Re: [Jikesrvm-core] Move Constants from VM_HeapLayoutConstants

[Steve B. <Ste...@an...>]

On 24/12/2006, at 10:16 AM, Eliot Moss wrote:

>>>>>> "Peter" == Peter Donald <pe...@re...> writes:
>
>     Peter> On 12/24/06, Eliot Moss <mo...@cs...> wrote:
>>> My question is this: Do we want all the class for all of these in
>>> the same boot image? This might lead to serious boot image bloat.  
>>> How do
>>> you avoid it?
>
>     Peter> I recorded what Dave said about the minimal Primordial  
> Class List at
>     Peter> [1]. So in theory if we can automagically determine this  
> minimal set
>     Peter> then the size of the boot image would be limited to that  
> size and
>     Peter> should be mostly stable across configurations. Each  
> configuration
>     Peter> could optionally include its related packages (ie **/ 
> *IA32*.class for
>     Peter> intel target architecture, org/ibm/jikesrvm/opt/**/ 
> *.class for
>     Peter> optimizing compiler etc)
>
>     Peter> [1] http://jikesrvm.sourceforge.net/wiki/index.php/ 
> Primordial_Class_List
>
> Yes, except some of the coding strategies you listed would require  
> all the
> classes to be present, because they're all mentioned.

I'm not entirely sure what you're referring to (Eliot), but let's be  
careful to distinguish presence at javac compile time and presence in  
the boot image.

These need not be the same thing and there are a variety of  
strategies for dealing with pruning.

Here you can see what I did a few days ago.   This was an entirely  
pragmatic step in the process of getting rid of the preprocessor.   
I'm very happy to recode it once we settle on a consistent pattern,  
so I'm totaly open to (constructive ;) critique.

In this case, VM_Configuration (which is "generated" code) nails down  
at build time a particular archHelper singleton, which will be an  
arch-specific subtype of VM_MachineSpecific.java:

http://svn.sourceforge.net/viewvc/jikesrvm/rvmroot/trunk/rvm/src/com/ 
ibm/jikesrvm/VM_MachineSpecific.java?view=markup
http://svn.sourceforge.net/viewvc/jikesrvm/rvmroot/trunk/rvm/src-ia32/ 
com/ibm/jikesrvm/VM_MachineSpecificIA.java?view=markup
http://svn.sourceforge.net/viewvc/jikesrvm/rvmroot/trunk/rvm/src-ppc/ 
com/ibm/jikesrvm/VM_MachineSpecificPowerPC.java?view=markup
http://svn.sourceforge.net/viewvc/jikesrvm/rvmroot/trunk/rvm/src- 
generated/vm-configuration/VM_Configuration.java? 
revision=11245&view=markup

Given the cleanup I'm undertaking right now, I'd move  
jikesrvm.VM_MachineSpecificIA to jikesrvm.ia.VM_MachineSpecific and  
VM_MachineSpecificPowerPC to jikesrvm.ppc.VM_MachineSpecific, while  
retaining the naming of the subclasses (modulo the details of "EM64T"  
v "IA64" v "x86-64" etc).

--Steve