Bug 1275043

3 files changed, 111 insertions, 17 deletions

diff --git a/ChangeLog b/ChangeLog
index de410c1..3137dc5 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,11 @@
+2005-08-29  Kevin Kenny  <kennykb@acm.org>
+
+	* generic/tclBasic.c (ExprMathFunc): Restored "round away from
+	* tests/expr.test (expr-39.*):       zero" behaviour to the
+	                                     "round" function.  Added 
+	test cases for the behavior, including the awkward case of a
+	number whose fractional part is 1/2-1/2ulp. [Bug 1275043]
+	
 2005-08-26  Andreas Kupries <andreask@activestate.com>
 
 	* generic/tclIO.c: Moved Tcl_{Cut,Splice}Channel to
diff --git a/generic/tclBasic.c b/generic/tclBasic.c
index df4abd8..a259130 100644
--- a/generic/tclBasic.c
+++ b/generic/tclBasic.c
@@ -13,7 +13,7 @@
  * See the file "license.terms" for information on usage and redistribution
  * of this file, and for a DISCLAIMER OF ALL WARRANTIES.
  *
- * RCS: @(#) $Id: tclBasic.c,v 1.167 2005/08/24 17:56:23 andreas_kupries Exp $
+ * RCS: @(#) $Id: tclBasic.c,v 1.168 2005/08/29 16:18:59 kennykb Exp $
  */
 
 #include "tclInt.h"
@@ -5425,7 +5425,7 @@ ExprRoundFunc(clientData, interp, objc, objv)
     Tcl_Obj *CONST *objv;	/* Parameter vector */
 {
     Tcl_Obj *valuePtr, *resPtr;
-    double d, a, f;
+    double d, i, f;
 
     /* Check the argument count. */
 
@@ -5448,31 +5448,43 @@ ExprRoundFunc(clientData, interp, objc, objv)
     GET_DOUBLE_VALUE(d, valuePtr, valuePtr->typePtr);
 
     /* 
-     * Round the number to the nearest integer.  I'd like to use rint()
-     * or nearbyint(), but they are far from universal.
+     * Round the number to the nearest integer.  I'd like to use round(),
+     * but it's C99 (or BSD), and not yet universal.
      */
 
-    a = fabs(d);
-    if (a < Tcl_WideAsDouble(LLONG_MAX) + 0.5) {
-	d = valuePtr->internalRep.doubleValue;
-	f = floor(d);
-	d -= f;
-	if (d > 0.5 || (d == 0.5 && fmod(f, 2.0) != 0.0)) {
-	    f = f + 1.0;
+    d = valuePtr->internalRep.doubleValue;
+    f = modf(d, &i);
+    if (d < 0.0) {
+	if (f <= -0.5) {
+	    i += -1.0;
 	}
-	if (f >= (double) LONG_MIN && f <= (double) LONG_MAX) {
-	    TclNewLongObj(resPtr, (long) f);
+	if (i <= Tcl_WideAsDouble(LLONG_MIN)) {
+	    goto tooLarge;
+	} else if (d <= (double) LONG_MIN) {
+	    resPtr = Tcl_NewWideIntObj(Tcl_DoubleAsWide(i));
 	} else {
-	    TclNewWideIntObj(resPtr, Tcl_DoubleAsWide(f));
+	    resPtr = Tcl_NewLongObj((long) i);
+	}			    
+    } else {
+	if (f >= 0.5) {
+	    i += 1.0;
+	}
+	if (i >= Tcl_WideAsDouble(LLONG_MAX)) {
+	    goto tooLarge;
+	} else if (i >= (double) LONG_MAX) {
+	    resPtr = Tcl_NewWideIntObj(Tcl_DoubleAsWide(i));
+	} else {
+	    resPtr = Tcl_NewLongObj((long) i);
 	}
-	Tcl_SetObjResult(interp, resPtr);
-	return TCL_OK;
     }
+    Tcl_SetObjResult(interp, resPtr);
+    return TCL_OK;
 
     /*
      * Error return: result cannot be represented as an integer.
      */
 
+  tooLarge:
     Tcl_SetObjResult(interp, Tcl_NewStringObj(
 	    "integer value too large to represent", -1));
     Tcl_SetErrorCode(interp, "ARITH", "IOVERFLOW",
diff --git a/tests/expr.test b/tests/expr.test
index 6089368..7b7c13b 100644
--- a/tests/expr.test
+++ b/tests/expr.test
@@ -10,7 +10,7 @@
 # See the file "license.terms" for information on usage and redistribution
 # of this file, and for a DISCLAIMER OF ALL WARRANTIES.
 #
-# RCS: @(#) $Id: expr.test,v 1.40 2005/08/24 23:36:56 dkf Exp $
+# RCS: @(#) $Id: expr.test,v 1.41 2005/08/29 16:18:59 kennykb Exp $
 
 if {[lsearch [namespace children] ::tcltest] == -1} {
     package require tcltest 2.1
@@ -6239,6 +6239,80 @@ test expr-38.1 {abs of smallest 32-bit integer [Bug 1241572]} {wideIs64bit} {
     expr {abs(-2147483648)}
 } 2147483648
 
+set tcl_precision 17
+
+test expr-39.1 {round() rounds to +-infinity} {
+    expr round(0.5)
+} 1
+test expr-39.2 {round() rounds to +-infinity} {
+    expr round(1.5)
+} 2
+test expr-39.3 {round() rounds to +-infinity} {
+    expr round(-0.5)
+} -1
+test expr-39.4 {round() rounds to +-infinity} {
+    expr round(-1.5)
+} -2
+test expr-39.5 {round() overflow} {
+    list [catch {expr round(9.2233720368547758e+018)} result] $result
+} {1 {integer value too large to represent}}
+test expr-39.6 {round() overflow} {
+    list [catch {expr round(-9.2233720368547758e+018)} result] $result
+} {1 {integer value too large to represent}}
+test expr-39.7 {round() bad value} {
+    set x trash
+    list [catch {expr {round($x)}} result] $result
+} {1 {argument to math function didn't have numeric value}}
+test expr-39.8 {round() already an integer} {
+    set x 123456789012
+    incr x
+    expr round($x)
+} 123456789013
+test expr-39.9 {round() boundary case - 1/2 - 1 ulp} {
+    set x 0.25
+    set bit 0.125
+    while 1 {
+	set newx [expr { $x + $bit }]
+	if { $newx == $x || $newx == 0.5 } break
+	set x $newx
+	set bit [expr { $bit / 2.0 }]
+    }
+    expr round($x)
+} 0
+test expr-39.10 {round() boundary case - 1/2 + 1 ulp} {
+    set x 0.75
+    set bit 0.125
+    while 1 {
+	set newx [expr { $x - $bit }]
+	if { $newx == $x || $newx == 0.5 } break
+	set x $newx
+	set bit [expr { $bit / 2.0 }]
+    }
+    expr round($x)
+} 1
+test expr-39.11 {round() boundary case - -1/2 - 1 ulp} {
+    set x -0.75
+    set bit 0.125
+    while 1 {
+	set newx [expr { $x + $bit }]
+	if { $newx == $x || $newx == -0.5 } break
+	set x $newx
+	set bit [expr { $bit / 2.0 }]
+    }
+    expr round($x)
+} -1
+test expr-39.10 {round() boundary case - -1/2 + 1 ulp} {
+    set x -0.25
+    set bit 0.125
+    while 1 {
+	set newx [expr { $x - $bit }]
+	if { $newx == $x || $newx == -0.5 } break
+	set x $newx
+	set bit [expr { $bit / 2.0 }]
+    }
+    expr round($x)
+} 0
+
 # cleanup
 if {[info exists a]} {
     unset a