https://pantsbuild.org/ logo
Powered by
#general
  • r

    rough-minister-58256

    07/30/2015, 6:34 PM
    re: general performance, not sure if you guys have seen https://lwn.net/Articles/646888/ but python 2.7.11 is supposed to improve general python performance by 15-20%.
  • h

    happy-kitchen-89482

    07/30/2015, 6:34 PM
    but I think john and I were expressing our support for the long-term goal of the daemon actually doing all the work
  • r

    rough-minister-58256

    07/30/2015, 6:34 PM
    beyond pantsd, think there are also some opportunities for perf optimization - we have done alot of research around that but mainly focusing on the biggest consumers.
  • h

    happy-kitchen-89482

    07/30/2015, 6:34 PM
    instead of passing serialized build graphs back to the cli
  • e

    enough-analyst-54434

    07/30/2015, 6:34 PM
    Yes, fast cli response is probably alot more critical than py perf enhancement
  • r

    rough-minister-58256

    07/30/2015, 6:35 PM
    my ultimately goal is to make pants as fast as humanly possible and get it competitive w/ buck/bazel et al
  • e

    enough-analyst-54434

    07/30/2015, 6:35 PM
    basically offline IO 1st, cli 2nd, rest 3rd
  • r

    rough-minister-58256

    07/30/2015, 6:36 PM
    I think there’s a definite play for that at the intersection of the engine rewrite + pantsd 2.0
  • r

    rough-minister-58256

    07/30/2015, 6:37 PM
    or sooner, potentially
  • r

    rough-minister-58256

    07/30/2015, 6:39 PM
    that==daemon doing all the work
  • h

    happy-kitchen-89482

    07/30/2015, 6:42 PM
    Yes
  • h

    happy-kitchen-89482

    07/30/2015, 6:42 PM
    hopefully the engine rewrite will at least make "all the work" something that's easy to bundle up
  • r

    rough-minister-58256

    07/30/2015, 6:45 PM
    sure, tho hijacking the main cli entry point and using a thin client approach also seems feasible prior to that if we want to go that route
  • r

    rough-minister-58256

    07/30/2015, 6:45 PM
    we could prototype it as an experiment
  • h

    happy-kitchen-89482

    07/30/2015, 6:52 PM
    true
  • e

    enough-analyst-54434

    07/30/2015, 7:00 PM
    @rough-minister-58256: so in your 0.0.39 / pex 1.0.1 upgrade did you have no users affected by the PEX-INFO trampling (https://rbcommons.com/s/twitter/r/2517/) or did you manually rig up a fixed pex package for internal use?
  • r

    rough-minister-58256

    07/30/2015, 7:31 PM
    @enough-analyst-54434: not sure it was specifically addressed; think we were generally trying to charge forward w/ the release due to other benefits and this was likely ignored due to no apparent squeaky wheels + the ability to workaround
  • r

    rough-minister-58256

    07/30/2015, 7:31 PM
    speaking of, let me ping wickman again on this one
  • r

    rough-minister-58256

    07/30/2015, 7:32 PM
    fwiw, we’ve just rolled back .39 internally
  • e

    enough-analyst-54434

    07/30/2015, 7:32 PM
    Thanks - I was hoping aurora re-deploys of not-zip-safe webapps might be the motivating factor to get Wickman shipping and releasing
  • r

    rough-minister-58256

    07/30/2015, 7:32 PM
    heh
  • h

    happy-kitchen-89482

    07/30/2015, 8:49 PM
    Question to the JVM shops out there:
  • h

    happy-kitchen-89482

    07/30/2015, 8:49 PM
    How do you work around the Java logging insanity?
  • h

    happy-kitchen-89482

    07/30/2015, 8:50 PM
    We have a blanket exclude of all logging frameworks in our 3rdparty/BUILD, and then our jvm_binary targets all depend explicitly on logback (which is our logging framework of choice)
  • h

    happy-kitchen-89482

    07/30/2015, 8:50 PM
    But this means we have no logging at all in our unit tests
  • a

    ancient-stone-50795

    07/30/2015, 8:50 PM
    com.squareup.logging … and we use log4j
  • h

    happy-kitchen-89482

    07/30/2015, 8:51 PM
    But how do you deal with the fact that 3rdparty deps introduce their own logging dependencies, that may collide?
  • e

    enough-analyst-54434

    07/30/2015, 8:51 PM
    I'm pretty sure there is no sane answer
  • a

    ancient-stone-50795

    07/30/2015, 8:52 PM
    we use the sjar and globally exclude a bunch of stuff.
    Copy code
    excludes:  [
   { "org": "org.slf4j",
     "name": "log4j-over-slf4j"},
  { "org": "commons-logging",
     "name": "commons-logging" },
    I don’t remember the details, only that logging is very tricky.
  • a

    ancient-stone-50795

    07/30/2015, 8:53 PM
    We had the opposite problem in pants for a while - there was too much logging. That was a problem with a dep from dropwizard getting in and overriding our desired logging config.
1...102103104...2420Latest