xref: /netbsd-src/external/bsd/nvi/dist/vi/v_txt.c (revision 926200353eed4c33d6b8b038ebfc47970dd04a39)
1 /*	$NetBSD: v_txt.c,v 1.7 2020/07/07 10:58:43 rin Exp $ */
2 /*-
3  * Copyright (c) 1993, 1994
4  *	The Regents of the University of California.  All rights reserved.
5  * Copyright (c) 1992, 1993, 1994, 1995, 1996
6  *	Keith Bostic.  All rights reserved.
7  *
8  * See the LICENSE file for redistribution information.
9  */
10 
11 #include "config.h"
12 
13 #include <sys/cdefs.h>
14 #if 0
15 #ifndef lint
16 static const char sccsid[] = "Id: v_txt.c,v 10.108 2003/07/18 21:27:42 skimo Exp  (Berkeley) Date: 2003/07/18 21:27:42 ";
17 #endif /* not lint */
18 #else
19 __RCSID("$NetBSD: v_txt.c,v 1.7 2020/07/07 10:58:43 rin Exp $");
20 #endif
21 
22 #include <sys/types.h>
23 #include <sys/queue.h>
24 #include <sys/stat.h>
25 #include <sys/time.h>
26 
27 #include <bitstring.h>
28 #include <ctype.h>
29 #include <errno.h>
30 #include <limits.h>
31 #include <stdio.h>
32 #include <stdlib.h>
33 #include <string.h>
34 #include <unistd.h>
35 
36 #include "../common/common.h"
37 #include "vi.h"
38 
39 static int	 txt_abbrev __P((SCR *, TEXT *, ARG_CHAR_T *, int, int *, int *));
40 static void	 txt_ai_resolve __P((SCR *, TEXT *, int *));
41 static TEXT	*txt_backup __P((SCR *, TEXTH *, TEXT *, u_int32_t *));
42 static int	 txt_dent __P((SCR *, TEXT *, int));
43 static int	 txt_emark __P((SCR *, TEXT *, size_t));
44 static void	 txt_err __P((SCR *, TEXTH *));
45 static int	 txt_fc __P((SCR *, TEXT *, int *));
46 static int	 txt_fc_col __P((SCR *, int, ARGS **));
47 static int	 txt_hex __P((SCR *, TEXT *));
48 static int	 txt_insch __P((SCR *, TEXT *, ARG_CHAR_T *, u_int));
49 static int	 txt_isrch __P((SCR *, VICMD *, TEXT *, u_int8_t *));
50 static int	 txt_map_end __P((SCR *));
51 static int	 txt_map_init __P((SCR *));
52 static int	 txt_margin __P((SCR *, TEXT *, TEXT *, int *, u_int32_t));
53 static void	 txt_nomorech __P((SCR *));
54 static void	 txt_Rresolve __P((SCR *, TEXTH *, TEXT *, const size_t));
55 static int	 txt_resolve __P((SCR *, TEXTH *, u_int32_t));
56 static int	 txt_showmatch __P((SCR *, TEXT *));
57 static void	 txt_unmap __P((SCR *, TEXT *, u_int32_t *));
58 
59 /* Cursor character (space is hard to track on the screen). */
60 #if defined(DEBUG) && 0
61 #undef	CH_CURSOR
62 #define	CH_CURSOR	'+'
63 #endif
64 
65 /*
66  * v_tcmd --
67  *	Fill a buffer from the terminal for vi.
68  *
69  * PUBLIC: int v_tcmd __P((SCR *, VICMD *, ARG_CHAR_T, u_int));
70  */
71 int
v_tcmd(SCR * sp,VICMD * vp,ARG_CHAR_T prompt,u_int flags)72 v_tcmd(SCR *sp, VICMD *vp, ARG_CHAR_T prompt, u_int flags)
73 {
74 	/* Normally, we end up where we started. */
75 	vp->m_final.lno = sp->lno;
76 	vp->m_final.cno = sp->cno;
77 
78 	/* Initialize the map. */
79 	if (txt_map_init(sp))
80 		return (1);
81 
82 	/* Move to the last line. */
83 	sp->lno = TMAP[0].lno;
84 	sp->cno = 0;
85 
86 	/* Don't update the modeline for now. */
87 	F_SET(sp, SC_TINPUT_INFO);
88 
89 	/* Set the input flags. */
90 	LF_SET(TXT_APPENDEOL |
91 	    TXT_CR | TXT_ESCAPE | TXT_INFOLINE | TXT_MAPINPUT);
92 	if (O_ISSET(sp, O_ALTWERASE))
93 		LF_SET(TXT_ALTWERASE);
94 	if (O_ISSET(sp, O_TTYWERASE))
95 		LF_SET(TXT_TTYWERASE);
96 
97 	/* Do the input thing. */
98 	if (v_txt(sp, vp, NULL, NULL, 0, prompt, 0, 1, flags))
99 		return (1);
100 
101 	/* Reenable the modeline updates. */
102 	F_CLR(sp, SC_TINPUT_INFO);
103 
104 	/* Clean up the map. */
105 	if (txt_map_end(sp))
106 		return (1);
107 
108 	if (IS_ONELINE(sp))
109 		F_SET(sp, SC_SCR_REDRAW);	/* XXX */
110 
111 	/* Set the cursor to the resulting position. */
112 	sp->lno = vp->m_final.lno;
113 	sp->cno = vp->m_final.cno;
114 
115 	return (0);
116 }
117 
118 /*
119  * txt_map_init
120  *	Initialize the screen map for colon command-line input.
121  */
122 static int
txt_map_init(SCR * sp)123 txt_map_init(SCR *sp)
124 {
125 	SMAP *esmp;
126 	VI_PRIVATE *vip;
127 
128 	vip = VIP(sp);
129 	if (!IS_ONELINE(sp)) {
130 		/*
131 		 * Fake like the user is doing input on the last line of the
132 		 * screen.  This makes all of the scrolling work correctly,
133 		 * and allows us the use of the vi text editing routines, not
134 		 * to mention practically infinite length ex commands.
135 		 *
136 		 * Save the current location.
137 		 */
138 		vip->sv_tm_lno = TMAP->lno;
139 		vip->sv_tm_soff = TMAP->soff;
140 		vip->sv_tm_coff = TMAP->coff;
141 		vip->sv_t_maxrows = sp->t_maxrows;
142 		vip->sv_t_minrows = sp->t_minrows;
143 		vip->sv_t_rows = sp->t_rows;
144 
145 		/*
146 		 * If it's a small screen, TMAP may be small for the screen.
147 		 * Fix it, filling in fake lines as we go.
148 		 */
149 		if (IS_SMALL(sp))
150 			for (esmp =
151 			    HMAP + (sp->t_maxrows - 1); TMAP < esmp; ++TMAP) {
152 				TMAP[1].lno = TMAP[0].lno + 1;
153 				TMAP[1].coff = HMAP->coff;
154 				TMAP[1].soff = 1;
155 			}
156 
157 		/* Build the fake entry. */
158 		TMAP[1].lno = TMAP[0].lno + 1;
159 		TMAP[1].soff = 1;
160 		TMAP[1].coff = 0;
161 		SMAP_FLUSH(&TMAP[1]);
162 		++TMAP;
163 
164 		/* Reset the screen information. */
165 		sp->t_rows = sp->t_minrows = ++sp->t_maxrows;
166 	}
167 	return (0);
168 }
169 
170 /*
171  * txt_map_end
172  *	Reset the screen map for colon command-line input.
173  */
174 static int
txt_map_end(SCR * sp)175 txt_map_end(SCR *sp)
176 {
177 	VI_PRIVATE *vip;
178 	size_t cnt;
179 
180 	vip = VIP(sp);
181 	if (!IS_ONELINE(sp)) {
182 		/* Restore the screen information. */
183 		sp->t_rows = vip->sv_t_rows;
184 		sp->t_minrows = vip->sv_t_minrows;
185 		sp->t_maxrows = vip->sv_t_maxrows;
186 
187 		/*
188 		 * If it's a small screen, TMAP may be wrong.  Clear any
189 		 * lines that might have been overwritten.
190 		 */
191 		if (IS_SMALL(sp)) {
192 			for (cnt = sp->t_rows; cnt <= sp->t_maxrows; ++cnt) {
193 				(void)sp->gp->scr_move(sp, cnt, 0);
194 				(void)sp->gp->scr_clrtoeol(sp);
195 			}
196 			TMAP = HMAP + (sp->t_rows - 1);
197 		} else
198 			--TMAP;
199 
200 		/*
201 		 * The map may be wrong if the user entered more than one
202 		 * (logical) line.  Fix it.  If the user entered a whole
203 		 * screen, this will be slow, but we probably don't care.
204 		 */
205 		if (!O_ISSET(sp, O_LEFTRIGHT))
206 			while (vip->sv_tm_lno != TMAP->lno ||
207 			    vip->sv_tm_soff != TMAP->soff)
208 				if (vs_sm_1down(sp))
209 					return (1);
210 	}
211 
212 	/*
213 	 * Invalidate the cursor and the line size cache, the line never
214 	 * really existed.  This fixes bugs where the user searches for
215 	 * the last line on the screen + 1 and the refresh routine thinks
216 	 * that's where we just were.
217 	 */
218 	VI_SCR_CFLUSH(vip);
219 	F_SET(vip, VIP_CUR_INVALID);
220 
221 	return (0);
222 }
223 
224 /*
225  * If doing input mapping on the colon command line, may need to unmap
226  * based on the command.
227  */
228 #define	UNMAP_TST							\
229 	FL_ISSET(ec_flags, EC_MAPINPUT) && LF_ISSET(TXT_INFOLINE)
230 
231 /*
232  * Internally, we maintain tp->lno and tp->cno, externally, everyone uses
233  * sp->lno and sp->cno.  Make them consistent as necessary.
234  */
235 #define	UPDATE_POSITION(sp, tp) {					\
236 	(sp)->lno = (tp)->lno;						\
237 	(sp)->cno = (tp)->cno;						\
238 }
239 
240 /*
241  * v_txt --
242  *	Vi text input.
243  *
244  * PUBLIC: int v_txt __P((SCR *, VICMD *, MARK *,
245  * PUBLIC:    const CHAR_T *, size_t, ARG_CHAR_T, db_recno_t, u_long, u_int32_t));
246  */
247 int
v_txt(SCR * sp,VICMD * vp,MARK * tm,const CHAR_T * lp,size_t len,ARG_CHAR_T prompt,db_recno_t ai_line,u_long rcount,u_int32_t flags)248 v_txt(SCR *sp, VICMD *vp, MARK *tm, const CHAR_T *lp, size_t len, ARG_CHAR_T prompt, db_recno_t ai_line, u_long rcount, u_int32_t flags)
249 
250 
251 	         		/* To MARK. */
252 	                 	/* Input line. */
253 	           		/* Input line length. */
254 	                  	/* Prompt to display. */
255 	                   	/* Line number to use for autoindent count. */
256 	              		/* Replay count. */
257 	                	/* TXT_* flags. */
258 {
259 	EVENT ev, *evp = NULL;	/* Current event. */
260 	EVENT fc;		/* File name completion event. */
261 	GS *gp;
262 	TEXT *ntp, *tp;		/* Input text structures. */
263 	TEXT ait;		/* Autoindent text structure. */
264 	TEXT wmt;		/* Wrapmargin text structure. */
265 	TEXTH *tiqh;
266 	VI_PRIVATE *vip;
267 	abb_t abb;		/* State of abbreviation checks. */
268 	carat_t carat;		/* State of the "[^0]^D" sequences. */
269 	quote_t quote;		/* State of quotation. */
270 	size_t owrite, insert;	/* Temporary copies of TEXT fields. */
271 	size_t margin;		/* Wrapmargin value. */
272 	size_t rcol;		/* 0-N: insert offset in the replay buffer. */
273 	size_t tcol;		/* Temporary column. */
274 	u_int32_t ec_flags;	/* Input mapping flags. */
275 #define	IS_RESTART	0x01	/* Reset the incremental search. */
276 #define	IS_RUNNING	0x02	/* Incremental search turned on. */
277 	u_int8_t is_flags;
278 	int abcnt, ab_turnoff;	/* Abbreviation character count, switch. */
279 	int filec_redraw;	/* Redraw after the file completion routine. */
280 	int hexcnt;		/* Hex character count. */
281 	int showmatch;		/* Showmatch set on this character. */
282 	int wm_set, wm_skip;	/* Wrapmargin happened, blank skip flags. */
283 	size_t max;
284 	int tmp;
285 	int nochange;
286 	CHAR_T *p;
287 
288 	gp = sp->gp;
289 	vip = VIP(sp);
290 	memset(&wmt, 0, sizeof(wmt));
291 
292 	/*
293 	 * Set the input flag, so tabs get displayed correctly
294 	 * and everyone knows that the text buffer is in use.
295 	 */
296 	F_SET(sp, SC_TINPUT);
297 
298 	/*
299 	 * Get one TEXT structure with some initial buffer space, reusing
300 	 * the last one if it's big enough.  (All TEXT bookkeeping fields
301 	 * default to 0 -- text_init() handles this.)  If changing a line,
302 	 * copy it into the TEXT buffer.
303 	 */
304 	tiqh = &sp->tiq;
305 	if (!TAILQ_EMPTY(tiqh)) {
306 		tp = TAILQ_FIRST(tiqh);
307 		if (TAILQ_NEXT(tp, q) != NULL || tp->lb_len < len + 32) {
308 			text_lfree(tiqh);
309 			goto newtp;
310 		}
311 		tp->ai = tp->insert = tp->offset = tp->owrite = 0;
312 		if (lp != NULL) {
313 			tp->len = len;
314 			BINC_RETW(sp, tp->lb, tp->lb_len, len);
315 			MEMMOVEW(tp->lb, lp, len);
316 		} else
317 			tp->len = 0;
318 	} else {
319 newtp:		if ((tp = text_init(sp, lp, len, len + 32)) == NULL)
320 			return (1);
321 		TAILQ_INSERT_HEAD(tiqh, tp, q);
322 	}
323 
324 	/* Set default termination condition. */
325 	tp->term = TERM_OK;
326 
327 	/* Set the starting line, column. */
328 	tp->lno = sp->lno;
329 	tp->cno = sp->cno;
330 
331 	/*
332 	 * Set the insert and overwrite counts.  If overwriting characters,
333 	 * do insertion afterward.  If not overwriting characters, assume
334 	 * doing insertion.  If change is to a mark, emphasize it with an
335 	 * CH_ENDMARK character.
336 	 */
337 	if (len) {
338 		if (LF_ISSET(TXT_OVERWRITE)) {
339 			tp->owrite = (tm->cno - tp->cno) + 1;
340 			tp->insert = (len - tm->cno) - 1;
341 		} else
342 			tp->insert = len - tp->cno;
343 
344 		if (LF_ISSET(TXT_EMARK) && txt_emark(sp, tp, tm->cno))
345 			return (1);
346 	}
347 
348 	/*
349 	 * Many of the special cases in text input are to handle autoindent
350 	 * support.  Somebody decided that it would be a good idea if "^^D"
351 	 * and "0^D" deleted all of the autoindented characters.  In an editor
352 	 * that takes single character input from the user, this beggars the
353 	 * imagination.  Note also, "^^D" resets the next lines' autoindent,
354 	 * but "0^D" doesn't.
355 	 *
356 	 * We assume that autoindent only happens on empty lines, so insert
357 	 * and overwrite will be zero.  If doing autoindent, figure out how
358 	 * much indentation we need and fill it in.  Update input column and
359 	 * screen cursor as necessary.
360 	 */
361 	if (LF_ISSET(TXT_AUTOINDENT) && ai_line != OOBLNO) {
362 		if (v_txt_auto(sp, ai_line, NULL, 0, tp))
363 			return (1);
364 		tp->cno = tp->ai;
365 	} else {
366 		/*
367 		 * The cc and S commands have a special feature -- leading
368 		 * <blank> characters are handled as autoindent characters.
369 		 * Beauty!
370 		 */
371 		if (LF_ISSET(TXT_AICHARS)) {
372 			tp->offset = 0;
373 			tp->ai = tp->cno;
374 		} else
375 			tp->offset = tp->cno;
376 	}
377 
378 	/* If getting a command buffer from the user, there may be a prompt. */
379 	if (LF_ISSET(TXT_PROMPT)) {
380 		tp->lb[tp->cno++] = prompt;
381 		++tp->len;
382 		++tp->offset;
383 	}
384 
385 	/*
386 	 * If appending after the end-of-line, add a space into the buffer
387 	 * and move the cursor right.  This space is inserted, i.e. pushed
388 	 * along, and then deleted when the line is resolved.  Assumes that
389 	 * the cursor is already positioned at the end of the line.  This
390 	 * avoids the nastiness of having the cursor reside on a magical
391 	 * column, i.e. a column that doesn't really exist.  The only down
392 	 * side is that we may wrap lines or scroll the screen before it's
393 	 * strictly necessary.  Not a big deal.
394 	 */
395 	if (LF_ISSET(TXT_APPENDEOL)) {
396 		tp->lb[tp->cno] = CH_CURSOR;
397 		++tp->len;
398 		++tp->insert;
399 		(void)vs_change(sp, tp->lno, LINE_RESET);
400 	}
401 
402 	/*
403 	 * Historic practice is that the wrapmargin value was a distance
404 	 * from the RIGHT-HAND margin, not the left.  It's more useful to
405 	 * us as a distance from the left-hand margin, i.e. the same as
406 	 * the wraplen value.  The wrapmargin option is historic practice.
407 	 * Nvi added the wraplen option so that it would be possible to
408 	 * edit files with consistent margins without knowing the number of
409 	 * columns in the window.
410 	 *
411 	 * XXX
412 	 * Setting margin causes a significant performance hit.  Normally
413 	 * we don't update the screen if there are keys waiting, but we
414 	 * have to if margin is set, otherwise the screen routines don't
415 	 * know where the cursor is.
416 	 *
417 	 * !!!
418 	 * Abbreviated keys were affected by the wrapmargin option in the
419 	 * historic 4BSD vi.  Mapped keys were usually, but sometimes not.
420 	 * See the comment in vi/v_text():set_txt_std for more information.
421 	 *
422 	 * !!!
423 	 * One more special case.  If an inserted <blank> character causes
424 	 * wrapmargin to split the line, the next user entered character is
425 	 * discarded if it's a <space> character.
426 	 */
427 	wm_set = wm_skip = 0;
428 	if (LF_ISSET(TXT_WRAPMARGIN))
429 		if ((margin = O_VAL(sp, O_WRAPMARGIN)) != 0)
430 			margin = sp->cols - margin;
431 		else
432 			margin = O_VAL(sp, O_WRAPLEN);
433 	else
434 		margin = 0;
435 
436 	/* Initialize abbreviation checks. */
437 	abcnt = ab_turnoff = 0;
438 	abb = F_ISSET(gp, G_ABBREV) &&
439 	    LF_ISSET(TXT_MAPINPUT) ? AB_INWORD : AB_NOTSET;
440 
441 	/*
442 	 * Set up the dot command.  Dot commands are done by saving the actual
443 	 * characters and then reevaluating them so that things like wrapmargin
444 	 * can change between the insert and the replay.
445 	 *
446 	 * !!!
447 	 * Historically, vi did not remap or reabbreviate replayed input.  (It
448 	 * did beep at you if you changed an abbreviation and then replayed the
449 	 * input.  We're not that compatible.)  We don't have to do anything to
450 	 * avoid remapping, as we're not getting characters from the terminal
451 	 * routines.  Turn the abbreviation check off.
452 	 *
453 	 * XXX
454 	 * It would be nice if we could swallow backspaces and such, but it's
455 	 * not all that easy to do.  What we can do is turn off the common
456 	 * error messages during the replay.  Otherwise, when the user enters
457 	 * an illegal command, e.g., "Ia<erase><erase><erase><erase>b<escape>",
458 	 * and then does a '.', they get a list of error messages after command
459 	 * completion.
460 	 */
461 	rcol = 0;
462 	if (LF_ISSET(TXT_REPLAY)) {
463 		abb = AB_NOTSET;
464 		LF_CLR(TXT_RECORD);
465 	}
466 
467 	/* Other text input mode setup. */
468 	quote = Q_NOTSET;
469 	carat = C_NOTSET;
470 	nochange = 0;
471 	FL_INIT(is_flags,
472 	    LF_ISSET(TXT_SEARCHINCR) ? IS_RESTART | IS_RUNNING : 0);
473 	filec_redraw = hexcnt = showmatch = 0;
474 
475 	/* Initialize input flags. */
476 	ec_flags = LF_ISSET(TXT_MAPINPUT) ? EC_MAPINPUT : 0;
477 
478 	/* Refresh the screen. */
479 	UPDATE_POSITION(sp, tp);
480 	if (vs_refresh(sp, 1))
481 		return (1);
482 
483 	/* If it's dot, just do it now. */
484 	if (F_ISSET(vp, VC_ISDOT))
485 		goto replay;
486 
487 	/* Get an event. */
488 	evp = &ev;
489 next:	if (v_event_get(sp, evp, 0, ec_flags))
490 		return (1);
491 
492 	/*
493 	 * If file completion overwrote part of the screen and nothing else has
494 	 * been displayed, clean up.  We don't do this as part of the normal
495 	 * message resolution because we know the user is on the colon command
496 	 * line and there's no reason to enter explicit characters to continue.
497 	 */
498 	if (filec_redraw && !F_ISSET(sp, SC_SCR_EXWROTE)) {
499 		filec_redraw = 0;
500 
501 		fc.e_event = E_REPAINT;
502 		fc.e_flno = vip->totalcount >=
503 		    sp->rows ? 1 : sp->rows - vip->totalcount;
504 		fc.e_tlno = sp->rows;
505 		vip->linecount = vip->lcontinue = vip->totalcount = 0;
506 		(void)v_erepaint(sp, &fc);
507 		(void)vs_refresh(sp, 1);
508 	}
509 
510 	/* Deal with all non-character events. */
511 	switch (evp->e_event) {
512 	case E_CHARACTER:
513 		break;
514 	case E_ERR:
515 	case E_EOF:
516 		F_SET(sp, SC_EXIT_FORCE);
517 		return (1);
518 	case E_INTERRUPT:
519 		/*
520 		 * !!!
521 		 * Historically, <interrupt> exited the user from text input
522 		 * mode or cancelled a colon command, and returned to command
523 		 * mode.  It also beeped the terminal, but that seems a bit
524 		 * excessive.
525 		 */
526 		goto k_escape;
527 	case E_REPAINT:
528 		if (v_erepaint(sp, &ev))
529 			return (1);
530 		goto next;
531 	case E_WRESIZE:
532 		/* <resize> interrupts the input mode. */
533 #ifdef IMCTRL
534 		sp->gp->scr_imctrl(sp, IMCTRL_OFF);
535 #endif
536 		v_emsg(sp, NULL, VIM_WRESIZE);
537 		goto k_escape;
538 	default:
539 		v_event_err(sp, evp);
540 		goto k_escape;
541 	}
542 
543 	/*
544 	 * !!!
545 	 * If the first character of the input is a nul, replay the previous
546 	 * input.  (Historically, it's okay to replay non-existent input.)
547 	 * This was not documented as far as I know, and is a great test of vi
548 	 * clones.
549 	 */
550 	if (rcol == 0 && !LF_ISSET(TXT_REPLAY) && evp->e_c == '\0') {
551 		if (vip->rep == NULL)
552 			goto done;
553 
554 		abb = AB_NOTSET;
555 		LF_CLR(TXT_RECORD);
556 		LF_SET(TXT_REPLAY);
557 		goto replay;
558 	}
559 
560 	/*
561 	 * File name completion and colon command-line editing.   We don't
562 	 * have enough meta characters, so we expect people to overload
563 	 * them.  If the two characters are the same, then we do file name
564 	 * completion if the cursor is past the first column, and do colon
565 	 * command-line editing if it's not.
566 	 */
567 	if (quote == Q_NOTSET) {
568 		int L__cedit, L__filec;
569 
570 		L__cedit = L__filec = 0;
571 		if (LF_ISSET(TXT_CEDIT) && O_STR(sp, O_CEDIT) != NULL &&
572 		    O_STR(sp, O_CEDIT)[0] == evp->e_c)
573 			L__cedit = 1;
574 		if (LF_ISSET(TXT_FILEC) && O_STR(sp, O_FILEC) != NULL &&
575 		    O_STR(sp, O_FILEC)[0] == evp->e_c)
576 			L__filec = 1;
577 		if (L__cedit == 1 && (L__filec == 0 || tp->cno == tp->offset)) {
578 			tp->term = TERM_CEDIT;
579 			goto k_escape;
580 		}
581 		if (L__filec == 1) {
582 			if (txt_fc(sp, tp, &filec_redraw))
583 				goto err;
584 			goto resolve;
585 		}
586 	}
587 
588 	/* Abbreviation overflow check.  See comment in txt_abbrev(). */
589 #define	MAX_ABBREVIATION_EXPANSION	256
590 	if (FL_ISSET(evp->e_flags, CH_ABBREVIATED)) {
591 		if (++abcnt > MAX_ABBREVIATION_EXPANSION) {
592 			if (v_event_flush(sp, CH_ABBREVIATED))
593 				msgq(sp, M_ERR,
594 "191|Abbreviation exceeded expansion limit: characters discarded");
595 			abcnt = 0;
596 			if (LF_ISSET(TXT_REPLAY))
597 				goto done;
598 			goto resolve;
599 		}
600 	} else
601 		abcnt = 0;
602 
603 	/* Check to see if the character fits into the replay buffers. */
604 	if (LF_ISSET(TXT_RECORD)) {
605 		BINC_GOTO(sp, EVENT, vip->rep,
606 		    vip->rep_len, (rcol + 1) * sizeof(EVENT));
607 		vip->rep[rcol++] = *evp;
608 	}
609 
610 replay:	if (LF_ISSET(TXT_REPLAY)) {
611 		if (rcol == vip->rep_cnt)
612 			goto k_escape;
613 		evp = vip->rep + rcol++;
614 	}
615 
616 	/* Wrapmargin check for leading space. */
617 	if (wm_skip) {
618 		wm_skip = 0;
619 		if (evp->e_c == ' ')
620 			goto resolve;
621 	}
622 
623 	/* If quoted by someone else, simply insert the character. */
624 	if (FL_ISSET(evp->e_flags, CH_QUOTED))
625 		goto insq_ch;
626 
627 	/*
628 	 * !!!
629 	 * If this character was quoted by a K_VLNEXT or a backslash, replace
630 	 * the placeholder (a carat or a backslash) with the new character.
631 	 * If it was quoted by a K_VLNEXT, we've already adjusted the cursor
632 	 * because it has to appear on top of the placeholder character.  If
633 	 * it was quoted by a backslash, adjust the cursor now, the cursor
634 	 * doesn't appear on top of it.  Historic practice in both cases.
635 	 *
636 	 * Skip tests for abbreviations; ":ab xa XA" followed by "ixa^V<space>"
637 	 * doesn't perform an abbreviation.  Special case, ^V^J (not ^V^M) is
638 	 * the same as ^J, historically.
639 	 */
640 	if (quote == Q_BTHIS || quote == Q_VTHIS) {
641 		FL_CLR(ec_flags, EC_QUOTED);
642 		if (LF_ISSET(TXT_MAPINPUT))
643 			FL_SET(ec_flags, EC_MAPINPUT);
644 
645 		if (quote == Q_BTHIS &&
646 		    (evp->e_value == K_VERASE || evp->e_value == K_VKILL)) {
647 			quote = Q_NOTSET;
648 			--tp->cno;
649 			++tp->owrite;
650 			goto insl_ch;
651 		}
652 		if (quote == Q_VTHIS && evp->e_value != K_NL) {
653 			quote = Q_NOTSET;
654 			goto insl_ch;
655 		}
656 		quote = Q_NOTSET;
657 	}
658 
659 	/*
660 	 * !!!
661 	 * Translate "<CH_HEX>[isxdigit()]*" to a character with a hex value:
662 	 * this test delimits the value by any non-hex character.  Offset by
663 	 * one, we use 0 to mean that we've found <CH_HEX>.
664 	 */
665 	if (hexcnt > 1 && !ISXDIGIT(evp->e_c)) {
666 		hexcnt = 0;
667 		if (txt_hex(sp, tp))
668 			goto err;
669 	}
670 
671 	switch (evp->e_value) {
672 	case K_CR:				/* Carriage return. */
673 	case K_NL:				/* New line. */
674 		/* Return in script windows and the command line. */
675 k_cr:		if (LF_ISSET(TXT_CR)) {
676 			static CHAR_T cr[] = { '\r', 0 };
677 			/*
678 			 * If this was a map, we may have not displayed
679 			 * the line.  Display it, just in case.
680 			 *
681 			 * If a script window and not the colon line,
682 			 * push a <cr> so it gets executed.
683 			 */
684 			if (LF_ISSET(TXT_INFOLINE)) {
685 				if (vs_change(sp, tp->lno, LINE_RESET))
686 					goto err;
687 			} else if (F_ISSET(sp, SC_SCRIPT))
688 				(void)v_event_push(sp, NULL, cr, 1, CH_NOMAP);
689 
690 			/* Set term condition: if empty. */
691 			if (tp->cno <= tp->offset)
692 				tp->term = TERM_CR;
693 			/*
694 			 * Set term condition: if searching incrementally and
695 			 * the user entered a pattern, return a completed
696 			 * search, regardless if the entire pattern was found.
697 			 */
698 			if (FL_ISSET(is_flags, IS_RUNNING) &&
699 			    tp->cno >= tp->offset + 1)
700 				tp->term = TERM_SEARCH;
701 
702 			goto k_escape;
703 		}
704 
705 #define	LINE_RESOLVE {							\
706 		/*							\
707 		 * Handle abbreviations.  If there was one, discard the	\
708 		 * replay characters.					\
709 		 */							\
710 		if (abb == AB_INWORD &&					\
711 		    !LF_ISSET(TXT_REPLAY) && F_ISSET(gp, G_ABBREV)) {	\
712 			if (txt_abbrev(sp, tp, &evp->e_c,		\
713 			    LF_ISSET(TXT_INFOLINE), &tmp,		\
714 			    &ab_turnoff))				\
715 				goto err;				\
716 			if (tmp) {					\
717 				if (LF_ISSET(TXT_RECORD))		\
718 					rcol -= tmp + 1;		\
719 				goto resolve;				\
720 			}						\
721 		}							\
722 		if (abb != AB_NOTSET)					\
723 			abb = AB_NOTWORD;				\
724 		if (UNMAP_TST)						\
725 			txt_unmap(sp, tp, &ec_flags);			\
726 		/*							\
727 		 * Delete any appended cursor.  It's possible to get in	\
728 		 * situations where TXT_APPENDEOL is set but tp->insert	\
729 		 * is 0 when using the R command and all the characters	\
730 		 * are tp->owrite characters.				\
731 		 */							\
732 		if (LF_ISSET(TXT_APPENDEOL) && tp->insert > 0) {	\
733 			--tp->len;					\
734 			--tp->insert;					\
735 		}							\
736 }
737 		LINE_RESOLVE;
738 
739 		/*
740 		 * Save the current line information for restoration in
741 		 * txt_backup(), and set the line final length.
742 		 */
743 		tp->sv_len = tp->len;
744 		tp->sv_cno = tp->cno;
745 		tp->len = tp->cno;
746 
747 		/* Update the old line. */
748 		if (vs_change(sp, tp->lno, LINE_RESET))
749 			goto err;
750 
751 		/*
752 		 * Historic practice, when the autoindent edit option was set,
753 		 * was to delete <blank> characters following the inserted
754 		 * newline.  This affected the 'R', 'c', and 's' commands; 'c'
755 		 * and 's' retained the insert characters only, 'R' moved the
756 		 * overwrite and insert characters into the next TEXT structure.
757 		 * We keep track of the number of characters erased for the 'R'
758 		 * command so that the final resolution of the line is correct.
759 		 */
760 		tp->R_erase = 0;
761 		owrite = tp->owrite;
762 		insert = tp->insert;
763 		if (LF_ISSET(TXT_REPLACE) && owrite != 0) {
764 			for (p = tp->lb + tp->cno; owrite > 0 && ISBLANK((UCHAR_T)*p);
765 			    ++p, --owrite, ++tp->R_erase);
766 			if (owrite == 0)
767 				for (; insert > 0 && ISBLANK((UCHAR_T)*p);
768 				    ++p, ++tp->R_erase, --insert);
769 		} else {
770 			p = tp->lb + tp->cno + owrite;
771 			if (O_ISSET(sp, O_AUTOINDENT))
772 				for (; insert > 0 &&
773 				    ISBLANK((UCHAR_T)*p); ++p, --insert);
774 			owrite = 0;
775 		}
776 
777 		/*
778 		 * !!!
779 		 * Create a new line and insert the new TEXT into the queue.
780 		 * DON'T insert until the old line has been updated, or the
781 		 * inserted line count in line.c:db_get() will be wrong.
782 		 */
783 		if ((ntp = text_init(sp, p,
784 		    insert + owrite, insert + owrite + 32)) == NULL)
785 			goto err;
786 		TAILQ_INSERT_TAIL(&sp->tiq, ntp, q);
787 
788 		/* Set up bookkeeping for the new line. */
789 		ntp->insert = insert;
790 		ntp->owrite = owrite;
791 		ntp->lno = tp->lno + 1;
792 
793 		/*
794 		 * Reset the autoindent line value.  0^D keeps the autoindent
795 		 * line from changing, ^D changes the level, even if there were
796 		 * no characters in the old line.  Note, if using the current
797 		 * tp structure, use the cursor as the length, the autoindent
798 		 * characters may have been erased.
799 		 */
800 		if (LF_ISSET(TXT_AUTOINDENT)) {
801 			if (nochange) {
802 				nochange = 0;
803 				if (v_txt_auto(sp, OOBLNO, &ait, ait.ai, ntp))
804 					goto err;
805 				FREE_SPACEW(sp, ait.lb, ait.lb_len);
806 			} else
807 				if (v_txt_auto(sp, OOBLNO, tp, tp->cno, ntp))
808 					goto err;
809 			carat = C_NOTSET;
810 		}
811 
812 		/* Reset the cursor. */
813 		ntp->cno = ntp->ai;
814 
815 		/*
816 		 * If we're here because wrapmargin was set and we've broken a
817 		 * line, there may be additional information (i.e. the start of
818 		 * a line) in the wmt structure.
819 		 */
820 		if (wm_set) {
821 			if (wmt.offset != 0 ||
822 			    wmt.owrite != 0 || wmt.insert != 0) {
823 #define	WMTSPACE	wmt.offset + wmt.owrite + wmt.insert
824 				BINC_GOTOW(sp, ntp->lb,
825 				    ntp->lb_len, ntp->len + WMTSPACE + 32);
826 				MEMMOVEW(ntp->lb + ntp->cno, wmt.lb, WMTSPACE);
827 				ntp->len += WMTSPACE;
828 				ntp->cno += wmt.offset;
829 				ntp->owrite = wmt.owrite;
830 				ntp->insert = wmt.insert;
831 			}
832 			wm_set = 0;
833 		}
834 
835 		/* New lines are TXT_APPENDEOL. */
836 		if (ntp->owrite == 0 && ntp->insert == 0) {
837 			BINC_GOTOW(sp, ntp->lb, ntp->lb_len, ntp->len + 1);
838 			LF_SET(TXT_APPENDEOL);
839 			ntp->lb[ntp->cno] = CH_CURSOR;
840 			++ntp->insert;
841 			++ntp->len;
842 		}
843 
844 		/* Swap old and new TEXT's, and update the new line. */
845 		tp = ntp;
846 		if (vs_change(sp, tp->lno, LINE_INSERT))
847 			goto err;
848 
849 		goto resolve;
850 	case K_ESCAPE:				/* Escape. */
851 		if (!LF_ISSET(TXT_ESCAPE))
852 			goto ins_ch;
853 
854 		/* If we have a count, start replaying the input. */
855 		if (rcount > 1) {
856 			--rcount;
857 
858 			vip->rep_cnt = rcol;
859 			rcol = 0;
860 			abb = AB_NOTSET;
861 			LF_CLR(TXT_RECORD);
862 			LF_SET(TXT_REPLAY);
863 
864 			/*
865 			 * Some commands (e.g. 'o') need a <newline> for each
866 			 * repetition.
867 			 */
868 			if (LF_ISSET(TXT_ADDNEWLINE))
869 				goto k_cr;
870 
871 			/*
872 			 * The R command turns into the 'a' command after the
873 			 * first repetition.
874 			 */
875 			if (LF_ISSET(TXT_REPLACE)) {
876 				tp->insert = tp->owrite;
877 				tp->owrite = 0;
878 				LF_CLR(TXT_REPLACE);
879 			}
880 			goto replay;
881 		}
882 
883 		/* Set term condition: if empty. */
884 		if (tp->cno <= tp->offset)
885 			tp->term = TERM_ESC;
886 		/*
887 		 * Set term condition: if searching incrementally and the user
888 		 * entered a pattern, return a completed search, regardless if
889 		 * the entire pattern was found.
890 		 */
891 		if (FL_ISSET(is_flags, IS_RUNNING) && tp->cno >= tp->offset + 1)
892 			tp->term = TERM_SEARCH;
893 
894 k_escape:	LINE_RESOLVE;
895 
896 		/*
897 		 * Clean up for the 'R' command, restoring overwrite
898 		 * characters, and making them into insert characters.
899 		 */
900 		if (LF_ISSET(TXT_REPLACE))
901 			txt_Rresolve(sp, &sp->tiq, tp, len);
902 
903 		/*
904 		 * If there are any overwrite characters, copy down
905 		 * any insert characters, and decrement the length.
906 		 */
907 		if (tp->owrite) {
908 			if (tp->insert)
909 				MEMMOVEW(tp->lb + tp->cno,
910 				    tp->lb + tp->cno + tp->owrite, tp->insert);
911 			tp->len -= tp->owrite;
912 		}
913 
914 		/*
915 		 * Optionally resolve the lines into the file.  If not
916 		 * resolving the lines into the file, end the line with
917 		 * a nul.  If the line is empty, then set the length to
918 		 * 0, the termination condition has already been set.
919 		 *
920 		 * XXX
921 		 * This is wrong, should pass back a length.
922 		 */
923 		if (LF_ISSET(TXT_RESOLVE)) {
924 			if (txt_resolve(sp, &sp->tiq, flags))
925 				goto err;
926 		} else {
927 			BINC_GOTOW(sp, tp->lb, tp->lb_len, tp->len + 1);
928 			tp->lb[tp->len] = '\0';
929 		}
930 
931 		/*
932 		 * Set the return cursor position to rest on the last
933 		 * inserted character.
934 		 */
935 		if (tp->cno != 0)
936 			--tp->cno;
937 
938 		/* Update the last line. */
939 		if (vs_change(sp, tp->lno, LINE_RESET))
940 			return (1);
941 		goto done;
942 	case K_CARAT:			/* Delete autoindent chars. */
943 		if (tp->cno <= tp->ai && LF_ISSET(TXT_AUTOINDENT))
944 			carat = C_CARATSET;
945 		goto ins_ch;
946 	case K_ZERO:			/* Delete autoindent chars. */
947 		if (tp->cno <= tp->ai && LF_ISSET(TXT_AUTOINDENT))
948 			carat = C_ZEROSET;
949 		goto ins_ch;
950 	case K_CNTRLD:			/* Delete autoindent char. */
951 		/*
952 		 * If in the first column or no characters to erase, ignore
953 		 * the ^D (this matches historic practice).  If not doing
954 		 * autoindent or already inserted non-ai characters, it's a
955 		 * literal.  The latter test is done in the switch, as the
956 		 * CARAT forms are N + 1, not N.
957 		 */
958 		if (!LF_ISSET(TXT_AUTOINDENT))
959 			goto ins_ch;
960 		if (tp->cno == 0)
961 			goto resolve;
962 
963 		switch (carat) {
964 		case C_CARATSET:	/* ^^D */
965 			if (tp->ai == 0 || tp->cno != tp->ai + tp->offset + 1)
966 				goto ins_ch;
967 
968 			/* Save the ai string for later. */
969 			ait.lb = NULL;
970 			ait.lb_len = 0;
971 			BINC_GOTOW(sp, ait.lb, ait.lb_len, tp->ai);
972 			MEMMOVEW(ait.lb, tp->lb, tp->ai);
973 			ait.ai = ait.len = tp->ai;
974 
975 			carat = C_NOTSET;
976 			nochange = 1;
977 			goto leftmargin;
978 		case C_ZEROSET:		/* 0^D */
979 			if (tp->ai == 0 || tp->cno != tp->ai + tp->offset + 1)
980 				goto ins_ch;
981 
982 			carat = C_NOTSET;
983 leftmargin:		tp->lb[tp->cno - 1] = ' ';
984 			tp->owrite += tp->cno - tp->offset;
985 			tp->cno = tp->offset;
986 			break;
987 		case C_NOTSET:		/* ^D */
988 			if (tp->ai == 0 || tp->cno != tp->ai + tp->offset)
989 				goto ins_ch;
990 
991 			(void)txt_dent(sp, tp, 0);
992 			break;
993 		default:
994 			abort();
995 		}
996 		break;
997 	case K_VERASE:			/* Erase the last character. */
998 		/* If can erase over the prompt, return. */
999 		if (tp->cno <= tp->offset && LF_ISSET(TXT_BS)) {
1000 			tp->term = TERM_BS;
1001 			goto done;
1002 		}
1003 
1004 		/*
1005 		 * If at the beginning of the line, try and drop back to a
1006 		 * previously inserted line.
1007 		 */
1008 		if (tp->cno == 0) {
1009 			if ((ntp =
1010 			    txt_backup(sp, &sp->tiq, tp, &flags)) == NULL)
1011 				goto err;
1012 			tp = ntp;
1013 			break;
1014 		}
1015 
1016 		/* If nothing to erase, bell the user. */
1017 		if (tp->cno <= tp->offset) {
1018 			if (!LF_ISSET(TXT_REPLAY))
1019 				txt_nomorech(sp);
1020 			break;
1021 		}
1022 
1023 		/* Drop back one character. */
1024 		--tp->cno;
1025 
1026 		/*
1027 		 * Historically, vi didn't replace the erased characters with
1028 		 * <blank>s, presumably because it's easier to fix a minor
1029 		 * typing mistake and continue on if the previous letters are
1030 		 * already there.  This is a problem for incremental searching,
1031 		 * because the user can no longer tell where they are in the
1032 		 * colon command line because the cursor is at the last search
1033 		 * point in the screen.  So, if incrementally searching, erase
1034 		 * the erased characters from the screen.
1035 		 */
1036 		if (FL_ISSET(is_flags, IS_RUNNING))
1037 			tp->lb[tp->cno] = ' ';
1038 
1039 		/*
1040 		 * Increment overwrite, decrement ai if deleted.
1041 		 *
1042 		 * !!!
1043 		 * Historic vi did not permit users to use erase characters
1044 		 * to delete autoindent characters.  We do.  Eat hot death,
1045 		 * POSIX.
1046 		 */
1047 		++tp->owrite;
1048 		if (tp->cno < tp->ai)
1049 			--tp->ai;
1050 
1051 		/* Reset if we deleted an incremental search character. */
1052 		if (FL_ISSET(is_flags, IS_RUNNING))
1053 			FL_SET(is_flags, IS_RESTART);
1054 		break;
1055 	case K_VWERASE:			/* Skip back one word. */
1056 		/*
1057 		 * If at the beginning of the line, try and drop back to a
1058 		 * previously inserted line.
1059 		 */
1060 		if (tp->cno == 0) {
1061 			if ((ntp =
1062 			    txt_backup(sp, &sp->tiq, tp, &flags)) == NULL)
1063 				goto err;
1064 			tp = ntp;
1065 		}
1066 
1067 		/*
1068 		 * If at offset, nothing to erase so bell the user.
1069 		 */
1070 		if (tp->cno <= tp->offset) {
1071 			if (!LF_ISSET(TXT_REPLAY))
1072 				txt_nomorech(sp);
1073 			break;
1074 		}
1075 
1076 		/*
1077 		 * The first werase goes back to any autoindent column and the
1078 		 * second werase goes back to the offset.
1079 		 *
1080 		 * !!!
1081 		 * Historic vi did not permit users to use erase characters to
1082 		 * delete autoindent characters.
1083 		 */
1084 		if (tp->ai && tp->cno > tp->ai)
1085 			max = tp->ai;
1086 		else {
1087 			tp->ai = 0;
1088 			max = tp->offset;
1089 		}
1090 
1091 		/* Skip over trailing space characters. */
1092 		while (tp->cno > max && ISBLANK((UCHAR_T)tp->lb[tp->cno - 1])) {
1093 			--tp->cno;
1094 			++tp->owrite;
1095 		}
1096 		if (tp->cno == max)
1097 			break;
1098 		/*
1099 		 * There are three types of word erase found on UNIX systems.
1100 		 * They can be identified by how the string /a/b/c is treated
1101 		 * -- as 1, 3, or 6 words.  Historic vi had two classes of
1102 		 * characters, and strings were delimited by them and
1103 		 * <blank>'s, so, 6 words.  The historic tty interface used
1104 		 * <blank>'s to delimit strings, so, 1 word.  The algorithm
1105 		 * offered in the 4.4BSD tty interface (as stty altwerase)
1106 		 * treats it as 3 words -- there are two classes of
1107 		 * characters, and strings are delimited by them and
1108 		 * <blank>'s.  The difference is that the type of the first
1109 		 * erased character erased is ignored, which is exactly right
1110 		 * when erasing pathname components.  The edit options
1111 		 * TXT_ALTWERASE and TXT_TTYWERASE specify the 4.4BSD tty
1112 		 * interface and the historic tty driver behavior,
1113 		 * respectively, and the default is the same as the historic
1114 		 * vi behavior.
1115 		 *
1116 		 * Overwrite erased characters if doing incremental search;
1117 		 * see comment above.
1118 		 */
1119 		if (LF_ISSET(TXT_TTYWERASE))
1120 			while (tp->cno > max) {
1121 				if (ISBLANK((UCHAR_T)tp->lb[tp->cno - 1]))
1122 					break;
1123 				--tp->cno;
1124 				++tp->owrite;
1125 				if (FL_ISSET(is_flags, IS_RUNNING))
1126 					tp->lb[tp->cno] = ' ';
1127 			}
1128 		else {
1129 			if (LF_ISSET(TXT_ALTWERASE)) {
1130 				--tp->cno; /* No worry for out of bounds. */
1131 				++tp->owrite;
1132 				if (FL_ISSET(is_flags, IS_RUNNING))
1133 					tp->lb[tp->cno] = ' ';
1134 			}
1135 			if (tp->cno > max)
1136 				tmp = inword((UCHAR_T)tp->lb[tp->cno - 1]);
1137 			while (tp->cno > max) {
1138 				if (tmp != inword((UCHAR_T)tp->lb[tp->cno - 1])
1139 				    || ISBLANK((UCHAR_T)tp->lb[tp->cno - 1]))
1140 					break;
1141 				--tp->cno;
1142 				++tp->owrite;
1143 				if (FL_ISSET(is_flags, IS_RUNNING))
1144 					tp->lb[tp->cno] = ' ';
1145 			}
1146 		}
1147 
1148 		/* Reset if we deleted an incremental search character. */
1149 		if (FL_ISSET(is_flags, IS_RUNNING))
1150 			FL_SET(is_flags, IS_RESTART);
1151 		break;
1152 	case K_VKILL:			/* Restart this line. */
1153 		/*
1154 		 * !!!
1155 		 * If at the beginning of the line, try and drop back to a
1156 		 * previously inserted line.  Historic vi did not permit
1157 		 * users to go back to previous lines.
1158 		 */
1159 		if (tp->cno == 0) {
1160 			if ((ntp =
1161 			    txt_backup(sp, &sp->tiq, tp, &flags)) == NULL)
1162 				goto err;
1163 			tp = ntp;
1164 		}
1165 
1166 		/* If at offset, nothing to erase so bell the user. */
1167 		if (tp->cno <= tp->offset) {
1168 			if (!LF_ISSET(TXT_REPLAY))
1169 				txt_nomorech(sp);
1170 			break;
1171 		}
1172 
1173 		/*
1174 		 * First kill goes back to any autoindent and second kill goes
1175 		 * back to the offset.
1176 		 *
1177 		 * !!!
1178 		 * Historic vi did not permit users to use erase characters to
1179 		 * delete autoindent characters.
1180 		 */
1181 		if (tp->ai && tp->cno > tp->ai)
1182 			max = tp->ai;
1183 		else {
1184 			tp->ai = 0;
1185 			max = tp->offset;
1186 		}
1187 		tp->owrite += tp->cno - max;
1188 
1189 		/*
1190 		 * Overwrite erased characters if doing incremental search;
1191 		 * see comment above.
1192 		 */
1193 		if (FL_ISSET(is_flags, IS_RUNNING))
1194 			do {
1195 				tp->lb[--tp->cno] = ' ';
1196 			} while (tp->cno > max);
1197 		else
1198 			tp->cno = max;
1199 
1200 		/* Reset if we deleted an incremental search character. */
1201 		if (FL_ISSET(is_flags, IS_RUNNING))
1202 			FL_SET(is_flags, IS_RESTART);
1203 		break;
1204 	case K_CNTRLT:			/* Add autoindent characters. */
1205 		if (!LF_ISSET(TXT_CNTRLT))
1206 			goto ins_ch;
1207 		if (txt_dent(sp, tp, 1))
1208 			goto err;
1209 		goto ebuf_chk;
1210 	case K_RIGHTBRACE:
1211 	case K_RIGHTPAREN:
1212 		if (LF_ISSET(TXT_SHOWMATCH))
1213 			showmatch = 1;
1214 		goto ins_ch;
1215 	case K_BACKSLASH:		/* Quote next erase/kill. */
1216 		/*
1217 		 * !!!
1218 		 * Historic vi tried to make abbreviations after a backslash
1219 		 * escape work.  If you did ":ab x y", and inserted "x\^H",
1220 		 * (assuming the erase character was ^H) you got "x^H", and
1221 		 * no abbreviation was done.  If you inserted "x\z", however,
1222 		 * it tried to back up and do the abbreviation, i.e. replace
1223 		 * 'x' with 'y'.  The problem was it got it wrong, and you
1224 		 * ended up with "zy\".
1225 		 *
1226 		 * This is really hard to do (you have to remember the
1227 		 * word/non-word state, for example), and doesn't make any
1228 		 * sense to me.  Both backslash and the characters it
1229 		 * (usually) escapes will individually trigger the
1230 		 * abbreviation, so I don't see why the combination of them
1231 		 * wouldn't.  I don't expect to get caught on this one,
1232 		 * particularly since it never worked right, but I've been
1233 		 * wrong before.
1234 		 *
1235 		 * Do the tests for abbreviations, so ":ab xa XA",
1236 		 * "ixa\<K_VERASE>" performs the abbreviation.
1237 		 */
1238 		quote = Q_BNEXT;
1239 		goto insq_ch;
1240 	case K_VLNEXT:			/* Quote next character. */
1241 		evp->e_c = '^';
1242 		quote = Q_VNEXT;
1243 		/*
1244 		 * Turn on the quote flag so that the underlying routines
1245 		 * quote the next character where it's possible. Turn off
1246 		 * the input mapbiting flag so that we don't remap the next
1247 		 * character.
1248 		 */
1249 		FL_SET(ec_flags, EC_QUOTED);
1250 		FL_CLR(ec_flags, EC_MAPINPUT);
1251 
1252 		/*
1253 		 * !!!
1254 		 * Skip the tests for abbreviations, so ":ab xa XA",
1255 		 * "ixa^V<space>" doesn't perform the abbreviation.
1256 		 */
1257 		goto insl_ch;
1258 	case K_HEXCHAR:
1259 		hexcnt = 1;
1260 		goto insq_ch;
1261 	default:			/* Insert the character. */
1262 ins_ch:		/*
1263 		 * Historically, vi eliminated nul's out of hand.  If the
1264 		 * beautify option was set, it also deleted any unknown
1265 		 * ASCII value less than space (040) and the del character
1266 		 * (0177), except for tabs.  Unknown is a key word here.
1267 		 * Most vi documentation claims that it deleted everything
1268 		 * but <tab>, <nl> and <ff>, as that's what the original
1269 		 * 4BSD documentation said.  This is obviously wrong,
1270 		 * however, as <esc> would be included in that list.  What
1271 		 * we do is eliminate any unquoted, iscntrl() character that
1272 		 * wasn't a replay and wasn't handled specially, except
1273 		 * <tab> or <ff>.
1274 		 */
1275 		if (LF_ISSET(TXT_BEAUTIFY) && ISCNTRL(evp->e_c) &&
1276 		    evp->e_value != K_FORMFEED && evp->e_value != K_TAB) {
1277 			msgq(sp, M_BERR,
1278 			    "192|Illegal character; quote to enter");
1279 			if (LF_ISSET(TXT_REPLAY))
1280 				goto done;
1281 			break;
1282 		}
1283 
1284 insq_ch:	/*
1285 		 * If entering a non-word character after a word, check for
1286 		 * abbreviations.  If there was one, discard replay characters.
1287 		 * If entering a blank character, check for unmap commands,
1288 		 * as well.
1289 		 */
1290 		if (!inword((UCHAR_T)evp->e_c)) {
1291 			if (abb == AB_INWORD &&
1292 			    !LF_ISSET(TXT_REPLAY) && F_ISSET(gp, G_ABBREV)) {
1293 				if (txt_abbrev(sp, tp, &evp->e_c,
1294 				    LF_ISSET(TXT_INFOLINE), &tmp, &ab_turnoff))
1295 					goto err;
1296 				if (tmp) {
1297 					if (LF_ISSET(TXT_RECORD))
1298 						rcol -= tmp + 1;
1299 					goto resolve;
1300 				}
1301 			}
1302 			if (ISBLANK((UCHAR_T)evp->e_c) && UNMAP_TST)
1303 				txt_unmap(sp, tp, &ec_flags);
1304 		}
1305 		if (abb != AB_NOTSET)
1306 			abb = inword((UCHAR_T)evp->e_c) ? AB_INWORD : AB_NOTWORD;
1307 
1308 insl_ch:	if (txt_insch(sp, tp, &evp->e_c, flags))
1309 			goto err;
1310 
1311 		/*
1312 		 * If we're using K_VLNEXT to quote the next character, then
1313 		 * we want the cursor to position itself on the ^ placeholder
1314 		 * we're displaying, to match historic practice.
1315 		 */
1316 		if (quote == Q_VNEXT) {
1317 			--tp->cno;
1318 			++tp->owrite;
1319 		}
1320 
1321 		/*
1322 		 * !!!
1323 		 * Translate "<CH_HEX>[isxdigit()]*" to a character with
1324 		 * a hex value: this test delimits the value by the max
1325 		 * number of hex bytes.  Offset by one, we use 0 to mean
1326 		 * that we've found <CH_HEX>.
1327 		 */
1328 		if (hexcnt != 0 && hexcnt++ == sizeof(CHAR_T) * 2 + 1) {
1329 			hexcnt = 0;
1330 			if (txt_hex(sp, tp))
1331 				goto err;
1332 		}
1333 
1334 		/*
1335 		 * Check to see if we've crossed the margin.
1336 		 *
1337 		 * !!!
1338 		 * In the historic vi, the wrapmargin value was figured out
1339 		 * using the display widths of the characters, i.e. <tab>
1340 		 * characters were counted as two characters if the list edit
1341 		 * option is set, but as the tabstop edit option number of
1342 		 * characters otherwise.  That's what the vs_column() function
1343 		 * gives us, so we use it.
1344 		 */
1345 		if (margin != 0) {
1346 			if (vs_column(sp, &tcol))
1347 				goto err;
1348 			if (tcol >= margin) {
1349 				if (txt_margin(sp, tp, &wmt, &tmp, flags))
1350 					goto err;
1351 				if (tmp) {
1352 					if (ISBLANK((UCHAR_T)evp->e_c))
1353 						wm_skip = 1;
1354 					wm_set = 1;
1355 					goto k_cr;
1356 				}
1357 			}
1358 		}
1359 
1360 		/*
1361 		 * If we've reached the end of the buffer, then we need to
1362 		 * switch into insert mode.  This happens when there's a
1363 		 * change to a mark and the user puts in more characters than
1364 		 * the length of the motion.
1365 		 */
1366 ebuf_chk:	if (tp->cno >= tp->len) {
1367 			BINC_GOTOW(sp, tp->lb, tp->lb_len, tp->len + 1);
1368 			LF_SET(TXT_APPENDEOL);
1369 
1370 			tp->lb[tp->cno] = CH_CURSOR;
1371 			++tp->insert;
1372 			++tp->len;
1373 		}
1374 
1375 		/* Step the quote state forward. */
1376 		if (quote != Q_NOTSET) {
1377 			if (quote == Q_BNEXT)
1378 				quote = Q_BTHIS;
1379 			if (quote == Q_VNEXT)
1380 				quote = Q_VTHIS;
1381 		}
1382 		break;
1383 	}
1384 
1385 #ifdef DEBUG
1386 	if (tp->cno + tp->insert + tp->owrite != tp->len) {
1387 		msgq(sp, M_ERR,
1388 		    "len %u != cno: %u ai: %u insert %u overwrite %u",
1389 		    tp->len, tp->cno, tp->ai, tp->insert, tp->owrite);
1390 		if (LF_ISSET(TXT_REPLAY))
1391 			goto done;
1392 		tp->len = tp->cno + tp->insert + tp->owrite;
1393 	}
1394 #endif
1395 
1396 resolve:/*
1397 	 * 1: If we don't need to know where the cursor really is and we're
1398 	 *    replaying text, keep going.
1399 	 */
1400 	if (margin == 0 && LF_ISSET(TXT_REPLAY))
1401 		goto replay;
1402 
1403 	/*
1404 	 * 2: Reset the line.  Don't bother unless we're about to wait on
1405 	 *    a character or we need to know where the cursor really is.
1406 	 *    We have to do this before showing matching characters so the
1407 	 *    user can see what they're matching.
1408 	 */
1409 	if ((margin != 0 || !KEYS_WAITING(sp)) &&
1410 	    vs_change(sp, tp->lno, LINE_RESET))
1411 		return (1);
1412 
1413 	/*
1414 	 * 3: If there aren't keys waiting, display the matching character.
1415 	 *    We have to do this before resolving any messages, otherwise
1416 	 *    the error message from a missing match won't appear correctly.
1417 	 */
1418 	if (showmatch) {
1419 		if (!KEYS_WAITING(sp) && txt_showmatch(sp, tp))
1420 			return (1);
1421 		showmatch = 0;
1422 	}
1423 
1424 	/*
1425 	 * 4: If there have been messages and we're not editing on the colon
1426 	 *    command line or doing file name completion, resolve them.
1427 	 */
1428 	if ((vip->totalcount != 0 || F_ISSET(gp, G_BELLSCHED)) &&
1429 	    !F_ISSET(sp, SC_TINPUT_INFO) && !filec_redraw &&
1430 	    vs_resolve(sp, NULL, 0))
1431 		return (1);
1432 
1433 	/*
1434 	 * 5: Refresh the screen if we're about to wait on a character or we
1435 	 *    need to know where the cursor really is.
1436 	 */
1437 	if (margin != 0 || !KEYS_WAITING(sp)) {
1438 		UPDATE_POSITION(sp, tp);
1439 		if (vs_refresh(sp, margin != 0))
1440 			return (1);
1441 	}
1442 
1443 	/* 6: Proceed with the incremental search. */
1444 	if (FL_ISSET(is_flags, IS_RUNNING) && txt_isrch(sp, vp, tp, &is_flags))
1445 		return (1);
1446 
1447 	/* 7: Next character... */
1448 	if (LF_ISSET(TXT_REPLAY))
1449 		goto replay;
1450 	goto next;
1451 
1452 done:	/* Leave input mode. */
1453 	F_CLR(sp, SC_TINPUT);
1454 
1455 	/* If recording for playback, save it. */
1456 	if (LF_ISSET(TXT_RECORD))
1457 		vip->rep_cnt = rcol;
1458 
1459 	/*
1460 	 * If not working on the colon command line, set the final cursor
1461 	 * position.
1462 	 */
1463 	if (!F_ISSET(sp, SC_TINPUT_INFO)) {
1464 		vp->m_final.lno = tp->lno;
1465 		vp->m_final.cno = tp->cno;
1466 	}
1467 	return (0);
1468 
1469 err:
1470 alloc_err:
1471 	F_CLR(sp, SC_TINPUT);
1472 	txt_err(sp, &sp->tiq);
1473 	return (1);
1474 }
1475 
1476 /*
1477  * txt_abbrev --
1478  *	Handle abbreviations.
1479  */
1480 static int
txt_abbrev(SCR * sp,TEXT * tp,ARG_CHAR_T * pushcp,int isinfoline,int * didsubp,int * turnoffp)1481 txt_abbrev(SCR *sp, TEXT *tp, ARG_CHAR_T *pushcp, int isinfoline, int *didsubp, int *turnoffp)
1482 {
1483 	CHAR_T ch, *p;
1484 	SEQ *qp;
1485 	size_t len, off;
1486 
1487 	/* Check to make sure we're not at the start of an append. */
1488 	*didsubp = 0;
1489 	if (tp->cno == tp->offset)
1490 		return (0);
1491 
1492 	/*
1493 	 * Find the start of the "word".
1494 	 *
1495 	 * !!!
1496 	 * We match historic practice, which, as far as I can tell, had an
1497 	 * off-by-one error.  The way this worked was that when the inserted
1498 	 * text switched from a "word" character to a non-word character,
1499 	 * vi would check for possible abbreviations.  It would then take the
1500 	 * type (i.e. word/non-word) of the character entered TWO characters
1501 	 * ago, and move backward in the text until reaching a character that
1502 	 * was not that type, or the beginning of the insert, the line, or
1503 	 * the file.  For example, in the string "abc<space>", when the <space>
1504 	 * character triggered the abbreviation check, the type of the 'b'
1505 	 * character was used for moving through the string.  Maybe there's a
1506 	 * reason for not using the first (i.e. 'c') character, but I can't
1507 	 * think of one.
1508 	 *
1509 	 * Terminate at the beginning of the insert or the character after the
1510 	 * offset character -- both can be tested for using tp->offset.
1511 	 */
1512 	off = tp->cno - 1;			/* Previous character. */
1513 	p = tp->lb + off;
1514 	len = 1;				/* One character test. */
1515 	if (off == tp->offset || ISBLANK((UCHAR_T)p[-1]))
1516 		goto search;
1517 	if (inword((UCHAR_T)p[-1]))			/* Move backward to change. */
1518 		for (;;) {
1519 			--off; --p; ++len;
1520 			if (off == tp->offset || !inword((UCHAR_T)p[-1]))
1521 				break;
1522 		}
1523 	else
1524 		for (;;) {
1525 			--off; --p; ++len;
1526 			if (off == tp->offset ||
1527 			    inword((UCHAR_T)p[-1]) || ISBLANK((UCHAR_T)p[-1]))
1528 				break;
1529 		}
1530 
1531 	/*
1532 	 * !!!
1533 	 * Historic vi exploded abbreviations on the command line.  This has
1534 	 * obvious problems in that unabbreviating the string can be extremely
1535 	 * tricky, particularly if the string has, say, an embedded escape
1536 	 * character.  Personally, I think it's a stunningly bad idea.  Other
1537 	 * examples of problems this caused in historic vi are:
1538 	 *	:ab foo bar
1539 	 *	:ab foo baz
1540 	 * results in "bar" being abbreviated to "baz", which wasn't what the
1541 	 * user had in mind at all.  Also, the commands:
1542 	 *	:ab foo bar
1543 	 *	:unab foo<space>
1544 	 * resulted in an error message that "bar" wasn't mapped.  Finally,
1545 	 * since the string was already exploded by the time the unabbreviate
1546 	 * command got it, all it knew was that an abbreviation had occurred.
1547 	 * Cleverly, it checked the replacement string for its unabbreviation
1548 	 * match, which meant that the commands:
1549 	 *	:ab foo1 bar
1550 	 *	:ab foo2 bar
1551 	 *	:unab foo2
1552 	 * unabbreviate "foo1", and the commands:
1553 	 *	:ab foo bar
1554 	 *	:ab bar baz
1555 	 * unabbreviate "foo"!
1556 	 *
1557 	 * Anyway, people neglected to first ask my opinion before they wrote
1558 	 * macros that depend on this stuff, so, we make this work as follows.
1559 	 * When checking for an abbreviation on the command line, if we get a
1560 	 * string which is <blank> terminated and which starts at the beginning
1561 	 * of the line, we check to see it is the abbreviate or unabbreviate
1562 	 * commands.  If it is, turn abbreviations off and return as if no
1563 	 * abbreviation was found.  Note also, minor trickiness, so that if
1564 	 * the user erases the line and starts another command, we turn the
1565 	 * abbreviations back on.
1566 	 *
1567 	 * This makes the layering look like a Nachos Supreme.
1568 	 */
1569 search:	if (isinfoline) {
1570 		if (off == tp->ai || off == tp->offset)
1571 			if (ex_is_abbrev(sp, p, len)) {
1572 				*turnoffp = 1;
1573 				return (0);
1574 			} else
1575 				*turnoffp = 0;
1576 		else
1577 			if (*turnoffp)
1578 				return (0);
1579 	}
1580 
1581 	/* Check for any abbreviations. */
1582 	if ((qp = seq_find(sp, NULL, NULL, p, len, SEQ_ABBREV, NULL)) == NULL)
1583 		return (0);
1584 
1585 	/*
1586 	 * Push the abbreviation onto the tty stack.  Historically, characters
1587 	 * resulting from an abbreviation expansion were themselves subject to
1588 	 * map expansions, O_SHOWMATCH matching etc.  This means the expanded
1589 	 * characters will be re-tested for abbreviations.  It's difficult to
1590 	 * know what historic practice in this case was, since abbreviations
1591 	 * were applied to :colon command lines, so entering abbreviations that
1592 	 * looped was tricky, although possible.  In addition, obvious loops
1593 	 * didn't work as expected.  (The command ':ab a b|ab b c|ab c a' will
1594 	 * silently only implement and/or display the last abbreviation.)
1595 	 *
1596 	 * This implementation doesn't recover well from such abbreviations.
1597 	 * The main input loop counts abbreviated characters, and, when it
1598 	 * reaches a limit, discards any abbreviated characters on the queue.
1599 	 * It's difficult to back up to the original position, as the replay
1600 	 * queue would have to be adjusted, and the line state when an initial
1601 	 * abbreviated character was received would have to be saved.
1602 	 */
1603 	ch = (UCHAR_T)*pushcp;
1604 	if (v_event_push(sp, NULL, &ch, 1, CH_ABBREVIATED))
1605 		return (1);
1606 	if (v_event_push(sp, NULL, qp->output, qp->olen, CH_ABBREVIATED))
1607 		return (1);
1608 
1609 	/*
1610 	 * If the size of the abbreviation is larger than or equal to the size
1611 	 * of the original text, move to the start of the replaced characters,
1612 	 * and add their length to the overwrite count.
1613 	 *
1614 	 * If the abbreviation is smaller than the original text, we have to
1615 	 * delete the additional overwrite characters and copy down any insert
1616 	 * characters.
1617 	 */
1618 	tp->cno -= len;
1619 	if (qp->olen >= len)
1620 		tp->owrite += len;
1621 	else {
1622 		if (tp->insert)
1623 			MEMMOVEW(tp->lb + tp->cno + qp->olen,
1624 			    tp->lb + tp->cno + tp->owrite + len, tp->insert);
1625 		tp->owrite += qp->olen;
1626 		tp->len -= len - qp->olen;
1627 	}
1628 
1629 	/*
1630 	 * We return the length of the abbreviated characters.  This is so
1631 	 * the calling routine can replace the replay characters with the
1632 	 * abbreviation.  This means that subsequent '.' commands will produce
1633 	 * the same text, regardless of intervening :[un]abbreviate commands.
1634 	 * This is historic practice.
1635 	 */
1636 	*didsubp = len;
1637 	return (0);
1638 }
1639 
1640 /*
1641  * txt_unmap --
1642  *	Handle the unmap command.
1643  */
1644 static void
txt_unmap(SCR * sp,TEXT * tp,u_int32_t * ec_flagsp)1645 txt_unmap(SCR *sp, TEXT *tp, u_int32_t *ec_flagsp)
1646 {
1647 	size_t len, off;
1648 	CHAR_T *p;
1649 
1650 	/* Find the beginning of this "word". */
1651 	for (off = tp->cno - 1, p = tp->lb + off, len = 0;; --p, --off) {
1652 		if (ISBLANK((UCHAR_T)*p)) {
1653 			++p;
1654 			break;
1655 		}
1656 		++len;
1657 		if (off == tp->ai || off == tp->offset)
1658 			break;
1659 	}
1660 
1661 	/*
1662 	 * !!!
1663 	 * Historic vi exploded input mappings on the command line.  See the
1664 	 * txt_abbrev() routine for an explanation of the problems inherent
1665 	 * in this.
1666 	 *
1667 	 * We make this work as follows.  If we get a string which is <blank>
1668 	 * terminated and which starts at the beginning of the line, we check
1669 	 * to see it is the unmap command.  If it is, we return that the input
1670 	 * mapping should be turned off.  Note also, minor trickiness, so that
1671 	 * if the user erases the line and starts another command, we go ahead
1672 	 * an turn mapping back on.
1673 	 */
1674 	if ((off == tp->ai || off == tp->offset) && ex_is_unmap(sp, p, len))
1675 		FL_CLR(*ec_flagsp, EC_MAPINPUT);
1676 	else
1677 		FL_SET(*ec_flagsp, EC_MAPINPUT);
1678 }
1679 
1680 /*
1681  * txt_ai_resolve --
1682  *	When a line is resolved by <esc>, review autoindent characters.
1683  */
1684 static void
txt_ai_resolve(SCR * sp,TEXT * tp,int * changedp)1685 txt_ai_resolve(SCR *sp, TEXT *tp, int *changedp)
1686 {
1687 	u_long ts;
1688 	int delc;
1689 	size_t cno, len, new, old, scno, spaces, tab_after_sp, tabs;
1690 	CHAR_T *p;
1691 
1692 	*changedp = 0;
1693 
1694 	/*
1695 	 * If the line is empty, has an offset, or no autoindent
1696 	 * characters, we're done.
1697 	 */
1698 	if (!tp->len || tp->offset || !tp->ai)
1699 		return;
1700 
1701 	/*
1702 	 * If the length is less than or equal to the autoindent
1703 	 * characters, delete them.
1704 	 */
1705 	if (tp->len <= tp->ai) {
1706 		tp->ai = tp->cno = tp->len = 0;
1707 		return;
1708 	}
1709 
1710 	/*
1711 	 * The autoindent characters plus any leading <blank> characters
1712 	 * in the line are resolved into the minimum number of characters.
1713 	 * Historic practice.
1714 	 */
1715 	ts = O_VAL(sp, O_TABSTOP);
1716 
1717 	/* Figure out the last <blank> screen column. */
1718 	for (p = tp->lb, scno = 0, len = tp->len,
1719 	    spaces = tab_after_sp = 0; len-- && ISBLANK((UCHAR_T)*p); ++p)
1720 		if (*p == '\t') {
1721 			if (spaces)
1722 				tab_after_sp = 1;
1723 			scno += COL_OFF(scno, ts);
1724 		} else {
1725 			++spaces;
1726 			++scno;
1727 		}
1728 
1729 	/*
1730 	 * If there are no spaces, or no tabs after spaces and less than
1731 	 * ts spaces, it's already minimal.
1732 	 * Keep analysing if expandtab is set.
1733 	 */
1734 	if ((!spaces || (!tab_after_sp && spaces < ts)) &&
1735 	    !O_ISSET(sp, O_EXPANDTAB))
1736 		return;
1737 
1738 	/* Count up spaces/tabs needed to get to the target. */
1739 	cno = 0;
1740 	tabs = 0;
1741 	if (!O_ISSET(sp, O_EXPANDTAB)) {
1742 		for (; cno + COL_OFF(cno, ts) <= scno; ++tabs)
1743 			cno += COL_OFF(cno, ts);
1744 	}
1745 	spaces = scno - cno;
1746 
1747 	/*
1748 	 * Figure out how many characters we're dropping -- if we're not
1749 	 * dropping any, it's already minimal, we're done.
1750 	 */
1751 	old = p - tp->lb;
1752 	new = spaces + tabs;
1753 	if (old == new)
1754 		return;
1755 
1756 	/* Shift the rest of the characters down, adjust the counts. */
1757 	delc = old - new;
1758 	MEMMOVEW(p - delc, p, tp->len - old);
1759 	tp->len -= delc;
1760 	tp->cno -= delc;
1761 
1762 	/* Fill in space/tab characters. */
1763 	for (p = tp->lb; tabs--;)
1764 		*p++ = '\t';
1765 	while (spaces--)
1766 		*p++ = ' ';
1767 	*changedp = 1;
1768 }
1769 
1770 /*
1771  * v_txt_auto --
1772  *	Handle autoindent.  If aitp isn't NULL, use it, otherwise,
1773  *	retrieve the line.
1774  *
1775  * PUBLIC: int v_txt_auto __P((SCR *, db_recno_t, TEXT *, size_t, TEXT *));
1776  */
1777 int
v_txt_auto(SCR * sp,db_recno_t lno,TEXT * aitp,size_t len,TEXT * tp)1778 v_txt_auto(SCR *sp, db_recno_t lno, TEXT *aitp, size_t len, TEXT *tp)
1779 {
1780 	size_t nlen;
1781 	CHAR_T *p, *t;
1782 
1783 	if (aitp == NULL) {
1784 		/*
1785 		 * If the ex append command is executed with an address of 0,
1786 		 * it's possible to get here with a line number of 0.  Return
1787 		 * an indent of 0.
1788 		 */
1789 		if (lno == 0) {
1790 			tp->ai = 0;
1791 			return (0);
1792 		}
1793 		if (db_get(sp, lno, DBG_FATAL, &t, &len))
1794 			return (1);
1795 	} else
1796 		t = aitp->lb;
1797 
1798 	/* Count whitespace characters. */
1799 	for (p = t; len > 0; ++p, --len)
1800 		if (!ISBLANK((UCHAR_T)*p))
1801 			break;
1802 
1803 	/* Set count, check for no indentation. */
1804 	if ((nlen = (p - t)) == 0)
1805 		return (0);
1806 
1807 	/* Make sure the buffer's big enough. */
1808 	BINC_RETW(sp, tp->lb, tp->lb_len, tp->len + nlen);
1809 
1810 	/* Copy the buffer's current contents up. */
1811 	if (tp->len != 0)
1812 		MEMMOVEW(tp->lb + nlen, tp->lb, tp->len);
1813 	tp->len += nlen;
1814 
1815 	/* Copy the indentation into the new buffer. */
1816 	MEMMOVEW(tp->lb, t, nlen);
1817 
1818 	/* Set the autoindent count. */
1819 	tp->ai = nlen;
1820 	return (0);
1821 }
1822 
1823 /*
1824  * txt_backup --
1825  *	Back up to the previously edited line.
1826  */
1827 static TEXT *
txt_backup(SCR * sp,TEXTH * tiqh,TEXT * tp,u_int32_t * flagsp)1828 txt_backup(SCR *sp, TEXTH *tiqh, TEXT *tp, u_int32_t *flagsp)
1829 {
1830 	TEXT *ntp;
1831 
1832 	/* Get a handle on the previous TEXT structure. */
1833 	if ((ntp = TAILQ_PREV(tp, _texth, q)) == NULL) {
1834 		if (!FL_ISSET(*flagsp, TXT_REPLAY))
1835 			msgq(sp, M_BERR,
1836 			    "193|Already at the beginning of the insert");
1837 		return (tp);
1838 	}
1839 
1840 	/* Bookkeeping. */
1841 	ntp->len = ntp->sv_len;
1842 
1843 	/* Handle appending to the line. */
1844 	if (ntp->owrite == 0 && ntp->insert == 0) {
1845 		ntp->lb[ntp->len] = CH_CURSOR;
1846 		++ntp->insert;
1847 		++ntp->len;
1848 		FL_SET(*flagsp, TXT_APPENDEOL);
1849 	} else
1850 		FL_CLR(*flagsp, TXT_APPENDEOL);
1851 
1852 	/* Release the current TEXT. */
1853 	TAILQ_REMOVE(tiqh, tp, q);
1854 	text_free(tp);
1855 
1856 	/* Update the old line on the screen. */
1857 	if (vs_change(sp, ntp->lno + 1, LINE_DELETE))
1858 		return (NULL);
1859 
1860 	/* Return the new/current TEXT. */
1861 	return (ntp);
1862 }
1863 
1864 /*
1865  * Text indentation is truly strange.  ^T and ^D do movements to the next or
1866  * previous shiftwidth value, i.e. for a 1-based numbering, with shiftwidth=3,
1867  * ^T moves a cursor on the 7th, 8th or 9th column to the 10th column, and ^D
1868  * moves it back.
1869  *
1870  * !!!
1871  * The ^T and ^D characters in historical vi had special meaning only when they
1872  * were the first characters entered after entering text input mode.  As normal
1873  * erase characters couldn't erase autoindent characters (^T in this case), it
1874  * meant that inserting text into previously existing text was strange -- ^T
1875  * only worked if it was the first keystroke(s), and then could only be erased
1876  * using ^D.  This implementation treats ^T specially anywhere it occurs in the
1877  * input, and permits the standard erase characters to erase the characters it
1878  * inserts.
1879  *
1880  * !!!
1881  * A fun test is to try:
1882  *	:se sw=4 ai list
1883  *	i<CR>^Tx<CR>^Tx<CR>^Tx<CR>^Dx<CR>^Dx<CR>^Dx<esc>
1884  * Historic vi loses some of the '$' marks on the line ends, but otherwise gets
1885  * it right.
1886  *
1887  * XXX
1888  * Technically, txt_dent should be part of the screen interface, as it requires
1889  * knowledge of character sizes, including <space>s, on the screen.  It's here
1890  * because it's a complicated little beast, and I didn't want to shove it down
1891  * into the screen.  It's probable that KEY_LEN will call into the screen once
1892  * there are screens with different character representations.
1893  *
1894  * txt_dent --
1895  *	Handle ^T indents, ^D outdents.
1896  *
1897  * If anything changes here, check the ex version to see if it needs similar
1898  * changes.
1899  */
1900 static int
txt_dent(SCR * sp,TEXT * tp,int isindent)1901 txt_dent(SCR *sp, TEXT *tp, int isindent)
1902 {
1903 	ARG_CHAR_T ch;
1904 	u_long sw, ts;
1905 	size_t cno, current, spaces, target, tabs;
1906 
1907 	ts = O_VAL(sp, O_TABSTOP);
1908 	sw = O_VAL(sp, O_SHIFTWIDTH);
1909 
1910 	/*
1911 	 * Since we don't know what precedes the character(s) being inserted
1912 	 * (or deleted), the preceding whitespace characters must be resolved.
1913 	 * An example is a <tab>, which doesn't need a full shiftwidth number
1914 	 * of columns because it's preceded by <space>s.  This is easy to get
1915 	 * if the user sets shiftwidth to a value less than tabstop (or worse,
1916 	 * something for which tabstop isn't a multiple) and then uses ^T to
1917 	 * indent, and ^D to outdent.
1918 	 *
1919 	 * Figure out the current and target screen columns.  In the historic
1920 	 * vi, the autoindent column was NOT determined using display widths
1921 	 * of characters as was the wrapmargin column.  For that reason, we
1922 	 * can't use the vs_column() function, but have to calculate it here.
1923 	 * This is slow, but it's normally only on the first few characters of
1924 	 * a line.
1925 	 */
1926 	for (current = cno = 0; cno < tp->cno; ++cno)
1927 		current += tp->lb[cno] == '\t' ?
1928 		    COL_OFF(current, ts) : KEY_COL(sp, tp->lb[cno]);
1929 
1930 	target = current;
1931 	if (isindent)
1932 		target += COL_OFF(target, sw);
1933 	else {
1934 		--target;
1935 		target -= target % sw;
1936 	}
1937 
1938 	/*
1939 	 * Back up over any previous <blank> characters, changing them into
1940 	 * overwrite characters (including any ai characters).  Then figure
1941 	 * out the current screen column.
1942 	 */
1943 	for (; tp->cno > tp->offset &&
1944 	    (tp->lb[tp->cno - 1] == ' ' || tp->lb[tp->cno - 1] == '\t');
1945 	    --tp->cno, ++tp->owrite);
1946 	for (current = cno = 0; cno < tp->cno; ++cno)
1947 		current += tp->lb[cno] == '\t' ?
1948 		    COL_OFF(current, ts) : KEY_COL(sp, tp->lb[cno]);
1949 
1950 	/*
1951 	 * If we didn't move up to or past the target, it's because there
1952 	 * weren't enough characters to delete, e.g. the first character
1953 	 * of the line was a tp->offset character, and the user entered
1954 	 * ^D to move to the beginning of a line.  An example of this is:
1955 	 *
1956 	 *	:set ai sw=4<cr>i<space>a<esc>i^T^D
1957 	 *
1958 	 * Otherwise, count up the total spaces/tabs needed to get from the
1959 	 * beginning of the line (or the last non-<blank> character) to the
1960 	 * target.
1961 	 */
1962 	if (current >= target)
1963 		spaces = tabs = 0;
1964 	else {
1965 		cno = current;
1966 		tabs = 0;
1967 		if (!O_ISSET(sp, O_EXPANDTAB)) {
1968 			for (; cno + COL_OFF(cno, ts) <= target; ++tabs)
1969 				cno += COL_OFF(cno, ts);
1970 		}
1971 		spaces = target - cno;
1972 	}
1973 
1974 	tp->ai = tabs + spaces;
1975 
1976 	/*
1977 	 * Call txt_insch() to insert each character, so that we get the
1978 	 * correct effect when we add a <tab> to replace N <spaces>.
1979 	 */
1980 	for (ch = '\t'; tabs > 0; --tabs)
1981 		(void)txt_insch(sp, tp, &ch, 0);
1982 	for (ch = ' '; spaces > 0; --spaces)
1983 		(void)txt_insch(sp, tp, &ch, 0);
1984 	return (0);
1985 }
1986 
1987 /*
1988  * txt_fc --
1989  *	File name completion.
1990  */
1991 static int
txt_fc(SCR * sp,TEXT * tp,int * redrawp)1992 txt_fc(SCR *sp, TEXT *tp, int *redrawp)
1993 {
1994 	struct stat sb;
1995 	ARGS **argv;
1996 	CHAR_T s_ch;
1997 	EXCMD cmd;
1998 	size_t indx, len, nlen, off;
1999 	int argc, trydir;
2000 	CHAR_T *p, *t;
2001 	const char *np;
2002 	size_t nplen;
2003 
2004 	trydir = 0;
2005 	*redrawp = 0;
2006 
2007 	/*
2008 	 * Find the beginning of this "word" -- if we're at the beginning
2009 	 * of the line, it's a special case.
2010 	 */
2011 	if (tp->cno == 1) {
2012 		len = 0;
2013 		p = tp->lb;
2014 	} else
2015 retry:		for (len = 0,
2016 		    off = tp->cno - 1, p = tp->lb + off;; --off, --p) {
2017 			if (ISBLANK((UCHAR_T)*p)) {
2018 				++p;
2019 				break;
2020 			}
2021 			++len;
2022 			if (off == tp->ai || off == tp->offset)
2023 				break;
2024 		}
2025 
2026 	/*
2027 	 * Get enough space for a wildcard character.
2028 	 *
2029 	 * XXX
2030 	 * This won't work for "foo\", since the \ will escape the expansion
2031 	 * character.  I'm not sure if that's a bug or not...
2032 	 */
2033 	off = p - tp->lb;
2034 	BINC_RETW(sp, tp->lb, tp->lb_len, tp->len + 1);
2035 	p = tp->lb + off;
2036 
2037 	s_ch = p[len];
2038 	p[len] = '*';
2039 
2040 	/* Build an ex command, and call the ex expansion routines. */
2041 	ex_cinit(sp, &cmd, 0, 0, OOBLNO, OOBLNO, 0);
2042 	if (argv_exp2(sp, &cmd, p, len + 1)) {
2043 		p[len] = s_ch;
2044 		return (0);
2045 	}
2046 	argc = cmd.argc;
2047 	argv = cmd.argv;
2048 
2049 	p[len] = s_ch;
2050 
2051 	switch (argc) {
2052 	case 0:				/* No matches. */
2053 		if (!trydir)
2054 			(void)sp->gp->scr_bell(sp);
2055 		return (0);
2056 	case 1:				/* One match. */
2057 		/* If something changed, do the exchange. */
2058 		nlen = STRLEN(cmd.argv[0]->bp);
2059 		if (len != nlen || MEMCMP(cmd.argv[0]->bp, p, len))
2060 			break;
2061 
2062 		/* If haven't done a directory test, do it now. */
2063 		INT2CHAR(sp, cmd.argv[0]->bp, cmd.argv[0]->len + 1,
2064 			 np, nplen);
2065 		if (!trydir &&
2066 		    !stat(np, &sb) && S_ISDIR(sb.st_mode)) {
2067 			p += len;
2068 			goto isdir;
2069 		}
2070 
2071 		/* If nothing changed, period, ring the bell. */
2072 		if (!trydir)
2073 			(void)sp->gp->scr_bell(sp);
2074 		return (0);
2075 	default:			/* Multiple matches. */
2076 		*redrawp = 1;
2077 		if (txt_fc_col(sp, argc, argv))
2078 			return (1);
2079 
2080 		/* Find the length of the shortest match. */
2081 		for (nlen = cmd.argv[0]->len; --argc > 0;) {
2082 			if (cmd.argv[argc]->len < nlen)
2083 				nlen = cmd.argv[argc]->len;
2084 			for (indx = 0; indx < nlen &&
2085 			    cmd.argv[argc]->bp[indx] == cmd.argv[0]->bp[indx];
2086 			    ++indx);
2087 			nlen = indx;
2088 		}
2089 		break;
2090 	}
2091 
2092 	/* Overwrite the expanded text first. */
2093 	for (t = cmd.argv[0]->bp; len > 0 && nlen > 0; --len, --nlen)
2094 		*p++ = *t++;
2095 
2096 	/* If lost text, make the remaining old text overwrite characters. */
2097 	if (len) {
2098 		tp->cno -= len;
2099 		tp->owrite += len;
2100 	}
2101 
2102 	/* Overwrite any overwrite characters next. */
2103 	for (; nlen > 0 && tp->owrite > 0; --nlen, --tp->owrite, ++tp->cno)
2104 		*p++ = *t++;
2105 
2106 	/* Shift remaining text up, and move the cursor to the end. */
2107 	if (nlen) {
2108 		off = p - tp->lb;
2109 		BINC_RETW(sp, tp->lb, tp->lb_len, tp->len + nlen);
2110 		p = tp->lb + off;
2111 
2112 		tp->cno += nlen;
2113 		tp->len += nlen;
2114 
2115 		if (tp->insert != 0)
2116 			(void)MEMMOVEW(p + nlen, p, tp->insert);
2117 		while (nlen--)
2118 			*p++ = *t++;
2119 	}
2120 
2121 	/* If a single match and it's a directory, retry it. */
2122 	INT2CHAR(sp, cmd.argv[0]->bp, cmd.argv[0]->len + 1, np, nplen);
2123 	if (argc == 1 && !stat(np, &sb) && S_ISDIR(sb.st_mode)) {
2124 isdir:		if (tp->owrite == 0) {
2125 			off = p - tp->lb;
2126 			BINC_RETW(sp, tp->lb, tp->lb_len, tp->len + 1);
2127 			p = tp->lb + off;
2128 			if (tp->insert != 0)
2129 				(void)MEMMOVEW(p + 1, p, tp->insert);
2130 			++tp->len;
2131 		} else
2132 			--tp->owrite;
2133 
2134 		++tp->cno;
2135 		*p++ = '/';
2136 
2137 		trydir = 1;
2138 		goto retry;
2139 	}
2140 	return (0);
2141 }
2142 
2143 /*
2144  * txt_fc_col --
2145  *	Display file names for file name completion.
2146  */
2147 static int
txt_fc_col(SCR * sp,int argc,ARGS ** argv)2148 txt_fc_col(SCR *sp, int argc, ARGS **argv)
2149 {
2150 	ARGS **av;
2151 	CHAR_T *p;
2152 	GS *gp;
2153 	size_t base, cnt, col, colwidth, numrows, numcols, prefix, row;
2154 	int ac, nf, reset;
2155 	const char *np;
2156 	char *pp;
2157 	size_t nlen;
2158 
2159 	gp = sp->gp;
2160 
2161 	/* Trim any directory prefix common to all of the files. */
2162 	INT2CHAR(sp, argv[0]->bp, argv[0]->len + 1, np, nlen);
2163 	if ((pp = strrchr(np, '/')) == NULL)
2164 		prefix = 0;
2165 	else {
2166 		prefix = (pp - np) + 1;
2167 		for (ac = argc - 1, av = argv + 1; ac > 0; --ac, ++av)
2168 			if (av[0]->len < prefix ||
2169 			    MEMCMP(av[0]->bp, argv[0]->bp,
2170 				   prefix)) {
2171 				prefix = 0;
2172 				break;
2173 			}
2174 	}
2175 
2176 	/*
2177 	 * Figure out the column width for the longest name.  Output is done on
2178 	 * 6 character "tab" boundaries for no particular reason.  (Since we
2179 	 * don't output tab characters, we ignore the terminal's tab settings.)
2180 	 * Ignore the user's tab setting because we have no idea how reasonable
2181 	 * it is.
2182 	 */
2183 	for (ac = argc, av = argv, colwidth = 0; ac > 0; --ac, ++av) {
2184 		for (col = 0, p = av[0]->bp + prefix; *p != '\0'; ++p)
2185 			col += KEY_LEN(sp, *p);
2186 		if (col > colwidth)
2187 			colwidth = col;
2188 	}
2189 	colwidth += COL_OFF(colwidth, 6);
2190 
2191 	/*
2192 	 * Writing to the bottom line of the screen is always turned off when
2193 	 * SC_TINPUT_INFO is set.  Turn it back on, we know what we're doing.
2194 	 */
2195 	if (F_ISSET(sp, SC_TINPUT_INFO)) {
2196 		reset = 1;
2197 		F_CLR(sp, SC_TINPUT_INFO);
2198 	} else
2199 		reset = 0;
2200 
2201 #define	CHK_INTR							\
2202 	if (F_ISSET(gp, G_INTERRUPTED))					\
2203 		goto intr;
2204 
2205 	/* If the largest file name is too large, just print them. */
2206 	if (colwidth > sp->cols) {
2207 		for (ac = argc, av = argv; ac > 0; --ac, ++av) {
2208 			INT2CHAR(sp, av[0]->bp+prefix, av[0]->len+1-prefix,
2209 				 np, nlen);
2210 			pp = msg_print(sp, np, &nf);
2211 			(void)ex_printf(sp, "%s\n", pp);
2212 			if (F_ISSET(gp, G_INTERRUPTED))
2213 				break;
2214 		}
2215 		if (nf)
2216 			FREE_SPACE(sp, pp, 0);
2217 		CHK_INTR;
2218 	} else {
2219 		/* Figure out the number of columns. */
2220 		numcols = (sp->cols - 1) / colwidth;
2221 		if ((size_t)argc > numcols) {
2222 			numrows = argc / numcols;
2223 			if (argc % numcols)
2224 				++numrows;
2225 		} else
2226 			numrows = 1;
2227 
2228 		/* Display the files in sorted order. */
2229 		for (row = 0; row < numrows; ++row) {
2230 			for (base = row, col = 0; col < numcols; ++col) {
2231 				INT2CHAR(sp, argv[base]->bp+prefix,
2232 					argv[base]->len+1-prefix, np, nlen);
2233 				pp = msg_print(sp, np, &nf);
2234 				cnt = ex_printf(sp, "%s", pp);
2235 				if (nf)
2236 					FREE_SPACE(sp, pp, 0);
2237 				CHK_INTR;
2238 				if ((base += numrows) >= (size_t)argc)
2239 					break;
2240 				(void)ex_printf(sp,
2241 				    "%*s", (int)(colwidth - cnt), "");
2242 				CHK_INTR;
2243 			}
2244 			(void)ex_puts(sp, "\n");
2245 			CHK_INTR;
2246 		}
2247 		(void)ex_puts(sp, "\n");
2248 		CHK_INTR;
2249 	}
2250 	(void)ex_fflush(sp);
2251 
2252 	if (0) {
2253 intr:		F_CLR(gp, G_INTERRUPTED);
2254 	}
2255 	if (reset)
2256 		F_SET(sp, SC_TINPUT_INFO);
2257 
2258 	return (0);
2259 }
2260 
2261 /*
2262  * txt_emark --
2263  *	Set the end mark on the line.
2264  */
2265 static int
txt_emark(SCR * sp,TEXT * tp,size_t cno)2266 txt_emark(SCR *sp, TEXT *tp, size_t cno)
2267 {
2268 	CHAR_T ch;
2269 	unsigned char *kp;
2270 	size_t chlen, nlen, olen;
2271 	CHAR_T *p;
2272 
2273 	ch = CH_ENDMARK;
2274 
2275 	/*
2276 	 * The end mark may not be the same size as the current character.
2277 	 * Don't let the line shift.
2278 	 */
2279 	nlen = KEY_COL(sp, ch);
2280 	if (tp->lb[cno] == '\t')
2281 		(void)vs_columns(sp, tp->lb, tp->lno, &cno, &olen);
2282 	else
2283 		olen = KEY_COL(sp, tp->lb[cno]);
2284 
2285 	/*
2286 	 * If the line got longer, well, it's weird, but it's easy.  If
2287 	 * it's the same length, it's easy.  If it got shorter, we have
2288 	 * to fix it up.
2289 	 */
2290 	if (olen > nlen) {
2291 		BINC_RETW(sp, tp->lb, tp->lb_len, tp->len + olen);
2292 		chlen = olen - nlen;
2293 		if (tp->insert != 0)
2294 			MEMMOVEW(tp->lb + cno + 1 + chlen,
2295 			    tp->lb + cno + 1, tp->insert);
2296 
2297 		tp->len += chlen;
2298 		tp->owrite += chlen;
2299 		p = tp->lb + cno;
2300 		if (tp->lb[cno] == '\t')
2301 			for (cno += chlen; chlen--;)
2302 				*p++ = ' ';
2303 		else if (INTISWIDE(tp->lb[cno])) {
2304 			/*
2305 			 * Do not put the end mark on a part of a multi-width
2306 			 * char, which results in screen corruption.
2307 			 */
2308 			for (; chlen >= nlen; chlen -= nlen) {
2309 				cno++;
2310 				*p++ = ch;
2311 			}
2312 			/* Paranoia: nlen is usually 1. */
2313 			for (cno += chlen; chlen--;)
2314 				*p++ = ' ';
2315 		} else
2316 			for (kp = KEY_NAME(sp, tp->lb[cno]),
2317 			    cno += chlen; chlen--;)
2318 				*p++ = *kp++;
2319 	}
2320 	tp->lb[cno] = ch;
2321 	return (vs_change(sp, tp->lno, LINE_RESET));
2322 }
2323 
2324 /*
2325  * txt_err --
2326  *	Handle an error during input processing.
2327  */
2328 static void
txt_err(SCR * sp,TEXTH * tiqh)2329 txt_err(SCR *sp, TEXTH *tiqh)
2330 {
2331 	db_recno_t lno;
2332 
2333 	/*
2334 	 * The problem with input processing is that the cursor is at an
2335 	 * indeterminate position since some input may have been lost due
2336 	 * to a malloc error.  So, try to go back to the place from which
2337 	 * the cursor started, knowing that it may no longer be available.
2338 	 *
2339 	 * We depend on at least one line number being set in the text
2340 	 * chain.
2341 	 */
2342 	for (lno = TAILQ_FIRST(tiqh)->lno;
2343 	    !db_exist(sp, lno) && lno > 0; --lno);
2344 
2345 	sp->lno = lno == 0 ? 1 : lno;
2346 	sp->cno = 0;
2347 
2348 	/* Redraw the screen, just in case. */
2349 	F_SET(sp, SC_SCR_REDRAW);
2350 }
2351 
2352 /*
2353  * txt_hex --
2354  *	Let the user insert any character value they want.
2355  *
2356  * !!!
2357  * This is an extension.  The pattern "^X[0-9a-fA-F]*" is a way
2358  * for the user to specify a character value which their keyboard
2359  * may not be able to enter.
2360  */
2361 static int
txt_hex(SCR * sp,TEXT * tp)2362 txt_hex(SCR *sp, TEXT *tp)
2363 {
2364 	CHAR_T savec;
2365 	size_t len, off;
2366 	long value;
2367 	CHAR_T *p, *wp;
2368 
2369 	/*
2370 	 * Null-terminate the string.  Since nul isn't a legal hex value,
2371 	 * this should be okay, and lets us use a local routine, which
2372 	 * presumably understands the character set, to convert the value.
2373 	 */
2374 	savec = tp->lb[tp->cno];
2375 	tp->lb[tp->cno] = 0;
2376 
2377 	/* Find the previous CH_HEX character. */
2378 	for (off = tp->cno - 1, p = tp->lb + off, len = 0;; --p, --off, ++len) {
2379 		if (*p == CH_HEX) {
2380 			wp = p + 1;
2381 			break;
2382 		}
2383 		/* Not on this line?  Shouldn't happen. */
2384 		if (off == tp->ai || off == tp->offset)
2385 			goto nothex;
2386 	}
2387 
2388 	/* If length of 0, then it wasn't a hex value. */
2389 	if (len == 0)
2390 		goto nothex;
2391 
2392 	/* Get the value. */
2393 	errno = 0;
2394 	value = STRTOL(wp, NULL, 16);
2395 	if (errno || value < 0 || value > 0xff) {
2396 nothex:		tp->lb[tp->cno] = savec;
2397 		return (0);
2398 	}
2399 
2400 	/* Restore the original character. */
2401 	tp->lb[tp->cno] = savec;
2402 
2403 	/* Adjust the bookkeeping. */
2404 	tp->cno -= len;
2405 	tp->len -= len;
2406 	tp->lb[tp->cno - 1] = value;
2407 
2408 	/* Copy down any overwrite characters. */
2409 	if (tp->owrite)
2410 		MEMMOVEW(tp->lb + tp->cno, tp->lb + tp->cno + len,
2411 		    tp->owrite);
2412 
2413 	/* Copy down any insert characters. */
2414 	if (tp->insert)
2415 		MEMMOVEW(tp->lb + tp->cno + tp->owrite,
2416 		    tp->lb + tp->cno + tp->owrite + len,
2417 		    tp->insert);
2418 
2419 	return (0);
2420 }
2421 
2422 /*
2423  * txt_insch --
2424  *
2425  * !!!
2426  * Historic vi did a special screen optimization for tab characters.  As an
2427  * example, for the keystrokes "iabcd<esc>0C<tab>", the tab overwrote the
2428  * rest of the string when it was displayed.
2429  *
2430  * Because early versions of this implementation redisplayed the entire line
2431  * on each keystroke, the "bcd" was pushed to the right as it ignored that
2432  * the user had "promised" to change the rest of the characters.  However,
2433  * the historic vi implementation had an even worse bug: given the keystrokes
2434  * "iabcd<esc>0R<tab><esc>", the "bcd" disappears, and magically reappears
2435  * on the second <esc> key.
2436  *
2437  * POSIX 1003.2 requires (will require) that this be fixed, specifying that
2438  * vi overwrite characters the user has committed to changing, on the basis
2439  * of the screen space they require, but that it not overwrite other characters.
2440  */
2441 static int
txt_insch(SCR * sp,TEXT * tp,ARG_CHAR_T * chp,u_int flags)2442 txt_insch(SCR *sp, TEXT *tp, ARG_CHAR_T *chp, u_int flags)
2443 {
2444 	unsigned char *kp;
2445 	CHAR_T savech;
2446 	size_t chlen, cno, copydown, olen, nlen;
2447 	CHAR_T *p;
2448 
2449 	/*
2450 	 * The 'R' command does one-for-one replacement, because there's
2451 	 * no way to know how many characters the user intends to replace.
2452 	 */
2453 	if (LF_ISSET(TXT_REPLACE)) {
2454 		if (tp->owrite) {
2455 			--tp->owrite;
2456 			tp->lb[tp->cno++] = (UCHAR_T)*chp;
2457 			return (0);
2458 		}
2459 	} else if (tp->owrite) {		/* Overwrite a character. */
2460 		cno = tp->cno;
2461 
2462 		/*
2463 		 * If the old or new characters are tabs, then the length of the
2464 		 * display depends on the character position in the display.  We
2465 		 * don't even try to handle this here, just ask the screen.
2466 		 */
2467 		if (*chp == '\t') {
2468 			savech = tp->lb[cno];
2469 			tp->lb[cno] = '\t';
2470 			(void)vs_columns(sp, tp->lb, tp->lno, &cno, &nlen);
2471 			tp->lb[cno] = savech;
2472 		} else
2473 			nlen = KEY_COL(sp, *chp);
2474 
2475 		/*
2476 		 * Eat overwrite characters until we run out of them or we've
2477 		 * handled the length of the new character.  If we only eat
2478 		 * part of an overwrite character, break it into its component
2479 		 * elements and display the remaining components.
2480 		 */
2481 		for (copydown = 0; nlen != 0 && tp->owrite != 0;) {
2482 			--tp->owrite;
2483 
2484 			if (tp->lb[cno] == '\t')
2485 				(void)vs_columns(sp,
2486 				    tp->lb, tp->lno, &cno, &olen);
2487 			else
2488 				olen = KEY_COL(sp, tp->lb[cno]);
2489 
2490 			if (olen == nlen) {
2491 				nlen = 0;
2492 				break;
2493 			}
2494 			if (olen < nlen) {
2495 				++copydown;
2496 				nlen -= olen;
2497 			} else {
2498 				BINC_RETW(sp,
2499 				    tp->lb, tp->lb_len, tp->len + olen);
2500 				chlen = olen - nlen;
2501 				MEMMOVEW(tp->lb + cno + 1 + chlen,
2502 				    tp->lb + cno + 1,
2503 				    tp->owrite + tp->insert);
2504 
2505 				tp->len += chlen;
2506 				tp->owrite += chlen;
2507 				/*
2508 				 * Do not rewrite a part of a multi-width char,
2509 				 * which results in screen corruption.
2510 				 */
2511 				if (tp->lb[cno] == '\t'
2512 				    || INTISWIDE(tp->lb[cno]))
2513 					for (p = tp->lb + cno + 1; chlen--;)
2514 						*p++ = ' ';
2515 				else
2516 					for (kp =
2517 					    KEY_NAME(sp, tp->lb[cno]) + nlen,
2518 					    p = tp->lb + cno + 1; chlen--;)
2519 						*p++ = *kp++;
2520 				nlen = 0;
2521 				break;
2522 			}
2523 		}
2524 
2525 		/*
2526 		 * If had to erase several characters, we adjust the total
2527 		 * count, and if there are any characters left, shift them
2528 		 * into position.
2529 		 */
2530 		if (copydown != 0 && (tp->len -= copydown) != 0)
2531 			MEMMOVEW(tp->lb + cno, tp->lb + cno + copydown,
2532 			    tp->owrite + tp->insert + copydown);
2533 
2534 		/* If we had enough overwrite characters, we're done. */
2535 		if (nlen == 0) {
2536 			tp->lb[tp->cno++] = (UCHAR_T)*chp;
2537 			return (0);
2538 		}
2539 	}
2540 
2541 	/* Check to see if the character fits into the input buffer. */
2542 	BINC_RETW(sp, tp->lb, tp->lb_len, tp->len + 1);
2543 
2544 	++tp->len;
2545 	if (tp->insert) {			/* Insert a character. */
2546 		if (tp->insert == 1)
2547 			tp->lb[tp->cno + 1] = tp->lb[tp->cno];
2548 		else
2549 			MEMMOVEW(tp->lb + tp->cno + 1,
2550 			    tp->lb + tp->cno, tp->owrite + tp->insert);
2551 	}
2552 	tp->lb[tp->cno++] = (UCHAR_T)*chp;
2553 	return (0);
2554 }
2555 
2556 /*
2557  * txt_isrch --
2558  *	Do an incremental search.
2559  */
2560 static int
txt_isrch(SCR * sp,VICMD * vp,TEXT * tp,u_int8_t * is_flagsp)2561 txt_isrch(SCR *sp, VICMD *vp, TEXT *tp, u_int8_t *is_flagsp)
2562 {
2563 	MARK start;
2564 	db_recno_t lno;
2565 	u_int sf;
2566 
2567 	/* If it's a one-line screen, we don't do incrementals. */
2568 	if (IS_ONELINE(sp)) {
2569 		FL_CLR(*is_flagsp, IS_RUNNING);
2570 		return (0);
2571 	}
2572 
2573 	/*
2574 	 * If the user erases back to the beginning of the buffer, there's
2575 	 * nothing to search for.  Reset the cursor to the starting point.
2576 	 */
2577 	if (tp->cno <= 1) {
2578 		vp->m_final = vp->m_start;
2579 		return (0);
2580 	}
2581 
2582 	/*
2583 	 * If it's an RE quote character, and not quoted, ignore it until
2584 	 * we get another character.
2585 	 */
2586 	if (tp->lb[tp->cno - 1] == '\\' &&
2587 	    (tp->cno == 2 || tp->lb[tp->cno - 2] != '\\'))
2588 		return (0);
2589 
2590 	/*
2591 	 * If it's a magic shell character, and not quoted, reset the cursor
2592 	 * to the starting point.
2593 	 */
2594 	if (strchr(O_STR(sp, O_SHELLMETA), tp->lb[tp->cno - 1]) != NULL &&
2595 	    (tp->cno == 2 || tp->lb[tp->cno - 2] != '\\'))
2596 		vp->m_final = vp->m_start;
2597 
2598 	/*
2599 	 * If we see the search pattern termination character, then quit doing
2600 	 * an incremental search.  There may be more, e.g., ":/foo/;/bar/",
2601 	 * and we can't handle that incrementally.  Also, reset the cursor to
2602 	 * the original location, the ex search routines don't know anything
2603 	 * about incremental searches.
2604 	 */
2605 	if (tp->lb[0] == tp->lb[tp->cno - 1] &&
2606 	    (tp->cno == 2 || tp->lb[tp->cno - 2] != '\\')) {
2607 		vp->m_final = vp->m_start;
2608 		FL_CLR(*is_flagsp, IS_RUNNING);
2609 		return (0);
2610 	}
2611 
2612 	/*
2613 	 * Remember the input line and discard the special input map,
2614 	 * but don't overwrite the input line on the screen.
2615 	 */
2616 	lno = tp->lno;
2617 	F_SET(VIP(sp), VIP_S_MODELINE);
2618 	F_CLR(sp, SC_TINPUT | SC_TINPUT_INFO);
2619 	if (txt_map_end(sp))
2620 		return (1);
2621 
2622 	/*
2623 	 * Specify a starting point and search.  If we find a match, move to
2624 	 * it and refresh the screen.  If we didn't find the match, then we
2625 	 * beep the screen.  When searching from the original cursor position,
2626 	 * we have to move the cursor, otherwise, we don't want to move the
2627 	 * cursor in case the text at the current position continues to match.
2628 	 */
2629 	if (FL_ISSET(*is_flagsp, IS_RESTART)) {
2630 		start = vp->m_start;
2631 		sf = SEARCH_SET;
2632 	} else {
2633 		start = vp->m_final;
2634 		sf = SEARCH_INCR | SEARCH_SET;
2635 	}
2636 
2637 	if (tp->lb[0] == '/' ?
2638 	    !f_search(sp,
2639 	    &start, &vp->m_final, tp->lb + 1, tp->cno - 1, NULL, sf) :
2640 	    !b_search(sp,
2641 	    &start, &vp->m_final, tp->lb + 1, tp->cno - 1, NULL, sf)) {
2642 		sp->lno = vp->m_final.lno;
2643 		sp->cno = vp->m_final.cno;
2644 		FL_CLR(*is_flagsp, IS_RESTART);
2645 
2646 		if (!KEYS_WAITING(sp) && vs_refresh(sp, 0))
2647 			return (1);
2648 	} else
2649 		FL_SET(*is_flagsp, IS_RESTART);
2650 
2651 	/* Reinstantiate the special input map. */
2652 	if (txt_map_init(sp))
2653 		return (1);
2654 	F_CLR(VIP(sp), VIP_S_MODELINE);
2655 	F_SET(sp, SC_TINPUT | SC_TINPUT_INFO);
2656 
2657 	/* Reset the line number of the input line. */
2658 	tp->lno = TMAP[0].lno;
2659 
2660 	/*
2661 	 * If the colon command-line moved, i.e. the screen scrolled,
2662 	 * refresh the input line.
2663 	 *
2664 	 * XXX
2665 	 * We shouldn't be calling vs_line, here -- we need dirty bits
2666 	 * on entries in the SMAP array.
2667 	 */
2668 	if (lno != TMAP[0].lno) {
2669 		if (vs_line(sp, &TMAP[0], NULL, NULL))
2670 			return (1);
2671 		(void)sp->gp->scr_refresh(sp, 0);
2672 	}
2673 	return (0);
2674 }
2675 
2676 /*
2677  * txt_resolve --
2678  *	Resolve the input text chain into the file.
2679  */
2680 static int
txt_resolve(SCR * sp,TEXTH * tiqh,u_int32_t flags)2681 txt_resolve(SCR *sp, TEXTH *tiqh, u_int32_t flags)
2682 {
2683 	TEXT *tp;
2684 	db_recno_t lno;
2685 	int changed;
2686 
2687 	/*
2688 	 * The first line replaces a current line, and all subsequent lines
2689 	 * are appended into the file.  Resolve autoindented characters for
2690 	 * each line before committing it.  If the latter causes the line to
2691 	 * change, we have to redisplay it, otherwise the information cached
2692 	 * about the line will be wrong.
2693 	 */
2694 	tp = TAILQ_FIRST(tiqh);
2695 
2696 	if (LF_ISSET(TXT_AUTOINDENT))
2697 		txt_ai_resolve(sp, tp, &changed);
2698 	else
2699 		changed = 0;
2700 	if (db_set(sp, tp->lno, tp->lb, tp->len) ||
2701 	    (changed && vs_change(sp, tp->lno, LINE_RESET)))
2702 		return (1);
2703 
2704 	for (lno = tp->lno; (tp = TAILQ_NEXT(tp, q)) != NULL; ++lno) {
2705 		if (LF_ISSET(TXT_AUTOINDENT))
2706 			txt_ai_resolve(sp, tp, &changed);
2707 		else
2708 			changed = 0;
2709 		if (db_append(sp, 0, lno, tp->lb, tp->len) ||
2710 		    (changed && vs_change(sp, tp->lno, LINE_RESET)))
2711 			return (1);
2712 	}
2713 
2714 	/*
2715 	 * Clear the input flag, the look-aside buffer is no longer valid.
2716 	 * Has to be done as part of text resolution, or upon return we'll
2717 	 * be looking at incorrect data.
2718 	 */
2719 	F_CLR(sp, SC_TINPUT);
2720 
2721 	return (0);
2722 }
2723 
2724 /*
2725  * txt_showmatch --
2726  *	Show a character match.
2727  *
2728  * !!!
2729  * Historic vi tried to display matches even in the :colon command line.
2730  * I think not.
2731  */
2732 static int
txt_showmatch(SCR * sp,TEXT * tp)2733 txt_showmatch(SCR *sp, TEXT *tp)
2734 {
2735 	VCS cs;
2736 	MARK m;
2737 	int cnt, endc, startc;
2738 
2739 	/*
2740 	 * Do a refresh first, in case we haven't done one in awhile,
2741 	 * so the user can see what we're complaining about.
2742 	 */
2743 	UPDATE_POSITION(sp, tp);
2744 	if (vs_refresh(sp, 1))
2745 		return (1);
2746 
2747 	/*
2748 	 * We don't display the match if it's not on the screen.  Find
2749 	 * out what the first character on the screen is.
2750 	 */
2751 	if (vs_sm_position(sp, &m, 0, P_TOP))
2752 		return (1);
2753 
2754 	/* Initialize the getc() interface. */
2755 	cs.cs_lno = tp->lno;
2756 	cs.cs_cno = tp->cno - 1;
2757 	if (cs_init(sp, &cs))
2758 		return (1);
2759 	startc = (endc = cs.cs_ch)  == ')' ? '(' : '{';
2760 
2761 	/* Search for the match. */
2762 	for (cnt = 1;;) {
2763 		if (cs_prev(sp, &cs))
2764 			return (1);
2765 		if (cs.cs_flags != 0) {
2766 			if (cs.cs_flags == CS_EOF || cs.cs_flags == CS_SOF) {
2767 				msgq(sp, M_BERR,
2768 				    "Unmatched %s", KEY_NAME(sp, endc));
2769 				return (0);
2770 			}
2771 			continue;
2772 		}
2773 		if (cs.cs_ch == endc)
2774 			++cnt;
2775 		else if (cs.cs_ch == startc && --cnt == 0)
2776 			break;
2777 	}
2778 
2779 	/* If the match is on the screen, move to it. */
2780 	if (cs.cs_lno < m.lno || (cs.cs_lno == m.lno && cs.cs_cno < m.cno))
2781 		return (0);
2782 	sp->lno = cs.cs_lno;
2783 	sp->cno = cs.cs_cno;
2784 	if (vs_refresh(sp, 1))
2785 		return (1);
2786 
2787 	/* Wait for timeout or character arrival. */
2788 	return (v_event_get(sp,
2789 	    NULL, O_VAL(sp, O_MATCHTIME) * 100, EC_TIMEOUT));
2790 }
2791 
2792 /*
2793  * txt_margin --
2794  *	Handle margin wrap.
2795  */
2796 static int
txt_margin(SCR * sp,TEXT * tp,TEXT * wmtp,int * didbreak,u_int32_t flags)2797 txt_margin(SCR *sp, TEXT *tp, TEXT *wmtp, int *didbreak, u_int32_t flags)
2798 {
2799 	size_t len, off;
2800 	CHAR_T *p;
2801 
2802 	/* Find the nearest previous blank. */
2803 	for (off = tp->cno - 1, p = tp->lb + off, len = 0;; --off, --p, ++len) {
2804 		if (ISBLANK((UCHAR_T)*p)) {
2805 			break;
2806 		}
2807 
2808 		/*
2809 		 * If reach the start of the line, there's nowhere to break.
2810 		 *
2811 		 * !!!
2812 		 * Historic vi belled each time a character was entered after
2813 		 * crossing the margin until a space was entered which could
2814 		 * be used to break the line.  I don't as it tends to wake the
2815 		 * cats.
2816 		 */
2817 		if (off == tp->ai || off == tp->offset) {
2818 			*didbreak = 0;
2819 			return (0);
2820 		}
2821 	}
2822 
2823 	/*
2824 	 * Store saved information about the rest of the line in the
2825 	 * wrapmargin TEXT structure.
2826 	 *
2827 	 * !!!
2828 	 * The offset field holds the length of the current characters
2829 	 * that the user entered, but which are getting split to the new
2830 	 * line -- it's going to be used to set the cursor value when we
2831 	 * move to the new line.
2832 	 */
2833 	wmtp->lb = p + 1;
2834 	wmtp->offset = len;
2835 	wmtp->insert = LF_ISSET(TXT_APPENDEOL) ?  tp->insert - 1 : tp->insert;
2836 	wmtp->owrite = tp->owrite;
2837 
2838 	/* Correct current bookkeeping information. */
2839 	tp->cno -= len;
2840 	if (LF_ISSET(TXT_APPENDEOL)) {
2841 		tp->len -= len + tp->owrite + (tp->insert - 1);
2842 		tp->insert = 1;
2843 	} else {
2844 		tp->len -= len + tp->owrite + tp->insert;
2845 		tp->insert = 0;
2846 	}
2847 	tp->owrite = 0;
2848 
2849 	/*
2850 	 * !!!
2851 	 * Delete any trailing whitespace from the current line.
2852 	 */
2853 	for (;; --p, --off) {
2854 		if (!ISBLANK((UCHAR_T)*p))
2855 			break;
2856 		--tp->cno;
2857 		--tp->len;
2858 		if (off == tp->ai || off == tp->offset)
2859 			break;
2860 	}
2861 	*didbreak = 1;
2862 	return (0);
2863 }
2864 
2865 /*
2866  * txt_Rresolve --
2867  *	Resolve the input line for the 'R' command.
2868  */
2869 static void
txt_Rresolve(SCR * sp,TEXTH * tiqh,TEXT * tp,const size_t orig_len)2870 txt_Rresolve(SCR *sp, TEXTH *tiqh, TEXT *tp, const size_t orig_len)
2871 {
2872 	TEXT *ttp;
2873 	size_t input_len, retain;
2874 	CHAR_T *p;
2875 
2876 	/*
2877 	 * Check to make sure that the cursor hasn't moved beyond
2878 	 * the end of the line.
2879 	 */
2880 	if (tp->owrite == 0)
2881 		return;
2882 
2883 	/*
2884 	 * Calculate how many characters the user has entered,
2885 	 * plus the blanks erased by <carriage-return>/<newline>s.
2886 	 */
2887 	for (ttp = TAILQ_FIRST(tiqh), input_len = 0;;) {
2888 		input_len += ttp == tp ? tp->cno : ttp->len + ttp->R_erase;
2889 		if ((ttp = TAILQ_NEXT(ttp, q)) == NULL)
2890 			break;
2891 	}
2892 
2893 	/*
2894 	 * If the user has entered less characters than the original line
2895 	 * was long, restore any overwriteable characters to the original
2896 	 * characters.  These characters are entered as "insert characters",
2897 	 * because they're after the cursor and we don't want to lose them.
2898 	 * (This is okay because the R command has no insert characters.)
2899 	 * We set owrite to 0 so that the insert characters don't get copied
2900 	 * to somewhere else, which means that the line and the length have
2901 	 * to be adjusted here as well.
2902 	 *
2903 	 * We have to retrieve the original line because the original pinned
2904 	 * page has long since been discarded.  If it doesn't exist, that's
2905 	 * okay, the user just extended the file.
2906 	 */
2907 	if (input_len < orig_len) {
2908 		retain = MIN(tp->owrite, orig_len - input_len);
2909 		if (db_get(sp,
2910 		    TAILQ_FIRST(tiqh)->lno, DBG_FATAL | DBG_NOCACHE, &p, NULL))
2911 			return;
2912 		MEMCPYW(tp->lb + tp->cno, p + input_len, retain);
2913 		tp->len -= tp->owrite - retain;
2914 		tp->owrite = 0;
2915 		tp->insert += retain;
2916 	}
2917 }
2918 
2919 /*
2920  * txt_nomorech --
2921  *	No more characters message.
2922  */
2923 static void
txt_nomorech(SCR * sp)2924 txt_nomorech(SCR *sp)
2925 {
2926 	msgq(sp, M_BERR, "194|No more characters to erase");
2927 }
2928