[gclist] Distributed Garbage Collection Revisited 1

[H.C.C.Rodrigues]

Matthew,

I wrote:

>> I think it is not necessary an infinite number of new incoming references
>> appear before the distributed trace terminates to not terminate the algorithm.
>> We can have a cycle of objects spawning two processes being reachable from a
>> root either in one process or in the other. The DGC may never reach a root.
>>
>
>We are definitely not seeing eye-to-eye here.  Could you develop a
>scenario with a sequence of messages which shows what you mean?
>
>> >1) A remotely rooted object.  If there are no garbage processes, we
>> >can just use this as proof the object is alive. 
>>
>>I think the problem I presented does not exist if you consider the objects
>>  being remotely
>> accessed live. In this case I imagine that the back-search would terminate.
>> What would happen when a "grey" object is remoteley accessed?
>
>Think about this.  When can a grey object be accessed, and what does
>its being accessed mean, at each point in time?

Sorry about this. Is very dificult to explain what I am thinking by eamil.
I am going to try to design an example:


                  ___________________
                 /                  |
                 |                  |
                 V                  |
         R     ScionA1              |
         |       |                  |
         V       V                  |
       ObjA2<--ObjA1                |
         |       |                  |
         |       V                  |
         \---->StubB1               |
                 |       Space A    |                    
                 |      ----------  |
                 V       Space B    |
         R     ScionB1              |
         |       |                  |
         V       V                  |
       ObjB2<--ObjB1                |
                 |                  |
                 V                  |
               StubA1               |
                 |                  |
                 \-------------------

This is the example I was thinking. If you start a back-serach say at object
A1, the algorithm will colour grey ScionA1, StubA1, ObjB1, ScionB1, say at
time 1. T time 2, ObjA2 invokes ObjB1 that copies <ObjB1,StubA1> to ObjB2.
ObjB1 was acessed and it is grey.
Am I write?

However, the problem I presented in my first message, it is not a problem.
Gustavo, Giuliano barriers resolve the problem.

Helena Rodrigues.