Issue #15992 has been updated by Eregon (Benoit Daloze).


Should `mon_enter` and `mon_exit` use `Thread.handle_interrupt` inside their method definitions, so they are safe when called directly too?

`Thread.handle_interrupt(Exception => :never)` seems too much for `mon_enter`, I think `Thread.handle_interrupt(Exception => :blocking)` would be better.
Otherwise, it's not possible to interrupt a thread trying to get a contended monitor with `Thread#raise`:
```
ruby -rmonitor -e 'o=Object.new.extend(MonitorMixin); Thread.new{ o.mon_enter; sleep }; sleep 0.1; Thread.new{sleep 3; p :kill; Thread.main.raise "bye"}; p o.synchronize { p :in; sleep }'
```
Never exits on trunk, but it does exit on 2.6.2 and before. Even Thread#kill does not help.

Ctrl+C does interrupt it, because signal traps are not yet supported by `handle_interrupt`, but that might change.

----------------------------------------
Bug #15992: An exception breaks monitor state and cause deadlock
https://bugs.ruby-lang.org/issues/15992#change-81063

* Author: naruse (Yui NARUSE)
* Status: Closed
* Priority: Normal
* Assignee: 
* Target version: 
* ruby -v: 
* Backport: 2.5: DONE, 2.6: DONE
----------------------------------------
lib/monitor.rb provides Monitor.
But its state handling is weak for interrupts caused by Thread.kill for example timeout libraries.

Timeout exception may happen everywhere. If it raised when the thread is executing

```ruby
  def mon_exit
    mon_check_owner
    @mon_count -=1
    if @mon_count == 0
      @mon_owner = nil
      # HERE!!!
      @mon_mutex.unlock
    end
  end
``` 

It breaks the state of the monitor and it causes deadlock.



-- 
https://bugs.ruby-lang.org/

Unsubscribe: <mailto:ruby-core-request / ruby-lang.org?subject=unsubscribe>
<http://lists.ruby-lang.org/cgi-bin/mailman/options/ruby-core>