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

    witty-crayon-22786

    03/18/2016, 4:53 PM
    the hadoop one is trickier
  • w

    witty-crayon-22786

    03/18/2016, 4:54 PM
    oh, tests of the annotation processor. gotcha.
  • w

    witty-crayon-22786

    03/18/2016, 4:54 PM
    but that does seem to be solved by indirection
  • w

    witty-crayon-22786

    03/18/2016, 4:54 PM
    hadoop less-so... or perhaps more-complicatedly so
  • w

    witty-crayon-22786

    03/18/2016, 4:55 PM
    i think it is still possible that the indirection is worth it in a large repo.
  • w

    witty-crayon-22786

    03/18/2016, 4:57 PM
    basically, it seems like the only time when it is more work to declare these properties on the dep is when less there are less than two dependents
  • w

    witty-crayon-22786

    03/18/2016, 4:57 PM
    @jolly-chef-92794: ^
  • w

    witty-crayon-22786

    03/18/2016, 4:58 PM
    as you showed in your example above with the indirection, it's not a huge amount of code, and you only do it once.
  • a

    ancient-stone-50795

    03/18/2016, 4:59 PM
    I'll be honest, one thing that makes me like doing it this way is that the mapping between maven and pants is straight forward. That's not a great design reason but it is good for adoption.
  • j

    jolly-chef-92794

    03/18/2016, 5:20 PM
    @witty-crayon-22786: I suppose that’s true if you only define the target in one place, eg, a 
    3rdparty/BUILD.scoped
    with hand-coded indirections
  • j

    jolly-chef-92794

    03/18/2016, 5:20 PM
    That still feels more cumbersome though
  • j

    jolly-chef-92794

    03/18/2016, 5:23 PM
    @witty-crayon-22786: In what situation do you think the indirection is worth it? Can you elaborate on how it makes things easier in a larger repo?
  • w

    witty-crayon-22786

    03/18/2016, 5:36 PM
    in a large repo, hundreds of things depend on hadoop
  • w

    witty-crayon-22786

    03/18/2016, 5:37 PM
    ie, the ratio of dependents to dependencies is higher
  • j

    jolly-chef-92794

    03/18/2016, 5:39 PM
    So.. the benefit is that in BUILD files you have:
    Copy code
    java_library(name=‘…’,
  dependencies=[
    ‘3rdparty:foobar-scoped-to-provided’,
  ],
)
    instead of:
    Copy code
    java_library(name=‘…’,
  scoped_dependencies=[
    scoped(‘provided’, ‘3rdparty:foobar’),
  ],
)
    in all your hadoop-related targets?
  • j

    jolly-chef-92794

    03/18/2016, 5:40 PM
    Those don’t seem particularly different in terms of maintainability, readability, or verbosity
  • j

    jolly-chef-92794

    03/18/2016, 5:41 PM
    And in the first case, if someone changed the scope of 
    '3rdparty:foobar-scoped-to-provided’
    without updating the name, it’d be pretty confusing
  • b

    bored-art-40741

    03/18/2016, 6:01 PM
    Hey @rough-minister-58256 what's the current status of a warm in-memory graph? I'd love to try that out for running buildgen
  • b

    bored-art-40741

    03/18/2016, 6:03 PM
    Also @mammoth-pharmacist-54215 what's the status on fixing 
    src/python/pants/backend/jvm/subsystems/scala_platform.py
    for using 
    custom
    build info?
  • m

    mammoth-pharmacist-54215

    03/18/2016, 6:04 PM
    @bored-art-40741: will take a look at that Monday, shouldn’t be to difficult to resolve.
  • r

    rough-minister-58256

    03/18/2016, 6:04 PM
    its coupled to the new engine atm. we’re still working up the integration into mainline pants + refining engine stuff
  • w

    wide-energy-11069

    03/18/2016, 6:43 PM
    @witty-crayon-22786: any complication if i store the 
    lastIntelliJBuild
    info under 
    <buildroot>/.plugin
    ? it is not covered by 
    clean-all
    but still consistent with status quo where e.g. deleting dist/ and run the test without code change will not trigger 
    export-classpath compile …
  • w

    witty-crayon-22786

    03/18/2016, 6:45 PM
    @jolly-chef-92794: i think most of the time you would not name it that way. whichever was the more common usage of the target would be the one with the shorter name
  • w

    witty-crayon-22786

    03/18/2016, 6:46 PM
    in the annotation processor case, everything except the tests for the annotation processor would use the "compile time only" definition
  • w

    witty-crayon-22786

    03/18/2016, 6:47 PM
    ... i still think the hadoop case is super weird for this feature though... all tests that touch hadoop would need to include the dep explicitly in order to get it
  • j

    jolly-chef-92794

    03/18/2016, 6:47 PM
    yes
  • j

    jolly-chef-92794

    03/18/2016, 6:48 PM
    Unless we added a 
    test
    scope
  • w

    witty-crayon-22786

    03/18/2016, 6:48 PM
    ...which i suppose is just the opposite of what twitter does now (all binaries that will run in hadoop use a 
    hadoop_binary
    with deploy_excludes)
  • w

    witty-crayon-22786

    03/18/2016, 6:49 PM
    it seems like scopes in maven are overloaded here
  • j

    jolly-chef-92794

    03/18/2016, 6:49 PM
    Yeah, we’re using deploy_excludes a lot right now. I would like to cut back on it
1...106410651066...2420Latest