From @AlexDaniel

Here's the output when it fails​:

ok 1 - Can do an atomic fetch from a Scalar container
ok 2 - Can do an atomic assign to a Scalar container; returns new value
ok 3 - Atomic fetch after atomic assign shows latest value
ok 4 - Updated value seen by non-atomic too
  1..2
  ok 1 - code dies
  ok 2 - right exception type (X​::TypeCheck​::Assignment)
ok 5 - Cannot atomic assign value of the wrong type
ok 6 - No hang due to incorrect lifting of atomic fetch out of loop
ok 7 - Can do an atomic fetch from an int container
ok 8 - Can do an atomic assign to an int container; returns new value
ok 9 - Atomic int fetch after atomic int assign shows latest value
ok 10 - Updated value seen by non-atomic too
A IntLexRef container does not know how to do an atomic load
  in block <unit> at t/spec/S17-lowlevel/atomic-ops.t line 41

# Looks like you planned 28 tests, but ran 10

I noticed that it fails more often when the system is under load (e.g. when running the spectest). Other than that I don't have any info.

From @AlexDaniel

Oh. A crude way to reproduce it is​:

while :; do clear; PERL6LIB=lib ./perl6-m t/spec/S17-lowlevel/atomic-ops.t || break; done

And just leave it running, it'll fail at some point. Get your system busy with something and it'll fail faster.

On 2017-09-05 00​:14​:11, alex.jakimenko@​gmail.com wrote​:

Here's the output when it fails​:

ok 1 - Can do an atomic fetch from a Scalar container
ok 2 - Can do an atomic assign to a Scalar container; returns new
value
ok 3 - Atomic fetch after atomic assign shows latest value
ok 4 - Updated value seen by non-atomic too
1..2
ok 1 - code dies
ok 2 - right exception type (X​::TypeCheck​::Assignment)
ok 5 - Cannot atomic assign value of the wrong type
ok 6 - No hang due to incorrect lifting of atomic fetch out of loop
ok 7 - Can do an atomic fetch from an int container
ok 8 - Can do an atomic assign to an int container; returns new value
ok 9 - Atomic int fetch after atomic int assign shows latest value
ok 10 - Updated value seen by non-atomic too
A IntLexRef container does not know how to do an atomic load
in block <unit> at t/spec/S17-lowlevel/atomic-ops.t line 41

# Looks like you planned 28 tests, but ran 10

I noticed that it fails more often when the system is under load (e.g.
when running the spectest). Other than that I don't have any info.

From @niner

A golfed version that reliably fails​:

{
  my Int $test-cont = 42;
  ⚛$test-cont;
}
{
  my atomicint $set = 0;
  start { sleep 1; $set ⚛= 1 };
  until ⚛$set { }
}

The important bit is the Int being read atomicly before reading the atomicint. The sleep makes it fail reliably.

From @jnthn

On Tue, 05 Sep 2017 00​:36​:34 -0700, nine@​detonation.org wrote​:

A golfed version that reliably fails​:

{
my Int $test-cont = 42;
⚛$test-cont;
}
{
my atomicint $set = 0;
start { sleep 1; $set ⚛= 1 };
until ⚛$set { }
}

The important bit is the Int being read atomicly before reading the
atomicint. The sleep makes it fail reliably.

And that in turn got it down to being a spesh bug (this golf reliably worked with MVM_SPESH_DISABLE=1), which was that it mis-optimized the call to the wrong multi candidate, due to some bogus logging. Fixed with MoarVM/MoarVM@5bf652e (though the spesh output it isn't too great, so I'll now see if I can do some further tweaks to get it inlining the prefix​:<⚛> for atomicint).

/jnthn

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

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

From @lizmat

FWIW, I couldn’t get this to fail on my MBP *while* running make spectest with TEST_JOBS=8

Perhaps this is OS / CPU / compiler dependent?

On 5 Sep 2017, at 09​:17, Aleks-Daniel Jakimenko-Aleksejev via RT <perl6-bugs-followup@​perl.org> wrote​:

Oh. A crude way to reproduce it is​:

while :; do clear; PERL6LIB=lib ./perl6-m t/spec/S17-lowlevel/atomic-ops.t ||
break; done

And just leave it running, it'll fail at some point. Get your system busy with
something and it'll fail faster.

On 2017-09-05 00​:14​:11, alex.jakimenko@​gmail.com wrote​:

Here's the output when it fails​:

ok 1 - Can do an atomic fetch from a Scalar container
ok 2 - Can do an atomic assign to a Scalar container; returns new
value
ok 3 - Atomic fetch after atomic assign shows latest value
ok 4 - Updated value seen by non-atomic too
1..2
ok 1 - code dies
ok 2 - right exception type (X​::TypeCheck​::Assignment)
ok 5 - Cannot atomic assign value of the wrong type
ok 6 - No hang due to incorrect lifting of atomic fetch out of loop
ok 7 - Can do an atomic fetch from an int container
ok 8 - Can do an atomic assign to an int container; returns new value
ok 9 - Atomic int fetch after atomic int assign shows latest value
ok 10 - Updated value seen by non-atomic too
A IntLexRef container does not know how to do an atomic load
in block <unit> at t/spec/S17-lowlevel/atomic-ops.t line 41

# Looks like you planned 28 tests, but ran 10

I noticed that it fails more often when the system is under load (e.g.
when running the spectest). Other than that I don't have any info.