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

    witty-crayon-22786

    02/16/2017, 1:20 AM
    yes
  • w

    witty-crayon-22786

    02/16/2017, 1:21 AM
    if a function returns, the local variables will all be decremented
  • f

    fancy-queen-20734

    02/16/2017, 1:21 AM
    so, the raw_roots were a local in the same function that they were iterated over. I then changed that so they were local to the fn where they were unpacked
  • w

    witty-crayon-22786

    02/16/2017, 1:21 AM
    this is why 
    cffi.gc
    wrapper
  • f

    fancy-queen-20734

    02/16/2017, 1:22 AM
    my guess is that that change made them have too short a life
  • w

    witty-crayon-22786

    02/16/2017, 1:22 AM
    not sure if i used it in this case...?
  • w

    witty-crayon-22786

    02/16/2017, 1:22 AM
    yea
  • w

    witty-crayon-22786

    02/16/2017, 1:22 AM
    Copy code
    raw_roots = self._native.gc(self._native.lib.execution_roots(self._scheduler), self._native.lib.nodes_destroy))
  • f

    fancy-queen-20734

    02/16/2017, 1:23 AM
    that’s the call
  • w

    witty-crayon-22786

    02/16/2017, 1:23 AM
    hm... oh, so something in the roots was still in use after the 
    raw_roots
    object had been dropped?
  • w

    witty-crayon-22786

    02/16/2017, 1:23 AM
    yea, gotta clone stuff out of there
  • w

    witty-crayon-22786

    02/16/2017, 1:24 AM
    cffi.buffer(cdata)[:]
  • f

    fancy-queen-20734

    02/16/2017, 1:24 AM
    I think between unpacking and use, the nodes were destroyed
  • w

    witty-crayon-22786

    02/16/2017, 1:24 AM
    copies data out of a buffer
  • f

    fancy-queen-20734

    02/16/2017, 1:24 AM
    unpack does?
  • w

    witty-crayon-22786

    02/16/2017, 1:24 AM
    (editted)
  • w

    witty-crayon-22786

    02/16/2017, 1:24 AM
    unpack doesn't...
  • w

    witty-crayon-22786

    02/16/2017, 1:25 AM
    unpack is "i have one pointer, gimme a bunch of pointers" (edit: ie, 
    cdata
    objects)
  • w

    witty-crayon-22786

    02/16/2017, 1:25 AM
    ...depending what is being unpacked i suppose. if its primitives, then you're ok
  • w

    witty-crayon-22786

    02/16/2017, 1:25 AM
    but if they contain allocations, you have to copy out of the allocation into a python object
  • w

    witty-crayon-22786

    02/16/2017, 1:25 AM
    basically, ensure that you're not still referencing a 
    cdata
    object
  • w

    witty-crayon-22786

    02/16/2017, 1:26 AM
    which is a pointer to memory
  • w

    witty-crayon-22786

    02/16/2017, 1:26 AM
    (editted the unpack above)
  • f

    fancy-queen-20734

    02/16/2017, 1:26 AM
    it’s interesting that a cdata object doesn’t survive a slice of it being kept
  • f

    fancy-queen-20734

    02/16/2017, 1:27 AM
    assuming the unpack docs are ~ equivalent to the impl
  • w

    witty-crayon-22786

    02/16/2017, 1:27 AM
    that's not how reference counting works, in general
  • w

    witty-crayon-22786

    02/16/2017, 1:27 AM
    ... but yea, i see what you mean
  • w

    witty-crayon-22786

    02/16/2017, 1:28 AM
    i think it would be pretty error prone
  • w

    witty-crayon-22786

    02/16/2017, 1:28 AM
    you think you're done with some allocation, when instead you're accidentally holding the whole thing still
  • f

    fancy-queen-20734

    02/16/2017, 1:28 AM
    true, though on the other hand,
1...237238239...906Latest