One of the core design principles of Sass has always been to understand CSS as little as possible. As a CSS preprocessor of course we have to understand the syntax of CSS, but as much as we can we try to avoid caring about the semantics—the meaning behind the styles. This means that Sass has no idea which properties are valid, which HTML elements actually exist, or even to a large extent what the syntax of most @-rules is.
We get a lot of benefit from this. The less built-in knowledge Sass has about CSS, the less likely it is to work poorly with new CSS features. Imagine having to file a feature request every time you want to use a new CSS property—that would suck! Instead, older versions of Sass will happily keep working unless the actual syntax changes, which is much rarer.
Because of this decoupling, we've never needed to worry much about browser compatibility. Sass just passes whatever CSS its given on through. It's up to the user to determine what works where, which gives them a lot of flexibility and gives us designers one fewer tough decision to make.
But despite this general policy, there are always a few cases where CSS knowledge turns out to be necessary. A big one is
@extend, which needs to know a lot about the meaning of selectors to properly unify them and weed out duplicates. Property values sometimes require semantic knowledge as well—we have to know how to interpret colors, for example.
One of those cases has leapt up to bite us. Long ago, we made the decision to always emit transparent colors as the keyword
transparent, because it was supported on IE6 through 8 and the alternative
rgba() syntax was not. But it turns out that the opposite is true for more recent versions: in IE10,
:hover styles aren't triggered for elements with
background-color: transparent but they are with
background-color: rgba(0, 0, 0, 0). Thanks, IE!
So we were faced with a dilemma. Keep the existing behavior which was compatible with old outdated browsers that no one uses, or choose a new behavior that works better with modern browsers? The choice was pretty clear: we decided to always emit
rgba(0, 0, 0, 0).
In addition, we wanted to come up with a general rule to guide us in determining which browsers we were willing to consider outdated, and which we would continue to support (whatever that meant for the behavior in question). We decided that if a change would negatively affect less than 2% of the global market share of browsers according to StatCounter GlobalStats, we were willing to make it.
This limit isn't set in stone. We reserve the right to change it in the future, and to make individual decisions that may affect more browsers. But this is the general guideline we're paying attention to, and we wanted you all to know.