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

    witty-crayon-22786

    03/02/2016, 9:01 PM
    most of the tools do a fetch-and-atomic rename thing.
  • w

    witty-crayon-22786

    03/02/2016, 9:01 PM
    but each tool shouldn't need to do it... seems like if we put them in the buildcache, then we can kill ~/.cache/pants directory
  • w

    witty-crayon-22786

    03/02/2016, 9:01 PM
    clean-all would wipe them, but they'd come back from the cache.
  • w

    witty-crayon-22786

    03/02/2016, 9:02 PM
    ...and then only one codepath for shared blobs
  • b

    bored-art-40741

    03/02/2016, 9:02 PM
    Sure, that's fine
  • b

    bored-art-40741

    03/02/2016, 9:02 PM
    But it doesn't actually explain this error
  • b

    bored-art-40741

    03/02/2016, 9:02 PM
    Stack is using the standard 
    BinaryUtil
  • b

    bored-art-40741

    03/02/2016, 9:02 PM
    So this must be a more general issue
  • w

    witty-crayon-22786

    03/02/2016, 9:03 PM
    my point is: it's not worth supporting concurrency
  • a

    ancient-stone-50795

    03/02/2016, 9:04 PM
    Maybe we could do the fetch as <tool>.tmp then atomic rename to <tool>
  • w

    witty-crayon-22786

    03/02/2016, 9:04 PM
    it's already supposed to be doing that
  • w

    witty-crayon-22786

    03/02/2016, 9:04 PM
    ...but it probably only does it for the initial 
    curl
  • w

    witty-crayon-22786

    03/02/2016, 9:05 PM
    or 
    requests
  • b

    bored-art-40741

    03/02/2016, 9:05 PM
    Still doesn't explain what's going on here
  • b

    bored-art-40741

    03/02/2016, 9:05 PM
    Supporting concurrency is trivial and it's silly not to
  • b

    bored-art-40741

    03/02/2016, 9:06 PM
    But I think we actually do in theory
  • b

    bored-art-40741

    03/02/2016, 9:06 PM
    Now I'm suspicious that this is just some sort of weird race condition
  • b

    bored-art-40741

    03/02/2016, 9:06 PM
    That has nothing to do with concurrency
  • b

    bored-art-40741

    03/02/2016, 9:06 PM
    Though I don't see how
  • w

    witty-crayon-22786

    03/02/2016, 9:07 PM
    after fetching the binary, the stack package does additional bootstrap steps.
  • w

    witty-crayon-22786

    03/02/2016, 9:07 PM
    so the fetch is protected, but i'm guessing the additional bootstrap steps are not.
  • b

    bored-art-40741

    03/02/2016, 9:08 PM
    Oh weird, a lot of the other shards failed with git/jenkins issues
  • b

    bored-art-40741

    03/02/2016, 9:08 PM
    Something I've also seen before, but only on startup
  • b

    bored-art-40741

    03/02/2016, 9:08 PM
    It's almost like the disk itself is shakey when it first comes up
  • w

    witty-crayon-22786

    03/02/2016, 9:10 PM
  • a

    ancient-stone-50795

    03/02/2016, 9:35 PM
    @witty-crayon-22786: I thought the whole reason for ~/.cache/pants was to not be in the workdir so it would survive a clean-all
  • w

    witty-crayon-22786

    03/02/2016, 9:35 PM
    right. as mentioned on the ticket, that'
  • w

    witty-crayon-22786

    03/02/2016, 9:35 PM
    s... what the build cache is for now
  • a

    ancient-stone-50795

    03/02/2016, 9:36 PM
    BUT the default build cache is... inside the workdir
  • w

    witty-crayon-22786

    03/02/2016, 9:36 PM
    clean-all doesn't delete $workdir/.cache
1...102610271028...2420Latest