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

    happy-kitchen-89482

    11/02/2015, 8:00 PM
    Now checkstyle plugins are only imported conditionally
  • e

    enough-analyst-54434

    11/02/2015, 8:00 PM
    You'd load the backends 1x
  • e

    enough-analyst-54434

    11/02/2015, 8:01 PM
    and not per CLI invocation
  • h

    happy-kitchen-89482

    11/02/2015, 8:01 PM
    yes
  • h

    happy-kitchen-89482

    11/02/2015, 8:01 PM
    the CLI should be very lightweight and fast
  • h

    happy-kitchen-89482

    11/02/2015, 8:01 PM
    Written in Go, perhaps
  • h

    happy-kitchen-89482

    11/02/2015, 8:01 PM
    If you want <10ms
  • e

    enough-analyst-54434

    11/02/2015, 8:01 PM
    I think Kris had a nailgun do-nothing client
  • h

    happy-kitchen-89482

    11/02/2015, 8:01 PM
    I think just firing up the python interpreter is at least 40ms
  • e

    enough-analyst-54434

    11/02/2015, 8:01 PM
    ack
  • h

    happy-kitchen-89482

    11/02/2015, 8:02 PM
    So yeah, I'm going to stop chasing diminishing returns now...
  • r

    rough-minister-58256

    11/02/2015, 8:02 PM
    yeah, I have things working with the native nailgun client (ng.c)
  • h

    happy-kitchen-89482

    11/02/2015, 8:02 PM
    but I do think that even in the daemon world, we want to be thinking about code loading costs
  • h

    happy-kitchen-89482

    11/02/2015, 8:04 PM
    Oh, funny, see if you can guess which python standard library function we spend 12% of that code-loading time in.
  • h

    happy-kitchen-89482

    11/02/2015, 8:04 PM
    The largest single chunk.
  • e

    enough-analyst-54434

    11/02/2015, 8:04 PM
    probably adjusting the sys.path
  • h

    happy-kitchen-89482

    11/02/2015, 8:04 PM
    re.compile()
  • r

    rough-minister-58256

    11/02/2015, 8:04 PM
    heh
  • e

    enough-analyst-54434

    11/02/2015, 8:04 PM
    what do we even use that for?
  • h

    happy-kitchen-89482

    11/02/2015, 8:04 PM
    we eagerly compile a lot of regexes apparently
  • e

    enough-analyst-54434

    11/02/2015, 8:05 PM
    oh - gotcha - across all backends
  • h

    happy-kitchen-89482

    11/02/2015, 8:05 PM
    Death of a thousand cuts
  • b

    bored-art-40741

    11/02/2015, 8:07 PM
    Funny enough, re has its own cache for compiled regexes
  • e

    enough-analyst-54434

    11/02/2015, 8:07 PM
    yup
  • b

    bored-art-40741

    11/02/2015, 8:07 PM
    Unless we were compiling I think ~100s, just doing the naive thing would actually have the same runtime performance
  • e

    enough-analyst-54434

    11/02/2015, 8:08 PM
    So - the npm failure flakes are disturbingly frequent ... in that I've hit this 2x today on master builds
  • e

    enough-analyst-54434

    11/02/2015, 8:08 PM
    Is this well known / can I rely on @bright-dinner-81701 to be sqaushing this or does it need more general love?
  • e

    enough-analyst-54434

    11/02/2015, 8:10 PM
    The latest looks like not-flake but real version drift that leads to an unsolvable resolve?: https://travis-ci.org/pantsbuild/pants/jobs/88856483
  • h

    happy-kitchen-89482

    11/02/2015, 8:10 PM
    Apparently we're calling it directly 336 times, yet there are only 66 calls to re.compile() in our codebase.
  • h

    happy-kitchen-89482

    11/02/2015, 8:10 PM
    So some loop somewhere
1...657658659...2420Latest