Re: [rvm-research] Is it possible to create a new TraceGenerator in a different file

[Zvi E. <ze...@cs...>]

Eliot Moss wrote:
> 
>> Eliot Moss wrote:
>>> And why do want the something different in the first place? (You may
>>> have
>>> very 
>>> good reasons, but it tends ot be helpful to know them :-) ...)
>>>
>> As I think about it today, I can't recall all of my reasons for thinking
>> it
>> might be useful to put different techniques in different classes.  What I
>> can recall is thinking that if every technique gets put into the same
>> file,
>> as more techniques are developed and implemented, the file will grow and
>> become more and more complex, which will make it much harder to work with
>> in
>> the future.  Also, switching between techniques will require modifying
>> the
>> TraceGenerator class (since the boolean for MERLIN_ANALYSIS is there),
>> whereas splitting different techniques into different classes could allow
>> the switching to be done in the plan by simply switching which class is
>> used.  I'm not sure how good those reasons are for warranting putting
>> different techniques in different classes, though.
> 
> Sure -- perhaps there should be an abstract class or an interface
> describing
> the API to trace generation, and then a concrete class that one uses. This
> is
> not unlike the setup for choosing which GC to use, so I suggest you get a
> handle on how that is done at build time, etc. I strongly agree withb the
> principles involved in making different techniques separate, yet putting
> their
> commonality into a superclass or other shared program structure.
> 
> Best wishes -- Eliot
> 

I've been working on a simple refactoring of the trace generation code
(which I think will prove to be temporary, since I'm not sure there isn't a
better way to accomplish it.  I'm currently working on reimplementing the
Merlin algorithm using that framework (since it needed to be fixed anyway,
as it is broken in 2.9.3).  I have the base alforithm working correctly, but
I noticed that org.mmtk.utility.TraceGenerator.java had a function
boot(Address bootStart) that seemed to add objects in the bootimage to the
trace, if I'm not mistaken.  In version 2.9.3, this code does not seem to be
called, so I was wondering if it was ever called, or if it was a leftover,
and if it used to be called when the trace generation worked, what was the
mechanism for calling it?

Zvi Effron
Harvey Mudd College
-- 
View this message in context: http://www.nabble.com/Is-it-possible-to-create-a-new-TraceGenerator-in-a-different-file-tp17993932p18450297.html
Sent from the jikesrvm-researchers mailing list archive at Nabble.com.