Now that all fields are NUL-terminated, we can eliminate $n copying in the interpreter.

2 files changed, 12 insertions, 6 deletions

diff --git a/ChangeLog b/ChangeLog
index 2699ef0c..1eb2fa0a 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,5 +1,15 @@
 2016-07-06         Andrew J. Schorr     <aschorr@telemetry-investments.com>
 
+	* interpret.h (Op_field_spec): Now that all $n field values are
+	NUL-terminated, there is no reason to call dupnode for $n where n > 0.
+	This saves malloc and copying overhead, thereby more than offsetting the
+	performance hit of the additional copying and NUL-termination in the
+	last patch to field.c. It also eliminates repeated parsing in cases
+	where $n, for n > 1, was accessed more than once in a numeric context,
+	so the new approach should be a performance win.
+
+2016-07-06         Andrew J. Schorr     <aschorr@telemetry-investments.com>
+
 	Make sure that all field values, and therefore all strings inside gawk,
 	are terminated with a '\0' character!
 	* field.c (databuf): New static struct to hold info about our buffer to
diff --git a/interpret.h b/interpret.h
index bbddd5a7..106367f7 100644
--- a/interpret.h
+++ b/interpret.h
@@ -363,12 +363,8 @@ uninitialized_scalar:
 			lhs = r_get_field(t1, (Func_ptr *) 0, true);
 			decr_sp();
 			DEREF(t1);
-			/* only for $0, up ref count */
-			if (*lhs == fields_arr[0]) {
-				r = *lhs;
-				UPREF(r);
-			} else
-				r = dupnode(*lhs);
+			r = *lhs;
+			UPREF(r);
 			PUSH(r);
 			break;