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

    bored-art-40741

    03/21/2016, 6:43 PM
    You don't need an internal "pypi"
  • b

    bored-art-40741

    03/21/2016, 6:44 PM
    You can consume from a simple structured directory served up by generic nginx or whatever
  • w

    witty-crayon-22786

    03/21/2016, 6:44 PM
    @wide-energy-11069: the 
    options
    command evaluates all scopes
  • w

    witty-crayon-22786

    03/21/2016, 6:45 PM
    @wide-energy-11069: so i can't actually reproduce the bug there, since it forces all of them
  • h

    happy-kitchen-89482

    03/21/2016, 6:45 PM
    It might be a big change
  • h

    happy-kitchen-89482

    03/21/2016, 6:45 PM
    Right now the arg splitter looks at those prefixes in order to assign flags to scopes
  • h

    happy-kitchen-89482

    03/21/2016, 6:46 PM
    Options are parsed on-demand, so only scopes that are used will expose an unknown option.
  • h

    happy-kitchen-89482

    03/21/2016, 6:46 PM
    Which I guess is the place to start
  • h

    happy-kitchen-89482

    03/21/2016, 6:46 PM
    at least that way you get an error
  • h

    happy-kitchen-89482

    03/21/2016, 6:47 PM
    But there are good performance reasons to only parse on demand
  • h

    happy-kitchen-89482

    03/21/2016, 6:47 PM
    that was a major part of the design
  • w

    witty-crayon-22786

    03/21/2016, 6:47 PM
    yea. forcing any scope for which you actually got an arg... doesn't sound huge...? (famous last words)
  • h

    happy-kitchen-89482

    03/21/2016, 6:47 PM
    Yeah, that seems fine
  • h

    happy-kitchen-89482

    03/21/2016, 6:47 PM
    Given the small number of flags any given invocation actually uses
  • w

    wide-energy-11069

    03/21/2016, 6:47 PM
    so start with force parsing all options regardless of usage?
  • h

    happy-kitchen-89482

    03/21/2016, 6:47 PM
    No
  • h

    happy-kitchen-89482

    03/21/2016, 6:48 PM
    Absolutely not
  • h

    happy-kitchen-89482

    03/21/2016, 6:48 PM
    That will be a performance hit
  • h

    happy-kitchen-89482

    03/21/2016, 6:48 PM
    But do parse scopes that have a flag assigned to them
  • h

    happy-kitchen-89482

    03/21/2016, 6:48 PM
    Today we parse on-demand, so only scopes that are actually used
  • h

    happy-kitchen-89482

    03/21/2016, 6:49 PM
    so we could say one of two things:
  • h

    happy-kitchen-89482

    03/21/2016, 6:49 PM
    or
  • h

    happy-kitchen-89482

    03/21/2016, 6:49 PM
    1. If a flag got assigned to a scope, then we parse that scope’s options
  • h

    happy-kitchen-89482

    03/21/2016, 6:49 PM
    2. If a flag got assigned to a scope, we treat it as if that scope was explicitly mentioned on the cmd line
  • h

    happy-kitchen-89482

    03/21/2016, 6:49 PM
    Oh, but that only works for task scopes, so never mind
  • h

    happy-kitchen-89482

    03/21/2016, 6:50 PM
    So, yeah, I guess we eagerly parse for scopes that had an explicit cmd-line flag assigned to them, and let the others continue to parse on-demand?
  • h

    happy-kitchen-89482

    03/21/2016, 6:52 PM
    That won’t solve the problem of scopes “stealing” options intended for other scopes, based on their prefix
  • h

    happy-kitchen-89482

    03/21/2016, 6:52 PM
    But it will at least give an error instead of silently ignoring the unknown option
  • w

    wide-energy-11069

    03/21/2016, 6:53 PM
    k. it will solve 
    ./pants bundle <target> --jvm-bundle-archive=zip
    but not 
    ./pants resolve.ivy --jvm-options='1 2 3’
  • w

    wide-energy-11069

    03/21/2016, 6:54 PM
    to re-iterate, with 
    --jvm-bundle-archive=zip
    , since 
    jvm
    is mentioned, we will actively parse 
    bundle-archive=zip
    within 
    jvm
    , thus erroring out?
1...107510761077...2420Latest