read.c revision 1.4 
1/*	$Vendor-Id: read.c,v 1.26 2011/11/07 01:24:40 schwarze Exp $ */
2/*
3 * Copyright (c) 2008, 2009, 2010, 2011 Kristaps Dzonsons <kristaps@bsd.lv>
4 * Copyright (c) 2010, 2011 Ingo Schwarze <schwarze@openbsd.org>
5 *
6 * Permission to use, copy, modify, and distribute this software for any
7 * purpose with or without fee is hereby granted, provided that the above
8 * copyright notice and this permission notice appear in all copies.
9 *
10 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
11 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
12 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
13 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
14 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
15 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
16 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
17 */
18#ifdef HAVE_CONFIG_H
19#include "config.h"
20#endif
21
22#ifdef HAVE_MMAP
23# include <sys/stat.h>
24# include <sys/mman.h>
25#endif
26
27#include <assert.h>
28#include <ctype.h>
29#include <fcntl.h>
30#include <stdarg.h>
31#include <stdio.h>
32#include <stdlib.h>
33#include <string.h>
34#include <unistd.h>
35
36#include "mandoc.h"
37#include "libmandoc.h"
38#include "mdoc.h"
39#include "man.h"
40
41#ifndef MAP_FILE
42#define	MAP_FILE	0
43#endif
44
45#define	REPARSE_LIMIT	1000
46
47struct	buf {
48	char	 	 *buf; /* binary input buffer */
49	size_t		  sz; /* size of binary buffer */
50};
51
52struct	mparse {
53	enum mandoclevel  file_status; /* status of current parse */
54	enum mandoclevel  wlevel; /* ignore messages below this */
55	int		  line; /* line number in the file */
56	enum mparset	  inttype; /* which parser to use */
57	struct man	 *pman; /* persistent man parser */
58	struct mdoc	 *pmdoc; /* persistent mdoc parser */
59	struct man	 *man; /* man parser */
60	struct mdoc	 *mdoc; /* mdoc parser */
61	struct roff	 *roff; /* roff parser (!NULL) */
62	int		  reparse_count; /* finite interp. stack */
63	mandocmsg	  mmsg; /* warning/error message handler */
64	void		 *arg; /* argument to mmsg */
65	const char	 *file;
66	struct buf	 *secondary;
67};
68
69static	void	  resize_buf(struct buf *, size_t);
70static	void	  mparse_buf_r(struct mparse *, struct buf, int);
71static	void	  mparse_readfd_r(struct mparse *, int, const char *, int);
72static	void	  pset(const char *, int, struct mparse *);
73static	void	  pdesc(struct mparse *, const char *, int);
74static	int	  read_whole_file(const char *, int, struct buf *, int *);
75static	void	  mparse_end(struct mparse *);
76
77static	const enum mandocerr	mandoclimits[MANDOCLEVEL_MAX] = {
78	MANDOCERR_OK,
79	MANDOCERR_WARNING,
80	MANDOCERR_WARNING,
81	MANDOCERR_ERROR,
82	MANDOCERR_FATAL,
83	MANDOCERR_MAX,
84	MANDOCERR_MAX
85};
86
87static	const char * const	mandocerrs[MANDOCERR_MAX] = {
88	"ok",
89
90	"generic warning",
91
92	/* related to the prologue */
93	"no title in document",
94	"document title should be all caps",
95	"unknown manual section",
96	"date missing, using today's date",
97	"cannot parse date, using it verbatim",
98	"prologue macros out of order",
99	"duplicate prologue macro",
100	"macro not allowed in prologue",
101	"macro not allowed in body",
102
103	/* related to document structure */
104	".so is fragile, better use ln(1)",
105	"NAME section must come first",
106	"bad NAME section contents",
107	"manual name not yet set",
108	"sections out of conventional order",
109	"duplicate section name",
110	"section not in conventional manual section",
111
112	/* related to macros and nesting */
113	"skipping obsolete macro",
114	"skipping paragraph macro",
115	"skipping no-space macro",
116	"blocks badly nested",
117	"child violates parent syntax",
118	"nested displays are not portable",
119	"already in literal mode",
120	"line scope broken",
121
122	/* related to missing macro arguments */
123	"skipping empty macro",
124	"argument count wrong",
125	"missing display type",
126	"list type must come first",
127	"tag lists require a width argument",
128	"missing font type",
129	"skipping end of block that is not open",
130
131	/* related to bad macro arguments */
132	"skipping argument",
133	"duplicate argument",
134	"duplicate display type",
135	"duplicate list type",
136	"unknown AT&T UNIX version",
137	"bad Boolean value",
138	"unknown font",
139	"unknown standard specifier",
140	"bad width argument",
141
142	/* related to plain text */
143	"blank line in non-literal context",
144	"tab in non-literal context",
145	"end of line whitespace",
146	"bad comment style",
147	"bad escape sequence",
148	"unterminated quoted string",
149
150	/* related to equations */
151	"unexpected literal in equation",
152
153	"generic error",
154
155	/* related to equations */
156	"unexpected equation scope closure",
157	"equation scope open on exit",
158	"overlapping equation scopes",
159	"unexpected end of equation",
160	"equation syntax error",
161
162	/* related to tables */
163	"bad table syntax",
164	"bad table option",
165	"bad table layout",
166	"no table layout cells specified",
167	"no table data cells specified",
168	"ignore data in cell",
169	"data block still open",
170	"ignoring extra data cells",
171
172	"input stack limit exceeded, infinite loop?",
173	"skipping bad character",
174	"escaped character not allowed in a name",
175	"skipping text before the first section header",
176	"skipping unknown macro",
177	"NOT IMPLEMENTED, please use groff: skipping request",
178	"argument count wrong",
179	"skipping end of block that is not open",
180	"missing end of block",
181	"scope open on exit",
182	"uname(3) system call failed",
183	"macro requires line argument(s)",
184	"macro requires body argument(s)",
185	"macro requires argument(s)",
186	"missing list type",
187	"line argument(s) will be lost",
188	"body argument(s) will be lost",
189
190	"generic fatal error",
191
192	"not a manual",
193	"column syntax is inconsistent",
194	"NOT IMPLEMENTED: .Bd -file",
195	"argument count wrong, violates syntax",
196	"child violates parent syntax",
197	"argument count wrong, violates syntax",
198	"NOT IMPLEMENTED: .so with absolute path or \"..\"",
199	"no document body",
200	"no document prologue",
201	"static buffer exhausted",
202};
203
204static	const char * const	mandoclevels[MANDOCLEVEL_MAX] = {
205	"SUCCESS",
206	"RESERVED",
207	"WARNING",
208	"ERROR",
209	"FATAL",
210	"BADARG",
211	"SYSERR"
212};
213
214static void
215resize_buf(struct buf *buf, size_t initial)
216{
217
218	buf->sz = buf->sz > initial/2 ? 2 * buf->sz : initial;
219	buf->buf = mandoc_realloc(buf->buf, buf->sz);
220}
221
222static void
223pset(const char *buf, int pos, struct mparse *curp)
224{
225	int		 i;
226
227	/*
228	 * Try to intuit which kind of manual parser should be used.  If
229	 * passed in by command-line (-man, -mdoc), then use that
230	 * explicitly.  If passed as -mandoc, then try to guess from the
231	 * line: either skip dot-lines, use -mdoc when finding `.Dt', or
232	 * default to -man, which is more lenient.
233	 *
234	 * Separate out pmdoc/pman from mdoc/man: the first persists
235	 * through all parsers, while the latter is used per-parse.
236	 */
237
238	if ('.' == buf[0] || '\'' == buf[0]) {
239		for (i = 1; buf[i]; i++)
240			if (' ' != buf[i] && '\t' != buf[i])
241				break;
242		if ('\0' == buf[i])
243			return;
244	}
245
246	switch (curp->inttype) {
247	case (MPARSE_MDOC):
248		if (NULL == curp->pmdoc)
249			curp->pmdoc = mdoc_alloc(curp->roff, curp);
250		assert(curp->pmdoc);
251		curp->mdoc = curp->pmdoc;
252		return;
253	case (MPARSE_MAN):
254		if (NULL == curp->pman)
255			curp->pman = man_alloc(curp->roff, curp);
256		assert(curp->pman);
257		curp->man = curp->pman;
258		return;
259	default:
260		break;
261	}
262
263	if (pos >= 3 && 0 == memcmp(buf, ".Dd", 3))  {
264		if (NULL == curp->pmdoc)
265			curp->pmdoc = mdoc_alloc(curp->roff, curp);
266		assert(curp->pmdoc);
267		curp->mdoc = curp->pmdoc;
268		return;
269	}
270
271	if (NULL == curp->pman)
272		curp->pman = man_alloc(curp->roff, curp);
273	assert(curp->pman);
274	curp->man = curp->pman;
275}
276
277/*
278 * Main parse routine for an opened file.  This is called for each
279 * opened file and simply loops around the full input file, possibly
280 * nesting (i.e., with `so').
281 */
282static void
283mparse_buf_r(struct mparse *curp, struct buf blk, int start)
284{
285	const struct tbl_span	*span;
286	struct buf	 ln;
287	enum rofferr	 rr;
288	int		 i, of, rc;
289	int		 pos; /* byte number in the ln buffer */
290	int		 lnn; /* line number in the real file */
291	unsigned char	 c;
292
293	memset(&ln, 0, sizeof(struct buf));
294
295	lnn = curp->line;
296	pos = 0;
297
298	for (i = 0; i < (int)blk.sz; ) {
299		if (0 == pos && '\0' == blk.buf[i])
300			break;
301
302		if (start) {
303			curp->line = lnn;
304			curp->reparse_count = 0;
305		}
306
307		while (i < (int)blk.sz && (start || '\0' != blk.buf[i])) {
308
309			/*
310			 * When finding an unescaped newline character,
311			 * leave the character loop to process the line.
312			 * Skip a preceding carriage return, if any.
313			 */
314
315			if ('\r' == blk.buf[i] && i + 1 < (int)blk.sz &&
316			    '\n' == blk.buf[i + 1])
317				++i;
318			if ('\n' == blk.buf[i]) {
319				++i;
320				++lnn;
321				break;
322			}
323
324			/*
325			 * Warn about bogus characters.  If you're using
326			 * non-ASCII encoding, you're screwing your
327			 * readers.  Since I'd rather this not happen,
328			 * I'll be helpful and drop these characters so
329			 * we don't display gibberish.  Note to manual
330			 * writers: use special characters.
331			 */
332
333			c = (unsigned char) blk.buf[i];
334
335			if ( ! (isascii(c) &&
336					(isgraph(c) || isblank(c)))) {
337				mandoc_msg(MANDOCERR_BADCHAR, curp,
338						curp->line, pos, "ignoring byte");
339				i++;
340				continue;
341			}
342
343			/* Trailing backslash = a plain char. */
344
345			if ('\\' != blk.buf[i] || i + 1 == (int)blk.sz) {
346				if (pos >= (int)ln.sz)
347					resize_buf(&ln, 256);
348				ln.buf[pos++] = blk.buf[i++];
349				continue;
350			}
351
352			/*
353			 * Found escape and at least one other character.
354			 * When it's a newline character, skip it.
355			 * When there is a carriage return in between,
356			 * skip that one as well.
357			 */
358
359			if ('\r' == blk.buf[i + 1] && i + 2 < (int)blk.sz &&
360			    '\n' == blk.buf[i + 2])
361				++i;
362			if ('\n' == blk.buf[i + 1]) {
363				i += 2;
364				++lnn;
365				continue;
366			}
367
368			if ('"' == blk.buf[i + 1] || '#' == blk.buf[i + 1]) {
369				i += 2;
370				/* Comment, skip to end of line */
371				for (; i < (int)blk.sz; ++i) {
372					if ('\n' == blk.buf[i]) {
373						++i;
374						++lnn;
375						break;
376					}
377				}
378
379				/* Backout trailing whitespaces */
380				for (; pos > 0; --pos) {
381					if (ln.buf[pos - 1] != ' ')
382						break;
383					if (pos > 2 && ln.buf[pos - 2] == '\\')
384						break;
385				}
386				break;
387			}
388
389			/* Some other escape sequence, copy & cont. */
390
391			if (pos + 1 >= (int)ln.sz)
392				resize_buf(&ln, 256);
393
394			ln.buf[pos++] = blk.buf[i++];
395			ln.buf[pos++] = blk.buf[i++];
396		}
397
398 		if (pos >= (int)ln.sz)
399			resize_buf(&ln, 256);
400
401		ln.buf[pos] = '\0';
402
403		/*
404		 * A significant amount of complexity is contained by
405		 * the roff preprocessor.  It's line-oriented but can be
406		 * expressed on one line, so we need at times to
407		 * readjust our starting point and re-run it.  The roff
408		 * preprocessor can also readjust the buffers with new
409		 * data, so we pass them in wholesale.
410		 */
411
412		of = 0;
413
414		/*
415		 * Maintain a lookaside buffer of all parsed lines.  We
416		 * only do this if mparse_keep() has been invoked (the
417		 * buffer may be accessed with mparse_getkeep()).
418		 */
419
420		if (curp->secondary) {
421			curp->secondary->buf =
422				mandoc_realloc
423				(curp->secondary->buf,
424				 curp->secondary->sz + pos + 2);
425			memcpy(curp->secondary->buf +
426					curp->secondary->sz,
427					ln.buf, pos);
428			curp->secondary->sz += pos;
429			curp->secondary->buf
430				[curp->secondary->sz] = '\n';
431			curp->secondary->sz++;
432			curp->secondary->buf
433				[curp->secondary->sz] = '\0';
434		}
435rerun:
436		rr = roff_parseln
437			(curp->roff, curp->line,
438			 &ln.buf, &ln.sz, of, &of);
439
440		switch (rr) {
441		case (ROFF_REPARSE):
442			if (REPARSE_LIMIT >= ++curp->reparse_count)
443				mparse_buf_r(curp, ln, 0);
444			else
445				mandoc_msg(MANDOCERR_ROFFLOOP, curp,
446					curp->line, pos, NULL);
447			pos = 0;
448			continue;
449		case (ROFF_APPEND):
450			pos = (int)strlen(ln.buf);
451			continue;
452		case (ROFF_RERUN):
453			goto rerun;
454		case (ROFF_IGN):
455			pos = 0;
456			continue;
457		case (ROFF_ERR):
458			assert(MANDOCLEVEL_FATAL <= curp->file_status);
459			break;
460		case (ROFF_SO):
461			/*
462			 * We remove `so' clauses from our lookaside
463			 * buffer because we're going to descend into
464			 * the file recursively.
465			 */
466			if (curp->secondary)
467				curp->secondary->sz -= pos + 1;
468			mparse_readfd_r(curp, -1, ln.buf + of, 1);
469			if (MANDOCLEVEL_FATAL <= curp->file_status)
470				break;
471			pos = 0;
472			continue;
473		default:
474			break;
475		}
476
477		/*
478		 * If we encounter errors in the recursive parse, make
479		 * sure we don't continue parsing.
480		 */
481
482		if (MANDOCLEVEL_FATAL <= curp->file_status)
483			break;
484
485		/*
486		 * If input parsers have not been allocated, do so now.
487		 * We keep these instanced between parsers, but set them
488		 * locally per parse routine since we can use different
489		 * parsers with each one.
490		 */
491
492		if ( ! (curp->man || curp->mdoc))
493			pset(ln.buf + of, pos - of, curp);
494
495		/*
496		 * Lastly, push down into the parsers themselves.  One
497		 * of these will have already been set in the pset()
498		 * routine.
499		 * If libroff returns ROFF_TBL, then add it to the
500		 * currently open parse.  Since we only get here if
501		 * there does exist data (see tbl_data.c), we're
502		 * guaranteed that something's been allocated.
503		 * Do the same for ROFF_EQN.
504		 */
505
506		rc = -1;
507
508		if (ROFF_TBL == rr)
509			while (NULL != (span = roff_span(curp->roff))) {
510				rc = curp->man ?
511					man_addspan(curp->man, span) :
512					mdoc_addspan(curp->mdoc, span);
513				if (0 == rc)
514					break;
515			}
516		else if (ROFF_EQN == rr)
517			rc = curp->mdoc ?
518				mdoc_addeqn(curp->mdoc,
519					roff_eqn(curp->roff)) :
520				man_addeqn(curp->man,
521					roff_eqn(curp->roff));
522		else if (curp->man || curp->mdoc)
523			rc = curp->man ?
524				man_parseln(curp->man,
525					curp->line, ln.buf, of) :
526				mdoc_parseln(curp->mdoc,
527					curp->line, ln.buf, of);
528
529		if (0 == rc) {
530			assert(MANDOCLEVEL_FATAL <= curp->file_status);
531			break;
532		}
533
534		/* Temporary buffers typically are not full. */
535
536		if (0 == start && '\0' == blk.buf[i])
537			break;
538
539		/* Start the next input line. */
540
541		pos = 0;
542	}
543
544	free(ln.buf);
545}
546
547static void
548pdesc(struct mparse *curp, const char *file, int fd)
549{
550	struct buf	 blk;
551	int		 with_mmap;
552
553	/*
554	 * Run for each opened file; may be called more than once for
555	 * each full parse sequence if the opened file is nested (i.e.,
556	 * from `so').  Simply sucks in the whole file and moves into
557	 * the parse phase for the file.
558	 */
559
560	if ( ! read_whole_file(file, fd, &blk, &with_mmap)) {
561		curp->file_status = MANDOCLEVEL_SYSERR;
562		return;
563	}
564
565	/* Line number is per-file. */
566
567	curp->line = 1;
568
569	mparse_buf_r(curp, blk, 1);
570
571#ifdef	HAVE_MMAP
572	if (with_mmap)
573		munmap(blk.buf, blk.sz);
574	else
575#endif
576		free(blk.buf);
577}
578
579static int
580read_whole_file(const char *file, int fd, struct buf *fb, int *with_mmap)
581{
582	size_t		 off;
583	ssize_t		 ssz;
584
585#ifdef	HAVE_MMAP
586	struct stat	 st;
587	if (-1 == fstat(fd, &st)) {
588		perror(file);
589		return(0);
590	}
591
592	/*
593	 * If we're a regular file, try just reading in the whole entry
594	 * via mmap().  This is faster than reading it into blocks, and
595	 * since each file is only a few bytes to begin with, I'm not
596	 * concerned that this is going to tank any machines.
597	 */
598
599	if (S_ISREG(st.st_mode)) {
600		if (st.st_size >= (1U << 31)) {
601			fprintf(stderr, "%s: input too large\n", file);
602			return(0);
603		}
604		*with_mmap = 1;
605		fb->sz = (size_t)st.st_size;
606		fb->buf = mmap(NULL, fb->sz, PROT_READ,
607				MAP_FILE|MAP_SHARED, fd, 0);
608		if (fb->buf != MAP_FAILED)
609			return(1);
610	}
611#endif
612
613	/*
614	 * If this isn't a regular file (like, say, stdin), then we must
615	 * go the old way and just read things in bit by bit.
616	 */
617
618	*with_mmap = 0;
619	off = 0;
620	fb->sz = 0;
621	fb->buf = NULL;
622	for (;;) {
623		if (off == fb->sz) {
624			if (fb->sz == (1U << 31)) {
625				fprintf(stderr, "%s: input too large\n", file);
626				break;
627			}
628			resize_buf(fb, 65536);
629		}
630		ssz = read(fd, fb->buf + (int)off, fb->sz - off);
631		if (ssz == 0) {
632			fb->sz = off;
633			return(1);
634		}
635		if (ssz == -1) {
636			perror(file);
637			break;
638		}
639		off += (size_t)ssz;
640	}
641
642	free(fb->buf);
643	fb->buf = NULL;
644	return(0);
645}
646
647static void
648mparse_end(struct mparse *curp)
649{
650
651	if (MANDOCLEVEL_FATAL <= curp->file_status)
652		return;
653
654	if (curp->mdoc && ! mdoc_endparse(curp->mdoc)) {
655		assert(MANDOCLEVEL_FATAL <= curp->file_status);
656		return;
657	}
658
659	if (curp->man && ! man_endparse(curp->man)) {
660		assert(MANDOCLEVEL_FATAL <= curp->file_status);
661		return;
662	}
663
664	if ( ! (curp->man || curp->mdoc)) {
665		mandoc_msg(MANDOCERR_NOTMANUAL, curp, 1, 0, NULL);
666		curp->file_status = MANDOCLEVEL_FATAL;
667		return;
668	}
669
670	roff_endparse(curp->roff);
671}
672
673static void
674mparse_readfd_r(struct mparse *curp, int fd, const char *file, int re)
675{
676	const char	*svfile;
677
678	if (-1 == fd)
679		if (-1 == (fd = open(file, O_RDONLY, 0))) {
680			perror(file);
681			curp->file_status = MANDOCLEVEL_SYSERR;
682			return;
683		}
684
685	svfile = curp->file;
686	curp->file = file;
687
688	pdesc(curp, file, fd);
689
690	if (0 == re && MANDOCLEVEL_FATAL > curp->file_status)
691		mparse_end(curp);
692
693	if (STDIN_FILENO != fd && -1 == close(fd))
694		perror(file);
695
696	curp->file = svfile;
697}
698
699enum mandoclevel
700mparse_readfd(struct mparse *curp, int fd, const char *file)
701{
702
703	mparse_readfd_r(curp, fd, file, 0);
704	return(curp->file_status);
705}
706
707struct mparse *
708mparse_alloc(enum mparset inttype, enum mandoclevel wlevel, mandocmsg mmsg, void *arg)
709{
710	struct mparse	*curp;
711
712	assert(wlevel <= MANDOCLEVEL_FATAL);
713
714	curp = mandoc_calloc(1, sizeof(struct mparse));
715
716	curp->wlevel = wlevel;
717	curp->mmsg = mmsg;
718	curp->arg = arg;
719	curp->inttype = inttype;
720
721	curp->roff = roff_alloc(curp);
722	return(curp);
723}
724
725void
726mparse_reset(struct mparse *curp)
727{
728
729	roff_reset(curp->roff);
730
731	if (curp->mdoc)
732		mdoc_reset(curp->mdoc);
733	if (curp->man)
734		man_reset(curp->man);
735	if (curp->secondary)
736		curp->secondary->sz = 0;
737
738	curp->file_status = MANDOCLEVEL_OK;
739	curp->mdoc = NULL;
740	curp->man = NULL;
741}
742
743void
744mparse_free(struct mparse *curp)
745{
746
747	if (curp->pmdoc)
748		mdoc_free(curp->pmdoc);
749	if (curp->pman)
750		man_free(curp->pman);
751	if (curp->roff)
752		roff_free(curp->roff);
753	if (curp->secondary)
754		free(curp->secondary->buf);
755
756	free(curp->secondary);
757	free(curp);
758}
759
760void
761mparse_result(struct mparse *curp, struct mdoc **mdoc, struct man **man)
762{
763
764	if (mdoc)
765		*mdoc = curp->mdoc;
766	if (man)
767		*man = curp->man;
768}
769
770void
771mandoc_vmsg(enum mandocerr t, struct mparse *m,
772		int ln, int pos, const char *fmt, ...)
773{
774	char		 buf[256];
775	va_list		 ap;
776
777	va_start(ap, fmt);
778	vsnprintf(buf, sizeof(buf) - 1, fmt, ap);
779	va_end(ap);
780
781	mandoc_msg(t, m, ln, pos, buf);
782}
783
784void
785mandoc_msg(enum mandocerr er, struct mparse *m,
786		int ln, int col, const char *msg)
787{
788	enum mandoclevel level;
789
790	level = MANDOCLEVEL_FATAL;
791	while (er < mandoclimits[level])
792		level--;
793
794	if (level < m->wlevel)
795		return;
796
797	if (m->mmsg)
798		(*m->mmsg)(er, level, m->file, ln, col, msg);
799
800	if (m->file_status < level)
801		m->file_status = level;
802}
803
804const char *
805mparse_strerror(enum mandocerr er)
806{
807
808	return(mandocerrs[er]);
809}
810
811const char *
812mparse_strlevel(enum mandoclevel lvl)
813{
814	return(mandoclevels[lvl]);
815}
816
817void
818mparse_keep(struct mparse *p)
819{
820
821	assert(NULL == p->secondary);
822	p->secondary = mandoc_calloc(1, sizeof(struct buf));
823}
824
825const char *
826mparse_getkeep(const struct mparse *p)
827{
828
829	assert(p->secondary);
830	return(p->secondary->sz ? p->secondary->buf : NULL);
831}
832