Platform/Layout/CSS Compatibility

CSS vendor-prefix compatibility

problem statement

Sites that have WebKit-specific content and back-up content for everyone else.

-webkit- properties are used so much on mobile content in particular that non-WebKit browsers face a Prisoner's Dilemma problem, analogous to past quirks battles (e.g. 2003-4 era innerHTML and undetected document.all).

data: https://bugzilla.mozilla.org/show_bug.cgi?id=708406

more problem details

Blog posts:

• http://hsivonen.iki.fi/vendor-prefixes/
• http://blogs.msdn.com/b/iemobile/archive/2010/05/10/javascript-and-css-changes-in-ie-mobile-for-windows-phone-7.aspx in particular:

  "We've also added support for the -webkit-text-size-adjust CSS selector. This selector allows you to control how text on the Web page is scaled to increase readability for the user (you can also use -ms-text-size-adjust, IE Mobile recognizes both).

  ... [one day later] ...

  "[Update 05/11/2010: based on community feedback, we will only be implementing the -ms- prefix, not the -webkit- one.]"

goals

The underlying open web goal:

• Open up the webkit-specific part of the web to other vendors in the same way that we had to be practical about what IE-proprietary or IE-only technologies to support.

straw proposal

• Consider implementing some -webkit- prefixed properties.

straw proposal downsides

• Unfortunately for the open web, implementing a -webkit- prefixed property (outside of WebKit) will nearly legitimize (make people assume they'll work forever) the use of -webkit- prefixed properties.
• ...

possible downside mitigation

• In the short term we can at least remove pain for web authors and users.
• In the long term we can ensure the unprefixed properties (in CR drafts) work and encourage authors to switch to them.
• ...

questions and methodology

• What are the thresholds (even approximate) for supporting an other-vendor prefixed property vs. not?
• How should we consider occurrence counts of -webkit- properties?
  • Weighted by PageRank or equivalent?
• Consider usability of page with/without the feature, not just how often it is used. E.g. tap-highlight-color does not affect the user's ability to use a website the same way text-size-adjust does.

parsing other vendor prefixes approaches

• parse other vendor prefixed properties only in conjunction with parsing the equivalent unprefixed properties
• only do it for environments where critically necessary, i.e. mobile not desktop, to encourage use of standard equivalents.

unprefixing principles

• unprefixing things early (before CR) should be an exceptional case
  • what is the methodology for "exceptional" unprefixing?
• unprefixing things must be evaluated carefully on case-by-case basis.
• unprefixing is not something to do routinely just to "go faster" by a few months.
  • put the energy first into contributing and passing test suites instead.
• ...

Data on vendor-specific prefixes

Here's a summary of the data collection and analysis that has been conducted regarding the use of various CSS vendor-specific prefixes.

The current datasets, collected by John Jensen, are:

Initial CSS properties dataset

• Completed in November 2011
• Summary of 88,000 CSS files from top 30,000 sites on the web, collected using Desktop FF 8.0 User-Agent
• Tables and summary reports in https://bugzilla.mozilla.org/show_bug.cgi?id=708406
• Written report and summary tables are attached to the ticket.
• Summary file, in CSV format, is 620MB compressed, 7.2GB uncompressed, available at http://people.mozilla.com/~jjensen/css.csv.gz

Q and A

• how many sites in your mobile Webkit browser crawl use at least one of 'transition', 'transition-timing-function', 'transition-duration', 'transition-property', 'transition-delay' (ignoring prefixes)?

1245 / 30087 = 4.13%

• how many use them only with -webkit prefixes (no -moz or unprefixed versions of the properties)?

336 / 30087 = 1.12%

• how many use them only with -webkit prefixes and unprefixed (no -moz versions of the properties)?

365 / 30087 = 1.21%

• For each CSS prefix for which there are both -moz- and -webkit- prefixes, what percentage of domains host CSS that uses only the -webkit- version and not the -moz- or unprefixed version?

text-size-adjust	510	1.70%
box-shadow	428	1.42%
border-radius	412	1.37%
appearance	379	1.26%
font-smoothing	285	0.95%
tap-highlight-color	250	0.83%
transform	75	0.25%
border-top-left-radius	72	0.24%
border-top-right-radius	72	0.24%
transition-duration	61	0.20%
animation-duration	56	0.19%
animation-name	56	0.19%
border-bottom-left-radius	55	0.18%
border-bottom-right-radius	55	0.18%
transition-property	49	0.16%
animation-iteration-count	45	0.15%
padding-start	45	0.15%
background-size	43	0.14%
animation-timing-function	42	0.14%
box-sizing	42	0.14%

Larger, as-yet-unprocessed datasets

• Raw data downloading completed in mid-January 2012, using these UAs:

1. latest Android Native Browser from ICS
2. latest Mobile Safari UA

• Includes all HTML, Javascript, CSS files in compressed format
• Roughly 1.1m files downloaded for each UA
• CSS file parsing is underway to produce more data