Re: [RFC] Number of insn slots used for method compilation

[Bryce McKinlay <mckinlay at redhat dot com>]

Tom Tromey wrote:
> Keith> Other than adding a variable to hold this information (a la the
> Keith> attached patch), is there another way that won't require using any
> Keith> more memory? I don't think that code_length can safely be changed to
> Keith> store this information.
>
> Actually, I think it would be safe to update this.  After compiling a
> method we don't use it any more, do we?
>   
I don't think we use it at present. However, the JDWP ByteCodes command 
provides access to the original bytecode for the method, so we need 
access to code_length somehow in order to implement that.

An alternative that would possibly save memory would be to perhaps 
simply re-parse the method attributes from the underlying bytecode when 
the ByteCodes op is used. Since this is used only by debuggers, 
performance isn't going to be an issue. But, of course, it'll be easier 
to implement by adding a separate field, and the extra memory 
consumption won't be all that significant in the context of the whole 
runtime.

Bryce