[Ironruby-core] Code Review: Thread#raise

Tomas Matousek Tomas.Matousek at microsoft.com
Sat Jan 3 14:52:03 EST 2009


I think we should at least report a warning when Thread#kill, Thread#raise or Thread#critical= is called if not eliminating them as Charlie proposed on his blog.

Tomas

-----Original Message-----
From: Curt Hagenlocher
Sent: Tuesday, December 30, 2008 10:05 PM
To: Shri Borde; IronRuby External Code Reviewers
Cc: ironruby-core at rubyforge.org
Subject: RE: Code Review: Thread#raise

Shouldn't the option "UseThreadAbortForSyncRaise" be called "...ForASyncRaise"?
I think that Thread.raise with no arguments should just inject a RuntimeError with no message as if $! were nil; this makes more sense than failing.  Trying to reference a "current exception" in another thread is a scary operation even if that's what MRI is doing.

Other than that, changes look really nice.  But anyone thinking of using this functionality should read Charlie's excellent piece from earlier in the year: http://blog.headius.com/2008/02/rubys-threadraise-threadkill-timeoutrb.html

-----Original Message-----
From: Shri Borde
Sent: Monday, December 29, 2008 3:00 PM
To: IronRuby External Code Reviewers
Cc: ironruby-core at rubyforge.org
Subject: Code Review: Thread#raise

  tfpt review "/shelveset:raise;REDMOND\sborde"
  Comment  :
  Implements Thread#raise using Thread.Abort, and added tests for it
  Implemented stress mode (RubyOptions.UseThreadAbortForSyncRaise) which found more issues. Fixed most but not all
  Enabled test for timeout as well
  Remaining work (not high pri for now)
   - Thread#raise without Exception parameters is not supported as it needs to access the active exception of the target thread. This is stored as a thread-local static, and so cannot be accessed from other threads. Can be fixed by not using ThreadStaticAttribute.
   - Adding probes (in generated code, in C# library code, etc) will help to raise the exception quicker as Thread.Abort can be delayed indefinitely. Ideally, we need both approaches. For now, using Thread.Abort goes a long way.
   - Ideally, we would add a try-catch to the IDynamicObject/MetaObject code paths so that when other languages called into Ruby code, they would get the expected user exception rather than a ThreadAbortException

  RunRSpec: supports -ruby to run with MRI. Its much faster than doing "rake ruby why_regression". Added support for -e to run a specific example




More information about the Ironruby-core mailing list