From: Boris Zbarsky <bzbarsky@mit.edu>

Date: Tue, 24 Jan 2006 11:03:46 -0600

Message-ID: <43D65DF2.2030105@mit.edu>

To: www-svg@w3.org

Date: Tue, 24 Jan 2006 11:03:46 -0600

Message-ID: <43D65DF2.2030105@mit.edu>

To: www-svg@w3.org

Transform normalization [1] seems to allow significant loss of precision (in this case, square root of 2 to 3 significant digits, wheareas I suspect any reasonable impl would have at least 5 significant digits (that's what 16.16 fixed-point gives you; floating point does much better for square root of 2)). If the desire is to avoid long strings of decimals (as would appear in a real-life implementation of such normalization, I should note), perhaps the example should use a rotation angle with rational sine and cosine (say 90 degrees). But I would prefer the example to have something resembling expected output, assuming either 16.16 fixed or floating point storage, so that people have an idea of what such normalization does. Possibly the example could even list both outputs so that it becomes very clear that such normalization makes the return value completely implementation-dependent. -Boris [1] http://www.w3.org/TR/SVGMobile12/svgudom.html#TransformNormalizationReceived on Tuesday, 24 January 2006 17:03:56 UTC

*
This archive was generated by hypermail 2.3.1
: Wednesday, 8 March 2017 09:47:06 UTC
*