Re: WebPlatform Browser Support Info

On Oct 19, 2013, at 12:15 PM, Ronald Mansveld <ronald@ronaldmansveld.nl> wrote:

> The data-exchange-format we use to transfer data is separated from the API that will be exposed in the end. I feel the data-exchange itself should be as accurate as possible, for, like Doug mentioned, it is possible to trim down layers, while we cannot add layers of accuracy.
> This goes especially for test-results: A test is run on a specific build of a UA on a specific OS (or firmware). Even though we as humans might have the knowledge that that very test-result can be extrapolated to other builds of the browser, or the same build of the browser on a different OS or different platform, does not mean that that other build has been actually tested. Therefore, scientifically speaking, a test-result for that other build is non-existent, and thus should not be in the data-exchange.
> Moreover: even though until now a test-result on build X means the feature will also work on build Y, doesn't mean this will hold true in the future. So, if the decision is made to collapse test-results in the data-exchange, you can never accommodate for such an event, unless you mark it up in a note with the test-result.
> To conclude: I strongly feel that test-results should really be about the specific version/build of a UA, where parties displaying that information can choose for themselves whether or not to extrapolate that data to other builds/versions. We should not decide that for them.

> For now the focus is on combining the data from several sources (quirksmode, html5test, mobilehtml5, caniuse) to provide as broad and accurate as possible test-results. 

I think there is a fundamental misunderstanding about the nature of the data that we actually have.


When I look at data from Caniuse, I do not see test results for individual browsers, but a general statement that for example "Firefox 20" supports a feature:
https://github.com/Fyrd/caniuse/blob/master/features-json/pagevisibility.json

When I look at the tables at Quirksmode I can see that "IE 9 and lower"  does not support a particular feature. I don't see individual tests for each version of IE:
http://quirksmode.org/css/images/

When I look at Mobile HTML 5 I can see that "Kindle Fire 1.0 to 2.0" support a feature. Again no individual tests:
http://mobilehtml5.org/#

And as for my own site: HTML5test. I am probably the only one to have both an enormous amount of raw data and conclusions. But at this time I am not comfortable to share the raw data. The data I am willing to share at this time are the verified conclusions, very similar to the data from the other sources.


When we are going to reuse data from these sources we need to let go of the idea of individual test results, but start thinking about exchanging and storing conclusions. Right now, that is the data we actually have.

If we created a way to exchange individual test results right now, we will not have any data to exchange. You could add individual test results later on and build your own conclusions based on that data. I can definitely understand the desire to have raw data and build your own conclusions. And it absolute fine to talk about how to store and exchange individual test results, but I really think we should focus on what we can do right now with the data we have right now.

And right now, the data we have are conclusions.

Cheers,

Niels
html5test.com