- From: Dirk Schulze <dschulze@adobe.com>
- Date: Thu, 4 Sep 2014 19:40:01 +0000
- To: Boris Zbarsky <bzbarsky@MIT.EDU>
- CC: "public-fx@w3.org" <public-fx@w3.org>
On Sep 4, 2014, at 9:22 PM, Boris Zbarsky <bzbarsky@MIT.EDU> wrote: > On 9/4/14, 2:41 PM, Dirk Schulze wrote: >> The question is, do we want to specify the algorithm in detail > > That's a separate issue. I just wanted us to be clear that not specifying one means observably different behavior, and not because UAs end up implementing incorrectly but just because that's the nature of the beast. These questions seems to be related. If you think it is necessary to avoid variances across browsers, then we need to specify the used algorithm. Not just for inverse() but all transformation functions. Even a simple translate could change the numerical output and therefore the result of an inverse() call afterwards. That said, with a rotation you won’t for sure get consistent results. The results could differ on the same browser and the same operation system but different hardware. IEEE did not require the same accuracy for trigonometric functions as it did for addition and subtraction. What I want to say is: variances are inevitable. Therefore, what ever algorithm a user agent choses, it should be optimized for the system it runs on and not for accuracy. Greetings, Dirk > > -Boris
Received on Thursday, 4 September 2014 19:40:32 UTC