On 10/02/2015 12:16 PM, Fabrice Desré wrote: > You have all the time you want if you don't put dogfooders at risk.
I don't believe we put dogfooders "at risk". Our preliminary testing of the NGA Music app indicated that the memory regression wasn't a significant issue in practice, unless you were testing on a 319MB Flame, and even then things weren't awful. We had several folks using the NGA version on a daily basis, and we only made the switch when a) the app was feature-complete, and b) we were able to use the app ourselves without any major problems. In any case, dogfooders should be expected to suffer. That's the whole point. If you're eating from the master branch, then you are *testing* the software. Among other things, dogfooders help test whether a memory regression in a (fairly-arbitrary) automated test actually impacts real-world usage in a meaningful way. If this isn't the goal of the dogfooding program, then we probably shouldn't be feeding them the master branch. > The problem we have is that most people don't care enough about > having a stable nightly, which is why we haven't updated dogfooders > for more than a month now. This isn't a question of stability. The music app isn't (very*) broken; it's just slower. It's only a matter of optimization (granted, excessive memory usage can lead to OOMs, but we weren't seeing that in our tests). Given my overall experience with attempting to dogfood, I'd love it if the worst problem I had was that an app used more memory. >From what you say, it sounds like you're treating the dogfooding program as more of a beta. Established practice elsewhere in Mozilla (Desktop, Thunderbird, etc) is that the development and beta trees are *separate*. If we want to ship beta-quality software to our dogfooders, then the standard Mozilla way of doing so is to let our patches ride the trains so they have time to get all the kinks out. I know we've been trying (and failing) to get Firefox OS to ride the trains for a long time now, but we don't have to copy Desktop. *Anything* that lets us provide a frequently-updated channel where patches have had some baking time would be a boon. Generally, I think we should be landing stuff to master as quickly as possible. It helps reduce merge conflicts, makes it easier for QA to test, and lets the more-adventurous dogfooders see the latest-greatest. You mentioned that we could have created a second app for people to try. We did. No one used it**. That also makes development harder, since switching out the apps will bitrot every in-progress patch. In the end, we need to have a release management strategy compatible with our quality expectations for the dogfood channel. From what you've said, it sounds like master isn't the best branch to be pulling from. - Jim * There are some significant bugs, but nothing that's caused me any serious stress, despite using the music app on a regular basis. ** Aside from the NGA Music developers and a couple of other people who tried it out. _______________________________________________ dev-fxos mailing list [email protected] https://lists.mozilla.org/listinfo/dev-fxos
