Trybot should compile all affected build targets even the target is not for test |
||
Issue descriptionMany of tree close due to compile failure on main waterfall were prevented if we compile all affected targets correctly. I think that will reduce sheriff's burden a lot. e.g. https://chromium-review.googlesource.com/c/chromium/src/+/1313170 https://chromium-review.googlesource.com/c/chromium/src/+/1292830 Can we always compile all in compile step or compile all executable affected by changes? We can extract all executables via `gn ls <build dir> --type=executable`. So how do you think extracting executable targets and use that as additional compile targets in 'analyze' steps?
,
Nov 2
I would need to be convinced on this; my guess is that it wouldn't be worth it. How often do we see failures like this? How much of sheriffs' burden is in addressing compile failures? How much would this cost us in cycle time? I'd guess the frequency and burden of compile failures like this is low -- findit found both CLs you linked as examples quickly -- while the impact on compile times would be high.
,
Nov 2
Historically, we've compiled 'all' on the configurations that would be able to do so and stay within the SLO (and we still do this). I think it would be fine to compile 'all' if it is worth it (catches enough failures to justify the additional cost) and we still stay within the SLO. To figure out the cost, you'd have to (a) figure out how often we get a failure on the waterfall, and how much slower it'd make the CQ. It would be okay to do some rough estimating here.
,
Nov 3
Like dirk and john point out, we'd need some analysis on how many more CLs this would likely catch vs the expected CQ slowdown. I doubt anyone around is motivated enough or has enough time to do that, so moving this into our queue. |
||
►
Sign in to add a comment |
||
Comment 1 by tikuta@chromium.org
, Nov 2