Fix porting inconsistency about rounding to even.

* doc/lispref/numbers.texi (Numeric Conversions, Rounding Operations):
Document that 'round' and 'fround' round to even.
* src/floatfns.c (emacs_rint) [!HAVE_RINT]: Round to even.
This way, the unusual !HAVE_RINT case acts like the usual
HAVE_RINT case, and we can fix the documentation accordingly.

doc/lispref/ChangeLog

+2014-03-19  Paul Eggert  <eggert@cs.ucla.edu>
+
+	Fix porting inconsistency about rounding to even.
+	* numbers.texi (Numeric Conversions, Rounding Operations):
+	Document that 'round' and 'fround' round to even.
+
 2014-03-18  Juanma Barranquero  <lekktu@gmail.com>
 
 	* customize.texi (Variable Definitions): Recommend avoiding

doc/lispref/numbers.texi

@@ -534,8 +534,7 @@ This returns @var{number}, converted to an integer by rounding upward
 @defun round number &optional divisor
 This returns @var{number}, converted to an integer by rounding towards the
 nearest integer.  Rounding a value equidistant between two integers
-may choose the integer closer to zero, or it may prefer an even integer,
-depending on your machine.
+returns the even integer.
 
 @example
 (round 1.2)
@@ -803,6 +802,7 @@ returns that value as a floating-point number.
 @defun fround float
 This function rounds @var{float} to the nearest integral value,
 and returns that value as a floating-point number.
+Rounding a value equidistant between two integers returns the even integer.
 @end defun
 
 @node Bitwise Operations

src/ChangeLog

+2014-03-19  Paul Eggert  <eggert@cs.ucla.edu>
+
+	Fix porting inconsistency about rounding to even.
+	* floatfns.c (emacs_rint) [!HAVE_RINT]: Round to even.
+	This way, the unusual !HAVE_RINT case acts like the usual
+	HAVE_RINT case, and we can fix the documentation accordingly.
+
 2014-03-19  Eli Zaretskii  <eliz@gnu.org>
 
 	* w32fns.c (reset_modifiers): Zero out keystate[] before using it.

src/floatfns.c

@@ -428,7 +428,9 @@ round2 (EMACS_INT i1, EMACS_INT i2)
 static double
 emacs_rint (double d)
 {
-  return floor (d + 0.5);
+  double d1 = d + 0.5;
+  double r = floor (d1);
+  return r - (r == d1 && fmod (r, 2) != 0);
 }
 #endif