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

    witty-crayon-22786

    02/05/2018, 7:25 PM
    but 1) users don't run full graph, 2) our expectation is that things are warm
  • r

    rough-minister-58256

    02/05/2018, 7:25 PM
    yep, makes sense
  • w

    witty-crayon-22786

    02/05/2018, 7:25 PM
    so things that strongly affect warm are the highest priority, imo
  • w

    witty-crayon-22786

    02/05/2018, 7:26 PM
    ...warm compiles and tests, in particular
  • w

    witty-crayon-22786

    02/05/2018, 7:28 PM
    @rough-minister-58256: BUT, i definitely spent some time profiling the larger cases this weekend.
  • r

    rough-minister-58256

    02/05/2018, 7:28 PM
    oh?
  • w

    witty-crayon-22786

    02/05/2018, 7:28 PM
    referred to it during standup
  • w

    witty-crayon-22786

    02/05/2018, 7:29 PM
    didn't find anything conclusive, but did get a few ideas.
  • r

    rough-minister-58256

    02/05/2018, 11:55 PM
    random thought: 1) would it make sense to support rust-based rule implementations? 2) if we have #1, is there a point in porting v1 tasks to v2 python tasks vs jumping directly to rust for that? (for perf)
  • r

    rough-minister-58256

    02/05/2018, 11:56 PM
    (at least for the “core” rules?)
  • w

    witty-crayon-22786

    02/05/2018, 11:58 PM
    rust-based rules would be a big, big step, because it would mean that end users couldn't implement them
  • w

    witty-crayon-22786

    02/05/2018, 11:59 PM
    (without some sort of runtime compilation of rust code)
  • w

    witty-crayon-22786

    02/06/2018, 12:04 AM
    i really hope we don't have to go there
  • r

    rough-minister-58256

    02/06/2018, 12:05 AM
    end users == Task/Rule authors?
  • r

    rough-minister-58256

    02/06/2018, 12:05 AM
    or as in, .bzl file authors writing tasks live in the repo?
  • r

    rough-minister-58256

    02/06/2018, 12:06 AM
    for pants “core” at least, I think we could get away with glomming onto build time compilation?
  • r

    rough-minister-58256

    02/06/2018, 12:06 AM
    but it would constrain authors to committers to pants - but if the perf wins are there..
  • w

    witty-crayon-22786

    02/06/2018, 12:06 AM
    Copy code
    end users == Task/Rule authors?
or as in, .bzl file authors writing tasks live in the repo?
    <- same thing, afaik
  • w

    witty-crayon-22786

    02/06/2018, 12:07 AM
    ...yea, what you said. sorry.
  • w

    witty-crayon-22786

    02/06/2018, 12:07 AM
    but yes: i want to be careful of that, because, for example, jvm compile is not implemented in skylark in bazel
  • w

    witty-crayon-22786

    02/06/2018, 12:07 AM
    and it means you cannot extend it.
  • r

    rough-minister-58256

    02/06/2018, 12:07 AM
    but thats skylark’s issue, not the engines?
  • w

    witty-crayon-22786

    02/06/2018, 12:07 AM
    which meant it took 3+ years to get basic scala support solid.
  • r

    rough-minister-58256

    02/06/2018, 12:07 AM
    if product requests all work the same way etc
  • w

    witty-crayon-22786

    02/06/2018, 12:08 AM
    ...i guess.
  • w

    witty-crayon-22786

    02/06/2018, 12:08 AM
    i don't know how much benefit you would have working on the rust side if everything still needed to be python objects though
  • w

    witty-crayon-22786

    02/06/2018, 12:09 AM
    basically, the temptation is to bend the rules (heh) and avoid actually creating python objects.
  • w

    witty-crayon-22786

    02/06/2018, 12:09 AM
    @rough-minister-58256: for the record, i don't expect that performance will be an issue anytime soon for these cases
  • w

    witty-crayon-22786

    02/06/2018, 12:10 AM
    it's a thing for "full graph"
  • r

    rough-minister-58256

    02/06/2018, 12:11 AM
    k
1...445446447...906Latest