xref: /openbsd/usr.bin/vi/ex/ex.c (revision 80ddc267)
1 /*	$OpenBSD: ex.c,v 1.23 2023/06/23 15:06:45 millert Exp $	*/
2 
3 /*-
4  * Copyright (c) 1992, 1993, 1994
5  *	The Regents of the University of California.  All rights reserved.
6  * Copyright (c) 1992, 1993, 1994, 1995, 1996
7  *	Keith Bostic.  All rights reserved.
8  *
9  * See the LICENSE file for redistribution information.
10  */
11 
12 #include "config.h"
13 
14 #include <sys/types.h>
15 #include <sys/queue.h>
16 #include <sys/stat.h>
17 #include <sys/time.h>
18 
19 #include <bitstring.h>
20 #include <ctype.h>
21 #include <errno.h>
22 #include <fcntl.h>
23 #include <limits.h>
24 #include <stdio.h>
25 #include <stdlib.h>
26 #include <string.h>
27 #include <unistd.h>
28 
29 #include "../common/common.h"
30 #include "../vi/vi.h"
31 
32 #if defined(DEBUG) && defined(COMLOG)
33 static void	ex_comlog(SCR *, EXCMD *);
34 #endif
35 static EXCMDLIST const *
36 		ex_comm_search(char *, size_t);
37 static int	ex_discard(SCR *);
38 static int	ex_line(SCR *, EXCMD *, MARK *, int *, int *);
39 static int	ex_load(SCR *);
40 static void	ex_unknown(SCR *, char *, size_t);
41 
42 /*
43  * ex --
44  *	Main ex loop.
45  *
46  * PUBLIC: int ex(SCR **);
47  */
48 int
ex(SCR ** spp)49 ex(SCR **spp)
50 {
51 	GS *gp;
52 	MSGS *mp;
53 	SCR *sp;
54 	TEXT *tp;
55 	u_int32_t flags;
56 
57 	sp = *spp;
58 	gp = sp->gp;
59 
60 	/* Start the ex screen. */
61 	if (ex_init(sp))
62 		return (1);
63 
64 	/* Flush any saved messages. */
65 	while ((mp = LIST_FIRST(&gp->msgq)) != NULL) {
66 		gp->scr_msg(sp, mp->mtype, mp->buf, mp->len);
67 		LIST_REMOVE(mp, q);
68 		free(mp->buf);
69 		free(mp);
70 	}
71 
72 	/* If reading from a file, errors should have name and line info. */
73 	if (F_ISSET(gp, G_SCRIPTED)) {
74 		gp->excmd.if_lno = 1;
75 		gp->excmd.if_name = "script";
76 	}
77 
78 	/*
79 	 * !!!
80 	 * Initialize the text flags.  The beautify edit option historically
81 	 * applied to ex command input read from a file.  In addition, the
82 	 * first time a ^H was discarded from the input, there was a message,
83 	 * "^H discarded", that was displayed.  We don't bother.
84 	 */
85 	LF_INIT(TXT_BACKSLASH | TXT_CNTRLD | TXT_CR);
86 	for (;; ++gp->excmd.if_lno) {
87 		/* Display status line and flush. */
88 		if (F_ISSET(sp, SC_STATUS)) {
89 			if (!F_ISSET(sp, SC_EX_SILENT))
90 				msgq_status(sp, sp->lno, 0);
91 			F_CLR(sp, SC_STATUS);
92 		}
93 		(void)ex_fflush(sp);
94 
95 		/* Set the flags the user can reset. */
96 		if (O_ISSET(sp, O_BEAUTIFY))
97 			LF_SET(TXT_BEAUTIFY);
98 		if (O_ISSET(sp, O_PROMPT))
99 			LF_SET(TXT_PROMPT);
100 
101 		/* Clear any current interrupts, and get a command. */
102 		CLR_INTERRUPT(sp);
103 		if (ex_txt(sp, &sp->tiq, ':', flags))
104 			return (1);
105 		if (INTERRUPTED(sp)) {
106 			(void)ex_puts(sp, "\n");
107 			(void)ex_fflush(sp);
108 			continue;
109 		}
110 
111 		/* Initialize the command structure. */
112 		CLEAR_EX_PARSER(&gp->excmd);
113 
114 		/*
115 		 * If the user entered a single carriage return, send
116 		 * ex_cmd() a separator -- it discards single newlines.
117 		 */
118 		tp = TAILQ_FIRST(&sp->tiq);
119 		if (tp->len == 0) {
120 			gp->excmd.cp = " ";	/* __TK__ why not |? */
121 			gp->excmd.clen = 1;
122 		} else {
123 			gp->excmd.cp = tp->lb;
124 			gp->excmd.clen = tp->len;
125 		}
126 		F_INIT(&gp->excmd, E_NRSEP);
127 
128 		if (ex_cmd(sp) && F_ISSET(gp, G_SCRIPTED))
129 			return (1);
130 
131 		if (INTERRUPTED(sp)) {
132 			CLR_INTERRUPT(sp);
133 			msgq(sp, M_ERR, "Interrupted");
134 		}
135 
136 		/* Sync recovery if changes were made. */
137 		if (F_ISSET(sp->ep, F_RCV_SYNC))
138 			rcv_sync(sp, 0);
139 
140 		/*
141 		 * If the last command caused a restart, or switched screens
142 		 * or into vi, return.
143 		 */
144 		if (F_ISSET(gp, G_SRESTART) || F_ISSET(sp, SC_SSWITCH | SC_VI)) {
145 			*spp = sp;
146 			break;
147 		}
148 
149 		/* If the last command switched files, we don't care. */
150 		F_CLR(sp, SC_FSWITCH);
151 
152 		/*
153 		 * If we're exiting this screen, move to the next one.  By
154 		 * definition, this means returning into vi, so return to the
155 		 * main editor loop.  The ordering is careful, don't discard
156 		 * the contents of sp until the end.
157 		 */
158 		if (F_ISSET(sp, SC_EXIT | SC_EXIT_FORCE)) {
159 			if (file_end(sp, NULL, F_ISSET(sp, SC_EXIT_FORCE)))
160 				return (1);
161 			*spp = screen_next(sp);
162 			return (screen_end(sp));
163 		}
164 	}
165 	return (0);
166 }
167 
168 /*
169  * ex_cmd --
170  *	The guts of the ex parser: parse and execute a string containing
171  *	ex commands.
172  *
173  * !!!
174  * This code MODIFIES the string that gets passed in, to delete quoting
175  * characters, etc.  The string cannot be readonly/text space, nor should
176  * you expect to use it again after ex_cmd() returns.
177  *
178  * !!!
179  * For the fun of it, if you want to see if a vi clone got the ex argument
180  * parsing right, try:
181  *
182  *	echo 'foo|bar' > file1; echo 'foo/bar' > file2;
183  *	vi
184  *	:edit +1|s/|/PIPE/|w file1| e file2|1 | s/\//SLASH/|wq
185  *
186  * or:	vi
187  *	:set|file|append|set|file
188  *
189  * For extra credit, try them in a startup .exrc file.
190  *
191  * PUBLIC: int ex_cmd(SCR *);
192  */
193 int
ex_cmd(SCR * sp)194 ex_cmd(SCR *sp)
195 {
196 	enum nresult nret;
197 	EX_PRIVATE *exp;
198 	EXCMD *ecp;
199 	GS *gp;
200 	MARK cur;
201 	recno_t lno;
202 	size_t arg1_len, discard, len;
203 	u_int32_t flags;
204 	long ltmp;
205 	int at_found, gv_found;
206 	int ch, cnt, delim, isaddr, namelen;
207 	int newscreen, notempty, tmp, vi_address;
208 	char *arg1, *p, *s, *t;
209 
210 	gp = sp->gp;
211 	exp = EXP(sp);
212 
213 	/*
214 	 * We always start running the command on the top of the stack.
215 	 * This means that *everything* must be resolved when we leave
216 	 * this function for any reason.
217 	 */
218 loop:	ecp = LIST_FIRST(&gp->ecq);
219 
220 	/* If we're reading a command from a file, set up error information. */
221 	if (ecp->if_name != NULL) {
222 		gp->if_lno = ecp->if_lno;
223 		gp->if_name = ecp->if_name;
224 	}
225 
226 	/*
227 	 * If a move to the end of the file is scheduled for this command,
228 	 * do it now.
229 	 */
230 	if (F_ISSET(ecp, E_MOVETOEND)) {
231 		if (db_last(sp, &sp->lno))
232 			goto rfail;
233 		sp->cno = 0;
234 		F_CLR(ecp, E_MOVETOEND);
235 	}
236 
237 	/* If we found a newline, increment the count now. */
238 	if (F_ISSET(ecp, E_NEWLINE)) {
239 		++gp->if_lno;
240 		++ecp->if_lno;
241 		F_CLR(ecp, E_NEWLINE);
242 	}
243 
244 	/* (Re)initialize the EXCMD structure, preserving some flags. */
245 	CLEAR_EX_CMD(ecp);
246 
247 	/* Initialize the argument structures. */
248 	if (argv_init(sp, ecp))
249 		goto err;
250 
251 	/* Initialize +cmd, saved command information. */
252 	arg1 = NULL;
253 	ecp->save_cmdlen = 0;
254 
255 	/* Skip <blank>s, empty lines.  */
256 	for (notempty = 0; ecp->clen > 0; ++ecp->cp, --ecp->clen)
257 		if ((ch = *ecp->cp) == '\n') {
258 			++gp->if_lno;
259 			++ecp->if_lno;
260 		} else if (isblank(ch))
261 			notempty = 1;
262 		else
263 			break;
264 
265 	/*
266 	 * !!!
267 	 * Permit extra colons at the start of the line.  Historically,
268 	 * ex/vi allowed a single extra one.  It's simpler not to count.
269 	 * The stripping is done here because, historically, any command
270 	 * could have preceding colons, e.g. ":g/pattern/:p" worked.
271 	 */
272 	if (ecp->clen != 0 && ch == ':') {
273 		notempty = 1;
274 		while (--ecp->clen > 0 && (ch = *++ecp->cp) == ':');
275 	}
276 
277 	/*
278 	 * Command lines that start with a double-quote are comments.
279 	 *
280 	 * !!!
281 	 * Historically, there was no escape or delimiter for a comment, e.g.
282 	 * :"foo|set was a single comment and nothing was output.  Since nvi
283 	 * permits users to escape <newline> characters into command lines, we
284 	 * have to check for that case.
285 	 */
286 	if (ecp->clen != 0 && ch == '"') {
287 		while (--ecp->clen > 0 && *++ecp->cp != '\n');
288 		if (*ecp->cp == '\n') {
289 			F_SET(ecp, E_NEWLINE);
290 			++ecp->cp;
291 			--ecp->clen;
292 		}
293 		goto loop;
294 	}
295 
296 	/* Skip whitespace. */
297 	for (; ecp->clen > 0; ++ecp->cp, --ecp->clen) {
298 		ch = *ecp->cp;
299 		if (!isblank(ch))
300 			break;
301 	}
302 
303 	/*
304 	 * The last point at which an empty line can mean do nothing.
305 	 *
306 	 * !!!
307 	 * Historically, in ex mode, lines containing only <blank> characters
308 	 * were the same as a single <carriage-return>, i.e. a default command.
309 	 * In vi mode, they were ignored.  In .exrc files this was a serious
310 	 * annoyance, as vi kept trying to treat them as print commands.  We
311 	 * ignore backward compatibility in this case, discarding lines that
312 	 * contain only <blank> characters from .exrc files.
313 	 *
314 	 * !!!
315 	 * This is where you end up when you're done a command, i.e. clen has
316 	 * gone to zero.  Continue if there are more commands to run.
317 	 */
318 	if (ecp->clen == 0 &&
319 	    (!notempty || F_ISSET(sp, SC_VI) || F_ISSET(ecp, E_BLIGNORE))) {
320 		if (ex_load(sp))
321 			goto rfail;
322 		ecp = LIST_FIRST(&gp->ecq);
323 		if (ecp->clen == 0)
324 			goto rsuccess;
325 		goto loop;
326 	}
327 
328 	/*
329 	 * Check to see if this is a command for which we may want to move
330 	 * the cursor back up to the previous line.  (The command :1<CR>
331 	 * wants a <newline> separator, but the command :<CR> wants to erase
332 	 * the command line.)  If the line is empty except for <blank>s,
333 	 * <carriage-return> or <eof>, we'll probably want to move up.  I
334 	 * don't think there's any way to get <blank> characters *after* the
335 	 * command character, but this is the ex parser, and I've been wrong
336 	 * before.
337 	 */
338 	if (F_ISSET(ecp, E_NRSEP) &&
339 	    ecp->clen != 0 && (ecp->clen != 1 || ecp->cp[0] != '\004'))
340 		F_CLR(ecp, E_NRSEP);
341 
342 	/* Parse command addresses. */
343 	if (ex_range(sp, ecp, &tmp))
344 		goto rfail;
345 	if (tmp)
346 		goto err;
347 
348 	/*
349 	 * Skip <blank>s and any more colons (the command :3,5:print
350 	 * worked, historically).
351 	 */
352 	for (; ecp->clen > 0; ++ecp->cp, --ecp->clen) {
353 		ch = *ecp->cp;
354 		if (!isblank(ch) && ch != ':')
355 			break;
356 	}
357 
358 	/*
359 	 * If no command, ex does the last specified of p, l, or #, and vi
360 	 * moves to the line.  Otherwise, determine the length of the command
361 	 * name by looking for the first non-alphabetic character.  (There
362 	 * are a few non-alphabetic characters in command names, but they're
363 	 * all single character commands.)  This isn't a great test, because
364 	 * it means that, for the command ":e +cut.c file", we'll report that
365 	 * the command "cut" wasn't known.  However, it makes ":e+35 file" work
366 	 * correctly.
367 	 *
368 	 * !!!
369 	 * Historically, lines with multiple adjacent (or <blank> separated)
370 	 * command separators were very strange.  For example, the command
371 	 * |||<carriage-return>, when the cursor was on line 1, displayed
372 	 * lines 2, 3 and 5 of the file.  In addition, the command "   |  "
373 	 * would only display the line after the next line, instead of the
374 	 * next two lines.  No ideas why.  It worked reasonably when executed
375 	 * from vi mode, and displayed lines 2, 3, and 4, so we do a default
376 	 * command for each separator.
377 	 */
378 #define	SINGLE_CHAR_COMMANDS	"\004!#&*<=>@~"
379 	newscreen = 0;
380 	if (ecp->clen != 0 && ecp->cp[0] != '|' && ecp->cp[0] != '\n') {
381 		if (strchr(SINGLE_CHAR_COMMANDS, *ecp->cp)) {
382 			p = ecp->cp;
383 			++ecp->cp;
384 			--ecp->clen;
385 			namelen = 1;
386 		} else {
387 			for (p = ecp->cp;
388 			    ecp->clen > 0; --ecp->clen, ++ecp->cp)
389 				if (!isalpha(*ecp->cp))
390 					break;
391 			if ((namelen = ecp->cp - p) == 0) {
392 				msgq(sp, M_ERR, "Unknown command name");
393 				goto err;
394 			}
395 		}
396 
397 		/*
398 		 * !!!
399 		 * Historic vi permitted flags to immediately follow any
400 		 * subset of the 'delete' command, but then did not permit
401 		 * further arguments (flag, buffer, count).  Make it work.
402 		 * Permit further arguments for the few shreds of dignity
403 		 * it offers.
404 		 *
405 		 * Adding commands that start with 'd', and match "delete"
406 		 * up to a l, p, +, - or # character can break this code.
407 		 *
408 		 * !!!
409 		 * Capital letters beginning the command names ex, edit,
410 		 * next, previous, tag and visual (in vi mode) indicate the
411 		 * command should happen in a new screen.
412 		 */
413 		switch (p[0]) {
414 		case 'd':
415 			for (s = p,
416 			    t = cmds[C_DELETE].name; *s == *t; ++s, ++t);
417 			if (s[0] == 'l' || s[0] == 'p' || s[0] == '+' ||
418 			    s[0] == '-' || s[0] == '^' || s[0] == '#') {
419 				len = (ecp->cp - p) - (s - p);
420 				ecp->cp -= len;
421 				ecp->clen += len;
422 				ecp->rcmd = cmds[C_DELETE];
423 				ecp->rcmd.syntax = "1bca1";
424 				ecp->cmd = &ecp->rcmd;
425 				goto skip_srch;
426 			}
427 			break;
428 		case 'E': case 'F': case 'N': case 'P': case 'T': case 'V':
429 			newscreen = 1;
430 			p[0] = tolower(p[0]);
431 			break;
432 		}
433 
434 		/*
435 		 * Search the table for the command.
436 		 *
437 		 * !!!
438 		 * Historic vi permitted the mark to immediately follow the
439 		 * 'k' in the 'k' command.  Make it work.
440 		 *
441 		 * !!!
442 		 * Historic vi permitted any flag to follow the s command, e.g.
443 		 * "s/e/E/|s|sgc3p" was legal.  Make the command "sgc" work.
444 		 * Since the following characters all have to be flags, i.e.
445 		 * alphabetics, we can let the s command routine return errors
446 		 * if it was some illegal command string.  This code will break
447 		 * if an "sg" or similar command is ever added.  The substitute
448 		 * code doesn't care if it's a "cgr" flag or a "#lp" flag that
449 		 * follows the 's', but we limit the choices here to "cgr" so
450 		 * that we get unknown command messages for wrong combinations.
451 		 */
452 		if ((ecp->cmd = ex_comm_search(p, namelen)) == NULL)
453 			switch (p[0]) {
454 			case 'k':
455 				if (namelen == 2) {
456 					ecp->cp -= namelen - 1;
457 					ecp->clen += namelen - 1;
458 					ecp->cmd = &cmds[C_K];
459 					break;
460 				}
461 				goto unknown;
462 			case 's':
463 				for (s = p + 1, cnt = namelen; --cnt; ++s)
464 					if (s[0] != 'c' &&
465 					    s[0] != 'g' && s[0] != 'r')
466 						break;
467 				if (cnt == 0) {
468 					ecp->cp -= namelen - 1;
469 					ecp->clen += namelen - 1;
470 					ecp->rcmd = cmds[C_SUBSTITUTE];
471 					ecp->rcmd.fn = ex_subagain;
472 					ecp->cmd = &ecp->rcmd;
473 					break;
474 				}
475 				/* FALLTHROUGH */
476 			default:
477 unknown:			if (newscreen)
478 					p[0] = toupper(p[0]);
479 				ex_unknown(sp, p, namelen);
480 				goto err;
481 			}
482 
483 		/*
484 		 * The visual command has a different syntax when called
485 		 * from ex than when called from a vi colon command.  FMH.
486 		 * Make the change now, before we test for the newscreen
487 		 * semantic, so that we're testing the right one.
488 		 */
489 skip_srch:	if (ecp->cmd == &cmds[C_VISUAL_EX] && F_ISSET(sp, SC_VI))
490 			ecp->cmd = &cmds[C_VISUAL_VI];
491 
492 		/*
493 		 * !!!
494 		 * Historic vi permitted a capital 'P' at the beginning of
495 		 * any command that started with 'p'.  Probably wanted the
496 		 * P[rint] command for backward compatibility, and the code
497 		 * just made Preserve and Put work by accident.  Nvi uses
498 		 * Previous to mean previous-in-a-new-screen, so be careful.
499 		 */
500 		if (newscreen && !F_ISSET(ecp->cmd, E_NEWSCREEN) &&
501 		    (ecp->cmd == &cmds[C_PRINT] ||
502 		    ecp->cmd == &cmds[C_PRESERVE]))
503 			newscreen = 0;
504 
505 		/* Test for a newscreen associated with this command. */
506 		if (newscreen && !F_ISSET(ecp->cmd, E_NEWSCREEN))
507 			goto unknown;
508 
509 		/* Secure means no shell access. */
510 		if (F_ISSET(ecp->cmd, E_SECURE) && O_ISSET(sp, O_SECURE)) {
511 			ex_emsg(sp, ecp->cmd->name, EXM_SECURE);
512 			goto err;
513 		}
514 
515 		/*
516 		 * Multiple < and > characters; another "feature".  Note,
517 		 * The string passed to the underlying function may not be
518 		 * nul terminated in this case.
519 		 */
520 		if ((ecp->cmd == &cmds[C_SHIFTL] && *p == '<') ||
521 		    (ecp->cmd == &cmds[C_SHIFTR] && *p == '>')) {
522 			for (ch = *p;
523 			    ecp->clen > 0; --ecp->clen, ++ecp->cp)
524 				if (*ecp->cp != ch)
525 					break;
526 			if (argv_exp0(sp, ecp, p, ecp->cp - p))
527 				goto err;
528 		}
529 
530 		/* Set the format style flags for the next command. */
531 		if (ecp->cmd == &cmds[C_HASH])
532 			exp->fdef = E_C_HASH;
533 		else if (ecp->cmd == &cmds[C_LIST])
534 			exp->fdef = E_C_LIST;
535 		else if (ecp->cmd == &cmds[C_PRINT])
536 			exp->fdef = E_C_PRINT;
537 		F_CLR(ecp, E_USELASTCMD);
538 	} else {
539 		/* Print is the default command. */
540 		ecp->cmd = &cmds[C_PRINT];
541 
542 		/* Set the saved format flags. */
543 		F_SET(ecp, exp->fdef);
544 
545 		/*
546 		 * !!!
547 		 * If no address was specified, and it's not a global command,
548 		 * we up the address by one.  (I have no idea why globals are
549 		 * exempted, but it's (ahem) historic practice.)
550 		 */
551 		if (ecp->addrcnt == 0 && !F_ISSET(sp, SC_EX_GLOBAL)) {
552 			ecp->addrcnt = 1;
553 			ecp->addr1.lno = sp->lno + 1;
554 			ecp->addr1.cno = sp->cno;
555 		}
556 
557 		F_SET(ecp, E_USELASTCMD);
558 	}
559 
560 	/*
561 	 * !!!
562 	 * Historically, the number option applied to both ex and vi.  One
563 	 * strangeness was that ex didn't switch display formats until a
564 	 * command was entered, e.g. <CR>'s after the set didn't change to
565 	 * the new format, but :1p would.
566 	 */
567 	if (O_ISSET(sp, O_NUMBER)) {
568 		F_SET(ecp, E_OPTNUM);
569 		FL_SET(ecp->iflags, E_C_HASH);
570 	} else
571 		F_CLR(ecp, E_OPTNUM);
572 
573 	/* Check for ex mode legality. */
574 	if (F_ISSET(sp, SC_EX) && (F_ISSET(ecp->cmd, E_VIONLY) || newscreen)) {
575 		msgq(sp, M_ERR,
576 		    "%s: command not available in ex mode", ecp->cmd->name);
577 		goto err;
578 	}
579 
580 	/* Add standard command flags. */
581 	F_SET(ecp, ecp->cmd->flags);
582 	if (!newscreen)
583 		F_CLR(ecp, E_NEWSCREEN);
584 
585 	/*
586 	 * There are three normal termination cases for an ex command.  They
587 	 * are the end of the string (ecp->clen), or unescaped (by <literal
588 	 * next> characters) <newline> or '|' characters.  As we're now past
589 	 * possible addresses, we can determine how long the command is, so we
590 	 * don't have to look for all the possible terminations.  Naturally,
591 	 * there are some exciting special cases:
592 	 *
593 	 * 1: The bang, global, v and the filter versions of the read and
594 	 *    write commands are delimited by <newline>s (they can contain
595 	 *    shell pipes).
596 	 * 2: The ex, edit, next and visual in vi mode commands all take ex
597 	 *    commands as their first arguments.
598 	 * 3: The s command takes an RE as its first argument, and wants it
599 	 *    to be specially delimited.
600 	 *
601 	 * Historically, '|' characters in the first argument of the ex, edit,
602 	 * next, vi visual, and s commands didn't delimit the command.  And,
603 	 * in the filter cases for read and write, and the bang, global and v
604 	 * commands, they did not delimit the command at all.
605 	 *
606 	 * For example, the following commands were legal:
607 	 *
608 	 *	:edit +25|s/abc/ABC/ file.c
609 	 *	:s/|/PIPE/
610 	 *	:read !spell % | columnate
611 	 *	:global/pattern/p|l
612 	 *
613 	 * It's not quite as simple as it sounds, however.  The command:
614 	 *
615 	 *	:s/a/b/|s/c/d|set
616 	 *
617 	 * was also legal, i.e. the historic ex parser (using the word loosely,
618 	 * since "parser" implies some regularity of syntax) delimited the RE's
619 	 * based on its delimiter and not anything so irretrievably vulgar as a
620 	 * command syntax.
621 	 *
622 	 * Anyhow, the following code makes this all work.  First, for the
623 	 * special cases we move past their special argument(s).  Then, we
624 	 * do normal command processing on whatever is left.  Barf-O-Rama.
625 	 */
626 	discard = 0;		/* Characters discarded from the command. */
627 	arg1_len = 0;
628 	ecp->save_cmd = ecp->cp;
629 	if (ecp->cmd == &cmds[C_EDIT] || ecp->cmd == &cmds[C_EX] ||
630 	    ecp->cmd == &cmds[C_NEXT] || ecp->cmd == &cmds[C_VISUAL_VI]) {
631 		/*
632 		 * Move to the next non-whitespace character.  A '!'
633 		 * immediately following the command is eaten as a
634 		 * force flag.
635 		 */
636 		if (ecp->clen > 0 && *ecp->cp == '!') {
637 			++ecp->cp;
638 			--ecp->clen;
639 			FL_SET(ecp->iflags, E_C_FORCE);
640 
641 			/* Reset, don't reparse. */
642 			ecp->save_cmd = ecp->cp;
643 		}
644 		for (; ecp->clen > 0; --ecp->clen, ++ecp->cp)
645 			if (!isblank(*ecp->cp))
646 				break;
647 		/*
648 		 * QUOTING NOTE:
649 		 *
650 		 * The historic implementation ignored all escape characters
651 		 * so there was no way to put a space or newline into the +cmd
652 		 * field.  We do a simplistic job of fixing it by moving to the
653 		 * first whitespace character that isn't escaped.  The escaping
654 		 * characters are stripped as no longer useful.
655 		 */
656 		if (ecp->clen > 0 && *ecp->cp == '+') {
657 			++ecp->cp;
658 			--ecp->clen;
659 			for (arg1 = p = ecp->cp;
660 			    ecp->clen > 0; --ecp->clen, ++ecp->cp) {
661 				ch = *ecp->cp;
662 				if (IS_ESCAPE(sp, ecp, ch) &&
663 				    ecp->clen > 1) {
664 					++discard;
665 					--ecp->clen;
666 					ch = *++ecp->cp;
667 				} else if (isblank(ch))
668 					break;
669 				*p++ = ch;
670 			}
671 			arg1_len = ecp->cp - arg1;
672 
673 			/* Reset, so the first argument isn't reparsed. */
674 			ecp->save_cmd = ecp->cp;
675 		}
676 	} else if (ecp->cmd == &cmds[C_BANG] ||
677 	    ecp->cmd == &cmds[C_GLOBAL] || ecp->cmd == &cmds[C_V]) {
678 		/*
679 		 * QUOTING NOTE:
680 		 *
681 		 * We use backslashes to escape <newline> characters, although
682 		 * this wasn't historic practice for the bang command.  It was
683 		 * for the global and v commands, and it's common usage when
684 		 * doing text insert during the command.  Escaping characters
685 		 * are stripped as no longer useful.
686 		 */
687 		for (p = ecp->cp; ecp->clen > 0; --ecp->clen, ++ecp->cp) {
688 			ch = *ecp->cp;
689 			if (ch == '\\' && ecp->clen > 1 && ecp->cp[1] == '\n') {
690 				++discard;
691 				--ecp->clen;
692 				ch = *++ecp->cp;
693 
694 				++gp->if_lno;
695 				++ecp->if_lno;
696 			} else if (ch == '\n')
697 				break;
698 			*p++ = ch;
699 		}
700 	} else if (ecp->cmd == &cmds[C_READ] || ecp->cmd == &cmds[C_WRITE]) {
701 		/*
702 		 * For write commands, if the next character is a <blank>, and
703 		 * the next non-blank character is a '!', it's a filter command
704 		 * and we want to eat everything up to the <newline>.  For read
705 		 * commands, if the next non-blank character is a '!', it's a
706 		 * filter command and we want to eat everything up to the next
707 		 * <newline>.  Otherwise, we're done.
708 		 */
709 		for (tmp = 0; ecp->clen > 0; --ecp->clen, ++ecp->cp) {
710 			ch = *ecp->cp;
711 			if (isblank(ch))
712 				tmp = 1;
713 			else
714 				break;
715 		}
716 		if (ecp->clen > 0 && ch == '!' &&
717 		    (ecp->cmd == &cmds[C_READ] || tmp))
718 			for (; ecp->clen > 0; --ecp->clen, ++ecp->cp)
719 				if (ecp->cp[0] == '\n')
720 					break;
721 	} else if (ecp->cmd == &cmds[C_SUBSTITUTE]) {
722 		/*
723 		 * Move to the next non-whitespace character, we'll use it as
724 		 * the delimiter.  If the character isn't an alphanumeric or
725 		 * a '|', it's the delimiter, so parse it.  Otherwise, we're
726 		 * into something like ":s g", so use the special s command.
727 		 */
728 		for (; ecp->clen > 0; --ecp->clen, ++ecp->cp)
729 			if (!isblank(ecp->cp[0]))
730 				break;
731 
732 		if (isalnum(ecp->cp[0]) || ecp->cp[0] == '|') {
733 			ecp->rcmd = cmds[C_SUBSTITUTE];
734 			ecp->rcmd.fn = ex_subagain;
735 			ecp->cmd = &ecp->rcmd;
736 		} else if (ecp->clen > 0) {
737 			/*
738 			 * QUOTING NOTE:
739 			 *
740 			 * Backslashes quote delimiter characters for RE's.
741 			 * The backslashes are NOT removed since they'll be
742 			 * used by the RE code.  Move to the third delimiter
743 			 * that's not escaped (or the end of the command).
744 			 */
745 			delim = *ecp->cp;
746 			++ecp->cp;
747 			--ecp->clen;
748 			for (cnt = 2; ecp->clen > 0 &&
749 			    cnt != 0; --ecp->clen, ++ecp->cp)
750 				if (ecp->cp[0] == '\\' &&
751 				    ecp->clen > 1) {
752 					++ecp->cp;
753 					--ecp->clen;
754 				} else if (ecp->cp[0] == delim)
755 					--cnt;
756 		}
757 	}
758 
759 	/*
760 	 * Use normal quoting and termination rules to find the end of this
761 	 * command.
762 	 *
763 	 * QUOTING NOTE:
764 	 *
765 	 * Historically, vi permitted ^V's to escape <newline>'s in the .exrc
766 	 * file.  It was almost certainly a bug, but that's what bug-for-bug
767 	 * compatibility means, Grasshopper.  Also, ^V's escape the command
768 	 * delimiters.  Literal next quote characters in front of the newlines,
769 	 * '|' characters or literal next characters are stripped as they're
770 	 * no longer useful.
771 	 */
772 	vi_address = ecp->clen != 0 && ecp->cp[0] != '\n';
773 	for (p = ecp->cp; ecp->clen > 0; --ecp->clen, ++ecp->cp) {
774 		ch = ecp->cp[0];
775 		if (IS_ESCAPE(sp, ecp, ch) && ecp->clen > 1) {
776 			tmp = ecp->cp[1];
777 			if (tmp == '\n' || tmp == '|') {
778 				if (tmp == '\n') {
779 					++gp->if_lno;
780 					++ecp->if_lno;
781 				}
782 				++discard;
783 				--ecp->clen;
784 				++ecp->cp;
785 				ch = tmp;
786 			}
787 		} else if (ch == '\n' || ch == '|') {
788 			if (ch == '\n')
789 				F_SET(ecp, E_NEWLINE);
790 			--ecp->clen;
791 			break;
792 		}
793 		*p++ = ch;
794 	}
795 
796 	/*
797 	 * Save off the next command information, go back to the
798 	 * original start of the command.
799 	 */
800 	p = ecp->cp + 1;
801 	ecp->cp = ecp->save_cmd;
802 	ecp->save_cmd = p;
803 	ecp->save_cmdlen = ecp->clen;
804 	ecp->clen = ((ecp->save_cmd - ecp->cp) - 1) - discard;
805 
806 	/*
807 	 * QUOTING NOTE:
808 	 *
809 	 * The "set tags" command historically used a backslash, not the
810 	 * user's literal next character, to escape whitespace.  Handle
811 	 * it here instead of complicating the argv_exp3() code.  Note,
812 	 * this isn't a particularly complex trap, and if backslashes were
813 	 * legal in set commands, this would have to be much more complicated.
814 	 */
815 	if (ecp->cmd == &cmds[C_SET])
816 		for (p = ecp->cp, len = ecp->clen; len > 0; --len, ++p)
817 			if (*p == '\\')
818 				*p = CH_LITERAL;
819 
820 	/*
821 	 * Set the default addresses.  It's an error to specify an address for
822 	 * a command that doesn't take them.  If two addresses are specified
823 	 * for a command that only takes one, lose the first one.  Two special
824 	 * cases here, some commands take 0 or 2 addresses.  For most of them
825 	 * (the E_ADDR2_ALL flag), 0 defaults to the entire file.  For one
826 	 * (the `!' command, the E_ADDR2_NONE flag), 0 defaults to no lines.
827 	 *
828 	 * Also, if the file is empty, some commands want to use an address of
829 	 * 0, i.e. the entire file is 0 to 0, and the default first address is
830 	 * 0.  Otherwise, an entire file is 1 to N and the default line is 1.
831 	 * Note, we also add the E_ADDR_ZERO flag to the command flags, for the
832 	 * case where the 0 address is only valid if it's a default address.
833 	 *
834 	 * Also, set a flag if we set the default addresses.  Some commands
835 	 * (ex: z) care if the user specified an address or if we just used
836 	 * the current cursor.
837 	 */
838 	switch (F_ISSET(ecp, E_ADDR1 | E_ADDR2 | E_ADDR2_ALL | E_ADDR2_NONE)) {
839 	case E_ADDR1:				/* One address: */
840 		switch (ecp->addrcnt) {
841 		case 0:				/* Default cursor/empty file. */
842 			ecp->addrcnt = 1;
843 			F_SET(ecp, E_ADDR_DEF);
844 			if (F_ISSET(ecp, E_ADDR_ZERODEF)) {
845 				if (db_last(sp, &lno))
846 					goto err;
847 				if (lno == 0) {
848 					ecp->addr1.lno = 0;
849 					F_SET(ecp, E_ADDR_ZERO);
850 				} else
851 					ecp->addr1.lno = sp->lno;
852 			} else
853 				ecp->addr1.lno = sp->lno;
854 			ecp->addr1.cno = sp->cno;
855 			break;
856 		case 1:
857 			break;
858 		case 2:				/* Lose the first address. */
859 			ecp->addrcnt = 1;
860 			ecp->addr1 = ecp->addr2;
861 		}
862 		break;
863 	case E_ADDR2_NONE:			/* Zero/two addresses: */
864 		if (ecp->addrcnt == 0)		/* Default to nothing. */
865 			break;
866 		goto two_addr;
867 	case E_ADDR2_ALL:			/* Zero/two addresses: */
868 		if (ecp->addrcnt == 0) {	/* Default entire/empty file. */
869 			F_SET(ecp, E_ADDR_DEF);
870 			ecp->addrcnt = 2;
871 			if (sp->ep == NULL)
872 				ecp->addr2.lno = 0;
873 			else if (db_last(sp, &ecp->addr2.lno))
874 				goto err;
875 			if (F_ISSET(ecp, E_ADDR_ZERODEF) &&
876 			    ecp->addr2.lno == 0) {
877 				ecp->addr1.lno = 0;
878 				F_SET(ecp, E_ADDR_ZERO);
879 			} else
880 				ecp->addr1.lno = 1;
881 			ecp->addr1.cno = ecp->addr2.cno = 0;
882 			F_SET(ecp, E_ADDR2_ALL);
883 			break;
884 		}
885 		/* FALLTHROUGH */
886 	case E_ADDR2:				/* Two addresses: */
887 two_addr:	switch (ecp->addrcnt) {
888 		case 0:				/* Default cursor/empty file. */
889 			ecp->addrcnt = 2;
890 			F_SET(ecp, E_ADDR_DEF);
891 			if (sp->lno == 1 &&
892 			    F_ISSET(ecp, E_ADDR_ZERODEF)) {
893 				if (db_last(sp, &lno))
894 					goto err;
895 				if (lno == 0) {
896 					ecp->addr1.lno = ecp->addr2.lno = 0;
897 					F_SET(ecp, E_ADDR_ZERO);
898 				} else
899 					ecp->addr1.lno =
900 					    ecp->addr2.lno = sp->lno;
901 			} else
902 				ecp->addr1.lno = ecp->addr2.lno = sp->lno;
903 			ecp->addr1.cno = ecp->addr2.cno = sp->cno;
904 			break;
905 		case 1:				/* Default to first address. */
906 			ecp->addrcnt = 2;
907 			ecp->addr2 = ecp->addr1;
908 			break;
909 		case 2:
910 			break;
911 		}
912 		break;
913 	default:
914 		if (ecp->addrcnt)		/* Error. */
915 			goto usage;
916 	}
917 
918 	/*
919 	 * !!!
920 	 * The ^D scroll command historically scrolled the value of the scroll
921 	 * option or to EOF.  It was an error if the cursor was already at EOF.
922 	 * (Leading addresses were permitted, but were then ignored.)
923 	 */
924 	if (ecp->cmd == &cmds[C_SCROLL]) {
925 		ecp->addrcnt = 2;
926 		ecp->addr1.lno = sp->lno + 1;
927 		ecp->addr2.lno = sp->lno + O_VAL(sp, O_SCROLL);
928 		ecp->addr1.cno = ecp->addr2.cno = sp->cno;
929 		if (db_last(sp, &lno))
930 			goto err;
931 		if (lno != 0 && lno > sp->lno && ecp->addr2.lno > lno)
932 			ecp->addr2.lno = lno;
933 	}
934 
935 	ecp->flagoff = 0;
936 	for (p = ecp->cmd->syntax; *p != '\0'; ++p) {
937 		/*
938 		 * The force flag is sensitive to leading whitespace, i.e.
939 		 * "next !" is different from "next!".  Handle it before
940 		 * skipping leading <blank>s.
941 		 */
942 		if (*p == '!') {
943 			if (ecp->clen > 0 && *ecp->cp == '!') {
944 				++ecp->cp;
945 				--ecp->clen;
946 				FL_SET(ecp->iflags, E_C_FORCE);
947 			}
948 			continue;
949 		}
950 
951 		/* Skip leading <blank>s. */
952 		for (; ecp->clen > 0; --ecp->clen, ++ecp->cp)
953 			if (!isblank(*ecp->cp))
954 				break;
955 		if (ecp->clen == 0)
956 			break;
957 
958 		switch (*p) {
959 		case '1':				/* +, -, #, l, p */
960 			/*
961 			 * !!!
962 			 * Historically, some flags were ignored depending
963 			 * on where they occurred in the command line.  For
964 			 * example, in the command, ":3+++p--#", historic vi
965 			 * acted on the '#' flag, but ignored the '-' flags.
966 			 * It's unambiguous what the flags mean, so we just
967 			 * handle them regardless of the stupidity of their
968 			 * location.
969 			 */
970 			for (; ecp->clen; --ecp->clen, ++ecp->cp)
971 				switch (*ecp->cp) {
972 				case '+':
973 					++ecp->flagoff;
974 					break;
975 				case '-':
976 				case '^':
977 					--ecp->flagoff;
978 					break;
979 				case '#':
980 					F_CLR(ecp, E_OPTNUM);
981 					FL_SET(ecp->iflags, E_C_HASH);
982 					exp->fdef |= E_C_HASH;
983 					break;
984 				case 'l':
985 					FL_SET(ecp->iflags, E_C_LIST);
986 					exp->fdef |= E_C_LIST;
987 					break;
988 				case 'p':
989 					FL_SET(ecp->iflags, E_C_PRINT);
990 					exp->fdef |= E_C_PRINT;
991 					break;
992 				default:
993 					goto end_case1;
994 				}
995 end_case1:		break;
996 		case '2':				/* -, ., +, ^ */
997 		case '3':				/* -, ., +, ^, = */
998 			for (; ecp->clen; --ecp->clen, ++ecp->cp)
999 				switch (*ecp->cp) {
1000 				case '-':
1001 					FL_SET(ecp->iflags, E_C_DASH);
1002 					break;
1003 				case '.':
1004 					FL_SET(ecp->iflags, E_C_DOT);
1005 					break;
1006 				case '+':
1007 					FL_SET(ecp->iflags, E_C_PLUS);
1008 					break;
1009 				case '^':
1010 					FL_SET(ecp->iflags, E_C_CARAT);
1011 					break;
1012 				case '=':
1013 					if (*p == '3') {
1014 						FL_SET(ecp->iflags, E_C_EQUAL);
1015 						break;
1016 					}
1017 					/* FALLTHROUGH */
1018 				default:
1019 					goto end_case23;
1020 				}
1021 end_case23:		break;
1022 		case 'b':				/* buffer */
1023 			/*
1024 			 * !!!
1025 			 * Historically, "d #" was a delete with a flag, not a
1026 			 * delete into the '#' buffer.  If the current command
1027 			 * permits a flag, don't use one as a buffer.  However,
1028 			 * the 'l' and 'p' flags were legal buffer names in the
1029 			 * historic ex, and were used as buffers, not flags.
1030 			 */
1031 			if ((ecp->cp[0] == '+' || ecp->cp[0] == '-' ||
1032 			    ecp->cp[0] == '^' || ecp->cp[0] == '#') &&
1033 			    strchr(p, '1') != NULL)
1034 				break;
1035 			/*
1036 			 * !!!
1037 			 * Digits can't be buffer names in ex commands, or the
1038 			 * command "d2" would be a delete into buffer '2', and
1039 			 * not a two-line deletion.
1040 			 */
1041 			if (!isdigit(ecp->cp[0])) {
1042 				ecp->buffer = *ecp->cp;
1043 				++ecp->cp;
1044 				--ecp->clen;
1045 				FL_SET(ecp->iflags, E_C_BUFFER);
1046 			}
1047 			break;
1048 		case 'c':				/* count [01+a] */
1049 			++p;
1050 			/* Validate any signed value. */
1051 			if (!isdigit(*ecp->cp) && (*p != '+' ||
1052 			    (*ecp->cp != '+' && *ecp->cp != '-')))
1053 				break;
1054 			/* If a signed value, set appropriate flags. */
1055 			if (*ecp->cp == '-')
1056 				FL_SET(ecp->iflags, E_C_COUNT_NEG);
1057 			else if (*ecp->cp == '+')
1058 				FL_SET(ecp->iflags, E_C_COUNT_POS);
1059 			if ((nret =
1060 			    nget_slong(&ltmp, ecp->cp, &t, 10)) != NUM_OK) {
1061 				ex_badaddr(sp, NULL, A_NOTSET, nret);
1062 				goto err;
1063 			}
1064 			if (ltmp == 0 && *p != '0') {
1065 				msgq(sp, M_ERR, "Count may not be zero");
1066 				goto err;
1067 			}
1068 			ecp->clen -= (t - ecp->cp);
1069 			ecp->cp = t;
1070 
1071 			/*
1072 			 * Counts as address offsets occur in commands taking
1073 			 * two addresses.  Historic vi practice was to use
1074 			 * the count as an offset from the *second* address.
1075 			 *
1076 			 * Set a count flag; some underlying commands (see
1077 			 * join) do different things with counts than with
1078 			 * line addresses.
1079 			 */
1080 			if (*p == 'a') {
1081 				ecp->addr1 = ecp->addr2;
1082 				ecp->addr2.lno = ecp->addr1.lno + ltmp - 1;
1083 			} else
1084 				ecp->count = ltmp;
1085 			FL_SET(ecp->iflags, E_C_COUNT);
1086 			break;
1087 		case 'f':				/* file */
1088 			if (argv_exp2(sp, ecp, ecp->cp, ecp->clen))
1089 				goto err;
1090 			goto arg_cnt_chk;
1091 		case 'l':				/* line */
1092 			/*
1093 			 * Get a line specification.
1094 			 *
1095 			 * If the line was a search expression, we may have
1096 			 * changed state during the call, and we're now
1097 			 * searching the file.  Push ourselves onto the state
1098 			 * stack.
1099 			 */
1100 			if (ex_line(sp, ecp, &cur, &isaddr, &tmp))
1101 				goto rfail;
1102 			if (tmp)
1103 				goto err;
1104 
1105 			/* Line specifications are always required. */
1106 			if (!isaddr) {
1107 				msgq_str(sp, M_ERR, ecp->cp,
1108 				     "%s: bad line specification");
1109 				goto err;
1110 			}
1111 			/*
1112 			 * The target line should exist for these commands,
1113 			 * but 0 is legal for them as well.
1114 			 */
1115 			if (cur.lno != 0 && !db_exist(sp, cur.lno)) {
1116 				ex_badaddr(sp, NULL, A_EOF, NUM_OK);
1117 				goto err;
1118 			}
1119 			ecp->lineno = cur.lno;
1120 			break;
1121 		case 'S':				/* string, file exp. */
1122 			if (ecp->clen != 0) {
1123 				if (argv_exp1(sp, ecp, ecp->cp,
1124 				    ecp->clen, ecp->cmd == &cmds[C_BANG]))
1125 					goto err;
1126 				goto addr_verify;
1127 			}
1128 			/* FALLTHROUGH */
1129 		case 's':				/* string */
1130 			if (argv_exp0(sp, ecp, ecp->cp, ecp->clen))
1131 				goto err;
1132 			goto addr_verify;
1133 		case 'W':				/* word string */
1134 			/*
1135 			 * QUOTING NOTE:
1136 			 *
1137 			 * Literal next characters escape the following
1138 			 * character.  Quoting characters are stripped here
1139 			 * since they are no longer useful.
1140 			 *
1141 			 * First there was the word.
1142 			 */
1143 			for (p = t = ecp->cp;
1144 			    ecp->clen > 0; --ecp->clen, ++ecp->cp) {
1145 				ch = *ecp->cp;
1146 				if (IS_ESCAPE(sp,
1147 				    ecp, ch) && ecp->clen > 1) {
1148 					--ecp->clen;
1149 					*p++ = *++ecp->cp;
1150 				} else if (isblank(ch)) {
1151 					++ecp->cp;
1152 					--ecp->clen;
1153 					break;
1154 				} else
1155 					*p++ = ch;
1156 			}
1157 			if (argv_exp0(sp, ecp, t, p - t))
1158 				goto err;
1159 
1160 			/* Delete intervening whitespace. */
1161 			for (; ecp->clen > 0;
1162 			    --ecp->clen, ++ecp->cp) {
1163 				ch = *ecp->cp;
1164 				if (!isblank(ch))
1165 					break;
1166 			}
1167 			if (ecp->clen == 0)
1168 				goto usage;
1169 
1170 			/* Followed by the string. */
1171 			for (p = t = ecp->cp; ecp->clen > 0;
1172 			    --ecp->clen, ++ecp->cp, ++p) {
1173 				ch = *ecp->cp;
1174 				if (IS_ESCAPE(sp,
1175 				    ecp, ch) && ecp->clen > 1) {
1176 					--ecp->clen;
1177 					*p = *++ecp->cp;
1178 				} else
1179 					*p = ch;
1180 			}
1181 			if (argv_exp0(sp, ecp, t, p - t))
1182 				goto err;
1183 			goto addr_verify;
1184 		case 'w':				/* word */
1185 			if (argv_exp3(sp, ecp, ecp->cp, ecp->clen))
1186 				goto err;
1187 arg_cnt_chk:		if (*++p != 'N') {		/* N */
1188 				/*
1189 				 * If a number is specified, must either be
1190 				 * 0 or that number, if optional, and that
1191 				 * number, if required.
1192 				 */
1193 				tmp = *p - '0';
1194 				if ((*++p != 'o' || exp->argsoff != 0) &&
1195 				    exp->argsoff != tmp)
1196 					goto usage;
1197 			}
1198 			goto addr_verify;
1199 		default:
1200 			msgq(sp, M_ERR,
1201 			    "Internal syntax table error (%s: %s)",
1202 			    ecp->cmd->name, KEY_NAME(sp, *p));
1203 		}
1204 	}
1205 
1206 	/* Skip trailing whitespace. */
1207 	for (; ecp->clen > 0; --ecp->clen) {
1208 		ch = *ecp->cp++;
1209 		if (!isblank(ch))
1210 			break;
1211 	}
1212 
1213 	/*
1214 	 * There shouldn't be anything left, and no more required fields,
1215 	 * i.e neither 'l' or 'r' in the syntax string.
1216 	 */
1217 	if (ecp->clen != 0 || strpbrk(p, "lr")) {
1218 usage:		msgq(sp, M_ERR, "Usage: %s", ecp->cmd->usage);
1219 		goto err;
1220 	}
1221 
1222 	/*
1223 	 * Verify that the addresses are legal.  Check the addresses here,
1224 	 * because this is a place where all ex addresses pass through.
1225 	 * (They don't all pass through ex_line(), for instance.)  We're
1226 	 * assuming that any non-existent line doesn't exist because it's
1227 	 * past the end-of-file.  That's a pretty good guess.
1228 	 *
1229 	 * If it's a "default vi command", an address of zero is okay.
1230 	 */
1231 addr_verify:
1232 	switch (ecp->addrcnt) {
1233 	case 2:
1234 		/*
1235 		 * Historic ex/vi permitted commands with counts to go past
1236 		 * EOF.  So, for example, if the file only had 5 lines, the
1237 		 * ex command "1,6>" would fail, but the command ">300"
1238 		 * would succeed.  Since we don't want to have to make all
1239 		 * of the underlying commands handle random line numbers,
1240 		 * fix it here.
1241 		 */
1242 		if (ecp->addr2.lno == 0) {
1243 			if (!F_ISSET(ecp, E_ADDR_ZERO) &&
1244 			    (F_ISSET(sp, SC_EX) ||
1245 			    !F_ISSET(ecp, E_USELASTCMD))) {
1246 				ex_badaddr(sp, ecp->cmd, A_ZERO, NUM_OK);
1247 				goto err;
1248 			}
1249 		} else if (!db_exist(sp, ecp->addr2.lno)) {
1250 			if (FL_ISSET(ecp->iflags, E_C_COUNT)) {
1251 				if (db_last(sp, &lno))
1252 					goto err;
1253 				ecp->addr2.lno = lno;
1254 			} else {
1255 				ex_badaddr(sp, NULL, A_EOF, NUM_OK);
1256 				goto err;
1257 			}
1258 		}
1259 		/* FALLTHROUGH */
1260 	case 1:
1261 		if (ecp->addr1.lno == 0) {
1262 			if (!F_ISSET(ecp, E_ADDR_ZERO) &&
1263 			    (F_ISSET(sp, SC_EX) ||
1264 			    !F_ISSET(ecp, E_USELASTCMD))) {
1265 				ex_badaddr(sp, ecp->cmd, A_ZERO, NUM_OK);
1266 				goto err;
1267 			}
1268 		} else if (!db_exist(sp, ecp->addr1.lno)) {
1269 			ex_badaddr(sp, NULL, A_EOF, NUM_OK);
1270 			goto err;
1271 		}
1272 		break;
1273 	}
1274 
1275 	/*
1276 	 * If doing a default command and there's nothing left on the line,
1277 	 * vi just moves to the line.  For example, ":3" and ":'a,'b" just
1278 	 * move to line 3 and line 'b, respectively, but ":3|" prints line 3.
1279 	 *
1280 	 * !!!
1281 	 * In addition, IF THE LINE CHANGES, move to the first nonblank of
1282 	 * the line.
1283 	 *
1284 	 * !!!
1285 	 * This is done before the absolute mark gets set; historically,
1286 	 * "/a/,/b/" did NOT set vi's absolute mark, but "/a/,/b/d" did.
1287 	 */
1288 	if ((F_ISSET(sp, SC_VI) || F_ISSET(ecp, E_NOPRDEF)) &&
1289 	    F_ISSET(ecp, E_USELASTCMD) && vi_address == 0) {
1290 		switch (ecp->addrcnt) {
1291 		case 2:
1292 			if (sp->lno !=
1293 			    (ecp->addr2.lno ? ecp->addr2.lno : 1)) {
1294 				sp->lno =
1295 				    ecp->addr2.lno ? ecp->addr2.lno : 1;
1296 				sp->cno = 0;
1297 				(void)nonblank(sp, sp->lno, &sp->cno);
1298 			}
1299 			break;
1300 		case 1:
1301 			if (sp->lno !=
1302 			    (ecp->addr1.lno ? ecp->addr1.lno : 1)) {
1303 				sp->lno =
1304 				    ecp->addr1.lno ? ecp->addr1.lno : 1;
1305 				sp->cno = 0;
1306 				(void)nonblank(sp, sp->lno, &sp->cno);
1307 			}
1308 			break;
1309 		}
1310 		ecp->cp = ecp->save_cmd;
1311 		ecp->clen = ecp->save_cmdlen;
1312 		goto loop;
1313 	}
1314 
1315 	/*
1316 	 * Set the absolute mark -- we have to set it for vi here, in case
1317 	 * it's a compound command, e.g. ":5p|6" should set the absolute
1318 	 * mark for vi.
1319 	 */
1320 	if (F_ISSET(ecp, E_ABSMARK)) {
1321 		cur.lno = sp->lno;
1322 		cur.cno = sp->cno;
1323 		F_CLR(ecp, E_ABSMARK);
1324 		if (mark_set(sp, ABSMARK1, &cur, 1))
1325 			goto err;
1326 	}
1327 
1328 #if defined(DEBUG) && defined(COMLOG)
1329 	ex_comlog(sp, ecp);
1330 #endif
1331 	/* Increment the command count if not called from vi. */
1332 	if (F_ISSET(sp, SC_EX))
1333 		++sp->ccnt;
1334 
1335 	/*
1336 	 * If file state available, and not doing a global command,
1337 	 * log the start of an action.
1338 	 */
1339 	if (sp->ep != NULL && !F_ISSET(sp, SC_EX_GLOBAL))
1340 		(void)log_cursor(sp);
1341 
1342 	/*
1343 	 * !!!
1344 	 * There are two special commands for the purposes of this code: the
1345 	 * default command (<carriage-return>) or the scrolling commands (^D
1346 	 * and <EOF>) as the first non-<blank> characters  in the line.
1347 	 *
1348 	 * If this is the first command in the command line, we received the
1349 	 * command from the ex command loop and we're talking to a tty, and
1350 	 * and there's nothing else on the command line, and it's one of the
1351 	 * special commands, we move back up to the previous line, and erase
1352 	 * the prompt character with the output.  Since ex runs in canonical
1353 	 * mode, we don't have to do anything else, a <newline> has already
1354 	 * been echoed by the tty driver.  It's OK if vi calls us -- we won't
1355 	 * be in ex mode so we'll do nothing.
1356 	 */
1357 	if (F_ISSET(ecp, E_NRSEP)) {
1358 		if (sp->ep != NULL &&
1359 		    F_ISSET(sp, SC_EX) && !F_ISSET(gp, G_SCRIPTED) &&
1360 		    (F_ISSET(ecp, E_USELASTCMD) || ecp->cmd == &cmds[C_SCROLL]))
1361 			gp->scr_ex_adjust(sp, EX_TERM_SCROLL);
1362 		F_CLR(ecp, E_NRSEP);
1363 	}
1364 
1365 	/*
1366 	 * Call the underlying function for the ex command.
1367 	 *
1368 	 * XXX
1369 	 * Interrupts behave like errors, for now.
1370 	 */
1371 	if (ecp->cmd->fn(sp, ecp) || INTERRUPTED(sp)) {
1372 		if (F_ISSET(gp, G_SCRIPTED))
1373 			F_SET(sp, SC_EXIT_FORCE);
1374 		goto err;
1375 	}
1376 
1377 #ifdef DEBUG
1378 	/* Make sure no function left global temporary space locked. */
1379 	if (F_ISSET(gp, G_TMP_INUSE)) {
1380 		F_CLR(gp, G_TMP_INUSE);
1381 		msgq(sp, M_ERR, "%s: temporary buffer not released",
1382 		    ecp->cmd->name);
1383 	}
1384 #endif
1385 	/*
1386 	 * Ex displayed the number of lines modified immediately after each
1387 	 * command, so the command "1,10d|1,10d" would display:
1388 	 *
1389 	 *	10 lines deleted
1390 	 *	10 lines deleted
1391 	 *	<autoprint line>
1392 	 *
1393 	 * Executing ex commands from vi only reported the final modified
1394 	 * lines message -- that's wrong enough that we don't match it.
1395 	 */
1396 	if (F_ISSET(sp, SC_EX))
1397 		mod_rpt(sp);
1398 
1399 	/*
1400 	 * Integrate any offset parsed by the underlying command, and make
1401 	 * sure the referenced line exists.
1402 	 *
1403 	 * XXX
1404 	 * May not match historic practice (which I've never been able to
1405 	 * completely figure out.)  For example, the '=' command from vi
1406 	 * mode often got the offset wrong, and complained it was too large,
1407 	 * but didn't seem to have a problem with the cursor.  If anyone
1408 	 * complains, ask them how it's supposed to work, they might know.
1409 	 */
1410 	if (sp->ep != NULL && ecp->flagoff) {
1411 		if (ecp->flagoff < 0) {
1412 			if (sp->lno <= -ecp->flagoff) {
1413 				msgq(sp, M_ERR,
1414 				    "Flag offset to before line 1");
1415 				goto err;
1416 			}
1417 		} else {
1418 			if (!NPFITS(MAX_REC_NUMBER, sp->lno, ecp->flagoff)) {
1419 				ex_badaddr(sp, NULL, A_NOTSET, NUM_OVER);
1420 				goto err;
1421 			}
1422 			if (!db_exist(sp, sp->lno + ecp->flagoff)) {
1423 				msgq(sp, M_ERR,
1424 				    "Flag offset past end-of-file");
1425 				goto err;
1426 			}
1427 		}
1428 		sp->lno += ecp->flagoff;
1429 	}
1430 
1431 	/*
1432 	 * If the command executed successfully, we may want to display a line
1433 	 * based on the autoprint option or an explicit print flag.  (Make sure
1434 	 * that there's a line to display.)  Also, the autoprint edit option is
1435 	 * turned off for the duration of global commands.
1436 	 */
1437 	if (F_ISSET(sp, SC_EX) && sp->ep != NULL && sp->lno != 0) {
1438 		/*
1439 		 * The print commands have already handled the `print' flags.
1440 		 * If so, clear them.
1441 		 */
1442 		if (FL_ISSET(ecp->iflags, E_CLRFLAG))
1443 			FL_CLR(ecp->iflags, E_C_HASH | E_C_LIST | E_C_PRINT);
1444 
1445 		/* If hash set only because of the number option, discard it. */
1446 		if (F_ISSET(ecp, E_OPTNUM))
1447 			FL_CLR(ecp->iflags, E_C_HASH);
1448 
1449 		/*
1450 		 * If there was an explicit flag to display the new cursor line,
1451 		 * or autoprint is set and a change was made, display the line.
1452 		 * If any print flags were set use them, else default to print.
1453 		 */
1454 		LF_INIT(FL_ISSET(ecp->iflags, E_C_HASH | E_C_LIST | E_C_PRINT));
1455 		if (!LF_ISSET(E_C_HASH | E_C_LIST | E_C_PRINT | E_NOAUTO) &&
1456 		    !F_ISSET(sp, SC_EX_GLOBAL) &&
1457 		    O_ISSET(sp, O_AUTOPRINT) && F_ISSET(ecp, E_AUTOPRINT)) {
1458 
1459 			/* Honor the number option if autoprint is set. */
1460 			if (F_ISSET(ecp, E_OPTNUM))
1461 				LF_INIT(E_C_HASH);
1462 			else
1463 				LF_INIT(E_C_PRINT);
1464 		}
1465 
1466 		if (LF_ISSET(E_C_HASH | E_C_LIST | E_C_PRINT)) {
1467 			cur.lno = sp->lno;
1468 			cur.cno = 0;
1469 			(void)ex_print(sp, ecp, &cur, &cur, flags);
1470 		}
1471 	}
1472 
1473 	/*
1474 	 * If the command had an associated "+cmd", it has to be executed
1475 	 * before we finish executing any more of this ex command.  For
1476 	 * example, consider a .exrc file that contains the following lines:
1477 	 *
1478 	 *	:set all
1479 	 *	:edit +25 file.c|s/abc/ABC/|1
1480 	 *	:3,5 print
1481 	 *
1482 	 * This can happen more than once -- the historic vi simply hung or
1483 	 * dropped core, of course.  Prepend the + command back into the
1484 	 * current command and continue.  We may have to add an additional
1485 	 * <literal next> character.  We know that it will fit because we
1486 	 * discarded at least one space and the + character.
1487 	 */
1488 	if (arg1_len != 0) {
1489 		/*
1490 		 * If the last character of the + command was a <literal next>
1491 		 * character, it would be treated differently because of the
1492 		 * append.  Quote it, if necessary.
1493 		 */
1494 		if (IS_ESCAPE(sp, ecp, arg1[arg1_len - 1])) {
1495 			*--ecp->save_cmd = CH_LITERAL;
1496 			++ecp->save_cmdlen;
1497 		}
1498 
1499 		ecp->save_cmd -= arg1_len;
1500 		ecp->save_cmdlen += arg1_len;
1501 		memmove(ecp->save_cmd, arg1, arg1_len);
1502 
1503 		/*
1504 		 * Any commands executed from a +cmd are executed starting at
1505 		 * the first column of the last line of the file -- NOT the
1506 		 * first nonblank.)  The main file startup code doesn't know
1507 		 * that a +cmd was set, however, so it may have put us at the
1508 		 * top of the file.  (Note, this is safe because we must have
1509 		 * switched files to get here.)
1510 		 */
1511 		F_SET(ecp, E_MOVETOEND);
1512 	}
1513 
1514 	/* Update the current command. */
1515 	ecp->cp = ecp->save_cmd;
1516 	ecp->clen = ecp->save_cmdlen;
1517 
1518 	/*
1519 	 * !!!
1520 	 * If we've changed screens or underlying files, any pending global or
1521 	 * v command, or @ buffer that has associated addresses, has to be
1522 	 * discarded.  This is historic practice for globals, and necessary for
1523 	 * @ buffers that had associated addresses.
1524 	 *
1525 	 * Otherwise, if we've changed underlying files, it's not a problem,
1526 	 * we continue with the rest of the ex command(s), operating on the
1527 	 * new file.  However, if we switch screens (either by exiting or by
1528 	 * an explicit command), we have no way of knowing where to put output
1529 	 * messages, and, since we don't control screens here, we could screw
1530 	 * up the upper layers, (e.g. we could exit/reenter a screen multiple
1531 	 * times).  So, return and continue after we've got a new screen.
1532 	 */
1533 	if (F_ISSET(sp, SC_EXIT | SC_EXIT_FORCE | SC_FSWITCH | SC_SSWITCH)) {
1534 		at_found = gv_found = 0;
1535 		LIST_FOREACH(ecp, &sp->gp->ecq, q)
1536 			switch (ecp->agv_flags) {
1537 			case 0:
1538 			case AGV_AT_NORANGE:
1539 				break;
1540 			case AGV_AT:
1541 				if (!at_found) {
1542 					at_found = 1;
1543 					msgq(sp, M_ERR,
1544 		"@ with range running when the file/screen changed");
1545 				}
1546 				break;
1547 			case AGV_GLOBAL:
1548 			case AGV_V:
1549 				if (!gv_found) {
1550 					gv_found = 1;
1551 					msgq(sp, M_ERR,
1552 		"Global/v command running when the file/screen changed");
1553 				}
1554 				break;
1555 			default:
1556 				abort();
1557 			}
1558 		if (at_found || gv_found)
1559 			goto discard;
1560 		if (F_ISSET(sp, SC_EXIT | SC_EXIT_FORCE | SC_SSWITCH))
1561 			goto rsuccess;
1562 	}
1563 
1564 	goto loop;
1565 	/* NOTREACHED */
1566 
1567 err:	/*
1568 	 * On command failure, we discard keys and pending commands remaining,
1569 	 * as well as any keys that were mapped and waiting.  The save_cmdlen
1570 	 * test is not necessarily correct.  If we fail early enough we don't
1571 	 * know if the entire string was a single command or not.  Guess, as
1572 	 * it's useful to know if commands other than the current one are being
1573 	 * discarded.
1574 	 */
1575 	if (ecp->save_cmdlen == 0)
1576 		for (; ecp->clen; --ecp->clen) {
1577 			ch = *ecp->cp++;
1578 			if (IS_ESCAPE(sp, ecp, ch) && ecp->clen > 1) {
1579 				--ecp->clen;
1580 				++ecp->cp;
1581 			} else if (ch == '\n' || ch == '|') {
1582 				if (ecp->clen > 1)
1583 					ecp->save_cmdlen = 1;
1584 				break;
1585 			}
1586 		}
1587 	if (ecp->save_cmdlen != 0 || LIST_FIRST(&gp->ecq) != &gp->excmd) {
1588 discard:	msgq(sp, M_BERR,
1589 		    "Ex command failed: pending commands discarded");
1590 		ex_discard(sp);
1591 	}
1592 	if (v_event_flush(sp, CH_MAPPED))
1593 		msgq(sp, M_BERR,
1594 		    "Ex command failed: mapped keys discarded");
1595 
1596 rfail:	tmp = 1;
1597 	if (0)
1598 rsuccess:	tmp = 0;
1599 
1600 	/* Turn off any file name error information. */
1601 	gp->if_name = NULL;
1602 
1603 	/* Turn off the global bit. */
1604 	F_CLR(sp, SC_EX_GLOBAL);
1605 
1606 	return (tmp);
1607 }
1608 
1609 /*
1610  * ex_range --
1611  *	Get a line range for ex commands, or perform a vi ex address search.
1612  *
1613  * PUBLIC: int ex_range(SCR *, EXCMD *, int *);
1614  */
1615 int
ex_range(SCR * sp,EXCMD * ecp,int * errp)1616 ex_range(SCR *sp, EXCMD *ecp, int *errp)
1617 {
1618 	enum { ADDR_FOUND, ADDR_NEED, ADDR_NONE } addr;
1619 	MARK m;
1620 	int isaddr;
1621 
1622 	*errp = 0;
1623 
1624 	/*
1625 	 * Parse comma or semi-colon delimited line specs.
1626 	 *
1627 	 * Semi-colon delimiters update the current address to be the last
1628 	 * address.  For example, the command
1629 	 *
1630 	 *	:3;/pattern/ecp->cp
1631 	 *
1632 	 * will search for pattern from line 3.  In addition, if ecp->cp
1633 	 * is not a valid command, the current line will be left at 3, not
1634 	 * at the original address.
1635 	 *
1636 	 * Extra addresses are discarded, starting with the first.
1637 	 *
1638 	 * !!!
1639 	 * If any addresses are missing, they default to the current line.
1640 	 * This was historically true for both leading and trailing comma
1641 	 * delimited addresses as well as for trailing semicolon delimited
1642 	 * addresses.  For consistency, we make it true for leading semicolon
1643 	 * addresses as well.
1644 	 */
1645 	for (addr = ADDR_NONE, ecp->addrcnt = 0; ecp->clen > 0;)
1646 		switch (*ecp->cp) {
1647 		case '%':		/* Entire file. */
1648 			/* Vi ex address searches didn't permit % signs. */
1649 			if (F_ISSET(ecp, E_VISEARCH))
1650 				goto ret;
1651 
1652 			/* It's an error if the file is empty. */
1653 			if (sp->ep == NULL) {
1654 				ex_badaddr(sp, NULL, A_EMPTY, NUM_OK);
1655 				*errp = 1;
1656 				return (0);
1657 			}
1658 			/*
1659 			 * !!!
1660 			 * A percent character addresses all of the lines in
1661 			 * the file.  Historically, it couldn't be followed by
1662 			 * any other address.  We do it as a text substitution
1663 			 * for simplicity.  POSIX 1003.2 is expected to follow
1664 			 * this practice.
1665 			 *
1666 			 * If it's an empty file, the first line is 0, not 1.
1667 			 */
1668 			if (addr == ADDR_FOUND) {
1669 				ex_badaddr(sp, NULL, A_COMBO, NUM_OK);
1670 				*errp = 1;
1671 				return (0);
1672 			}
1673 			if (db_last(sp, &ecp->addr2.lno))
1674 				return (1);
1675 			ecp->addr1.lno = ecp->addr2.lno == 0 ? 0 : 1;
1676 			ecp->addr1.cno = ecp->addr2.cno = 0;
1677 			ecp->addrcnt = 2;
1678 			addr = ADDR_FOUND;
1679 			++ecp->cp;
1680 			--ecp->clen;
1681 			break;
1682 		case ',':               /* Comma delimiter. */
1683 			/* Vi ex address searches didn't permit commas. */
1684 			if (F_ISSET(ecp, E_VISEARCH))
1685 				goto ret;
1686 			/* FALLTHROUGH */
1687 		case ';':               /* Semi-colon delimiter. */
1688 			if (sp->ep == NULL) {
1689 				ex_badaddr(sp, NULL, A_EMPTY, NUM_OK);
1690 				*errp = 1;
1691 				return (0);
1692 			}
1693 			if (addr != ADDR_FOUND)
1694 				switch (ecp->addrcnt) {
1695 				case 0:
1696 					ecp->addr1.lno = sp->lno;
1697 					ecp->addr1.cno = sp->cno;
1698 					ecp->addrcnt = 1;
1699 					break;
1700 				case 2:
1701 					ecp->addr1 = ecp->addr2;
1702 					/* FALLTHROUGH */
1703 				case 1:
1704 					ecp->addr2.lno = sp->lno;
1705 					ecp->addr2.cno = sp->cno;
1706 					ecp->addrcnt = 2;
1707 					break;
1708 				}
1709 			if (*ecp->cp == ';')
1710 				switch (ecp->addrcnt) {
1711 				case 0:
1712 					abort();
1713 					/* NOTREACHED */
1714 				case 1:
1715 					sp->lno = ecp->addr1.lno;
1716 					sp->cno = ecp->addr1.cno;
1717 					break;
1718 				case 2:
1719 					sp->lno = ecp->addr2.lno;
1720 					sp->cno = ecp->addr2.cno;
1721 					break;
1722 				}
1723 			addr = ADDR_NEED;
1724 			/* FALLTHROUGH */
1725 		case ' ':		/* Whitespace. */
1726 		case '\t':		/* Whitespace. */
1727 			++ecp->cp;
1728 			--ecp->clen;
1729 			break;
1730 		default:
1731 			/* Get a line specification. */
1732 			if (ex_line(sp, ecp, &m, &isaddr, errp))
1733 				return (1);
1734 			if (*errp)
1735 				return (0);
1736 			if (!isaddr)
1737 				goto ret;
1738 			if (addr == ADDR_FOUND) {
1739 				ex_badaddr(sp, NULL, A_COMBO, NUM_OK);
1740 				*errp = 1;
1741 				return (0);
1742 			}
1743 			switch (ecp->addrcnt) {
1744 			case 0:
1745 				ecp->addr1 = m;
1746 				ecp->addrcnt = 1;
1747 				break;
1748 			case 1:
1749 				ecp->addr2 = m;
1750 				ecp->addrcnt = 2;
1751 				break;
1752 			case 2:
1753 				ecp->addr1 = ecp->addr2;
1754 				ecp->addr2 = m;
1755 				break;
1756 			}
1757 			addr = ADDR_FOUND;
1758 			break;
1759 		}
1760 
1761 	/*
1762 	 * !!!
1763 	 * Vi ex address searches are indifferent to order or trailing
1764 	 * semi-colons.
1765 	 */
1766 ret:	if (F_ISSET(ecp, E_VISEARCH))
1767 		return (0);
1768 
1769 	if (addr == ADDR_NEED)
1770 		switch (ecp->addrcnt) {
1771 		case 0:
1772 			ecp->addr1.lno = sp->lno;
1773 			ecp->addr1.cno = sp->cno;
1774 			ecp->addrcnt = 1;
1775 			break;
1776 		case 2:
1777 			ecp->addr1 = ecp->addr2;
1778 			/* FALLTHROUGH */
1779 		case 1:
1780 			ecp->addr2.lno = sp->lno;
1781 			ecp->addr2.cno = sp->cno;
1782 			ecp->addrcnt = 2;
1783 			break;
1784 		}
1785 
1786 	if (ecp->addrcnt == 2 && ecp->addr2.lno < ecp->addr1.lno) {
1787 		msgq(sp, M_ERR,
1788 		    "The second address is smaller than the first");
1789 		*errp = 1;
1790 	}
1791 	return (0);
1792 }
1793 
1794 /*
1795  * ex_line --
1796  *	Get a single line address specifier.
1797  *
1798  * The way the "previous context" mark worked was that any "non-relative"
1799  * motion set it.  While ex/vi wasn't totally consistent about this, ANY
1800  * numeric address, search pattern, '$', or mark reference in an address
1801  * was considered non-relative, and set the value.  Which should explain
1802  * why we're hacking marks down here.  The problem was that the mark was
1803  * only set if the command was called, i.e. we have to set a flag and test
1804  * it later.
1805  *
1806  * XXX
1807  * This is probably still not exactly historic practice, although I think
1808  * it's fairly close.
1809  */
1810 static int
ex_line(SCR * sp,EXCMD * ecp,MARK * mp,int * isaddrp,int * errp)1811 ex_line(SCR *sp, EXCMD *ecp, MARK *mp, int *isaddrp, int *errp)
1812 {
1813 	enum nresult nret;
1814 	long total, val;
1815 	int isneg;
1816 	int (*sf)(SCR *, MARK *, MARK *, char *, size_t, char **, u_int);
1817 	char *endp;
1818 
1819 	*isaddrp = *errp = 0;
1820 	F_CLR(ecp, E_DELTA);
1821 
1822 	/* No addresses permitted until a file has been read in. */
1823 	if (sp->ep == NULL && strchr("$0123456789'\\/?.+-^", *ecp->cp)) {
1824 		ex_badaddr(sp, NULL, A_EMPTY, NUM_OK);
1825 		*errp = 1;
1826 		return (0);
1827 	}
1828 
1829 	switch (*ecp->cp) {
1830 	case '$':				/* Last line in the file. */
1831 		*isaddrp = 1;
1832 		F_SET(ecp, E_ABSMARK);
1833 
1834 		mp->cno = 0;
1835 		if (db_last(sp, &mp->lno))
1836 			return (1);
1837 		++ecp->cp;
1838 		--ecp->clen;
1839 		break;				/* Absolute line number. */
1840 	case '0': case '1': case '2': case '3': case '4':
1841 	case '5': case '6': case '7': case '8': case '9':
1842 		*isaddrp = 1;
1843 		F_SET(ecp, E_ABSMARK);
1844 
1845 		if ((nret = nget_slong(&val, ecp->cp, &endp, 10)) != NUM_OK) {
1846 			ex_badaddr(sp, NULL, A_NOTSET, nret);
1847 			*errp = 1;
1848 			return (0);
1849 		}
1850 		if (!NPFITS(MAX_REC_NUMBER, 0, val)) {
1851 			ex_badaddr(sp, NULL, A_NOTSET, NUM_OVER);
1852 			*errp = 1;
1853 			return (0);
1854 		}
1855 		mp->lno = val;
1856 		mp->cno = 0;
1857 		ecp->clen -= (endp - ecp->cp);
1858 		ecp->cp = endp;
1859 		break;
1860 	case '\'':				/* Use a mark. */
1861 		*isaddrp = 1;
1862 		F_SET(ecp, E_ABSMARK);
1863 
1864 		if (ecp->clen == 1) {
1865 			msgq(sp, M_ERR, "No mark name supplied");
1866 			*errp = 1;
1867 			return (0);
1868 		}
1869 		if (mark_get(sp, ecp->cp[1], mp, M_ERR)) {
1870 			*errp = 1;
1871 			return (0);
1872 		}
1873 		ecp->cp += 2;
1874 		ecp->clen -= 2;
1875 		break;
1876 	case '\\':				/* Search: forward/backward. */
1877 		/*
1878 		 * !!!
1879 		 * I can't find any difference between // and \/ or between
1880 		 * ?? and \?.  Mark Horton doesn't remember there being any
1881 		 * difference.  C'est la vie.
1882 		 */
1883 		if (ecp->clen < 2 ||
1884 		    (ecp->cp[1] != '/' && ecp->cp[1] != '?')) {
1885 			msgq(sp, M_ERR, "\\ not followed by / or ?");
1886 			*errp = 1;
1887 			return (0);
1888 		}
1889 		++ecp->cp;
1890 		--ecp->clen;
1891 		sf = ecp->cp[0] == '/' ? f_search : b_search;
1892 		goto search;
1893 	case '/':				/* Search forward. */
1894 		sf = f_search;
1895 		goto search;
1896 	case '?':				/* Search backward. */
1897 		sf = b_search;
1898 
1899 search:		mp->lno = sp->lno;
1900 		mp->cno = sp->cno;
1901 		if (sf(sp, mp, mp, ecp->cp, ecp->clen, &endp,
1902 		    SEARCH_MSG | SEARCH_PARSE | SEARCH_SET |
1903 		    (F_ISSET(ecp, E_SEARCH_WMSG) ? SEARCH_WMSG : 0))) {
1904 			*errp = 1;
1905 			return (0);
1906 		}
1907 
1908 		/* Fix up the command pointers. */
1909 		ecp->clen -= (endp - ecp->cp);
1910 		ecp->cp = endp;
1911 
1912 		*isaddrp = 1;
1913 		F_SET(ecp, E_ABSMARK);
1914 		break;
1915 	case '.':				/* Current position. */
1916 		*isaddrp = 1;
1917 		mp->cno = sp->cno;
1918 
1919 		/* If an empty file, then '.' is 0, not 1. */
1920 		if (sp->lno == 1) {
1921 			if (db_last(sp, &mp->lno))
1922 				return (1);
1923 			if (mp->lno != 0)
1924 				mp->lno = 1;
1925 		} else
1926 			mp->lno = sp->lno;
1927 
1928 		/*
1929 		 * !!!
1930 		 * Historically, .<number> was the same as .+<number>, i.e.
1931 		 * the '+' could be omitted.  (This feature is found in ed
1932 		 * as well.)
1933 		 */
1934 		if (ecp->clen > 1 && isdigit(ecp->cp[1]))
1935 			*ecp->cp = '+';
1936 		else {
1937 			++ecp->cp;
1938 			--ecp->clen;
1939 		}
1940 		break;
1941 	}
1942 
1943 	/* Skip trailing <blank>s. */
1944 	for (; ecp->clen > 0 &&
1945 	    isblank(ecp->cp[0]); ++ecp->cp, --ecp->clen);
1946 
1947 	/*
1948 	 * Evaluate any offset.  If no address yet found, the offset
1949 	 * is relative to ".".
1950 	 */
1951 	total = 0;
1952 	if (ecp->clen != 0 && (isdigit(ecp->cp[0]) ||
1953 	    ecp->cp[0] == '+' || ecp->cp[0] == '-' ||
1954 	    ecp->cp[0] == '^')) {
1955 		if (!*isaddrp) {
1956 			*isaddrp = 1;
1957 			mp->lno = sp->lno;
1958 			mp->cno = sp->cno;
1959 		}
1960 		/*
1961 		 * Evaluate an offset, defined as:
1962 		 *
1963 		 *		[+-^<blank>]*[<blank>]*[0-9]*
1964 		 *
1965 		 * The rough translation is any number of signs, optionally
1966 		 * followed by numbers, or a number by itself, all <blank>
1967 		 * separated.
1968 		 *
1969 		 * !!!
1970 		 * All address offsets were additive, e.g. "2 2 3p" was the
1971 		 * same as "7p", or, "/ZZZ/ 2" was the same as "/ZZZ/+2".
1972 		 * Note, however, "2 /ZZZ/" was an error.  It was also legal
1973 		 * to insert signs without numbers, so "3 - 2" was legal, and
1974 		 * equal to 4.
1975 		 *
1976 		 * !!!
1977 		 * Offsets were historically permitted for any line address,
1978 		 * e.g. the command "1,2 copy 2 2 2 2" copied lines 1,2 after
1979 		 * line 8.
1980 		 *
1981 		 * !!!
1982 		 * Offsets were historically permitted for search commands,
1983 		 * and handled as addresses: "/pattern/2 2 2" was legal, and
1984 		 * referenced the 6th line after pattern.
1985 		 */
1986 		F_SET(ecp, E_DELTA);
1987 		for (;;) {
1988 			for (; ecp->clen > 0 && isblank(ecp->cp[0]);
1989 			    ++ecp->cp, --ecp->clen);
1990 			if (ecp->clen == 0 || (!isdigit(ecp->cp[0]) &&
1991 			    ecp->cp[0] != '+' && ecp->cp[0] != '-' &&
1992 			    ecp->cp[0] != '^'))
1993 				break;
1994 			if (!isdigit(ecp->cp[0]) &&
1995 			    !isdigit(ecp->cp[1])) {
1996 				total += ecp->cp[0] == '+' ? 1 : -1;
1997 				--ecp->clen;
1998 				++ecp->cp;
1999 			} else {
2000 				if (ecp->cp[0] == '-' ||
2001 				    ecp->cp[0] == '^') {
2002 					++ecp->cp;
2003 					--ecp->clen;
2004 					isneg = 1;
2005 				} else
2006 					isneg = 0;
2007 
2008 				/* Get a signed long, add it to the total. */
2009 				if ((nret = nget_slong(&val,
2010 				    ecp->cp, &endp, 10)) != NUM_OK ||
2011 				    (nret = NADD_SLONG(total, val)) != NUM_OK) {
2012 					ex_badaddr(sp, NULL, A_NOTSET, nret);
2013 					*errp = 1;
2014 					return (0);
2015 				}
2016 				total += isneg ? -val : val;
2017 				ecp->clen -= (endp - ecp->cp);
2018 				ecp->cp = endp;
2019 			}
2020 		}
2021 	}
2022 
2023 	/*
2024 	 * Any value less than 0 is an error.  Make sure that the new value
2025 	 * will fit into a recno_t.
2026 	 */
2027 	if (*isaddrp && total != 0) {
2028 		if (total < 0) {
2029 			if (-total > mp->lno) {
2030 				msgq(sp, M_ERR,
2031 			    "Reference to a line number less than 0");
2032 				*errp = 1;
2033 				return (0);
2034 			}
2035 		} else
2036 			if (!NPFITS(MAX_REC_NUMBER, mp->lno, total)) {
2037 				ex_badaddr(sp, NULL, A_NOTSET, NUM_OVER);
2038 				*errp = 1;
2039 				return (0);
2040 			}
2041 		mp->lno += total;
2042 	}
2043 	return (0);
2044 }
2045 
2046 
2047 /*
2048  * ex_load --
2049  *	Load up the next command, which may be an @ buffer or global command.
2050  */
2051 static int
ex_load(SCR * sp)2052 ex_load(SCR *sp)
2053 {
2054 	GS *gp;
2055 	EXCMD *ecp;
2056 	RANGE *rp;
2057 
2058 	F_CLR(sp, SC_EX_GLOBAL);
2059 
2060 	/*
2061 	 * Lose any exhausted commands.  We know that the first command
2062 	 * can't be an AGV command, which makes things a bit easier.
2063 	 */
2064 	for (gp = sp->gp;;) {
2065 		/*
2066 		 * If we're back to the original structure, leave it around,
2067 		 * but discard any allocated source name, we've returned to
2068 		 * the beginning of the command stack.
2069 		 */
2070 		if ((ecp = LIST_FIRST(&gp->ecq)) == &gp->excmd) {
2071 			if (F_ISSET(ecp, E_NAMEDISCARD)) {
2072 				free(ecp->if_name);
2073 				ecp->if_name = NULL;
2074 			}
2075 			return (0);
2076 		}
2077 
2078 		/*
2079 		 * ecp->clen will be 0 for the first discarded command, but
2080 		 * may not be 0 for subsequent ones, e.g. if the original
2081 		 * command was ":g/xx/@a|s/b/c/", then when we discard the
2082 		 * command pushed on the stack by the @a, we have to resume
2083 		 * the global command which included the substitute command.
2084 		 */
2085 		if (ecp->clen != 0)
2086 			return (0);
2087 
2088 		/*
2089 		 * If it's an @, global or v command, we may need to continue
2090 		 * the command on a different line.
2091 		 */
2092 		if (FL_ISSET(ecp->agv_flags, AGV_ALL)) {
2093 			/* Discard any exhausted ranges. */
2094 			while ((rp = TAILQ_FIRST(&ecp->rq))) {
2095 				if (rp->start > rp->stop) {
2096 					TAILQ_REMOVE(&ecp->rq, rp, q);
2097 					free(rp);
2098 				} else
2099 					break;
2100 			}
2101 
2102 			/* If there's another range, continue with it. */
2103 			if (rp)
2104 				break;
2105 
2106 			/* If it's a global/v command, fix up the last line. */
2107 			if (FL_ISSET(ecp->agv_flags,
2108 			    AGV_GLOBAL | AGV_V) && ecp->range_lno != OOBLNO) {
2109 				if (db_exist(sp, ecp->range_lno))
2110 					sp->lno = ecp->range_lno;
2111 				else {
2112 					if (db_last(sp, &sp->lno))
2113 						return (1);
2114 					if (sp->lno == 0)
2115 						sp->lno = 1;
2116 				}
2117 			}
2118 			free(ecp->o_cp);
2119 		}
2120 
2121 		/* Discard the EXCMD. */
2122 		LIST_REMOVE(ecp, q);
2123 		free(ecp);
2124 	}
2125 
2126 	/*
2127 	 * We only get here if it's an active @, global or v command.  Set
2128 	 * the current line number, and get a new copy of the command for
2129 	 * the parser.  Note, the original pointer almost certainly moved,
2130 	 * so we have play games.
2131 	 */
2132 	ecp->cp = ecp->o_cp;
2133 	memcpy(ecp->cp, ecp->cp + ecp->o_clen, ecp->o_clen);
2134 	ecp->clen = ecp->o_clen;
2135 	ecp->range_lno = sp->lno = rp->start++;
2136 
2137 	if (FL_ISSET(ecp->agv_flags, AGV_GLOBAL | AGV_V))
2138 		F_SET(sp, SC_EX_GLOBAL);
2139 	return (0);
2140 }
2141 
2142 /*
2143  * ex_discard --
2144  *	Discard any pending ex commands.
2145  */
2146 static int
ex_discard(SCR * sp)2147 ex_discard(SCR *sp)
2148 {
2149 	GS *gp;
2150 	EXCMD *ecp;
2151 	RANGE *rp;
2152 
2153 	/*
2154 	 * We know the first command can't be an AGV command, so we don't
2155 	 * process it specially.  We do, however, nail the command itself.
2156 	 */
2157 	for (gp = sp->gp; (ecp = LIST_FIRST(&gp->ecq)) != &gp->excmd;) {
2158 		if (FL_ISSET(ecp->agv_flags, AGV_ALL)) {
2159 			while ((rp = TAILQ_FIRST(&ecp->rq))) {
2160 				TAILQ_REMOVE(&ecp->rq, rp, q);
2161 				free(rp);
2162 			}
2163 			free(ecp->o_cp);
2164 		}
2165 		LIST_REMOVE(ecp, q);
2166 		free(ecp);
2167 	}
2168 	LIST_FIRST(&gp->ecq)->clen = 0;
2169 	return (0);
2170 }
2171 
2172 /*
2173  * ex_unknown --
2174  *	Display an unknown command name.
2175  */
2176 static void
ex_unknown(SCR * sp,char * cmd,size_t len)2177 ex_unknown(SCR *sp, char *cmd, size_t len)
2178 {
2179 	size_t blen;
2180 	char *bp;
2181 
2182 	GET_SPACE_GOTO(sp, bp, blen, len + 1);
2183 	bp[len] = '\0';
2184 	memcpy(bp, cmd, len);
2185 	msgq_str(sp, M_ERR, bp, "The %s command is unknown");
2186 	FREE_SPACE(sp, bp, blen);
2187 
2188 alloc_err:
2189 	return;
2190 }
2191 
2192 /*
2193  * ex_is_abbrev -
2194  *	The vi text input routine needs to know if ex thinks this is an
2195  *	[un]abbreviate command, so it can turn off abbreviations.  See
2196  *	the usual ranting in the vi/v_txt_ev.c:txt_abbrev() routine.
2197  *
2198  * PUBLIC: int ex_is_abbrev(char *, size_t);
2199  */
2200 int
ex_is_abbrev(char * name,size_t len)2201 ex_is_abbrev(char *name, size_t len)
2202 {
2203 	EXCMDLIST const *cp;
2204 
2205 	return ((cp = ex_comm_search(name, len)) != NULL &&
2206 	    (cp == &cmds[C_ABBR] || cp == &cmds[C_UNABBREVIATE]));
2207 }
2208 
2209 /*
2210  * ex_is_unmap -
2211  *	The vi text input routine needs to know if ex thinks this is an
2212  *	unmap command, so it can turn off input mapping.  See the usual
2213  *	ranting in the vi/v_txt_ev.c:txt_unmap() routine.
2214  *
2215  * PUBLIC: int ex_is_unmap(char *, size_t);
2216  */
2217 int
ex_is_unmap(char * name,size_t len)2218 ex_is_unmap(char *name, size_t len)
2219 {
2220 	EXCMDLIST const *cp;
2221 
2222 	/*
2223 	 * The command the vi input routines are really interested in
2224 	 * is "unmap!", not just unmap.
2225 	 */
2226 	if (name[len - 1] != '!')
2227 		return (0);
2228 	--len;
2229 	return ((cp = ex_comm_search(name, len)) != NULL &&
2230 	    cp == &cmds[C_UNMAP]);
2231 }
2232 
2233 /*
2234  * ex_comm_search --
2235  *	Search for a command name.
2236  */
2237 static EXCMDLIST const *
ex_comm_search(char * name,size_t len)2238 ex_comm_search(char *name, size_t len)
2239 {
2240 	EXCMDLIST const *cp;
2241 
2242 	for (cp = cmds; cp->name != NULL; ++cp) {
2243 		if (cp->name[0] > name[0])
2244 			return (NULL);
2245 		if (cp->name[0] != name[0])
2246 			continue;
2247 		if (!memcmp(name, cp->name, len))
2248 			return (cp);
2249 	}
2250 	return (NULL);
2251 }
2252 
2253 /*
2254  * ex_badaddr --
2255  *	Display a bad address message.
2256  *
2257  * PUBLIC: void ex_badaddr
2258  * PUBLIC:(SCR *, EXCMDLIST const *, enum badaddr, enum nresult);
2259  */
2260 void
ex_badaddr(SCR * sp,EXCMDLIST const * cp,enum badaddr ba,enum nresult nret)2261 ex_badaddr(SCR *sp, EXCMDLIST const *cp, enum badaddr ba, enum nresult nret)
2262 {
2263 	recno_t lno;
2264 
2265 	switch (nret) {
2266 	case NUM_OK:
2267 		break;
2268 	case NUM_ERR:
2269 		msgq(sp, M_SYSERR, NULL);
2270 		return;
2271 	case NUM_OVER:
2272 		msgq(sp, M_ERR, "Address value overflow");
2273 		return;
2274 	case NUM_UNDER:
2275 		msgq(sp, M_ERR, "Address value underflow");
2276 		return;
2277 	}
2278 
2279 	/*
2280 	 * When encountering an address error, tell the user if there's no
2281 	 * underlying file, that's the real problem.
2282 	 */
2283 	if (sp->ep == NULL) {
2284 		ex_emsg(sp, cp != NULL ? cp->name : NULL, EXM_NOFILEYET);
2285 		return;
2286 	}
2287 
2288 	switch (ba) {
2289 	case A_COMBO:
2290 		msgq(sp, M_ERR, "Illegal address combination");
2291 		break;
2292 	case A_EOF:
2293 		if (db_last(sp, &lno))
2294 			return;
2295 		if (lno != 0) {
2296 			msgq(sp, M_ERR,
2297 			    "Illegal address: only %lu lines in the file",
2298 			    lno);
2299 			break;
2300 		}
2301 		/* FALLTHROUGH */
2302 	case A_EMPTY:
2303 		msgq(sp, M_ERR, "Illegal address: the file is empty");
2304 		break;
2305 	case A_NOTSET:
2306 		abort();
2307 		/* NOTREACHED */
2308 	case A_ZERO:
2309 		msgq(sp, M_ERR,
2310 		    "The %s command doesn't permit an address of 0",
2311 		    cp->name);
2312 		break;
2313 	}
2314 	return;
2315 }
2316 
2317 #if defined(DEBUG) && defined(COMLOG)
2318 /*
2319  * ex_comlog --
2320  *	Log ex commands.
2321  */
2322 static void
ex_comlog(SCR * sp,EXCMD * ecp)2323 ex_comlog(SCR *sp, EXCMD *ecp)
2324 {
2325 	TRACE(sp, "ecmd: %s", ecp->cmd->name);
2326 	if (ecp->addrcnt > 0) {
2327 		TRACE(sp, " a1 %d", ecp->addr1.lno);
2328 		if (ecp->addrcnt > 1)
2329 			TRACE(sp, " a2: %d", ecp->addr2.lno);
2330 	}
2331 	if (ecp->lineno)
2332 		TRACE(sp, " line %d", ecp->lineno);
2333 	if (ecp->flags)
2334 		TRACE(sp, " flags 0x%x", ecp->flags);
2335 	if (F_ISSET(&exc, E_BUFFER))
2336 		TRACE(sp, " buffer %c", ecp->buffer);
2337 	if (ecp->argc)
2338 		for (cnt = 0; cnt < ecp->argc; ++cnt)
2339 			TRACE(sp, " arg %d: {%s}", cnt, ecp->argv[cnt]->bp);
2340 	TRACE(sp, "\n");
2341 }
2342 #endif
2343