Ironruby core - Mar 2010 - What appears to be a really sneaky bug related to loading assemblies (System::Action is temporarily nil)

Intro: Using IronRuby 1.0 rc2; .NET 3.5SP1 on windows XP sp3
I have some code which is using DynamicMethod / GetILGenerator to create
statically typed methods at runtime, so I can call IUnknown-based COM
objects from IronRuby.

Here is a snippet

  # ... piles of code which figures out what the method signature should be
and generates it ...
  func_type_params = [System::Object] + someotherstuff

  # generate a .NET delegate to wrap the new DynamicMethod
  puts "DEBUG: Action= #{System::Action.inspect}"
  puts "DEBUG: Params= #{func_type_params.inspect}"
  puts "DEBUG: F= #{System::Action.of(*func_type_params)}"
  del_type = System::Action.of(*func_type_params).to_clr_type

This has been working reliably for some time, and still works with most
input. However, we''ve recently seen random failures. Here''s
some of the
output from those Debug lines:

  # successful
  DEBUG: Action= #<TypeGroup: System::Action, System::Action[T],
System::Action[T1, T2], System::Action[T1, T2, T3], System::Action[T1, T2,
T3, T4]>
  DEBUG: Params= [Object, Object, Object]
  DEBUG: F= System::Action[Object, Object, Object]

  # and then the next one fails
  DEBUG: Action= nil
  DEBUG: Params= [Object, Fixnum, Interop::FieldID[], System::Object[]]
  NoMethodError: undefined method `of'' for NilClass

As far as that tells me, System::Action has somehow been set to nil.
Note: The point at which System::Action gets nil''ed out is NOT always
the
same point.

If I run the code under iirb (copy/pasting my original script in), and then
after the exception inspect System::Action, it reports this:

=> #<TypeGroup: System::Action, System::Action[T], System::Action[T1, T2],
System::Action[T1, T2, T3], System::Action[T1, T2, T3, T4],
System::Action[T1, T2, T3, T4, T5], System::Action[T1, T2, T3, T4, T5, T6],
System::Action[T1, T2, T3, T4, T5, T6,
T7], System::Action[T1, T2, T3, T4, T5, T6, T7, T8], System::Action[T1, T2,
T3, T4, T5, T6, T7, T8, T9], System::Action[T1, T2, T3, T4, T5, T6, T7, T8,
T9, T10], System::Action[T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11],
System::Action[T1, T2, T3,
 T4, T5, T6, T7, T8, T9, T10, T11, T12], System::Action[T1, T2, T3, T4, T5,
T6, T7, T8, T9, T10, T11, T12, T13], System::Action[T1, T2, T3, T4, T5, T6,
T7, T8, T9, T10, T11, T12, T13, T14], System::Action[T1, T2, T3, T4, T5, T6,
T7, T8, T9, T10, T11,
 T12, T13, T14, T15], System::Action[T1, T2, T3, T4, T5, T6, T7, T8, T9,
T10, T11, T12, T13, T14, T15, T16]>

As you can see, System::Action is no longer nil, but has now been given a
bunch more overloads. These overloads come from System.CoreEx (part of the
.NET reactive framework). Putting require ''System.CoreEx'' at
the top of my
script to pre-load it makes this error go away.
>From what I can tell, something causes the CLR to on-demand load
System.CoreEx.dll. This in turn adds more overloads to System::Action.
IronRuby appears to temporarily nil it out while loading, thus causing the
crash.

Am I on the right track? As mentioned I think I can work around it by
preloading all the dll''s I know I''m going to need, but this
doesn''t seem to
be a reliable long-term solution.

Cheers, Orion
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://rubyforge.org/pipermail/ironruby-core/attachments/20100311/abeeb1de/attachment-0001.html>

Could you try to write a simple repro in a .rb file that demonstrates the issue?

Tomas

From: ironruby-core-bounces at rubyforge.org [mailto:ironruby-core-bounces at
rubyforge.org] On Behalf Of Orion Edwards
Sent: Wednesday, March 10, 2010 12:42 PM
To: ironruby-core at rubyforge.org
Subject: [Ironruby-core] What appears to be a really sneaky bug related to
loading assemblies (System::Action is temporarily nil)

Intro: Using IronRuby 1.0 rc2; .NET 3.5SP1 on windows XP sp3
I have some code which is using DynamicMethod / GetILGenerator to create
statically typed methods at runtime, so I can call IUnknown-based COM objects
from IronRuby.

Here is a snippet

  # ... piles of code which figures out what the method signature should be and
generates it ...
  func_type_params = [System::Object] + someotherstuff

  # generate a .NET delegate to wrap the new DynamicMethod
  puts "DEBUG: Action= #{System::Action.inspect}"
  puts "DEBUG: Params= #{func_type_params.inspect}"
  puts "DEBUG: F= #{System::Action.of(*func_type_params)}"
  del_type = System::Action.of(*func_type_params).to_clr_type

This has been working reliably for some time, and still works with most input.
However, we''ve recently seen random failures. Here''s some of
the output from those Debug lines:

  # successful
  DEBUG: Action= #<TypeGroup: System::Action, System::Action[T],
System::Action[T1, T2], System::Action[T1, T2, T3], System::Action[T1, T2, T3,
T4]>
  DEBUG: Params= [Object, Object, Object]
  DEBUG: F= System::Action[Object, Object, Object]

  # and then the next one fails
  DEBUG: Action= nil
  DEBUG: Params= [Object, Fixnum, Interop::FieldID[], System::Object[]]
  NoMethodError: undefined method `of'' for NilClass

As far as that tells me, System::Action has somehow been set to nil.
Note: The point at which System::Action gets nil''ed out is NOT always
the same point.

If I run the code under iirb (copy/pasting my original script in), and then
after the exception inspect System::Action, it reports this:

=> #<TypeGroup: System::Action, System::Action[T], System::Action[T1, T2],
System::Action[T1, T2, T3], System::Action[T1, T2, T3, T4], System::Action[T1,
T2, T3, T4, T5], System::Action[T1, T2, T3, T4, T5, T6], System::Action[T1, T2,
T3, T4, T5, T6,
T7], System::Action[T1, T2, T3, T4, T5, T6, T7, T8], System::Action[T1, T2, T3,
T4, T5, T6, T7, T8, T9], System::Action[T1, T2, T3, T4, T5, T6, T7, T8, T9,
T10], System::Action[T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11],
System::Action[T1, T2, T3,
 T4, T5, T6, T7, T8, T9, T10, T11, T12], System::Action[T1, T2, T3, T4, T5, T6,
T7, T8, T9, T10, T11, T12, T13], System::Action[T1, T2, T3, T4, T5, T6, T7, T8,
T9, T10, T11, T12, T13, T14], System::Action[T1, T2, T3, T4, T5, T6, T7, T8, T9,
T10, T11,
 T12, T13, T14, T15], System::Action[T1, T2, T3, T4, T5, T6, T7, T8, T9, T10,
T11, T12, T13, T14, T15, T16]>

As you can see, System::Action is no longer nil, but has now been given a bunch
more overloads. These overloads come from System.CoreEx (part of the .NET
reactive framework). Putting require ''System.CoreEx'' at the
top of my script to pre-load it makes this error go away.
>From what I can tell, something causes the CLR to on-demand load
System.CoreEx.dll. This in turn adds more overloads to System::Action.
IronRuby appears to temporarily nil it out while loading, thus causing the
crash.

Am I on the right track? As mentioned I think I can work around it by preloading
all the dll''s I know I''m going to need, but this
doesn''t seem to be a reliable long-term solution.

Cheers, Orion
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://rubyforge.org/pipermail/ironruby-core/attachments/20100310/12e5e65a/attachment.html>