Enhance FIELDWIDTHS syntax to support a skip prefix, and unify logic with API field parsing.

12 files changed, 226 insertions, 134 deletions

diff --git a/ChangeLog b/ChangeLog
index 176142c9..feea67af 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,43 @@
+2017-03-21         Andrew J. Schorr     <aschorr@telemetry-investments.com>
+
+	* gawkapi.h (awk_fieldwidth_info_t): Define new structure to contain
+	API field parsing info, replacing the previous awk_input_field_info_t
+	array.
+	(awk_fieldwidth_info_size): Define macro to calculate size of the
+	variable-length awk_fieldwidth_info_t structure.
+	(awk_input_buf_t): Update get_record prototype to update the type
+	of the final field_width argument from 'const awk_input_field_info_t **'
+	to 'const awk_fieldwidth_info_t **'.
+	* awk.h (set_record): Change 3rd argument from
+	'const awk_input_field_info_t *' to 'const awk_fieldwidth_info_t *'.
+	* io.c (inrec, do_getline_redir, do_getline): Change field_width type
+	from 'const awk_input_field_info_t *' to
+	'const awk_fieldwidth_info_t *'.
+	(get_a_record): Change field_width argument type from
+	'const awk_input_field_info_t **' to 'const awk_fieldwidth_info_t **'.
+	* field.c (api_parser_override): Define new boolean to track whether
+	API parsing is currently overriding default parsing behavior.
+	(api_fw): Change type from 'const awk_input_field_info_t *'
+	to 'const awk_fieldwidth_info_t *'.
+	(FIELDWIDTHS): Change type from 'int *' to 'awk_fieldwidth_info_t *'.
+	(set_record): Use new boolean api_parser_override to track whether
+	API parsing override is in effect, since we can no longer discern
+	this from the value of parse_field -- FIELDWIDTHS parsing uses the
+	same function.
+	(calc_mbslen): New function to calculate the length of a multi-byte 
+	string.
+	(fw_parse_field): Enhance to support the awk_fieldwidth_info_t 
+	structure instead of simply using an array of integer field widths.
+	(api_parse_field): Remove function no longer needed since fw_parse_field
+	now supports both FIELDWIDTHS and API parsing.
+	(set_parser): Use api_parser_override instead of comparing parse_field
+	to api_parse_field.
+	(set_FIELDWIDTHS): Enhance to use new awk_fieldwidth_info_t structure
+	and parse new skip prefix for each field.
+	(current_field_sep): Use api_parser_override flag instead of comparing
+	to api_parse_field.
+	(current_field_sep_str): Ditto.
+
 2017-03-20         Arnold D. Robbins     <arnold@skeeve.com>
 
 	Improve handling of EPIPE. Problems reported by
diff --git a/awk.h b/awk.h
index 1935534c..43d2713c 100644
--- a/awk.h
+++ b/awk.h
@@ -1510,7 +1510,7 @@ extern NODE *get_actual_argument(NODE *, int, bool);
 #endif
 /* field.c */
 extern void init_fields(void);
-extern void set_record(const char *buf, int cnt, const awk_input_field_info_t *);
+extern void set_record(const char *buf, int cnt, const awk_fieldwidth_info_t *);
 extern void reset_record(void);
 extern void rebuild_record(void);
 extern void set_NF(void);
diff --git a/extension/ChangeLog b/extension/ChangeLog
index bb99f9d8..7aeb8aaa 100644
--- a/extension/ChangeLog
+++ b/extension/ChangeLog
@@ -1,3 +1,14 @@
+2017-03-21         Andrew J. Schorr     <aschorr@telemetry-investments.com>
+
+	* readdir_test.c (open_directory_t): Replace field_width array
+	with new awk_fieldwidth_info_t structure. Wrap it in a union so
+	we can allocate the proper size.
+	(dir_get_record): Update field_width type from
+	'const awk_input_field_info_t **' to 'const awk_fieldwidth_info_t **'.
+	Update new fieldwidth parsing info appropriately.
+	(dir_take_control_of): Populate new fieldwidth parsing structure
+	with initial values.
+
 2017-03-09         Andrew J. Schorr     <aschorr@telemetry-investments.com>
 
 	* readdir_test.c (open_directory_t): Update field_width type from an
diff --git a/extension/readdir_test.c b/extension/readdir_test.c
index e023b67c..d21b4e97 100644
--- a/extension/readdir_test.c
+++ b/extension/readdir_test.c
@@ -85,8 +85,12 @@ int plugin_is_GPL_compatible;
 typedef struct open_directory {
 	DIR *dp;
 	char *buf;
-	awk_input_field_info_t field_width[4];
+	union {
+		awk_fieldwidth_info_t fw;
+		char buf[awk_fieldwidth_info_size(3)];
+	} u;
 } open_directory_t;
+#define fw u.fw
 
 /* ftype --- return type of file as a single character string */
 
@@ -170,7 +174,7 @@ get_inode(struct dirent *entry, const char *dirname)
 static int
 dir_get_record(char **out, awk_input_buf_t *iobuf, int *errcode,
 		char **rt_start, size_t *rt_len,
-		const awk_input_field_info_t **field_width)
+		const awk_fieldwidth_info_t **field_width)
 {
 	DIR *dp;
 	struct dirent *dirent;
@@ -207,20 +211,20 @@ dir_get_record(char **out, awk_input_buf_t *iobuf, int *errcode,
 #else
 	len = sprintf(the_dir->buf, "%llu", ino);
 #endif
-	the_dir->field_width[0].len = len;
+	the_dir->fw.fields[0].len = len;
 	len += (flen = sprintf(the_dir->buf + len, "/%s", dirent->d_name));
-	the_dir->field_width[1].len = flen-1;
+	the_dir->fw.fields[1].len = flen-1;
 
 	ftstr = ftype(dirent, iobuf->name);
 	len += (flen = sprintf(the_dir->buf + len, "/%s", ftstr));
-	the_dir->field_width[2].len = flen-1;
+	the_dir->fw.fields[2].len = flen-1;
 
 	*out = the_dir->buf;
 
 	*rt_start = NULL;
 	*rt_len = 0;	/* set RT to "" */
 	if (field_width)
-		*field_width = the_dir->field_width;
+		*field_width = & the_dir->fw;
 	return len;
 }
 
@@ -284,11 +288,12 @@ dir_take_control_of(awk_input_buf_t *iobuf)
 
 	emalloc(the_dir, open_directory_t *, sizeof(open_directory_t), "dir_take_control_of");
 	the_dir->dp = dp;
-	/* pre-populate the field_width array with constant values: */
-	the_dir->field_width[0].skip = 0;	/* no leading space */
-	the_dir->field_width[1].skip = 1;	/* single '/' separator */
-	the_dir->field_width[2].skip = 1;	/* single '/' separator */
-	the_dir->field_width[3].skip = -1;	/* terminate after 3 fields */
+	/* pre-populate the field_width struct with constant values: */
+	the_dir->fw.use_chars = awk_false;
+	the_dir->fw.nf = 3;
+	the_dir->fw.fields[0].skip = 0;	/* no leading space */
+	the_dir->fw.fields[1].skip = 1;	/* single '/' separator */
+	the_dir->fw.fields[2].skip = 1;	/* single '/' separator */
 	size = sizeof(struct dirent) + 21 /* max digits in inode */ + 2 /* slashes */;
 	emalloc(the_dir->buf, char *, size, "dir_take_control_of");
 
diff --git a/field.c b/field.c
index 5ef4d74b..bd333c92 100644
--- a/field.c
+++ b/field.c
@@ -38,6 +38,8 @@ is_blank(int c)
 
 typedef void (* Setfunc)(long, char *, long, NODE *);
 
+/* is the API currently overriding the default parsing mechanism? */
+static bool api_parser_override = false;
 static long (*parse_field)(long, char **, int, NODE *,
 			     Regexp *, Setfunc, NODE *, NODE *, bool);
 /*
@@ -57,9 +59,7 @@ static long sc_parse_field(long, char **, int, NODE *,
 			     Regexp *, Setfunc, NODE *, NODE *, bool);
 static long fw_parse_field(long, char **, int, NODE *,
 			     Regexp *, Setfunc, NODE *, NODE *, bool);
-static long api_parse_field(long, char **, int, NODE *,
-			     Regexp *, Setfunc, NODE *, NODE *, bool);
-static const awk_input_field_info_t *api_fw = NULL;
+static const awk_fieldwidth_info_t *api_fw = NULL;
 static long fpat_parse_field(long, char **, int, NODE *,
 			     Regexp *, Setfunc, NODE *, NODE *, bool);
 static void set_element(long num, char * str, long len, NODE *arr);
@@ -74,7 +74,7 @@ static bool resave_fs;
 static NODE *save_FS;		/* save current value of FS when line is read,
 				 * to be used in deferred parsing
 				 */
-static int *FIELDWIDTHS = NULL;
+static awk_fieldwidth_info_t *FIELDWIDTHS = NULL;
 
 NODE **fields_arr;		/* array of pointers to the field nodes */
 bool field0_valid;		/* $(>0) has not been changed yet */
@@ -262,7 +262,7 @@ rebuild_record()
  * but better correct than fast.
  */
 void
-set_record(const char *buf, int cnt, const awk_input_field_info_t *fw)
+set_record(const char *buf, int cnt, const awk_fieldwidth_info_t *fw)
 {
 	NODE *n;
 	static char *databuf;
@@ -318,11 +318,13 @@ set_record(const char *buf, int cnt, const awk_input_field_info_t *fw)
 	fields_arr[0] = n;
 	if (fw != api_fw) {
 		if ((api_fw = fw) != NULL) {
-			if (parse_field != api_parse_field) {
-				parse_field = api_parse_field;
+			if (! api_parser_override) {
+				api_parser_override = true;
+				parse_field = fw_parse_field;
 				update_PROCINFO_str("FS", "API");
 			}
-		} else if (parse_field != normal_parse_field) {
+		} else if (api_parser_override) {
+			api_parser_override = false;
 			parse_field = normal_parse_field;
 			update_PROCINFO_str("FS", current_field_sep_str());
 		}
@@ -712,83 +714,38 @@ sc_parse_field(long up_to,	/* parse only up to this field number */
 }
 
 /*
- * fw_parse_field --- field parsing using FIELDWIDTHS spec
- *
- * This is called from get_field() via (*parse_field)().
- * This variation is for fields are fixed widths.
+ * calc_mbslen --- calculate the length in bytes of a multi-byte string
+ * containing len characters.
  */
-static long
-fw_parse_field(long up_to,	/* parse only up to this field number */
-	char **buf,	/* on input: string to parse; on output: point to start next */
-	int len,
-	NODE *fs ATTRIBUTE_UNUSED,
-	Regexp *rp ATTRIBUTE_UNUSED,
-	Setfunc set,	/* routine to set the value of the parsed field */
-	NODE *n,
-	NODE *dummy ATTRIBUTE_UNUSED, /* sep_arr not needed here: hence dummy */
-	bool in_middle ATTRIBUTE_UNUSED)
+
+static size_t
+calc_mbslen(char *scan, char *end, size_t len, mbstate_t *mbs)
 {
-	char *scan = *buf;
-	long nf = parse_high_water;
-	char *end = scan + len;
-	int nmbc;
-	size_t mbclen;
-	size_t mbslen;
-	size_t lenrest;
-	char *mbscan;
-	mbstate_t mbs;
 
-	memset(&mbs, 0, sizeof(mbstate_t));
+	size_t mbclen;
+	char *mbscan = scan;
 
-	if (up_to == UNLIMITED)
-		nf = 0;
-	if (len == 0)
-		return nf;
-	for (; nf < up_to && (len = FIELDWIDTHS[nf+1]) != -1; ) {
-		if (gawk_mb_cur_max > 1) {
-			nmbc = 0;
-			mbslen = 0;
-			mbscan = scan;
-			lenrest = end - scan;
-			while (nmbc < len && mbslen < lenrest) {
-				mbclen = mbrlen(mbscan, end - mbscan, &mbs);
-				if (   mbclen == 1
-				    || mbclen == (size_t) -1
-				    || mbclen == (size_t) -2
-				    || mbclen == 0) {
-					/* We treat it as a singlebyte character.  */
-		    			mbclen = 1;
-				}
-				if (mbclen <= end - mbscan) {
-					mbscan += mbclen;
-		    			mbslen += mbclen;
-		    			++nmbc;
-				}
-	    		}
-			(*set)(++nf, scan, (long) mbslen, n);
-			scan += mbslen;
-		} else {
-			if (len > end - scan)
-				len = end - scan;
-			(*set)(++nf, scan, (long) len, n);
-			scan += len;
-		}
+	while (len-- > 0 && mbscan < end) {
+		mbclen = mbrlen(mbscan, end - mbscan, mbs);
+		if (!(mbclen > 0 && mbclen <= (size_t)(end - mbscan)))
+			/*
+			 * We treat it as a singlebyte character. This should
+			 * catch error codes 0, (size_t) -1, and (size_t) -2.
+			 */
+			mbclen = 1;
+		mbscan += mbclen;
 	}
-	if (len == -1)
-		*buf = end;
-	else
-		*buf = scan;
-	return nf;
+	return mbscan - scan;
 }
 
 /*
- * api_parse_field --- field parsing using field widths returned by API parser.
+ * fw_parse_field --- field parsing using FIELDWIDTHS spec
  *
  * This is called from get_field() via (*parse_field)().
+ * This variation is for fields are fixed widths.
  */
-
 static long
-api_parse_field(long up_to,	/* parse only up to this field number */
+fw_parse_field(long up_to,	/* parse only up to this field number */
 	char **buf,	/* on input: string to parse; on output: point to start next */
 	int len,
 	NODE *fs ATTRIBUTE_UNUSED,
@@ -801,26 +758,50 @@ api_parse_field(long up_to,	/* parse only up to this field number */
 	char *scan = *buf;
 	long nf = parse_high_water;
 	char *end = scan + len;
-	int skiplen;
+	const awk_fieldwidth_info_t *fw;
+	mbstate_t mbs;
+	size_t skiplen;
 	size_t flen;
 
+	fw = (api_parser_override ? api_fw : FIELDWIDTHS);
+
 	if (up_to == UNLIMITED)
 		nf = 0;
 	if (len == 0)
 		return nf;
-	while (nf < up_to) {
-		if ((skiplen = api_fw[nf].skip) < 0) {
-		    	*buf = end;
-			return nf;
+	if (gawk_mb_cur_max > 1 && fw->use_chars) {
+		/*
+		 * XXX This may be a bug. Most likely, shift state should
+		 * persist across all fields in a record, if not across record
+		 * boundaries as well.
+		 */
+		memset(&mbs, 0, sizeof(mbstate_t));
+		while (nf < up_to) {
+			if (nf >= fw->nf) {
+				*buf = end;
+				return nf;
+			}
+			scan += calc_mbslen(scan, end, fw->fields[nf].skip, &mbs);
+			flen = calc_mbslen(scan, end, fw->fields[nf].len, &mbs);
+			(*set)(++nf, scan, (long) flen, n);
+			scan += flen;
+		}
+	} else {
+		while (nf < up_to) {
+			if (nf >= fw->nf) {
+				*buf = end;
+				return nf;
+			}
+			skiplen = fw->fields[nf].skip;
+			if (skiplen > end - scan)
+				skiplen = end - scan;
+			scan += skiplen;
+			flen = fw->fields[nf].len;
+			if (flen > end - scan)
+				flen = end - scan;
+			(*set)(++nf, scan, (long) flen, n);
+			scan += flen;
 		}
-		if (skiplen > end - scan)
-			skiplen = end - scan;
-		scan += skiplen;
-		flen = api_fw[nf].len;
-		if (flen > end - scan)
-			flen = end - scan;
-		(*set)(++nf, scan, (long) flen, n);
-		scan += flen;
 	}
 	*buf = scan;
 	return nf;
@@ -1129,7 +1110,7 @@ static void
 set_parser(long (*func)(long, char **, int, NODE *, Regexp *, Setfunc, NODE *, NODE *, bool))
 {
 	normal_parse_field = func;
-	if (parse_field != api_parse_field && parse_field != func) {
+	if (! api_parser_override && parse_field != func) {
 		parse_field = func;
 	        update_PROCINFO_str("FS", current_field_sep_str());
 	}
@@ -1156,7 +1137,7 @@ set_FIELDWIDTHS()
 		return;
 
 	/*
-	 * If changing the way fields are split, obey least-suprise
+	 * If changing the way fields are split, obey least-surprise
 	 * semantics, and force $0 to be split totally.
 	 */
 	if (fields_arr != NULL)
@@ -1166,17 +1147,17 @@ set_FIELDWIDTHS()
 	tmp = force_string(FIELDWIDTHS_node->var_value);
 	scan = tmp->stptr;
 
-	if (FIELDWIDTHS == NULL)
-		emalloc(FIELDWIDTHS, int *, fw_alloc * sizeof(int), "set_FIELDWIDTHS");
-	FIELDWIDTHS[0] = 0;
-	for (i = 1; ; i++) {
+	if (FIELDWIDTHS == NULL) {
+		emalloc(FIELDWIDTHS, awk_fieldwidth_info_t *, awk_fieldwidth_info_size(fw_alloc), "set_FIELDWIDTHS");
+		FIELDWIDTHS->use_chars = awk_true;
+	}
+	FIELDWIDTHS->nf = 0;
+	for (i = 0; ; i++) {
 		unsigned long int tmp;
-		if (i + 1 >= fw_alloc) {
+		if (i >= fw_alloc) {
 			fw_alloc *= 2;
-			erealloc(FIELDWIDTHS, int *, fw_alloc * sizeof(int), "set_FIELDWIDTHS");
+			erealloc(FIELDWIDTHS, awk_fieldwidth_info_t *, awk_fieldwidth_info_size(fw_alloc), "set_FIELDWIDTHS");
 		}
-		/* Initialize value to be end of list */
-		FIELDWIDTHS[i] = -1;
 		/* Ensure that there is no leading `-' sign.  Otherwise,
 		   strtoul would accept it and return a bogus result.  */
 		while (is_blank(*scan)) {
@@ -1194,6 +1175,13 @@ set_FIELDWIDTHS()
 		   or a value that is not in the range [1..INT_MAX].  */
 		errno = 0;
 		tmp = strtoul(scan, &end, 10);
+		if (errno == 0 && *end == ':' && (0 < tmp && tmp <= INT_MAX)) {
+			FIELDWIDTHS->fields[i].skip = tmp;
+			scan = end + 1;
+			tmp = strtoul(scan, &end, 10);
+		}
+		else
+			FIELDWIDTHS->fields[i].skip = 0;
 		if (errno != 0
 		    	|| (*end != '\0' && ! is_blank(*end))
 				|| !(0 < tmp && tmp <= INT_MAX)
@@ -1201,7 +1189,8 @@ set_FIELDWIDTHS()
 			fatal_error = true;
 			break;
 		}
-		FIELDWIDTHS[i] = tmp;
+		FIELDWIDTHS->fields[i].len = tmp;
+		FIELDWIDTHS->nf = i+1;
 		scan = end;
 		/* Skip past any trailing blanks.  */
 		while (is_blank(*scan)) {
@@ -1210,7 +1199,6 @@ set_FIELDWIDTHS()
 		if (*scan == '\0')
 			break;
 	}
-	FIELDWIDTHS[i+1] = -1;
 
 	if (fatal_error)
 		fatal(_("invalid FIELDWIDTHS value, near `%s'"),
@@ -1354,12 +1342,12 @@ choose_fs_function:
 field_sep_type
 current_field_sep()
 {
-	if (parse_field == fw_parse_field)
+	if (api_parser_override)
+		return Using_API;
+	else if (parse_field == fw_parse_field)
 		return Using_FIELDWIDTHS;
 	else if (parse_field == fpat_parse_field)
 		return Using_FPAT;
-	else if (parse_field == api_parse_field)
-		return Using_API;
 	else
 		return Using_FS;
 }
@@ -1369,12 +1357,12 @@ current_field_sep()
 const char *
 current_field_sep_str()
 {
-	if (parse_field == fw_parse_field)
+	if (api_parser_override)
+		return "API";
+	else if (parse_field == fw_parse_field)
 		return "FIELDWIDTHS";
 	else if (parse_field == fpat_parse_field)
 		return "FPAT";
-	else if (parse_field == api_parse_field)
-		return "API";
 	else
 		return "FS";
 }
diff --git a/gawkapi.h b/gawkapi.h
index e744a0fc..1ea067f4 100644
--- a/gawkapi.h
+++ b/gawkapi.h
@@ -119,16 +119,27 @@ typedef enum awk_bool {
 
 /*
  * If the input parser would like to specify the field positions in the input
- * record, it may populate an array of awk_input_field_info_t structures
- * to indicate the location of each field. The 0th array element contains
- * the information about field $1, and the NFth element should set skip
- * to a negative value. For both skip and len, the value should be in
- * bytes, not (potentially multi-byte) characters.
+ * record, it may populate an awk_fieldwidth_info_t structure to indicate
+ * the location of each field. The use_chars boolean controls whether the
+ * field lengths are specified in terms of bytes or potentially multi-byte
+ * characters. Performance will be better if the values are supplied in
+ * terms of bytes. The fields[0].skip value indicates how many bytes (or
+ * characters to skip) before $1, and fields[0].len is the length of $1, etc.
  */
 typedef struct {
-	int skip;	/* # of bytes to skip before field starts */
-	size_t len;	/* # of bytes in field */
-} awk_input_field_info_t;
+	awk_bool_t	use_chars;	/* false ==> use bytes */
+	size_t		nf;
+	struct awk_field_info {
+		size_t	skip;	/* # to skip before field starts */
+		size_t len;	/* length of field */
+	} fields[1];	/* actual dimension should be nf */
+} awk_fieldwidth_info_t;
+/*
+ * This macro calculates the total struct size needed. This is useful when
+ * calling malloc or realloc.
+ */
+#define awk_fieldwidth_info_size(NF) (sizeof(awk_fieldwidth_info_t) + \
+			(((NF)-1) * sizeof(struct awk_field_info)))
 
 /* The information about input files that input parsers need to know: */
 typedef struct awk_input {
@@ -174,7 +185,7 @@ typedef struct awk_input {
 	 */
 	int (*get_record)(char **out, struct awk_input *iobuf, int *errcode,
 			char **rt_start, size_t *rt_len,
-			const awk_input_field_info_t **field_width);
+			const awk_fieldwidth_info_t **field_width);
 
 	/*
 	 * No argument prototype on read_func to allow for older systems
diff --git a/io.c b/io.c
index d1033fcd..25b6bc93 100644
--- a/io.c
+++ b/io.c
@@ -287,7 +287,7 @@ static RECVALUE rsrescan(IOBUF *iop, struct recmatch *recm, SCANSTATE *state);
 
 static RECVALUE (*matchrec)(IOBUF *iop, struct recmatch *recm, SCANSTATE *state) = rs1scan;
 
-static int get_a_record(char **out, IOBUF *iop, int *errcode, const awk_input_field_info_t **field_width);
+static int get_a_record(char **out, IOBUF *iop, int *errcode, const awk_fieldwidth_info_t **field_width);
 
 static void free_rp(struct redirect *rp);
 
@@ -590,7 +590,7 @@ inrec(IOBUF *iop, int *errcode)
 	char *begin;
 	int cnt;
 	bool retval = true;
-	const awk_input_field_info_t *field_width = NULL;
+	const awk_fieldwidth_info_t *field_width = NULL;
 
 	if (at_eof(iop) && no_data_left(iop))
 		cnt = EOF;
@@ -2638,7 +2638,7 @@ do_getline_redir(int into_variable, enum redirval redirtype)
 	NODE *redir_exp = NULL;
 	NODE **lhs = NULL;
 	int redir_error = 0;
-	const awk_input_field_info_t *field_width = NULL;
+	const awk_fieldwidth_info_t *field_width = NULL;
 
 	if (into_variable)
 		lhs = POP_ADDRESS();
@@ -2707,7 +2707,7 @@ do_getline(int into_variable, IOBUF *iop)
 	int cnt = EOF;
 	char *s = NULL;
 	int errcode;
-	const awk_input_field_info_t *field_width = NULL;
+	const awk_fieldwidth_info_t *field_width = NULL;
 
 	if (iop == NULL) {	/* end of input */
 		if (into_variable)
@@ -3676,8 +3676,8 @@ static int
 get_a_record(char **out,        /* pointer to pointer to data */
         IOBUF *iop,             /* input IOP */
         int *errcode,           /* pointer to error variable */
-        const awk_input_field_info_t **field_width)
-				/* pointer to pointer to field_width array */
+        const awk_fieldwidth_info_t **field_width)
+				/* pointer to pointer to field_width info */
 {
 	struct recmatch recm;
 	SCANSTATE state;
diff --git a/test/ChangeLog b/test/ChangeLog
index ad1b35b1..46c52626 100644
--- a/test/ChangeLog
+++ b/test/ChangeLog
@@ -1,3 +1,8 @@
+2017-03-21         Andrew J. Schorr     <aschorr@telemetry-investments.com>
+
+	* Makefile.am (fwtest2b): Add new test of enhanced FIELDWIDTHS syntax.
+	* fwtest2b.awk, fwtest2b.ok: New files.
+
 2017-03-19         Andrew J. Schorr     <aschorr@telemetry-investments.com>
 
 	* Makefile.am (argarray): Always copy argarray.in to the local
diff --git a/test/Makefile.am b/test/Makefile.am
index 855958f1..b46f36e3 100644
--- a/test/Makefile.am
+++ b/test/Makefile.am
@@ -385,6 +385,8 @@ EXTRA_DIST = \
 	fwtest2.awk \
 	fwtest2.in \
 	fwtest2.ok \
+	fwtest2b.awk \
+	fwtest2b.ok \
 	fwtest3.awk \
 	fwtest3.in \
 	fwtest3.ok \
@@ -1221,7 +1223,7 @@ GAWK_EXT_TESTS = \
 	crlf dbugeval dbugeval2 dbugtypedre1 dbugtypedre2 delsub \
 	devfd devfd1 devfd2 dumpvars errno exit \
 	fieldwdth forcenum fpat1 fpat2 fpat3 fpat4 fpat5 fpatnull fsfwfs funlen \
-	functab1 functab2 functab3 fwtest fwtest2 fwtest3 \
+	functab1 functab2 functab3 fwtest fwtest2 fwtest2b fwtest3 \
 	genpot gensub gensub2 gensub3 getlndir gnuops2 gnuops3 gnureops gsubind \
 	icasefs icasers id igncdym igncfs ignrcas2 ignrcas4 ignrcase \
 	incdupe incdupe2 incdupe3 incdupe4 incdupe5 incdupe6 incdupe7 \
@@ -2373,6 +2375,11 @@ arrdbg:
 	@-$(CMP) "$(srcdir)"/$@.ok _$@ && rm -f _$@ "$(srcdir)"/$@.ok
 #	@-$(CMP) "$(srcdir)"/$@.ok _$@ && rm -f _$@ "$(srcdir)"/$@.ok || exit 0
 
+fwtest2b:
+	@echo $@
+	@AWKPATH="$(srcdir)" $(AWK) -f $@.awk  < "$(srcdir)"/fwtest2.in >_$@ 2>&1 || echo EXIT CODE: $$? >>_$@
+	@-$(CMP) "$(srcdir)"/$@.ok _$@ && rm -f _$@
+
 # Targets generated for other tests:
 include Maketests
 
diff --git a/test/Makefile.in b/test/Makefile.in
index c23156df..fe449734 100644
--- a/test/Makefile.in
+++ b/test/Makefile.in
@@ -643,6 +643,8 @@ EXTRA_DIST = \
 	fwtest2.awk \
 	fwtest2.in \
 	fwtest2.ok \
+	fwtest2b.awk \
+	fwtest2b.ok \
 	fwtest3.awk \
 	fwtest3.in \
 	fwtest3.ok \
@@ -1478,7 +1480,7 @@ GAWK_EXT_TESTS = \
 	crlf dbugeval dbugeval2 dbugtypedre1 dbugtypedre2 delsub \
 	devfd devfd1 devfd2 dumpvars errno exit \
 	fieldwdth forcenum fpat1 fpat2 fpat3 fpat4 fpat5 fpatnull fsfwfs funlen \
-	functab1 functab2 functab3 fwtest fwtest2 fwtest3 \
+	functab1 functab2 functab3 fwtest fwtest2 fwtest2b fwtest3 \
 	genpot gensub gensub2 gensub3 getlndir gnuops2 gnuops3 gnureops gsubind \
 	icasefs icasers id igncdym igncfs ignrcas2 ignrcas4 ignrcase \
 	incdupe incdupe2 incdupe3 incdupe4 incdupe5 incdupe6 incdupe7 \
@@ -2809,6 +2811,12 @@ arrdbg:
 	@echo $@
 	@$(AWK) -v "okfile=$(srcdir)/$@.ok" -f "$(srcdir)"/$@.awk | grep array_f >_$@
 	@-$(CMP) "$(srcdir)"/$@.ok _$@ && rm -f _$@ "$(srcdir)"/$@.ok
+#	@-$(CMP) "$(srcdir)"/$@.ok _$@ && rm -f _$@ "$(srcdir)"/$@.ok || exit 0
+
+fwtest2b:
+	@echo $@
+	@AWKPATH="$(srcdir)" $(AWK) -f $@.awk  < "$(srcdir)"/fwtest2.in >_$@ 2>&1 || echo EXIT CODE: $$? >>_$@
+	@-$(CMP) "$(srcdir)"/$@.ok _$@ && rm -f _$@
 Gt-dummy:
 # file Maketests, generated from Makefile.am by the Gentests program
 addcomma:
@@ -4463,7 +4471,6 @@ time:
 	@-$(CMP) "$(srcdir)"/$@.ok _$@ && rm -f _$@
 
 # end of file Maketests
-#	@-$(CMP) "$(srcdir)"/$@.ok _$@ && rm -f _$@ "$(srcdir)"/$@.ok || exit 0
 
 # Targets generated for other tests:
 
diff --git a/test/fwtest2b.awk b/test/fwtest2b.awk
new file mode 100644
index 00000000..5e96c1aa
--- /dev/null
+++ b/test/fwtest2b.awk
@@ -0,0 +1,6 @@
+BEGIN {
+   FIELDWIDTHS = "2:13 2:13 2:13";
+}
+{
+   printf "%s|%s|%s\n", $1, $2, $3
+}
diff --git a/test/fwtest2b.ok b/test/fwtest2b.ok
new file mode 100644
index 00000000..f4d28232
--- /dev/null
+++ b/test/fwtest2b.ok
@@ -0,0 +1,12 @@
+ 0.4867373206| 1.3206333033|-0.2333178127
+ 0.5668176165| 1.3711756314|-0.2193558040
+ 0.4325251781| 1.3399488722|-0.1568307497
+ 0.4900487563| 1.3295759570|-0.2217392402
+-0.6790064191| 1.2536623801|-0.2955415433
+-0.6311440220| 1.2966579993|-0.2246692210
+-0.7209390351| 1.1783407099|-0.2539408209
+-0.6782473356| 1.2495242556|-0.2811436366
+-0.7062054082| 1.1223820964|-1.1619805834
+-0.6491590119| 1.1248946162|-1.0851579675
+-0.7948856821| 1.1208852325|-1.1259821556
+-0.7102549262| 1.1225121126|-1.1475381286