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

    ancient-stone-50795

    03/08/2016, 5:56 PM
    I don't think its generally the case that the resource used by an annotation processor should be excluded from the runtime classpath, but once we solve the provided dependencies problem we can choose to ignore them if we want.
  • w

    witty-crayon-22786

    03/08/2016, 5:57 PM
    the annotation processor's subgraph should not be on the runtime classpath, correct?
  • a

    ancient-stone-50795

    03/08/2016, 5:57 PM
    That could be true, but I don't think that was the question.
  • b

    bored-art-40741

    03/08/2016, 5:57 PM
    Okay so re: that RB being a bugfix that doesn't break anything any further, I agree @witty-crayon-22786
  • b

    bored-art-40741

    03/08/2016, 5:58 PM
    However I think the original change was backed out
  • w

    witty-crayon-22786

    03/08/2016, 5:58 PM
    it was not
  • a

    ancient-stone-50795

    03/08/2016, 5:58 PM
    If you are going to remove fingerprinting from resources invoking compile, that will break us, but you can land it behind a flag and that'd be fine.
  • b

    bored-art-40741

    03/08/2016, 5:58 PM
    Ah you're right, it wasn't
  • w

    witty-crayon-22786

    03/08/2016, 5:58 PM
    er... that's what's in master
  • b

    bored-art-40741

    03/08/2016, 5:58 PM
    @ancient-stone-50795: That's the state of the world today
  • b

    bored-art-40741

    03/08/2016, 5:59 PM
    I called it out in the CR you commented on since I was just fixing a bug in that original change to remove Resource fingerprinting
  • b

    bored-art-40741

    03/08/2016, 5:59 PM
    But that change was never backed out
  • b

    bored-art-40741

    03/08/2016, 5:59 PM
    See 
    src/python/pants/backend/jvm/tasks/jvm_compile/jvm_compile.py:46
  • w

    witty-crayon-22786

    03/08/2016, 5:59 PM
    to be clear about what i'm suggesting:
  • w

    witty-crayon-22786

    03/08/2016, 5:59 PM
    Copy code
    java_library(
  name='uses_an_annotation_processor',
  dependencies=[...],
  compile_dependencies=[
    'src/resources/thing_annotation_processor_needs',
    '3rdparty/jvm/autovalue',
  ],
)
  • w

    witty-crayon-22786

    03/08/2016, 5:59 PM
    something like
  • b

    bored-art-40741

    03/08/2016, 6:00 PM
    Actually I'm kind of confused. We definitely see inappropriate invalidation as the result to changes in transitive resources() dependencies
  • w

    witty-crayon-22786

    03/08/2016, 6:00 PM
    that's your bugfix.
  • b

    bored-art-40741

    03/08/2016, 6:00 PM
    No
  • b

    bored-art-40741

    03/08/2016, 6:00 PM
    My bugfix is only correcting the issue where nothing changed but the structure of the graph
  • b

    bored-art-40741

    03/08/2016, 6:01 PM
    I claim that if you have 
    A
    depends on 
    R
    where 
    R
    is some trivial 
    resources()
    , that changing the contents of some file owned by 
    R
    will invalidate 
    A
  • b

    bored-art-40741

    03/08/2016, 6:01 PM
    In master today
  • b

    bored-art-40741

    03/08/2016, 6:01 PM
    Let me see if I can back that up
  • a

    ancient-stone-50795

    03/08/2016, 6:01 PM
    @witty-crayon-22786: the comment says, "doesnothing" but I think it does something
  • a

    ancient-stone-50795

    03/08/2016, 6:01 PM
    The change was to return None in that case
  • w

    witty-crayon-22786

    03/08/2016, 6:02 PM
    an empty sha1
  • b

    bored-art-40741

    03/08/2016, 6:02 PM
    Oh Eric is right
  • b

    bored-art-40741

    03/08/2016, 6:02 PM
    I see it now
  • a

    ancient-stone-50795

    03/08/2016, 6:02 PM
    Could it be that it still inherits the default behavior of fingerprinting sources in a superclass?
  • b

    bored-art-40741

    03/08/2016, 6:02 PM
    Yes
1...104410451046...2420Latest