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

    witty-crayon-22786

    07/29/2015, 9:10 PM
    @enough-analyst-54434: would 
    ls -R
    for the chroot and a report of the content of init.py files for all 
    com/twitter/ml
    subdirectories work?
  • e

    enough-analyst-54434

    07/29/2015, 9:10 PM
    That plus said-same for all .deps/*com.twitter.ml*.whl exploded
  • e

    enough-analyst-54434

    07/29/2015, 9:11 PM
    if any one of those bits fails to declare a namespace then you;re at the whim of import loading order
  • w

    witty-crayon-22786

    07/29/2015, 9:12 PM
    i definitely see one potential case.
  • w

    witty-crayon-22786

    07/29/2015, 9:12 PM
    what do you mean by exploded though? i only see loose files in those subdirectories
  • e

    enough-analyst-54434

    07/29/2015, 9:12 PM
    The wheels should be namespaced
  • e

    enough-analyst-54434

    07/29/2015, 9:12 PM
    Aha - my misconception
  • e

    enough-analyst-54434

    07/29/2015, 9:12 PM
    so just scan those whl dirs too
  • w

    witty-crayon-22786

    07/29/2015, 9:12 PM
    k
  • e

    enough-analyst-54434

    07/29/2015, 9:14 PM
    If there is a missing namespace in the chroot root files but not the whls this should be any easy repro for me to set up and solve
  • e

    enough-analyst-54434

    07/29/2015, 9:14 PM
    If the whls are missing the namespacing I'll be more stumped
  • w

    witty-crayon-22786

    07/29/2015, 9:22 PM
    @enough-analyst-54434: the wrapping is unfortunate, but i included the content of init.py entries under 
    com/twitter/ml
    directories
  • e

    enough-analyst-54434

    07/29/2015, 9:23 PM
    yeah - so, I'm not sure how this ever worked before
  • e

    enough-analyst-54434

    07/29/2015, 9:23 PM
    I'll focus on this in a few minutes, but I see no way to support nested disconnected thrift namespaces in py given the fact all is emitted in the init.ys by the thrift compiler
  • e

    enough-analyst-54434

    07/29/2015, 9:24 PM
    parallel would work
  • e

    enough-analyst-54434

    07/29/2015, 9:24 PM
    its the fact that ml is in the middle of a deeper package
  • e

    enough-analyst-54434

    07/29/2015, 9:24 PM
    basically, the Foursquare rule of non non-leaf packages would be winning here
  • w

    witty-crayon-22786

    07/29/2015, 9:24 PM
    is there a way to fail for this case?
  • e

    enough-analyst-54434

    07/29/2015, 9:25 PM
    But pants should fail fast if there is no way for it to support this / I need to figure out how pants made this work before
  • e

    enough-analyst-54434

    07/29/2015, 9:25 PM
    Definitely can fail this sort of case, yes
  • w

    witty-crayon-22786

    07/29/2015, 9:25 PM
    i don't completely understand why init.py files are ever necessary, but it would be neat to just disallow manually adding them and generate them instead
  • e

    enough-analyst-54434

    07/29/2015, 9:26 PM
    That is not the issue
  • w

    witty-crayon-22786

    07/29/2015, 9:26 PM
    anywho, for another day i guess.
  • e

    enough-analyst-54434

    07/29/2015, 9:26 PM
    The issue is thrift creates these
  • e

    enough-analyst-54434

    07/29/2015, 9:26 PM
    no human does
  • e

    enough-analyst-54434

    07/29/2015, 9:26 PM
    But all roots of a package must agree on init.py contents
  • e

    enough-analyst-54434

    07/29/2015, 9:26 PM
    to have repeatable package loading
  • e

    enough-analyst-54434

    07/29/2015, 9:27 PM
    So the best pants can do is fail fast on inconsistent package init.py contents
  • w

    witty-crayon-22786

    07/29/2015, 9:28 PM
    or always generate a namespace package?
  • e

    enough-analyst-54434

    07/29/2015, 9:28 PM
    no - it can't
1...949596...2420Latest