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

    rough-minister-58256

    10/21/2015, 11:26 PM
    technically this await_socket() call should guard against metadata staleness until the new process is launched: https://github.com/pantsbuild/pants/blob/master/src/python/pants/backend/core/tasks/reporting_server.py#L52
  • r

    rough-minister-58256

    10/21/2015, 11:26 PM
    then its just passthrough of ProcessManager.{pid,socket}
  • r

    rough-minister-58256

    10/21/2015, 11:29 PM
    ah, hmm. I bet the await_socket() is passing immediately based on the old pidfile existing, but it still gets daemonized
  • r

    rough-minister-58256

    10/21/2015, 11:30 PM
    I have a strawman branch that purges metadata more aggressively, can test that against your repro
  • r

    rough-minister-58256

    10/21/2015, 11:30 PM
    I think this is probably the same metadata staleness issue you described in your HP Lovecraft rendition a few weeks ago
  • f

    fancy-queen-20734

    10/21/2015, 11:39 PM
    lol
  • f

    freezing-motorcycle-14222

    10/21/2015, 11:44 PM
    So I was learning a bit more about how 
    pants
    invalidation works and I was wondering why the 
    TaskBase.invalidated
    function defaults 
    invalidate_dependents
    to 
    False
    . Wouldn't the common case be to transitively invalidate any reverse dependencies?
  • f

    freezing-motorcycle-14222

    10/21/2015, 11:47 PM
    The only reason I'm asking this is not to argue that the default should be changed. Rather, I just want to make sure that there isn't some other parallel mechanism for invalidating reverse dependencies that I'm not aware of
  • w

    witty-crayon-22786

    10/22/2015, 12:15 AM
    @limited-terabyte-34878: looks like there is an actual failure
  • b

    bored-art-40741

    10/22/2015, 12:24 AM
    @freezing-motorcycle-14222: I believe the answer here is somewhat historical. I believe if we were writing this in a vacuum today (sans any new engine influence), we would default that to 
    True
    , but the thrash in changing the default is too large for it to be worthwhile
  • b

    bored-art-40741

    10/22/2015, 12:24 AM
    And in general, for any task that doesn't actually care about transitive invalidation, it avoids a potentially costly topological sort
  • l

    limited-country-20626

    10/22/2015, 2:23 PM
    We have noticed that wrong server port as well, @fancy-queen-20734. Also, @full-kilobyte-93429 I am testing 3003 internally today…we were a couple releases behind so I had to port our codebase up to HEAD.
  • f

    freezing-motorcycle-14222

    10/22/2015, 3:56 PM
    @bored-art-40741: Thanks for the explanation! Sorry, I got pulled away yesterday
  • f

    full-kilobyte-93429

    10/22/2015, 4:02 PM
    @limited-country-20626: thanks i was about to suggest test this before shipping. there are two additional code changes needed, user code
  • f

    full-kilobyte-93429

    10/22/2015, 4:03 PM
    one is similar to repl but calling from test, the other is weka library (has its own classCache)
  • w

    witty-crayon-22786

    10/22/2015, 6:29 PM
    @enough-analyst-54434: hey... any thoughts on how we might
  • w

    witty-crayon-22786

    10/22/2015, 6:29 PM
    differentiate "stable/public" products from private products
  • w

    witty-crayon-22786

    10/22/2015, 6:30 PM
    currently, everything under 
    dist
    is public, and has an assumed stable layout/api
  • w

    witty-crayon-22786

    10/22/2015, 6:32 PM
    the first usecase is that the intellij plugin needs to be able to consume the output classes on a per target basis, and currently it does that by looking in 
    .pants.d
    , which has broken a bunch of times.
  • w

    witty-crayon-22786

    10/22/2015, 6:33 PM
    but now it has been well and truly broken (intentionally) by https://rbcommons.com/s/twitter/r/2991/
  • e

    enough-analyst-54434

    10/22/2015, 6:34 PM
    Use a different dir then, say .dist/ or ~/.cache/pants/...
  • w

    witty-crayon-22786

    10/22/2015, 6:34 PM
    a strawman proposal: a 
    dist/${product_name}/${target.id}/$stuff
  • e

    enough-analyst-54434

    10/22/2015, 6:35 PM
    If it's private it should not go in dist imo
  • w

    witty-crayon-22786

    10/22/2015, 6:35 PM
    i guess what i'm saying is: this isn't private
  • w

    witty-crayon-22786

    10/22/2015, 6:35 PM
    concretely, this might be: 
    dist/runtime_classpath/${target.id}/CLASSPATH
  • w

    witty-crayon-22786

    10/22/2015, 6:36 PM
    (again, strawman)
  • w

    witty-crayon-22786

    10/22/2015, 6:36 PM
    and bundle bundles would go under there, etc.
  • e

    enough-analyst-54434

    10/22/2015, 6:36 PM
    I still don't understand "private" can I ignore that word?
  • w

    witty-crayon-22786

    10/22/2015, 6:36 PM
    and to admint prior art, i think this is approximately what buck does, except the outputs are rule-name-based
  • w

    witty-crayon-22786

    10/22/2015, 6:37 PM
    @enough-analyst-54434: some products aren't intended to be relied on by external processes... some are.
1...577578579...2420Latest