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

    enough-analyst-54434

    09/24/2015, 10:26 PM
    Just a few days ago he claimed this
  • h

    happy-kitchen-89482

    09/24/2015, 10:27 PM
    About 50% of that startup is in 
    BuildConfiguration.initialize_parse_state
  • h

    happy-kitchen-89482

    09/24/2015, 10:27 PM
    Not sure where backend loading and argparse setup happen though
  • e

    enough-analyst-54434

    09/24/2015, 10:28 PM
    Thats surprising
  • h

    happy-kitchen-89482

    09/24/2015, 10:28 PM
    Hypothesis: we incur that initialize_parse_state cost once for each bootstrap BUILD file, and then once for all other BUILD files together...
  • e

    enough-analyst-54434

    09/24/2015, 10:28 PM
    all small map copying and thats about it in that method
  • e

    enough-analyst-54434

    09/24/2015, 10:29 PM
    well - that's proven true by the signature!: def initialize_parse_state(self, build_file)
  • e

    enough-analyst-54434

    09/24/2015, 10:29 PM
    its per-BUILD parse state
  • h

    happy-kitchen-89482

    09/24/2015, 10:30 PM
    per explicitly-specified BUILD, yes. But the parser follows dependencies without having to re-init
  • h

    happy-kitchen-89482

    09/24/2015, 10:31 PM
    I think?
  • e

    enough-analyst-54434

    09/24/2015, 10:31 PM
    That can't be right
  • e

    enough-analyst-54434

    09/24/2015, 10:31 PM
    no
  • f

    freezing-motorcycle-14222

    09/24/2015, 10:31 PM
    Actually, no
  • e

    enough-analyst-54434

    09/24/2015, 10:31 PM
    the callback is setup up in there so address mapper can have 1 map per BUILD (family)
  • f

    freezing-motorcycle-14222

    09/24/2015, 10:31 PM
    I did some more investigation
  • f

    freezing-motorcycle-14222

    09/24/2015, 10:31 PM
    This was because I was doing 
    ./pants goals
  • f

    freezing-motorcycle-14222

    09/24/2015, 10:31 PM
    That was forcing everything to load
  • f

    freezing-motorcycle-14222

    09/24/2015, 10:31 PM
    There is some logic in there already to be more lazy about loading things, which I discovered when I studied this more closely
  • h

    happy-kitchen-89482

    09/24/2015, 10:32 PM
    Hmm no
  • e

    enough-analyst-54434

    09/24/2015, 10:32 PM
    ah k - thanks @freezing-motorcycle-14222
  • h

    happy-kitchen-89482

    09/24/2015, 10:32 PM
    Yeah, the init gets called for every BUILD file (looking at call counts in the profile)
  • h

    happy-kitchen-89482

    09/24/2015, 10:32 PM
    But for some reason it gets less expensive per BUILD file if you have many
  • h

    happy-kitchen-89482

    09/24/2015, 10:32 PM
    caching of something?
  • f

    freezing-motorcycle-14222

    09/24/2015, 10:32 PM
    I also verified this by guarding the 
    CustomArgumentParser
    behind a lazy val. There was no performance improvement
  • f

    freezing-motorcycle-14222

    09/24/2015, 10:33 PM
    Even when testing commands that don't load everything
  • e

    enough-analyst-54434

    09/24/2015, 10:33 PM
    Well - the 1st call would walk all symbol import graphs
  • f

    freezing-motorcycle-14222

    09/24/2015, 10:33 PM
    There's also another issue
  • e

    enough-analyst-54434

    09/24/2015, 10:33 PM
    and force loading of JavaLibrary - and its imports, etc
  • f

    freezing-motorcycle-14222

    09/24/2015, 10:33 PM
    There is one thing which I think might be strict (but I haven't verified)
  • e

    enough-analyst-54434

    09/24/2015, 10:33 PM
    subsequent initilizes would not do this
1...457458459...2420Latest