- From: Alan Stearns <stearns@adobe.com>
- Date: Tue, 27 Jan 2015 19:32:26 +0000
- To: Zack Weinberg <zackw@panix.com>, Tab Atkins Jr. <jackalmage@gmail.com>
- CC: Florian Rivoal <florian@rivoal.net>, "www-style@w3.org" <www-style@w3.org>, Randy Edmunds <redmunds@adobe.com>
On 1/27/15, 11:00 AM, "Zack Weinberg" <zackw@panix.com> wrote: > >On Tue, Jan 27, 2015 at 1:19 PM, Tab Atkins Jr. <jackalmage@gmail.com> >wrote: >> On Tue, Jan 27, 2015 at 12:04 AM, Florian Rivoal <florian@rivoal.net> >>wrote: >>>> On 26 Jan 2015, at 23:55, Alan Stearns <stearns@adobe.com> wrote: >>>> [...] >>>> the deviation from the average >>>> [...] >>> >>> Is this intentionally left vague, opening differences in terms of >>>quality of implementation? >> >> That's not vague. Deviation is a numerical value that you can compare >> across lines. It's just linear difference. >> >>> I would clarify 'deviation' into 'standard deviation'. Compared to the >>>mean absolute deviation, the std dev is preferable, as it punishes >>>extreme values more. For example, on a four line block, if you have >>>between each line being 1 character off the average, or 2 lines being >>>exactly at the average and 2 lines being off by 2 characters, the >>>standard deviation would prefer the first situation, while the mean >>>absolute deviation would be indifferent. >> >> I agree that minimizing the squared deviation is better. It's almost >> always better for these kinds of "get close to the average" things. >> >> ~TJ >> > >I am concerned that this is getting to the level of detail where we >might end up (possibly implicitly) requiring the use of a specific >algorithm. I don't think that would be appropriate here. > I don’t think that changing the method of evaluation from 'deviation' to 'standard deviation' gets us too close to specifying an algorithm (which I would definitely resist). I’ve made that change, noted that balancing comes before justification adjustments, and added a clause that balancing occurs *only if* it’s possible to get better balance than normal line breaking. There will be some basic, testable cases where all reasonable attempts at balancing will succeed. But line-breaking algorithms differ and it’s easy to assemble content into difficult cases for balancing. So I assume there will be middle-difficulty cases where one browser balances better than another, and tough cases where one browser might find a way to balance where others fail. Thanks, Alan
Received on Tuesday, 27 January 2015 19:33:00 UTC