From @hoelzro

See the attached script.

The worker(Any, Int) sub is called from within a start block with only a single Any argument; I would expect this to either fail at compile time, throw an exception in the await'ing thread, or set the statuses of the Promises in @​workers to a failed state. Instead, different behavior occurs across repeated execution non-deterministically, namely​:

  - Sometimes await() never returns and just blocks forever.
  - Sometimes "Too few positionals passed" is raised during execution.
  - Sometimes I see a "cannot write to readonly value warning"
  - I used to see "Type check failed in binding exception; expected 'Any' but got 'Mu'", but that's not appearing in my latest tests.

From @hoelzro

use v6;

sub worker(Any $a, Int $b) {}

my $value = Any;

my @​workers = (^2).map​: {
  start { worker($value) }
};

await @​workers;

From @jnthn

On Sat Jun 06 21​:01​:10 2015, rob@​hoelz.ro wrote​:

See the attached script.

The worker(Any, Int) sub is called from within a start block with only
a single Any argument; I would expect this to either fail at compile
time, throw an exception in the await'ing thread, or set the statuses
of the Promises in @​workers to a failed state. Instead, different
behavior occurs across repeated execution non-deterministically,
namely​:

- Sometimes await() never returns and just blocks forever.
- Sometimes "Too few positionals passed" is raised during execution.
- Sometimes I see a "cannot write to readonly value warning"
- I used to see "Type check failed in binding exception; expected
'Any' but got 'Mu'", but that's not appearing in my latest tests.

Changing the await line to​:

loop {
  sleep 1;
  say @​workers[0].status ~ ' ' ~ @​workers[1].status;
}

Shows that sometimes the first Promise ends up Planned and second Broken, when both should be Broken. This explains the hang case. As for why, it golfs to this​:

my @​log = [] xx 4;
for ^4 -> $t {
  Thread.start({
  for ^10000 {
  try {
  die 'omg';
  CATCH {
  default {
  push @​log[$t], $*THREAD.id();
  }
  }
  }
  }
  });
}
say [==] @​$_ for @​log;

Which should always print four Trues, but instead can come out with some of them False. This means it's breaking the wrong Promise. Almost certainly an issue with how we compile CATCH.

The RT System itself - Status changed from 'new' to 'open'

From @jnthn

On Sat Jun 06 21​:01​:10 2015, rob@​hoelz.ro wrote​:

See the attached script.

The worker(Any, Int) sub is called from within a start block with only
a single Any argument; I would expect this to either fail at compile
time, throw an exception in the await'ing thread, or set the statuses
of the Promises in @​workers to a failed state. Instead, different
behavior occurs across repeated execution non-deterministically,
namely​:

- Sometimes await() never returns and just blocks forever.
- Sometimes "Too few positionals passed" is raised during execution.
- Sometimes I see a "cannot write to readonly value warning"
- I used to see "Type check failed in binding exception; expected
'Any' but got 'Mu'", but that's not appearing in my latest tests.

This got fixed; it was a code-gen bug in the when/default stuff that is used inside of CATCH in the end. Just added a test, that seems to reliably pass, to S17-promise/start.t.

@jnthn - Status changed from 'open' to 'resolved'