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

    rough-minister-58256

    07/20/2016, 8:41 PM
    what does that gain us?
  • f

    fancy-queen-20734

    07/20/2016, 8:41 PM
    Files are not consistent enough
  • w

    witty-crayon-22786

    07/20/2016, 8:42 PM
    @rough-minister-58256: fewer 
    Nodes
    potentially
  • f

    fancy-queen-20734

    07/20/2016, 8:42 PM
    if we rely on snapshots, we can be certain that contents won’t change out from under
  • f

    fancy-queen-20734

    07/20/2016, 8:42 PM
    during execution
  • r

    rough-minister-58256

    07/20/2016, 8:42 PM
    ic
  • w

    witty-crayon-22786

    07/20/2016, 8:42 PM
    a lot is reified as Nodes that could be tracked as a single request for a snapshot.
  • f

    fancy-queen-20734

    07/20/2016, 8:42 PM
    this makes me have a funny thought about python chroots
  • w

    witty-crayon-22786

    07/20/2016, 8:42 PM
    (maybe? i think you'd still want to dedupe filesystem operations, so... maybe not)
  • r

    rough-minister-58256

    07/20/2016, 8:43 PM
    how would invalidation work?
  • r

    rough-minister-58256

    07/20/2016, 8:43 PM
    the entire snapshot would get invalidated by changing a single file?
  • w

    witty-crayon-22786

    07/20/2016, 8:43 PM
    yea, still need to track all of the operations. but it might just be that a Snapshot is labelled with all of the operations we took
  • f

    fancy-queen-20734

    07/20/2016, 8:43 PM
    I have this issue sometimes where the chroot contents aren’t updated because, I think, the fingerprinting races the chroot creation somehow
  • w

    witty-crayon-22786

    07/20/2016, 8:43 PM
    @rough-minister-58256: it already does (EDIT: get invalidated when you change a single file)
  • r

    rough-minister-58256

    07/20/2016, 8:44 PM
    hm true
  • f

    fancy-queen-20734

    07/20/2016, 8:44 PM
    the scoot idea is that snapshots represent file state at some point in time
  • w

    witty-crayon-22786

    07/20/2016, 8:44 PM
    file(s) state
  • f

    fancy-queen-20734

    07/20/2016, 8:44 PM
    identity for snapshots is either based on contents or on initial snapshot contents + list of operations
  • f

    fancy-queen-20734

    07/20/2016, 8:45 PM
    where the second one is updated to point at a contents based ref after it’s been fulfilled
  • w

    witty-crayon-22786

    07/20/2016, 8:45 PM
    well, how the identity is computed is orthogonal i think
  • f

    fancy-queen-20734

    07/20/2016, 8:45 PM
    identity is how we get invalidation though
  • w

    witty-crayon-22786

    07/20/2016, 8:46 PM
    @fancy-queen-20734: yes. but how it is computed is orthogonal
  • f

    fancy-queen-20734

    07/20/2016, 8:47 PM
    I see what you mean
  • w

    witty-crayon-22786

    07/20/2016, 8:47 PM
    (how it is computed in scoot at least)
  • r

    rough-minister-58256

    07/20/2016, 8:47 PM
    tho I’d imagine snapshot reconstruction would be more expensive without more granularly invalidatable file objects below it
  • w

    witty-crayon-22786

    07/20/2016, 8:47 PM
    @rough-minister-58256: no; that's part of why scoot doesn't guarantee that it will be a fingerprint indefinitely
  • r

    rough-minister-58256

    07/20/2016, 8:48 PM
    Glob -> (FileA, FileB, FileC) -> SnapshotA with invalidation->FileB vs Glob -> SnapshotA
  • r

    rough-minister-58256

    07/20/2016, 8:48 PM
    hm?
  • w

    witty-crayon-22786

    07/20/2016, 8:48 PM
    because in the second case nick mentioned the id is not a fingerprint
  • w

    witty-crayon-22786

    07/20/2016, 8:48 PM
    anyway, that part is a long way off
1...121122123...906Latest