[webkit-dev] New runtime setting: css3TextDecoration

[Julien Chaffraix]

On Tue, Aug 14, 2012 at 12:06 PM, Ryosuke Niwa <rniwa at webkit.org> wrote:
> On Tue, Aug 14, 2012 at 11:56 AM, Julien Chaffraix
> <julien.chaffraix at gmail.com> wrote:
>>
>> > I don't think it's appropriate to add settings for CSS features that are
>> > under development,
>> > for a number of reasons:
>> >
>> > * If we did this for every feature, we'd end up with hundreds of
>> > Settings.
>> > * Traditionally, Settings don't tend to get removed, resulting in an
>> > ever-accumulating number of Settings.
>>
>> ENABLE has a slightly better track of record but I don't think we
>> should push back on runtime flags just because of that.
>
>
> Having a runtime flag incurs runtime cost.

Performance is one of our core goal and any WebKit hacker will agree
that it's important. Here you forgot to mention that we care about is
if it has a *notable* runtime cost in which case there is no proof
that the current flags incur such a cost. We know of patches from
feature under a flag that have been rolled out because of measurable
performance regression [1] so we would know if that was the case. It's
also better to catch such slow-down earlier rather than later.

The most significant performance impact is probably the increase in
the binary size which I won't deny and don't have insights into.

>> Having tons of #ifdef's is a lot more worrying from my perspective,
>
>
> Having a runtime flag is significantly worse in that it affects end users.
> Having a compile flag is a painful for developers but has absolutely zero
> cost for end users.

You will need to define "end users" here. WebKit is meant to be
embedded and as such you can think of different end users. People
embedding WebKit are expected to know about our flags. Browsers' end
users are only impacted insofar as browsers vendors implement the
right UI and this claim is making strong assumption about how this is
done.

Now, both options with respect to the flag have their trade-offs but
there seems to be majority in favor of ENABLE flags. Brushing aside
one option based on FUD is not really something I am supportive of
though.

Thanks,
Julien

[1] for example, see https://bugs.webkit.org/show_bug.cgi?id=76064