#general
- w
wide-energy-11069
11/02/2015, 6:48 PMI cannot be added to People under reviewers in RB. probably because I am not in pants-review group? Can someone help with that? - w
witty-crayon-22786
11/02/2015, 6:50 PM@wide-energy-11069: unfortunately, our rbcommons account has a limited number of reviewers. we constrain to committers. see http://pantsbuild.github.io/howto_contribute.html#code-review - r
rough-minister-58256
11/02/2015, 6:51 PManyone have time to look at https://rbcommons.com/s/twitter/r/3054/ ? - e
enough-analyst-54434
11/02/2015, 6:54 PMI'll take a look here in ~15 minutes - r
rough-minister-58256
11/02/2015, 6:55 PMappreciate it! - f
fast-megabyte-33579
11/02/2015, 7:34 PMthanks for all the links guys, I'll work through them - h
happy-kitchen-89482
11/02/2015, 7:44 PMUpdate: My recent changes have roughly halved the no-op
time, from ~1.2 seconds to ~0.6 seconds. That's still too long though../pants - e
enough-analyst-54434
11/02/2015, 7:53 PMI'd be hoping for ~10ms but that won't come w/o daemon - what is your presumably 1.0 / pre-daemon goal? - h
happy-kitchen-89482
11/02/2015, 7:55 PMNot sure - h
happy-kitchen-89482
11/02/2015, 7:55 PMNow we're down to the bottleneck being code loading - h
happy-kitchen-89482
11/02/2015, 7:56 PMWe call a lot of register.py (in pantsbuild/pants at least) - h
happy-kitchen-89482
11/02/2015, 7:56 PMand each of those causes the entire backend codebase to be loaded - h
happy-kitchen-89482
11/02/2015, 7:56 PMEven though only a small fraction of that code will actually be executed in a typical run. - h
happy-kitchen-89482
11/02/2015, 7:57 PMBut to fix that we'd need to restructure a lot of stuff - h
happy-kitchen-89482
11/02/2015, 7:57 PMAnd I'm not sure if that's worth it, pre-daemon - b
bored-art-40741
11/02/2015, 7:57 PMIt's almost impossible to avoid that code loading while we still support python BUILD files - b
bored-art-40741
11/02/2015, 7:57 PMYou never properly know what might be needed in a BUILD context - h
happy-kitchen-89482
11/02/2015, 7:57 PMWell, we could at least not load the task code until it's needed - h
happy-kitchen-89482
11/02/2015, 7:58 PMI suspect that's where most of the bloat is - h
happy-kitchen-89482
11/02/2015, 7:58 PMtarget code is typically small and self-contained - b
bored-art-40741
11/02/2015, 7:58 PMBut how would you know, when tasks can request changes to the target roots? - h
happy-kitchen-89482
11/02/2015, 7:58 PMYou can't, in general - b
bored-art-40741
11/02/2015, 7:58 PMI think <1s is fine pre-engine-refactor - b
bored-art-40741
11/02/2015, 7:59 PMAnything else is going to be chasing difficult to notice marginal gains while doing a lot of refactoring that will need to be redone for the engine anyway - h
happy-kitchen-89482
11/02/2015, 7:59 PMI mean, there might be ways, but it's probably not worth the effort - b
bored-art-40741
11/02/2015, 7:59 PMyea - h
happy-kitchen-89482
11/02/2015, 7:59 PMMy point is - when writing the new engine, please think about this simple smile - h
happy-kitchen-89482
11/02/2015, 8:00 PMWe need a way to not load the vast majority of backend code until it's actually truly needed - h
happy-kitchen-89482
11/02/2015, 8:00 PMI did a version of this specifically for the python checkstyle backend, because it was a particularly expensive one. See here: 8802c8c07af07a5fe81ba3f463ebdf1162a62e1b - e
enough-analyst-54434
11/02/2015, 8:00 PMWell, daemon as crack makes this not a concern