
3639     beep_flush();
3640 
3641 theend:
3642     vim_free(tofree);
3643     convert_setup(&vimconv, NULL, NULL);
3644 }
3645 
3646 #if defined(FEAT_GUI_GTK) || defined(PROTO)
3647 /*
3648  * Return TRUE if string "s" is a valid utf-8 string.
3649  * When "end" is NULL stop at the first NUL.
3650  * When "end" is positive stop there.
3651  */
3652     int
3653 utf_valid_string(s, end)
3654     char_u	*s;
3655     char_u	*end;
3656 {
3657     int		l;
3658     char_u	*p = s;
3659 
3660     while (end == NULL ? *p != NUL : p < end)
3661     {
3662 	l = utf8len_tab_zero[*p];
3663 	if (l == 0)
3664 	    return FALSE;	/* invalid lead byte */
3665 	if (end != NULL && p + l > end)
3666 	    return FALSE;	/* incomplete byte sequence */
3667 	++p;
3668 	while (--l > 0)
3669 	    if ((*p++ & 0xc0) != 0x80)
3670 		return FALSE;	/* invalid trail byte */
3671     }
3672     return TRUE;
3673 }
3674 #endif
3675 
3676 #if defined(FEAT_GUI) || defined(PROTO)
3677 /*
3678  * Special version of mb_tail_off() for use in ScreenLines[].
3679  */
3680     int
3681 dbcs_screen_tail_off(base, p)
3682     char_u	*base;
3683     char_u	*p;
3684 {
3685     /* It can't be the first byte if a double-byte when not using DBCS, at the
3686      * end of the string or the byte can't start a double-byte.
3687      * For euc-jp an 0x8e byte always means we have a lead byte in the current
3688      * cell. */
3689     if (*p == NUL || p[1] == NUL
3690 	    || (enc_dbcs == DBCS_JPNU && *p == 0x8e)
3691 	    || MB_BYTE2LEN(*p) == 1)
3692 	return 0;
3693 
3694     /* Return 1 when on the lead byte, 0 when on the tail byte. */
3695     return 1 - dbcs_screen_head_off(base, p);
3696 }
3697 #endif
3698 
3699 /*
3700  * If the cursor moves on an trail byte, set the cursor on the lead byte.
3701  * Thus it moves left if necessary.
3702  * Return TRUE when the cursor was adjusted.
3703  */
3704     void
3705 mb_adjust_cursor()
3706 {
3707     mb_adjustpos(curbuf, &curwin->w_cursor);
3708 }
3709 
3710 /*
3711  * Adjust position "*lp" to point to the first byte of a multi-byte character.
3712  * If it points to a tail byte it's moved backwards to the head byte.
3713  */
3714     void
3715 mb_adjustpos(buf, lp)
3716     buf_T	*buf;
3717     pos_T	*lp;
3718 {
3719     char_u	*p;
3720 
3721     if (lp->col > 0
3722 #ifdef FEAT_VIRTUALEDIT
3723 	    || lp->coladd > 1
3724 #endif
3725 	    )
3726     {
3727 	p = ml_get_buf(buf, lp->lnum, FALSE);
3728 	lp->col -= (*mb_head_off)(p, p + lp->col);
3729 #ifdef FEAT_VIRTUALEDIT
3730 	/* Reset "coladd" when the cursor would be on the right half of a
3731 	 * double-wide character. */
3732 	if (lp->coladd == 1
3733 		&& p[lp->col] != TAB
3734 		&& vim_isprintc((*mb_ptr2char)(p + lp->col))
3735 		&& ptr2cells(p + lp->col) > 1)
3736 	    lp->coladd = 0;
3737 #endif
3738     }
3739 }
3740 
3741 /*
3742  * Return a pointer to the character before "*p", if there is one.
3743  */
3744     char_u *
3745 mb_prevptr(line, p)
3746     char_u *line;	/* start of the string */
3747     char_u *p;
3748 {
3749     if (p > line)
3750 	mb_ptr_back(line, p);
3751     return p;
3752 }
3753 
3754 /*
3755  * Return the character length of "str".  Each multi-byte character (with
3756  * following composing characters) counts as one.
3757  */
3758     int
3759 mb_charlen(str)
3760     char_u	*str;
3761 {
3762     char_u	*p = str;
3763     int		count;
3764 
3765     if (p == NULL)
3766 	return 0;
3767 
3768     for (count = 0; *p != NUL; count++)
3769 	p += (*mb_ptr2len)(p);
3770 
3771     return count;
3772 }
3773 
3774 #if defined(FEAT_SPELL) || defined(PROTO)
3775 /*
3776  * Like mb_charlen() but for a string with specified length.
3777  */
3778     int
3779 mb_charlen_len(str, len)
3780     char_u	*str;
3781     int		len;
3782 {
3783     char_u	*p = str;
3784     int		count;
3785 
3786     for (count = 0; *p != NUL && p < str + len; count++)
3787 	p += (*mb_ptr2len)(p);
3788 
3789     return count;
3790 }
3791 #endif
3792 
3793 /*
3794  * Try to un-escape a multi-byte character.
3795  * Used for the "to" and "from" part of a mapping.
3796  * Return the un-escaped string if it is a multi-byte character, and advance
3797  * "pp" to just after the bytes that formed it.
3798  * Return NULL if no multi-byte char was found.
3799  */
3800     char_u *
3801 mb_unescape(pp)
3802     char_u **pp;
3803 {
3804     static char_u	buf[6];
3805     int			n;
3806     int			m = 0;
3807     char_u		*str = *pp;
3808 
3809     /* Must translate K_SPECIAL KS_SPECIAL KE_FILLER to K_SPECIAL and CSI
3810      * KS_EXTRA KE_CSI to CSI.
3811      * Maximum length of a utf-8 character is 4 bytes. */
3812     for (n = 0; str[n] != NUL && m < 4; ++n)
3813     {
3814 	if (str[n] == K_SPECIAL
3815 		&& str[n + 1] == KS_SPECIAL
3816 		&& str[n + 2] == KE_FILLER)
3817 	{
3818 	    buf[m++] = K_SPECIAL;
3819 	    n += 2;
3820 	}
3821 	else if ((str[n] == K_SPECIAL
3822 # ifdef FEAT_GUI
3823 		    || str[n] == CSI
3824 # endif
3825 		 )
3826 		&& str[n + 1] == KS_EXTRA
3827 		&& str[n + 2] == (int)KE_CSI)
3828 	{
3829 	    buf[m++] = CSI;
3830 	    n += 2;
3831 	}
3832 	else if (str[n] == K_SPECIAL
3833 # ifdef FEAT_GUI
3834 		|| str[n] == CSI
3835 # endif
3836 		)
3837 	    break;		/* a special key can't be a multibyte char */
3838 	else
3839 	    buf[m++] = str[n];
3840 	buf[m] = NUL;
3841 
3842 	/* Return a multi-byte character if it's found.  An illegal sequence
3843 	 * will result in a 1 here. */
3844 	if ((*mb_ptr2len)(buf) > 1)
3845 	{
3846 	    *pp = str + n + 1;
3847 	    return buf;
3848 	}
3849 
3850 	/* Bail out quickly for ASCII. */
3851 	if (buf[0] < 128)
3852 	    break;
3853     }
3854     return NULL;
3855 }
3856 
3857 /*
3858  * Return TRUE if the character at "row"/"col" on the screen is the left side
3859  * of a double-width character.
3860  * Caller must make sure "row" and "col" are not invalid!
3861  */
3862     int
3863 mb_lefthalve(row, col)
3864     int	    row;
3865     int	    col;
3866 {
3867 #ifdef FEAT_HANGULIN
3868     if (composing_hangul)
3869 	return TRUE;
3870 #endif
3871     return (*mb_off2cells)(LineOffset[row] + col,
3872 					LineOffset[row] + screen_Columns) > 1;
3873 }
3874 
3875 /*
3876  * Correct a position on the screen, if it's the right half of a double-wide
3877  * char move it to the left half.  Returns the corrected column.
3878  */
3879     int
3880 mb_fix_col(col, row)
3881     int		col;
3882     int		row;
3883 {
3884     col = check_col(col);
3885     row = check_row(row);
3886     if (has_mbyte && ScreenLines != NULL && col > 0
3887 	    && ((enc_dbcs
3888 		    && ScreenLines[LineOffset[row] + col] != NUL
3889 		    && dbcs_screen_head_off(ScreenLines + LineOffset[row],
3890 					 ScreenLines + LineOffset[row] + col))
3891 		|| (enc_utf8 && ScreenLines[LineOffset[row] + col] == 0)))
3892 	return col - 1;
3893     return col;
3894 }
3895 #endif
3896 
3897 #if defined(FEAT_MBYTE) || defined(FEAT_POSTSCRIPT) || defined(PROTO)
3898 static int enc_alias_search __ARGS((char_u *name));
3899 
3900 /*
3901  * Skip the Vim specific head of a 'encoding' name.
3902  */
3903     char_u *
3904 enc_skip(p)
3905     char_u	*p;
3906 {
3907     if (STRNCMP(p, "2byte-", 6) == 0)
3908 	return p + 6;
3909     if (STRNCMP(p, "8bit-", 5) == 0)
3910 	return p + 5;
3911     return p;
3912 }
3913 
3914 /*
3915  * Find the canonical name for encoding "enc".
3916  * When the name isn't recognized, returns "enc" itself, but with all lower
3917  * case characters and '_' replaced with '-'.
3918  * Returns an allocated string.  NULL for out-of-memory.
3919  */
3920     char_u *
3921 enc_canonize(enc)
3922     char_u	*enc;
3923 {
3924     char_u	*r;
3925     char_u	*p, *s;
3926     int		i;
3927 
3928 # ifdef FEAT_MBYTE
3929     if (STRCMP(enc, "default") == 0)
3930     {
3931 	/* Use the default encoding as it's found by set_init_1(). */
3932 	r = get_encoding_default();
3933 	if (r == NULL)
3934 	    r = (char_u *)"latin1";
3935 	return vim_strsave(r);
3936     }
3937 # endif
3938 
3939     /* copy "enc" to allocated memory, with room for two '-' */
3940     r = alloc((unsigned)(STRLEN(enc) + 3));
3941     if (r != NULL)
3942     {
3943 	/* Make it all lower case and replace '_' with '-'. */
3944 	p = r;
3945 	for (s = enc; *s != NUL; ++s)
3946 	{
3947 	    if (*s == '_')
3948 		*p++ = '-';
3949 	    else
3950 		*p++ = TOLOWER_ASC(*s);
3951 	}
3952 	*p = NUL;
3953 
3954 	/* Skip "2byte-" and "8bit-". */
3955 	p = enc_skip(r);
3956 
3957 	/* Change "microsoft-cp" to "cp".  Used in some spell files. */
3958 	if (STRNCMP(p, "microsoft-cp", 12) == 0)
3959 	    STRMOVE(p, p + 10);
3960 
3961 	/* "iso8859" -> "iso-8859" */
3962 	if (STRNCMP(p, "iso8859", 7) == 0)
3963 	{
3964 	    STRMOVE(p + 4, p + 3);
3965 	    p[3] = '-';
3966 	}
3967 
3968 	/* "iso-8859n" -> "iso-8859-n" */
3969 	if (STRNCMP(p, "iso-8859", 8) == 0 && p[8] != '-')
3970 	{
3971 	    STRMOVE(p + 9, p + 8);
3972 	    p[8] = '-';
3973 	}
3974 
3975 	/* "latin-N" -> "latinN" */
3976 	if (STRNCMP(p, "latin-", 6) == 0)
3977 	    STRMOVE(p + 5, p + 6);
3978 
3979 	if (enc_canon_search(p) >= 0)
3980 	{
3981 	    /* canonical name can be used unmodified */
3982 	    if (p != r)
3983 		STRMOVE(r, p);
3984 	}
3985 	else if ((i = enc_alias_search(p)) >= 0)
3986 	{
3987 	    /* alias recognized, get canonical name */
3988 	    vim_free(r);
3989 	    r = vim_strsave((char_u *)enc_canon_table[i].name);
3990 	}
3991     }
3992     return r;
3993 }
3994 
3995 /*
3996  * Search for an encoding alias of "name".
3997  * Returns -1 when not found.
3998  */
3999     static int
4000 enc_alias_search(name)
4001     char_u	*name;
4002 {
4003     int		i;
4004 
4005     for (i = 0; enc_alias_table[i].name != NULL; ++i)
4006 	if (STRCMP(name, enc_alias_table[i].name) == 0)
4007 	    return enc_alias_table[i].canon;
4008     return -1;
4009 }
4010 #endif
4011 
4012 #if defined(FEAT_MBYTE) || defined(PROTO)
4013 
4014 #ifdef HAVE_LANGINFO_H
4015 # include <langinfo.h>
4016 #endif
4017 
4018 /*
4019  * Get the canonicalized encoding of the current locale.
4020  * Returns an allocated string when successful, NULL when not.
4021  */
4022     char_u *
4023 enc_locale()
4024 {
4025 #ifndef WIN3264
4026     char	*s;
4027     char	*p;
4028     int		i;
4029 #endif
4030     char	buf[50];
4031 #ifdef WIN3264
4032     long	acp = GetACP();
4033 
4034     if (acp == 1200)
4035 	STRCPY(buf, "ucs-2le");
4036     else if (acp == 1252)	    /* cp1252 is used as latin1 */
4037 	STRCPY(buf, "latin1");
4038     else
4039 	sprintf(buf, "cp%ld", acp);
4040 #else
4041 # ifdef HAVE_NL_LANGINFO_CODESET
4042     if ((s = nl_langinfo(CODESET)) == NULL || *s == NUL)
4043 # endif
4044 #  if defined(HAVE_LOCALE_H) || defined(X_LOCALE)
4045 	if ((s = setlocale(LC_CTYPE, NULL)) == NULL || *s == NUL)
4046 #  endif
4047 	    if ((s = getenv("LC_ALL")) == NULL || *s == NUL)
4048 		if ((s = getenv("LC_CTYPE")) == NULL || *s == NUL)
4049 		    s = getenv("LANG");
4050 
4051     if (s == NULL || *s == NUL)
4052 	return FAIL;
4053 
4054     /* The most generic locale format is:
4055      * language[_territory][.codeset][@modifier][+special][,[sponsor][_revision]]
4056      * If there is a '.' remove the part before it.
4057      * if there is something after the codeset, remove it.
4058      * Make the name lowercase and replace '_' with '-'.
4059      * Exception: "ja_JP.EUC" == "euc-jp", "zh_CN.EUC" = "euc-cn",
4060      * "ko_KR.EUC" == "euc-kr"
4061      */
4062     if ((p = (char *)vim_strchr((char_u *)s, '.')) != NULL)
4063     {
4064 	if (p > s + 2 && STRNICMP(p + 1, "EUC", 3) == 0
4065 			&& !isalnum((int)p[4]) && p[4] != '-' && p[-3] == '_')
4066 	{
4067 	    /* copy "XY.EUC" to "euc-XY" to buf[10] */
4068 	    STRCPY(buf + 10, "euc-");
4069 	    buf[14] = p[-2];
4070 	    buf[15] = p[-1];
4071 	    buf[16] = 0;
4072 	    s = buf + 10;
4073 	}
4074 	else
4075 	    s = p + 1;
4076     }
4077     for (i = 0; s[i] != NUL && i < (int)sizeof(buf) - 1; ++i)
4078     {
4079 	if (s[i] == '_' || s[i] == '-')
4080 	    buf[i] = '-';
4081 	else if (isalnum((int)s[i]))
4082 	    buf[i] = TOLOWER_ASC(s[i]);
4083 	else
4084 	    break;
4085     }
4086     buf[i] = NUL;
4087 #endif
4088 
4089     return enc_canonize((char_u *)buf);
4090 }
4091 
4092 #if defined(WIN3264) || defined(PROTO) || defined(FEAT_CYGWIN_WIN32_CLIPBOARD)
4093 /*
4094  * Convert an encoding name to an MS-Windows codepage.
4095  * Returns zero if no codepage can be figured out.
4096  */
4097     int
4098 encname2codepage(name)
4099     char_u	*name;
4100 {
4101     int		cp;
4102     char_u	*p = name;
4103     int		idx;
4104 
4105     if (STRNCMP(p, "8bit-", 5) == 0)
4106 	p += 5;
4107     else if (STRNCMP(p_enc, "2byte-", 6) == 0)
4108 	p += 6;
4109 
4110     if (p[0] == 'c' && p[1] == 'p')
4111 	cp = atoi(p + 2);
4112     else if ((idx = enc_canon_search(p)) >= 0)
4113 	cp = enc_canon_table[idx].codepage;
4114     else
4115 	return 0;
4116     if (IsValidCodePage(cp))
4117 	return cp;
4118     return 0;
4119 }
4120 #endif
4121 
4122 # if defined(USE_ICONV) || defined(PROTO)
4123 
4124 static char_u *iconv_string __ARGS((vimconv_T *vcp, char_u *str, int slen, int *unconvlenp, int *resultlenp));
4125 
4126 /*
4127  * Call iconv_open() with a check if iconv() works properly (there are broken
4128  * versions).
4129  * Returns (void *)-1 if failed.
4130  * (should return iconv_t, but that causes problems with prototypes).
4131  */
4132     void *
4133 my_iconv_open(to, from)
4134     char_u	*to;
4135     char_u	*from;
4136 {
4137     iconv_t	fd;
4138 #define ICONV_TESTLEN 400
4139     char_u	tobuf[ICONV_TESTLEN];
4140     char	*p;
4141     size_t	tolen;
4142     static int	iconv_ok = -1;
4143 
4144     if (iconv_ok == FALSE)
4145 	return (void *)-1;	/* detected a broken iconv() previously */
4146 
4147 #ifdef DYNAMIC_ICONV
4148     /* Check if the iconv.dll can be found. */
4149     if (!iconv_enabled(TRUE))
4150 	return (void *)-1;
4151 #endif
4152 
4153     fd = iconv_open((char *)enc_skip(to), (char *)enc_skip(from));
4154 
4155     if (fd != (iconv_t)-1 && iconv_ok == -1)
4156     {
4157 	/*
4158 	 * Do a dummy iconv() call to check if it actually works.  There is a
4159 	 * version of iconv() on Linux that is broken.  We can't ignore it,
4160 	 * because it's wide-spread.  The symptoms are that after outputting
4161 	 * the initial shift state the "to" pointer is NULL and conversion
4162 	 * stops for no apparent reason after about 8160 characters.
4163 	 */
4164 	p = (char *)tobuf;
4165 	tolen = ICONV_TESTLEN;
4166 	(void)iconv(fd, NULL, NULL, &p, &tolen);
4167 	if (p == NULL)
4168 	{
4169 	    iconv_ok = FALSE;
4170 	    iconv_close(fd);
4171 	    fd = (iconv_t)-1;
4172 	}
4173 	else
4174 	    iconv_ok = TRUE;
4175     }
4176 
4177     return (void *)fd;
4178 }
4179 
4180 /*
4181  * Convert the string "str[slen]" with iconv().
4182  * If "unconvlenp" is not NULL handle the string ending in an incomplete
4183  * sequence and set "*unconvlenp" to the length of it.
4184  * Returns the converted string in allocated memory.  NULL for an error.
4185  * If resultlenp is not NULL, sets it to the result length in bytes.
4186  */
4187     static char_u *
4188 iconv_string(vcp, str, slen, unconvlenp, resultlenp)
4189     vimconv_T	*vcp;
4190     char_u	*str;
4191     int		slen;
4192     int		*unconvlenp;
4193     int		*resultlenp;
4194 {
4195     const char	*from;
4196     size_t	fromlen;
4197     char	*to;
4198     size_t	tolen;
4199     size_t	len = 0;
4200     size_t	done = 0;
4201     char_u	*result = NULL;
4202     char_u	*p;
4203     int		l;
4204 
4205     from = (char *)str;
4206     fromlen = slen;
4207     for (;;)
4208     {
4209 	if (len == 0 || ICONV_ERRNO == ICONV_E2BIG)
4210 	{
4211 	    /* Allocate enough room for most conversions.  When re-allocating
4212 	     * increase the buffer size. */
4213 	    len = len + fromlen * 2 + 40;
4214 	    p = alloc((unsigned)len);
4215 	    if (p != NULL && done > 0)
4216 		mch_memmove(p, result, done);
4217 	    vim_free(result);
4218 	    result = p;
4219 	    if (result == NULL)	/* out of memory */
4220 		break;
4221 	}
4222 
4223 	to = (char *)result + done;
4224 	tolen = len - done - 2;
4225 	/* Avoid a warning for systems with a wrong iconv() prototype by
4226 	 * casting the second argument to void *. */
4227 	if (iconv(vcp->vc_fd, (void *)&from, &fromlen, &to, &tolen)
4228 								!= (size_t)-1)
4229 	{
4230 	    /* Finished, append a NUL. */
4231 	    *to = NUL;
4232 	    break;
4233 	}
4234 
4235 	/* Check both ICONV_EINVAL and EINVAL, because the dynamically loaded
4236 	 * iconv library may use one of them. */
4237 	if (!vcp->vc_fail && unconvlenp != NULL
4238 		&& (ICONV_ERRNO == ICONV_EINVAL || ICONV_ERRNO == EINVAL))
4239 	{
4240 	    /* Handle an incomplete sequence at the end. */
4241 	    *to = NUL;
4242 	    *unconvlenp = (int)fromlen;
4243 	    break;
4244 	}
4245 
4246 	/* Check both ICONV_EILSEQ and EILSEQ, because the dynamically loaded
4247 	 * iconv library may use one of them. */
4248 	else if (!vcp->vc_fail
4249 		&& (ICONV_ERRNO == ICONV_EILSEQ || ICONV_ERRNO == EILSEQ
4250 		    || ICONV_ERRNO == ICONV_EINVAL || ICONV_ERRNO == EINVAL))
4251 	{
4252 	    /* Can't convert: insert a '?' and skip a character.  This assumes
4253 	     * conversion from 'encoding' to something else.  In other
4254 	     * situations we don't know what to skip anyway. */
4255 	    *to++ = '?';
4256 	    if ((*mb_ptr2cells)((char_u *)from) > 1)
4257 		*to++ = '?';
4258 	    if (enc_utf8)
4259 		l = utfc_ptr2len_len((char_u *)from, (int)fromlen);
4260 	    else
4261 	    {
4262 		l = (*mb_ptr2len)((char_u *)from);
4263 		if (l > (int)fromlen)
4264 		    l = (int)fromlen;
4265 	    }
4266 	    from += l;
4267 	    fromlen -= l;
4268 	}
4269 	else if (ICONV_ERRNO != ICONV_E2BIG)
4270 	{
4271 	    /* conversion failed */
4272 	    vim_free(result);
4273 	    result = NULL;
4274 	    break;
4275 	}
4276 	/* Not enough room or skipping illegal sequence. */
4277 	done = to - (char *)result;
4278     }
4279 
4280     if (resultlenp != NULL && result != NULL)
4281 	*resultlenp = (int)(to - (char *)result);
4282     return result;
4283 }
4284 
4285 #  if defined(DYNAMIC_ICONV) || defined(PROTO)
4286 /*
4287  * Dynamically load the "iconv.dll" on Win32.
4288  */
4289 
4290 #ifndef DYNAMIC_ICONV	    /* just generating prototypes */
4291 # define HINSTANCE int
4292 #endif
4293 static HINSTANCE hIconvDLL = 0;
4294 static HINSTANCE hMsvcrtDLL = 0;
4295 
4296 #  ifndef DYNAMIC_ICONV_DLL
4297 #   define DYNAMIC_ICONV_DLL "iconv.dll"
4298 #   define DYNAMIC_ICONV_DLL_ALT "libiconv.dll"
4299 #  endif
4300 #  ifndef DYNAMIC_MSVCRT_DLL
4301 #   define DYNAMIC_MSVCRT_DLL "msvcrt.dll"
4302 #  endif
4303 
4304 /*
4305  * Get the address of 'funcname' which is imported by 'hInst' DLL.
4306  */
4307     static void *
4308 get_iconv_import_func(HINSTANCE hInst, const char *funcname)
4309 {
4310     PBYTE			pImage = (PBYTE)hInst;
4311     PIMAGE_DOS_HEADER		pDOS = (PIMAGE_DOS_HEADER)hInst;
4312     PIMAGE_NT_HEADERS		pPE;
4313     PIMAGE_IMPORT_DESCRIPTOR	pImpDesc;
4314     PIMAGE_THUNK_DATA		pIAT;	    /* Import Address Table */
4315     PIMAGE_THUNK_DATA		pINT;	    /* Import Name Table */
4316     PIMAGE_IMPORT_BY_NAME	pImpName;
4317 
4318     if (pDOS->e_magic != IMAGE_DOS_SIGNATURE)
4319 	return NULL;
4320     pPE = (PIMAGE_NT_HEADERS)(pImage + pDOS->e_lfanew);
4321     if (pPE->Signature != IMAGE_NT_SIGNATURE)
4322 	return NULL;
4323     pImpDesc = (PIMAGE_IMPORT_DESCRIPTOR)(pImage
4324 	    + pPE->OptionalHeader.DataDirectory[IMAGE_DIRECTORY_ENTRY_IMPORT]
4325 							    .VirtualAddress);
4326     for (; pImpDesc->FirstThunk; ++pImpDesc)
4327     {
4328 	if (!pImpDesc->OriginalFirstThunk)
4329 	    continue;
4330 	pIAT = (PIMAGE_THUNK_DATA)(pImage + pImpDesc->FirstThunk);
4331 	pINT = (PIMAGE_THUNK_DATA)(pImage + pImpDesc->OriginalFirstThunk);
4332 	for (; pIAT->u1.Function; ++pIAT, ++pINT)
4333 	{
4334 	    if (IMAGE_SNAP_BY_ORDINAL(pINT->u1.Ordinal))
4335 		continue;
4336 	    pImpName = (PIMAGE_IMPORT_BY_NAME)(pImage
4337 					+ (UINT_PTR)(pINT->u1.AddressOfData));
4338 	    if (strcmp(pImpName->Name, funcname) == 0)
4339 		return (void *)pIAT->u1.Function;
4340 	}
4341     }
4342     return NULL;
4343 }
4344 
4345 /*
4346  * Try opening the iconv.dll and return TRUE if iconv() can be used.
4347  */
4348     int
4349 iconv_enabled(verbose)
4350     int		verbose;
4351 {
4352     if (hIconvDLL != 0 && hMsvcrtDLL != 0)
4353 	return TRUE;
4354     hIconvDLL = vimLoadLib(DYNAMIC_ICONV_DLL);
4355     if (hIconvDLL == 0)		/* sometimes it's called libiconv.dll */
4356 	hIconvDLL = vimLoadLib(DYNAMIC_ICONV_DLL_ALT);
4357     if (hIconvDLL != 0)
4358 	hMsvcrtDLL = vimLoadLib(DYNAMIC_MSVCRT_DLL);
4359     if (hIconvDLL == 0 || hMsvcrtDLL == 0)
4360     {
4361 	/* Only give the message when 'verbose' is set, otherwise it might be
4362 	 * done whenever a conversion is attempted. */
4363 	if (verbose && p_verbose > 0)
4364 	{
4365 	    verbose_enter();
4366 	    EMSG2(_(e_loadlib),
4367 		    hIconvDLL == 0 ? DYNAMIC_ICONV_DLL : DYNAMIC_MSVCRT_DLL);
4368 	    verbose_leave();
4369 	}
4370 	iconv_end();
4371 	return FALSE;
4372     }
4373 
4374     iconv	= (void *)GetProcAddress(hIconvDLL, "libiconv");
4375     iconv_open	= (void *)GetProcAddress(hIconvDLL, "libiconv_open");
4376     iconv_close	= (void *)GetProcAddress(hIconvDLL, "libiconv_close");
4377     iconvctl	= (void *)GetProcAddress(hIconvDLL, "libiconvctl");
4378     iconv_errno	= get_iconv_import_func(hIconvDLL, "_errno");
4379     if (iconv_errno == NULL)
4380 	iconv_errno = (void *)GetProcAddress(hMsvcrtDLL, "_errno");
4381     if (iconv == NULL || iconv_open == NULL || iconv_close == NULL
4382 	    || iconvctl == NULL || iconv_errno == NULL)
4383     {
4384 	iconv_end();
4385 	if (verbose && p_verbose > 0)
4386 	{
4387 	    verbose_enter();
4388 	    EMSG2(_(e_loadfunc), "for libiconv");
4389 	    verbose_leave();
4390 	}
4391 	return FALSE;
4392     }
4393     return TRUE;
4394 }
4395 
4396     void
4397 iconv_end()
4398 {
4399     /* Don't use iconv() when inputting or outputting characters. */
4400     if (input_conv.vc_type == CONV_ICONV)
4401 	convert_setup(&input_conv, NULL, NULL);
4402     if (output_conv.vc_type == CONV_ICONV)
4403 	convert_setup(&output_conv, NULL, NULL);
4404 
4405     if (hIconvDLL != 0)
4406 	FreeLibrary(hIconvDLL);
4407     if (hMsvcrtDLL != 0)
4408 	FreeLibrary(hMsvcrtDLL);
4409     hIconvDLL = 0;
4410     hMsvcrtDLL = 0;
4411 }
4412 #  endif /* DYNAMIC_ICONV */
4413 # endif /* USE_ICONV */
4414 
4415 #endif /* FEAT_MBYTE */
4416 
4417 #if defined(FEAT_XIM) || defined(PROTO)
4418 
4419 # if defined(FEAT_GUI_GTK) || defined(PROTO)
4420 static int xim_has_preediting INIT(= FALSE);  /* IM current status */
4421 
4422 /*
4423  * Set preedit_start_col to the current cursor position.
4424  */
4425     static void
4426 init_preedit_start_col(void)
4427 {
4428     if (State & CMDLINE)
4429 	preedit_start_col = cmdline_getvcol_cursor();
4430     else if (curwin != NULL)
4431 	getvcol(curwin, &curwin->w_cursor, &preedit_start_col, NULL, NULL);
4432     /* Prevent that preediting marks the buffer as changed. */
4433     xim_changed_while_preediting = curbuf->b_changed;
4434 }
4435 
4436 static int im_is_active	       = FALSE;	/* IM is enabled for current mode    */
4437 static int preedit_is_active   = FALSE;
4438 static int im_preedit_cursor   = 0;	/* cursor offset in characters       */
4439 static int im_preedit_trailing = 0;	/* number of characters after cursor */
4440 
4441 static unsigned long im_commit_handler_id  = 0;
4442 static unsigned int  im_activatekey_keyval = GDK_VoidSymbol;
4443 static unsigned int  im_activatekey_state  = 0;
4444 
4445     void
4446 im_set_active(int active)
4447 {
4448     int was_active;
4449 
4450     was_active = !!im_is_active;
4451     im_is_active = (active && !p_imdisable);
4452 
4453     if (im_is_active != was_active)
4454 	xim_reset();
4455 }
4456 
4457     void
4458 xim_set_focus(int focus)
4459 {
4460     if (xic != NULL)
4461     {
4462 	if (focus)
4463 	    gtk_im_context_focus_in(xic);
4464 	else
4465 	    gtk_im_context_focus_out(xic);
4466     }
4467 }
4468 
4469     void
4470 im_set_position(int row, int col)
4471 {
4472     if (xic != NULL)
4473     {
4474 	GdkRectangle area;
4475 
4476 	area.x = FILL_X(col);
4477 	area.y = FILL_Y(row);
4478 	area.width  = gui.char_width * (mb_lefthalve(row, col) ? 2 : 1);
4479 	area.height = gui.char_height;
4480 
4481 	gtk_im_context_set_cursor_location(xic, &area);
4482     }
4483 }
4484 
4485 #  if 0 || defined(PROTO) /* apparently only used in gui_x11.c */
4486     void
4487 xim_set_preedit(void)
4488 {
4489     im_set_position(gui.row, gui.col);
4490 }
4491 #  endif
4492 
4493     static void
4494 im_add_to_input(char_u *str, int len)
4495 {
4496     /* Convert from 'termencoding' (always "utf-8") to 'encoding' */
4497     if (input_conv.vc_type != CONV_NONE)
4498     {
4499 	str = string_convert(&input_conv, str, &len);
4500 	g_return_if_fail(str != NULL);
4501     }
4502 
4503     add_to_input_buf_csi(str, len);
4504 
4505     if (input_conv.vc_type != CONV_NONE)
4506 	vim_free(str);
4507 
4508     if (p_mh) /* blank out the pointer if necessary */
4509 	gui_mch_mousehide(TRUE);
4510 }
4511 
4512     static void
4513 im_delete_preedit(void)
4514 {
4515     char_u bskey[]  = {CSI, 'k', 'b'};
4516     char_u delkey[] = {CSI, 'k', 'D'};
4517 
4518     if (State & NORMAL)
4519     {
4520 	im_preedit_cursor = 0;
4521 	return;
4522     }
4523     for (; im_preedit_cursor > 0; --im_preedit_cursor)
4524 	add_to_input_buf(bskey, (int)sizeof(bskey));
4525 
4526     for (; im_preedit_trailing > 0; --im_preedit_trailing)
4527 	add_to_input_buf(delkey, (int)sizeof(delkey));
4528 }
4529 
4530 /*
4531  * Move the cursor left by "num_move_back" characters.
4532  * Note that ins_left() checks im_is_preediting() to avoid breaking undo for
4533  * these K_LEFT keys.
4534  */
4535     static void
4536 im_correct_cursor(int num_move_back)
4537 {
4538     char_u backkey[] = {CSI, 'k', 'l'};
4539 
4540     if (State & NORMAL)
4541 	return;
4542 #  ifdef FEAT_RIGHTLEFT
4543     if ((State & CMDLINE) == 0 && curwin != NULL && curwin->w_p_rl)
4544 	backkey[2] = 'r';
4545 #  endif
4546     for (; num_move_back > 0; --num_move_back)
4547 	add_to_input_buf(backkey, (int)sizeof(backkey));
4548 }
4549 
4550 static int xim_expected_char = NUL;
4551 static int xim_ignored_char = FALSE;
4552 
4553 /*
4554  * Update the mode and cursor while in an IM callback.
4555  */
4556     static void
4557 im_show_info(void)
4558 {
4559     int	    old_vgetc_busy;
4560 
4561     old_vgetc_busy = vgetc_busy;
4562     vgetc_busy = TRUE;
4563     showmode();
4564     vgetc_busy = old_vgetc_busy;
4565     if ((State & NORMAL) || (State & INSERT))
4566 	setcursor();
4567     out_flush();
4568 }
4569 
4570 /*
4571  * Callback invoked when the user finished preediting.
4572  * Put the final string into the input buffer.
4573  */
4574     static void
4575 im_commit_cb(GtkIMContext *context UNUSED,
4576 	     const gchar *str,
4577 	     gpointer data UNUSED)
4578 {
4579     int		slen = (int)STRLEN(str);
4580     int		add_to_input = TRUE;
4581     int		clen;
4582     int		len = slen;
4583     int		commit_with_preedit = TRUE;
4584     char_u	*im_str;
4585 
4586 #ifdef XIM_DEBUG
4587     xim_log("im_commit_cb(): %s\n", str);
4588 #endif
4589 
4590     /* The imhangul module doesn't reset the preedit string before
4591      * committing.  Call im_delete_preedit() to work around that. */
4592     im_delete_preedit();
4593 
4594     /* Indicate that preediting has finished. */
4595     if (preedit_start_col == MAXCOL)
4596     {
4597 	init_preedit_start_col();
4598 	commit_with_preedit = FALSE;
4599     }
4600 
4601     /* The thing which setting "preedit_start_col" to MAXCOL means that
4602      * "preedit_start_col" will be set forcedly when calling
4603      * preedit_changed_cb() next time.
4604      * "preedit_start_col" should not reset with MAXCOL on this part. Vim
4605      * is simulating the preediting by using add_to_input_str(). when
4606      * preedit begin immediately before committed, the typebuf is not
4607      * flushed to screen, then it can't get correct "preedit_start_col".
4608      * Thus, it should calculate the cells by adding cells of the committed
4609      * string. */
4610     if (input_conv.vc_type != CONV_NONE)
4611     {
4612 	im_str = string_convert(&input_conv, (char_u *)str, &len);
4613 	g_return_if_fail(im_str != NULL);
4614     }
4615     else
4616 	im_str = (char_u *)str;
4617 
4618     clen = mb_string2cells(im_str, len);
4619 
4620     if (input_conv.vc_type != CONV_NONE)
4621 	vim_free(im_str);
4622     preedit_start_col += clen;
4623 
4624     /* Is this a single character that matches a keypad key that's just
4625      * been pressed?  If so, we don't want it to be entered as such - let
4626      * us carry on processing the raw keycode so that it may be used in
4627      * mappings as <kSomething>. */
4628     if (xim_expected_char != NUL)
4629     {
4630 	/* We're currently processing a keypad or other special key */
4631 	if (slen == 1 && str[0] == xim_expected_char)
4632 	{
4633 	    /* It's a match - don't do it here */
4634 	    xim_ignored_char = TRUE;
4635 	    add_to_input = FALSE;
4636 	}
4637 	else
4638 	{
4639 	    /* Not a match */
4640 	    xim_ignored_char = FALSE;
4641 	}
4642     }
4643 
4644     if (add_to_input)
4645 	im_add_to_input((char_u *)str, slen);
4646 
4647     /* Inserting chars while "im_is_active" is set does not cause a change of
4648      * buffer.  When the chars are committed the buffer must be marked as
4649      * changed. */
4650     if (!commit_with_preedit)
4651 	preedit_start_col = MAXCOL;
4652 
4653     /* This flag is used in changed() at next call. */
4654     xim_changed_while_preediting = TRUE;
4655 
4656     if (gtk_main_level() > 0)
4657 	gtk_main_quit();
4658 }
4659 
4660 /*
4661  * Callback invoked after start to the preedit.
4662  */
4663     static void
4664 im_preedit_start_cb(GtkIMContext *context UNUSED, gpointer data UNUSED)
4665 {
4666 #ifdef XIM_DEBUG
4667     xim_log("im_preedit_start_cb()\n");
4668 #endif
4669 
4670     im_is_active = TRUE;
4671     preedit_is_active = TRUE;
4672     gui_update_cursor(TRUE, FALSE);
4673     im_show_info();
4674 }
4675 
4676 /*
4677  * Callback invoked after end to the preedit.
4678  */
4679     static void
4680 im_preedit_end_cb(GtkIMContext *context UNUSED, gpointer data UNUSED)
4681 {
4682 #ifdef XIM_DEBUG
4683     xim_log("im_preedit_end_cb()\n");
4684 #endif
4685     im_delete_preedit();
4686 
4687     /* Indicate that preediting has finished */
4688     preedit_start_col = MAXCOL;
4689     xim_has_preediting = FALSE;
4690 
4691 #if 0
4692     /* Removal of this line suggested by Takuhiro Nishioka.  Fixes that IM was
4693      * switched off unintentionally.  We now use preedit_is_active (added by
4694      * SungHyun Nam). */
4695     im_is_active = FALSE;
4696 #endif
4697     preedit_is_active = FALSE;
4698     gui_update_cursor(TRUE, FALSE);
4699     im_show_info();
4700 }
4701 
4702 /*
4703  * Callback invoked after changes to the preedit string.  If the preedit
4704  * string was empty before, remember the preedit start column so we know
4705  * where to apply feedback attributes.  Delete the previous preedit string
4706  * if there was one, save the new preedit cursor offset, and put the new
4707  * string into the input buffer.
4708  *
4709  * TODO: The pragmatic "put into input buffer" approach used here has
4710  *       several fundamental problems:
4711  *
4712  * - The characters in the preedit string are subject to remapping.
4713  *   That's broken, only the finally committed string should be remapped.
4714  *
4715  * - There is a race condition involved:  The retrieved value for the
4716  *   current cursor position will be wrong if any unprocessed characters
4717  *   are still queued in the input buffer.
4718  *
4719  * - Due to the lack of synchronization between the file buffer in memory
4720  *   and any typed characters, it's practically impossible to implement the
4721  *   "retrieve_surrounding" and "delete_surrounding" signals reliably.  IM
4722  *   modules for languages such as Thai are likely to rely on this feature
4723  *   for proper operation.
4724  *
4725  * Conclusions:  I think support for preediting needs to be moved to the
4726  * core parts of Vim.  Ideally, until it has been committed, the preediting
4727  * string should only be displayed and not affect the buffer content at all.
4728  * The question how to deal with the synchronization issue still remains.
4729  * Circumventing the input buffer is probably not desirable.  Anyway, I think
4730  * implementing "retrieve_surrounding" is the only hard problem.
4731  *
4732  * One way to solve all of this in a clean manner would be to queue all key
4733  * press/release events "as is" in the input buffer, and apply the IM filtering
4734  * at the receiving end of the queue.  This, however, would have a rather large
4735  * impact on the code base.  If there is an easy way to force processing of all
4736  * remaining input from within the "retrieve_surrounding" signal handler, this
4737  * might not be necessary.  Gotta ask on vim-dev for opinions.
4738  */
4739     static void
4740 im_preedit_changed_cb(GtkIMContext *context, gpointer data UNUSED)
4741 {
4742     char    *preedit_string = NULL;
4743     int	    cursor_index    = 0;
4744     int	    num_move_back   = 0;
4745     char_u  *str;
4746     char_u  *p;
4747     int	    i;
4748 
4749     gtk_im_context_get_preedit_string(context,
4750 				      &preedit_string, NULL,
4751 				      &cursor_index);
4752 
4753 #ifdef XIM_DEBUG
4754     xim_log("im_preedit_changed_cb(): %s\n", preedit_string);
4755 #endif
4756 
4757     g_return_if_fail(preedit_string != NULL); /* just in case */
4758 
4759     /* If preedit_start_col is MAXCOL set it to the current cursor position. */
4760     if (preedit_start_col == MAXCOL && preedit_string[0] != '\0')
4761     {
4762 	xim_has_preediting = TRUE;
4763 
4764 	/* Urgh, this breaks if the input buffer isn't empty now */
4765 	init_preedit_start_col();
4766     }
4767     else if (cursor_index == 0 && preedit_string[0] == '\0')
4768     {
4769 	xim_has_preediting = FALSE;
4770 
4771 	/* If at the start position (after typing backspace)
4772 	 * preedit_start_col must be reset. */
4773 	preedit_start_col = MAXCOL;
4774     }
4775 
4776     im_delete_preedit();
4777 
4778     /*
4779      * Compute the end of the preediting area: "preedit_end_col".
4780      * According to the documentation of gtk_im_context_get_preedit_string(),
4781      * the cursor_pos output argument returns the offset in bytes.  This is
4782      * unfortunately not true -- real life shows the offset is in characters,
4783      * and the GTK+ source code agrees with me.  Will file a bug later.
4784      */
4785     if (preedit_start_col != MAXCOL)
4786 	preedit_end_col = preedit_start_col;
4787     str = (char_u *)preedit_string;
4788     for (p = str, i = 0; *p != NUL; p += utf_byte2len(*p), ++i)
4789     {
4790 	int is_composing;
4791 
4792 	is_composing = ((*p & 0x80) != 0 && utf_iscomposing(utf_ptr2char(p)));
4793 	/*
4794 	 * These offsets are used as counters when generating <BS> and <Del>
4795 	 * to delete the preedit string.  So don't count composing characters
4796 	 * unless 'delcombine' is enabled.
4797 	 */
4798 	if (!is_composing || p_deco)
4799 	{
4800 	    if (i < cursor_index)
4801 		++im_preedit_cursor;
4802 	    else
4803 		++im_preedit_trailing;
4804 	}
4805 	if (!is_composing && i >= cursor_index)
4806 	{
4807 	    /* This is essentially the same as im_preedit_trailing, except
4808 	     * composing characters are not counted even if p_deco is set. */
4809 	    ++num_move_back;
4810 	}
4811 	if (preedit_start_col != MAXCOL)
4812 	    preedit_end_col += utf_ptr2cells(p);
4813     }
4814 
4815     if (p > str)
4816     {
4817 	im_add_to_input(str, (int)(p - str));
4818 	im_correct_cursor(num_move_back);
4819     }
4820 
4821     g_free(preedit_string);
4822 
4823     if (gtk_main_level() > 0)
4824 	gtk_main_quit();
4825 }
4826 
4827 /*
4828  * Translate the Pango attributes at iter to Vim highlighting attributes.
4829  * Ignore attributes not supported by Vim highlighting.  This shouldn't have
4830  * too much impact -- right now we handle even more attributes than necessary
4831  * for the IM modules I tested with.
4832  */
4833     static int
4834 translate_pango_attributes(PangoAttrIterator *iter)
4835 {
4836     PangoAttribute  *attr;
4837     int		    char_attr = HL_NORMAL;
4838 
4839     attr = pango_attr_iterator_get(iter, PANGO_ATTR_UNDERLINE);
4840     if (attr != NULL && ((PangoAttrInt *)attr)->value
4841 						 != (int)PANGO_UNDERLINE_NONE)
4842 	char_attr |= HL_UNDERLINE;
4843 
4844     attr = pango_attr_iterator_get(iter, PANGO_ATTR_WEIGHT);
4845     if (attr != NULL && ((PangoAttrInt *)attr)->value >= (int)PANGO_WEIGHT_BOLD)
4846 	char_attr |= HL_BOLD;
4847 
4848     attr = pango_attr_iterator_get(iter, PANGO_ATTR_STYLE);
4849     if (attr != NULL && ((PangoAttrInt *)attr)->value
4850 						   != (int)PANGO_STYLE_NORMAL)
4851 	char_attr |= HL_ITALIC;
4852 
4853     attr = pango_attr_iterator_get(iter, PANGO_ATTR_BACKGROUND);
4854     if (attr != NULL)
4855     {
4856 	const PangoColor *color = &((PangoAttrColor *)attr)->color;
4857 
4858 	/* Assume inverse if black background is requested */
4859 	if ((color->red | color->green | color->blue) == 0)
4860 	    char_attr |= HL_INVERSE;
4861     }
4862 
4863     return char_attr;
4864 }
4865 
4866 /*
4867  * Retrieve the highlighting attributes at column col in the preedit string.
4868  * Return -1 if not in preediting mode or if col is out of range.
4869  */
4870     int
4871 im_get_feedback_attr(int col)
4872 {
4873     char	    *preedit_string = NULL;
4874     PangoAttrList   *attr_list	    = NULL;
4875     int		    char_attr	    = -1;
4876 
4877     if (xic == NULL)
4878 	return char_attr;
4879 
4880     gtk_im_context_get_preedit_string(xic, &preedit_string, &attr_list, NULL);
4881 
4882     if (preedit_string != NULL && attr_list != NULL)
4883     {
4884 	int idx;
4885 
4886 	/* Get the byte index as used by PangoAttrIterator */
4887 	for (idx = 0; col > 0 && preedit_string[idx] != '\0'; --col)
4888 	    idx += utfc_ptr2len((char_u *)preedit_string + idx);
4889 
4890 	if (preedit_string[idx] != '\0')
4891 	{
4892 	    PangoAttrIterator	*iter;
4893 	    int			start, end;
4894 
4895 	    char_attr = HL_NORMAL;
4896 	    iter = pango_attr_list_get_iterator(attr_list);
4897 
4898 	    /* Extract all relevant attributes from the list. */
4899 	    do
4900 	    {
4901 		pango_attr_iterator_range(iter, &start, &end);
4902 
4903 		if (idx >= start && idx < end)
4904 		    char_attr |= translate_pango_attributes(iter);
4905 	    }
4906 	    while (pango_attr_iterator_next(iter));
4907 
4908 	    pango_attr_iterator_destroy(iter);
4909 	}
4910     }
4911 
4912     if (attr_list != NULL)
4913 	pango_attr_list_unref(attr_list);
4914     g_free(preedit_string);
4915 
4916     return char_attr;
4917 }
4918 
4919     void
4920 xim_init(void)
4921 {
4922 #ifdef XIM_DEBUG
4923     xim_log("xim_init()\n");
4924 #endif
4925 
4926     g_return_if_fail(gui.drawarea != NULL);
4927     g_return_if_fail(gui.drawarea->window != NULL);
4928 
4929     xic = gtk_im_multicontext_new();
4930     g_object_ref(xic);
4931 
4932     im_commit_handler_id = g_signal_connect(G_OBJECT(xic), "commit",
4933 					    G_CALLBACK(&im_commit_cb), NULL);
4934     g_signal_connect(G_OBJECT(xic), "preedit_changed",
4935 		     G_CALLBACK(&im_preedit_changed_cb), NULL);
4936     g_signal_connect(G_OBJECT(xic), "preedit_start",
4937 		     G_CALLBACK(&im_preedit_start_cb), NULL);
4938     g_signal_connect(G_OBJECT(xic), "preedit_end",
4939 		     G_CALLBACK(&im_preedit_end_cb), NULL);
4940 
4941     gtk_im_context_set_client_window(xic, gui.drawarea->window);
4942 }
4943 
4944     void
4945 im_shutdown(void)
4946 {
4947 #ifdef XIM_DEBUG
4948     xim_log("im_shutdown()\n");
4949 #endif
4950 
4951     if (xic != NULL)
4952     {
4953 	gtk_im_context_focus_out(xic);
4954 	g_object_unref(xic);
4955 	xic = NULL;
4956     }
4957     im_is_active = FALSE;
4958     im_commit_handler_id = 0;
4959     preedit_start_col = MAXCOL;
4960     xim_has_preediting = FALSE;
4961 }
4962 
4963 /*
4964  * Convert the string argument to keyval and state for GdkEventKey.
4965  * If str is valid return TRUE, otherwise FALSE.
4966  *
4967  * See 'imactivatekey' for documentation of the format.
4968  */
4969     static int
4970 im_string_to_keyval(const char *str, unsigned int *keyval, unsigned int *state)
4971 {
4972     const char	    *mods_end;
4973     unsigned	    tmp_keyval;
4974     unsigned	    tmp_state = 0;
4975 
4976     mods_end = strrchr(str, '-');
4977     mods_end = (mods_end != NULL) ? mods_end + 1 : str;
4978 
4979     /* Parse modifier keys */
4980     while (str < mods_end)
4981 	switch (*str++)
4982 	{
4983 	    case '-':							break;
4984 	    case 'S': case 's': tmp_state |= (unsigned)GDK_SHIFT_MASK;	break;
4985 	    case 'L': case 'l': tmp_state |= (unsigned)GDK_LOCK_MASK;	break;
4986 	    case 'C': case 'c': tmp_state |= (unsigned)GDK_CONTROL_MASK;break;
4987 	    case '1':		tmp_state |= (unsigned)GDK_MOD1_MASK;	break;
4988 	    case '2':		tmp_state |= (unsigned)GDK_MOD2_MASK;	break;
4989 	    case '3':		tmp_state |= (unsigned)GDK_MOD3_MASK;	break;
4990 	    case '4':		tmp_state |= (unsigned)GDK_MOD4_MASK;	break;
4991 	    case '5':		tmp_state |= (unsigned)GDK_MOD5_MASK;	break;
4992 	    default:
4993 		return FALSE;
4994 	}
4995 
4996     tmp_keyval = gdk_keyval_from_name(str);
4997 
4998     if (tmp_keyval == 0 || tmp_keyval == GDK_VoidSymbol)
4999 	return FALSE;
5000 
5001     if (keyval != NULL)
5002 	*keyval = tmp_keyval;
5003     if (state != NULL)
5004 	*state = tmp_state;
5005 
5006     return TRUE;
5007 }
5008 
5009 /*
5010  * Return TRUE if p_imak is valid, otherwise FALSE.  As a special case, an
5011  * empty string is also regarded as valid.
5012  *
5013  * Note: The numerical key value of p_imak is cached if it was valid; thus
5014  * boldly assuming im_xim_isvalid_imactivate() will always be called whenever
5015  * 'imak' changes.  This is currently the case but not obvious -- should
5016  * probably rename the function for clarity.
5017  */
5018     int
5019 im_xim_isvalid_imactivate(void)
5020 {
5021     if (p_imak[0] == NUL)
5022     {
5023 	im_activatekey_keyval = GDK_VoidSymbol;
5024 	im_activatekey_state  = 0;
5025 	return TRUE;
5026     }
5027 
5028     return im_string_to_keyval((const char *)p_imak,
5029 			       &im_activatekey_keyval,
5030 			       &im_activatekey_state);
5031 }
5032 
5033     static void
5034 im_synthesize_keypress(unsigned int keyval, unsigned int state)
5035 {
5036     GdkEventKey *event;
5037 
5038 #  ifdef HAVE_GTK_MULTIHEAD
5039     event = (GdkEventKey *)gdk_event_new(GDK_KEY_PRESS);
5040     g_object_ref(gui.drawarea->window); /* unreffed by gdk_event_free() */
5041 #  else
5042     event = (GdkEventKey *)g_malloc0((gulong)sizeof(GdkEvent));
5043     event->type = GDK_KEY_PRESS;
5044 #  endif
5045     event->window = gui.drawarea->window;
5046     event->send_event = TRUE;
5047     event->time = GDK_CURRENT_TIME;
5048     event->state  = state;
5049     event->keyval = keyval;
5050     event->hardware_keycode = /* needed for XIM */
5051 	XKeysymToKeycode(GDK_WINDOW_XDISPLAY(event->window), (KeySym)keyval);
5052     event->length = 0;
5053     event->string = NULL;
5054 
5055     gtk_im_context_filter_keypress(xic, event);
5056 
5057     /* For consistency, also send the corresponding release event. */
5058     event->type = GDK_KEY_RELEASE;
5059     event->send_event = FALSE;
5060     gtk_im_context_filter_keypress(xic, event);
5061 
5062 #  ifdef HAVE_GTK_MULTIHEAD
5063     gdk_event_free((GdkEvent *)event);
5064 #  else
5065     g_free(event);
5066 #  endif
5067 }
5068 
5069     void
5070 xim_reset(void)
5071 {
5072     if (xic != NULL)
5073     {
5074 	/*
5075 	 * The third-party imhangul module (and maybe others too) ignores
5076 	 * gtk_im_context_reset() or at least doesn't reset the active state.
5077 	 * Thus sending imactivatekey would turn it off if it was on before,
5078 	 * which is clearly not what we want.  Fortunately we can work around
5079 	 * that for imhangul by sending GDK_Escape, but I don't know if it
5080 	 * works with all IM modules that support an activation key :/
5081 	 *
5082 	 * An alternative approach would be to destroy the IM context and
5083 	 * recreate it.  But that means loading/unloading the IM module on
5084 	 * every mode switch, which causes a quite noticeable delay even on
5085 	 * my rather fast box...
5086 	 * *
5087 	 * Moreover, there are some XIM which cannot respond to
5088 	 * im_synthesize_keypress(). we hope that they reset by
5089 	 * xim_shutdown().
5090 	 */
5091 	if (im_activatekey_keyval != GDK_VoidSymbol && im_is_active)
5092 	    im_synthesize_keypress(GDK_Escape, 0U);
5093 
5094 	gtk_im_context_reset(xic);
5095 
5096 	/*
5097 	 * HACK for Ami: This sequence of function calls makes Ami handle
5098 	 * the IM reset graciously, without breaking loads of other stuff.
5099 	 * It seems to force English mode as well, which is exactly what we
5100 	 * want because it makes the Ami status display work reliably.
5101 	 */
5102 	gtk_im_context_set_use_preedit(xic, FALSE);
5103 
5104 	if (p_imdisable)
5105 	    im_shutdown();
5106 	else
5107 	{
5108 	    gtk_im_context_set_use_preedit(xic, TRUE);
5109 	    xim_set_focus(gui.in_focus);
5110 
5111 	    if (im_activatekey_keyval != GDK_VoidSymbol)
5112 	    {
5113 		if (im_is_active)
5114 		{
5115 		    g_signal_handler_block(xic, im_commit_handler_id);
5116 		    im_synthesize_keypress(im_activatekey_keyval,
5117 						    im_activatekey_state);
5118 		    g_signal_handler_unblock(xic, im_commit_handler_id);
5119 		}
5120 	    }
5121 	    else
5122 	    {
5123 		im_shutdown();
5124 		xim_init();
5125 		xim_set_focus(gui.in_focus);
5126 	    }
5127 	}
5128     }
5129 
5130     preedit_start_col = MAXCOL;
5131     xim_has_preediting = FALSE;
5132 }
5133 
5134     int
5135 xim_queue_key_press_event(GdkEventKey *event, int down)
5136 {
5137     if (down)
5138     {
5139 	/*
5140 	 * Workaround GTK2 XIM 'feature' that always converts keypad keys to
5141 	 * chars., even when not part of an IM sequence (ref. feature of
5142 	 * gdk/gdkkeyuni.c).
5143 	 * Flag any keypad keys that might represent a single char.
5144 	 * If this (on its own - i.e., not part of an IM sequence) is
5145 	 * committed while we're processing one of these keys, we can ignore
5146 	 * that commit and go ahead & process it ourselves.  That way we can
5147 	 * still distinguish keypad keys for use in mappings.
5148 	 * Also add GDK_space to make <S-Space> work.
5149 	 */
5150 	switch (event->keyval)
5151 	{
5152 	    case GDK_KP_Add:      xim_expected_char = '+';  break;
5153 	    case GDK_KP_Subtract: xim_expected_char = '-';  break;
5154 	    case GDK_KP_Divide:   xim_expected_char = '/';  break;
5155 	    case GDK_KP_Multiply: xim_expected_char = '*';  break;
5156 	    case GDK_KP_Decimal:  xim_expected_char = '.';  break;
5157 	    case GDK_KP_Equal:    xim_expected_char = '=';  break;
5158 	    case GDK_KP_0:	  xim_expected_char = '0';  break;
5159 	    case GDK_KP_1:	  xim_expected_char = '1';  break;
5160 	    case GDK_KP_2:	  xim_expected_char = '2';  break;
5161 	    case GDK_KP_3:	  xim_expected_char = '3';  break;
5162 	    case GDK_KP_4:	  xim_expected_char = '4';  break;
5163 	    case GDK_KP_5:	  xim_expected_char = '5';  break;
5164 	    case GDK_KP_6:	  xim_expected_char = '6';  break;
5165 	    case GDK_KP_7:	  xim_expected_char = '7';  break;
5166 	    case GDK_KP_8:	  xim_expected_char = '8';  break;
5167 	    case GDK_KP_9:	  xim_expected_char = '9';  break;
5168 	    case GDK_space:	  xim_expected_char = ' ';  break;
5169 	    default:		  xim_expected_char = NUL;
5170 	}
5171 	xim_ignored_char = FALSE;
5172     }
5173 
5174     /*
5175      * When typing fFtT, XIM may be activated. Thus it must pass
5176      * gtk_im_context_filter_keypress() in Normal mode.
5177      * And while doing :sh too.
5178      */
5179     if (xic != NULL && !p_imdisable
5180 		    && (State & (INSERT | CMDLINE | NORMAL | EXTERNCMD)) != 0)
5181     {
5182 	/*
5183 	 * Filter 'imactivatekey' and map it to CTRL-^.  This way, Vim is
5184 	 * always aware of the current status of IM, and can even emulate
5185 	 * the activation key for modules that don't support one.
5186 	 */
5187 	if (event->keyval == im_activatekey_keyval
5188 	     && (event->state & im_activatekey_state) == im_activatekey_state)
5189 	{
5190 	    unsigned int state_mask;
5191 
5192 	    /* Require the state of the 3 most used modifiers to match exactly.
5193 	     * Otherwise e.g. <S-C-space> would be unusable for other purposes
5194 	     * if the IM activate key is <S-space>. */
5195 	    state_mask  = im_activatekey_state;
5196 	    state_mask |= ((int)GDK_SHIFT_MASK | (int)GDK_CONTROL_MASK
5197 							| (int)GDK_MOD1_MASK);
5198 
5199 	    if ((event->state & state_mask) != im_activatekey_state)
5200 		return FALSE;
5201 
5202 	    /* Don't send it a second time on GDK_KEY_RELEASE. */
5203 	    if (event->type != GDK_KEY_PRESS)
5204 		return TRUE;
5205 
5206 	    if (map_to_exists_mode((char_u *)"", LANGMAP, FALSE))
5207 	    {
5208 		im_set_active(FALSE);
5209 
5210 		/* ":lmap" mappings exists, toggle use of mappings. */
5211 		State ^= LANGMAP;
5212 		if (State & LANGMAP)
5213 		{
5214 		    curbuf->b_p_iminsert = B_IMODE_NONE;
5215 		    State &= ~LANGMAP;
5216 		}
5217 		else
5218 		{
5219 		    curbuf->b_p_iminsert = B_IMODE_LMAP;
5220 		    State |= LANGMAP;
5221 		}
5222 		return TRUE;
5223 	    }
5224 
5225 	    return gtk_im_context_filter_keypress(xic, event);
5226 	}
5227 
5228 	/* Don't filter events through the IM context if IM isn't active
5229 	 * right now.  Unlike with GTK+ 1.2 we cannot rely on the IM module
5230 	 * not doing anything before the activation key was sent. */
5231 	if (im_activatekey_keyval == GDK_VoidSymbol || im_is_active)
5232 	{
5233 	    int imresult = gtk_im_context_filter_keypress(xic, event);
5234 
5235 	    /* Some XIM send following sequence:
5236 	     * 1. preedited string.
5237 	     * 2. committed string.
5238 	     * 3. line changed key.
5239 	     * 4. preedited string.
5240 	     * 5. remove preedited string.
5241 	     * if 3, Vim can't move back the above line for 5.
5242 	     * thus, this part should not parse the key. */
5243 	    if (!imresult && preedit_start_col != MAXCOL
5244 					       && event->keyval == GDK_Return)
5245 	    {
5246 		im_synthesize_keypress(GDK_Return, 0U);
5247 		return FALSE;
5248 	    }
5249 
5250 	    /* If XIM tried to commit a keypad key as a single char.,
5251 	     * ignore it so we can use the keypad key 'raw', for mappings. */
5252 	    if (xim_expected_char != NUL && xim_ignored_char)
5253 		/* We had a keypad key, and XIM tried to thieve it */
5254 		return FALSE;
5255 
5256 	    /* This is supposed to fix a problem with iBus, that space
5257 	     * characters don't work in input mode. */
5258 	    xim_expected_char = NUL;
5259 
5260 	    /* Normal processing */
5261 	    return imresult;
5262 	}
5263     }
5264 
5265     return FALSE;
5266 }
5267 
5268     int
5269 im_get_status(void)
5270 {
5271     return im_is_active;
5272 }
5273 
5274     int
5275 preedit_get_status(void)
5276 {
5277     return preedit_is_active;
5278 }
5279 
5280     int
5281 im_is_preediting()
5282 {
5283     return xim_has_preediting;
5284 }
5285 
5286 # else /* !FEAT_GUI_GTK */
5287 
5288 static int	xim_is_active = FALSE;  /* XIM should be active in the current
5289 					   mode */
5290 static int	xim_has_focus = FALSE;	/* XIM is really being used for Vim */
5291 #ifdef FEAT_GUI_X11
5292 static XIMStyle	input_style;
5293 static int	status_area_enabled = TRUE;
5294 #endif
5295 
5296 /*
5297  * Switch using XIM on/off.  This is used by the code that changes "State".
5298  */
5299     void
5300 im_set_active(active)
5301     int		active;
5302 {
5303     if (xic == NULL)
5304 	return;
5305 
5306     /* If 'imdisable' is set, XIM is never active. */
5307     if (p_imdisable)
5308 	active = FALSE;
5309 #if !defined(FEAT_GUI_GTK)
5310     else if (input_style & XIMPreeditPosition)
5311 	/* There is a problem in switching XIM off when preediting is used,
5312 	 * and it is not clear how this can be solved.  For now, keep XIM on
5313 	 * all the time, like it was done in Vim 5.8. */
5314 	active = TRUE;
5315 #endif
5316 
5317     /* Remember the active state, it is needed when Vim gets keyboard focus. */
5318     xim_is_active = active;
5319     xim_set_preedit();
5320 }
5321 
5322 /*
5323  * Adjust using XIM for gaining or losing keyboard focus.  Also called when
5324  * "xim_is_active" changes.
5325  */
5326     void
5327 xim_set_focus(focus)
5328     int		focus;
5329 {
5330     if (xic == NULL)
5331 	return;
5332 
5333     /*
5334      * XIM only gets focus when the Vim window has keyboard focus and XIM has
5335      * been set active for the current mode.
5336      */
5337     if (focus && xim_is_active)
5338     {
5339 	if (!xim_has_focus)
5340 	{
5341 	    xim_has_focus = TRUE;
5342 	    XSetICFocus(xic);
5343 	}
5344     }
5345     else
5346     {
5347 	if (xim_has_focus)
5348 	{
5349 	    xim_has_focus = FALSE;
5350 	    XUnsetICFocus(xic);
5351 	}
5352     }
5353 }
5354 
5355     void
5356 im_set_position(row, col)
5357     int		row UNUSED;
5358     int		col UNUSED;
5359 {
5360     xim_set_preedit();
5361 }
5362 
5363 /*
5364  * Set the XIM to the current cursor position.
5365  */
5366     void
5367 xim_set_preedit()
5368 {
5369     XVaNestedList attr_list;
5370     XRectangle spot_area;
5371     XPoint over_spot;
5372     int line_space;
5373 
5374     if (xic == NULL)
5375 	return;
5376 
5377     xim_set_focus(TRUE);
5378 
5379     if (!xim_has_focus)
5380     {
5381 	/* hide XIM cursor */
5382 	over_spot.x = 0;
5383 	over_spot.y = -100; /* arbitrary invisible position */
5384 	attr_list = (XVaNestedList) XVaCreateNestedList(0,
5385 							XNSpotLocation,
5386 							&over_spot,
5387 							NULL);
5388 	XSetICValues(xic, XNPreeditAttributes, attr_list, NULL);
5389 	XFree(attr_list);
5390 	return;
5391     }
5392 
5393     if (input_style & XIMPreeditPosition)
5394     {
5395 	if (xim_fg_color == INVALCOLOR)
5396 	{
5397 	    xim_fg_color = gui.def_norm_pixel;
5398 	    xim_bg_color = gui.def_back_pixel;
5399 	}
5400 	over_spot.x = TEXT_X(gui.col);
5401 	over_spot.y = TEXT_Y(gui.row);
5402 	spot_area.x = 0;
5403 	spot_area.y = 0;
5404 	spot_area.height = gui.char_height * Rows;
5405 	spot_area.width  = gui.char_width * Columns;
5406 	line_space = gui.char_height;
5407 	attr_list = (XVaNestedList) XVaCreateNestedList(0,
5408 					XNSpotLocation, &over_spot,
5409 					XNForeground, (Pixel) xim_fg_color,
5410 					XNBackground, (Pixel) xim_bg_color,
5411 					XNArea, &spot_area,
5412 					XNLineSpace, line_space,
5413 					NULL);
5414 	if (XSetICValues(xic, XNPreeditAttributes, attr_list, NULL))
5415 	    EMSG(_("E284: Cannot set IC values"));
5416 	XFree(attr_list);
5417     }
5418 }
5419 
5420 #if defined(FEAT_GUI_X11)
5421 static char e_xim[] = N_("E285: Failed to create input context");
5422 #endif
5423 
5424 #if defined(FEAT_GUI_X11) || defined(PROTO)
5425 # if defined(XtSpecificationRelease) && XtSpecificationRelease >= 6 && !defined(sun)
5426 #  define USE_X11R6_XIM
5427 # endif
5428 
5429 static int xim_real_init __ARGS((Window x11_window, Display *x11_display));
5430 
5431 
5432 #ifdef USE_X11R6_XIM
5433 static void xim_instantiate_cb __ARGS((Display *display, XPointer client_data, XPointer	call_data));
5434 static void xim_destroy_cb __ARGS((XIM im, XPointer client_data, XPointer call_data));
5435 
5436     static void
5437 xim_instantiate_cb(display, client_data, call_data)
5438     Display	*display;
5439     XPointer	client_data UNUSED;
5440     XPointer	call_data UNUSED;
5441 {
5442     Window	x11_window;
5443     Display	*x11_display;
5444 
5445 #ifdef XIM_DEBUG
5446     xim_log("xim_instantiate_cb()\n");
5447 #endif
5448 
5449     gui_get_x11_windis(&x11_window, &x11_display);
5450     if (display != x11_display)
5451 	return;
5452 
5453     xim_real_init(x11_window, x11_display);
5454     gui_set_shellsize(FALSE, FALSE, RESIZE_BOTH);
5455     if (xic != NULL)
5456 	XUnregisterIMInstantiateCallback(x11_display, NULL, NULL, NULL,
5457 					 xim_instantiate_cb, NULL);
5458 }
5459 
5460     static void
5461 xim_destroy_cb(im, client_data, call_data)
5462     XIM		im UNUSED;
5463     XPointer	client_data UNUSED;
5464     XPointer	call_data UNUSED;
5465 {
5466     Window	x11_window;
5467     Display	*x11_display;
5468 
5469 #ifdef XIM_DEBUG
5470     xim_log("xim_destroy_cb()\n");
5471 #endif
5472     gui_get_x11_windis(&x11_window, &x11_display);
5473 
5474     xic = NULL;
5475     status_area_enabled = FALSE;
5476 
5477     gui_set_shellsize(FALSE, FALSE, RESIZE_BOTH);
5478 
5479     XRegisterIMInstantiateCallback(x11_display, NULL, NULL, NULL,
5480 				   xim_instantiate_cb, NULL);
5481 }
5482 #endif
5483 
5484     void
5485 xim_init()
5486 {
5487     Window	x11_window;
5488     Display	*x11_display;
5489 
5490 #ifdef XIM_DEBUG
5491     xim_log("xim_init()\n");
5492 #endif
5493 
5494     gui_get_x11_windis(&x11_window, &x11_display);
5495 
5496     xic = NULL;
5497 
5498     if (xim_real_init(x11_window, x11_display))
5499 	return;
5500 
5501     gui_set_shellsize(FALSE, FALSE, RESIZE_BOTH);
5502 
5503 #ifdef USE_X11R6_XIM
5504     XRegisterIMInstantiateCallback(x11_display, NULL, NULL, NULL,
5505 				   xim_instantiate_cb, NULL);
5506 #endif
5507 }
5508 
5509     static int
5510 xim_real_init(x11_window, x11_display)
5511     Window	x11_window;
5512     Display	*x11_display;
5513 {
5514     int		i;
5515     char	*p,
5516 		*s,
5517 		*ns,
5518 		*end,
5519 		tmp[1024];
5520 #define IMLEN_MAX 40
5521     char	buf[IMLEN_MAX + 7];
5522     XIM		xim = NULL;
5523     XIMStyles	*xim_styles;
5524     XIMStyle	this_input_style = 0;
5525     Boolean	found;
5526     XPoint	over_spot;
5527     XVaNestedList preedit_list, status_list;
5528 
5529     input_style = 0;
5530     status_area_enabled = FALSE;
5531 
5532     if (xic != NULL)
5533 	return FALSE;
5534 
5535     if (gui.rsrc_input_method != NULL && *gui.rsrc_input_method != NUL)
5536     {
5537 	strcpy(tmp, gui.rsrc_input_method);
5538 	for (ns = s = tmp; ns != NULL && *s != NUL;)
5539 	{
5540 	    s = (char *)skipwhite((char_u *)s);
5541 	    if (*s == NUL)
5542 		break;
5543 	    if ((ns = end = strchr(s, ',')) == NULL)
5544 		end = s + strlen(s);
5545 	    while (isspace(((char_u *)end)[-1]))
5546 		end--;
5547 	    *end = NUL;
5548 
5549 	    if (strlen(s) <= IMLEN_MAX)
5550 	    {
5551 		strcpy(buf, "@im=");
5552 		strcat(buf, s);
5553 		if ((p = XSetLocaleModifiers(buf)) != NULL && *p != NUL
5554 			&& (xim = XOpenIM(x11_display, NULL, NULL, NULL))
5555 								      != NULL)
5556 		    break;
5557 	    }
5558 
5559 	    s = ns + 1;
5560 	}
5561     }
5562 
5563     if (xim == NULL && (p = XSetLocaleModifiers("")) != NULL && *p != NUL)
5564 	xim = XOpenIM(x11_display, NULL, NULL, NULL);
5565 
5566     /* This is supposed to be useful to obtain characters through
5567      * XmbLookupString() without really using a XIM. */
5568     if (xim == NULL && (p = XSetLocaleModifiers("@im=none")) != NULL
5569 								 && *p != NUL)
5570 	xim = XOpenIM(x11_display, NULL, NULL, NULL);
5571 
5572     if (xim == NULL)
5573     {
5574 	/* Only give this message when verbose is set, because too many people
5575 	 * got this message when they didn't want to use a XIM. */
5576 	if (p_verbose > 0)
5577 	{
5578 	    verbose_enter();
5579 	    EMSG(_("E286: Failed to open input method"));
5580 	    verbose_leave();
5581 	}
5582 	return FALSE;
5583     }
5584 
5585 #ifdef USE_X11R6_XIM
5586     {
5587 	XIMCallback destroy_cb;
5588 
5589 	destroy_cb.callback = xim_destroy_cb;
5590 	destroy_cb.client_data = NULL;
5591 	if (XSetIMValues(xim, XNDestroyCallback, &destroy_cb, NULL))
5592 	    EMSG(_("E287: Warning: Could not set destroy callback to IM"));
5593     }
5594 #endif
5595 
5596     if (XGetIMValues(xim, XNQueryInputStyle, &xim_styles, NULL) || !xim_styles)
5597     {
5598 	EMSG(_("E288: input method doesn't support any style"));
5599 	XCloseIM(xim);
5600 	return FALSE;
5601     }
5602 
5603     found = False;
5604     strcpy(tmp, gui.rsrc_preedit_type_name);
5605     for (s = tmp; s && !found; )
5606     {
5607 	while (*s && isspace((unsigned char)*s))
5608 	    s++;
5609 	if (!*s)
5610 	    break;
5611 	if ((ns = end = strchr(s, ',')) != 0)
5612 	    ns++;
5613 	else
5614 	    end = s + strlen(s);
5615 	while (isspace((unsigned char)*end))
5616 	    end--;
5617 	*end = '\0';
5618 
5619 	if (!strcmp(s, "OverTheSpot"))
5620 	    this_input_style = (XIMPreeditPosition | XIMStatusArea);
5621 	else if (!strcmp(s, "OffTheSpot"))
5622 	    this_input_style = (XIMPreeditArea | XIMStatusArea);
5623 	else if (!strcmp(s, "Root"))
5624 	    this_input_style = (XIMPreeditNothing | XIMStatusNothing);
5625 
5626 	for (i = 0; (unsigned short)i < xim_styles->count_styles; i++)
5627 	{
5628 	    if (this_input_style == xim_styles->supported_styles[i])
5629 	    {
5630 		found = True;
5631 		break;
5632 	    }
5633 	}
5634 	if (!found)
5635 	    for (i = 0; (unsigned short)i < xim_styles->count_styles; i++)
5636 	    {
5637 		if ((xim_styles->supported_styles[i] & this_input_style)
5638 			== (this_input_style & ~XIMStatusArea))
5639 		{
5640 		    this_input_style &= ~XIMStatusArea;
5641 		    found = True;
5642 		    break;
5643 		}
5644 	    }
5645 
5646 	s = ns;
5647     }
5648     XFree(xim_styles);
5649 
5650     if (!found)
5651     {
5652 	/* Only give this message when verbose is set, because too many people
5653 	 * got this message when they didn't want to use a XIM. */
5654 	if (p_verbose > 0)
5655 	{
5656 	    verbose_enter();
5657 	    EMSG(_("E289: input method doesn't support my preedit type"));
5658 	    verbose_leave();
5659 	}
5660 	XCloseIM(xim);
5661 	return FALSE;
5662     }
5663 
5664     over_spot.x = TEXT_X(gui.col);
5665     over_spot.y = TEXT_Y(gui.row);
5666     input_style = this_input_style;
5667 
5668     /* A crash was reported when trying to pass gui.norm_font as XNFontSet,
5669      * thus that has been removed.  Hopefully the default works... */
5670 #ifdef FEAT_XFONTSET
5671     if (gui.fontset != NOFONTSET)
5672     {
5673 	preedit_list = XVaCreateNestedList(0,
5674 				XNSpotLocation, &over_spot,
5675 				XNForeground, (Pixel)gui.def_norm_pixel,
5676 				XNBackground, (Pixel)gui.def_back_pixel,
5677 				XNFontSet, (XFontSet)gui.fontset,
5678 				NULL);
5679 	status_list = XVaCreateNestedList(0,
5680 				XNForeground, (Pixel)gui.def_norm_pixel,
5681 				XNBackground, (Pixel)gui.def_back_pixel,
5682 				XNFontSet, (XFontSet)gui.fontset,
5683 				NULL);
5684     }
5685     else
5686 #endif
5687     {
5688 	preedit_list = XVaCreateNestedList(0,
5689 				XNSpotLocation, &over_spot,
5690 				XNForeground, (Pixel)gui.def_norm_pixel,
5691 				XNBackground, (Pixel)gui.def_back_pixel,
5692 				NULL);
5693 	status_list = XVaCreateNestedList(0,
5694 				XNForeground, (Pixel)gui.def_norm_pixel,
5695 				XNBackground, (Pixel)gui.def_back_pixel,
5696 				NULL);
5697     }
5698 
5699     xic = XCreateIC(xim,
5700 		    XNInputStyle, input_style,
5701 		    XNClientWindow, x11_window,
5702 		    XNFocusWindow, gui.wid,
5703 		    XNPreeditAttributes, preedit_list,
5704 		    XNStatusAttributes, status_list,
5705 		    NULL);
5706     XFree(status_list);
5707     XFree(preedit_list);
5708     if (xic != NULL)
5709     {
5710 	if (input_style & XIMStatusArea)
5711 	{
5712 	    xim_set_status_area();
5713 	    status_area_enabled = TRUE;
5714 	}
5715 	else
5716 	    gui_set_shellsize(FALSE, FALSE, RESIZE_BOTH);
5717     }
5718     else
5719     {
5720 	EMSG(_(e_xim));
5721 	XCloseIM(xim);
5722 	return FALSE;
5723     }
5724 
5725     return TRUE;
5726 }
5727 
5728 #endif /* FEAT_GUI_X11 */
5729 
5730 /*
5731  * Get IM status.  When IM is on, return TRUE.  Else return FALSE.
5732  * FIXME: This doesn't work correctly: Having focus doesn't always mean XIM is
5733  * active, when not having focus XIM may still be active (e.g., when using a
5734  * tear-off menu item).
5735  */
5736     int
5737 im_get_status()
5738 {
5739     return xim_has_focus;
5740 }
5741 
5742 # endif /* !FEAT_GUI_GTK */
5743 
5744 # if !defined(FEAT_GUI_GTK) || defined(PROTO)
5745 /*
5746  * Set up the status area.
5747  *
5748  * This should use a separate Widget, but that seems not possible, because
5749  * preedit_area and status_area should be set to the same window as for the
5750  * text input.  Unfortunately this means the status area pollutes the text
5751  * window...
5752  */
5753     void
5754 xim_set_status_area()
5755 {
5756     XVaNestedList preedit_list = 0, status_list = 0, list = 0;
5757     XRectangle pre_area, status_area;
5758 
5759     if (xic == NULL)
5760 	return;
5761 
5762     if (input_style & XIMStatusArea)
5763     {
5764 	if (input_style & XIMPreeditArea)
5765 	{
5766 	    XRectangle *needed_rect;
5767 
5768 	    /* to get status_area width */
5769 	    status_list = XVaCreateNestedList(0, XNAreaNeeded,
5770 					      &needed_rect, NULL);
5771 	    XGetICValues(xic, XNStatusAttributes, status_list, NULL);
5772 	    XFree(status_list);
5773 
5774 	    status_area.width = needed_rect->width;
5775 	}
5776 	else
5777 	    status_area.width = gui.char_width * Columns;
5778 
5779 	status_area.x = 0;
5780 	status_area.y = gui.char_height * Rows + gui.border_offset;
5781 	if (gui.which_scrollbars[SBAR_BOTTOM])
5782 	    status_area.y += gui.scrollbar_height;
5783 #ifdef FEAT_MENU
5784 	if (gui.menu_is_active)
5785 	    status_area.y += gui.menu_height;
5786 #endif
5787 	status_area.height = gui.char_height;
5788 	status_list = XVaCreateNestedList(0, XNArea, &status_area, NULL);
5789     }
5790     else
5791     {
5792 	status_area.x = 0;
5793 	status_area.y = gui.char_height * Rows + gui.border_offset;
5794 	if (gui.which_scrollbars[SBAR_BOTTOM])
5795 	    status_area.y += gui.scrollbar_height;
5796 #ifdef FEAT_MENU
5797 	if (gui.menu_is_active)
5798 	    status_area.y += gui.menu_height;
5799 #endif
5800 	status_area.width = 0;
5801 	status_area.height = gui.char_height;
5802     }
5803 
5804     if (input_style & XIMPreeditArea)   /* off-the-spot */
5805     {
5806 	pre_area.x = status_area.x + status_area.width;
5807 	pre_area.y = gui.char_height * Rows + gui.border_offset;
5808 	pre_area.width = gui.char_width * Columns - pre_area.x;
5809 	if (gui.which_scrollbars[SBAR_BOTTOM])
5810 	    pre_area.y += gui.scrollbar_height;
5811 #ifdef FEAT_MENU
5812 	if (gui.menu_is_active)
5813 	    pre_area.y += gui.menu_height;
5814 #endif
5815 	pre_area.height = gui.char_height;
5816 	preedit_list = XVaCreateNestedList(0, XNArea, &pre_area, NULL);
5817     }
5818     else if (input_style & XIMPreeditPosition)   /* over-the-spot */
5819     {
5820 	pre_area.x = 0;
5821 	pre_area.y = 0;
5822 	pre_area.height = gui.char_height * Rows;
5823 	pre_area.width = gui.char_width * Columns;
5824 	preedit_list = XVaCreateNestedList(0, XNArea, &pre_area, NULL);
5825     }
5826 
5827     if (preedit_list && status_list)
5828 	list = XVaCreateNestedList(0, XNPreeditAttributes, preedit_list,
5829 				   XNStatusAttributes, status_list, NULL);
5830     else if (preedit_list)
5831 	list = XVaCreateNestedList(0, XNPreeditAttributes, preedit_list,
5832 				   NULL);
5833     else if (status_list)
5834 	list = XVaCreateNestedList(0, XNStatusAttributes, status_list,
5835 				   NULL);
5836     else
5837 	list = NULL;
5838 
5839     if (list)
5840     {
5841 	XSetICValues(xic, XNVaNestedList, list, NULL);
5842 	XFree(list);
5843     }
5844     if (status_list)
5845 	XFree(status_list);
5846     if (preedit_list)
5847 	XFree(preedit_list);
5848 }
5849 
5850     int
5851 xim_get_status_area_height()
5852 {
5853     if (status_area_enabled)
5854 	return gui.char_height;
5855     return 0;
5856 }
5857 # endif
5858 
5859 #endif /* FEAT_XIM */
5860 
5861 #if defined(FEAT_MBYTE) || defined(PROTO)
5862 
5863 /*
5864  * Setup "vcp" for conversion from "from" to "to".
5865  * The names must have been made canonical with enc_canonize().
5866  * vcp->vc_type must have been initialized to CONV_NONE.
5867  * Note: cannot be used for conversion from/to ucs-2 and ucs-4 (will use utf-8
5868  * instead).
5869  * Afterwards invoke with "from" and "to" equal to NULL to cleanup.
5870  * Return FAIL when conversion is not supported, OK otherwise.
5871  */
5872     int
5873 convert_setup(vcp, from, to)
5874     vimconv_T	*vcp;
5875     char_u	*from;
5876     char_u	*to;
5877 {
5878     return convert_setup_ext(vcp, from, TRUE, to, TRUE);
5879 }
5880 
5881 /*
5882  * As convert_setup(), but only when from_unicode_is_utf8 is TRUE will all
5883  * "from" unicode charsets be considered utf-8.  Same for "to".
5884  */
5885     int
5886 convert_setup_ext(vcp, from, from_unicode_is_utf8, to, to_unicode_is_utf8)
5887     vimconv_T	*vcp;
5888     char_u	*from;
5889     int		from_unicode_is_utf8;
5890     char_u	*to;
5891     int		to_unicode_is_utf8;
5892 {
5893     int		from_prop;
5894     int		to_prop;
5895     int		from_is_utf8;
5896     int		to_is_utf8;
5897 
5898     /* Reset to no conversion. */
5899 # ifdef USE_ICONV
5900     if (vcp->vc_type == CONV_ICONV && vcp->vc_fd != (iconv_t)-1)
5901 	iconv_close(vcp->vc_fd);
5902 # endif
5903     vcp->vc_type = CONV_NONE;
5904     vcp->vc_factor = 1;
5905     vcp->vc_fail = FALSE;
5906 
5907     /* No conversion when one of the names is empty or they are equal. */
5908     if (from == NULL || *from == NUL || to == NULL || *to == NUL
5909 						     || STRCMP(from, to) == 0)
5910 	return OK;
5911 
5912     from_prop = enc_canon_props(from);
5913     to_prop = enc_canon_props(to);
5914     if (from_unicode_is_utf8)
5915 	from_is_utf8 = from_prop & ENC_UNICODE;
5916     else
5917 	from_is_utf8 = from_prop == ENC_UNICODE;
5918     if (to_unicode_is_utf8)
5919 	to_is_utf8 = to_prop & ENC_UNICODE;
5920     else
5921 	to_is_utf8 = to_prop == ENC_UNICODE;
5922 
5923     if ((from_prop & ENC_LATIN1) && to_is_utf8)
5924     {
5925 	/* Internal latin1 -> utf-8 conversion. */
5926 	vcp->vc_type = CONV_TO_UTF8;
5927 	vcp->vc_factor = 2;	/* up to twice as long */
5928     }
5929     else if ((from_prop & ENC_LATIN9) && to_is_utf8)
5930     {
5931 	/* Internal latin9 -> utf-8 conversion. */
5932 	vcp->vc_type = CONV_9_TO_UTF8;
5933 	vcp->vc_factor = 3;	/* up to three as long (euro sign) */
5934     }
5935     else if (from_is_utf8 && (to_prop & ENC_LATIN1))
5936     {
5937 	/* Internal utf-8 -> latin1 conversion. */
5938 	vcp->vc_type = CONV_TO_LATIN1;
5939     }
5940     else if (from_is_utf8 && (to_prop & ENC_LATIN9))
5941     {
5942 	/* Internal utf-8 -> latin9 conversion. */
5943 	vcp->vc_type = CONV_TO_LATIN9;
5944     }
5945 #ifdef WIN3264
5946     /* Win32-specific codepage <-> codepage conversion without iconv. */
5947     else if ((from_is_utf8 || encname2codepage(from) > 0)
5948 	    && (to_is_utf8 || encname2codepage(to) > 0))
5949     {
5950 	vcp->vc_type = CONV_CODEPAGE;
5951 	vcp->vc_factor = 2;	/* up to twice as long */
5952 	vcp->vc_cpfrom = from_is_utf8 ? 0 : encname2codepage(from);
5953 	vcp->vc_cpto = to_is_utf8 ? 0 : encname2codepage(to);
5954     }
5955 #endif
5956 #ifdef MACOS_X
5957     else if ((from_prop & ENC_MACROMAN) && (to_prop & ENC_LATIN1))
5958     {
5959 	vcp->vc_type = CONV_MAC_LATIN1;
5960     }
5961     else if ((from_prop & ENC_MACROMAN) && to_is_utf8)
5962     {
5963 	vcp->vc_type = CONV_MAC_UTF8;
5964 	vcp->vc_factor = 2;	/* up to twice as long */
5965     }
5966     else if ((from_prop & ENC_LATIN1) && (to_prop & ENC_MACROMAN))
5967     {
5968 	vcp->vc_type = CONV_LATIN1_MAC;
5969     }
5970     else if (from_is_utf8 && (to_prop & ENC_MACROMAN))
5971     {
5972 	vcp->vc_type = CONV_UTF8_MAC;
5973     }
5974 #endif
5975 # ifdef USE_ICONV
5976     else
5977     {
5978 	/* Use iconv() for conversion. */
5979 	vcp->vc_fd = (iconv_t)my_iconv_open(
5980 		to_is_utf8 ? (char_u *)"utf-8" : to,
5981 		from_is_utf8 ? (char_u *)"utf-8" : from);
5982 	if (vcp->vc_fd != (iconv_t)-1)
5983 	{
5984 	    vcp->vc_type = CONV_ICONV;
5985 	    vcp->vc_factor = 4;	/* could be longer too... */
5986 	}
5987     }
5988 # endif
5989     if (vcp->vc_type == CONV_NONE)
5990 	return FAIL;
5991 
5992     return OK;
5993 }
5994 
5995 #if defined(FEAT_GUI) || defined(AMIGA) || defined(WIN3264) \
5996 	|| defined(MSDOS) || defined(PROTO)
5997 /*
5998  * Do conversion on typed input characters in-place.
5999  * The input and output are not NUL terminated!
6000  * Returns the length after conversion.
6001  */
6002     int
6003 convert_input(ptr, len, maxlen)
6004     char_u	*ptr;
6005     int		len;
6006     int		maxlen;
6007 {
6008     return convert_input_safe(ptr, len, maxlen, NULL, NULL);
6009 }
6010 #endif
6011 
6012 /*
6013  * Like convert_input(), but when there is an incomplete byte sequence at the
6014  * end return that as an allocated string in "restp" and set "*restlenp" to
6015  * the length.  If "restp" is NULL it is not used.
6016  */
6017     int
6018 convert_input_safe(ptr, len, maxlen, restp, restlenp)
6019     char_u	*ptr;
6020     int		len;
6021     int		maxlen;
6022     char_u	**restp;
6023     int		*restlenp;
6024 {
6025     char_u	*d;
6026     int		dlen = len;
6027     int		unconvertlen = 0;
6028 
6029     d = string_convert_ext(&input_conv, ptr, &dlen,
6030 					restp == NULL ? NULL : &unconvertlen);
6031     if (d != NULL)
6032     {
6033 	if (dlen <= maxlen)
6034 	{
6035 	    if (unconvertlen > 0)
6036 	    {
6037 		/* Move the unconverted characters to allocated memory. */
6038 		*restp = alloc(unconvertlen);
6039 		if (*restp != NULL)
6040 		    mch_memmove(*restp, ptr + len - unconvertlen, unconvertlen);
6041 		*restlenp = unconvertlen;
6042 	    }
6043 	    mch_memmove(ptr, d, dlen);
6044 	}
6045 	else
6046 	    /* result is too long, keep the unconverted text (the caller must
6047 	     * have done something wrong!) */
6048 	    dlen = len;
6049 	vim_free(d);
6050     }
6051     return dlen;
6052 }
6053 
6054 /*
6055  * Convert text "ptr[*lenp]" according to "vcp".
6056  * Returns the result in allocated memory and sets "*lenp".
6057  * When "lenp" is NULL, use NUL terminated strings.
6058  * Illegal chars are often changed to "?", unless vcp->vc_fail is set.
6059  * When something goes wrong, NULL is returned and "*lenp" is unchanged.
6060  */
6061     char_u *
6062 string_convert(vcp, ptr, lenp)
6063     vimconv_T	*vcp;
6064     char_u	*ptr;
6065     int		*lenp;
6066 {
6067     return string_convert_ext(vcp, ptr, lenp, NULL);
6068 }
6069 
6070 /*
6071  * Like string_convert(), but when "unconvlenp" is not NULL and there are is
6072  * an incomplete sequence at the end it is not converted and "*unconvlenp" is
6073  * set to the number of remaining bytes.
6074  */
6075     char_u *
6076 string_convert_ext(vcp, ptr, lenp, unconvlenp)
6077     vimconv_T	*vcp;
6078     char_u	*ptr;
6079     int		*lenp;
6080     int		*unconvlenp;
6081 {
6082     char_u	*retval = NULL;
6083     char_u	*d;
6084     int		len;
6085     int		i;
6086     int		l;
6087     int		c;
6088 
6089     if (lenp == NULL)
6090 	len = (int)STRLEN(ptr);
6091     else
6092 	len = *lenp;
6093     if (len == 0)
6094 	return vim_strsave((char_u *)"");
6095 
6096     switch (vcp->vc_type)
6097     {
6098 	case CONV_TO_UTF8:	/* latin1 to utf-8 conversion */
6099 	    retval = alloc(len * 2 + 1);
6100 	    if (retval == NULL)
6101 		break;
6102 	    d = retval;
6103 	    for (i = 0; i < len; ++i)
6104 	    {
6105 		c = ptr[i];
6106 		if (c < 0x80)
6107 		    *d++ = c;
6108 		else
6109 		{
6110 		    *d++ = 0xc0 + ((unsigned)c >> 6);
6111 		    *d++ = 0x80 + (c & 0x3f);
6112 		}
6113 	    }
6114 	    *d = NUL;
6115 	    if (lenp != NULL)
6116 		*lenp = (int)(d - retval);
6117 	    break;
6118 
6119 	case CONV_9_TO_UTF8:	/* latin9 to utf-8 conversion */
6120 	    retval = alloc(len * 3 + 1);
6121 	    if (retval == NULL)
6122 		break;
6123 	    d = retval;
6124 	    for (i = 0; i < len; ++i)
6125 	    {
6126 		c = ptr[i];
6127 		switch (c)
6128 		{
6129 		    case 0xa4: c = 0x20ac; break;   /* euro */
6130 		    case 0xa6: c = 0x0160; break;   /* S hat */
6131 		    case 0xa8: c = 0x0161; break;   /* S -hat */
6132 		    case 0xb4: c = 0x017d; break;   /* Z hat */
6133 		    case 0xb8: c = 0x017e; break;   /* Z -hat */
6134 		    case 0xbc: c = 0x0152; break;   /* OE */
6135 		    case 0xbd: c = 0x0153; break;   /* oe */
6136 		    case 0xbe: c = 0x0178; break;   /* Y */
6137 		}
6138 		d += utf_char2bytes(c, d);
6139 	    }
6140 	    *d = NUL;
6141 	    if (lenp != NULL)
6142 		*lenp = (int)(d - retval);
6143 	    break;
6144 
6145 	case CONV_TO_LATIN1:	/* utf-8 to latin1 conversion */
6146 	case CONV_TO_LATIN9:	/* utf-8 to latin9 conversion */
6147 	    retval = alloc(len + 1);
6148 	    if (retval == NULL)
6149 		break;
6150 	    d = retval;
6151 	    for (i = 0; i < len; ++i)
6152 	    {
6153 		l = utf_ptr2len_len(ptr + i, len - i);
6154 		if (l == 0)
6155 		    *d++ = NUL;
6156 		else if (l == 1)
6157 		{
6158 		    int l_w = utf8len_tab_zero[ptr[i]];
6159 
6160 		    if (l_w == 0)
6161 		    {
6162 			/* Illegal utf-8 byte cannot be converted */
6163 			vim_free(retval);
6164 			return NULL;
6165 		    }
6166 		    if (unconvlenp != NULL && l_w > len - i)
6167 		    {
6168 			/* Incomplete sequence at the end. */
6169 			*unconvlenp = len - i;
6170 			break;
6171 		    }
6172 		    *d++ = ptr[i];
6173 		}
6174 		else
6175 		{
6176 		    c = utf_ptr2char(ptr + i);
6177 		    if (vcp->vc_type == CONV_TO_LATIN9)
6178 			switch (c)
6179 			{
6180 			    case 0x20ac: c = 0xa4; break;   /* euro */
6181 			    case 0x0160: c = 0xa6; break;   /* S hat */
6182 			    case 0x0161: c = 0xa8; break;   /* S -hat */
6183 			    case 0x017d: c = 0xb4; break;   /* Z hat */
6184 			    case 0x017e: c = 0xb8; break;   /* Z -hat */
6185 			    case 0x0152: c = 0xbc; break;   /* OE */
6186 			    case 0x0153: c = 0xbd; break;   /* oe */
6187 			    case 0x0178: c = 0xbe; break;   /* Y */
6188 			    case 0xa4:
6189 			    case 0xa6:
6190 			    case 0xa8:
6191 			    case 0xb4:
6192 			    case 0xb8:
6193 			    case 0xbc:
6194 			    case 0xbd:
6195 			    case 0xbe: c = 0x100; break; /* not in latin9 */
6196 			}
6197 		    if (!utf_iscomposing(c))	/* skip composing chars */
6198 		    {
6199 			if (c < 0x100)
6200 			    *d++ = c;
6201 			else if (vcp->vc_fail)
6202 			{
6203 			    vim_free(retval);
6204 			    return NULL;
6205 			}
6206 			else
6207 			{
6208 			    *d++ = 0xbf;
6209 			    if (utf_char2cells(c) > 1)
6210 				*d++ = '?';
6211 			}
6212 		    }
6213 		    i += l - 1;
6214 		}
6215 	    }
6216 	    *d = NUL;
6217 	    if (lenp != NULL)
6218 		*lenp = (int)(d - retval);
6219 	    break;
6220 
6221 # ifdef MACOS_CONVERT
6222 	case CONV_MAC_LATIN1:
6223 	    retval = mac_string_convert(ptr, len, lenp, vcp->vc_fail,
6224 					'm', 'l', unconvlenp);
6225 	    break;
6226 
6227 	case CONV_LATIN1_MAC:
6228 	    retval = mac_string_convert(ptr, len, lenp, vcp->vc_fail,
6229 					'l', 'm', unconvlenp);
6230 	    break;
6231 
6232 	case CONV_MAC_UTF8:
6233 	    retval = mac_string_convert(ptr, len, lenp, vcp->vc_fail,
6234 					'm', 'u', unconvlenp);
6235 	    break;
6236 
6237 	case CONV_UTF8_MAC:
6238 	    retval = mac_string_convert(ptr, len, lenp, vcp->vc_fail,
6239 					'u', 'm', unconvlenp);
6240 	    break;
6241 # endif
6242 
6243 # ifdef USE_ICONV
6244 	case CONV_ICONV:	/* conversion with output_conv.vc_fd */
6245 	    retval = iconv_string(vcp, ptr, len, unconvlenp, lenp);
6246 	    break;
6247 # endif
6248 # ifdef WIN3264
6249 	case CONV_CODEPAGE:		/* codepage -> codepage */
6250 	{
6251 	    int		retlen;
6252 	    int		tmp_len;
6253 	    short_u	*tmp;
6254 
6255 	    /* 1. codepage/UTF-8  ->  ucs-2. */
6256 	    if (vcp->vc_cpfrom == 0)
6257 		tmp_len = utf8_to_utf16(ptr, len, NULL, NULL);
6258 	    else
6259 	    {
6260 		tmp_len = MultiByteToWideChar(vcp->vc_cpfrom,
6261 					unconvlenp ? MB_ERR_INVALID_CHARS : 0,
6262 					ptr, len, 0, 0);
6263 		if (tmp_len == 0
6264 			&& GetLastError() == ERROR_NO_UNICODE_TRANSLATION)
6265 		{
6266 		    if (lenp != NULL)
6267 			*lenp = 0;
6268 		    if (unconvlenp != NULL)
6269 			*unconvlenp = len;
6270 		    retval = alloc(1);
6271 		    if (retval)
6272 			retval[0] = NUL;
6273 		    return retval;
6274 		}
6275 	    }
6276 	    tmp = (short_u *)alloc(sizeof(short_u) * tmp_len);
6277 	    if (tmp == NULL)
6278 		break;
6279 	    if (vcp->vc_cpfrom == 0)
6280 		utf8_to_utf16(ptr, len, tmp, unconvlenp);
6281 	    else
6282 		MultiByteToWideChar(vcp->vc_cpfrom, 0, ptr, len, tmp, tmp_len);
6283 
6284 	    /* 2. ucs-2  ->  codepage/UTF-8. */
6285 	    if (vcp->vc_cpto == 0)
6286 		retlen = utf16_to_utf8(tmp, tmp_len, NULL);
6287 	    else
6288 		retlen = WideCharToMultiByte(vcp->vc_cpto, 0,
6289 						    tmp, tmp_len, 0, 0, 0, 0);
6290 	    retval = alloc(retlen + 1);
6291 	    if (retval != NULL)
6292 	    {
6293 		if (vcp->vc_cpto == 0)
6294 		    utf16_to_utf8(tmp, tmp_len, retval);
6295 		else
6296 		    WideCharToMultiByte(vcp->vc_cpto, 0,
6297 					  tmp, tmp_len, retval, retlen, 0, 0);
6298 		retval[retlen] = NUL;
6299 		if (lenp != NULL)
6300 		    *lenp = retlen;
6301 	    }
6302 	    vim_free(tmp);
6303 	    break;
6304 	}
6305 # endif
6306     }
6307 
6308     return retval;
6309 }
6310 #endif
6311