xref: /freebsd/contrib/nvi/ex/ex.c (revision d0b2dbfa0ecf2bbc9709efc5e20baf8e4b44bbbf)
1 /*-
2  * Copyright (c) 1992, 1993, 1994
3  *	The Regents of the University of California.  All rights reserved.
4  * Copyright (c) 1992, 1993, 1994, 1995, 1996
5  *	Keith Bostic.  All rights reserved.
6  *
7  * See the LICENSE file for redistribution information.
8  */
9 
10 #include "config.h"
11 
12 #include <sys/types.h>
13 #include <sys/queue.h>
14 #include <sys/stat.h>
15 
16 #include <bitstring.h>
17 #include <ctype.h>
18 #include <errno.h>
19 #include <fcntl.h>
20 #include <limits.h>
21 #include <stdio.h>
22 #include <stdlib.h>
23 #include <string.h>
24 #include <unistd.h>
25 
26 #include "../common/common.h"
27 #include "../vi/vi.h"
28 
29 #if defined(DEBUG) && defined(COMLOG)
30 static void	ex_comlog(SCR *, EXCMD *);
31 #endif
32 static EXCMDLIST const *
33 		ex_comm_search(CHAR_T *, size_t);
34 static int	ex_discard(SCR *);
35 static int	ex_line(SCR *, EXCMD *, MARK *, int *, int *);
36 static int	ex_load(SCR *);
37 static void	ex_unknown(SCR *, CHAR_T *, size_t);
38 
39 /*
40  * ex --
41  *	Main ex loop.
42  *
43  * PUBLIC: int ex(SCR **);
44  */
45 int
46 ex(SCR **spp)
47 {
48 	EX_PRIVATE *exp;
49 	GS *gp;
50 	MSGS *mp;
51 	SCR *sp;
52 	TEXT *tp;
53 	u_int32_t flags;
54 
55 	sp = *spp;
56 	gp = sp->gp;
57 	exp = EXP(sp);
58 
59 	/* Start the ex screen. */
60 	if (ex_init(sp))
61 		return (1);
62 
63 	/* Flush any saved messages. */
64 	while ((mp = SLIST_FIRST(gp->msgq)) != NULL) {
65 		gp->scr_msg(sp, mp->mtype, mp->buf, mp->len);
66 		SLIST_REMOVE_HEAD(gp->msgq, q);
67 		free(mp->buf);
68 		free(mp);
69 	}
70 
71 	/* If reading from a file, errors should have name and line info. */
72 	if (F_ISSET(gp, G_SCRIPTED)) {
73 		gp->excmd.if_lno = 1;
74 		gp->excmd.if_name = "script";
75 	}
76 
77 	/*
78 	 * !!!
79 	 * Initialize the text flags.  The beautify edit option historically
80 	 * applied to ex command input read from a file.  In addition, the
81 	 * first time a ^H was discarded from the input, there was a message,
82 	 * "^H discarded", that was displayed.  We don't bother.
83 	 */
84 	LF_INIT(TXT_BACKSLASH | TXT_CNTRLD | TXT_CR);
85 	for (;; ++gp->excmd.if_lno) {
86 		/* Display status line and flush. */
87 		if (F_ISSET(sp, SC_STATUS)) {
88 			if (!F_ISSET(sp, SC_EX_SILENT))
89 				msgq_status(sp, sp->lno, 0);
90 			F_CLR(sp, SC_STATUS);
91 		}
92 		(void)ex_fflush(sp);
93 
94 		/* Set the flags the user can reset. */
95 		if (O_ISSET(sp, O_BEAUTIFY))
96 			LF_SET(TXT_BEAUTIFY);
97 		if (O_ISSET(sp, O_PROMPT))
98 			LF_SET(TXT_PROMPT);
99 
100 		/* Clear any current interrupts, and get a command. */
101 		CLR_INTERRUPT(sp);
102 		if (ex_txt(sp, sp->tiq, ':', flags))
103 			return (1);
104 		if (INTERRUPTED(sp)) {
105 			(void)ex_puts(sp, "\n");
106 			(void)ex_fflush(sp);
107 			continue;
108 		}
109 
110 		/* Initialize the command structure. */
111 		CLEAR_EX_PARSER(&gp->excmd);
112 
113 		/*
114 		 * If the user entered a single carriage return, send
115 		 * ex_cmd() a separator -- it discards single newlines.
116 		 */
117 		tp = TAILQ_FIRST(sp->tiq);
118 		if (tp->len == 0) {
119 			gp->excmd.cp = L(" ");	/* __TK__ why not |? */
120 			gp->excmd.clen = 1;
121 		} else {
122 			gp->excmd.cp = tp->lb;
123 			gp->excmd.clen = tp->len;
124 		}
125 		F_INIT(&gp->excmd, E_NRSEP);
126 
127 		if (ex_cmd(sp) && F_ISSET(gp, G_SCRIPTED))
128 			return (1);
129 
130 		if (INTERRUPTED(sp)) {
131 			CLR_INTERRUPT(sp);
132 			msgq(sp, M_ERR, "170|Interrupted");
133 		}
134 
135 		/*
136 		 * If the last command caused a restart, or switched screens
137 		 * or into vi, return.
138 		 */
139 		if (F_ISSET(gp, G_SRESTART) || F_ISSET(sp, SC_SSWITCH | SC_VI)) {
140 			*spp = sp;
141 			break;
142 		}
143 
144 		/* If the last command switched files, we don't care. */
145 		F_CLR(sp, SC_FSWITCH);
146 
147 		/*
148 		 * If we're exiting this screen, move to the next one.  By
149 		 * definition, this means returning into vi, so return to the
150 		 * main editor loop.  The ordering is careful, don't discard
151 		 * the contents of sp until the end.
152 		 */
153 		if (F_ISSET(sp, SC_EXIT | SC_EXIT_FORCE)) {
154 			if (file_end(sp, NULL, F_ISSET(sp, SC_EXIT_FORCE)))
155 				return (1);
156 			*spp = screen_next(sp);
157 			return (screen_end(sp));
158 		}
159 	}
160 	return (0);
161 }
162 
163 /*
164  * ex_cmd --
165  *	The guts of the ex parser: parse and execute a string containing
166  *	ex commands.
167  *
168  * !!!
169  * This code MODIFIES the string that gets passed in, to delete quoting
170  * characters, etc.  The string cannot be readonly/text space, nor should
171  * you expect to use it again after ex_cmd() returns.
172  *
173  * !!!
174  * For the fun of it, if you want to see if a vi clone got the ex argument
175  * parsing right, try:
176  *
177  *	echo 'foo|bar' > file1; echo 'foo/bar' > file2;
178  *	vi
179  *	:edit +1|s/|/PIPE/|w file1| e file2|1 | s/\//SLASH/|wq
180  *
181  * or:	vi
182  *	:set|file|append|set|file
183  *
184  * For extra credit, try them in a startup .exrc file.
185  *
186  * PUBLIC: int ex_cmd(SCR *);
187  */
188 int
189 ex_cmd(SCR *sp)
190 {
191 	enum nresult nret;
192 	EX_PRIVATE *exp;
193 	EXCMD *ecp;
194 	GS *gp;
195 	MARK cur;
196 	recno_t lno;
197 	size_t arg1_len, discard, len;
198 	u_int32_t flags;
199 	long ltmp;
200 	int at_found, gv_found;
201 	int cnt, delim, isaddr, namelen;
202 	int newscreen, notempty, tmp, vi_address;
203 	CHAR_T *arg1, *s, *p, *t;
204 	CHAR_T ch = '\0';
205 	CHAR_T *n;
206 	char *np;
207 
208 	gp = sp->gp;
209 	exp = EXP(sp);
210 
211 	/*
212 	 * We always start running the command on the top of the stack.
213 	 * This means that *everything* must be resolved when we leave
214 	 * this function for any reason.
215 	 */
216 loop:	ecp = SLIST_FIRST(gp->ecq);
217 
218 	/* If we're reading a command from a file, set up error information. */
219 	if (ecp->if_name != NULL) {
220 		gp->if_lno = ecp->if_lno;
221 		gp->if_name = ecp->if_name;
222 	}
223 
224 	/*
225 	 * If a move to the end of the file is scheduled for this command,
226 	 * do it now.
227 	 */
228 	if (F_ISSET(ecp, E_MOVETOEND)) {
229 		if (db_last(sp, &sp->lno))
230 			goto rfail;
231 		sp->cno = 0;
232 		F_CLR(ecp, E_MOVETOEND);
233 	}
234 
235 	/* If we found a newline, increment the count now. */
236 	if (F_ISSET(ecp, E_NEWLINE)) {
237 		++gp->if_lno;
238 		++ecp->if_lno;
239 		F_CLR(ecp, E_NEWLINE);
240 	}
241 
242 	/* (Re)initialize the EXCMD structure, preserving some flags. */
243 	CLEAR_EX_CMD(ecp);
244 
245 	/* Initialize the argument structures. */
246 	if (argv_init(sp, ecp))
247 		goto err;
248 
249 	/* Initialize +cmd, saved command information. */
250 	arg1 = NULL;
251 	ecp->save_cmdlen = 0;
252 
253 	/* Skip <blank>s, empty lines.  */
254 	for (notempty = 0; ecp->clen > 0; ++ecp->cp, --ecp->clen)
255 		if ((ch = *ecp->cp) == '\n') {
256 			++gp->if_lno;
257 			++ecp->if_lno;
258 		} else if (cmdskip(ch))
259 			notempty = 1;
260 		else
261 			break;
262 
263 	/*
264 	 * !!!
265 	 * Permit extra colons at the start of the line.  Historically,
266 	 * ex/vi allowed a single extra one.  It's simpler not to count.
267 	 * The stripping is done here because, historically, any command
268 	 * could have preceding colons, e.g. ":g/pattern/:p" worked.
269 	 */
270 	if (ecp->clen != 0 && ch == ':') {
271 		notempty = 1;
272 		while (--ecp->clen > 0 && (ch = *++ecp->cp) == ':');
273 	}
274 
275 	/*
276 	 * Command lines that start with a double-quote are comments.
277 	 *
278 	 * !!!
279 	 * Historically, there was no escape or delimiter for a comment, e.g.
280 	 * :"foo|set was a single comment and nothing was output.  Since nvi
281 	 * permits users to escape <newline> characters into command lines, we
282 	 * have to check for that case.
283 	 */
284 	if (ecp->clen != 0 && ch == '"') {
285 		while (--ecp->clen > 0 && *++ecp->cp != '\n');
286 		if (*ecp->cp == '\n') {
287 			F_SET(ecp, E_NEWLINE);
288 			++ecp->cp;
289 			--ecp->clen;
290 		}
291 		goto loop;
292 	}
293 
294 	/* Skip whitespace. */
295 	for (; ecp->clen > 0; ++ecp->cp, --ecp->clen) {
296 		ch = *ecp->cp;
297 		if (!cmdskip(ch))
298 			break;
299 	}
300 
301 	/*
302 	 * The last point at which an empty line can mean do nothing.
303 	 *
304 	 * !!!
305 	 * Historically, in ex mode, lines containing only <blank> characters
306 	 * were the same as a single <carriage-return>, i.e. a default command.
307 	 * In vi mode, they were ignored.  In .exrc files this was a serious
308 	 * annoyance, as vi kept trying to treat them as print commands.  We
309 	 * ignore backward compatibility in this case, discarding lines that
310 	 * contain only <blank> characters from .exrc files.
311 	 *
312 	 * !!!
313 	 * This is where you end up when you're done a command, i.e. clen has
314 	 * gone to zero.  Continue if there are more commands to run.
315 	 */
316 	if (ecp->clen == 0 &&
317 	    (!notempty || F_ISSET(sp, SC_VI) || F_ISSET(ecp, E_BLIGNORE))) {
318 		if (ex_load(sp))
319 			goto rfail;
320 		ecp = SLIST_FIRST(gp->ecq);
321 		if (ecp->clen == 0)
322 			goto rsuccess;
323 		goto loop;
324 	}
325 
326 	/*
327 	 * Check to see if this is a command for which we may want to move
328 	 * the cursor back up to the previous line.  (The command :1<CR>
329 	 * wants a <newline> separator, but the command :<CR> wants to erase
330 	 * the command line.)  If the line is empty except for <blank>s,
331 	 * <carriage-return> or <eof>, we'll probably want to move up.  I
332 	 * don't think there's any way to get <blank> characters *after* the
333 	 * command character, but this is the ex parser, and I've been wrong
334 	 * before.
335 	 */
336 	if (F_ISSET(ecp, E_NRSEP) &&
337 	    ecp->clen != 0 && (ecp->clen != 1 || ecp->cp[0] != '\004'))
338 		F_CLR(ecp, E_NRSEP);
339 
340 	/* Parse command addresses. */
341 	if (ex_range(sp, ecp, &tmp))
342 		goto rfail;
343 	if (tmp)
344 		goto err;
345 
346 	/*
347 	 * Skip <blank>s and any more colons (the command :3,5:print
348 	 * worked, historically).
349 	 */
350 	for (; ecp->clen > 0; ++ecp->cp, --ecp->clen) {
351 		ch = *ecp->cp;
352 		if (!cmdskip(ch) && ch != ':')
353 			break;
354 	}
355 
356 	/*
357 	 * If no command, ex does the last specified of p, l, or #, and vi
358 	 * moves to the line.  Otherwise, determine the length of the command
359 	 * name by looking for the first non-alphabetic character.  (There
360 	 * are a few non-alphabetic characters in command names, but they're
361 	 * all single character commands.)  This isn't a great test, because
362 	 * it means that, for the command ":e +cut.c file", we'll report that
363 	 * the command "cut" wasn't known.  However, it makes ":e+35 file" work
364 	 * correctly.
365 	 *
366 	 * !!!
367 	 * Historically, lines with multiple adjacent (or <blank> separated)
368 	 * command separators were very strange.  For example, the command
369 	 * |||<carriage-return>, when the cursor was on line 1, displayed
370 	 * lines 2, 3 and 5 of the file.  In addition, the command "   |  "
371 	 * would only display the line after the next line, instead of the
372 	 * next two lines.  No ideas why.  It worked reasonably when executed
373 	 * from vi mode, and displayed lines 2, 3, and 4, so we do a default
374 	 * command for each separator.
375 	 */
376 #define	SINGLE_CHAR_COMMANDS	L("\004!#&*<=>@~")
377 	newscreen = 0;
378 	if (ecp->clen != 0 && ecp->cp[0] != '|' && ecp->cp[0] != '\n') {
379 		if (STRCHR(SINGLE_CHAR_COMMANDS, *ecp->cp)) {
380 			p = ecp->cp;
381 			++ecp->cp;
382 			--ecp->clen;
383 			namelen = 1;
384 		} else {
385 			for (p = ecp->cp;
386 			    ecp->clen > 0; --ecp->clen, ++ecp->cp)
387 				if (!isazAZ(*ecp->cp))
388 					break;
389 			if ((namelen = ecp->cp - p) == 0) {
390 				msgq(sp, M_ERR, "080|Unknown command name");
391 				goto err;
392 			}
393 		}
394 
395 		/*
396 		 * !!!
397 		 * Historic vi permitted flags to immediately follow any
398 		 * subset of the 'delete' command, but then did not permit
399 		 * further arguments (flag, buffer, count).  Make it work.
400 		 * Permit further arguments for the few shreds of dignity
401 		 * it offers.
402 		 *
403 		 * Adding commands that start with 'd', and match "delete"
404 		 * up to a l, p, +, - or # character can break this code.
405 		 *
406 		 * !!!
407 		 * Capital letters beginning the command names ex, edit,
408 		 * next, previous, tag and visual (in vi mode) indicate the
409 		 * command should happen in a new screen.
410 		 */
411 		switch (p[0]) {
412 		case 'd':
413 			for (s = p,
414 			    n = cmds[C_DELETE].name; *s == *n; ++s, ++n);
415 			if (s[0] == 'l' || s[0] == 'p' || s[0] == '+' ||
416 			    s[0] == '-' || s[0] == '^' || s[0] == '#') {
417 				len = (ecp->cp - p) - (s - p);
418 				ecp->cp -= len;
419 				ecp->clen += len;
420 				ecp->rcmd = cmds[C_DELETE];
421 				ecp->rcmd.syntax = "1bca1";
422 				ecp->cmd = &ecp->rcmd;
423 				goto skip_srch;
424 			}
425 			break;
426 		case 'E': case 'F': case 'N': case 'P': case 'T': case 'V':
427 			newscreen = 1;
428 			p[0] = tolower(p[0]);
429 			break;
430 		}
431 
432 		/*
433 		 * Search the table for the command.
434 		 *
435 		 * !!!
436 		 * Historic vi permitted the mark to immediately follow the
437 		 * 'k' in the 'k' command.  Make it work.
438 		 *
439 		 * !!!
440 		 * Historic vi permitted any flag to follow the s command, e.g.
441 		 * "s/e/E/|s|sgc3p" was legal.  Make the command "sgc" work.
442 		 * Since the following characters all have to be flags, i.e.
443 		 * alphabetics, we can let the s command routine return errors
444 		 * if it was some illegal command string.  This code will break
445 		 * if an "sg" or similar command is ever added.  The substitute
446 		 * code doesn't care if it's a "cgr" flag or a "#lp" flag that
447 		 * follows the 's', but we limit the choices here to "cgr" so
448 		 * that we get unknown command messages for wrong combinations.
449 		 */
450 		if ((ecp->cmd = ex_comm_search(p, namelen)) == NULL)
451 			switch (p[0]) {
452 			case 'k':
453 				if (namelen == 2) {
454 					ecp->cp -= namelen - 1;
455 					ecp->clen += namelen - 1;
456 					ecp->cmd = &cmds[C_K];
457 					break;
458 				}
459 				goto unknown;
460 			case 's':
461 				for (s = p + 1, cnt = namelen; --cnt; ++s)
462 					if (s[0] != 'c' &&
463 					    s[0] != 'g' && s[0] != 'r')
464 						break;
465 				if (cnt == 0) {
466 					ecp->cp -= namelen - 1;
467 					ecp->clen += namelen - 1;
468 					ecp->rcmd = cmds[C_SUBSTITUTE];
469 					ecp->rcmd.fn = ex_subagain;
470 					ecp->cmd = &ecp->rcmd;
471 					break;
472 				}
473 				/* FALLTHROUGH */
474 			default:
475 unknown:			if (newscreen)
476 					p[0] = toupper(p[0]);
477 				ex_unknown(sp, p, namelen);
478 				goto err;
479 			}
480 
481 		/*
482 		 * The visual command has a different syntax when called
483 		 * from ex than when called from a vi colon command.  FMH.
484 		 * Make the change now, before we test for the newscreen
485 		 * semantic, so that we're testing the right one.
486 		 */
487 skip_srch:	if (ecp->cmd == &cmds[C_VISUAL_EX] && F_ISSET(sp, SC_VI))
488 			ecp->cmd = &cmds[C_VISUAL_VI];
489 
490 		/*
491 		 * !!!
492 		 * Historic vi permitted a capital 'P' at the beginning of
493 		 * any command that started with 'p'.  Probably wanted the
494 		 * P[rint] command for backward compatibility, and the code
495 		 * just made Preserve and Put work by accident.  Nvi uses
496 		 * Previous to mean previous-in-a-new-screen, so be careful.
497 		 */
498 		if (newscreen && !F_ISSET(ecp->cmd, E_NEWSCREEN) &&
499 		    (ecp->cmd == &cmds[C_PRINT] ||
500 		    ecp->cmd == &cmds[C_PRESERVE]))
501 			newscreen = 0;
502 
503 		/* Test for a newscreen associated with this command. */
504 		if (newscreen && !F_ISSET(ecp->cmd, E_NEWSCREEN))
505 			goto unknown;
506 
507 		/* Secure means no shell access. */
508 		if (F_ISSET(ecp->cmd, E_SECURE) && O_ISSET(sp, O_SECURE)) {
509 			ex_wemsg(sp, ecp->cmd->name, EXM_SECURE);
510 			goto err;
511 		}
512 
513 		/*
514 		 * Multiple < and > characters; another "feature".  Note,
515 		 * The string passed to the underlying function may not be
516 		 * nul terminated in this case.
517 		 */
518 		if ((ecp->cmd == &cmds[C_SHIFTL] && *p == '<') ||
519 		    (ecp->cmd == &cmds[C_SHIFTR] && *p == '>')) {
520 			for (ch = *p;
521 			    ecp->clen > 0; --ecp->clen, ++ecp->cp)
522 				if (*ecp->cp != ch)
523 					break;
524 			if (argv_exp0(sp, ecp, p, ecp->cp - p))
525 				goto err;
526 		}
527 
528 		/* Set the format style flags for the next command. */
529 		if (ecp->cmd == &cmds[C_HASH])
530 			exp->fdef = E_C_HASH;
531 		else if (ecp->cmd == &cmds[C_LIST])
532 			exp->fdef = E_C_LIST;
533 		else if (ecp->cmd == &cmds[C_PRINT])
534 			exp->fdef = E_C_PRINT;
535 		F_CLR(ecp, E_USELASTCMD);
536 	} else {
537 		/* Print is the default command. */
538 		ecp->cmd = &cmds[C_PRINT];
539 
540 		/* Set the saved format flags. */
541 		F_SET(ecp, exp->fdef);
542 
543 		/*
544 		 * !!!
545 		 * If no address was specified, and it's not a global command,
546 		 * we up the address by one.  (I have no idea why globals are
547 		 * exempted, but it's (ahem) historic practice.)
548 		 */
549 		if (ecp->addrcnt == 0 && !F_ISSET(sp, SC_EX_GLOBAL)) {
550 			ecp->addrcnt = 1;
551 			ecp->addr1.lno = sp->lno + 1;
552 			ecp->addr1.cno = sp->cno;
553 		}
554 
555 		F_SET(ecp, E_USELASTCMD);
556 	}
557 
558 	/*
559 	 * !!!
560 	 * Historically, the number option applied to both ex and vi.  One
561 	 * strangeness was that ex didn't switch display formats until a
562 	 * command was entered, e.g. <CR>'s after the set didn't change to
563 	 * the new format, but :1p would.
564 	 */
565 	if (O_ISSET(sp, O_NUMBER)) {
566 		F_SET(ecp, E_OPTNUM);
567 		FL_SET(ecp->iflags, E_C_HASH);
568 	} else
569 		F_CLR(ecp, E_OPTNUM);
570 
571 	/* Check for ex mode legality. */
572 	if (F_ISSET(sp, SC_EX) && (F_ISSET(ecp->cmd, E_VIONLY) || newscreen)) {
573 		msgq_wstr(sp, M_ERR, ecp->cmd->name,
574 		    "082|%s: command not available in ex mode");
575 		goto err;
576 	}
577 
578 	/* Add standard command flags. */
579 	F_SET(ecp, ecp->cmd->flags);
580 	if (!newscreen)
581 		F_CLR(ecp, E_NEWSCREEN);
582 
583 	/*
584 	 * There are three normal termination cases for an ex command.  They
585 	 * are the end of the string (ecp->clen), or unescaped (by <literal
586 	 * next> characters) <newline> or '|' characters.  As we're now past
587 	 * possible addresses, we can determine how long the command is, so we
588 	 * don't have to look for all the possible terminations.  Naturally,
589 	 * there are some exciting special cases:
590 	 *
591 	 * 1: The bang, global, v and the filter versions of the read and
592 	 *    write commands are delimited by <newline>s (they can contain
593 	 *    shell pipes).
594 	 * 2: The ex, edit, next and visual in vi mode commands all take ex
595 	 *    commands as their first arguments.
596 	 * 3: The s command takes an RE as its first argument, and wants it
597 	 *    to be specially delimited.
598 	 *
599 	 * Historically, '|' characters in the first argument of the ex, edit,
600 	 * next, vi visual, and s commands didn't delimit the command.  And,
601 	 * in the filter cases for read and write, and the bang, global and v
602 	 * commands, they did not delimit the command at all.
603 	 *
604 	 * For example, the following commands were legal:
605 	 *
606 	 *	:edit +25|s/abc/ABC/ file.c
607 	 *	:s/|/PIPE/
608 	 *	:read !spell % | columnate
609 	 *	:global/pattern/p|l
610 	 *
611 	 * It's not quite as simple as it sounds, however.  The command:
612 	 *
613 	 *	:s/a/b/|s/c/d|set
614 	 *
615 	 * was also legal, i.e. the historic ex parser (using the word loosely,
616 	 * since "parser" implies some regularity of syntax) delimited the RE's
617 	 * based on its delimiter and not anything so irretrievably vulgar as a
618 	 * command syntax.
619 	 *
620 	 * Anyhow, the following code makes this all work.  First, for the
621 	 * special cases we move past their special argument(s).  Then, we
622 	 * do normal command processing on whatever is left.  Barf-O-Rama.
623 	 */
624 	discard = 0;		/* Characters discarded from the command. */
625 	arg1_len = 0;
626 	ecp->save_cmd = ecp->cp;
627 	if (ecp->cmd == &cmds[C_EDIT] || ecp->cmd == &cmds[C_EX] ||
628 	    ecp->cmd == &cmds[C_NEXT] || ecp->cmd == &cmds[C_VISUAL_VI] ||
629 	    ecp->cmd == &cmds[C_VSPLIT]) {
630 		/*
631 		 * Move to the next non-whitespace character.  A '!'
632 		 * immediately following the command is eaten as a
633 		 * force flag.
634 		 */
635 		if (ecp->clen > 0 && *ecp->cp == '!') {
636 			++ecp->cp;
637 			--ecp->clen;
638 			FL_SET(ecp->iflags, E_C_FORCE);
639 
640 			/* Reset, don't reparse. */
641 			ecp->save_cmd = ecp->cp;
642 		}
643 		for (; ecp->clen > 0; --ecp->clen, ++ecp->cp)
644 			if (!cmdskip(*ecp->cp))
645 				break;
646 		/*
647 		 * QUOTING NOTE:
648 		 *
649 		 * The historic implementation ignored all escape characters
650 		 * so there was no way to put a space or newline into the +cmd
651 		 * field.  We do a simplistic job of fixing it by moving to the
652 		 * first whitespace character that isn't escaped.  The escaping
653 		 * characters are stripped as no longer useful.
654 		 */
655 		if (ecp->clen > 0 && *ecp->cp == '+') {
656 			++ecp->cp;
657 			--ecp->clen;
658 			for (arg1 = p = ecp->cp;
659 			    ecp->clen > 0; --ecp->clen, ++ecp->cp) {
660 				ch = *ecp->cp;
661 				if (IS_ESCAPE(sp, ecp, ch) &&
662 				    ecp->clen > 1) {
663 					++discard;
664 					--ecp->clen;
665 					ch = *++ecp->cp;
666 				} else if (cmdskip(ch))
667 					break;
668 				*p++ = ch;
669 			}
670 			arg1_len = ecp->cp - arg1;
671 
672 			/* Reset, so the first argument isn't reparsed. */
673 			ecp->save_cmd = ecp->cp;
674 		}
675 	} else if (ecp->cmd == &cmds[C_BANG] ||
676 	    ecp->cmd == &cmds[C_GLOBAL] || ecp->cmd == &cmds[C_V]) {
677 		/*
678 		 * QUOTING NOTE:
679 		 *
680 		 * We use backslashes to escape <newline> characters, although
681 		 * this wasn't historic practice for the bang command.  It was
682 		 * for the global and v commands, and it's common usage when
683 		 * doing text insert during the command.  Escaping characters
684 		 * are stripped as no longer useful.
685 		 */
686 		for (p = ecp->cp; ecp->clen > 0; --ecp->clen, ++ecp->cp) {
687 			ch = *ecp->cp;
688 			if (ch == '\\' && ecp->clen > 1 && ecp->cp[1] == '\n') {
689 				++discard;
690 				--ecp->clen;
691 				ch = *++ecp->cp;
692 
693 				++gp->if_lno;
694 				++ecp->if_lno;
695 			} else if (ch == '\n')
696 				break;
697 			*p++ = ch;
698 		}
699 	} else if (ecp->cmd == &cmds[C_READ] || ecp->cmd == &cmds[C_WRITE]) {
700 		/*
701 		 * For write commands, if the next character is a <blank>, and
702 		 * the next non-blank character is a '!', it's a filter command
703 		 * and we want to eat everything up to the <newline>.  For read
704 		 * commands, if the next non-blank character is a '!', it's a
705 		 * filter command and we want to eat everything up to the next
706 		 * <newline>.  Otherwise, we're done.
707 		 */
708 		for (tmp = 0; ecp->clen > 0; --ecp->clen, ++ecp->cp) {
709 			ch = *ecp->cp;
710 			if (cmdskip(ch))
711 				tmp = 1;
712 			else
713 				break;
714 		}
715 		if (ecp->clen > 0 && ch == '!' &&
716 		    (ecp->cmd == &cmds[C_READ] || tmp))
717 			for (; ecp->clen > 0; --ecp->clen, ++ecp->cp)
718 				if (ecp->cp[0] == '\n')
719 					break;
720 	} else if (ecp->cmd == &cmds[C_SUBSTITUTE]) {
721 		/*
722 		 * Move to the next non-whitespace character, we'll use it as
723 		 * the delimiter.  If the character isn't an alphanumeric or
724 		 * a '|', it's the delimiter, so parse it.  Otherwise, we're
725 		 * into something like ":s g", so use the special s command.
726 		 */
727 		for (; ecp->clen > 0; --ecp->clen, ++ecp->cp)
728 			if (!cmdskip(ecp->cp[0]))
729 				break;
730 
731 		if (is09azAZ(ecp->cp[0]) || ecp->cp[0] == '|') {
732 			ecp->rcmd = cmds[C_SUBSTITUTE];
733 			ecp->rcmd.fn = ex_subagain;
734 			ecp->cmd = &ecp->rcmd;
735 		} else if (ecp->clen > 0) {
736 			/*
737 			 * QUOTING NOTE:
738 			 *
739 			 * Backslashes quote delimiter characters for RE's.
740 			 * The backslashes are NOT removed since they'll be
741 			 * used by the RE code.  Move to the third delimiter
742 			 * that's not escaped (or the end of the command).
743 			 */
744 			delim = *ecp->cp;
745 			++ecp->cp;
746 			--ecp->clen;
747 			for (cnt = 2; ecp->clen > 0 &&
748 			    cnt != 0; --ecp->clen, ++ecp->cp)
749 				if (ecp->cp[0] == '\\' &&
750 				    ecp->clen > 1) {
751 					++ecp->cp;
752 					--ecp->clen;
753 				} else if (ecp->cp[0] == delim)
754 					--cnt;
755 		}
756 	}
757 
758 	/*
759 	 * Use normal quoting and termination rules to find the end of this
760 	 * command.
761 	 *
762 	 * QUOTING NOTE:
763 	 *
764 	 * Historically, vi permitted ^V's to escape <newline>'s in the .exrc
765 	 * file.  It was almost certainly a bug, but that's what bug-for-bug
766 	 * compatibility means, Grasshopper.  Also, ^V's escape the command
767 	 * delimiters.  Literal next quote characters in front of the newlines,
768 	 * '|' characters or literal next characters are stripped as they're
769 	 * no longer useful.
770 	 */
771 	vi_address = ecp->clen != 0 && ecp->cp[0] != '\n';
772 	for (p = ecp->cp; ecp->clen > 0; --ecp->clen, ++ecp->cp) {
773 		ch = ecp->cp[0];
774 		if (IS_ESCAPE(sp, ecp, ch) && ecp->clen > 1) {
775 			CHAR_T tmp = ecp->cp[1];
776 			if (tmp == '\n' || tmp == '|') {
777 				if (tmp == '\n') {
778 					++gp->if_lno;
779 					++ecp->if_lno;
780 				}
781 				++discard;
782 				--ecp->clen;
783 				++ecp->cp;
784 				ch = tmp;
785 			}
786 		} else if (ch == '\n' || ch == '|') {
787 			if (ch == '\n')
788 				F_SET(ecp, E_NEWLINE);
789 			--ecp->clen;
790 			break;
791 		}
792 		*p++ = ch;
793 	}
794 
795 	/*
796 	 * Save off the next command information, go back to the
797 	 * original start of the command.
798 	 */
799 	p = ecp->cp + 1;
800 	ecp->cp = ecp->save_cmd;
801 	ecp->save_cmd = p;
802 	ecp->save_cmdlen = ecp->clen;
803 	ecp->clen = ((ecp->save_cmd - ecp->cp) - 1) - discard;
804 
805 	/*
806 	 * QUOTING NOTE:
807 	 *
808 	 * The "set tags" command historically used a backslash, not the
809 	 * user's literal next character, to escape whitespace.  Handle
810 	 * it here instead of complicating the argv_exp3() code.  Note,
811 	 * this isn't a particularly complex trap, and if backslashes were
812 	 * legal in set commands, this would have to be much more complicated.
813 	 */
814 	if (ecp->cmd == &cmds[C_SET]) {
815 		for (p = ecp->cp, len = ecp->clen; len > 0; --len, ++p)
816 			if (IS_ESCAPE(sp, ecp, *p) && len > 1) {
817 				--len;
818 				++p;
819 			} else if (*p == '\\')
820 				*p = CH_LITERAL;
821 	}
822 
823 	/*
824 	 * Set the default addresses.  It's an error to specify an address for
825 	 * a command that doesn't take them.  If two addresses are specified
826 	 * for a command that only takes one, lose the first one.  Two special
827 	 * cases here, some commands take 0 or 2 addresses.  For most of them
828 	 * (the E_ADDR2_ALL flag), 0 defaults to the entire file.  For one
829 	 * (the `!' command, the E_ADDR2_NONE flag), 0 defaults to no lines.
830 	 *
831 	 * Also, if the file is empty, some commands want to use an address of
832 	 * 0, i.e. the entire file is 0 to 0, and the default first address is
833 	 * 0.  Otherwise, an entire file is 1 to N and the default line is 1.
834 	 * Note, we also add the E_ADDR_ZERO flag to the command flags, for the
835 	 * case where the 0 address is only valid if it's a default address.
836 	 *
837 	 * Also, set a flag if we set the default addresses.  Some commands
838 	 * (ex: z) care if the user specified an address or if we just used
839 	 * the current cursor.
840 	 */
841 	switch (F_ISSET(ecp, E_ADDR1 | E_ADDR2 | E_ADDR2_ALL | E_ADDR2_NONE)) {
842 	case E_ADDR1:				/* One address: */
843 		switch (ecp->addrcnt) {
844 		case 0:				/* Default cursor/empty file. */
845 			ecp->addrcnt = 1;
846 			F_SET(ecp, E_ADDR_DEF);
847 			if (F_ISSET(ecp, E_ADDR_ZERODEF)) {
848 				if (db_last(sp, &lno))
849 					goto err;
850 				if (lno == 0) {
851 					ecp->addr1.lno = 0;
852 					F_SET(ecp, E_ADDR_ZERO);
853 				} else
854 					ecp->addr1.lno = sp->lno;
855 			} else
856 				ecp->addr1.lno = sp->lno;
857 			ecp->addr1.cno = sp->cno;
858 			break;
859 		case 1:
860 			break;
861 		case 2:				/* Lose the first address. */
862 			ecp->addrcnt = 1;
863 			ecp->addr1 = ecp->addr2;
864 		}
865 		break;
866 	case E_ADDR2_NONE:			/* Zero/two addresses: */
867 		if (ecp->addrcnt == 0)		/* Default to nothing. */
868 			break;
869 		goto two_addr;
870 	case E_ADDR2_ALL:			/* Zero/two addresses: */
871 		if (ecp->addrcnt == 0) {	/* Default entire/empty file. */
872 			F_SET(ecp, E_ADDR_DEF);
873 			ecp->addrcnt = 2;
874 			if (sp->ep == NULL)
875 				ecp->addr2.lno = 0;
876 			else if (db_last(sp, &ecp->addr2.lno))
877 				goto err;
878 			if (F_ISSET(ecp, E_ADDR_ZERODEF) &&
879 			    ecp->addr2.lno == 0) {
880 				ecp->addr1.lno = 0;
881 				F_SET(ecp, E_ADDR_ZERO);
882 			} else
883 				ecp->addr1.lno = 1;
884 			ecp->addr1.cno = ecp->addr2.cno = 0;
885 			F_SET(ecp, E_ADDR2_ALL);
886 			break;
887 		}
888 		/* FALLTHROUGH */
889 	case E_ADDR2:				/* Two addresses: */
890 two_addr:	switch (ecp->addrcnt) {
891 		case 0:				/* Default cursor/empty file. */
892 			ecp->addrcnt = 2;
893 			F_SET(ecp, E_ADDR_DEF);
894 			if (sp->lno == 1 &&
895 			    F_ISSET(ecp, E_ADDR_ZERODEF)) {
896 				if (db_last(sp, &lno))
897 					goto err;
898 				if (lno == 0) {
899 					ecp->addr1.lno = ecp->addr2.lno = 0;
900 					F_SET(ecp, E_ADDR_ZERO);
901 				} else
902 					ecp->addr1.lno =
903 					    ecp->addr2.lno = sp->lno;
904 			} else
905 				ecp->addr1.lno = ecp->addr2.lno = sp->lno;
906 			ecp->addr1.cno = ecp->addr2.cno = sp->cno;
907 			break;
908 		case 1:				/* Default to first address. */
909 			ecp->addrcnt = 2;
910 			ecp->addr2 = ecp->addr1;
911 			break;
912 		case 2:
913 			break;
914 		}
915 		break;
916 	default:
917 		if (ecp->addrcnt)		/* Error. */
918 			goto usage;
919 	}
920 
921 	/*
922 	 * !!!
923 	 * The ^D scroll command historically scrolled the value of the scroll
924 	 * option or to EOF.  It was an error if the cursor was already at EOF.
925 	 * (Leading addresses were permitted, but were then ignored.)
926 	 */
927 	if (ecp->cmd == &cmds[C_SCROLL]) {
928 		ecp->addrcnt = 2;
929 		ecp->addr1.lno = sp->lno + 1;
930 		ecp->addr2.lno = sp->lno + O_VAL(sp, O_SCROLL);
931 		ecp->addr1.cno = ecp->addr2.cno = sp->cno;
932 		if (db_last(sp, &lno))
933 			goto err;
934 		if (lno != 0 && lno > sp->lno && ecp->addr2.lno > lno)
935 			ecp->addr2.lno = lno;
936 	}
937 
938 	ecp->flagoff = 0;
939 	for (np = ecp->cmd->syntax; *np != '\0'; ++np) {
940 		/*
941 		 * The force flag is sensitive to leading whitespace, i.e.
942 		 * "next !" is different from "next!".  Handle it before
943 		 * skipping leading <blank>s.
944 		 */
945 		if (*np == '!') {
946 			if (ecp->clen > 0 && *ecp->cp == '!') {
947 				++ecp->cp;
948 				--ecp->clen;
949 				FL_SET(ecp->iflags, E_C_FORCE);
950 			}
951 			continue;
952 		}
953 
954 		/* Skip leading <blank>s. */
955 		for (; ecp->clen > 0; --ecp->clen, ++ecp->cp)
956 			if (!cmdskip(*ecp->cp))
957 				break;
958 		if (ecp->clen == 0)
959 			break;
960 
961 		switch (*np) {
962 		case '1':				/* +, -, #, l, p */
963 			/*
964 			 * !!!
965 			 * Historically, some flags were ignored depending
966 			 * on where they occurred in the command line.  For
967 			 * example, in the command, ":3+++p--#", historic vi
968 			 * acted on the '#' flag, but ignored the '-' flags.
969 			 * It's unambiguous what the flags mean, so we just
970 			 * handle them regardless of the stupidity of their
971 			 * location.
972 			 */
973 			for (; ecp->clen; --ecp->clen, ++ecp->cp)
974 				switch (*ecp->cp) {
975 				case '+':
976 					++ecp->flagoff;
977 					break;
978 				case '-':
979 				case '^':
980 					--ecp->flagoff;
981 					break;
982 				case '#':
983 					F_CLR(ecp, E_OPTNUM);
984 					FL_SET(ecp->iflags, E_C_HASH);
985 					exp->fdef |= E_C_HASH;
986 					break;
987 				case 'l':
988 					FL_SET(ecp->iflags, E_C_LIST);
989 					exp->fdef |= E_C_LIST;
990 					break;
991 				case 'p':
992 					FL_SET(ecp->iflags, E_C_PRINT);
993 					exp->fdef |= E_C_PRINT;
994 					break;
995 				default:
996 					goto end_case1;
997 				}
998 end_case1:		break;
999 		case '2':				/* -, ., +, ^ */
1000 		case '3':				/* -, ., +, ^, = */
1001 			for (; ecp->clen; --ecp->clen, ++ecp->cp)
1002 				switch (*ecp->cp) {
1003 				case '-':
1004 					FL_SET(ecp->iflags, E_C_DASH);
1005 					break;
1006 				case '.':
1007 					FL_SET(ecp->iflags, E_C_DOT);
1008 					break;
1009 				case '+':
1010 					FL_SET(ecp->iflags, E_C_PLUS);
1011 					break;
1012 				case '^':
1013 					FL_SET(ecp->iflags, E_C_CARAT);
1014 					break;
1015 				case '=':
1016 					if (*np == '3') {
1017 						FL_SET(ecp->iflags, E_C_EQUAL);
1018 						break;
1019 					}
1020 					/* FALLTHROUGH */
1021 				default:
1022 					goto end_case23;
1023 				}
1024 end_case23:		break;
1025 		case 'b':				/* buffer */
1026 			/*
1027 			 * !!!
1028 			 * Historically, "d #" was a delete with a flag, not a
1029 			 * delete into the '#' buffer.  If the current command
1030 			 * permits a flag, don't use one as a buffer.  However,
1031 			 * the 'l' and 'p' flags were legal buffer names in the
1032 			 * historic ex, and were used as buffers, not flags.
1033 			 */
1034 			if ((ecp->cp[0] == '+' || ecp->cp[0] == '-' ||
1035 			    ecp->cp[0] == '^' || ecp->cp[0] == '#') &&
1036 			    strchr(np, '1') != NULL)
1037 				break;
1038 			/*
1039 			 * !!!
1040 			 * Digits can't be buffer names in ex commands, or the
1041 			 * command "d2" would be a delete into buffer '2', and
1042 			 * not a two-line deletion.
1043 			 */
1044 			if (!ISDIGIT(ecp->cp[0])) {
1045 				ecp->buffer = *ecp->cp;
1046 				++ecp->cp;
1047 				--ecp->clen;
1048 				FL_SET(ecp->iflags, E_C_BUFFER);
1049 			}
1050 			break;
1051 		case 'c':				/* count [01+a] */
1052 			++np;
1053 			/* Validate any signed value. */
1054 			if (!ISDIGIT(*ecp->cp) && (*np != '+' ||
1055 			    (*ecp->cp != '+' && *ecp->cp != '-')))
1056 				break;
1057 			/* If a signed value, set appropriate flags. */
1058 			if (*ecp->cp == '-')
1059 				FL_SET(ecp->iflags, E_C_COUNT_NEG);
1060 			else if (*ecp->cp == '+')
1061 				FL_SET(ecp->iflags, E_C_COUNT_POS);
1062 			if ((nret =
1063 			    nget_slong(&ltmp, ecp->cp, &t, 10)) != NUM_OK) {
1064 				ex_badaddr(sp, NULL, A_NOTSET, nret);
1065 				goto err;
1066 			}
1067 			if (ltmp == 0 && *np != '0') {
1068 				msgq(sp, M_ERR, "083|Count may not be zero");
1069 				goto err;
1070 			}
1071 			ecp->clen -= (t - ecp->cp);
1072 			ecp->cp = t;
1073 
1074 			/*
1075 			 * Counts as address offsets occur in commands taking
1076 			 * two addresses.  Historic vi practice was to use
1077 			 * the count as an offset from the *second* address.
1078 			 *
1079 			 * Set a count flag; some underlying commands (see
1080 			 * join) do different things with counts than with
1081 			 * line addresses.
1082 			 */
1083 			if (*np == 'a') {
1084 				ecp->addr1 = ecp->addr2;
1085 				ecp->addr2.lno = ecp->addr1.lno + ltmp - 1;
1086 			} else
1087 				ecp->count = ltmp;
1088 			FL_SET(ecp->iflags, E_C_COUNT);
1089 			break;
1090 		case 'f':				/* file */
1091 			if (argv_exp2(sp, ecp, ecp->cp, ecp->clen))
1092 				goto err;
1093 			goto arg_cnt_chk;
1094 		case 'l':				/* line */
1095 			/*
1096 			 * Get a line specification.
1097 			 *
1098 			 * If the line was a search expression, we may have
1099 			 * changed state during the call, and we're now
1100 			 * searching the file.  Push ourselves onto the state
1101 			 * stack.
1102 			 */
1103 			if (ex_line(sp, ecp, &cur, &isaddr, &tmp))
1104 				goto rfail;
1105 			if (tmp)
1106 				goto err;
1107 
1108 			/* Line specifications are always required. */
1109 			if (!isaddr) {
1110 				msgq_wstr(sp, M_ERR, ecp->cp,
1111 				     "084|%s: bad line specification");
1112 				goto err;
1113 			}
1114 			/*
1115 			 * The target line should exist for these commands,
1116 			 * but 0 is legal for them as well.
1117 			 */
1118 			if (cur.lno != 0 && !db_exist(sp, cur.lno)) {
1119 				ex_badaddr(sp, NULL, A_EOF, NUM_OK);
1120 				goto err;
1121 			}
1122 			ecp->lineno = cur.lno;
1123 			break;
1124 		case 'S':				/* string, file exp. */
1125 			if (ecp->clen != 0) {
1126 				if (argv_exp1(sp, ecp, ecp->cp,
1127 				    ecp->clen, ecp->cmd == &cmds[C_BANG]))
1128 					goto err;
1129 				goto addr_verify;
1130 			}
1131 			/* FALLTHROUGH */
1132 		case 's':				/* string */
1133 			if (argv_exp0(sp, ecp, ecp->cp, ecp->clen))
1134 				goto err;
1135 			goto addr_verify;
1136 		case 'W':				/* word string */
1137 			/*
1138 			 * QUOTING NOTE:
1139 			 *
1140 			 * Literal next characters escape the following
1141 			 * character.  Quoting characters are stripped here
1142 			 * since they are no longer useful.
1143 			 *
1144 			 * First there was the word.
1145 			 */
1146 			for (p = t = ecp->cp;
1147 			    ecp->clen > 0; --ecp->clen, ++ecp->cp) {
1148 				ch = *ecp->cp;
1149 				if (IS_ESCAPE(sp,
1150 				    ecp, ch) && ecp->clen > 1) {
1151 					--ecp->clen;
1152 					*p++ = *++ecp->cp;
1153 				} else if (cmdskip(ch)) {
1154 					++ecp->cp;
1155 					--ecp->clen;
1156 					break;
1157 				} else
1158 					*p++ = ch;
1159 			}
1160 			if (argv_exp0(sp, ecp, t, p - t))
1161 				goto err;
1162 
1163 			/* Delete intervening whitespace. */
1164 			for (; ecp->clen > 0;
1165 			    --ecp->clen, ++ecp->cp) {
1166 				ch = *ecp->cp;
1167 				if (!cmdskip(ch))
1168 					break;
1169 			}
1170 			if (ecp->clen == 0)
1171 				goto usage;
1172 
1173 			/* Followed by the string. */
1174 			for (p = t = ecp->cp; ecp->clen > 0;
1175 			    --ecp->clen, ++ecp->cp, ++p) {
1176 				ch = *ecp->cp;
1177 				if (IS_ESCAPE(sp,
1178 				    ecp, ch) && ecp->clen > 1) {
1179 					--ecp->clen;
1180 					*p = *++ecp->cp;
1181 				} else
1182 					*p = ch;
1183 			}
1184 			if (argv_exp0(sp, ecp, t, p - t))
1185 				goto err;
1186 			goto addr_verify;
1187 		case 'w':				/* word */
1188 			if (argv_exp3(sp, ecp, ecp->cp, ecp->clen))
1189 				goto err;
1190 arg_cnt_chk:		if (*++np != 'N') {		/* N */
1191 				/*
1192 				 * If a number is specified, must either be
1193 				 * 0 or that number, if optional, and that
1194 				 * number, if required.
1195 				 */
1196 				tmp = *np - '0';
1197 				if ((*++np != 'o' || exp->argsoff != 0) &&
1198 				    exp->argsoff != tmp)
1199 					goto usage;
1200 			}
1201 			goto addr_verify;
1202 		default: {
1203 			size_t nlen;
1204 			char *nstr;
1205 
1206 			INT2CHAR(sp, ecp->cmd->name, STRLEN(ecp->cmd->name) + 1,
1207 			    nstr, nlen);
1208 			msgq(sp, M_ERR,
1209 			    "085|Internal syntax table error (%s: %s)",
1210 			    nstr, KEY_NAME(sp, *np));
1211 		}
1212 		}
1213 	}
1214 
1215 	/* Skip trailing whitespace. */
1216 	for (; ecp->clen > 0; --ecp->clen) {
1217 		ch = *ecp->cp++;
1218 		if (!cmdskip(ch))
1219 			break;
1220 	}
1221 
1222 	/*
1223 	 * There shouldn't be anything left, and no more required fields,
1224 	 * i.e neither 'l' or 'r' in the syntax string.
1225 	 */
1226 	if (ecp->clen != 0 || strpbrk(np, "lr")) {
1227 usage:		msgq(sp, M_ERR, "086|Usage: %s", ecp->cmd->usage);
1228 		goto err;
1229 	}
1230 
1231 	/*
1232 	 * Verify that the addresses are legal.  Check the addresses here,
1233 	 * because this is a place where all ex addresses pass through.
1234 	 * (They don't all pass through ex_line(), for instance.)  We're
1235 	 * assuming that any non-existent line doesn't exist because it's
1236 	 * past the end-of-file.  That's a pretty good guess.
1237 	 *
1238 	 * If it's a "default vi command", an address of zero is okay.
1239 	 */
1240 addr_verify:
1241 	switch (ecp->addrcnt) {
1242 	case 2:
1243 		/*
1244 		 * Historic ex/vi permitted commands with counts to go past
1245 		 * EOF.  So, for example, if the file only had 5 lines, the
1246 		 * ex command "1,6>" would fail, but the command ">300"
1247 		 * would succeed.  Since we don't want to have to make all
1248 		 * of the underlying commands handle random line numbers,
1249 		 * fix it here.
1250 		 */
1251 		if (ecp->addr2.lno == 0) {
1252 			if (!F_ISSET(ecp, E_ADDR_ZERO) &&
1253 			    (F_ISSET(sp, SC_EX) ||
1254 			    !F_ISSET(ecp, E_USELASTCMD))) {
1255 				ex_badaddr(sp, ecp->cmd, A_ZERO, NUM_OK);
1256 				goto err;
1257 			}
1258 		} else if (!db_exist(sp, ecp->addr2.lno)) {
1259 			if (FL_ISSET(ecp->iflags, E_C_COUNT)) {
1260 				if (db_last(sp, &lno))
1261 					goto err;
1262 				ecp->addr2.lno = lno;
1263 			} else {
1264 				ex_badaddr(sp, NULL, A_EOF, NUM_OK);
1265 				goto err;
1266 			}
1267 		}
1268 		/* FALLTHROUGH */
1269 	case 1:
1270 		if (ecp->addr1.lno == 0) {
1271 			if (!F_ISSET(ecp, E_ADDR_ZERO) &&
1272 			    (F_ISSET(sp, SC_EX) ||
1273 			    !F_ISSET(ecp, E_USELASTCMD))) {
1274 				ex_badaddr(sp, ecp->cmd, A_ZERO, NUM_OK);
1275 				goto err;
1276 			}
1277 		} else if (!db_exist(sp, ecp->addr1.lno)) {
1278 			ex_badaddr(sp, NULL, A_EOF, NUM_OK);
1279 			goto err;
1280 		}
1281 		break;
1282 	}
1283 
1284 	/*
1285 	 * If doing a default command and there's nothing left on the line,
1286 	 * vi just moves to the line.  For example, ":3" and ":'a,'b" just
1287 	 * move to line 3 and line 'b, respectively, but ":3|" prints line 3.
1288 	 *
1289 	 * !!!
1290 	 * In addition, IF THE LINE CHANGES, move to the first nonblank of
1291 	 * the line.
1292 	 *
1293 	 * !!!
1294 	 * This is done before the absolute mark gets set; historically,
1295 	 * "/a/,/b/" did NOT set vi's absolute mark, but "/a/,/b/d" did.
1296 	 */
1297 	if ((F_ISSET(sp, SC_VI) || F_ISSET(ecp, E_NOPRDEF)) &&
1298 	    F_ISSET(ecp, E_USELASTCMD) && vi_address == 0) {
1299 		switch (ecp->addrcnt) {
1300 		case 2:
1301 			if (sp->lno !=
1302 			    (ecp->addr2.lno ? ecp->addr2.lno : 1)) {
1303 				sp->lno =
1304 				    ecp->addr2.lno ? ecp->addr2.lno : 1;
1305 				sp->cno = 0;
1306 				(void)nonblank(sp, sp->lno, &sp->cno);
1307 			}
1308 			break;
1309 		case 1:
1310 			if (sp->lno !=
1311 			    (ecp->addr1.lno ? ecp->addr1.lno : 1)) {
1312 				sp->lno =
1313 				    ecp->addr1.lno ? ecp->addr1.lno : 1;
1314 				sp->cno = 0;
1315 				(void)nonblank(sp, sp->lno, &sp->cno);
1316 			}
1317 			break;
1318 		}
1319 		ecp->cp = ecp->save_cmd;
1320 		ecp->clen = ecp->save_cmdlen;
1321 		goto loop;
1322 	}
1323 
1324 	/*
1325 	 * Set the absolute mark -- we have to set it for vi here, in case
1326 	 * it's a compound command, e.g. ":5p|6" should set the absolute
1327 	 * mark for vi.
1328 	 */
1329 	if (F_ISSET(ecp, E_ABSMARK)) {
1330 		cur.lno = sp->lno;
1331 		cur.cno = sp->cno;
1332 		F_CLR(ecp, E_ABSMARK);
1333 		if (mark_set(sp, ABSMARK1, &cur, 1))
1334 			goto err;
1335 	}
1336 
1337 #if defined(DEBUG) && defined(COMLOG)
1338 	ex_comlog(sp, ecp);
1339 #endif
1340 	/* Increment the command count if not called from vi. */
1341 	if (F_ISSET(sp, SC_EX))
1342 		++sp->ccnt;
1343 
1344 	/*
1345 	 * If file state available, and not doing a global command,
1346 	 * log the start of an action.
1347 	 */
1348 	if (sp->ep != NULL && !F_ISSET(sp, SC_EX_GLOBAL))
1349 		(void)log_cursor(sp);
1350 
1351 	/*
1352 	 * !!!
1353 	 * There are two special commands for the purposes of this code: the
1354 	 * default command (<carriage-return>) or the scrolling commands (^D
1355 	 * and <EOF>) as the first non-<blank> characters  in the line.
1356 	 *
1357 	 * If this is the first command in the command line, we received the
1358 	 * command from the ex command loop and we're talking to a tty, and
1359 	 * and there's nothing else on the command line, and it's one of the
1360 	 * special commands, we move back up to the previous line, and erase
1361 	 * the prompt character with the output.  Since ex runs in canonical
1362 	 * mode, we don't have to do anything else, a <newline> has already
1363 	 * been echoed by the tty driver.  It's OK if vi calls us -- we won't
1364 	 * be in ex mode so we'll do nothing.
1365 	 */
1366 	if (F_ISSET(ecp, E_NRSEP)) {
1367 		if (sp->ep != NULL &&
1368 		    F_ISSET(sp, SC_EX) && !F_ISSET(gp, G_SCRIPTED) &&
1369 		    (F_ISSET(ecp, E_USELASTCMD) || ecp->cmd == &cmds[C_SCROLL]))
1370 			gp->scr_ex_adjust(sp, EX_TERM_SCROLL);
1371 		F_CLR(ecp, E_NRSEP);
1372 	}
1373 
1374 	/*
1375 	 * Call the underlying function for the ex command.
1376 	 *
1377 	 * XXX
1378 	 * Interrupts behave like errors, for now.
1379 	 */
1380 	if (ecp->cmd->fn(sp, ecp) || INTERRUPTED(sp)) {
1381 		if (F_ISSET(gp, G_SCRIPTED))
1382 			F_SET(sp, SC_EXIT_FORCE);
1383 		goto err;
1384 	}
1385 
1386 #ifdef DEBUG
1387 	/* Make sure no function left global temporary space locked. */
1388 	if (F_ISSET(gp, G_TMP_INUSE)) {
1389 		F_CLR(gp, G_TMP_INUSE);
1390 		msgq_wstr(sp, M_ERR, ecp->cmd->name,
1391 		    "087|%s: temporary buffer not released");
1392 	}
1393 #endif
1394 	/*
1395 	 * Ex displayed the number of lines modified immediately after each
1396 	 * command, so the command "1,10d|1,10d" would display:
1397 	 *
1398 	 *	10 lines deleted
1399 	 *	10 lines deleted
1400 	 *	<autoprint line>
1401 	 *
1402 	 * Executing ex commands from vi only reported the final modified
1403 	 * lines message -- that's wrong enough that we don't match it.
1404 	 */
1405 	if (F_ISSET(sp, SC_EX))
1406 		mod_rpt(sp);
1407 
1408 	/*
1409 	 * Integrate any offset parsed by the underlying command, and make
1410 	 * sure the referenced line exists.
1411 	 *
1412 	 * XXX
1413 	 * May not match historic practice (which I've never been able to
1414 	 * completely figure out.)  For example, the '=' command from vi
1415 	 * mode often got the offset wrong, and complained it was too large,
1416 	 * but didn't seem to have a problem with the cursor.  If anyone
1417 	 * complains, ask them how it's supposed to work, they might know.
1418 	 */
1419 	if (sp->ep != NULL && ecp->flagoff) {
1420 		if (ecp->flagoff < 0) {
1421 			if (sp->lno <= -ecp->flagoff) {
1422 				msgq(sp, M_ERR,
1423 				    "088|Flag offset to before line 1");
1424 				goto err;
1425 			}
1426 		} else {
1427 			if (!NPFITS(MAX_REC_NUMBER, sp->lno, ecp->flagoff)) {
1428 				ex_badaddr(sp, NULL, A_NOTSET, NUM_OVER);
1429 				goto err;
1430 			}
1431 			if (!db_exist(sp, sp->lno + ecp->flagoff)) {
1432 				msgq(sp, M_ERR,
1433 				    "089|Flag offset past end-of-file");
1434 				goto err;
1435 			}
1436 		}
1437 		sp->lno += ecp->flagoff;
1438 	}
1439 
1440 	/*
1441 	 * If the command executed successfully, we may want to display a line
1442 	 * based on the autoprint option or an explicit print flag.  (Make sure
1443 	 * that there's a line to display.)  Also, the autoprint edit option is
1444 	 * turned off for the duration of global commands.
1445 	 */
1446 	if (F_ISSET(sp, SC_EX) && sp->ep != NULL && sp->lno != 0) {
1447 		/*
1448 		 * The print commands have already handled the `print' flags.
1449 		 * If so, clear them.
1450 		 */
1451 		if (FL_ISSET(ecp->iflags, E_CLRFLAG))
1452 			FL_CLR(ecp->iflags, E_C_HASH | E_C_LIST | E_C_PRINT);
1453 
1454 		/* If hash set only because of the number option, discard it. */
1455 		if (F_ISSET(ecp, E_OPTNUM))
1456 			FL_CLR(ecp->iflags, E_C_HASH);
1457 
1458 		/*
1459 		 * If there was an explicit flag to display the new cursor line,
1460 		 * or autoprint is set and a change was made, display the line.
1461 		 * If any print flags were set use them, else default to print.
1462 		 */
1463 		LF_INIT(FL_ISSET(ecp->iflags, E_C_HASH | E_C_LIST | E_C_PRINT));
1464 		if (!LF_ISSET(E_C_HASH | E_C_LIST | E_C_PRINT | E_NOAUTO) &&
1465 		    !F_ISSET(sp, SC_EX_GLOBAL) &&
1466 		    O_ISSET(sp, O_AUTOPRINT) && F_ISSET(ecp, E_AUTOPRINT))
1467 			LF_INIT(E_C_PRINT);
1468 
1469 		if (LF_ISSET(E_C_HASH | E_C_LIST | E_C_PRINT)) {
1470 			cur.lno = sp->lno;
1471 			cur.cno = 0;
1472 			(void)ex_print(sp, ecp, &cur, &cur, flags);
1473 		}
1474 	}
1475 
1476 	/*
1477 	 * If the command had an associated "+cmd", it has to be executed
1478 	 * before we finish executing any more of this ex command.  For
1479 	 * example, consider a .exrc file that contains the following lines:
1480 	 *
1481 	 *	:set all
1482 	 *	:edit +25 file.c|s/abc/ABC/|1
1483 	 *	:3,5 print
1484 	 *
1485 	 * This can happen more than once -- the historic vi simply hung or
1486 	 * dropped core, of course.  Prepend the + command back into the
1487 	 * current command and continue.  We may have to add an additional
1488 	 * <literal next> character.  We know that it will fit because we
1489 	 * discarded at least one space and the + character.
1490 	 */
1491 	if (arg1_len != 0) {
1492 		/*
1493 		 * If the last character of the + command was a <literal next>
1494 		 * character, it would be treated differently because of the
1495 		 * append.  Quote it, if necessary.
1496 		 */
1497 		if (IS_ESCAPE(sp, ecp, arg1[arg1_len - 1])) {
1498 			*--ecp->save_cmd = CH_LITERAL;
1499 			++ecp->save_cmdlen;
1500 		}
1501 
1502 		ecp->save_cmd -= arg1_len;
1503 		ecp->save_cmdlen += arg1_len;
1504 		MEMMOVE(ecp->save_cmd, arg1, arg1_len);
1505 
1506 		/*
1507 		 * Any commands executed from a +cmd are executed starting at
1508 		 * the first column of the last line of the file -- NOT the
1509 		 * first nonblank.)  The main file startup code doesn't know
1510 		 * that a +cmd was set, however, so it may have put us at the
1511 		 * top of the file.  (Note, this is safe because we must have
1512 		 * switched files to get here.)
1513 		 */
1514 		F_SET(ecp, E_MOVETOEND);
1515 	}
1516 
1517 	/* Update the current command. */
1518 	ecp->cp = ecp->save_cmd;
1519 	ecp->clen = ecp->save_cmdlen;
1520 
1521 	/*
1522 	 * !!!
1523 	 * If we've changed screens or underlying files, any pending global or
1524 	 * v command, or @ buffer that has associated addresses, has to be
1525 	 * discarded.  This is historic practice for globals, and necessary for
1526 	 * @ buffers that had associated addresses.
1527 	 *
1528 	 * Otherwise, if we've changed underlying files, it's not a problem,
1529 	 * we continue with the rest of the ex command(s), operating on the
1530 	 * new file.  However, if we switch screens (either by exiting or by
1531 	 * an explicit command), we have no way of knowing where to put output
1532 	 * messages, and, since we don't control screens here, we could screw
1533 	 * up the upper layers, (e.g. we could exit/reenter a screen multiple
1534 	 * times).  So, return and continue after we've got a new screen.
1535 	 */
1536 	if (F_ISSET(sp, SC_EXIT | SC_EXIT_FORCE | SC_FSWITCH | SC_SSWITCH)) {
1537 		at_found = gv_found = 0;
1538 		SLIST_FOREACH(ecp, sp->gp->ecq, q)
1539 			switch (ecp->agv_flags) {
1540 			case 0:
1541 			case AGV_AT_NORANGE:
1542 				break;
1543 			case AGV_AT:
1544 				if (!at_found) {
1545 					at_found = 1;
1546 					msgq(sp, M_ERR,
1547 		"090|@ with range running when the file/screen changed");
1548 				}
1549 				break;
1550 			case AGV_GLOBAL:
1551 			case AGV_V:
1552 				if (!gv_found) {
1553 					gv_found = 1;
1554 					msgq(sp, M_ERR,
1555 		"091|Global/v command running when the file/screen changed");
1556 				}
1557 				break;
1558 			default:
1559 				abort();
1560 			}
1561 		if (at_found || gv_found)
1562 			goto discard;
1563 		if (F_ISSET(sp, SC_EXIT | SC_EXIT_FORCE | SC_SSWITCH))
1564 			goto rsuccess;
1565 	}
1566 
1567 	goto loop;
1568 	/* NOTREACHED */
1569 
1570 err:	/*
1571 	 * On command failure, we discard keys and pending commands remaining,
1572 	 * as well as any keys that were mapped and waiting.  The save_cmdlen
1573 	 * test is not necessarily correct.  If we fail early enough we don't
1574 	 * know if the entire string was a single command or not.  Guess, as
1575 	 * it's useful to know if commands other than the current one are being
1576 	 * discarded.
1577 	 */
1578 	if (ecp->save_cmdlen == 0)
1579 		for (; ecp->clen; --ecp->clen) {
1580 			ch = *ecp->cp++;
1581 			if (IS_ESCAPE(sp, ecp, ch) && ecp->clen > 1) {
1582 				--ecp->clen;
1583 				++ecp->cp;
1584 			} else if (ch == '\n' || ch == '|') {
1585 				if (ecp->clen > 1)
1586 					ecp->save_cmdlen = 1;
1587 				break;
1588 			}
1589 		}
1590 	if (ecp->save_cmdlen != 0 || SLIST_FIRST(gp->ecq) != &gp->excmd) {
1591 discard:	msgq(sp, M_BERR,
1592 		    "092|Ex command failed: pending commands discarded");
1593 		ex_discard(sp);
1594 	}
1595 	if (v_event_flush(sp, CH_MAPPED))
1596 		msgq(sp, M_BERR,
1597 		    "093|Ex command failed: mapped keys discarded");
1598 
1599 rfail:	tmp = 1;
1600 	if (0)
1601 rsuccess:	tmp = 0;
1602 
1603 	/* Turn off any file name error information. */
1604 	gp->if_name = NULL;
1605 
1606 	/* Turn off the global bit. */
1607 	F_CLR(sp, SC_EX_GLOBAL);
1608 
1609 	return (tmp);
1610 }
1611 
1612 /*
1613  * ex_range --
1614  *	Get a line range for ex commands, or perform a vi ex address search.
1615  *
1616  * PUBLIC: int ex_range(SCR *, EXCMD *, int *);
1617  */
1618 int
1619 ex_range(SCR *sp, EXCMD *ecp, int *errp)
1620 {
1621 	enum { ADDR_FOUND, ADDR_NEED, ADDR_NONE } addr;
1622 	GS *gp;
1623 	EX_PRIVATE *exp;
1624 	MARK m;
1625 	int isaddr;
1626 
1627 	*errp = 0;
1628 
1629 	/*
1630 	 * Parse comma or semi-colon delimited line specs.
1631 	 *
1632 	 * Semi-colon delimiters update the current address to be the last
1633 	 * address.  For example, the command
1634 	 *
1635 	 *	:3;/pattern/ecp->cp
1636 	 *
1637 	 * will search for pattern from line 3.  In addition, if ecp->cp
1638 	 * is not a valid command, the current line will be left at 3, not
1639 	 * at the original address.
1640 	 *
1641 	 * Extra addresses are discarded, starting with the first.
1642 	 *
1643 	 * !!!
1644 	 * If any addresses are missing, they default to the current line.
1645 	 * This was historically true for both leading and trailing comma
1646 	 * delimited addresses as well as for trailing semicolon delimited
1647 	 * addresses.  For consistency, we make it true for leading semicolon
1648 	 * addresses as well.
1649 	 */
1650 	gp = sp->gp;
1651 	exp = EXP(sp);
1652 	for (addr = ADDR_NONE, ecp->addrcnt = 0; ecp->clen > 0;)
1653 		switch (*ecp->cp) {
1654 		case '%':		/* Entire file. */
1655 			/* Vi ex address searches didn't permit % signs. */
1656 			if (F_ISSET(ecp, E_VISEARCH))
1657 				goto ret;
1658 
1659 			/* It's an error if the file is empty. */
1660 			if (sp->ep == NULL) {
1661 				ex_badaddr(sp, NULL, A_EMPTY, NUM_OK);
1662 				*errp = 1;
1663 				return (0);
1664 			}
1665 			/*
1666 			 * !!!
1667 			 * A percent character addresses all of the lines in
1668 			 * the file.  Historically, it couldn't be followed by
1669 			 * any other address.  We do it as a text substitution
1670 			 * for simplicity.  POSIX 1003.2 is expected to follow
1671 			 * this practice.
1672 			 *
1673 			 * If it's an empty file, the first line is 0, not 1.
1674 			 */
1675 			if (addr == ADDR_FOUND) {
1676 				ex_badaddr(sp, NULL, A_COMBO, NUM_OK);
1677 				*errp = 1;
1678 				return (0);
1679 			}
1680 			if (db_last(sp, &ecp->addr2.lno))
1681 				return (1);
1682 			ecp->addr1.lno = ecp->addr2.lno == 0 ? 0 : 1;
1683 			ecp->addr1.cno = ecp->addr2.cno = 0;
1684 			ecp->addrcnt = 2;
1685 			addr = ADDR_FOUND;
1686 			++ecp->cp;
1687 			--ecp->clen;
1688 			break;
1689 		case ',':	       /* Comma delimiter. */
1690 			/* Vi ex address searches didn't permit commas. */
1691 			if (F_ISSET(ecp, E_VISEARCH))
1692 				goto ret;
1693 			/* FALLTHROUGH */
1694 		case ';':	       /* Semi-colon delimiter. */
1695 			if (sp->ep == NULL) {
1696 				ex_badaddr(sp, NULL, A_EMPTY, NUM_OK);
1697 				*errp = 1;
1698 				return (0);
1699 			}
1700 			if (addr != ADDR_FOUND)
1701 				switch (ecp->addrcnt) {
1702 				case 0:
1703 					ecp->addr1.lno = sp->lno;
1704 					ecp->addr1.cno = sp->cno;
1705 					ecp->addrcnt = 1;
1706 					break;
1707 				case 2:
1708 					ecp->addr1 = ecp->addr2;
1709 					/* FALLTHROUGH */
1710 				case 1:
1711 					ecp->addr2.lno = sp->lno;
1712 					ecp->addr2.cno = sp->cno;
1713 					ecp->addrcnt = 2;
1714 					break;
1715 				}
1716 			if (*ecp->cp == ';')
1717 				switch (ecp->addrcnt) {
1718 				case 0:
1719 					abort();
1720 					/* NOTREACHED */
1721 				case 1:
1722 					sp->lno = ecp->addr1.lno;
1723 					sp->cno = ecp->addr1.cno;
1724 					break;
1725 				case 2:
1726 					sp->lno = ecp->addr2.lno;
1727 					sp->cno = ecp->addr2.cno;
1728 					break;
1729 				}
1730 			addr = ADDR_NEED;
1731 			/* FALLTHROUGH */
1732 		case ' ':		/* Whitespace. */
1733 		case '\t':		/* Whitespace. */
1734 			++ecp->cp;
1735 			--ecp->clen;
1736 			break;
1737 		default:
1738 			/* Get a line specification. */
1739 			if (ex_line(sp, ecp, &m, &isaddr, errp))
1740 				return (1);
1741 			if (*errp)
1742 				return (0);
1743 			if (!isaddr)
1744 				goto ret;
1745 			if (addr == ADDR_FOUND) {
1746 				ex_badaddr(sp, NULL, A_COMBO, NUM_OK);
1747 				*errp = 1;
1748 				return (0);
1749 			}
1750 			switch (ecp->addrcnt) {
1751 			case 0:
1752 				ecp->addr1 = m;
1753 				ecp->addrcnt = 1;
1754 				break;
1755 			case 1:
1756 				ecp->addr2 = m;
1757 				ecp->addrcnt = 2;
1758 				break;
1759 			case 2:
1760 				ecp->addr1 = ecp->addr2;
1761 				ecp->addr2 = m;
1762 				break;
1763 			}
1764 			addr = ADDR_FOUND;
1765 			break;
1766 		}
1767 
1768 	/*
1769 	 * !!!
1770 	 * Vi ex address searches are indifferent to order or trailing
1771 	 * semi-colons.
1772 	 */
1773 ret:	if (F_ISSET(ecp, E_VISEARCH))
1774 		return (0);
1775 
1776 	if (addr == ADDR_NEED)
1777 		switch (ecp->addrcnt) {
1778 		case 0:
1779 			ecp->addr1.lno = sp->lno;
1780 			ecp->addr1.cno = sp->cno;
1781 			ecp->addrcnt = 1;
1782 			break;
1783 		case 2:
1784 			ecp->addr1 = ecp->addr2;
1785 			/* FALLTHROUGH */
1786 		case 1:
1787 			ecp->addr2.lno = sp->lno;
1788 			ecp->addr2.cno = sp->cno;
1789 			ecp->addrcnt = 2;
1790 			break;
1791 		}
1792 
1793 	if (ecp->addrcnt == 2 && ecp->addr2.lno < ecp->addr1.lno) {
1794 		msgq(sp, M_ERR,
1795 		    "094|The second address is smaller than the first");
1796 		*errp = 1;
1797 	}
1798 	return (0);
1799 }
1800 
1801 /*
1802  * ex_line --
1803  *	Get a single line address specifier.
1804  *
1805  * The way the "previous context" mark worked was that any "non-relative"
1806  * motion set it.  While ex/vi wasn't totally consistent about this, ANY
1807  * numeric address, search pattern, '$', or mark reference in an address
1808  * was considered non-relative, and set the value.  Which should explain
1809  * why we're hacking marks down here.  The problem was that the mark was
1810  * only set if the command was called, i.e. we have to set a flag and test
1811  * it later.
1812  *
1813  * XXX
1814  * This is probably still not exactly historic practice, although I think
1815  * it's fairly close.
1816  */
1817 static int
1818 ex_line(SCR *sp, EXCMD *ecp, MARK *mp, int *isaddrp, int *errp)
1819 {
1820 	enum nresult nret;
1821 	EX_PRIVATE *exp;
1822 	GS *gp;
1823 	long total, val;
1824 	int isneg;
1825 	int (*sf)(SCR *, MARK *, MARK *, CHAR_T *, size_t, CHAR_T **, u_int);
1826 	CHAR_T *endp;
1827 
1828 	gp = sp->gp;
1829 	exp = EXP(sp);
1830 
1831 	*isaddrp = *errp = 0;
1832 	F_CLR(ecp, E_DELTA);
1833 
1834 	/* No addresses permitted until a file has been read in. */
1835 	if (sp->ep == NULL && STRCHR(L("$0123456789'\\/?.+-^"), *ecp->cp)) {
1836 		ex_badaddr(sp, NULL, A_EMPTY, NUM_OK);
1837 		*errp = 1;
1838 		return (0);
1839 	}
1840 
1841 	switch (*ecp->cp) {
1842 	case '$':				/* Last line in the file. */
1843 		*isaddrp = 1;
1844 		F_SET(ecp, E_ABSMARK);
1845 
1846 		mp->cno = 0;
1847 		if (db_last(sp, &mp->lno))
1848 			return (1);
1849 		++ecp->cp;
1850 		--ecp->clen;
1851 		break;				/* Absolute line number. */
1852 	case '0': case '1': case '2': case '3': case '4':
1853 	case '5': case '6': case '7': case '8': case '9':
1854 		*isaddrp = 1;
1855 		F_SET(ecp, E_ABSMARK);
1856 
1857 		if ((nret = nget_slong(&val, ecp->cp, &endp, 10)) != NUM_OK) {
1858 			ex_badaddr(sp, NULL, A_NOTSET, nret);
1859 			*errp = 1;
1860 			return (0);
1861 		}
1862 		if (!NPFITS(MAX_REC_NUMBER, 0, val)) {
1863 			ex_badaddr(sp, NULL, A_NOTSET, NUM_OVER);
1864 			*errp = 1;
1865 			return (0);
1866 		}
1867 		mp->lno = val;
1868 		mp->cno = 0;
1869 		ecp->clen -= (endp - ecp->cp);
1870 		ecp->cp = endp;
1871 		break;
1872 	case '\'':				/* Use a mark. */
1873 		*isaddrp = 1;
1874 		F_SET(ecp, E_ABSMARK);
1875 
1876 		if (ecp->clen == 1) {
1877 			msgq(sp, M_ERR, "095|No mark name supplied");
1878 			*errp = 1;
1879 			return (0);
1880 		}
1881 		if (mark_get(sp, ecp->cp[1], mp, M_ERR)) {
1882 			*errp = 1;
1883 			return (0);
1884 		}
1885 		ecp->cp += 2;
1886 		ecp->clen -= 2;
1887 		break;
1888 	case '\\':				/* Search: forward/backward. */
1889 		/*
1890 		 * !!!
1891 		 * I can't find any difference between // and \/ or between
1892 		 * ?? and \?.  Mark Horton doesn't remember there being any
1893 		 * difference.  C'est la vie.
1894 		 */
1895 		if (ecp->clen < 2 ||
1896 		    (ecp->cp[1] != '/' && ecp->cp[1] != '?')) {
1897 			msgq(sp, M_ERR, "096|\\ not followed by / or ?");
1898 			*errp = 1;
1899 			return (0);
1900 		}
1901 		++ecp->cp;
1902 		--ecp->clen;
1903 		sf = ecp->cp[0] == '/' ? f_search : b_search;
1904 		goto search;
1905 	case '/':				/* Search forward. */
1906 		sf = f_search;
1907 		goto search;
1908 	case '?':				/* Search backward. */
1909 		sf = b_search;
1910 
1911 search:		mp->lno = sp->lno;
1912 		mp->cno = sp->cno;
1913 		if (sf(sp, mp, mp, ecp->cp, ecp->clen, &endp,
1914 		    SEARCH_MSG | SEARCH_PARSE | SEARCH_SET |
1915 		    (F_ISSET(ecp, E_SEARCH_WMSG) ? SEARCH_WMSG : 0))) {
1916 			*errp = 1;
1917 			return (0);
1918 		}
1919 
1920 		/* Fix up the command pointers. */
1921 		ecp->clen -= (endp - ecp->cp);
1922 		ecp->cp = endp;
1923 
1924 		*isaddrp = 1;
1925 		F_SET(ecp, E_ABSMARK);
1926 		break;
1927 	case '.':				/* Current position. */
1928 		*isaddrp = 1;
1929 		mp->cno = sp->cno;
1930 
1931 		/* If an empty file, then '.' is 0, not 1. */
1932 		if (sp->lno == 1) {
1933 			if (db_last(sp, &mp->lno))
1934 				return (1);
1935 			if (mp->lno != 0)
1936 				mp->lno = 1;
1937 		} else
1938 			mp->lno = sp->lno;
1939 
1940 		/*
1941 		 * !!!
1942 		 * Historically, .<number> was the same as .+<number>, i.e.
1943 		 * the '+' could be omitted.  (This feature is found in ed
1944 		 * as well.)
1945 		 */
1946 		if (ecp->clen > 1 && ISDIGIT(ecp->cp[1]))
1947 			*ecp->cp = '+';
1948 		else {
1949 			++ecp->cp;
1950 			--ecp->clen;
1951 		}
1952 		break;
1953 	}
1954 
1955 	/* Skip trailing <blank>s. */
1956 	for (; ecp->clen > 0 &&
1957 	    cmdskip(ecp->cp[0]); ++ecp->cp, --ecp->clen);
1958 
1959 	/*
1960 	 * Evaluate any offset.  If no address yet found, the offset
1961 	 * is relative to ".".
1962 	 */
1963 	total = 0;
1964 	if (ecp->clen != 0 && (ISDIGIT(ecp->cp[0]) ||
1965 	    ecp->cp[0] == '+' || ecp->cp[0] == '-' ||
1966 	    ecp->cp[0] == '^')) {
1967 		if (!*isaddrp) {
1968 			*isaddrp = 1;
1969 			mp->lno = sp->lno;
1970 			mp->cno = sp->cno;
1971 		}
1972 		/*
1973 		 * Evaluate an offset, defined as:
1974 		 *
1975 		 *		[+-^<blank>]*[<blank>]*[0-9]*
1976 		 *
1977 		 * The rough translation is any number of signs, optionally
1978 		 * followed by numbers, or a number by itself, all <blank>
1979 		 * separated.
1980 		 *
1981 		 * !!!
1982 		 * All address offsets were additive, e.g. "2 2 3p" was the
1983 		 * same as "7p", or, "/ZZZ/ 2" was the same as "/ZZZ/+2".
1984 		 * Note, however, "2 /ZZZ/" was an error.  It was also legal
1985 		 * to insert signs without numbers, so "3 - 2" was legal, and
1986 		 * equal to 4.
1987 		 *
1988 		 * !!!
1989 		 * Offsets were historically permitted for any line address,
1990 		 * e.g. the command "1,2 copy 2 2 2 2" copied lines 1,2 after
1991 		 * line 8.
1992 		 *
1993 		 * !!!
1994 		 * Offsets were historically permitted for search commands,
1995 		 * and handled as addresses: "/pattern/2 2 2" was legal, and
1996 		 * referenced the 6th line after pattern.
1997 		 */
1998 		F_SET(ecp, E_DELTA);
1999 		for (;;) {
2000 			for (; ecp->clen > 0 && cmdskip(ecp->cp[0]);
2001 			    ++ecp->cp, --ecp->clen);
2002 			if (ecp->clen == 0 || (!ISDIGIT(ecp->cp[0]) &&
2003 			    ecp->cp[0] != '+' && ecp->cp[0] != '-' &&
2004 			    ecp->cp[0] != '^'))
2005 				break;
2006 			if (!ISDIGIT(ecp->cp[0]) &&
2007 			    !ISDIGIT(ecp->cp[1])) {
2008 				total += ecp->cp[0] == '+' ? 1 : -1;
2009 				--ecp->clen;
2010 				++ecp->cp;
2011 			} else {
2012 				if (ecp->cp[0] == '-' ||
2013 				    ecp->cp[0] == '^') {
2014 					++ecp->cp;
2015 					--ecp->clen;
2016 					isneg = 1;
2017 				} else
2018 					isneg = 0;
2019 
2020 				/* Get a signed long, add it to the total. */
2021 				if ((nret = nget_slong(&val,
2022 				    ecp->cp, &endp, 10)) != NUM_OK ||
2023 				    (nret = NADD_SLONG(sp,
2024 				    total, val)) != NUM_OK) {
2025 					ex_badaddr(sp, NULL, A_NOTSET, nret);
2026 					*errp = 1;
2027 					return (0);
2028 				}
2029 				total += isneg ? -val : val;
2030 				ecp->clen -= (endp - ecp->cp);
2031 				ecp->cp = endp;
2032 			}
2033 		}
2034 	}
2035 
2036 	/*
2037 	 * Any value less than 0 is an error.  Make sure that the new value
2038 	 * will fit into a recno_t.
2039 	 */
2040 	if (*isaddrp && total != 0) {
2041 		if (total < 0) {
2042 			if (-total > mp->lno) {
2043 				msgq(sp, M_ERR,
2044 			    "097|Reference to a line number less than 0");
2045 				*errp = 1;
2046 				return (0);
2047 			}
2048 		} else
2049 			if (!NPFITS(MAX_REC_NUMBER, mp->lno, total)) {
2050 				ex_badaddr(sp, NULL, A_NOTSET, NUM_OVER);
2051 				*errp = 1;
2052 				return (0);
2053 			}
2054 		mp->lno += total;
2055 	}
2056 	return (0);
2057 }
2058 
2059 
2060 /*
2061  * ex_load --
2062  *	Load up the next command, which may be an @ buffer or global command.
2063  */
2064 static int
2065 ex_load(SCR *sp)
2066 {
2067 	GS *gp;
2068 	EXCMD *ecp;
2069 	RANGE *rp;
2070 
2071 	F_CLR(sp, SC_EX_GLOBAL);
2072 
2073 	/*
2074 	 * Lose any exhausted commands.  We know that the first command
2075 	 * can't be an AGV command, which makes things a bit easier.
2076 	 */
2077 	for (gp = sp->gp;;) {
2078 		ecp = SLIST_FIRST(gp->ecq);
2079 
2080 		/* Discard the allocated source name as requested. */
2081 		if (F_ISSET(ecp, E_NAMEDISCARD))
2082 			free(ecp->if_name);
2083 
2084 		/*
2085 		 * If we're back to the original structure, leave it around,
2086 		 * since we've returned to the beginning of the command stack.
2087 		 */
2088 		if (ecp == &gp->excmd) {
2089 			ecp->if_name = NULL;
2090 			return (0);
2091 		}
2092 
2093 		/*
2094 		 * ecp->clen will be 0 for the first discarded command, but
2095 		 * may not be 0 for subsequent ones, e.g. if the original
2096 		 * command was ":g/xx/@a|s/b/c/", then when we discard the
2097 		 * command pushed on the stack by the @a, we have to resume
2098 		 * the global command which included the substitute command.
2099 		 */
2100 		if (ecp->clen != 0)
2101 			return (0);
2102 
2103 		/*
2104 		 * If it's an @, global or v command, we may need to continue
2105 		 * the command on a different line.
2106 		 */
2107 		if (FL_ISSET(ecp->agv_flags, AGV_ALL)) {
2108 			/* Discard any exhausted ranges. */
2109 			while ((rp = TAILQ_FIRST(ecp->rq)) != NULL)
2110 				if (rp->start > rp->stop) {
2111 					TAILQ_REMOVE(ecp->rq, rp, q);
2112 					free(rp);
2113 				} else
2114 					break;
2115 
2116 			/* If there's another range, continue with it. */
2117 			if (rp != NULL)
2118 				break;
2119 
2120 			/* If it's a global/v command, fix up the last line. */
2121 			if (FL_ISSET(ecp->agv_flags,
2122 			    AGV_GLOBAL | AGV_V) && ecp->range_lno != OOBLNO) {
2123 				if (db_exist(sp, ecp->range_lno))
2124 					sp->lno = ecp->range_lno;
2125 				else {
2126 					if (db_last(sp, &sp->lno))
2127 						return (1);
2128 					if (sp->lno == 0)
2129 						sp->lno = 1;
2130 				}
2131 			}
2132 			free(ecp->o_cp);
2133 		}
2134 
2135 		/* Discard the EXCMD. */
2136 		SLIST_REMOVE_HEAD(gp->ecq, q);
2137 		free(ecp);
2138 	}
2139 
2140 	/*
2141 	 * We only get here if it's an active @, global or v command.  Set
2142 	 * the current line number, and get a new copy of the command for
2143 	 * the parser.  Note, the original pointer almost certainly moved,
2144 	 * so we have play games.
2145 	 */
2146 	ecp->cp = ecp->o_cp;
2147 	MEMCPY(ecp->cp, ecp->cp + ecp->o_clen, ecp->o_clen);
2148 	ecp->clen = ecp->o_clen;
2149 	ecp->range_lno = sp->lno = rp->start++;
2150 
2151 	if (FL_ISSET(ecp->agv_flags, AGV_GLOBAL | AGV_V))
2152 		F_SET(sp, SC_EX_GLOBAL);
2153 	return (0);
2154 }
2155 
2156 /*
2157  * ex_discard --
2158  *	Discard any pending ex commands.
2159  */
2160 static int
2161 ex_discard(SCR *sp)
2162 {
2163 	GS *gp;
2164 	EXCMD *ecp;
2165 	RANGE *rp;
2166 
2167 	/*
2168 	 * We know the first command can't be an AGV command, so we don't
2169 	 * process it specially.  We do, however, nail the command itself.
2170 	 */
2171 	for (gp = sp->gp;;) {
2172 		ecp = SLIST_FIRST(gp->ecq);
2173 		if (F_ISSET(ecp, E_NAMEDISCARD))
2174 			free(ecp->if_name);
2175 		/* Reset the last command without dropping it. */
2176 		if (ecp == &gp->excmd)
2177 			break;
2178 		if (FL_ISSET(ecp->agv_flags, AGV_ALL)) {
2179 			while ((rp = TAILQ_FIRST(ecp->rq)) != NULL) {
2180 				TAILQ_REMOVE(ecp->rq, rp, q);
2181 				free(rp);
2182 			}
2183 			free(ecp->o_cp);
2184 		}
2185 		SLIST_REMOVE_HEAD(gp->ecq, q);
2186 		free(ecp);
2187 	}
2188 
2189 	ecp->if_name = NULL;
2190 	ecp->clen = 0;
2191 	return (0);
2192 }
2193 
2194 /*
2195  * ex_unknown --
2196  *	Display an unknown command name.
2197  */
2198 static void
2199 ex_unknown(SCR *sp, CHAR_T *cmd, size_t len)
2200 {
2201 	size_t blen;
2202 	CHAR_T *bp;
2203 
2204 	GET_SPACE_GOTOW(sp, bp, blen, len + 1);
2205 	bp[len] = '\0';
2206 	MEMCPY(bp, cmd, len);
2207 	msgq_wstr(sp, M_ERR, bp, "098|The %s command is unknown");
2208 	FREE_SPACEW(sp, bp, blen);
2209 
2210 alloc_err:
2211 	return;
2212 }
2213 
2214 /*
2215  * ex_is_abbrev -
2216  *	The vi text input routine needs to know if ex thinks this is an
2217  *	[un]abbreviate command, so it can turn off abbreviations.  See
2218  *	the usual ranting in the vi/v_txt_ev.c:txt_abbrev() routine.
2219  *
2220  * PUBLIC: int ex_is_abbrev(CHAR_T *, size_t);
2221  */
2222 int
2223 ex_is_abbrev(CHAR_T *name, size_t len)
2224 {
2225 	EXCMDLIST const *cp;
2226 
2227 	return ((cp = ex_comm_search(name, len)) != NULL &&
2228 	    (cp == &cmds[C_ABBR] || cp == &cmds[C_UNABBREVIATE]));
2229 }
2230 
2231 /*
2232  * ex_is_unmap -
2233  *	The vi text input routine needs to know if ex thinks this is an
2234  *	unmap command, so it can turn off input mapping.  See the usual
2235  *	ranting in the vi/v_txt_ev.c:txt_unmap() routine.
2236  *
2237  * PUBLIC: int ex_is_unmap(CHAR_T *, size_t);
2238  */
2239 int
2240 ex_is_unmap(CHAR_T *name, size_t len)
2241 {
2242 	EXCMDLIST const *cp;
2243 
2244 	/*
2245 	 * The command the vi input routines are really interested in
2246 	 * is "unmap!", not just unmap.
2247 	 */
2248 	if (name[len - 1] != '!')
2249 		return (0);
2250 	--len;
2251 	return ((cp = ex_comm_search(name, len)) != NULL &&
2252 	    cp == &cmds[C_UNMAP]);
2253 }
2254 
2255 /*
2256  * ex_comm_search --
2257  *	Search for a command name.
2258  */
2259 static EXCMDLIST const *
2260 ex_comm_search(CHAR_T *name, size_t len)
2261 {
2262 	EXCMDLIST const *cp;
2263 
2264 	for (cp = cmds; cp->name != NULL; ++cp) {
2265 		if (cp->name[0] > name[0])
2266 			return (NULL);
2267 		if (cp->name[0] != name[0])
2268 			continue;
2269 		if (STRLEN(cp->name) >= len &&
2270 			!MEMCMP(name, cp->name, len))
2271 			return (cp);
2272 	}
2273 	return (NULL);
2274 }
2275 
2276 /*
2277  * ex_badaddr --
2278  *	Display a bad address message.
2279  *
2280  * PUBLIC: void ex_badaddr
2281  * PUBLIC:   (SCR *, EXCMDLIST const *, enum badaddr, enum nresult);
2282  */
2283 void
2284 ex_badaddr(SCR *sp, const EXCMDLIST *cp, enum badaddr ba, enum nresult nret)
2285 {
2286 	recno_t lno;
2287 
2288 	switch (nret) {
2289 	case NUM_OK:
2290 		break;
2291 	case NUM_ERR:
2292 		msgq(sp, M_SYSERR, NULL);
2293 		return;
2294 	case NUM_OVER:
2295 		msgq(sp, M_ERR, "099|Address value overflow");
2296 		return;
2297 	case NUM_UNDER:
2298 		msgq(sp, M_ERR, "100|Address value underflow");
2299 		return;
2300 	}
2301 
2302 	/*
2303 	 * When encountering an address error, tell the user if there's no
2304 	 * underlying file, that's the real problem.
2305 	 */
2306 	if (sp->ep == NULL) {
2307 		ex_wemsg(sp, cp ? cp->name : NULL, EXM_NOFILEYET);
2308 		return;
2309 	}
2310 
2311 	switch (ba) {
2312 	case A_COMBO:
2313 		msgq(sp, M_ERR, "101|Illegal address combination");
2314 		break;
2315 	case A_EOF:
2316 		if (db_last(sp, &lno))
2317 			return;
2318 		if (lno != 0) {
2319 			msgq(sp, M_ERR,
2320 			    "102|Illegal address: only %lu lines in the file",
2321 			    (u_long)lno);
2322 			break;
2323 		}
2324 		/* FALLTHROUGH */
2325 	case A_EMPTY:
2326 		msgq(sp, M_ERR, "103|Illegal address: the file is empty");
2327 		break;
2328 	case A_NOTSET:
2329 		abort();
2330 		/* NOTREACHED */
2331 	case A_ZERO:
2332 		msgq_wstr(sp, M_ERR, cp->name,
2333 		    "104|The %s command doesn't permit an address of 0");
2334 		break;
2335 	}
2336 	return;
2337 }
2338 
2339 #if defined(DEBUG) && defined(COMLOG)
2340 /*
2341  * ex_comlog --
2342  *	Log ex commands.
2343  */
2344 static void
2345 ex_comlog(sp, ecp)
2346 	SCR *sp;
2347 	EXCMD *ecp;
2348 {
2349 	TRACE(sp, "ecmd: "WS, ecp->cmd->name);
2350 	if (ecp->addrcnt > 0) {
2351 		TRACE(sp, " a1 %d", ecp->addr1.lno);
2352 		if (ecp->addrcnt > 1)
2353 			TRACE(sp, " a2: %d", ecp->addr2.lno);
2354 	}
2355 	if (ecp->lineno)
2356 		TRACE(sp, " line %d", ecp->lineno);
2357 	if (ecp->flags)
2358 		TRACE(sp, " flags 0x%x", ecp->flags);
2359 	if (FL_ISSET(ecp->iflags, E_C_BUFFER))
2360 		TRACE(sp, " buffer "WC, ecp->buffer);
2361 	if (ecp->argc) {
2362 		int cnt;
2363 		for (cnt = 0; cnt < ecp->argc; ++cnt)
2364 			TRACE(sp, " arg %d: {"WS"}", cnt, ecp->argv[cnt]->bp);
2365 	}
2366 	TRACE(sp, "\n");
2367 }
2368 #endif
2369