Split the document parsing sequence out of main.c and into read.c,

putting the interface into mandoc.h.  This effectively makes the
function of main.c be command-line handling, invoking the parser, and
sending its output to the output handler.  The sequence of parsing
(pfile(), pdesc(), etc.) has changed very little but for clean-up of
some state variables (curp->fd, etc.).

4 files changed, 709 insertions, 608 deletions

diff --git a/Makefile b/Makefile
index 72f6e31a..d4874058 100644
--- a/Makefile
+++ b/Makefile
@@ -66,15 +66,15 @@ MANSRCS	   = man_macro.c man.c man_hash.c man_validate.c \
 MAINLNS	   = main.ln mdoc_term.ln chars.ln term.ln tree.ln \
 	     compat.ln man_term.ln html.ln mdoc_html.ln \
 	     man_html.ln out.ln term_ps.ln term_ascii.ln \
-	     tbl_term.ln tbl_html.ln
+	     tbl_term.ln tbl_html.ln read.ln
 
 MAINOBJS   = main.o mdoc_term.o chars.o term.o tree.o compat.o \
 	     man_term.o html.o mdoc_html.o man_html.o out.o \
-	     term_ps.o term_ascii.o tbl_term.o tbl_html.o
+	     term_ps.o term_ascii.o tbl_term.o tbl_html.o read.o
 
 MAINSRCS   = main.c mdoc_term.c chars.c term.c tree.c compat.c \
 	     man_term.c html.c mdoc_html.c man_html.c out.c \
-	     term_ps.c term_ascii.c tbl_term.c tbl_html.c
+	     term_ps.c term_ascii.c tbl_term.c tbl_html.c read.c
 
 LLNS	   = llib-llibmdoc.ln llib-llibman.ln llib-lmandoc.ln \
 	     llib-llibmandoc.ln llib-llibroff.ln
diff --git a/main.c b/main.c
index fc0ccb95..4aac64e5 100644
--- a/main.c
+++ b/main.c
@@ -19,12 +19,7 @@
 #include "config.h"
 #endif
 
-#include <sys/mman.h>
-#include <sys/stat.h>
-
 #include <assert.h>
-#include <ctype.h>
-#include <fcntl.h>
 #include <stdio.h>
 #include <stdint.h>
 #include <stdlib.h>
@@ -37,14 +32,6 @@
 #include "man.h"
 #include "roff.h"
 
-#ifndef MAP_FILE
-#define	MAP_FILE	0
-#endif
-
-#define	REPARSE_LIMIT	1000
-
-/* FIXME: Intel's compiler?  LLVM?  pcc?  */
-
 #if !defined(__GNUC__) || (__GNUC__ < 2)
 # if !defined(lint)
 #  define __attribute__(x)
@@ -55,43 +42,21 @@ typedef	void		(*out_mdoc)(void *, const struct mdoc *);
 typedef	void		(*out_man)(void *, const struct man *);
 typedef	void		(*out_free)(void *);
 
-struct	buf {
-	char	 	 *buf;
-	size_t		  sz;
-};
-
-enum	intt {
-	INTT_AUTO,
-	INTT_MDOC,
-	INTT_MAN
-};
-
 enum	outt {
-	OUTT_ASCII = 0,
-	OUTT_TREE,
-	OUTT_HTML,
-	OUTT_XHTML,
-	OUTT_LINT,
-	OUTT_PS,
-	OUTT_PDF
+	OUTT_ASCII = 0,	/* -Tascii */
+	OUTT_TREE,	/* -Ttree */
+	OUTT_HTML,	/* -Thtml */
+	OUTT_XHTML,	/* -Txhtml */
+	OUTT_LINT,	/* -Tlint */
+	OUTT_PS,	/* -Tps */
+	OUTT_PDF	/* -Tpdf */
 };
 
 struct	curparse {
-	enum mandoclevel  exit_status;	/* status of all file parses */
+	struct mparse	 *mp;
 	const char	 *file;		/* current file-name */
-	enum mandoclevel  file_status;	/* error status of current parse */
-	int		  fd;		/* current file-descriptor */
-	int		  line;		/* line number in the file */
 	enum mandoclevel  wlevel;	/* ignore messages below this */
 	int		  wstop;	/* stop after a file with a warning */
-	enum intt	  inttype;	/* which parser to use */
-	struct man	 *pman;		/* persistent man parser */
-	struct mdoc	 *pmdoc;	/* persistent mdoc parser */
-	struct man	 *man;		/* man parser */
-	struct mdoc	 *mdoc;		/* mdoc parser */
-	struct roff	 *roff;		/* roff parser (!NULL) */
-	struct regset	  regs;		/* roff registers */
-	int		  reparse_count; /* finite interpolation stack */
 	enum outt	  outtype; 	/* which output to use */
 	out_mdoc	  outmdoc;	/* mdoc output ptr */
 	out_man	  	  outman;	/* man output ptr */
@@ -227,15 +192,13 @@ static	const char * const	mandocerrs[MANDOCERR_MAX] = {
 	"static buffer exhausted",
 };
 
-static	void		  parsebuf(struct curparse *, struct buf, int);
-static	void		  pdesc(struct curparse *);
-static	void		  fdesc(struct curparse *);
-static	void		  ffile(const char *, struct curparse *);
-static	int		  pfile(const char *, struct curparse *);
-static	int		  moptions(enum intt *, char *);
+static	void		  evt_close(void *, const char *);
+static	int		  evt_open(void *, const char *);
+static	int		  moptions(enum mparset *, char *);
 static	void		  mmsg(enum mandocerr, void *, 
 				int, int, const char *);
-static	void		  pset(const char *, int, struct curparse *);
+static	void		  parse(struct curparse *, int, 
+				const char *, enum mandoclevel *);
 static	int		  toptions(struct curparse *, char *);
 static	void		  usage(void) __attribute__((noreturn));
 static	void		  version(void) __attribute__((noreturn));
@@ -248,6 +211,8 @@ main(int argc, char *argv[])
 {
 	int		 c;
 	struct curparse	 curp;
+	enum mparset	 type;
+	enum mandoclevel rc;
 
 	progname = strrchr(argv[0], '/');
 	if (progname == NULL)
@@ -257,16 +222,15 @@ main(int argc, char *argv[])
 
 	memset(&curp, 0, sizeof(struct curparse));
 
-	curp.inttype = INTT_AUTO;
+	type = MPARSE_AUTO;
 	curp.outtype = OUTT_ASCII;
 	curp.wlevel  = MANDOCLEVEL_FATAL;
-	curp.exit_status = MANDOCLEVEL_OK;
 
 	/* LINTED */
 	while (-1 != (c = getopt(argc, argv, "m:O:T:VW:")))
 		switch (c) {
 		case ('m'):
-			if ( ! moptions(&curp.inttype, optarg))
+			if ( ! moptions(&type, optarg))
 				return((int)MANDOCLEVEL_BADARG);
 			break;
 		case ('O'):
@@ -289,50 +253,44 @@ main(int argc, char *argv[])
 			/* NOTREACHED */
 		}
 
+	curp.mp = mparse_alloc(type, evt_open, evt_close, mmsg, &curp);
+
 	argc -= optind;
 	argv += optind;
 
-	if (NULL == *argv) {
-		curp.file = "<stdin>";
-		curp.fd = STDIN_FILENO;
+	rc = MANDOCLEVEL_OK;
 
-		fdesc(&curp);
-	}
+	if (NULL == *argv)
+		parse(&curp, STDIN_FILENO, "<stdin>", &rc);
 
 	while (*argv) {
-		ffile(*argv, &curp);
-		if (MANDOCLEVEL_OK != curp.exit_status && curp.wstop)
+		parse(&curp, -1, *argv, &rc);
+		if (MANDOCLEVEL_OK != rc && curp.wstop)
 			break;
 		++argv;
 	}
 
 	if (curp.outfree)
 		(*curp.outfree)(curp.outdata);
-	if (curp.pmdoc)
-		mdoc_free(curp.pmdoc);
-	if (curp.pman)
-		man_free(curp.pman);
-	if (curp.roff)
-		roff_free(curp.roff);
-
-	return((int)curp.exit_status);
-}
+	if (curp.mp)
+		mparse_free(curp.mp);
 
+	return((int)rc);
+}
 
 static void
 version(void)
 {
 
-	(void)printf("%s %s\n", progname, VERSION);
+	printf("%s %s\n", progname, VERSION);
 	exit((int)MANDOCLEVEL_OK);
 }
 
-
 static void
 usage(void)
 {
 
-	(void)fprintf(stderr, "usage: %s "
+	fprintf(stderr, "usage: %s "
 			"[-V] "
 			"[-foption] "
 			"[-mformat] "
@@ -345,202 +303,49 @@ usage(void)
 	exit((int)MANDOCLEVEL_BADARG);
 }
 
-static void
-ffile(const char *file, struct curparse *curp)
-{
-
-	/*
-	 * Called once per input file.  Get the file ready for reading,
-	 * pass it through to the parser-driver, then close it out.
-	 * XXX: don't do anything special as this is only called for
-	 * files; stdin goes directly to fdesc().
-	 */
-
-	curp->file = file;
-
-	if (-1 == (curp->fd = open(curp->file, O_RDONLY, 0))) {
-		perror(curp->file);
-		curp->exit_status = MANDOCLEVEL_SYSERR;
-		return;
-	}
-
-	fdesc(curp);
-
-	if (-1 == close(curp->fd))
-		perror(curp->file);
-}
-
 static int
-pfile(const char *file, struct curparse *curp)
+evt_open(void *arg, const char *file)
 {
-	const char	*savefile;
-	int		 fd, savefd;
 
-	if (-1 == (fd = open(file, O_RDONLY, 0))) {
-		perror(file);
-		curp->file_status = MANDOCLEVEL_SYSERR;
-		return(0);
-	}
-
-	savefile = curp->file;
-	savefd = curp->fd;
-
-	curp->file = file;
-	curp->fd = fd;
-
-	pdesc(curp);
-
-	curp->file = savefile;
-	curp->fd = savefd;
-
-	if (-1 == close(fd))
-		perror(file);
-
-	return(MANDOCLEVEL_FATAL > curp->file_status ? 1 : 0);
+	evt_close(arg, file);
+	return(1);
 }
 
-
 static void
-resize_buf(struct buf *buf, size_t initial)
-{
-
-	buf->sz = buf->sz > initial/2 ? 2 * buf->sz : initial;
-	buf->buf = mandoc_realloc(buf->buf, buf->sz);
-}
-
-
-static int
-read_whole_file(struct curparse *curp, struct buf *fb, int *with_mmap)
+evt_close(void *arg, const char *file)
 {
-	struct stat	 st;
-	size_t		 off;
-	ssize_t		 ssz;
-
-	if (-1 == fstat(curp->fd, &st)) {
-		perror(curp->file);
-		return(0);
-	}
-
-	/*
-	 * If we're a regular file, try just reading in the whole entry
-	 * via mmap().  This is faster than reading it into blocks, and
-	 * since each file is only a few bytes to begin with, I'm not
-	 * concerned that this is going to tank any machines.
-	 */
-
-	if (S_ISREG(st.st_mode)) {
-		if (st.st_size >= (1U << 31)) {
-			fprintf(stderr, "%s: input too large\n", 
-					curp->file);
-			return(0);
-		}
-		*with_mmap = 1;
-		fb->sz = (size_t)st.st_size;
-		fb->buf = mmap(NULL, fb->sz, PROT_READ, 
-				MAP_FILE|MAP_SHARED, curp->fd, 0);
-		if (fb->buf != MAP_FAILED)
-			return(1);
-	}
-
-	/*
-	 * If this isn't a regular file (like, say, stdin), then we must
-	 * go the old way and just read things in bit by bit.
-	 */
-
-	*with_mmap = 0;
-	off = 0;
-	fb->sz = 0;
-	fb->buf = NULL;
-	for (;;) {
-		if (off == fb->sz) {
-			if (fb->sz == (1U << 31)) {
-				fprintf(stderr, "%s: input too large\n", 
-						curp->file);
-				break;
-			}
-			resize_buf(fb, 65536);
-		}
-		ssz = read(curp->fd, fb->buf + (int)off, fb->sz - off);
-		if (ssz == 0) {
-			fb->sz = off;
-			return(1);
-		}
-		if (ssz == -1) {
-			perror(curp->file);
-			break;
-		}
-		off += (size_t)ssz;
-	}
+	struct curparse	*p;
 
-	free(fb->buf);
-	fb->buf = NULL;
-	return(0);
+	p = (struct curparse *)arg;
+	p->file = file;
 }
 
-
 static void
-fdesc(struct curparse *curp)
+parse(struct curparse *curp, int fd, 
+		const char *file, enum mandoclevel *level)
 {
+	enum mandoclevel  rc;
+	struct mdoc	 *mdoc;
+	struct man	 *man;
 
-	/*
-	 * Called once per file with an opened file descriptor.  All
-	 * pre-file-parse operations (whether stdin or a file) should go
-	 * here.
-	 *
-	 * This calls down into the nested parser, which drills down and
-	 * fully parses a file and all its dependences (i.e., `so').  It
-	 * then runs the cleanup validators and pushes to output.
-	 */
-
-	/* Zero the parse type. */
-
-	curp->mdoc = NULL;
-	curp->man = NULL;
-	curp->file_status = MANDOCLEVEL_OK;
+	/* Begin by parsing the file itself. */
 
-	/* Make sure the mandotory roff parser is initialised. */
-
-	if (NULL == curp->roff) {
-		curp->roff = roff_alloc(&curp->regs, curp, mmsg);
-		assert(curp->roff);
-	}
-
-	/* Fully parse the file. */
-
-	pdesc(curp);
-
-	if (MANDOCLEVEL_FATAL <= curp->file_status)
-		goto cleanup;
-
-	/* NOTE a parser may not have been assigned, yet. */
-
-	if ( ! (curp->man || curp->mdoc)) {
-		fprintf(stderr, "%s: Not a manual\n", curp->file);
-		curp->file_status = MANDOCLEVEL_FATAL;
-		goto cleanup;
-	}
+	assert(file);
+	assert(fd >= -1);
 
-	/* Clean up the parse routine ASTs. */
+	rc = mparse_readfd(curp->mp, fd, file);
 
-	if (curp->mdoc && ! mdoc_endparse(curp->mdoc)) {
-		assert(MANDOCLEVEL_FATAL <= curp->file_status);
-		goto cleanup;
-	}
+	/* Stop immediately if the parse has failed. */
 
-	if (curp->man && ! man_endparse(curp->man)) {
-		assert(MANDOCLEVEL_FATAL <= curp->file_status);
+	if (MANDOCLEVEL_FATAL <= rc)
 		goto cleanup;
-	}
-
-	assert(curp->roff);
-	roff_endparse(curp->roff);
 
 	/*
-	 * With -Wstop and warnings or errors of at least
-	 * the requested level, do not produce output.
+	 * With -Wstop and warnings or errors of at least the requested
+	 * level, do not produce output.
 	 */
 
-	if (MANDOCLEVEL_OK != curp->file_status && curp->wstop)
+	if (MANDOCLEVEL_OK != rc && curp->wstop)
 		goto cleanup;
 
 	/* If unset, allocate output dev now (if applicable). */
@@ -594,373 +399,33 @@ fdesc(struct curparse *curp)
 		}
 	}
 
+	mparse_result(curp->mp, &mdoc, &man);
+
 	/* Execute the out device, if it exists. */
 
-	if (curp->man && curp->outman)
-		(*curp->outman)(curp->outdata, curp->man);
-	if (curp->mdoc && curp->outmdoc)
-		(*curp->outmdoc)(curp->outdata, curp->mdoc);
+	if (man && curp->outman)
+		(*curp->outman)(curp->outdata, man);
+	if (mdoc && curp->outmdoc)
+		(*curp->outmdoc)(curp->outdata, mdoc);
 
  cleanup:
 
-	memset(&curp->regs, 0, sizeof(struct regset));
-
-	/* Reset the current-parse compilers. */
-
-	if (curp->mdoc)
-		mdoc_reset(curp->mdoc);
-	if (curp->man)
-		man_reset(curp->man);
-
-	assert(curp->roff);
-	roff_reset(curp->roff);
-
-	if (curp->exit_status < curp->file_status)
-		curp->exit_status = curp->file_status;
-
-	return;
-}
-
-static void
-pdesc(struct curparse *curp)
-{
-	struct buf	 blk;
-	int		 with_mmap;
-
-	/*
-	 * Run for each opened file; may be called more than once for
-	 * each full parse sequence if the opened file is nested (i.e.,
-	 * from `so').  Simply sucks in the whole file and moves into
-	 * the parse phase for the file.
-	 */
-
-	if ( ! read_whole_file(curp, &blk, &with_mmap)) {
-		curp->file_status = MANDOCLEVEL_SYSERR;
-		return;
-	}
-
-	/* Line number is per-file. */
-
-	curp->line = 1;
-
-	parsebuf(curp, blk, 1);
-
-	if (with_mmap)
-		munmap(blk.buf, blk.sz);
-	else
-		free(blk.buf);
-}
-
-/*
- * Main parse routine for an opened file.  This is called for each
- * opened file and simply loops around the full input file, possibly
- * nesting (i.e., with `so').
- */
-static void
-parsebuf(struct curparse *curp, struct buf blk, int start)
-{
-	const struct tbl_span	*span;
-	struct buf	 ln;
-	enum rofferr	 rr;
-	int		 i, of, rc;
-	int		 pos; /* byte number in the ln buffer */
-	int		 lnn; /* line number in the real file */
-	unsigned char	 c;
-
-	memset(&ln, 0, sizeof(struct buf));
-
-	lnn = curp->line; 
-	pos = 0; 
-
-	for (i = 0; i < (int)blk.sz; ) {
-		if (0 == pos && '\0' == blk.buf[i])
-			break;
-
-		if (start) {
-			curp->line = lnn;
-			curp->reparse_count = 0;
-		}
-
-		while (i < (int)blk.sz && (start || '\0' != blk.buf[i])) {
-
-			/*
-			 * When finding an unescaped newline character,
-			 * leave the character loop to process the line.
-			 * Skip a preceding carriage return, if any.
-			 */
-
-			if ('\r' == blk.buf[i] && i + 1 < (int)blk.sz &&
-			    '\n' == blk.buf[i + 1])
-				++i;
-			if ('\n' == blk.buf[i]) {
-				++i;
-				++lnn;
-				break;
-			}
-
-			/* 
-			 * Warn about bogus characters.  If you're using
-			 * non-ASCII encoding, you're screwing your
-			 * readers.  Since I'd rather this not happen,
-			 * I'll be helpful and drop these characters so
-			 * we don't display gibberish.  Note to manual
-			 * writers: use special characters.
-			 */
-
-			c = (unsigned char) blk.buf[i];
-
-			if ( ! (isascii(c) && 
-					(isgraph(c) || isblank(c)))) {
-				mmsg(MANDOCERR_BADCHAR, curp, 
-				    curp->line, pos, "ignoring byte");
-				i++;
-				continue;
-			}
-
-			/* Trailing backslash = a plain char. */
-
-			if ('\\' != blk.buf[i] || i + 1 == (int)blk.sz) {
-				if (pos >= (int)ln.sz)
-					resize_buf(&ln, 256);
-				ln.buf[pos++] = blk.buf[i++];
-				continue;
-			}
-
-			/*
-			 * Found escape and at least one other character.
-			 * When it's a newline character, skip it.
-			 * When there is a carriage return in between,
-			 * skip that one as well.
-			 */
-
-			if ('\r' == blk.buf[i + 1] && i + 2 < (int)blk.sz &&
-			    '\n' == blk.buf[i + 2])
-				++i;
-			if ('\n' == blk.buf[i + 1]) {
-				i += 2;
-				++lnn;
-				continue;
-			}
-
-			if ('"' == blk.buf[i + 1]) {
-				i += 2;
-				/* Comment, skip to end of line */
-				for (; i < (int)blk.sz; ++i) {
-					if ('\n' == blk.buf[i]) {
-						++i;
-						++lnn;
-						break;
-					}
-				}
-
-				/* Backout trailing whitespaces */
-				for (; pos > 0; --pos) {
-					if (ln.buf[pos - 1] != ' ')
-						break;
-					if (pos > 2 && ln.buf[pos - 2] == '\\')
-						break;
-				}
-				break;
-			}
-
-			/* Some other escape sequence, copy & cont. */
-
-			if (pos + 1 >= (int)ln.sz)
-				resize_buf(&ln, 256);
-
-			ln.buf[pos++] = blk.buf[i++];
-			ln.buf[pos++] = blk.buf[i++];
-		}
-
- 		if (pos >= (int)ln.sz)
-			resize_buf(&ln, 256);
-
-		ln.buf[pos] = '\0';
-
-		/*
-		 * A significant amount of complexity is contained by
-		 * the roff preprocessor.  It's line-oriented but can be
-		 * expressed on one line, so we need at times to
-		 * readjust our starting point and re-run it.  The roff
-		 * preprocessor can also readjust the buffers with new
-		 * data, so we pass them in wholesale.
-		 */
-
-		of = 0;
-
-rerun:
-		rr = roff_parseln
-			(curp->roff, curp->line, 
-			 &ln.buf, &ln.sz, of, &of);
-
-		switch (rr) {
-		case (ROFF_REPARSE):
-			if (REPARSE_LIMIT >= ++curp->reparse_count)
-				parsebuf(curp, ln, 0);
-			else
-				mmsg(MANDOCERR_ROFFLOOP, curp, 
-				    curp->line, pos, NULL);
-			pos = 0;
-			continue;
-		case (ROFF_APPEND):
-			pos = (int)strlen(ln.buf);
-			continue;
-		case (ROFF_RERUN):
-			goto rerun;
-		case (ROFF_IGN):
-			pos = 0;
-			continue;
-		case (ROFF_ERR):
-			assert(MANDOCLEVEL_FATAL <= curp->file_status);
-			break;
-		case (ROFF_SO):
-			if (pfile(ln.buf + of, curp)) {
-				pos = 0;
-				continue;
-			} else
-				break;
-		default:
-			break;
-		}
-
-		/*
-		 * If we encounter errors in the recursive parsebuf()
-		 * call, make sure we don't continue parsing.
-		 */
-
-		if (MANDOCLEVEL_FATAL <= curp->file_status)
-			break;
-
-		/*
-		 * If input parsers have not been allocated, do so now.
-		 * We keep these instanced betwen parsers, but set them
-		 * locally per parse routine since we can use different
-		 * parsers with each one.
-		 */
-
-		if ( ! (curp->man || curp->mdoc))
-			pset(ln.buf + of, pos - of, curp);
-
-		/* 
-		 * Lastly, push down into the parsers themselves.  One
-		 * of these will have already been set in the pset()
-		 * routine.
-		 * If libroff returns ROFF_TBL, then add it to the
-		 * currently open parse.  Since we only get here if
-		 * there does exist data (see tbl_data.c), we're
-		 * guaranteed that something's been allocated.
-		 * Do the same for ROFF_EQN.
-		 */
-
-		rc = -1;
-
-		if (ROFF_TBL == rr)
-			while (NULL != (span = roff_span(curp->roff))) {
-				rc = curp->man ?
-					man_addspan(curp->man, span) :
-					mdoc_addspan(curp->mdoc, span);
-				if (0 == rc)
-					break;
-			}
-		else if (ROFF_EQN == rr)
-			rc = curp->mdoc ? 
-				mdoc_addeqn(curp->mdoc, 
-					roff_eqn(curp->roff)) :
-				man_addeqn(curp->man,
-					roff_eqn(curp->roff));
-		else if (curp->man || curp->mdoc)
-			rc = curp->man ?
-				man_parseln(curp->man, 
-					curp->line, ln.buf, of) :
-				mdoc_parseln(curp->mdoc, 
-					curp->line, ln.buf, of);
-
-		if (0 == rc) {
-			assert(MANDOCLEVEL_FATAL <= curp->file_status);
-			break;
-		}
-
-		/* Temporary buffers typically are not full. */
-
-		if (0 == start && '\0' == blk.buf[i])
-			break;
-
-		/* Start the next input line. */
-
-		pos = 0;
-	}
-
-	free(ln.buf);
-}
-
-static void
-pset(const char *buf, int pos, struct curparse *curp)
-{
-	int		 i;
-
-	/*
-	 * Try to intuit which kind of manual parser should be used.  If
-	 * passed in by command-line (-man, -mdoc), then use that
-	 * explicitly.  If passed as -mandoc, then try to guess from the
-	 * line: either skip dot-lines, use -mdoc when finding `.Dt', or
-	 * default to -man, which is more lenient.
-	 *
-	 * Separate out pmdoc/pman from mdoc/man: the first persists
-	 * through all parsers, while the latter is used per-parse.
-	 */
-
-	if ('.' == buf[0] || '\'' == buf[0]) {
-		for (i = 1; buf[i]; i++)
-			if (' ' != buf[i] && '\t' != buf[i])
-				break;
-		if ('\0' == buf[i])
-			return;
-	}
+	mparse_reset(curp->mp);
 
-	switch (curp->inttype) {
-	case (INTT_MDOC):
-		if (NULL == curp->pmdoc) 
-			curp->pmdoc = mdoc_alloc
-				(&curp->regs, curp, mmsg);
-		assert(curp->pmdoc);
-		curp->mdoc = curp->pmdoc;
-		return;
-	case (INTT_MAN):
-		if (NULL == curp->pman) 
-			curp->pman = man_alloc
-				(&curp->regs, curp, mmsg);
-		assert(curp->pman);
-		curp->man = curp->pman;
-		return;
-	default:
-		break;
-	}
-
-	if (pos >= 3 && 0 == memcmp(buf, ".Dd", 3))  {
-		if (NULL == curp->pmdoc) 
-			curp->pmdoc = mdoc_alloc
-				(&curp->regs, curp, mmsg);
-		assert(curp->pmdoc);
-		curp->mdoc = curp->pmdoc;
-		return;
-	} 
-
-	if (NULL == curp->pman) 
-		curp->pman = man_alloc(&curp->regs, curp, mmsg);
-	assert(curp->pman);
-	curp->man = curp->pman;
+	if (*level < rc)
+		*level = rc;
 }
 
 static int
-moptions(enum intt *tflags, char *arg)
+moptions(enum mparset *tflags, char *arg)
 {
 
 	if (0 == strcmp(arg, "doc"))
-		*tflags = INTT_MDOC;
+		*tflags = MPARSE_MDOC;
 	else if (0 == strcmp(arg, "andoc"))
-		*tflags = INTT_AUTO;
+		*tflags = MPARSE_AUTO;
 	else if (0 == strcmp(arg, "an"))
-		*tflags = INTT_MAN;
+		*tflags = MPARSE_MAN;
 	else {
 		fprintf(stderr, "%s: Bad argument\n", arg);
 		return(0);
@@ -978,8 +443,7 @@ toptions(struct curparse *curp, char *arg)
 	else if (0 == strcmp(arg, "lint")) {
 		curp->outtype = OUTT_LINT;
 		curp->wlevel  = MANDOCLEVEL_WARNING;
-	}
-	else if (0 == strcmp(arg, "tree"))
+	} else if (0 == strcmp(arg, "tree"))
 		curp->outtype = OUTT_TREE;
 	else if (0 == strcmp(arg, "html"))
 		curp->outtype = OUTT_HTML;
@@ -1051,12 +515,13 @@ mmsg(enum mandocerr t, void *arg, int ln, int col, const char *msg)
 	if (level < cp->wlevel)
 		return;
 
-	fprintf(stderr, "%s:%d:%d: %s: %s",
-	    cp->file, ln, col + 1, mandoclevels[level], mandocerrs[t]);
+	fprintf(stderr, "%s:%d:%d: %s: %s", cp->file, ln, col + 1, 
+			mandoclevels[level], mandocerrs[t]);
+
 	if (msg)
 		fprintf(stderr, ": %s", msg);
+
 	fputc('\n', stderr);
 
-	if (cp->file_status < level)
-		cp->file_status = level;
+	mparse_setstatus(cp->mp, level);
 }
diff --git a/mandoc.h b/mandoc.h
index 0726c531..ee835a84 100644
--- a/mandoc.h
+++ b/mandoc.h
@@ -328,11 +328,36 @@ enum	mdelim {
 	DELIM_CLOSE
 };
 
+/*
+ * The type of parse sequence.  This value is usually passed via the
+ * mandoc(1) command line of -man and -mdoc.  It's almost exclusively
+ * -mandoc but the others have been retained for compatibility.
+ */
+enum	mparset {
+	MPARSE_AUTO, /* magically determine the document type */
+	MPARSE_MDOC, /* assume -mdoc */
+	MPARSE_MAN /* assume -man */
+};
+
 typedef	void	(*mandocmsg)(enum mandocerr, void *,
 			int, int, const char *);
+typedef	int	(*mevt_open)(void *, const char *);
+typedef	void	(*mevt_close)(void *, const char *);
+
+struct	mparse;
+struct	mdoc;
+struct	man;
 
 __BEGIN_DECLS
 
+void		  mparse_free(struct mparse *);
+void		  mparse_reset(struct mparse *);
+struct mparse	 *mparse_alloc(enum mparset, mevt_open, 
+			mevt_close, mandocmsg, void *);
+void		  mparse_setstatus(struct mparse *, enum mandoclevel);
+enum mandoclevel  mparse_readfd(struct mparse *, int, const char *);
+void		  mparse_result(struct mparse *, struct mdoc **, struct man **);
+
 void		 *mandoc_calloc(size_t, size_t);
 void		 *mandoc_malloc(size_t);
 void		 *mandoc_realloc(void *, size_t);
diff --git a/read.c b/read.c
new file mode 100644
index 00000000..42deb754
--- /dev/null
+++ b/read.c
@@ -0,0 +1,611 @@
+/*	$Id$ */
+/*
+ * Copyright (c) 2008, 2009, 2010, 2011 Kristaps Dzonsons <kristaps@bsd.lv>
+ * Copyright (c) 2010, 2011 Ingo Schwarze <schwarze@openbsd.org>
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ */
+#include <sys/mman.h>
+#include <sys/stat.h>
+
+#include <assert.h>
+#include <ctype.h>
+#include <fcntl.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <unistd.h>
+
+#include "mandoc.h"
+#include "mdoc.h"
+#include "man.h"
+#include "roff.h"
+
+#ifndef MAP_FILE
+#define	MAP_FILE	0
+#endif
+
+#define	REPARSE_LIMIT	1000
+
+struct	buf {
+	char	 	 *buf; /* binary input buffer */
+	size_t		  sz; /* size of binary buffer */
+};
+
+struct	mparse {
+	enum mandoclevel  file_status; /* status of current parse */
+	int		  line; /* line number in the file */
+	enum mparset	  inttype; /* which parser to use */
+	struct man	 *pman; /* persistent man parser */
+	struct mdoc	 *pmdoc; /* persistent mdoc parser */
+	struct man	 *man; /* man parser */
+	struct mdoc	 *mdoc; /* mdoc parser */
+	struct roff	 *roff; /* roff parser (!NULL) */
+	struct regset	  regs; /* roff registers */
+	int		  reparse_count; /* finite interp. stack */
+	mandocmsg	  mmsg; /* warning/error message handler */
+	void		 *arg; /* argument to mmsg */
+	mevt_open	  evt_open; /* file-open event */
+	mevt_close	  evt_close; /* file-close event */
+	const char	 *svfile; 
+};
+
+static	void	  resize_buf(struct buf *, size_t);
+static	void	  mparse_buf_r(struct mparse *, struct buf, int);
+static	void	  mparse_readfd_r(struct mparse *, int, const char *, int);
+static	void	  pset(const char *, int, struct mparse *);
+static	void	  pdesc(struct mparse *, const char *, int);
+static	int	  read_whole_file(const char *, int, struct buf *, int *);
+static	void	  mparse_end(struct mparse *);
+
+static void
+resize_buf(struct buf *buf, size_t initial)
+{
+
+	buf->sz = buf->sz > initial/2 ? 2 * buf->sz : initial;
+	buf->buf = mandoc_realloc(buf->buf, buf->sz);
+}
+
+static void
+pset(const char *buf, int pos, struct mparse *curp)
+{
+	int		 i;
+
+	/*
+	 * Try to intuit which kind of manual parser should be used.  If
+	 * passed in by command-line (-man, -mdoc), then use that
+	 * explicitly.  If passed as -mandoc, then try to guess from the
+	 * line: either skip dot-lines, use -mdoc when finding `.Dt', or
+	 * default to -man, which is more lenient.
+	 *
+	 * Separate out pmdoc/pman from mdoc/man: the first persists
+	 * through all parsers, while the latter is used per-parse.
+	 */
+
+	if ('.' == buf[0] || '\'' == buf[0]) {
+		for (i = 1; buf[i]; i++)
+			if (' ' != buf[i] && '\t' != buf[i])
+				break;
+		if ('\0' == buf[i])
+			return;
+	}
+
+	switch (curp->inttype) {
+	case (MPARSE_MDOC):
+		if (NULL == curp->pmdoc) 
+			curp->pmdoc = mdoc_alloc
+				(&curp->regs, curp->arg, curp->mmsg);
+		assert(curp->pmdoc);
+		curp->mdoc = curp->pmdoc;
+		return;
+	case (MPARSE_MAN):
+		if (NULL == curp->pman) 
+			curp->pman = man_alloc
+				(&curp->regs, curp->arg, curp->mmsg);
+		assert(curp->pman);
+		curp->man = curp->pman;
+		return;
+	default:
+		break;
+	}
+
+	if (pos >= 3 && 0 == memcmp(buf, ".Dd", 3))  {
+		if (NULL == curp->pmdoc) 
+			curp->pmdoc = mdoc_alloc
+				(&curp->regs, curp->arg, curp->mmsg);
+		assert(curp->pmdoc);
+		curp->mdoc = curp->pmdoc;
+		return;
+	} 
+
+	if (NULL == curp->pman) 
+		curp->pman = man_alloc
+			(&curp->regs, curp->arg, curp->mmsg);
+	assert(curp->pman);
+	curp->man = curp->pman;
+}
+
+/*
+ * Main parse routine for an opened file.  This is called for each
+ * opened file and simply loops around the full input file, possibly
+ * nesting (i.e., with `so').
+ */
+static void
+mparse_buf_r(struct mparse *curp, struct buf blk, int start)
+{
+	const struct tbl_span	*span;
+	struct buf	 ln;
+	enum rofferr	 rr;
+	int		 i, of, rc;
+	int		 pos; /* byte number in the ln buffer */
+	int		 lnn; /* line number in the real file */
+	unsigned char	 c;
+
+	memset(&ln, 0, sizeof(struct buf));
+
+	lnn = curp->line; 
+	pos = 0; 
+
+	for (i = 0; i < (int)blk.sz; ) {
+		if (0 == pos && '\0' == blk.buf[i])
+			break;
+
+		if (start) {
+			curp->line = lnn;
+			curp->reparse_count = 0;
+		}
+
+		while (i < (int)blk.sz && (start || '\0' != blk.buf[i])) {
+
+			/*
+			 * When finding an unescaped newline character,
+			 * leave the character loop to process the line.
+			 * Skip a preceding carriage return, if any.
+			 */
+
+			if ('\r' == blk.buf[i] && i + 1 < (int)blk.sz &&
+			    '\n' == blk.buf[i + 1])
+				++i;
+			if ('\n' == blk.buf[i]) {
+				++i;
+				++lnn;
+				break;
+			}
+
+			/* 
+			 * Warn about bogus characters.  If you're using
+			 * non-ASCII encoding, you're screwing your
+			 * readers.  Since I'd rather this not happen,
+			 * I'll be helpful and drop these characters so
+			 * we don't display gibberish.  Note to manual
+			 * writers: use special characters.
+			 */
+
+			c = (unsigned char) blk.buf[i];
+
+			if ( ! (isascii(c) && 
+					(isgraph(c) || isblank(c)))) {
+				curp->mmsg(MANDOCERR_BADCHAR, curp->arg, 
+						curp->line, pos, "ignoring byte");
+				i++;
+				continue;
+			}
+
+			/* Trailing backslash = a plain char. */
+
+			if ('\\' != blk.buf[i] || i + 1 == (int)blk.sz) {
+				if (pos >= (int)ln.sz)
+					resize_buf(&ln, 256);
+				ln.buf[pos++] = blk.buf[i++];
+				continue;
+			}
+
+			/*
+			 * Found escape and at least one other character.
+			 * When it's a newline character, skip it.
+			 * When there is a carriage return in between,
+			 * skip that one as well.
+			 */
+
+			if ('\r' == blk.buf[i + 1] && i + 2 < (int)blk.sz &&
+			    '\n' == blk.buf[i + 2])
+				++i;
+			if ('\n' == blk.buf[i + 1]) {
+				i += 2;
+				++lnn;
+				continue;
+			}
+
+			if ('"' == blk.buf[i + 1]) {
+				i += 2;
+				/* Comment, skip to end of line */
+				for (; i < (int)blk.sz; ++i) {
+					if ('\n' == blk.buf[i]) {
+						++i;
+						++lnn;
+						break;
+					}
+				}
+
+				/* Backout trailing whitespaces */
+				for (; pos > 0; --pos) {
+					if (ln.buf[pos - 1] != ' ')
+						break;
+					if (pos > 2 && ln.buf[pos - 2] == '\\')
+						break;
+				}
+				break;
+			}
+
+			/* Some other escape sequence, copy & cont. */
+
+			if (pos + 1 >= (int)ln.sz)
+				resize_buf(&ln, 256);
+
+			ln.buf[pos++] = blk.buf[i++];
+			ln.buf[pos++] = blk.buf[i++];
+		}
+
+ 		if (pos >= (int)ln.sz)
+			resize_buf(&ln, 256);
+
+		ln.buf[pos] = '\0';
+
+		/*
+		 * A significant amount of complexity is contained by
+		 * the roff preprocessor.  It's line-oriented but can be
+		 * expressed on one line, so we need at times to
+		 * readjust our starting point and re-run it.  The roff
+		 * preprocessor can also readjust the buffers with new
+		 * data, so we pass them in wholesale.
+		 */
+
+		of = 0;
+
+rerun:
+		rr = roff_parseln
+			(curp->roff, curp->line, 
+			 &ln.buf, &ln.sz, of, &of);
+
+		switch (rr) {
+		case (ROFF_REPARSE):
+			if (REPARSE_LIMIT >= ++curp->reparse_count)
+				mparse_buf_r(curp, ln, 0);
+			else
+				curp->mmsg(MANDOCERR_ROFFLOOP, curp->arg, 
+					curp->line, pos, NULL);
+			pos = 0;
+			continue;
+		case (ROFF_APPEND):
+			pos = (int)strlen(ln.buf);
+			continue;
+		case (ROFF_RERUN):
+			goto rerun;
+		case (ROFF_IGN):
+			pos = 0;
+			continue;
+		case (ROFF_ERR):
+			assert(MANDOCLEVEL_FATAL <= curp->file_status);
+			break;
+		case (ROFF_SO):
+			mparse_readfd_r(curp, -1, ln.buf + of, 1);
+			if (MANDOCLEVEL_FATAL <= curp->file_status)
+				break;
+			pos = 0;
+			continue;
+		default:
+			break;
+		}
+
+		/*
+		 * If we encounter errors in the recursive parse, make
+		 * sure we don't continue parsing.
+		 */
+
+		if (MANDOCLEVEL_FATAL <= curp->file_status)
+			break;
+
+		/*
+		 * If input parsers have not been allocated, do so now.
+		 * We keep these instanced betwen parsers, but set them
+		 * locally per parse routine since we can use different
+		 * parsers with each one.
+		 */
+
+		if ( ! (curp->man || curp->mdoc))
+			pset(ln.buf + of, pos - of, curp);
+
+		/* 
+		 * Lastly, push down into the parsers themselves.  One
+		 * of these will have already been set in the pset()
+		 * routine.
+		 * If libroff returns ROFF_TBL, then add it to the
+		 * currently open parse.  Since we only get here if
+		 * there does exist data (see tbl_data.c), we're
+		 * guaranteed that something's been allocated.
+		 * Do the same for ROFF_EQN.
+		 */
+
+		rc = -1;
+
+		if (ROFF_TBL == rr)
+			while (NULL != (span = roff_span(curp->roff))) {
+				rc = curp->man ?
+					man_addspan(curp->man, span) :
+					mdoc_addspan(curp->mdoc, span);
+				if (0 == rc)
+					break;
+			}
+		else if (ROFF_EQN == rr)
+			rc = curp->mdoc ? 
+				mdoc_addeqn(curp->mdoc, 
+					roff_eqn(curp->roff)) :
+				man_addeqn(curp->man,
+					roff_eqn(curp->roff));
+		else if (curp->man || curp->mdoc)
+			rc = curp->man ?
+				man_parseln(curp->man, 
+					curp->line, ln.buf, of) :
+				mdoc_parseln(curp->mdoc, 
+					curp->line, ln.buf, of);
+
+		if (0 == rc) {
+			assert(MANDOCLEVEL_FATAL <= curp->file_status);
+			break;
+		}
+
+		/* Temporary buffers typically are not full. */
+
+		if (0 == start && '\0' == blk.buf[i])
+			break;
+
+		/* Start the next input line. */
+
+		pos = 0;
+	}
+
+	free(ln.buf);
+}
+
+static void
+pdesc(struct mparse *curp, const char *file, int fd)
+{
+	struct buf	 blk;
+	int		 with_mmap;
+
+	/*
+	 * Run for each opened file; may be called more than once for
+	 * each full parse sequence if the opened file is nested (i.e.,
+	 * from `so').  Simply sucks in the whole file and moves into
+	 * the parse phase for the file.
+	 */
+
+	if ( ! read_whole_file(file, fd, &blk, &with_mmap)) {
+		curp->file_status = MANDOCLEVEL_SYSERR;
+		return;
+	}
+
+	/* Line number is per-file. */
+
+	curp->line = 1;
+
+	mparse_buf_r(curp, blk, 1);
+
+	if (with_mmap)
+		munmap(blk.buf, blk.sz);
+	else
+		free(blk.buf);
+}
+
+static int
+read_whole_file(const char *file, int fd, struct buf *fb, int *with_mmap)
+{
+	struct stat	 st;
+	size_t		 off;
+	ssize_t		 ssz;
+
+	if (-1 == fstat(fd, &st)) {
+		perror(file);
+		return(0);
+	}
+
+	/*
+	 * If we're a regular file, try just reading in the whole entry
+	 * via mmap().  This is faster than reading it into blocks, and
+	 * since each file is only a few bytes to begin with, I'm not
+	 * concerned that this is going to tank any machines.
+	 */
+
+	if (S_ISREG(st.st_mode)) {
+		if (st.st_size >= (1U << 31)) {
+			fprintf(stderr, "%s: input too large\n", file);
+			return(0);
+		}
+		*with_mmap = 1;
+		fb->sz = (size_t)st.st_size;
+		fb->buf = mmap(NULL, fb->sz, PROT_READ, 
+				MAP_FILE|MAP_SHARED, fd, 0);
+		if (fb->buf != MAP_FAILED)
+			return(1);
+	}
+
+	/*
+	 * If this isn't a regular file (like, say, stdin), then we must
+	 * go the old way and just read things in bit by bit.
+	 */
+
+	*with_mmap = 0;
+	off = 0;
+	fb->sz = 0;
+	fb->buf = NULL;
+	for (;;) {
+		if (off == fb->sz) {
+			if (fb->sz == (1U << 31)) {
+				fprintf(stderr, "%s: input too large\n", file);
+				break;
+			}
+			resize_buf(fb, 65536);
+		}
+		ssz = read(fd, fb->buf + (int)off, fb->sz - off);
+		if (ssz == 0) {
+			fb->sz = off;
+			return(1);
+		}
+		if (ssz == -1) {
+			perror(file);
+			break;
+		}
+		off += (size_t)ssz;
+	}
+
+	free(fb->buf);
+	fb->buf = NULL;
+	return(0);
+}
+
+static void
+mparse_end(struct mparse *curp)
+{
+
+	if (MANDOCLEVEL_FATAL <= curp->file_status)
+		return;
+
+	if (curp->mdoc && ! mdoc_endparse(curp->mdoc)) {
+		assert(MANDOCLEVEL_FATAL <= curp->file_status);
+		return;
+	}
+
+	if (curp->man && ! man_endparse(curp->man)) {
+		assert(MANDOCLEVEL_FATAL <= curp->file_status);
+		return;
+	}
+
+#if 0
+	/* NOTE a parser may not have been assigned, yet. */
+
+	if ( ! (curp->man || curp->mdoc)) {
+		/* FIXME: make into an mandoc.h error. */
+		fprintf(stderr, "%s: Not a manual\n", curp->file);
+		curp->file_status = MANDOCLEVEL_FATAL;
+		goto cleanup;
+	}
+#endif
+
+	roff_endparse(curp->roff);
+}
+
+static void
+mparse_readfd_r(struct mparse *curp, int fd, const char *file, int re)
+{
+	const char	*svfile;
+
+	if ( ! (*curp->evt_open)(curp->arg, file)) {
+		curp->file_status = MANDOCLEVEL_SYSERR;
+		return;
+	}
+
+	if (-1 == fd)
+		if (-1 == (fd = open(file, O_RDONLY, 0))) {
+			perror(file);
+			curp->file_status = MANDOCLEVEL_SYSERR;
+			return;
+		}
+
+	svfile = curp->svfile;
+	curp->svfile = file;
+
+	pdesc(curp, file, fd);
+
+	if (0 == re && MANDOCLEVEL_FATAL > curp->file_status)
+		mparse_end(curp);
+
+	if (STDIN_FILENO != fd && -1 == close(fd))
+		perror(file);
+
+	(*curp->evt_close)(curp->arg, svfile);
+	curp->svfile = svfile;
+}
+
+enum mandoclevel
+mparse_readfd(struct mparse *curp, int fd, const char *file)
+{
+
+	mparse_readfd_r(curp, fd, file, 0);
+	return(curp->file_status);
+}
+
+void
+mparse_setstatus(struct mparse *curp, enum mandoclevel lvl)
+{
+
+	if (curp->file_status < lvl)
+		curp->file_status = lvl;
+}
+
+struct mparse *
+mparse_alloc(enum mparset inttype, mevt_open eopen, 
+		mevt_close eclose, mandocmsg mmsg, void *arg)
+{
+	struct mparse	*curp;
+
+	curp = mandoc_calloc(1, sizeof(struct mparse));
+
+	curp->mmsg = mmsg;
+	curp->arg = arg;
+	curp->inttype = inttype;
+	curp->evt_open = eopen;
+	curp->evt_close = eclose;
+
+	curp->roff = roff_alloc(&curp->regs, arg, mmsg);
+	return(curp);
+}
+
+void
+mparse_reset(struct mparse *curp)
+{
+
+	memset(&curp->regs, 0, sizeof(struct regset));
+
+	roff_reset(curp->roff);
+
+	if (curp->mdoc)
+		mdoc_reset(curp->mdoc);
+	if (curp->man)
+		man_reset(curp->man);
+
+	curp->file_status = MANDOCLEVEL_OK;
+	curp->mdoc = NULL;
+	curp->man = NULL;
+}
+
+void
+mparse_free(struct mparse *curp)
+{
+
+	if (curp->pmdoc)
+		mdoc_free(curp->pmdoc);
+	if (curp->pman)
+		man_free(curp->pman);
+	if (curp->roff)
+		roff_free(curp->roff);
+
+	free(curp);
+}
+
+void
+mparse_result(struct mparse *curp, struct mdoc **mdoc, struct man **man)
+{
+
+	*mdoc = curp->mdoc;
+	*man = curp->man;
+}