xref: /src/contrib/nvi/ex/ex.c (revision 16cef5f7a65588def71db4fdfa961f959847e3b6)
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 			if (*spp) {
158 				F_CLR(*spp, SC_SCR_VI);
159 				F_SET(*spp, SC_SCR_EX);
160 			}
161 			return (screen_end(sp));
162 		}
163 	}
164 	return (0);
165 }
166 
167 /*
168  * ex_cmd --
169  *	The guts of the ex parser: parse and execute a string containing
170  *	ex commands.
171  *
172  * !!!
173  * This code MODIFIES the string that gets passed in, to delete quoting
174  * characters, etc.  The string cannot be readonly/text space, nor should
175  * you expect to use it again after ex_cmd() returns.
176  *
177  * !!!
178  * For the fun of it, if you want to see if a vi clone got the ex argument
179  * parsing right, try:
180  *
181  *	echo 'foo|bar' > file1; echo 'foo/bar' > file2;
182  *	vi
183  *	:edit +1|s/|/PIPE/|w file1| e file2|1 | s/\//SLASH/|wq
184  *
185  * or:	vi
186  *	:set|file|append|set|file
187  *
188  * For extra credit, try them in a startup .exrc file.
189  *
190  * PUBLIC: int ex_cmd(SCR *);
191  */
192 int
193 ex_cmd(SCR *sp)
194 {
195 	enum nresult nret;
196 	EX_PRIVATE *exp;
197 	EXCMD *ecp;
198 	GS *gp;
199 	MARK cur;
200 	recno_t lno;
201 	size_t arg1_len, discard, len;
202 	u_int32_t flags;
203 	long ltmp;
204 	int at_found, gv_found;
205 	int cnt, delim, isaddr, namelen;
206 	int newscreen, notempty, tmp, vi_address;
207 	CHAR_T *arg1, *s, *p, *t;
208 	CHAR_T ch = '\0';
209 	CHAR_T *n;
210 	char *np;
211 
212 	gp = sp->gp;
213 	exp = EXP(sp);
214 
215 	/*
216 	 * We always start running the command on the top of the stack.
217 	 * This means that *everything* must be resolved when we leave
218 	 * this function for any reason.
219 	 */
220 loop:	ecp = SLIST_FIRST(gp->ecq);
221 
222 	/* If we're reading a command from a file, set up error information. */
223 	if (ecp->if_name != NULL) {
224 		gp->if_lno = ecp->if_lno;
225 		gp->if_name = ecp->if_name;
226 	}
227 
228 	/*
229 	 * If a move to the end of the file is scheduled for this command,
230 	 * do it now.
231 	 */
232 	if (F_ISSET(ecp, E_MOVETOEND)) {
233 		if (db_last(sp, &sp->lno))
234 			goto rfail;
235 		sp->cno = 0;
236 		F_CLR(ecp, E_MOVETOEND);
237 	}
238 
239 	/* If we found a newline, increment the count now. */
240 	if (F_ISSET(ecp, E_NEWLINE)) {
241 		++gp->if_lno;
242 		++ecp->if_lno;
243 		F_CLR(ecp, E_NEWLINE);
244 	}
245 
246 	/* (Re)initialize the EXCMD structure, preserving some flags. */
247 	CLEAR_EX_CMD(ecp);
248 
249 	/* Initialize the argument structures. */
250 	if (argv_init(sp, ecp))
251 		goto err;
252 
253 	/* Initialize +cmd, saved command information. */
254 	arg1 = NULL;
255 	ecp->save_cmdlen = 0;
256 
257 	/* Skip <blank>s, empty lines.  */
258 	for (notempty = 0; ecp->clen > 0; ++ecp->cp, --ecp->clen)
259 		if ((ch = *ecp->cp) == '\n') {
260 			++gp->if_lno;
261 			++ecp->if_lno;
262 		} else if (cmdskip(ch))
263 			notempty = 1;
264 		else
265 			break;
266 
267 	/*
268 	 * !!!
269 	 * Permit extra colons at the start of the line.  Historically,
270 	 * ex/vi allowed a single extra one.  It's simpler not to count.
271 	 * The stripping is done here because, historically, any command
272 	 * could have preceding colons, e.g. ":g/pattern/:p" worked.
273 	 */
274 	if (ecp->clen != 0 && ch == ':') {
275 		notempty = 1;
276 		while (--ecp->clen > 0 && (ch = *++ecp->cp) == ':');
277 	}
278 
279 	/*
280 	 * Command lines that start with a double-quote are comments.
281 	 *
282 	 * !!!
283 	 * Historically, there was no escape or delimiter for a comment, e.g.
284 	 * :"foo|set was a single comment and nothing was output.  Since nvi
285 	 * permits users to escape <newline> characters into command lines, we
286 	 * have to check for that case.
287 	 */
288 	if (ecp->clen != 0 && ch == '"') {
289 		while (--ecp->clen > 0 && *++ecp->cp != '\n');
290 		if (*ecp->cp == '\n') {
291 			F_SET(ecp, E_NEWLINE);
292 			++ecp->cp;
293 			--ecp->clen;
294 		}
295 		goto loop;
296 	}
297 
298 	/* Skip whitespace. */
299 	for (; ecp->clen > 0; ++ecp->cp, --ecp->clen) {
300 		ch = *ecp->cp;
301 		if (!cmdskip(ch))
302 			break;
303 	}
304 
305 	/*
306 	 * The last point at which an empty line can mean do nothing.
307 	 *
308 	 * !!!
309 	 * Historically, in ex mode, lines containing only <blank> characters
310 	 * were the same as a single <carriage-return>, i.e. a default command.
311 	 * In vi mode, they were ignored.  In .exrc files this was a serious
312 	 * annoyance, as vi kept trying to treat them as print commands.  We
313 	 * ignore backward compatibility in this case, discarding lines that
314 	 * contain only <blank> characters from .exrc files.
315 	 *
316 	 * !!!
317 	 * This is where you end up when you're done a command, i.e. clen has
318 	 * gone to zero.  Continue if there are more commands to run.
319 	 */
320 	if (ecp->clen == 0 &&
321 	    (!notempty || F_ISSET(sp, SC_VI) || F_ISSET(ecp, E_BLIGNORE))) {
322 		if (ex_load(sp))
323 			goto rfail;
324 		ecp = SLIST_FIRST(gp->ecq);
325 		if (ecp->clen == 0)
326 			goto rsuccess;
327 		goto loop;
328 	}
329 
330 	/*
331 	 * Check to see if this is a command for which we may want to move
332 	 * the cursor back up to the previous line.  (The command :1<CR>
333 	 * wants a <newline> separator, but the command :<CR> wants to erase
334 	 * the command line.)  If the line is empty except for <blank>s,
335 	 * <carriage-return> or <eof>, we'll probably want to move up.  I
336 	 * don't think there's any way to get <blank> characters *after* the
337 	 * command character, but this is the ex parser, and I've been wrong
338 	 * before.
339 	 */
340 	if (F_ISSET(ecp, E_NRSEP) &&
341 	    ecp->clen != 0 && (ecp->clen != 1 || ecp->cp[0] != '\004'))
342 		F_CLR(ecp, E_NRSEP);
343 
344 	/* Parse command addresses. */
345 	if (ex_range(sp, ecp, &tmp))
346 		goto rfail;
347 	if (tmp)
348 		goto err;
349 
350 	/*
351 	 * Skip <blank>s and any more colons (the command :3,5:print
352 	 * worked, historically).
353 	 */
354 	for (; ecp->clen > 0; ++ecp->cp, --ecp->clen) {
355 		ch = *ecp->cp;
356 		if (!cmdskip(ch) && ch != ':')
357 			break;
358 	}
359 
360 	/*
361 	 * If no command, ex does the last specified of p, l, or #, and vi
362 	 * moves to the line.  Otherwise, determine the length of the command
363 	 * name by looking for the first non-alphabetic character.  (There
364 	 * are a few non-alphabetic characters in command names, but they're
365 	 * all single character commands.)  This isn't a great test, because
366 	 * it means that, for the command ":e +cut.c file", we'll report that
367 	 * the command "cut" wasn't known.  However, it makes ":e+35 file" work
368 	 * correctly.
369 	 *
370 	 * !!!
371 	 * Historically, lines with multiple adjacent (or <blank> separated)
372 	 * command separators were very strange.  For example, the command
373 	 * |||<carriage-return>, when the cursor was on line 1, displayed
374 	 * lines 2, 3 and 5 of the file.  In addition, the command "   |  "
375 	 * would only display the line after the next line, instead of the
376 	 * next two lines.  No ideas why.  It worked reasonably when executed
377 	 * from vi mode, and displayed lines 2, 3, and 4, so we do a default
378 	 * command for each separator.
379 	 */
380 #define	SINGLE_CHAR_COMMANDS	L("\004!#&*<=>@~")
381 	newscreen = 0;
382 	if (ecp->clen != 0 && ecp->cp[0] != '|' && ecp->cp[0] != '\n') {
383 		if (STRCHR(SINGLE_CHAR_COMMANDS, *ecp->cp)) {
384 			p = ecp->cp;
385 			++ecp->cp;
386 			--ecp->clen;
387 			namelen = 1;
388 		} else {
389 			for (p = ecp->cp;
390 			    ecp->clen > 0; --ecp->clen, ++ecp->cp)
391 				if (!isazAZ(*ecp->cp))
392 					break;
393 			if ((namelen = ecp->cp - p) == 0) {
394 				msgq(sp, M_ERR, "080|Unknown command name");
395 				goto err;
396 			}
397 		}
398 
399 		/*
400 		 * !!!
401 		 * Historic vi permitted flags to immediately follow any
402 		 * subset of the 'delete' command, but then did not permit
403 		 * further arguments (flag, buffer, count).  Make it work.
404 		 * Permit further arguments for the few shreds of dignity
405 		 * it offers.
406 		 *
407 		 * Adding commands that start with 'd', and match "delete"
408 		 * up to a l, p, +, - or # character can break this code.
409 		 *
410 		 * !!!
411 		 * Capital letters beginning the command names ex, edit,
412 		 * next, previous, tag and visual (in vi mode) indicate the
413 		 * command should happen in a new screen.
414 		 */
415 		switch (p[0]) {
416 		case 'd':
417 			for (s = p,
418 			    n = cmds[C_DELETE].name; *s == *n; ++s, ++n);
419 			if (s[0] == 'l' || s[0] == 'p' || s[0] == '+' ||
420 			    s[0] == '-' || s[0] == '^' || s[0] == '#') {
421 				len = (ecp->cp - p) - (s - p);
422 				ecp->cp -= len;
423 				ecp->clen += len;
424 				ecp->rcmd = cmds[C_DELETE];
425 				ecp->rcmd.syntax = "1bca1";
426 				ecp->cmd = &ecp->rcmd;
427 				goto skip_srch;
428 			}
429 			break;
430 		case 'E': case 'F': case 'N': case 'P': case 'T': case 'V':
431 			newscreen = 1;
432 			p[0] = tolower(p[0]);
433 			break;
434 		}
435 
436 		/*
437 		 * Search the table for the command.
438 		 *
439 		 * !!!
440 		 * Historic vi permitted the mark to immediately follow the
441 		 * 'k' in the 'k' command.  Make it work.
442 		 *
443 		 * !!!
444 		 * Historic vi permitted any flag to follow the s command, e.g.
445 		 * "s/e/E/|s|sgc3p" was legal.  Make the command "sgc" work.
446 		 * Since the following characters all have to be flags, i.e.
447 		 * alphabetics, we can let the s command routine return errors
448 		 * if it was some illegal command string.  This code will break
449 		 * if an "sg" or similar command is ever added.  The substitute
450 		 * code doesn't care if it's a "cgr" flag or a "#lp" flag that
451 		 * follows the 's', but we limit the choices here to "cgr" so
452 		 * that we get unknown command messages for wrong combinations.
453 		 */
454 		if ((ecp->cmd = ex_comm_search(p, namelen)) == NULL)
455 			switch (p[0]) {
456 			case 'k':
457 				if (namelen == 2) {
458 					ecp->cp -= namelen - 1;
459 					ecp->clen += namelen - 1;
460 					ecp->cmd = &cmds[C_K];
461 					break;
462 				}
463 				goto unknown;
464 			case 's':
465 				for (s = p + 1, cnt = namelen; --cnt; ++s)
466 					if (s[0] != 'c' &&
467 					    s[0] != 'g' && s[0] != 'r')
468 						break;
469 				if (cnt == 0) {
470 					ecp->cp -= namelen - 1;
471 					ecp->clen += namelen - 1;
472 					ecp->rcmd = cmds[C_SUBSTITUTE];
473 					ecp->rcmd.fn = ex_subagain;
474 					ecp->cmd = &ecp->rcmd;
475 					break;
476 				}
477 				/* FALLTHROUGH */
478 			default:
479 unknown:			if (newscreen)
480 					p[0] = toupper(p[0]);
481 				ex_unknown(sp, p, namelen);
482 				goto err;
483 			}
484 
485 		/*
486 		 * The visual command has a different syntax when called
487 		 * from ex than when called from a vi colon command.  FMH.
488 		 * Make the change now, before we test for the newscreen
489 		 * semantic, so that we're testing the right one.
490 		 */
491 skip_srch:	if (ecp->cmd == &cmds[C_VISUAL_EX] && F_ISSET(sp, SC_VI))
492 			ecp->cmd = &cmds[C_VISUAL_VI];
493 
494 		/*
495 		 * !!!
496 		 * Historic vi permitted a capital 'P' at the beginning of
497 		 * any command that started with 'p'.  Probably wanted the
498 		 * P[rint] command for backward compatibility, and the code
499 		 * just made Preserve and Put work by accident.  Nvi uses
500 		 * Previous to mean previous-in-a-new-screen, so be careful.
501 		 */
502 		if (newscreen && !F_ISSET(ecp->cmd, E_NEWSCREEN) &&
503 		    (ecp->cmd == &cmds[C_PRINT] ||
504 		    ecp->cmd == &cmds[C_PRESERVE]))
505 			newscreen = 0;
506 
507 		/* Test for a newscreen associated with this command. */
508 		if (newscreen && !F_ISSET(ecp->cmd, E_NEWSCREEN))
509 			goto unknown;
510 
511 		/* Secure means no shell access. */
512 		if (F_ISSET(ecp->cmd, E_SECURE) && O_ISSET(sp, O_SECURE)) {
513 			ex_wemsg(sp, ecp->cmd->name, EXM_SECURE);
514 			goto err;
515 		}
516 
517 		/*
518 		 * Multiple < and > characters; another "feature".  Note,
519 		 * The string passed to the underlying function may not be
520 		 * nul terminated in this case.
521 		 */
522 		if ((ecp->cmd == &cmds[C_SHIFTL] && *p == '<') ||
523 		    (ecp->cmd == &cmds[C_SHIFTR] && *p == '>')) {
524 			for (ch = *p;
525 			    ecp->clen > 0; --ecp->clen, ++ecp->cp)
526 				if (*ecp->cp != ch)
527 					break;
528 			if (argv_exp0(sp, ecp, p, ecp->cp - p))
529 				goto err;
530 		}
531 
532 		/* Set the format style flags for the next command. */
533 		if (ecp->cmd == &cmds[C_HASH])
534 			exp->fdef = E_C_HASH;
535 		else if (ecp->cmd == &cmds[C_LIST])
536 			exp->fdef = E_C_LIST;
537 		else if (ecp->cmd == &cmds[C_PRINT])
538 			exp->fdef = E_C_PRINT;
539 		F_CLR(ecp, E_USELASTCMD);
540 	} else {
541 		/* Print is the default command. */
542 		ecp->cmd = &cmds[C_PRINT];
543 
544 		/* Set the saved format flags. */
545 		F_SET(ecp, exp->fdef);
546 
547 		/*
548 		 * !!!
549 		 * If no address was specified, and it's not a global command,
550 		 * we up the address by one.  (I have no idea why globals are
551 		 * exempted, but it's (ahem) historic practice.)
552 		 */
553 		if (ecp->addrcnt == 0 && !F_ISSET(sp, SC_EX_GLOBAL)) {
554 			ecp->addrcnt = 1;
555 			ecp->addr1.lno = sp->lno + 1;
556 			ecp->addr1.cno = sp->cno;
557 		}
558 
559 		F_SET(ecp, E_USELASTCMD);
560 	}
561 
562 	/*
563 	 * !!!
564 	 * Historically, the number option applied to both ex and vi.  One
565 	 * strangeness was that ex didn't switch display formats until a
566 	 * command was entered, e.g. <CR>'s after the set didn't change to
567 	 * the new format, but :1p would.
568 	 */
569 	if (O_ISSET(sp, O_NUMBER)) {
570 		F_SET(ecp, E_OPTNUM);
571 		FL_SET(ecp->iflags, E_C_HASH);
572 	} else
573 		F_CLR(ecp, E_OPTNUM);
574 
575 	/* Check for ex mode legality. */
576 	if (F_ISSET(sp, SC_EX) && (F_ISSET(ecp->cmd, E_VIONLY) || newscreen)) {
577 		msgq_wstr(sp, M_ERR, ecp->cmd->name,
578 		    "082|%s: command not available in ex mode");
579 		goto err;
580 	}
581 
582 	/* Add standard command flags. */
583 	F_SET(ecp, ecp->cmd->flags);
584 	if (!newscreen)
585 		F_CLR(ecp, E_NEWSCREEN);
586 
587 	/*
588 	 * There are three normal termination cases for an ex command.  They
589 	 * are the end of the string (ecp->clen), or unescaped (by <literal
590 	 * next> characters) <newline> or '|' characters.  As we're now past
591 	 * possible addresses, we can determine how long the command is, so we
592 	 * don't have to look for all the possible terminations.  Naturally,
593 	 * there are some exciting special cases:
594 	 *
595 	 * 1: The bang, global, v and the filter versions of the read and
596 	 *    write commands are delimited by <newline>s (they can contain
597 	 *    shell pipes).
598 	 * 2: The ex, edit, next and visual in vi mode commands all take ex
599 	 *    commands as their first arguments.
600 	 * 3: The s command takes an RE as its first argument, and wants it
601 	 *    to be specially delimited.
602 	 *
603 	 * Historically, '|' characters in the first argument of the ex, edit,
604 	 * next, vi visual, and s commands didn't delimit the command.  And,
605 	 * in the filter cases for read and write, and the bang, global and v
606 	 * commands, they did not delimit the command at all.
607 	 *
608 	 * For example, the following commands were legal:
609 	 *
610 	 *	:edit +25|s/abc/ABC/ file.c
611 	 *	:s/|/PIPE/
612 	 *	:read !spell % | columnate
613 	 *	:global/pattern/p|l
614 	 *
615 	 * It's not quite as simple as it sounds, however.  The command:
616 	 *
617 	 *	:s/a/b/|s/c/d|set
618 	 *
619 	 * was also legal, i.e. the historic ex parser (using the word loosely,
620 	 * since "parser" implies some regularity of syntax) delimited the RE's
621 	 * based on its delimiter and not anything so irretrievably vulgar as a
622 	 * command syntax.
623 	 *
624 	 * Anyhow, the following code makes this all work.  First, for the
625 	 * special cases we move past their special argument(s).  Then, we
626 	 * do normal command processing on whatever is left.  Barf-O-Rama.
627 	 */
628 	discard = 0;		/* Characters discarded from the command. */
629 	arg1_len = 0;
630 	ecp->save_cmd = ecp->cp;
631 	if (ecp->cmd == &cmds[C_EDIT] || ecp->cmd == &cmds[C_EX] ||
632 	    ecp->cmd == &cmds[C_NEXT] || ecp->cmd == &cmds[C_VISUAL_VI] ||
633 	    ecp->cmd == &cmds[C_VSPLIT]) {
634 		/*
635 		 * Move to the next non-whitespace character.  A '!'
636 		 * immediately following the command is eaten as a
637 		 * force flag.
638 		 */
639 		if (ecp->clen > 0 && *ecp->cp == '!') {
640 			++ecp->cp;
641 			--ecp->clen;
642 			FL_SET(ecp->iflags, E_C_FORCE);
643 
644 			/* Reset, don't reparse. */
645 			ecp->save_cmd = ecp->cp;
646 		}
647 		for (; ecp->clen > 0; --ecp->clen, ++ecp->cp)
648 			if (!cmdskip(*ecp->cp))
649 				break;
650 		/*
651 		 * QUOTING NOTE:
652 		 *
653 		 * The historic implementation ignored all escape characters
654 		 * so there was no way to put a space or newline into the +cmd
655 		 * field.  We do a simplistic job of fixing it by moving to the
656 		 * first whitespace character that isn't escaped.  The escaping
657 		 * characters are stripped as no longer useful.
658 		 */
659 		if (ecp->clen > 0 && *ecp->cp == '+') {
660 			++ecp->cp;
661 			--ecp->clen;
662 			for (arg1 = p = ecp->cp;
663 			    ecp->clen > 0; --ecp->clen, ++ecp->cp) {
664 				ch = *ecp->cp;
665 				if (IS_ESCAPE(sp, ecp, ch) &&
666 				    ecp->clen > 1) {
667 					++discard;
668 					--ecp->clen;
669 					ch = *++ecp->cp;
670 				} else if (cmdskip(ch))
671 					break;
672 				*p++ = ch;
673 			}
674 			arg1_len = ecp->cp - arg1;
675 
676 			/* Reset, so the first argument isn't reparsed. */
677 			ecp->save_cmd = ecp->cp;
678 		}
679 	} else if (ecp->cmd == &cmds[C_BANG] ||
680 	    ecp->cmd == &cmds[C_GLOBAL] || ecp->cmd == &cmds[C_V]) {
681 		/*
682 		 * QUOTING NOTE:
683 		 *
684 		 * We use backslashes to escape <newline> characters, although
685 		 * this wasn't historic practice for the bang command.  It was
686 		 * for the global and v commands, and it's common usage when
687 		 * doing text insert during the command.  Escaping characters
688 		 * are stripped as no longer useful.
689 		 */
690 		for (p = ecp->cp; ecp->clen > 0; --ecp->clen, ++ecp->cp) {
691 			ch = *ecp->cp;
692 			if (ch == '\\' && ecp->clen > 1 && ecp->cp[1] == '\n') {
693 				++discard;
694 				--ecp->clen;
695 				ch = *++ecp->cp;
696 
697 				++gp->if_lno;
698 				++ecp->if_lno;
699 			} else if (ch == '\n')
700 				break;
701 			*p++ = ch;
702 		}
703 	} else if (ecp->cmd == &cmds[C_READ] || ecp->cmd == &cmds[C_WRITE]) {
704 		/*
705 		 * For write commands, if the next character is a <blank>, and
706 		 * the next non-blank character is a '!', it's a filter command
707 		 * and we want to eat everything up to the <newline>.  For read
708 		 * commands, if the next non-blank character is a '!', it's a
709 		 * filter command and we want to eat everything up to the next
710 		 * <newline>.  Otherwise, we're done.
711 		 */
712 		for (tmp = 0; ecp->clen > 0; --ecp->clen, ++ecp->cp) {
713 			ch = *ecp->cp;
714 			if (cmdskip(ch))
715 				tmp = 1;
716 			else
717 				break;
718 		}
719 		if (ecp->clen > 0 && ch == '!' &&
720 		    (ecp->cmd == &cmds[C_READ] || tmp))
721 			for (; ecp->clen > 0; --ecp->clen, ++ecp->cp)
722 				if (ecp->cp[0] == '\n')
723 					break;
724 	} else if (ecp->cmd == &cmds[C_SUBSTITUTE]) {
725 		/*
726 		 * Move to the next non-whitespace character, we'll use it as
727 		 * the delimiter.  If the character isn't an alphanumeric or
728 		 * a '|', it's the delimiter, so parse it.  Otherwise, we're
729 		 * into something like ":s g", so use the special s command.
730 		 */
731 		for (; ecp->clen > 0; --ecp->clen, ++ecp->cp)
732 			if (!cmdskip(ecp->cp[0]))
733 				break;
734 
735 		if (is09azAZ(ecp->cp[0]) || ecp->cp[0] == '|') {
736 			ecp->rcmd = cmds[C_SUBSTITUTE];
737 			ecp->rcmd.fn = ex_subagain;
738 			ecp->cmd = &ecp->rcmd;
739 		} else if (ecp->clen > 0) {
740 			/*
741 			 * QUOTING NOTE:
742 			 *
743 			 * Backslashes quote delimiter characters for RE's.
744 			 * The backslashes are NOT removed since they'll be
745 			 * used by the RE code.  Move to the third delimiter
746 			 * that's not escaped (or the end of the command).
747 			 */
748 			delim = *ecp->cp;
749 			++ecp->cp;
750 			--ecp->clen;
751 			for (cnt = 2; ecp->clen > 0 &&
752 			    cnt != 0; --ecp->clen, ++ecp->cp)
753 				if (ecp->cp[0] == '\\' &&
754 				    ecp->clen > 1) {
755 					++ecp->cp;
756 					--ecp->clen;
757 				} else if (ecp->cp[0] == delim)
758 					--cnt;
759 		}
760 	}
761 
762 	/*
763 	 * Use normal quoting and termination rules to find the end of this
764 	 * command.
765 	 *
766 	 * QUOTING NOTE:
767 	 *
768 	 * Historically, vi permitted ^V's to escape <newline>'s in the .exrc
769 	 * file.  It was almost certainly a bug, but that's what bug-for-bug
770 	 * compatibility means, Grasshopper.  Also, ^V's escape the command
771 	 * delimiters.  Literal next quote characters in front of the newlines,
772 	 * '|' characters or literal next characters are stripped as they're
773 	 * no longer useful.
774 	 */
775 	vi_address = ecp->clen != 0 && ecp->cp[0] != '\n';
776 	ecp->trailing = 0;
777 	for (p = ecp->cp; ecp->clen > 0; --ecp->clen, ++ecp->cp) {
778 		ch = ecp->cp[0];
779 		if (IS_ESCAPE(sp, ecp, ch) && ecp->clen > 1) {
780 			CHAR_T tmp = ecp->cp[1];
781 			if (tmp == '\n' || tmp == '|') {
782 				if (tmp == '\n') {
783 					++gp->if_lno;
784 					++ecp->if_lno;
785 				}
786 				++discard;
787 				--ecp->clen;
788 				++ecp->cp;
789 				ch = tmp;
790 			}
791 		} else if (ch == '\n' || ch == '|') {
792 			ecp->trailing = 1;
793 			if (ch == '\n')
794 				F_SET(ecp, E_NEWLINE);
795 			--ecp->clen;
796 			break;
797 		}
798 		*p++ = ch;
799 	}
800 
801 	/*
802 	 * Save off the next command information, go back to the
803 	 * original start of the command.
804 	 */
805 	p = ecp->cp + 1;
806 	ecp->cp = ecp->save_cmd;
807 	ecp->save_cmd = p;
808 	ecp->save_cmdlen = ecp->clen;
809 	ecp->clen = ((ecp->save_cmd - ecp->cp) - 1) - discard;
810 
811 	/*
812 	 * QUOTING NOTE:
813 	 *
814 	 * The "set tags" command historically used a backslash, not the
815 	 * user's literal next character, to escape whitespace.  Handle
816 	 * it here instead of complicating the argv_exp3() code.  Note,
817 	 * this isn't a particularly complex trap, and if backslashes were
818 	 * legal in set commands, this would have to be much more complicated.
819 	 */
820 	if (ecp->cmd == &cmds[C_SET]) {
821 		for (p = ecp->cp, len = ecp->clen; len > 0; --len, ++p)
822 			if (IS_ESCAPE(sp, ecp, *p) && len > 1) {
823 				--len;
824 				++p;
825 			} else if (*p == '\\')
826 				*p = CH_LITERAL;
827 	}
828 
829 	/*
830 	 * Set the default addresses.  It's an error to specify an address for
831 	 * a command that doesn't take them.  If two addresses are specified
832 	 * for a command that only takes one, lose the first one.  Two special
833 	 * cases here, some commands take 0 or 2 addresses.  For most of them
834 	 * (the E_ADDR2_ALL flag), 0 defaults to the entire file.  For one
835 	 * (the `!' command, the E_ADDR2_NONE flag), 0 defaults to no lines.
836 	 *
837 	 * Also, if the file is empty, some commands want to use an address of
838 	 * 0, i.e. the entire file is 0 to 0, and the default first address is
839 	 * 0.  Otherwise, an entire file is 1 to N and the default line is 1.
840 	 * Note, we also add the E_ADDR_ZERO flag to the command flags, for the
841 	 * case where the 0 address is only valid if it's a default address.
842 	 *
843 	 * Also, set a flag if we set the default addresses.  Some commands
844 	 * (ex: z) care if the user specified an address or if we just used
845 	 * the current cursor.
846 	 */
847 	switch (F_ISSET(ecp, E_ADDR1 | E_ADDR2 | E_ADDR2_ALL | E_ADDR2_NONE)) {
848 	case E_ADDR1:				/* One address: */
849 		switch (ecp->addrcnt) {
850 		case 0:				/* Default cursor/empty file. */
851 			ecp->addrcnt = 1;
852 			F_SET(ecp, E_ADDR_DEF);
853 			if (F_ISSET(ecp, E_ADDR_ZERODEF)) {
854 				if (db_last(sp, &lno))
855 					goto err;
856 				if (lno == 0) {
857 					ecp->addr1.lno = 0;
858 					F_SET(ecp, E_ADDR_ZERO);
859 				} else
860 					ecp->addr1.lno = sp->lno;
861 			} else
862 				ecp->addr1.lno = sp->lno;
863 			ecp->addr1.cno = sp->cno;
864 			break;
865 		case 1:
866 			break;
867 		case 2:				/* Lose the first address. */
868 			ecp->addrcnt = 1;
869 			ecp->addr1 = ecp->addr2;
870 		}
871 		break;
872 	case E_ADDR2_NONE:			/* Zero/two addresses: */
873 		if (ecp->addrcnt == 0)		/* Default to nothing. */
874 			break;
875 		goto two_addr;
876 	case E_ADDR2_ALL:			/* Zero/two addresses: */
877 		if (ecp->addrcnt == 0) {	/* Default entire/empty file. */
878 			F_SET(ecp, E_ADDR_DEF);
879 			ecp->addrcnt = 2;
880 			if (sp->ep == NULL)
881 				ecp->addr2.lno = 0;
882 			else if (db_last(sp, &ecp->addr2.lno))
883 				goto err;
884 			if (F_ISSET(ecp, E_ADDR_ZERODEF) &&
885 			    ecp->addr2.lno == 0) {
886 				ecp->addr1.lno = 0;
887 				F_SET(ecp, E_ADDR_ZERO);
888 			} else
889 				ecp->addr1.lno = 1;
890 			ecp->addr1.cno = ecp->addr2.cno = 0;
891 			F_SET(ecp, E_ADDR2_ALL);
892 			break;
893 		}
894 		/* FALLTHROUGH */
895 	case E_ADDR2:				/* Two addresses: */
896 two_addr:	switch (ecp->addrcnt) {
897 		case 0:				/* Default cursor/empty file. */
898 			ecp->addrcnt = 2;
899 			F_SET(ecp, E_ADDR_DEF);
900 			if (sp->lno == 1 &&
901 			    F_ISSET(ecp, E_ADDR_ZERODEF)) {
902 				if (db_last(sp, &lno))
903 					goto err;
904 				if (lno == 0) {
905 					ecp->addr1.lno = ecp->addr2.lno = 0;
906 					F_SET(ecp, E_ADDR_ZERO);
907 				} else
908 					ecp->addr1.lno =
909 					    ecp->addr2.lno = sp->lno;
910 			} else
911 				ecp->addr1.lno = ecp->addr2.lno = sp->lno;
912 			ecp->addr1.cno = ecp->addr2.cno = sp->cno;
913 			break;
914 		case 1:				/* Default to first address. */
915 			ecp->addrcnt = 2;
916 			ecp->addr2 = ecp->addr1;
917 			break;
918 		case 2:
919 			break;
920 		}
921 		break;
922 	default:
923 		if (ecp->addrcnt)		/* Error. */
924 			goto usage;
925 	}
926 
927 	/*
928 	 * !!!
929 	 * The ^D scroll command historically scrolled the value of the scroll
930 	 * option or to EOF.  It was an error if the cursor was already at EOF.
931 	 * (Leading addresses were permitted, but were then ignored.)
932 	 */
933 	if (ecp->cmd == &cmds[C_SCROLL]) {
934 		ecp->addrcnt = 2;
935 		ecp->addr1.lno = sp->lno + 1;
936 		ecp->addr2.lno = sp->lno + O_VAL(sp, O_SCROLL);
937 		ecp->addr1.cno = ecp->addr2.cno = sp->cno;
938 		if (db_last(sp, &lno))
939 			goto err;
940 		if (lno != 0 && lno > sp->lno && ecp->addr2.lno > lno)
941 			ecp->addr2.lno = lno;
942 	}
943 
944 	ecp->flagoff = 0;
945 	for (np = ecp->cmd->syntax; *np != '\0'; ++np) {
946 		/*
947 		 * The force flag is sensitive to leading whitespace, i.e.
948 		 * "next !" is different from "next!".  Handle it before
949 		 * skipping leading <blank>s.
950 		 */
951 		if (*np == '!') {
952 			if (ecp->clen > 0 && *ecp->cp == '!') {
953 				++ecp->cp;
954 				--ecp->clen;
955 				FL_SET(ecp->iflags, E_C_FORCE);
956 			}
957 			continue;
958 		}
959 
960 		/* Skip leading <blank>s. */
961 		for (; ecp->clen > 0; --ecp->clen, ++ecp->cp)
962 			if (!cmdskip(*ecp->cp))
963 				break;
964 		if (ecp->clen == 0)
965 			break;
966 
967 		switch (*np) {
968 		case '1':				/* +, -, #, l, p */
969 			/*
970 			 * !!!
971 			 * Historically, some flags were ignored depending
972 			 * on where they occurred in the command line.  For
973 			 * example, in the command, ":3+++p--#", historic vi
974 			 * acted on the '#' flag, but ignored the '-' flags.
975 			 * It's unambiguous what the flags mean, so we just
976 			 * handle them regardless of the stupidity of their
977 			 * location.
978 			 */
979 			for (; ecp->clen; --ecp->clen, ++ecp->cp)
980 				switch (*ecp->cp) {
981 				case '+':
982 					++ecp->flagoff;
983 					break;
984 				case '-':
985 				case '^':
986 					--ecp->flagoff;
987 					break;
988 				case '#':
989 					F_CLR(ecp, E_OPTNUM);
990 					FL_SET(ecp->iflags, E_C_HASH);
991 					exp->fdef |= E_C_HASH;
992 					break;
993 				case 'l':
994 					FL_SET(ecp->iflags, E_C_LIST);
995 					exp->fdef |= E_C_LIST;
996 					break;
997 				case 'p':
998 					FL_SET(ecp->iflags, E_C_PRINT);
999 					exp->fdef |= E_C_PRINT;
1000 					break;
1001 				default:
1002 					goto end_case1;
1003 				}
1004 end_case1:		break;
1005 		case '2':				/* -, ., +, ^ */
1006 		case '3':				/* -, ., +, ^, = */
1007 			for (; ecp->clen; --ecp->clen, ++ecp->cp)
1008 				switch (*ecp->cp) {
1009 				case '-':
1010 					FL_SET(ecp->iflags, E_C_DASH);
1011 					break;
1012 				case '.':
1013 					FL_SET(ecp->iflags, E_C_DOT);
1014 					break;
1015 				case '+':
1016 					FL_SET(ecp->iflags, E_C_PLUS);
1017 					break;
1018 				case '^':
1019 					FL_SET(ecp->iflags, E_C_CARAT);
1020 					break;
1021 				case '=':
1022 					if (*np == '3') {
1023 						FL_SET(ecp->iflags, E_C_EQUAL);
1024 						break;
1025 					}
1026 					/* FALLTHROUGH */
1027 				default:
1028 					goto end_case23;
1029 				}
1030 end_case23:		break;
1031 		case 'b':				/* buffer */
1032 			/*
1033 			 * !!!
1034 			 * Historically, "d #" was a delete with a flag, not a
1035 			 * delete into the '#' buffer.  If the current command
1036 			 * permits a flag, don't use one as a buffer.  However,
1037 			 * the 'l' and 'p' flags were legal buffer names in the
1038 			 * historic ex, and were used as buffers, not flags.
1039 			 */
1040 			if ((ecp->cp[0] == '+' || ecp->cp[0] == '-' ||
1041 			    ecp->cp[0] == '^' || ecp->cp[0] == '#') &&
1042 			    strchr(np, '1') != NULL)
1043 				break;
1044 			/*
1045 			 * !!!
1046 			 * Digits can't be buffer names in ex commands, or the
1047 			 * command "d2" would be a delete into buffer '2', and
1048 			 * not a two-line deletion.
1049 			 */
1050 			if (!ISDIGIT(ecp->cp[0])) {
1051 				ecp->buffer = *ecp->cp;
1052 				++ecp->cp;
1053 				--ecp->clen;
1054 				FL_SET(ecp->iflags, E_C_BUFFER);
1055 			}
1056 			break;
1057 		case 'c':				/* count [01+a] */
1058 			++np;
1059 			/* Validate any signed value. */
1060 			if (!ISDIGIT(*ecp->cp) && (*np != '+' ||
1061 			    (*ecp->cp != '+' && *ecp->cp != '-')))
1062 				break;
1063 			/* If a signed value, set appropriate flags. */
1064 			if (*ecp->cp == '-')
1065 				FL_SET(ecp->iflags, E_C_COUNT_NEG);
1066 			else if (*ecp->cp == '+')
1067 				FL_SET(ecp->iflags, E_C_COUNT_POS);
1068 			if ((nret =
1069 			    nget_slong(&ltmp, ecp->cp, &t, 10)) != NUM_OK) {
1070 				ex_badaddr(sp, NULL, A_NOTSET, nret);
1071 				goto err;
1072 			}
1073 			if (ltmp == 0 && *np != '0') {
1074 				msgq(sp, M_ERR, "083|Count may not be zero");
1075 				goto err;
1076 			}
1077 			ecp->clen -= (t - ecp->cp);
1078 			ecp->cp = t;
1079 
1080 			/*
1081 			 * Counts as address offsets occur in commands taking
1082 			 * two addresses.  Historic vi practice was to use
1083 			 * the count as an offset from the *second* address.
1084 			 *
1085 			 * Set a count flag; some underlying commands (see
1086 			 * join) do different things with counts than with
1087 			 * line addresses.
1088 			 */
1089 			if (*np == 'a') {
1090 				ecp->addr1 = ecp->addr2;
1091 				ecp->addr2.lno = ecp->addr1.lno + ltmp - 1;
1092 			} else
1093 				ecp->count = ltmp;
1094 			FL_SET(ecp->iflags, E_C_COUNT);
1095 			break;
1096 		case 'f':				/* file */
1097 			if (argv_exp2(sp, ecp, ecp->cp, ecp->clen))
1098 				goto err;
1099 			goto arg_cnt_chk;
1100 		case 'l':				/* line */
1101 			/*
1102 			 * Get a line specification.
1103 			 *
1104 			 * If the line was a search expression, we may have
1105 			 * changed state during the call, and we're now
1106 			 * searching the file.  Push ourselves onto the state
1107 			 * stack.
1108 			 */
1109 			if (ex_line(sp, ecp, &cur, &isaddr, &tmp))
1110 				goto rfail;
1111 			if (tmp)
1112 				goto err;
1113 
1114 			/* Line specifications are always required. */
1115 			if (!isaddr) {
1116 				msgq_wstr(sp, M_ERR, ecp->cp,
1117 				     "084|%s: bad line specification");
1118 				goto err;
1119 			}
1120 			/*
1121 			 * The target line should exist for these commands,
1122 			 * but 0 is legal for them as well.
1123 			 */
1124 			if (cur.lno != 0 && !db_exist(sp, cur.lno)) {
1125 				ex_badaddr(sp, NULL, A_EOF, NUM_OK);
1126 				goto err;
1127 			}
1128 			ecp->lineno = cur.lno;
1129 			break;
1130 		case 'S':				/* string, file exp. */
1131 			if (ecp->clen != 0) {
1132 				if (argv_exp1(sp, ecp, ecp->cp,
1133 				    ecp->clen, ecp->cmd == &cmds[C_BANG]))
1134 					goto err;
1135 				goto addr_verify;
1136 			}
1137 			/* FALLTHROUGH */
1138 		case 's':				/* string */
1139 			if (argv_exp0(sp, ecp, ecp->cp, ecp->clen))
1140 				goto err;
1141 			goto addr_verify;
1142 		case 'W':				/* word string */
1143 			/*
1144 			 * QUOTING NOTE:
1145 			 *
1146 			 * Literal next characters escape the following
1147 			 * character.  Quoting characters are stripped here
1148 			 * since they are no longer useful.
1149 			 *
1150 			 * First there was the word.
1151 			 */
1152 			for (p = t = ecp->cp;
1153 			    ecp->clen > 0; --ecp->clen, ++ecp->cp) {
1154 				ch = *ecp->cp;
1155 				if (IS_ESCAPE(sp,
1156 				    ecp, ch) && ecp->clen > 1) {
1157 					--ecp->clen;
1158 					*p++ = *++ecp->cp;
1159 				} else if (cmdskip(ch)) {
1160 					++ecp->cp;
1161 					--ecp->clen;
1162 					break;
1163 				} else
1164 					*p++ = ch;
1165 			}
1166 			if (argv_exp0(sp, ecp, t, p - t))
1167 				goto err;
1168 
1169 			/* Delete intervening whitespace. */
1170 			for (; ecp->clen > 0;
1171 			    --ecp->clen, ++ecp->cp) {
1172 				ch = *ecp->cp;
1173 				if (!cmdskip(ch))
1174 					break;
1175 			}
1176 			if (ecp->clen == 0)
1177 				goto usage;
1178 
1179 			/* Followed by the string. */
1180 			for (p = t = ecp->cp; ecp->clen > 0;
1181 			    --ecp->clen, ++ecp->cp, ++p) {
1182 				ch = *ecp->cp;
1183 				if (IS_ESCAPE(sp,
1184 				    ecp, ch) && ecp->clen > 1) {
1185 					--ecp->clen;
1186 					*p = *++ecp->cp;
1187 				} else
1188 					*p = ch;
1189 			}
1190 			if (argv_exp0(sp, ecp, t, p - t))
1191 				goto err;
1192 			goto addr_verify;
1193 		case 'w':				/* word */
1194 			if (argv_exp3(sp, ecp, ecp->cp, ecp->clen))
1195 				goto err;
1196 arg_cnt_chk:		if (*++np != 'N') {		/* N */
1197 				/*
1198 				 * If a number is specified, must either be
1199 				 * 0 or that number, if optional, and that
1200 				 * number, if required.
1201 				 */
1202 				tmp = *np - '0';
1203 				if ((*++np != 'o' || exp->argsoff != 0) &&
1204 				    exp->argsoff != tmp)
1205 					goto usage;
1206 			}
1207 			goto addr_verify;
1208 		default: {
1209 			size_t nlen;
1210 			char *nstr;
1211 
1212 			INT2CHAR(sp, ecp->cmd->name, STRLEN(ecp->cmd->name) + 1,
1213 			    nstr, nlen);
1214 			msgq(sp, M_ERR,
1215 			    "085|Internal syntax table error (%s: %s)",
1216 			    nstr, KEY_NAME(sp, *np));
1217 		}
1218 		}
1219 	}
1220 
1221 	/* Skip trailing whitespace. */
1222 	for (; ecp->clen > 0; --ecp->clen) {
1223 		ch = *ecp->cp++;
1224 		if (!cmdskip(ch))
1225 			break;
1226 	}
1227 
1228 	/*
1229 	 * There shouldn't be anything left, and no more required fields,
1230 	 * i.e neither 'l' or 'r' in the syntax string.
1231 	 */
1232 	if (ecp->clen != 0 || strpbrk(np, "lr")) {
1233 usage:		msgq(sp, M_ERR, "086|Usage: %s", ecp->cmd->usage);
1234 		goto err;
1235 	}
1236 
1237 	/*
1238 	 * Verify that the addresses are legal.  Check the addresses here,
1239 	 * because this is a place where all ex addresses pass through.
1240 	 * (They don't all pass through ex_line(), for instance.)  We're
1241 	 * assuming that any non-existent line doesn't exist because it's
1242 	 * past the end-of-file.  That's a pretty good guess.
1243 	 *
1244 	 * If it's a "default vi command", an address of zero is okay.
1245 	 */
1246 addr_verify:
1247 	switch (ecp->addrcnt) {
1248 	case 2:
1249 		/*
1250 		 * Historic ex/vi permitted commands with counts to go past
1251 		 * EOF.  So, for example, if the file only had 5 lines, the
1252 		 * ex command "1,6>" would fail, but the command ">300"
1253 		 * would succeed.  Since we don't want to have to make all
1254 		 * of the underlying commands handle random line numbers,
1255 		 * fix it here.
1256 		 */
1257 		if (ecp->addr2.lno == 0) {
1258 			if (!F_ISSET(ecp, E_ADDR_ZERO) &&
1259 			    (F_ISSET(sp, SC_EX) ||
1260 			    !F_ISSET(ecp, E_USELASTCMD))) {
1261 				ex_badaddr(sp, ecp->cmd, A_ZERO, NUM_OK);
1262 				goto err;
1263 			}
1264 		} else if (!db_exist(sp, ecp->addr2.lno)) {
1265 			if (FL_ISSET(ecp->iflags, E_C_COUNT)) {
1266 				if (db_last(sp, &lno))
1267 					goto err;
1268 				ecp->addr2.lno = lno;
1269 			} else {
1270 				ex_badaddr(sp, NULL, A_EOF, NUM_OK);
1271 				goto err;
1272 			}
1273 		}
1274 		/* FALLTHROUGH */
1275 	case 1:
1276 		if (ecp->addr1.lno == 0) {
1277 			if (!F_ISSET(ecp, E_ADDR_ZERO) &&
1278 			    (F_ISSET(sp, SC_EX) ||
1279 			    !F_ISSET(ecp, E_USELASTCMD))) {
1280 				ex_badaddr(sp, ecp->cmd, A_ZERO, NUM_OK);
1281 				goto err;
1282 			}
1283 		} else if (!db_exist(sp, ecp->addr1.lno)) {
1284 			ex_badaddr(sp, NULL, A_EOF, NUM_OK);
1285 			goto err;
1286 		}
1287 		break;
1288 	}
1289 
1290 	/*
1291 	 * If doing a default command and there's nothing left on the line,
1292 	 * vi just moves to the line.  For example, ":3" and ":'a,'b" just
1293 	 * move to line 3 and line 'b, respectively, but ":3|" prints line 3.
1294 	 *
1295 	 * !!!
1296 	 * In addition, IF THE LINE CHANGES, move to the first nonblank of
1297 	 * the line.
1298 	 *
1299 	 * !!!
1300 	 * This is done before the absolute mark gets set; historically,
1301 	 * "/a/,/b/" did NOT set vi's absolute mark, but "/a/,/b/d" did.
1302 	 */
1303 	if ((F_ISSET(sp, SC_VI) || F_ISSET(ecp, E_NOPRDEF)) &&
1304 	    F_ISSET(ecp, E_USELASTCMD) && vi_address == 0) {
1305 		switch (ecp->addrcnt) {
1306 		case 2:
1307 			if (sp->lno !=
1308 			    (ecp->addr2.lno ? ecp->addr2.lno : 1)) {
1309 				sp->lno =
1310 				    ecp->addr2.lno ? ecp->addr2.lno : 1;
1311 				sp->cno = 0;
1312 				(void)nonblank(sp, sp->lno, &sp->cno);
1313 			}
1314 			break;
1315 		case 1:
1316 			if (sp->lno !=
1317 			    (ecp->addr1.lno ? ecp->addr1.lno : 1)) {
1318 				sp->lno =
1319 				    ecp->addr1.lno ? ecp->addr1.lno : 1;
1320 				sp->cno = 0;
1321 				(void)nonblank(sp, sp->lno, &sp->cno);
1322 			}
1323 			break;
1324 		}
1325 		ecp->cp = ecp->save_cmd;
1326 		ecp->clen = ecp->save_cmdlen;
1327 		goto loop;
1328 	}
1329 
1330 	/*
1331 	 * Set the absolute mark -- we have to set it for vi here, in case
1332 	 * it's a compound command, e.g. ":5p|6" should set the absolute
1333 	 * mark for vi.
1334 	 */
1335 	if (F_ISSET(ecp, E_ABSMARK)) {
1336 		cur.lno = sp->lno;
1337 		cur.cno = sp->cno;
1338 		F_CLR(ecp, E_ABSMARK);
1339 		if (mark_set(sp, ABSMARK1, &cur, 1))
1340 			goto err;
1341 	}
1342 
1343 #if defined(DEBUG) && defined(COMLOG)
1344 	ex_comlog(sp, ecp);
1345 #endif
1346 	/* Increment the command count if not called from vi. */
1347 	if (F_ISSET(sp, SC_EX))
1348 		++sp->ccnt;
1349 
1350 	/*
1351 	 * If file state available, and not doing a global command,
1352 	 * log the start of an action.
1353 	 */
1354 	if (sp->ep != NULL && !F_ISSET(sp, SC_EX_GLOBAL))
1355 		(void)log_cursor(sp);
1356 
1357 	/*
1358 	 * !!!
1359 	 * There are two special commands for the purposes of this code: the
1360 	 * default command (<carriage-return>) or the scrolling commands (^D
1361 	 * and <EOF>) as the first non-<blank> characters  in the line.
1362 	 *
1363 	 * If this is the first command in the command line, we received the
1364 	 * command from the ex command loop and we're talking to a tty, and
1365 	 * and there's nothing else on the command line, and it's one of the
1366 	 * special commands, we move back up to the previous line, and erase
1367 	 * the prompt character with the output.  Since ex runs in canonical
1368 	 * mode, we don't have to do anything else, a <newline> has already
1369 	 * been echoed by the tty driver.  It's OK if vi calls us -- we won't
1370 	 * be in ex mode so we'll do nothing.
1371 	 */
1372 	if (F_ISSET(ecp, E_NRSEP)) {
1373 		if (sp->ep != NULL &&
1374 		    F_ISSET(sp, SC_EX) && !F_ISSET(gp, G_SCRIPTED) &&
1375 		    (F_ISSET(ecp, E_USELASTCMD) || ecp->cmd == &cmds[C_SCROLL]))
1376 			gp->scr_ex_adjust(sp, EX_TERM_SCROLL);
1377 		F_CLR(ecp, E_NRSEP);
1378 	}
1379 
1380 	/*
1381 	 * Call the underlying function for the ex command.
1382 	 *
1383 	 * XXX
1384 	 * Interrupts behave like errors, for now.
1385 	 */
1386 	if (ecp->cmd->fn(sp, ecp) || INTERRUPTED(sp)) {
1387 		if (F_ISSET(gp, G_SCRIPTED))
1388 			F_SET(sp, SC_EXIT_FORCE);
1389 		goto err;
1390 	}
1391 
1392 #ifdef DEBUG
1393 	/* Make sure no function left global temporary space locked. */
1394 	if (F_ISSET(gp, G_TMP_INUSE)) {
1395 		F_CLR(gp, G_TMP_INUSE);
1396 		msgq_wstr(sp, M_ERR, ecp->cmd->name,
1397 		    "087|%s: temporary buffer not released");
1398 	}
1399 #endif
1400 	/*
1401 	 * Ex displayed the number of lines modified immediately after each
1402 	 * command, so the command "1,10d|1,10d" would display:
1403 	 *
1404 	 *	10 lines deleted
1405 	 *	10 lines deleted
1406 	 *	<autoprint line>
1407 	 *
1408 	 * Executing ex commands from vi only reported the final modified
1409 	 * lines message -- that's wrong enough that we don't match it.
1410 	 */
1411 	if (F_ISSET(sp, SC_EX))
1412 		mod_rpt(sp);
1413 
1414 	/*
1415 	 * Integrate any offset parsed by the underlying command, and make
1416 	 * sure the referenced line exists.
1417 	 *
1418 	 * XXX
1419 	 * May not match historic practice (which I've never been able to
1420 	 * completely figure out.)  For example, the '=' command from vi
1421 	 * mode often got the offset wrong, and complained it was too large,
1422 	 * but didn't seem to have a problem with the cursor.  If anyone
1423 	 * complains, ask them how it's supposed to work, they might know.
1424 	 */
1425 	if (sp->ep != NULL && ecp->flagoff) {
1426 		if (ecp->flagoff < 0) {
1427 			if (sp->lno <= -ecp->flagoff) {
1428 				msgq(sp, M_ERR,
1429 				    "088|Flag offset to before line 1");
1430 				goto err;
1431 			}
1432 		} else {
1433 			if (!NPFITS(MAX_REC_NUMBER, sp->lno, ecp->flagoff)) {
1434 				ex_badaddr(sp, NULL, A_NOTSET, NUM_OVER);
1435 				goto err;
1436 			}
1437 			if (!db_exist(sp, sp->lno + ecp->flagoff)) {
1438 				msgq(sp, M_ERR,
1439 				    "089|Flag offset past end-of-file");
1440 				goto err;
1441 			}
1442 		}
1443 		sp->lno += ecp->flagoff;
1444 	}
1445 
1446 	/*
1447 	 * If the command executed successfully, we may want to display a line
1448 	 * based on the autoprint option or an explicit print flag.  (Make sure
1449 	 * that there's a line to display.)  Also, the autoprint edit option is
1450 	 * turned off for the duration of global commands.
1451 	 */
1452 	if (F_ISSET(sp, SC_EX) && sp->ep != NULL && sp->lno != 0) {
1453 		/*
1454 		 * The print commands have already handled the `print' flags.
1455 		 * If so, clear them.
1456 		 */
1457 		if (FL_ISSET(ecp->iflags, E_CLRFLAG))
1458 			FL_CLR(ecp->iflags, E_C_HASH | E_C_LIST | E_C_PRINT);
1459 
1460 		/* If hash set only because of the number option, discard it. */
1461 		if (F_ISSET(ecp, E_OPTNUM))
1462 			FL_CLR(ecp->iflags, E_C_HASH);
1463 
1464 		/*
1465 		 * If there was an explicit flag to display the new cursor line,
1466 		 * or autoprint is set and a change was made, display the line.
1467 		 * If any print flags were set use them, else default to print.
1468 		 */
1469 		LF_INIT(FL_ISSET(ecp->iflags, E_C_HASH | E_C_LIST | E_C_PRINT));
1470 		if (!LF_ISSET(E_C_HASH | E_C_LIST | E_C_PRINT | E_NOAUTO) &&
1471 		    !F_ISSET(sp, SC_EX_GLOBAL) &&
1472 		    O_ISSET(sp, O_AUTOPRINT) && F_ISSET(ecp, E_AUTOPRINT)) {
1473 			/* Honor the number option if autoprint is set. */
1474 			if (F_ISSET(ecp, E_OPTNUM))
1475 				LF_INIT(E_C_HASH);
1476 			else
1477 				LF_INIT(E_C_PRINT);
1478 		}
1479 
1480 		if (LF_ISSET(E_C_HASH | E_C_LIST | E_C_PRINT)) {
1481 			cur.lno = sp->lno;
1482 			cur.cno = 0;
1483 			(void)ex_print(sp, ecp, &cur, &cur, flags);
1484 		}
1485 	}
1486 
1487 	/*
1488 	 * If the command had an associated "+cmd", it has to be executed
1489 	 * before we finish executing any more of this ex command.  For
1490 	 * example, consider a .exrc file that contains the following lines:
1491 	 *
1492 	 *	:set all
1493 	 *	:edit +25 file.c|s/abc/ABC/|1
1494 	 *	:3,5 print
1495 	 *
1496 	 * This can happen more than once -- the historic vi simply hung or
1497 	 * dropped core, of course.  Prepend the + command back into the
1498 	 * current command and continue.  We may have to add an additional
1499 	 * <literal next> character.  We know that it will fit because we
1500 	 * discarded at least one space and the + character.
1501 	 */
1502 	if (arg1_len != 0) {
1503 		/*
1504 		 * If the last character of the + command was a <literal next>
1505 		 * character, it would be treated differently because of the
1506 		 * append.  Quote it, if necessary.
1507 		 */
1508 		if (IS_ESCAPE(sp, ecp, arg1[arg1_len - 1])) {
1509 			*--ecp->save_cmd = CH_LITERAL;
1510 			++ecp->save_cmdlen;
1511 		}
1512 
1513 		ecp->save_cmd -= arg1_len;
1514 		ecp->save_cmdlen += arg1_len;
1515 		MEMMOVE(ecp->save_cmd, arg1, arg1_len);
1516 
1517 		/*
1518 		 * Any commands executed from a +cmd are executed starting at
1519 		 * the first column of the last line of the file -- NOT the
1520 		 * first nonblank.)  The main file startup code doesn't know
1521 		 * that a +cmd was set, however, so it may have put us at the
1522 		 * top of the file.  (Note, this is safe because we must have
1523 		 * switched files to get here.)
1524 		 */
1525 		F_SET(ecp, E_MOVETOEND);
1526 	}
1527 
1528 	/* Update the current command. */
1529 	ecp->cp = ecp->save_cmd;
1530 	ecp->clen = ecp->save_cmdlen;
1531 
1532 	/*
1533 	 * !!!
1534 	 * If we've changed screens or underlying files, any pending global or
1535 	 * v command, or @ buffer that has associated addresses, has to be
1536 	 * discarded.  This is historic practice for globals, and necessary for
1537 	 * @ buffers that had associated addresses.
1538 	 *
1539 	 * Otherwise, if we've changed underlying files, it's not a problem,
1540 	 * we continue with the rest of the ex command(s), operating on the
1541 	 * new file.  However, if we switch screens (either by exiting or by
1542 	 * an explicit command), we have no way of knowing where to put output
1543 	 * messages, and, since we don't control screens here, we could screw
1544 	 * up the upper layers, (e.g. we could exit/reenter a screen multiple
1545 	 * times).  So, return and continue after we've got a new screen.
1546 	 */
1547 	if (F_ISSET(sp, SC_EXIT | SC_EXIT_FORCE | SC_FSWITCH | SC_SSWITCH)) {
1548 		at_found = gv_found = 0;
1549 		SLIST_FOREACH(ecp, sp->gp->ecq, q)
1550 			switch (ecp->agv_flags) {
1551 			case 0:
1552 			case AGV_AT_NORANGE:
1553 				break;
1554 			case AGV_AT:
1555 				if (!at_found) {
1556 					at_found = 1;
1557 					msgq(sp, M_ERR,
1558 		"090|@ with range running when the file/screen changed");
1559 				}
1560 				break;
1561 			case AGV_GLOBAL:
1562 			case AGV_V:
1563 				if (!gv_found) {
1564 					gv_found = 1;
1565 					msgq(sp, M_ERR,
1566 		"091|Global/v command running when the file/screen changed");
1567 				}
1568 				break;
1569 			default:
1570 				abort();
1571 			}
1572 		if (at_found || gv_found)
1573 			goto discard;
1574 		if (F_ISSET(sp, SC_EXIT | SC_EXIT_FORCE | SC_SSWITCH))
1575 			goto rsuccess;
1576 	}
1577 
1578 	goto loop;
1579 	/* NOTREACHED */
1580 
1581 err:	/*
1582 	 * On command failure, we discard keys and pending commands remaining,
1583 	 * as well as any keys that were mapped and waiting.  The save_cmdlen
1584 	 * test is not necessarily correct.  If we fail early enough we don't
1585 	 * know if the entire string was a single command or not.  Guess, as
1586 	 * it's useful to know if commands other than the current one are being
1587 	 * discarded.
1588 	 */
1589 	if (ecp->save_cmdlen == 0)
1590 		for (; ecp->clen; --ecp->clen) {
1591 			ch = *ecp->cp++;
1592 			if (IS_ESCAPE(sp, ecp, ch) && ecp->clen > 1) {
1593 				--ecp->clen;
1594 				++ecp->cp;
1595 			} else if (ch == '\n' || ch == '|') {
1596 				if (ecp->clen > 1)
1597 					ecp->save_cmdlen = 1;
1598 				break;
1599 			}
1600 		}
1601 	if (ecp->save_cmdlen != 0 || SLIST_FIRST(gp->ecq) != &gp->excmd) {
1602 discard:	msgq(sp, M_BERR,
1603 		    "092|Ex command failed: pending commands discarded");
1604 		ex_discard(sp);
1605 	}
1606 	if (v_event_flush(sp, CH_MAPPED))
1607 		msgq(sp, M_BERR,
1608 		    "093|Ex command failed: mapped keys discarded");
1609 
1610 rfail:	tmp = 1;
1611 	if (0)
1612 rsuccess:	tmp = 0;
1613 
1614 	/* Turn off any file name error information. */
1615 	gp->if_name = NULL;
1616 
1617 	/* Turn off the global bit. */
1618 	F_CLR(sp, SC_EX_GLOBAL);
1619 
1620 	return (tmp);
1621 }
1622 
1623 /*
1624  * ex_range --
1625  *	Get a line range for ex commands, or perform a vi ex address search.
1626  *
1627  * PUBLIC: int ex_range(SCR *, EXCMD *, int *);
1628  */
1629 int
1630 ex_range(SCR *sp, EXCMD *ecp, int *errp)
1631 {
1632 	enum { ADDR_FOUND, ADDR_NEED, ADDR_NONE } addr;
1633 	GS *gp;
1634 	EX_PRIVATE *exp;
1635 	MARK m;
1636 	int isaddr;
1637 
1638 	*errp = 0;
1639 
1640 	/*
1641 	 * Parse comma or semi-colon delimited line specs.
1642 	 *
1643 	 * Semi-colon delimiters update the current address to be the last
1644 	 * address.  For example, the command
1645 	 *
1646 	 *	:3;/pattern/ecp->cp
1647 	 *
1648 	 * will search for pattern from line 3.  In addition, if ecp->cp
1649 	 * is not a valid command, the current line will be left at 3, not
1650 	 * at the original address.
1651 	 *
1652 	 * Extra addresses are discarded, starting with the first.
1653 	 *
1654 	 * !!!
1655 	 * If any addresses are missing, they default to the current line.
1656 	 * This was historically true for both leading and trailing comma
1657 	 * delimited addresses as well as for trailing semicolon delimited
1658 	 * addresses.  For consistency, we make it true for leading semicolon
1659 	 * addresses as well.
1660 	 */
1661 	gp = sp->gp;
1662 	exp = EXP(sp);
1663 	for (addr = ADDR_NONE, ecp->addrcnt = 0; ecp->clen > 0;)
1664 		switch (*ecp->cp) {
1665 		case '%':		/* Entire file. */
1666 			/* Vi ex address searches didn't permit % signs. */
1667 			if (F_ISSET(ecp, E_VISEARCH))
1668 				goto ret;
1669 
1670 			/* It's an error if the file is empty. */
1671 			if (sp->ep == NULL) {
1672 				ex_badaddr(sp, NULL, A_EMPTY, NUM_OK);
1673 				*errp = 1;
1674 				return (0);
1675 			}
1676 			/*
1677 			 * !!!
1678 			 * A percent character addresses all of the lines in
1679 			 * the file.  Historically, it couldn't be followed by
1680 			 * any other address.  We do it as a text substitution
1681 			 * for simplicity.  POSIX 1003.2 is expected to follow
1682 			 * this practice.
1683 			 *
1684 			 * If it's an empty file, the first line is 0, not 1.
1685 			 */
1686 			if (addr == ADDR_FOUND) {
1687 				ex_badaddr(sp, NULL, A_COMBO, NUM_OK);
1688 				*errp = 1;
1689 				return (0);
1690 			}
1691 			if (db_last(sp, &ecp->addr2.lno))
1692 				return (1);
1693 			ecp->addr1.lno = ecp->addr2.lno == 0 ? 0 : 1;
1694 			ecp->addr1.cno = ecp->addr2.cno = 0;
1695 			ecp->addrcnt = 2;
1696 			addr = ADDR_FOUND;
1697 			++ecp->cp;
1698 			--ecp->clen;
1699 			break;
1700 		case ',':	       /* Comma delimiter. */
1701 			/* Vi ex address searches didn't permit commas. */
1702 			if (F_ISSET(ecp, E_VISEARCH))
1703 				goto ret;
1704 			/* FALLTHROUGH */
1705 		case ';':	       /* Semi-colon delimiter. */
1706 			if (sp->ep == NULL) {
1707 				ex_badaddr(sp, NULL, A_EMPTY, NUM_OK);
1708 				*errp = 1;
1709 				return (0);
1710 			}
1711 			if (addr != ADDR_FOUND)
1712 				switch (ecp->addrcnt) {
1713 				case 0:
1714 					ecp->addr1.lno = sp->lno;
1715 					ecp->addr1.cno = sp->cno;
1716 					ecp->addrcnt = 1;
1717 					break;
1718 				case 2:
1719 					ecp->addr1 = ecp->addr2;
1720 					/* FALLTHROUGH */
1721 				case 1:
1722 					ecp->addr2.lno = sp->lno;
1723 					ecp->addr2.cno = sp->cno;
1724 					ecp->addrcnt = 2;
1725 					break;
1726 				}
1727 			if (*ecp->cp == ';')
1728 				switch (ecp->addrcnt) {
1729 				case 0:
1730 					abort();
1731 					/* NOTREACHED */
1732 				case 1:
1733 					sp->lno = ecp->addr1.lno;
1734 					sp->cno = ecp->addr1.cno;
1735 					break;
1736 				case 2:
1737 					sp->lno = ecp->addr2.lno;
1738 					sp->cno = ecp->addr2.cno;
1739 					break;
1740 				}
1741 			addr = ADDR_NEED;
1742 			/* FALLTHROUGH */
1743 		case ' ':		/* Whitespace. */
1744 		case '\t':		/* Whitespace. */
1745 			++ecp->cp;
1746 			--ecp->clen;
1747 			break;
1748 		default:
1749 			/* Get a line specification. */
1750 			if (ex_line(sp, ecp, &m, &isaddr, errp))
1751 				return (1);
1752 			if (*errp)
1753 				return (0);
1754 			if (!isaddr)
1755 				goto ret;
1756 			if (addr == ADDR_FOUND) {
1757 				ex_badaddr(sp, NULL, A_COMBO, NUM_OK);
1758 				*errp = 1;
1759 				return (0);
1760 			}
1761 			switch (ecp->addrcnt) {
1762 			case 0:
1763 				ecp->addr1 = m;
1764 				ecp->addrcnt = 1;
1765 				break;
1766 			case 1:
1767 				ecp->addr2 = m;
1768 				ecp->addrcnt = 2;
1769 				break;
1770 			case 2:
1771 				ecp->addr1 = ecp->addr2;
1772 				ecp->addr2 = m;
1773 				break;
1774 			}
1775 			addr = ADDR_FOUND;
1776 			break;
1777 		}
1778 
1779 	/*
1780 	 * !!!
1781 	 * Vi ex address searches are indifferent to order or trailing
1782 	 * semi-colons.
1783 	 */
1784 ret:	if (F_ISSET(ecp, E_VISEARCH))
1785 		return (0);
1786 
1787 	if (addr == ADDR_NEED)
1788 		switch (ecp->addrcnt) {
1789 		case 0:
1790 			ecp->addr1.lno = sp->lno;
1791 			ecp->addr1.cno = sp->cno;
1792 			ecp->addrcnt = 1;
1793 			break;
1794 		case 2:
1795 			ecp->addr1 = ecp->addr2;
1796 			/* FALLTHROUGH */
1797 		case 1:
1798 			ecp->addr2.lno = sp->lno;
1799 			ecp->addr2.cno = sp->cno;
1800 			ecp->addrcnt = 2;
1801 			break;
1802 		}
1803 
1804 	if (ecp->addrcnt == 2 && ecp->addr2.lno < ecp->addr1.lno) {
1805 		msgq(sp, M_ERR,
1806 		    "094|The second address is smaller than the first");
1807 		*errp = 1;
1808 	}
1809 	return (0);
1810 }
1811 
1812 /*
1813  * ex_line --
1814  *	Get a single line address specifier.
1815  *
1816  * The way the "previous context" mark worked was that any "non-relative"
1817  * motion set it.  While ex/vi wasn't totally consistent about this, ANY
1818  * numeric address, search pattern, '$', or mark reference in an address
1819  * was considered non-relative, and set the value.  Which should explain
1820  * why we're hacking marks down here.  The problem was that the mark was
1821  * only set if the command was called, i.e. we have to set a flag and test
1822  * it later.
1823  *
1824  * XXX
1825  * This is probably still not exactly historic practice, although I think
1826  * it's fairly close.
1827  */
1828 static int
1829 ex_line(SCR *sp, EXCMD *ecp, MARK *mp, int *isaddrp, int *errp)
1830 {
1831 	enum nresult nret;
1832 	EX_PRIVATE *exp;
1833 	GS *gp;
1834 	long total, val;
1835 	int isneg;
1836 	int (*sf)(SCR *, MARK *, MARK *, CHAR_T *, size_t, CHAR_T **, u_int);
1837 	CHAR_T *endp;
1838 
1839 	gp = sp->gp;
1840 	exp = EXP(sp);
1841 
1842 	*isaddrp = *errp = 0;
1843 	F_CLR(ecp, E_DELTA);
1844 
1845 	/* No addresses permitted until a file has been read in. */
1846 	if (sp->ep == NULL && STRCHR(L("$0123456789'\\/?.+-^"), *ecp->cp)) {
1847 		ex_badaddr(sp, NULL, A_EMPTY, NUM_OK);
1848 		*errp = 1;
1849 		return (0);
1850 	}
1851 
1852 	switch (*ecp->cp) {
1853 	case '$':				/* Last line in the file. */
1854 		*isaddrp = 1;
1855 		F_SET(ecp, E_ABSMARK);
1856 
1857 		mp->cno = 0;
1858 		if (db_last(sp, &mp->lno))
1859 			return (1);
1860 		++ecp->cp;
1861 		--ecp->clen;
1862 		break;				/* Absolute line number. */
1863 	case '0': case '1': case '2': case '3': case '4':
1864 	case '5': case '6': case '7': case '8': case '9':
1865 		*isaddrp = 1;
1866 		F_SET(ecp, E_ABSMARK);
1867 
1868 		if ((nret = nget_slong(&val, ecp->cp, &endp, 10)) != NUM_OK) {
1869 			ex_badaddr(sp, NULL, A_NOTSET, nret);
1870 			*errp = 1;
1871 			return (0);
1872 		}
1873 		if (!NPFITS(MAX_REC_NUMBER, 0, val)) {
1874 			ex_badaddr(sp, NULL, A_NOTSET, NUM_OVER);
1875 			*errp = 1;
1876 			return (0);
1877 		}
1878 		mp->lno = val;
1879 		mp->cno = 0;
1880 		ecp->clen -= (endp - ecp->cp);
1881 		ecp->cp = endp;
1882 		break;
1883 	case '\'':				/* Use a mark. */
1884 		*isaddrp = 1;
1885 		F_SET(ecp, E_ABSMARK);
1886 
1887 		if (ecp->clen == 1) {
1888 			msgq(sp, M_ERR, "095|No mark name supplied");
1889 			*errp = 1;
1890 			return (0);
1891 		}
1892 		if (mark_get(sp, ecp->cp[1], mp, M_ERR)) {
1893 			*errp = 1;
1894 			return (0);
1895 		}
1896 		ecp->cp += 2;
1897 		ecp->clen -= 2;
1898 		break;
1899 	case '\\':				/* Search: forward/backward. */
1900 		/*
1901 		 * !!!
1902 		 * I can't find any difference between // and \/ or between
1903 		 * ?? and \?.  Mark Horton doesn't remember there being any
1904 		 * difference.  C'est la vie.
1905 		 */
1906 		if (ecp->clen < 2 ||
1907 		    (ecp->cp[1] != '/' && ecp->cp[1] != '?')) {
1908 			msgq(sp, M_ERR, "096|\\ not followed by / or ?");
1909 			*errp = 1;
1910 			return (0);
1911 		}
1912 		++ecp->cp;
1913 		--ecp->clen;
1914 		sf = ecp->cp[0] == '/' ? f_search : b_search;
1915 		goto search;
1916 	case '/':				/* Search forward. */
1917 		sf = f_search;
1918 		goto search;
1919 	case '?':				/* Search backward. */
1920 		sf = b_search;
1921 
1922 search:		mp->lno = sp->lno;
1923 		mp->cno = sp->cno;
1924 		if (sf(sp, mp, mp, ecp->cp, ecp->clen, &endp,
1925 		    SEARCH_MSG | SEARCH_PARSE | SEARCH_SET |
1926 		    (F_ISSET(ecp, E_SEARCH_WMSG) ? SEARCH_WMSG : 0))) {
1927 			*errp = 1;
1928 			return (0);
1929 		}
1930 
1931 		/* Fix up the command pointers. */
1932 		ecp->clen -= (endp - ecp->cp);
1933 		ecp->cp = endp;
1934 
1935 		*isaddrp = 1;
1936 		F_SET(ecp, E_ABSMARK);
1937 		break;
1938 	case '.':				/* Current position. */
1939 		*isaddrp = 1;
1940 		mp->cno = sp->cno;
1941 
1942 		/* If an empty file, then '.' is 0, not 1. */
1943 		if (sp->lno == 1) {
1944 			if (db_last(sp, &mp->lno))
1945 				return (1);
1946 			if (mp->lno != 0)
1947 				mp->lno = 1;
1948 		} else
1949 			mp->lno = sp->lno;
1950 
1951 		/*
1952 		 * !!!
1953 		 * Historically, .<number> was the same as .+<number>, i.e.
1954 		 * the '+' could be omitted.  (This feature is found in ed
1955 		 * as well.)
1956 		 */
1957 		if (ecp->clen > 1 && ISDIGIT(ecp->cp[1]))
1958 			*ecp->cp = '+';
1959 		else {
1960 			++ecp->cp;
1961 			--ecp->clen;
1962 		}
1963 		break;
1964 	}
1965 
1966 	/* Skip trailing <blank>s. */
1967 	for (; ecp->clen > 0 &&
1968 	    cmdskip(ecp->cp[0]); ++ecp->cp, --ecp->clen);
1969 
1970 	/*
1971 	 * Evaluate any offset.  If no address yet found, the offset
1972 	 * is relative to ".".
1973 	 */
1974 	total = 0;
1975 	if (ecp->clen != 0 && (ISDIGIT(ecp->cp[0]) ||
1976 	    ecp->cp[0] == '+' || ecp->cp[0] == '-' ||
1977 	    ecp->cp[0] == '^')) {
1978 		if (!*isaddrp) {
1979 			*isaddrp = 1;
1980 			mp->lno = sp->lno;
1981 			mp->cno = sp->cno;
1982 		}
1983 		/*
1984 		 * Evaluate an offset, defined as:
1985 		 *
1986 		 *		[+-^<blank>]*[<blank>]*[0-9]*
1987 		 *
1988 		 * The rough translation is any number of signs, optionally
1989 		 * followed by numbers, or a number by itself, all <blank>
1990 		 * separated.
1991 		 *
1992 		 * !!!
1993 		 * All address offsets were additive, e.g. "2 2 3p" was the
1994 		 * same as "7p", or, "/ZZZ/ 2" was the same as "/ZZZ/+2".
1995 		 * Note, however, "2 /ZZZ/" was an error.  It was also legal
1996 		 * to insert signs without numbers, so "3 - 2" was legal, and
1997 		 * equal to 4.
1998 		 *
1999 		 * !!!
2000 		 * Offsets were historically permitted for any line address,
2001 		 * e.g. the command "1,2 copy 2 2 2 2" copied lines 1,2 after
2002 		 * line 8.
2003 		 *
2004 		 * !!!
2005 		 * Offsets were historically permitted for search commands,
2006 		 * and handled as addresses: "/pattern/2 2 2" was legal, and
2007 		 * referenced the 6th line after pattern.
2008 		 */
2009 		F_SET(ecp, E_DELTA);
2010 		for (;;) {
2011 			for (; ecp->clen > 0 && cmdskip(ecp->cp[0]);
2012 			    ++ecp->cp, --ecp->clen);
2013 			if (ecp->clen == 0 || (!ISDIGIT(ecp->cp[0]) &&
2014 			    ecp->cp[0] != '+' && ecp->cp[0] != '-' &&
2015 			    ecp->cp[0] != '^'))
2016 				break;
2017 			if (!ISDIGIT(ecp->cp[0]) &&
2018 			    !ISDIGIT(ecp->cp[1])) {
2019 				total += ecp->cp[0] == '+' ? 1 : -1;
2020 				--ecp->clen;
2021 				++ecp->cp;
2022 			} else {
2023 				if (ecp->cp[0] == '-' ||
2024 				    ecp->cp[0] == '^') {
2025 					++ecp->cp;
2026 					--ecp->clen;
2027 					isneg = 1;
2028 				} else
2029 					isneg = 0;
2030 
2031 				/* Get a signed long, add it to the total. */
2032 				if ((nret = nget_slong(&val,
2033 				    ecp->cp, &endp, 10)) != NUM_OK ||
2034 				    (nret = NADD_SLONG(sp,
2035 				    total, val)) != NUM_OK) {
2036 					ex_badaddr(sp, NULL, A_NOTSET, nret);
2037 					*errp = 1;
2038 					return (0);
2039 				}
2040 				total += isneg ? -val : val;
2041 				ecp->clen -= (endp - ecp->cp);
2042 				ecp->cp = endp;
2043 			}
2044 		}
2045 	}
2046 
2047 	/*
2048 	 * Any value less than 0 is an error.  Make sure that the new value
2049 	 * will fit into a recno_t.
2050 	 */
2051 	if (*isaddrp && total != 0) {
2052 		if (total < 0) {
2053 			if (-total > mp->lno) {
2054 				msgq(sp, M_ERR,
2055 			    "097|Reference to a line number less than 0");
2056 				*errp = 1;
2057 				return (0);
2058 			}
2059 		} else
2060 			if (!NPFITS(MAX_REC_NUMBER, mp->lno, total)) {
2061 				ex_badaddr(sp, NULL, A_NOTSET, NUM_OVER);
2062 				*errp = 1;
2063 				return (0);
2064 			}
2065 		mp->lno += total;
2066 	}
2067 	return (0);
2068 }
2069 
2070 
2071 /*
2072  * ex_load --
2073  *	Load up the next command, which may be an @ buffer or global command.
2074  */
2075 static int
2076 ex_load(SCR *sp)
2077 {
2078 	GS *gp;
2079 	EXCMD *ecp;
2080 	RANGE *rp;
2081 
2082 	F_CLR(sp, SC_EX_GLOBAL);
2083 
2084 	/*
2085 	 * Lose any exhausted commands.  We know that the first command
2086 	 * can't be an AGV command, which makes things a bit easier.
2087 	 */
2088 	for (gp = sp->gp;;) {
2089 		ecp = SLIST_FIRST(gp->ecq);
2090 
2091 		/* Discard the allocated source name as requested. */
2092 		if (F_ISSET(ecp, E_NAMEDISCARD))
2093 			free(ecp->if_name);
2094 
2095 		/*
2096 		 * If we're back to the original structure, leave it around,
2097 		 * since we've returned to the beginning of the command stack.
2098 		 */
2099 		if (ecp == &gp->excmd) {
2100 			ecp->if_name = NULL;
2101 			return (0);
2102 		}
2103 
2104 		/*
2105 		 * ecp->clen will be 0 for the first discarded command, but
2106 		 * may not be 0 for subsequent ones, e.g. if the original
2107 		 * command was ":g/xx/@a|s/b/c/", then when we discard the
2108 		 * command pushed on the stack by the @a, we have to resume
2109 		 * the global command which included the substitute command.
2110 		 */
2111 		if (ecp->clen != 0)
2112 			return (0);
2113 
2114 		/*
2115 		 * If it's an @, global or v command, we may need to continue
2116 		 * the command on a different line.
2117 		 */
2118 		if (FL_ISSET(ecp->agv_flags, AGV_ALL)) {
2119 			/* Discard any exhausted ranges. */
2120 			while ((rp = TAILQ_FIRST(ecp->rq)) != NULL)
2121 				if (rp->start > rp->stop) {
2122 					TAILQ_REMOVE(ecp->rq, rp, q);
2123 					free(rp);
2124 				} else
2125 					break;
2126 
2127 			/* If there's another range, continue with it. */
2128 			if (rp != NULL)
2129 				break;
2130 
2131 			/* If it's a global/v command, fix up the last line. */
2132 			if (FL_ISSET(ecp->agv_flags,
2133 			    AGV_GLOBAL | AGV_V) && ecp->range_lno != OOBLNO) {
2134 				if (db_exist(sp, ecp->range_lno))
2135 					sp->lno = ecp->range_lno;
2136 				else {
2137 					if (db_last(sp, &sp->lno))
2138 						return (1);
2139 					if (sp->lno == 0)
2140 						sp->lno = 1;
2141 				}
2142 			}
2143 			free(ecp->o_cp);
2144 		}
2145 
2146 		/* Discard the EXCMD. */
2147 		SLIST_REMOVE_HEAD(gp->ecq, q);
2148 		free(ecp);
2149 	}
2150 
2151 	/*
2152 	 * We only get here if it's an active @, global or v command.  Set
2153 	 * the current line number, and get a new copy of the command for
2154 	 * the parser.  Note, the original pointer almost certainly moved,
2155 	 * so we have play games.
2156 	 */
2157 	ecp->cp = ecp->o_cp;
2158 	MEMCPY(ecp->cp, ecp->cp + ecp->o_clen, ecp->o_clen);
2159 	ecp->clen = ecp->o_clen;
2160 	ecp->range_lno = sp->lno = rp->start++;
2161 
2162 	if (FL_ISSET(ecp->agv_flags, AGV_GLOBAL | AGV_V))
2163 		F_SET(sp, SC_EX_GLOBAL);
2164 	return (0);
2165 }
2166 
2167 /*
2168  * ex_discard --
2169  *	Discard any pending ex commands.
2170  */
2171 static int
2172 ex_discard(SCR *sp)
2173 {
2174 	GS *gp;
2175 	EXCMD *ecp;
2176 	RANGE *rp;
2177 
2178 	/*
2179 	 * We know the first command can't be an AGV command, so we don't
2180 	 * process it specially.  We do, however, nail the command itself.
2181 	 */
2182 	for (gp = sp->gp;;) {
2183 		ecp = SLIST_FIRST(gp->ecq);
2184 		if (F_ISSET(ecp, E_NAMEDISCARD))
2185 			free(ecp->if_name);
2186 		/* Reset the last command without dropping it. */
2187 		if (ecp == &gp->excmd)
2188 			break;
2189 		if (FL_ISSET(ecp->agv_flags, AGV_ALL)) {
2190 			while ((rp = TAILQ_FIRST(ecp->rq)) != NULL) {
2191 				TAILQ_REMOVE(ecp->rq, rp, q);
2192 				free(rp);
2193 			}
2194 			free(ecp->o_cp);
2195 		}
2196 		SLIST_REMOVE_HEAD(gp->ecq, q);
2197 		free(ecp);
2198 	}
2199 
2200 	ecp->if_name = NULL;
2201 	ecp->clen = 0;
2202 	return (0);
2203 }
2204 
2205 /*
2206  * ex_unknown --
2207  *	Display an unknown command name.
2208  */
2209 static void
2210 ex_unknown(SCR *sp, CHAR_T *cmd, size_t len)
2211 {
2212 	size_t blen;
2213 	CHAR_T *bp;
2214 
2215 	GET_SPACE_GOTOW(sp, bp, blen, len + 1);
2216 	bp[len] = '\0';
2217 	MEMCPY(bp, cmd, len);
2218 	msgq_wstr(sp, M_ERR, bp, "098|The %s command is unknown");
2219 	FREE_SPACEW(sp, bp, blen);
2220 
2221 alloc_err:
2222 	return;
2223 }
2224 
2225 /*
2226  * ex_is_abbrev -
2227  *	The vi text input routine needs to know if ex thinks this is an
2228  *	[un]abbreviate command, so it can turn off abbreviations.  See
2229  *	the usual ranting in the vi/v_txt_ev.c:txt_abbrev() routine.
2230  *
2231  * PUBLIC: int ex_is_abbrev(CHAR_T *, size_t);
2232  */
2233 int
2234 ex_is_abbrev(CHAR_T *name, size_t len)
2235 {
2236 	EXCMDLIST const *cp;
2237 
2238 	return ((cp = ex_comm_search(name, len)) != NULL &&
2239 	    (cp == &cmds[C_ABBR] || cp == &cmds[C_UNABBREVIATE]));
2240 }
2241 
2242 /*
2243  * ex_is_unmap -
2244  *	The vi text input routine needs to know if ex thinks this is an
2245  *	unmap command, so it can turn off input mapping.  See the usual
2246  *	ranting in the vi/v_txt_ev.c:txt_unmap() routine.
2247  *
2248  * PUBLIC: int ex_is_unmap(CHAR_T *, size_t);
2249  */
2250 int
2251 ex_is_unmap(CHAR_T *name, size_t len)
2252 {
2253 	EXCMDLIST const *cp;
2254 
2255 	/*
2256 	 * The command the vi input routines are really interested in
2257 	 * is "unmap!", not just unmap.
2258 	 */
2259 	if (name[len - 1] != '!')
2260 		return (0);
2261 	--len;
2262 	return ((cp = ex_comm_search(name, len)) != NULL &&
2263 	    cp == &cmds[C_UNMAP]);
2264 }
2265 
2266 /*
2267  * ex_comm_search --
2268  *	Search for a command name.
2269  */
2270 static EXCMDLIST const *
2271 ex_comm_search(CHAR_T *name, size_t len)
2272 {
2273 	EXCMDLIST const *cp;
2274 
2275 	for (cp = cmds; cp->name != NULL; ++cp) {
2276 		if (cp->name[0] > name[0])
2277 			return (NULL);
2278 		if (cp->name[0] != name[0])
2279 			continue;
2280 		if (STRLEN(cp->name) >= len &&
2281 			!MEMCMP(name, cp->name, len))
2282 			return (cp);
2283 	}
2284 	return (NULL);
2285 }
2286 
2287 /*
2288  * ex_badaddr --
2289  *	Display a bad address message.
2290  *
2291  * PUBLIC: void ex_badaddr
2292  * PUBLIC:   (SCR *, EXCMDLIST const *, enum badaddr, enum nresult);
2293  */
2294 void
2295 ex_badaddr(SCR *sp, const EXCMDLIST *cp, enum badaddr ba, enum nresult nret)
2296 {
2297 	recno_t lno;
2298 
2299 	switch (nret) {
2300 	case NUM_OK:
2301 		break;
2302 	case NUM_ERR:
2303 		msgq(sp, M_SYSERR, NULL);
2304 		return;
2305 	case NUM_OVER:
2306 		msgq(sp, M_ERR, "099|Address value overflow");
2307 		return;
2308 	case NUM_UNDER:
2309 		msgq(sp, M_ERR, "100|Address value underflow");
2310 		return;
2311 	}
2312 
2313 	/*
2314 	 * When encountering an address error, tell the user if there's no
2315 	 * underlying file, that's the real problem.
2316 	 */
2317 	if (sp->ep == NULL) {
2318 		ex_wemsg(sp, cp ? cp->name : NULL, EXM_NOFILEYET);
2319 		return;
2320 	}
2321 
2322 	switch (ba) {
2323 	case A_COMBO:
2324 		msgq(sp, M_ERR, "101|Illegal address combination");
2325 		break;
2326 	case A_EOF:
2327 		if (db_last(sp, &lno))
2328 			return;
2329 		if (lno != 0) {
2330 			msgq(sp, M_ERR,
2331 			    "102|Illegal address: only %lu lines in the file",
2332 			    (u_long)lno);
2333 			break;
2334 		}
2335 		/* FALLTHROUGH */
2336 	case A_EMPTY:
2337 		msgq(sp, M_ERR, "103|Illegal address: the file is empty");
2338 		break;
2339 	case A_NOTSET:
2340 		abort();
2341 		/* NOTREACHED */
2342 	case A_ZERO:
2343 		msgq_wstr(sp, M_ERR, cp->name,
2344 		    "104|The %s command doesn't permit an address of 0");
2345 		break;
2346 	}
2347 	return;
2348 }
2349 
2350 #if defined(DEBUG) && defined(COMLOG)
2351 /*
2352  * ex_comlog --
2353  *	Log ex commands.
2354  */
2355 static void
2356 ex_comlog(sp, ecp)
2357 	SCR *sp;
2358 	EXCMD *ecp;
2359 {
2360 	TRACE(sp, "ecmd: "WS, ecp->cmd->name);
2361 	if (ecp->addrcnt > 0) {
2362 		TRACE(sp, " a1 %d", ecp->addr1.lno);
2363 		if (ecp->addrcnt > 1)
2364 			TRACE(sp, " a2: %d", ecp->addr2.lno);
2365 	}
2366 	if (ecp->lineno)
2367 		TRACE(sp, " line %d", ecp->lineno);
2368 	if (ecp->flags)
2369 		TRACE(sp, " flags 0x%x", ecp->flags);
2370 	if (FL_ISSET(ecp->iflags, E_C_BUFFER))
2371 		TRACE(sp, " buffer "WC, ecp->buffer);
2372 	if (ecp->argc) {
2373 		int cnt;
2374 		for (cnt = 0; cnt < ecp->argc; ++cnt)
2375 			TRACE(sp, " arg %d: {"WS"}", cnt, ecp->argv[cnt]->bp);
2376 	}
2377 	TRACE(sp, "\n");
2378 }
2379 #endif
2380