From b46368059445b51d28d8899ffb8f457c1642b6cb Mon Sep 17 00:00:00 2001
From: Cody Peter Mello <melloc@writev.io>
Date: Fri, 19 Oct 2018 15:07:53 -0700
Subject: [PATCH] Update field-splitting behaviour to match POSIX definition

---
 lib.c   | 24 +++++++++++++++++-------
 proto.h |  1 +
 tran.c  |  2 ++
 3 files changed, 20 insertions(+), 7 deletions(-)

diff --git a/lib.c b/lib.c
index a365245..00359ac 100644
--- a/lib.c
+++ b/lib.c
@@ -104,6 +104,22 @@ void initgetrec(void)
 	infile = stdin;		/* no filenames, so use stdin */
 }
 
+/*
+ * POSIX specifies that fields are supposed to be evaluated as if they were
+ * split using the value of FS at the time that the record's value ($0) was
+ * read.
+ *
+ * Since field-splitting is done lazily, we save the current value of FS
+ * whenever a new record is read in (implicitly or via getline), or when
+ * a new value is assigned to $0.
+ */
+void savefs(void)
+{
+	if (strlen(getsval(fsloc)) >= sizeof (inputFS))
+		FATAL("field separator %.10s... is too long", *FS);
+	strcpy(inputFS, *FS);
+}
+
 static int firsttime = 1;
 
 int getrec(char **pbuf, int *pbufsize, int isrecord)	/* get next input record */
@@ -122,6 +138,7 @@ int getrec(char **pbuf, int *pbufsize, int isrecord)	/* get next input record */
 	if (isrecord) {
 		donefld = 0;
 		donerec = 1;
+		savefs();
 	}
 	saveb0 = buf[0];
 	buf[0] = 0;
@@ -191,10 +208,6 @@ int readrec(char **pbuf, int *pbufsize, FILE *inf)	/* read one record into buf *
 	int bufsize = *pbufsize;
 	char *rs = getsval(rsloc);
 
-	if (strlen(getsval(fsloc)) >= sizeof (inputFS))
-		FATAL("field separator %.10s... is too long", *FS);
-	/*fflush(stdout); avoids some buffering problem but makes it 25% slower*/
-	strcpy(inputFS, *FS);	/* for subsequent field splitting */
 	if ((sep = *rs) == 0) {
 		sep = '\n';
 		while ((c=getc(inf)) == '\n' && c != EOF)	/* skip leading \n's */
@@ -284,9 +297,6 @@ void fldbld(void)	/* create fields from current record */
 	}
 	fr = fields;
 	i = 0;	/* number of fields accumulated here */
-	if (strlen(getsval(fsloc)) >= sizeof (inputFS))
-		FATAL("field separator %.10s... is too long", *FS);
-	strcpy(inputFS, *FS);
 	if (strlen(inputFS) > 1) {	/* it's a regular expression */
 		i = refldbld(r, inputFS);
 	} else if ((sep = *inputFS) == ' ') {	/* default whitespace */
diff --git a/proto.h b/proto.h
index ad6f2e8..e44370d 100644
--- a/proto.h
+++ b/proto.h
@@ -116,6 +116,7 @@ extern	void	recinit(unsigned int);
 extern	void	initgetrec(void);
 extern	void	makefields(int, int);
 extern	void	growfldtab(int n);
+extern	void	savefs(void);
 extern	int	getrec(char **, int *, int);
 extern	void	nextfile(void);
 extern	int	readrec(char **buf, int *bufsize, FILE *inf);
diff --git a/tran.c b/tran.c
index d1dfe2b..ca103f9 100644
--- a/tran.c
+++ b/tran.c
@@ -318,6 +318,7 @@ Awkfloat setfval(Cell *vp, Awkfloat f)	/* set float val of a Cell */
 	} else if (isrec(vp)) {
 		donefld = 0;	/* mark $1... invalid */
 		donerec = 1;
+		savefs();
 	} else if (vp == ofsloc) {
 		if (donerec == 0)
 			recbld();
@@ -362,6 +363,7 @@ char *setsval(Cell *vp, const char *s)	/* set string val of a Cell */
 	} else if (isrec(vp)) {
 		donefld = 0;	/* mark $1... invalid */
 		donerec = 1;
+		savefs();
 	} else if (vp == ofsloc) {
 		if (donerec == 0)
 			recbld();