Re: [bug] serious memory leak + continuations

Michael Neumann wrote:
> Hi,
> 
> ruby 1.9.0 (2004-12-20) [i386-freebsd5.2.1]
> gcc version 3.3.3 [FreeBSD] 20031106
> 
> In my application, if I change:
> 
>    # (1) this works
>    callback_stream.with_callbacks_for(self, :action) { |callback, val|
>       catch(:something) { callback.call }
>       throw :wee_back_to_session
>     }
> 
> into:
> 
>    # (2) this leaks
>    callback_stream.with_callbacks_for(self, :action) { |callback, val|
>       res = catch(:something) { callback.call }
>       throw :wee_back_to_session
>     }
> 
> or into:
> 
>     # (3) this leaks
>     callback_stream.with_callbacks_for(self, :action) { |callback, val|
>       callback.call
>       throw :wee_back_to_session
>     }
> 
> memory consumption is unbounded! BTW, ":something" is never thrown. It 
> seems that if I reference the return value of callback.call, the memory 
> leak appears. The leak also happens if I change "res = " in (2) into 
> "@res = " or "$res = ". Note that inside callback.call continuations are 
> created.
> 
> And if I return "nil" from the block (callback.call), it leaks, too, 
> even if I use code sample (1).
> 
> I tried above example also with ruby-stable compiled with 2.95.4 (-O0 
> enabled), but there all three examples leak memory.
> 
> It seems to be related to the "unknown node type 0 bug", as it appears 
> to be pretty random (at least in my eyes):
> 
> http://blade.nagaokaut.ac.jp/cgi-bin/scat.rb/ruby/ruby-talk/123694
> 
> 
> Help! I can send the whole sources to someone who is interested. Only 
> requirements are rubygems, installed narf gem (webunit) and gnuplot to 
> display memory consumption, and of course *nix.

The memory leak probably results from the fact that the continuations 
are not garbage collected. The number of them keep growing in (2) and 
(3), whereas in (1) it stays constant.

Regards,

   Michael