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

    witty-crayon-22786

    07/12/2018, 4:20 PM
    ....ah, right, right.
  • a

    average-vr-56795

    07/12/2018, 4:20 PM
    I guess we can rely on the walk being top-down?
  • w

    witty-crayon-22786

    07/12/2018, 4:22 PM
    it ... isn't, quite. but could be
  • w

    witty-crayon-22786

    07/12/2018, 4:22 PM
    ... yea, i'm fine with a big comment here.
  • a

    average-vr-56795

    07/12/2018, 4:22 PM
    I think the important relationship we’re trying to establish is: “If the graph is batch-dirtying things, we will never view our dependencies as being after the batch starts, and then have them be invalidated in that batch”
  • w

    witty-crayon-22786

    07/12/2018, 4:22 PM
    (it's a comment either way)
  • w

    witty-crayon-22786

    07/12/2018, 4:22 PM
    right. so 
    complete
    needs to acquire the graph lock.
  • a

    average-vr-56795

    07/12/2018, 4:22 PM
    Yeah
  • w

    witty-crayon-22786

    07/12/2018, 4:23 PM
    ... i wonder what fraction of the total time under the lock that will be.
  • w

    witty-crayon-22786

    07/12/2018, 4:24 PM
    it only needs 
    read
    access on the graph though, which is nice.
  • w

    witty-crayon-22786

    07/12/2018, 4:24 PM
    so perhaps reason to switch to an RwLock there at some point.
  • a

    average-vr-56795

    07/12/2018, 4:25 PM
    Yeah; that’s why I ended up pulling out a separate 
    RwLock
    for just dirty state in the thing I linked to up there ^^; but I think that’s more complex than it needs to be 😞
  • w

    witty-crayon-22786

    07/12/2018, 4:25 PM
    i've tried using an RwLock for the graph itself a bunch of times, but didn't have enough code running in the read sections to make it worthwhile. maybe this would be
  • a

    average-vr-56795

    07/12/2018, 4:25 PM
    Yeah, just needing a 
    &mut Graph
    feels pretty nice
  • w

    witty-crayon-22786

    07/12/2018, 4:26 PM
    yea. let's do that for now
  • w

    witty-crayon-22786

    07/12/2018, 4:26 PM
    the potential changes are endless, so just finding a good place to wrap up is fine.
  • a

    average-vr-56795

    07/12/2018, 4:27 PM
    Yeah 🙂 I’ll put some comments up and send something your way
  • w

    witty-crayon-22786

    07/12/2018, 4:27 PM
    thanks a lot.
  • a

    average-vr-56795

    07/12/2018, 4:27 PM
    Will do some actual performance looking at some point too; got a little distracted today 🙂
  • w

    witty-crayon-22786

    07/12/2018, 4:28 PM
    🤞
  • a

    average-vr-56795

    07/12/2018, 4:34 PM
    Ugh, I can’t do the same trick with 
    dirty
    without cloning Entries on dirtying 😞
  • w

    witty-crayon-22786

    07/12/2018, 4:45 PM
    hm?
  • w

    witty-crayon-22786

    07/12/2018, 4:47 PM
    @average-vr-56795: what do you mean
  • a

    average-vr-56795

    07/12/2018, 5:00 PM
    The place that InnerGraph calls dirty, the Entry is mutably borrowed from the InnerGraph, so can't pass &mut self as an arg to dirty without cloning the Entry to end the mut self borrow
  • w

    witty-crayon-22786

    07/12/2018, 5:02 PM
    not sure how that is different from 
    complete
    ... but yea, got it.
  • w

    witty-crayon-22786

    07/12/2018, 5:04 PM
    (i remain fine with two big comments here)
  • a

    average-vr-56795

    07/12/2018, 5:11 PM
    Complete happens in Graph not InnerGraph, and doesn't have an InnerGraph lock
  • a

    average-vr-56795

    07/12/2018, 5:11 PM
    Anywho, updated the PR :)
  • a

    average-vr-56795

    07/12/2018, 6:52 PM
    @witty-crayon-22786 Any objection to publishing the boxfuture crate?
  • w

    witty-crayon-22786

    07/12/2018, 6:52 PM
    no, fine with that.
1...558559560...906Latest