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

    aloof-angle-91616

    09/23/2018, 4:41 PM
    because (1) i literally have done this but with coffeescript (2) was just thinking if i could get auto-
    Debug
    into rustc i could be twice as obnoxious
  • a

    aloof-angle-91616

    09/23/2018, 4:42 PM
    maybe withoutboats will think i'm cool if i dye my hair i've been going to do it for a while
  • a

    aloof-angle-91616

    09/23/2018, 4:44 PM
    oh and i'm absolutely going to make all my errors PoisonError type like a pokemon from now on
  • a

    aloof-angle-91616

    09/23/2018, 4:46 PM
    ok also this is way confusing documentation at https://doc.rust-lang.org/std/sync/struct.RwLock.html#method.write
  • a

    aloof-angle-91616

    09/23/2018, 4:47 PM
    This function will return an error if the RwLock is poisoned. An RwLock is poisoned whenever a writer panics while holding an exclusive lock. An error will be returned when the lock is acquired.
  • a

    aloof-angle-91616

    09/23/2018, 4:47 PM
    if you look into the source it means there are two error cases, one of which is being poisoned: https://github.com/rust-lang/rust/blob/be91c35f3466843713b9b6c5a7238ba83aef3602/src/libstd/sync/rwlock.rs#L436
  • a

    aloof-angle-91616

    09/23/2018, 4:48 PM
    and it's clear the "errors" section just doesn't work for methods that return `Result`s that aren't just exceptions which is fine but it's confusing and we're gr9 now
  • a

    aloof-angle-91616

    09/23/2018, 4:51 PM
    or maybe it's just me as usual
  • a

    aloof-angle-91616

    09/23/2018, 5:01 PM
    also racer (emacs rust support) is crazy because cargo is so good
    • 1
    • 1
  • a

    aloof-angle-91616

    09/23/2018, 5:46 PM
    did indeed cause a panic on attempted re-entrant usage of this resettable and then also got:
    Copy code
    thread panicked while processing panic. aborting.
Fatal Python error: Illegal instruction
    • 1
    • 2
  • a

    aloof-angle-91616

    09/23/2018, 6:03 PM
    that's more like it
    Copy code
    thread '<unnamed>' panicked at 'we aren't re-entrant, but are being re-entered: "WouldBlock" (write)', resettable/src/lib.rs:69:21
  • a

    aloof-angle-91616

    09/23/2018, 6:44 PM
    the panic was because the read lock i inserted wasn't let go before the write lock i inserted so that's good
  • a

    aloof-angle-91616

    09/23/2018, 6:52 PM
    ...which also fixed the deadlock on 
    clean-all
    . (not?) very mysterious
  • a

    aloof-angle-91616

    09/23/2018, 6:54 PM
    (the 
    clean-all
    deadlock was introduced when i made resettable lazy (again))
  • w

    witty-crayon-22786

    09/24/2018, 4:35 PM
    @aloof-angle-91616: was your original hang https://github.com/pantsbuild/pants/issues/6510 ?
  • a

    aloof-angle-91616

    09/24/2018, 4:36 PM
    yes after one second of seeing that link, looking again
  • a

    aloof-angle-91616

    09/24/2018, 4:38 PM
    i can probably make a stacktrace to corroborate that as mentioned in the ticket if needed. one of the things i tried was to raise on an attempted recursive write lock, but thinking about it now i don’t know what try_write() is supposed to do in that situation and assumed it would error nicely but perhaps not. i don’t think it reproes on os x so i could have gotten the other PR up in the meantime probably
  • w

    witty-crayon-22786

    09/24/2018, 4:39 PM
    i can probably make a stacktrace to corroborate that as mentioned in the ticket if needed
    it would be good to confirm that it is the same issue, yes.
  • a

    aloof-angle-91616

    09/24/2018, 4:39 PM
    great
  • w

    witty-crayon-22786

    09/24/2018, 4:39 PM
    it already raises on a recursive write lock
  • w

    witty-crayon-22786

    09/24/2018, 4:40 PM
    (afaik... i've certainly seen a bunch of them)
  • a

    aloof-angle-91616

    09/24/2018, 4:40 PM
    oh ok
  • w

    witty-crayon-22786

    09/24/2018, 4:40 PM
    get the stack traces, if you haven't
  • w

    witty-crayon-22786

    09/24/2018, 4:41 PM
    gdb -p $pid
    and then 
    thread all apply backtrace
    👍 1
    a
    • 2
    • 9
  • a

    aloof-angle-91616

    09/24/2018, 4:41 PM
    i think there might be some info to dump here but it think the stack traces are all that’s needed to serialize this progress for now
  • a

    aloof-angle-91616

    09/24/2018, 4:41 PM
    doing that now
  • w

    witty-crayon-22786

    09/24/2018, 5:07 PM
    er... does your comment say that gdb segfaulted...?
  • a

    aloof-angle-91616

    09/24/2018, 5:12 PM
    yep
  • a

    aloof-angle-91616

    09/24/2018, 5:13 PM
    not sure if any known issues w that w this version of gdb
  • a

    aloof-angle-91616

    09/24/2018, 5:23 PM
    note that the pants process didn’t die at all
1...576577578...906Latest