Re: [blink-dev] PSA: Risky changes should be guarded by a Runtime Enabled Feature

2023-04-05 Thread Rick Byers
Right. I've heard some debate and confusion on this. While performance work and major new user-facing browser features tend to go through finch-based rollout and A/B testing to validate quality, platform API work tends instead to rely on developers being in control of their own A/B testing as they

Re: [blink-dev] PSA: Risky changes should be guarded by a Runtime Enabled Feature

2023-04-04 Thread Charles Harrison
For me, I really just needed the link to this guidance which you posted: https://chromium.googlesource.com/chromium/src/+/main/docs/flag_guarding_guidelines.md#what-type-of-flag-rollout-to-use This positions the "default" rollout to be enabled-by-default with a killswitch, rather than a finch-enab

Re: [blink-dev] PSA: Risky changes should be guarded by a Runtime Enabled Feature

2023-04-04 Thread Rick Byers
Oh that's interesting. It's possible there are cases where disabled-by-default is the better choice. What do you see as the main trade offs? Obviously it's important that we're testing the default configuration - so either way, we shouldn't be relying on finch to change behavior except in the emerg

Re: [blink-dev] PSA: Risky changes should be guarded by a Runtime Enabled Feature

2023-04-04 Thread Charles Harrison
Thanks Rick this is really helpful. I have been involved in a few changes where API OWNERS said "flag-gated" and the implementer chose to use a disabled-by-default flag which I didn't push back on. I think just I needed this recalibration :) On Tue, Apr 4, 2023 at 1:16 PM Rick Byers wrote: > Tha

Re: [blink-dev] PSA: Risky changes should be guarded by a Runtime Enabled Feature

2023-04-04 Thread Rick Byers
Thanks Charlie. Yes I should have been specific, I'm really talking about enabled-by-default flags here (eg. Runtime Enabled Feature with status=stable). Note that our process for flipping a kill switch via finch includes first landing a CL that disables the flag (confirming it doesn't break any t

Re: [blink-dev] PSA: Risky changes should be guarded by a Runtime Enabled Feature

2023-03-29 Thread Charles Harrison
Thanks for posting this Rick. One call-out that I think is worth mentioning is the distinction between enabled-by-default flags and disabled-by-default flags. The general guidelines for choosing between these is in the links you posted, but given that it can be sometimes a subjective call it might

[blink-dev] PSA: Risky changes should be guarded by a Runtime Enabled Feature

2023-03-29 Thread Rick Byers
Hey blink-dev, Within Google we’ve had a lot of discussion over the past year about how we should make increasing use of flags and kill switches to reduce the risk of breaking changes, but I realized there has