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

    straight-waiter-86090

    04/07/2017, 3:48 PM
    that’s right. ant runs javac once per top level project, including all java files recursively. with pants we have some (many) packages with just one or two files.
  • f

    full-kilobyte-93429

    04/07/2017, 3:50 PM
    i wonder how much of this one hour is in javac and how much is in python
  • s

    straight-waiter-86090

    04/07/2017, 3:53 PM
    I’m also curious how much overhead is added by the zinc scala code.
  • f

    full-kilobyte-93429

    04/07/2017, 3:55 PM
    yea, they can all be profiled. enable 
    PANTS_PROFILE
    will dump to python profile, for zinc you will need a java debugger, like yourkit
  • s

    straight-waiter-86090

    04/07/2017, 3:56 PM
    ok. I’ll poke around and see if anything comes up.
  • w

    witty-crayon-22786

    04/07/2017, 5:00 PM
    @high-advantage-47550: per-target will only be available in 1.3.x
    h
    • 2
    • 1
  • w

    witty-crayon-22786

    04/07/2017, 5:01 PM
    @straight-waiter-86090 : how many [compile.zinc] workers is your machine choosing?
  • w

    witty-crayon-22786

    04/07/2017, 5:02 PM
    ...actually, just dumping 
    ./pants options | grep 'compile.zinc'
    would probably be the easiest way forward.
  • w

    witty-crayon-22786

    04/07/2017, 5:03 PM
    because the next questions are: 1) how much memory, and 2) zinc analysis cache size
  • s

    straight-waiter-86090

    04/07/2017, 5:03 PM
    compile.zinc.worker_count = 8 (from HARDCODED)
  • s

    straight-waiter-86090

    04/07/2017, 5:04 PM
    I’m running on a Macbook Pro.
  • s

    straight-waiter-86090

    04/07/2017, 5:04 PM
    I haven’t changed memory size for zinc, but I could try that.
  • s

    straight-waiter-86090

    04/07/2017, 5:04 PM
    compile.zinc.jvm_options = ['-Dfile.encoding=UTF-8', '-Dzinc.analysis.cache.limit=1000', '-Djava.awt.headless=true', '-Xmx2g'] (from HARDCODED)
  • w

    witty-crayon-22786

    04/07/2017, 5:05 PM
    we use 
    -Xmx6g -Xms6g
    internally
  • s

    straight-waiter-86090

    04/07/2017, 5:06 PM
    I would set this with 
    compile.zinc.jvm_options
    ?
  • w

    witty-crayon-22786

    04/07/2017, 5:06 PM
    yea
  • w

    witty-crayon-22786

    04/07/2017, 5:06 PM
    you've got 
    -Xmx2g
    in there now
  • w

    witty-crayon-22786

    04/07/2017, 5:09 PM
    @straight-waiter-86090 : the rest of 
    ./pants options | grep 'compile.zinc'
    would be good too, scrubbed if necessary
  • s

    straight-waiter-86090

    04/07/2017, 5:10 PM
  • w

    witty-crayon-22786

    04/07/2017, 5:13 PM
    ok, nothing else jumping out at me there.
  • w

    witty-crayon-22786

    04/07/2017, 5:14 PM
    thanks.
  • s

    straight-waiter-86090

    04/07/2017, 5:14 PM
    I’ll try more heap, see how that goes.
  • w

    witty-crayon-22786

    04/07/2017, 5:14 PM
    should be significant. not 5x, but.
  • w

    witty-crayon-22786

    04/07/2017, 5:15 PM
    you didn't use strict deps with ant, did you?
  • w

    witty-crayon-22786

    04/07/2017, 5:15 PM
    ie, only explicitly declared deps on the compile classpath?
  • w

    witty-crayon-22786

    04/07/2017, 5:15 PM
    @straight-waiter-86090 : ^
  • s

    straight-waiter-86090

    04/07/2017, 5:16 PM
    no. we use ivy and allow ivy:cachepath to generate the classpath.
  • w

    witty-crayon-22786

    04/07/2017, 5:17 PM
    there are two or three "2-3x" improvement type differences... one of them is 
    java.strict_deps = True
    , but it takes a lot of effort to apply.
  • w

    witty-crayon-22786

    04/07/2017, 5:17 PM
    the other is 
    compile.zinc.use_classpath_jars = True
    ... but it essentially disables incremental compiles
  • w

    witty-crayon-22786

    04/07/2017, 5:17 PM
    (cross-target incremental compiles)
1...169816991700...2420Latest