3346 2006-06-18 20:14:08 +0000 K-NumericAdd-5, K-NumericAdd-6: floating point assumption wrong 2006-06-27 15:48:06 +0000 1 1 1 Unclassified XML Query Test Suite XML Query Test Suite 0.9.4 PC Windows XP CLOSED FIXED P2 normal --- 1 mike frans.englich simeon oldest_to_newest 10087 0 mike 2006-06-18 20:14:08 +0000 The test is: xs:double(1.1) + xs:double(2.2) eq 3.3 I believe the correct answer is false(). The internal representation of double 1.1 is x3ff199999999999a The exact decimal value of this is 1.100000000000000088817841970012523233890533447265625 This is the correct representation because the value is closer to 1.1 than the next lower value in the value space of double, whose exact decimal value is 1.0999999999999998667732370449812151491641998291015625 The internal representation of double 2.2 is x400199999999999a The exact decimal value of this is 2.20000000000000017763568394002504646778106689453125 This is the correct representation because the value is closer to 2.2 than the next lower value in the value space of double, whose exact decimal value is 2.199999999999999733546474089962430298328399658203125 The internal representation of double 3.3 is x400a666666666666 The exact decimal value of this is 3.29999999999999982236431605997495353221893310546875 This is the correct representation because the value is closer to 3.3 than the next higher value in the value space of double, whose exact decimal value is 3.300000000000000266453525910037569701671600341796875 The result of adding the doubles 1.1 and 2.2 is the double whose internal representation is x400a666666666667, whose exact decimal value is 3.300000000000000266453525910037569701671600341796875 This is not the same as the internal representation of the double 3.3. So 1.1 + 2.2 != 3.3. QED. I believe the same reasoning applies to K-NumericAdd-6, which uses float arithmetic, though I have not done the detailed analysis for this case. 10288 1 frans.englich 2006-06-27 15:25:21 +0000 *** Bug 3382 has been marked as a duplicate of this bug. *** 10289 2 frans.englich 2006-06-27 15:47:34 +0000 K-NumericAdd-5 was changed to: xs:double(1.1) + xs:double(2.2) ne 3.3 K-NumericAdd-6 was changed to: xs:float(1.1) + xs:float(2.2) ne 3.3 Nice analysis. Might very well come in handy in other situations. Since this erport was fixed as suggested, I will change status to CLOSED. Frans