De-chunking of text removed from parsers. This is a significant change (and I don't really like it), but it's what groff does.

Distinction of ARGS_PHRASE and ARGS_PPHRASE in backend (not yet used).

5 files changed, 51 insertions, 131 deletions

diff --git a/man.7 b/man.7
index af75bff3..5a3b22dc 100644
--- a/man.7
+++ b/man.7
@@ -887,11 +887,6 @@ In quoted literals, GNU troff allowed pair-wise double-quotes to produce
 a standalone double-quote in formatted output.  It is not known whether
 this behaviour is exhibited by other formatters.
 .It
-Blocks of whitespace are stripped from macro and free-form text lines
-(except when in literal mode) in mandoc.  This is not the case for GNU
-troff: for maximum portability, whitespace sensitive blocks should be
-enclosed in literal contexts.
-.It
 The
 .Sx \&sp
 macro does not accept negative values in mandoc.  In GNU troff, this
diff --git a/man.c b/man.c
index 878963c1..b83684f3 100644
--- a/man.c
+++ b/man.c
@@ -83,8 +83,6 @@ static	int		 man_ptext(struct man *, int, char *);
 static	int		 man_pmacro(struct man *, int, char *);
 static	void		 man_free1(struct man *);
 static	void		 man_alloc1(struct man *);
-static	int		 pstring(struct man *, int, int, 
-				const char *, size_t);
 static	int		 macrowarn(struct man *, int, const char *);
 
 
@@ -318,35 +316,29 @@ man_block_alloc(struct man *m, int line, int pos, enum mant tok)
 }
 
 
-static int
-pstring(struct man *m, int line, int pos, 
-		const char *p, size_t len)
+int
+man_word_alloc(struct man *m, int line, int pos, const char *word)
 {
 	struct man_node	*n;
-	size_t		 sv;
+	size_t		 sv, len;
+
+	len = strlen(word);
 
 	n = man_node_alloc(line, pos, MAN_TEXT, MAN_MAX);
 	n->string = mandoc_malloc(len + 1);
-	sv = strlcpy(n->string, p, len + 1);
+	sv = strlcpy(n->string, word, len + 1);
 
 	/* Prohibit truncation. */
 	assert(sv < len + 1);
 
 	if ( ! man_node_append(m, n))
 		return(0);
+
 	m->next = MAN_NEXT_SIBLING;
 	return(1);
 }
 
 
-int
-man_word_alloc(struct man *m, int line, int pos, const char *word)
-{
-
-	return(pstring(m, line, pos, word, strlen(word)));
-}
-
-
 /*
  * Free all of the resources held by a node.  This does NOT unlink a
  * node from its context; for that, see man_node_unlink().
@@ -376,8 +368,7 @@ man_node_delete(struct man *m, struct man_node *p)
 static int
 man_ptext(struct man *m, int line, char *buf)
 {
-	int		 i, j;
-	char		 sv;
+	int		 i;
 
 	/* Ignore bogus comments. */
 
@@ -392,61 +383,32 @@ man_ptext(struct man *m, int line, char *buf)
 		goto descope;
 	}
 
-	/* First de-chunk and allocate words. */
+	/* Pump blank lines directly into the backend. */
 
 	for (i = 0; ' ' == buf[i]; i++)
 		/* Skip leading whitespace. */ ;
 
 	if ('\0' == buf[i]) {
-		/* Trailing whitespace? */
-		if (i && ' ' == buf[i - 1])
-			if ( ! man_pwarn(m, line, i - 1, WTSPACE))
-				return(0);
-		if ( ! pstring(m, line, 0, &buf[i], 0))
+		/* Allocate a blank entry. */
+		if ( ! man_word_alloc(m, line, 0, ""))
 			return(0);
 		goto descope;
 	}
 
-	for (j = i; buf[i]; i++) {
-		if (' ' != buf[i])
-			continue;
+	/* Warn if the last un-escaped character is whitespace. */
 
-		/* Escaped whitespace. */
-		if (i && ' ' == buf[i] && '\\' == buf[i - 1])
-			continue;
+	i = (int)strlen(buf);
+	assert(i);
 
-		sv = buf[i];
-		buf[i++] = '\0';
-
-		if ( ! pstring(m, line, j, &buf[j], (size_t)(i - j)))
-			return(0);
-
-		/* Trailing whitespace?  Check at overwritten byte. */
-
-		if (' ' == sv && '\0' == buf[i])
+	if (' ' == buf[i - 1] || '\t' == buf[i - 1])
+		if (1 == i || ('\\' != buf[i - 2]))
 			if ( ! man_pwarn(m, line, i - 1, WTSPACE))
 				return(0);
 
-		for ( ; ' ' == buf[i]; i++)
-			/* Skip trailing whitespace. */ ;
-
-		j = i;
-
-		/* Trailing whitespace? */
-
-		if (' ' == buf[i - 1] && '\0' == buf[i])
-			if ( ! man_pwarn(m, line, i - 1, WTSPACE))
-				return(0);
-
-		if ('\0' == buf[i])
-			break;
-	}
-
-	if (j != i && ! pstring(m, line, j, &buf[j], (size_t)(i - j)))
+	if ( ! man_word_alloc(m, line, 0, buf))
 		return(0);
 
 descope:
-
 	/*
 	 * Co-ordinate what happens with having a next-line scope open:
 	 * first close out the element scope (if applicable), then close
diff --git a/mdoc.7 b/mdoc.7
index c354a3b5..1b807efa 100644
--- a/mdoc.7
+++ b/mdoc.7
@@ -1824,10 +1824,6 @@ are aliases, as are
 and
 .Fl unfilled .
 .It
-In mandoc, blocks of whitespace are stripped from both macro and
-free-form text lines (except when in literal mode); groff would retain
-whitespace in free-form text lines.
-.It
 Historic groff has many un-callable macros.  Most of these (excluding
 some block-level macros) are now callable.
 .It
diff --git a/mdoc.c b/mdoc.c
index 10aad780..f5e9ab7d 100644
--- a/mdoc.c
+++ b/mdoc.c
@@ -154,8 +154,7 @@ static	int		  node_append(struct mdoc *,
 static	int		  mdoc_ptext(struct mdoc *, int, char *);
 static	int		  mdoc_pmacro(struct mdoc *, int, char *);
 static	int		  macrowarn(struct mdoc *, int, const char *);
-static	int		  pstring(struct mdoc *, int, int, 
-				const char *, size_t);
+
 
 const struct mdoc_node *
 mdoc_node(const struct mdoc *m)
@@ -535,11 +534,13 @@ mdoc_elem_alloc(struct mdoc *m, int line, int pos,
 }
 
 
-static int
-pstring(struct mdoc *m, int line, int pos, const char *p, size_t len)
+int
+mdoc_word_alloc(struct mdoc *m, int line, int pos, const char *p)
 {
 	struct mdoc_node *n;
-	size_t		  sv;
+	size_t		  sv, len;
+
+	len = strlen(p);
 
 	n = node_alloc(m, line, pos, -1, MDOC_TEXT);
 	n->string = mandoc_malloc(len + 1);
@@ -550,19 +551,12 @@ pstring(struct mdoc *m, int line, int pos, const char *p, size_t len)
 
 	if ( ! node_append(m, n))
 		return(0);
+
 	m->next = MDOC_NEXT_SIBLING;
 	return(1);
 }
 
 
-int
-mdoc_word_alloc(struct mdoc *m, int line, int pos, const char *p)
-{
-
-	return(pstring(m, line, pos, p, strlen(p)));
-}
-
-
 void
 mdoc_node_free(struct mdoc_node *p)
 {
@@ -633,88 +627,57 @@ mdoc_node_delete(struct mdoc *m, struct mdoc_node *p)
 static int
 mdoc_ptext(struct mdoc *m, int line, char *buf)
 {
-	int		 i, j;
-	char		 sv;
+	int		 i;
 
 	/* Ignore bogus comments. */
 
 	if ('\\' == buf[0] && '.' == buf[1] && '\"' == buf[2])
 		return(mdoc_pwarn(m, line, 0, EBADCOMMENT));
 
+	/* No text before an initial macro. */
+
 	if (SEC_NONE == m->lastnamed)
 		return(mdoc_perr(m, line, 0, ETEXTPROL));
-	
-	/*
-	 * If in literal mode, then pass the buffer directly to the
-	 * back-end, as it should be preserved as a single term.
-	 */
 
+	/* Literal just gets pulled in as-is. */
+	
 	if (MDOC_LITERAL & m->flags)
 		return(mdoc_word_alloc(m, line, 0, buf));
 
-	/* Disallow blank/white-space lines in non-literal mode. */
+	/* Check for a blank line, which may also consist of spaces. */
 
 	for (i = 0; ' ' == buf[i]; i++)
-		/* Skip leading whitespace. */ ;
+		/* Skip to first non-space. */ ;
 
 	if ('\0' == buf[i]) {
 		if ( ! mdoc_pwarn(m, line, 0, ENOBLANK))
 			return(0);
+
 		/*
-		 * Assume that a `Pp' should be inserted in the case of
-		 * a blank line.  Technically, blank lines aren't
-		 * allowed, but enough manuals assume this behaviour
-		 * that we want to work around it.
+		 * Insert a `Pp' in the case of a blank line.  Technically,
+		 * blank lines aren't allowed, but enough manuals assume this
+		 * behaviour that we want to work around it.
 		 */
 		if ( ! mdoc_elem_alloc(m, line, 0, MDOC_Pp, NULL))
 			return(0);
-	}
-
-	/*
-	 * Break apart a free-form line into tokens.  Spaces are
-	 * stripped out of the input.
-	 */
 
-	for (j = i; buf[i]; i++) {
-		if (' ' != buf[i])
-			continue;
-
-		/* Escaped whitespace. */
-		if (i && ' ' == buf[i] && '\\' == buf[i - 1])
-			continue;
-
-		sv = buf[i];
-		buf[i++] = '\0';
-
-		if ( ! pstring(m, line, j, &buf[j], (size_t)(i - j)))
-			return(0);
-
-		/* Trailing whitespace?  Check at overwritten byte. */
-
-		if (' ' == sv && '\0' == buf[i])
-			if ( ! mdoc_pwarn(m, line, i - 1, ETAILWS))
-				return(0);
-
-		for ( ; ' ' == buf[i]; i++)
-			/* Skip trailing whitespace. */ ;
+		m->next = MDOC_NEXT_SIBLING;
+		return(1);
+	}
 
-		j = i;
+	/* Warn if the last un-escaped character is whitespace. */
 
-		/* Trailing whitespace? */
+	i = (int)strlen(buf);
+	assert(i);
 
-		if (' ' == buf[i - 1] && '\0' == buf[i])
+	if (' ' == buf[i - 1] || '\t' == buf[i - 1])
+		if (1 == i || ('\\' != buf[i - 2]))
 			if ( ! mdoc_pwarn(m, line, i - 1, ETAILWS))
 				return(0);
 
-		if ('\0' == buf[i])
-			break;
-	}
-
-	if (j != i && ! pstring(m, line, j, &buf[j], (size_t)(i - j)))
-		return(0);
+	/* Allocate the whole word. */
 
-	m->next = MDOC_NEXT_SIBLING;
-	return(1);
+	return(mdoc_word_alloc(m, line, 0, buf));
 }
 
 
diff --git a/mdoc_argv.c b/mdoc_argv.c
index 0ab8d9d6..5519d07f 100644
--- a/mdoc_argv.c
+++ b/mdoc_argv.c
@@ -398,7 +398,7 @@ args(struct mdoc *m, int line, int *pos,
 	assert(*pos);
 	assert(' ' != buf[*pos]);
 
-	if (0 == buf[*pos])
+	if ('\0' == buf[*pos])
 		return(ARGS_EOLN);
 
 	/* 
@@ -441,7 +441,6 @@ args(struct mdoc *m, int line, int *pos,
 	 */
 
 	if (ARGS_TABSEP & fl) {
-		rc = ARGS_PHRASE;
 		/* Scan ahead to tab (can't be escaped). */
 		p = strchr(*v, '\t');
 
@@ -455,14 +454,19 @@ args(struct mdoc *m, int line, int *pos,
 				break;
 		}
 
+		/* By default, assume a phrase. */
+		rc = ARGS_PHRASE;
+
 		/* 
 		 * Adjust new-buffer position to be beyond delimiter
 		 * mark (e.g., Ta -> end + 2).
 		 */
 		if (p && pp) {
 			*pos += pp < p ? 2 : 1;
+			rc = pp < p ? ARGS_PHRASE : ARGS_PPHRASE;
 			p = pp < p ? pp : p;
 		} else if (p && ! pp) {
+			rc = ARGS_PPHRASE;
 			*pos += 1;
 		} else if (pp && ! p) {
 			p = pp;