- From: Stefan Håkansson LK <stefan.lk.hakansson@ericsson.com>
- Date: Thu, 11 Sep 2014 17:55:05 +0000
- To: cowwoc <cowwoc@bbs.darktech.org>, "public-media-capture@w3.org" <public-media-capture@w3.org>
On 2014-09-11 19:48, cowwoc wrote: > On 11/09/2014 12:21 PM, bugzilla@jessica.w3.org wrote: >> https://www.w3.org/Bugs/Public/show_bug.cgi?id=26526 >> >> Jan-Ivar Bruaroey [:jib] <jib@mozilla.com> changed: >> >> What |Removed |Added >> ---------------------------------------------------------------------------- >> CC| |jib@mozilla.com >> >> --- Comment #3 from Jan-Ivar Bruaroey [:jib] <jib@mozilla.com> --- >> I think the existing epsilon covers inaccuracies in double just fine, and that >> arguments to change it were in the opposite direction, chasing the problem of >> interpreting expectations correctly of people entering decimals by hand. >> >> I believe Harald proposed 1/1000 [1], and I indicated perhaps 1/100 as being >> better, citing wikipedia [2] as evidence that accuracy-needs on aspect don't >> rise with higher resolutions. >> >> [1] http://lists.w3.org/Archives/Public/public-media-capture/2014Aug/0056.html >> [2] http://lists.w3.org/Archives/Public/public-media-capture/2014Aug/0084.html >> > > I believe your reasoning is still flawed. You shouldn't evaluate the > "goodness" of an epsilon by looking at a handful of resolutions as > you've done. You need to evaluate the epsilon against the entire > resolution space from 1x1 through 16k. > > For example, what happens when a user asks for an aspect ratio of 8/12? > He might be aiming for 800x1200 but you'll give him 788x1200 for an > epsilon of 1/100. That's a huge error margin. I agree, and I think we could go with Cullen's proposal, i.e. the spec default (which currently means 1e-10. Everyone has a calculator at hand I believe, so getting 10 decimals should not be hard.
Received on Thursday, 11 September 2014 17:55:32 UTC