Intermittent crash of /css-backgrounds-3_dev/html4/background-016.htm

[asajeffrey]

  ▶ CRASH [expected PASS] /css-backgrounds-3_dev/html4/background-016.htm
  │ 
  │ Shutting down the Constellation after generating an output file or exit flag specified
  │ thread 'ScriptThread PipelineId { namespace_id: PipelineNamespaceId(0), index: PipelineIndex(0) }' panicked at 'called `Result::unwrap()` on an `Err` value: Error { repr: Custom(Custom { kind: NotFound, error: StringError("Bogus destination port.") }) }', ../src/libcore/result.rs:746
  │ stack backtrace:
  │    1:        0x1051072e8 - std::sys::backtrace::tracing::imp::write::h714760a4c8c0cdd8
  │    2:        0x10510af75 - std::panicking::default_hook::_$u7b$$u7b$closure$u7d$$u7d$::hff309ab1d83ffd90
  │    3:        0x10510ab8c - std::panicking::default_hook::h08ad3bb09872855b
  │    4:        0x1050f54b8 - std::sys_common::unwind::begin_unwind_inner::h406d5f1a330b854b
  │    5:        0x1050f676e - std::sys_common::unwind::begin_unwind_fmt::h57ea3fbee1a40196
  │    6:        0x1051065b7 - rust_begin_unwind
  │    7:        0x1051329b0 - core::panicking::panic_fmt::ha6b3c19493c123b3
  │    8:        0x1047d2b0e - core::result::unwrap_failed::he41fdbb4a3e1914b
  │    9:        0x1047d277d - profile_traits::mem::ProfilerChan::send::ha3c06ebf43268c25
  │   10:        0x1041a12dc - std::sys_common::unwind::try::try_fn::h7685a677ea889c26
  │   11:        0x10510654b - __rust_try
  │   12:        0x1051064d3 - std::sys_common::unwind::inner_try::h4e97625a08807651
  │   13:        0x1041a248a - _<F as std..boxed..FnBox<A>>::call_box::h8bac8015289dad49
  │   14:        0x105109fe8 - std::sys::thread::Thread::new::thread_start::h74af400293164137
  │   15:     0x7fff86bcf059 - _pthread_body
  │   16:     0x7fff86bcefd6 - _pthread_start
  │ thread 'ScriptThread PipelineId { namespace_id: PipelineNamespaceId(0), index: PipelineIndex(0) }' panicked at 'called `Option::unwrap()` on a `None` value', ../src/libcore/option.rs:326
  │ stack backtrace:
  │    1:        0x1051072e8 - std::sys::backtrace::tracing::imp::write::h714760a4c8c0cdd8
  │    2:        0x10510af75 - std::panicking::default_hook::_$u7b$$u7b$closure$u7d$$u7d$::hff309ab1d83ffd90
  │    3:        0x10510ab8c - std::panicking::default_hook::h08ad3bb09872855b
  │    4:        0x1050f54b8 - std::sys_common::unwind::begin_unwind_inner::h406d5f1a330b854b
  │    5:        0x1050f676e - std::sys_common::unwind::begin_unwind_fmt::h57ea3fbee1a40196
  │    6:        0x1051065b7 - rust_begin_unwind
  │    7:        0x1051329b0 - core::panicking::panic_fmt::ha6b3c19493c123b3
  │    8:        0x105132cac - core::panicking::panic::h188c6c6a0fe5463c
  │    9:        0x103e29ff9 - script::page::Page::window::h010a05681584eec2
  │   10:        0x10419ef02 - _<script_thread..ScriptMemoryFailsafe<'a> as core..ops..Drop>::drop::h74f0f1b638973d0e
  │   11:        0x1041a21fe - std::sys_common::unwind::try::try_fn::h7685a677ea889c26
  │   12:        0x10510654b - __rust_try
  │   13:        0x1051064d3 - std::sys_common::unwind::inner_try::h4e97625a08807651
  │   14:        0x1041a248a - _<F as std..boxed..FnBox<A>>::call_box::h8bac8015289dad49
  │   15:        0x105109fe8 - std::sys::thread::Thread::new::thread_start::h74af400293164137
  │   16:     0x7fff86bcf059 - _pthread_body
  │   17:     0x7fff86bcefd6 - _pthread_start
  └ thread panicked while panicking. aborting.

[asajeffrey]

Some IRC chat about this between myself (ajeffrey) and @danlrobertson (dlr): http://logs.glob.uno/?c=mozilla%23servo#c402478 including a theory http://logs.glob.uno/?c=mozilla%23servo#c402621:

• Pipeline::exit is asynchronous: there is no ack channel,
• so Constellation::handle_exit doen't wait for the pipelines to finish closing down before sending a ShutdownComplete message to the compositor,
• so Compositor::finish_shutting_down might shut down the profiler before each pipeline is closed,
• so ProfilerChan::send can end up being called by a script thread after the profiler has shut down.

This theory is compatible with the stack trace we're seeing.

[asajeffrey]

Choices for what to do about it:

a) Pipeline::exit without ack (makes profiling lossy)
b) Pipeline::exit with ack (may cause shutdown to deadlock waiting for an ack)
c) Pipeline::exit with ack but timeout (yet another source of nondeterminism)

None of these fill me with glee.

[asajeffrey]

Suggestion due to @metajack: normal execution is async, but we add in a sync option for the cases where we care about accurate profiling information during shutdown.

[asajeffrey]

This should be fixed by #11585.

[asajeffrey]

Not been reported for a while.