===================================================================
RCS file: /cvs/pod2mdoc/pod2mdoc.c,v
retrieving revision 1.3
retrieving revision 1.7
diff -u -p -r1.3 -r1.7
--- pod2mdoc/pod2mdoc.c	2014/03/20 15:18:56	1.3
+++ pod2mdoc/pod2mdoc.c	2014/03/23 23:35:59	1.7
@@ -1,4 +1,4 @@
-/*	$Id: pod2mdoc.c,v 1.3 2014/03/20 15:18:56 schwarze Exp $ */
+/*	$Id: pod2mdoc.c,v 1.7 2014/03/23 23:35:59 kristaps Exp $ */
 /*
  * Copyright (c) 2014 Kristaps Dzonsons <kristaps@bsd.lv>
  *
@@ -32,12 +32,22 @@ struct	args {
 	const char	*section; /* override "Dt" section */
 };
 
+enum	list {
+	LIST_BULLET = 0,
+	LIST_ENUM,
+	LIST_TAG,
+	LIST__MAX
+};
+
 struct	state {
 	int		 parsing; /* after =cut of before command */
 	int		 paused; /* in =begin and before =end */
 	int		 haspar; /* in paragraph: do we need Pp? */
 	int		 isname; /* are we the NAME section? */
 	const char	*fname; /* file being parsed */
+#define	LIST_STACKSZ	 128
+	enum list	 lstack[LIST_STACKSZ]; /* open lists */
+	size_t		 lpos; /* where in list stack */
 };
 
 enum	fmt {
@@ -98,6 +108,8 @@ static	const char fmts[FMT__MAX] = {
 	'Z'		/* FMT_NULL */
 };
 
+static	int 	last;
+
 /*
  * Given buf[*start] is at the start of an escape name, read til the end
  * of the escape ('>') then try to do something with it.
@@ -131,7 +143,7 @@ formatescape(const char *buf, size_t *start, size_t en
 	 * TODO: right now, we only recognise the named escapes.
 	 * Just let the rest of them go. 
 	 */
-	if (0 == strcmp(esc, "lt"))
+	if (0 == strcmp(esc, "lt")) 
 		printf("\\(la");
 	else if (0 == strcmp(esc, "gt"))
 		printf("\\(ra");
@@ -139,17 +151,24 @@ formatescape(const char *buf, size_t *start, size_t en
 		printf("\\(ba");
 	else if (0 == strcmp(esc, "sol"))
 		printf("\\(sl");
+	else
+		return;
+
+	last = 'a';
 }
 
 /*
  * Skip space characters.
  */
-static void
+static int
 skipspace(const char *buf, size_t *start, size_t end)
 {
+	size_t		 sv = *start;
 
 	while (*start < end && ' ' == buf[*start])
 		(*start)++;
+
+	return(*start > sv);
 }
 
 /*
@@ -159,49 +178,90 @@ skipspace(const char *buf, size_t *start, size_t end)
  * the end of matched production.
  * If "reentrant", then we're being called after a macro has already
  * been printed to the current line.
- * "last" is set to the last read character: this is used to determine
- * whether we should buffer with space or not.
- * If "nomacro", then we don't print any macros, just contained data.
+ * If "nomacro", then we don't print any macros, just contained data
+ * (e.g., following "Sh" or "Nm").
+ * Return whether we've printed a macro or not--in other words, whether
+ * this should trigger a subsequent newline (this should be ignored when
+ * reentrant).
  */
 static int
 formatcode(const char *buf, size_t *start, 
-	size_t end, int reentrant, int last, int nomacro)
+	size_t end, int reentrant, int nomacro)
 {
 	enum fmt	 fmt;
+	size_t		 i, j, dsz;
 
 	assert(*start + 1 < end);
 	assert('<' == buf[*start + 1]);
 
+	/* 
+	 * First, look up the format code. 
+	 * If it's not valid, then exit immediately.
+	 */
 	for (fmt = 0; fmt < FMT__MAX; fmt++) 
 		if (buf[*start] == fmts[fmt])
 			break;
 
-	/* Invalid macros are just regular text. */
-
 	if (FMT__MAX == fmt) {
-		putchar(buf[*start]);
-		(*start)++;
+		putchar(last = buf[(*start)++]);
 		return(0);
 	}
 
-	*start += 2;
+	/*
+	 * Determine whether we're overriding our delimiter.
+	 * According to POD, if we have more than one '<' followed by a
+	 * space, then we need a space followed by matching '>' to close
+	 * the expression.
+	 * Otherwise we use the usual '<' and '>' matched pair.
+	 */
+	i = *start + 1;
+	while (i < end && '<' == buf[i])
+		i++;
+	assert(i > *start + 1);
+	dsz = i - (*start + 1);
+	if (dsz > 1 && (i >= end || ' ' != buf[i]))
+		dsz = 1;
 
+	/* Remember, if dsz>1, to jump the trailing space. */
+	*start += dsz + 1 + (dsz > 1 ? 1 : 0);
+
 	/*
-	 * Escapes don't print macro sequences, so just output them like
-	 * normal text before processing for macros.
+	 * Escapes and ignored codes (NULL and INDEX) don't print macro
+	 * sequences, so just output them like normal text before
+	 * processing for real macros.
 	 */
 	if (FMT_ESCAPE == fmt) {
 		formatescape(buf, start, end);
 		return(0);
 	} else if (FMT_NULL == fmt || FMT_INDEX == fmt) {
-		/* For indices and nulls, just consume. */
-		while (*start < end && '>' != buf[*start])
-			(*start)++;
-		if (*start < end)
-			(*start)++;
+		/* 
+		 * Just consume til the end delimiter, accounting for
+		 * whether it's a custom one.
+		 */
+		for ( ; *start < end; (*start)++) {
+			if ('>' != buf[*start])
+				continue;
+			else if (dsz == 1)
+				break;
+			assert(*start > 0);
+			if (' ' != buf[*start - 1])
+				continue;
+			i = *start;
+			for (j = 0; i < end && j < dsz; j++) 
+				if ('>' != buf[i++])
+					break;
+			if (dsz != j) 
+				continue;
+			(*start) += dsz;
+			break;
+		}
 		return(0);
 	}
 
+	/*
+	 * Check whether we're supposed to print macro stuff (this is
+	 * suppressed in, e.g., "Nm" and "Sh" macros).
+	 */
 	if ( ! nomacro) {
 		/*
 		 * Print out the macro describing this format code.
@@ -210,19 +270,21 @@ formatcode(const char *buf, size_t *start, 
 		 * indicator.
 		 * Otherwise, offset us with a space.
 		 */
-		if ( ! reentrant && last != '\n')
-			putchar('\n');
-		if ( ! reentrant)
+		if ( ! reentrant) {
+			if (last != '\n')
+				putchar('\n');
 			putchar('.');
-		else
+		} else 
 			putchar(' ');
 		
 		/*
-		 * If we don't have whitespace before us, then suppress
-		 * macro whitespace with Ns.
+		 * If we don't have whitespace before us (and none after
+		 * the opening delimiter), then suppress macro
+		 * whitespace with Pf.
 		 */
-		if (' ' != last)
-			printf("Ns ");
+		if (' ' != last && '\n' != last && ' ' != buf[*start])
+			printf("Pf ");
+
 		switch (fmt) {
 		case (FMT_ITALIC):
 			printf("Em ");
@@ -249,68 +311,77 @@ formatcode(const char *buf, size_t *start, 
 	}
 
 	/*
-	 * Read until we reach the end market ('>') or until we find a
-	 * nested format code.
+	 * Process until we reach the end marker (e.g., '>') or until we
+	 * find a nested format code.
 	 * Don't emit any newlines: since we're on a macro line, we
 	 * don't want to break the line.
 	 */
 	while (*start < end) {
-		if ('>' == buf[*start]) {
+		if ('>' == buf[*start] && 1 == dsz) {
 			(*start)++;
 			break;
+		} else if ('>' == buf[*start] && 
+				' ' == buf[*start - 1]) {
+			/*
+			 * Handle custom delimiters.
+			 * These require a certain number of
+			 * space-preceded carrots before we're really at
+			 * the end.
+			 */
+			i = *start;
+			for (j = 0; i < end && j < dsz; j++)
+				if ('>' != buf[i++])
+					break;
+			if (dsz == j) {
+				*start += dsz;
+				break;
+			}
 		}
 		if (*start + 1 < end && '<' == buf[*start + 1]) {
-			formatcode(buf, start, end, 1, last, nomacro);
+			formatcode(buf, start, end, 1, nomacro);
 			continue;
 		}
 
-			/*
-			 * Make sure that any macro-like words (or
-			 * really any word starting with a capital
-			 * letter) is assumed to be a macro that must be
-			 * escaped.
-			 * XXX: should this be isalpha()?
-			 */
-			if ((' ' == last || '\n' == last) && 
-				isupper(buf[*start]))
-				printf("\\&");
+		/*
+		 * Make sure that any macro-like words (or
+		 * really any word starting with a capital
+		 * letter) is assumed to be a macro that must be
+		 * escaped.
+		 * This matches "Xx " and "XxEOLN".
+		 */
+		if ((' ' == last || '\n' == last) && 
+				end - *start > 1 &&
+				isupper((int)buf[*start]) &&
+				islower((int)buf[*start + 1]) &&
+				(end - *start == 2 ||
+				 ' ' == buf[*start + 2]))
+			printf("\\&");
 
-		last = buf[*start];
-		if ('\n' == last)
-			last = ' ';
-		putchar(last);
+		/* Suppress newline. */
+		if ('\n' == buf[*start])
+			putchar(last = ' ');
+		else
+			putchar(last = buf[*start]);
 
 		(*start)++;
+
+		if (' ' == last)
+			while (*start < end && ' ' == buf[*start])
+				(*start)++;
 	}
 
 	if ( ! nomacro && FMT_CODE == fmt)
 		printf(" Qc ");
 
-	if (reentrant)
-		return(1);
-
 	/*
-	 * If we're not reentrant, we want to put ending punctuation on
-	 * the macro line so that it's properly handled by being
-	 * smooshed against the terminal word.
+	 * We're now subsequent the format code.
+	 * If there isn't a space (or newline) here, and we haven't just
+	 * printed a space, then suppress space.
 	 */
-	skipspace(buf, start, end);
-	if (',' != buf[*start] && '.' != buf[*start] &&
-		'!' != buf[*start] && '?' != buf[*start] &&
-		')' != buf[*start])
-		return(1);
-	while (*start < end) {
-		if (',' != buf[*start] &&
-			'.' != buf[*start] &&
-			'!' != buf[*start] &&
-			'?' != buf[*start] &&
-			')' != buf[*start])
-			break;
-		putchar(' ');
-		putchar(buf[*start]);
-		(*start)++;
-	}
-	skipspace(buf, start, end);
+	if ( ! nomacro && ' ' != last)
+		if (' ' != buf[*start] && '\n' != buf[*start])
+			printf(" Ns ");
+
 	return(1);
 }
 
@@ -320,21 +391,58 @@ formatcode(const char *buf, size_t *start, 
 static void
 formatcodeln(const char *buf, size_t *start, size_t end, int nomacro)
 {
-	int		 last;
 
-	last = '\n';
+	last = ' ';
 	while (*start < end)  {
 		if (*start + 1 < end && '<' == buf[*start + 1]) {
-			formatcode(buf, start, end, 1, last, nomacro);
+			formatcode(buf, start, end, 1, nomacro);
 			continue;
 		}
+		/*
+		 * Since we're already on a macro line, we want to make
+		 * sure that we don't inadvertently invoke a macro.
+		 * We need to do this carefully because section names
+		 * are used in troff and we don't want to escape
+		 * something that needn't be escaped.
+		 */
+		if (' ' == last && end - *start > 1 &&
+				isupper((int)buf[*start]) &&
+				islower((int)buf[*start + 1]) &&
+				(end - *start == 2 ||
+				 ' ' == buf[*start + 2]))
+			printf("\\&");
+
 		if ('\n' != buf[*start])
 			putchar(last = buf[*start]);
+		else
+			putchar(last = ' ');
 		(*start)++;
 	}
 }
 
 /*
+ * Guess at what kind of list we are.
+ * These are taken straight from the POD manual.
+ * I don't know what people do in real life.
+ */
+static enum list
+listguess(const char *buf, size_t start, size_t end)
+{
+	size_t		 len = end - start;
+
+	assert(end >= start);
+
+	if (len == 1 && '*' == buf[start])
+		return(LIST_BULLET);
+	if (len == 2 && '1' == buf[start] && '.' == buf[start + 1])
+		return(LIST_ENUM);
+	else if (len == 1 && '1' == buf[start])
+		return(LIST_ENUM);
+	else
+		return(LIST_TAG);
+}
+
+/*
  * A command paragraph, as noted in the perlpod manual, just indicates
  * that we should do something, optionally with some text to print as
  * well.
@@ -411,21 +519,78 @@ command(struct state *st, const char *buf, size_t star
 		st->haspar = 1;
 		break;
 	case (CMD_OVER):
-		/*
-		 * TODO: we should be doing this after we process the
-		 * first =item to see whether we'll do an -enum,
-		 * -bullet, or something else.
+		/* 
+		 * If we have an existing list that hasn't had an =item
+		 * yet, then make sure that we open it now.
+		 * We use the default list type, but that can't be
+		 * helped (we haven't seen any items yet).
 		 */
-		puts(".Bl -tag -width Ds");
+		if (st->lpos > 0)
+			if (LIST__MAX == st->lstack[st->lpos - 1]) {
+				st->lstack[st->lpos - 1] = LIST_TAG;
+				puts(".Bl -tag -width Ds");
+			}
+		st->lpos++;
+		assert(st->lpos < LIST_STACKSZ);
+		st->lstack[st->lpos - 1] = LIST__MAX;
 		break;
 	case (CMD_ITEM):
-		printf(".It ");
-		formatcodeln(buf, &start, end, 0);
-		putchar('\n');
+		if (0 == st->lpos) {
+			/* 
+			 * Bad markup.
+			 * Try to compensate.
+			 */
+			st->lstack[st->lpos] = LIST__MAX;
+			st->lpos++;
+		}
+		assert(st->lpos > 0);
+		/*
+		 * If we're the first =item, guess at what our content
+		 * will be: "*" is a bullet list, "1." is a numbered
+		 * list, and everything is tagged.
+		 */
+		if (LIST__MAX == st->lstack[st->lpos - 1]) {
+			st->lstack[st->lpos - 1] = 
+				listguess(buf, start, end);
+			switch (st->lstack[st->lpos - 1]) {
+			case (LIST_BULLET):
+				puts(".Bl -bullet");
+				break;
+			case (LIST_ENUM):
+				puts(".Bl -enum");
+				break;
+			default:
+				puts(".Bl -tag -width Ds");
+				break;
+			}
+		}
+		switch (st->lstack[st->lpos - 1]) {
+		case (LIST_TAG):
+			printf(".It ");
+			formatcodeln(buf, &start, end, 0);
+			putchar('\n');
+			break;
+		case (LIST_ENUM):
+			/* FALLTHROUGH */
+		case (LIST_BULLET):
+			/*
+			 * Abandon the remainder of the paragraph
+			 * because we're going to be a bulletted or
+			 * numbered list.
+			 */
+			puts(".It");
+			break;
+		default:
+			abort();
+		}
 		st->haspar = 1;
 		break;
 	case (CMD_BACK):
-		puts(".El");
+		/* Make sure we don't back over the stack. */
+		if (st->lpos > 0) {
+			st->lpos--;
+			puts(".El");
+		}
 		break;
 	case (CMD_BEGIN):
 		/* 
@@ -459,12 +624,19 @@ command(struct state *st, const char *buf, size_t star
 static void
 verbatim(struct state *st, const char *buf, size_t start, size_t end)
 {
+	size_t		 sv = start;
 
 	if ( ! st->parsing || st->paused)
 		return;
 
 	puts(".Bd -literal");
-	printf("%.*s\n", (int)(end - start), &buf[start]);
+	while (start < end) {
+		if (start > sv && '\n' == buf[start - 1])
+			if ('.' == buf[start] || '\'' == buf[start])
+				printf("\\&");
+		putchar(buf[start++]);
+	}
+	putchar('\n');
 	puts(".Ed");
 }
 
@@ -479,7 +651,6 @@ verbatim(struct state *st, const char *buf, size_t sta
 static void
 ordinary(struct state *st, const char *buf, size_t start, size_t end)
 {
-	int		last;
 	size_t		i, j;
 
 	if ( ! st->parsing || st->paused)
@@ -501,10 +672,13 @@ ordinary(struct state *st, const char *buf, size_t sta
 			for ( ; i > start; i--)
 				if ('-' != buf[i])
 					break;
-			printf(".Nm %.*s\n", 
-				(int)((i + 1) - start), &buf[start]);
-			printf(".Nd %.*s\n", 
-				(int)(end - (j + 1)), &buf[j + 1]);
+			printf(".Nm ");
+			formatcodeln(buf, &start, i + 1, 1);
+			putchar('\n');
+			start = j + 1;
+			printf(".Nd ");
+			formatcodeln(buf, &start, end, 1);
+			putchar('\n');
 			return;
 		}
 	}
@@ -540,8 +714,11 @@ ordinary(struct state *st, const char *buf, size_t sta
 			 * Then print the (possibly nested) macros and
 			 * following that, a newline.
 			 */
-			if (formatcode(buf, &start, end, 0, last, 0))
+			if (formatcode(buf, &start, end, 0, 0)) {
 				putchar(last = '\n');
+				while (start < end && ' ' == buf[start])
+					start++;
+			}
 		} else if (start < end && '\n' == buf[start]) {
 			/*
 			 * Print the newline only if we haven't already