RFR: 8241825: Make compressed oops and compressed class pointers independent on x86_64

[Erik Österlund]

Thanks Coleen. Will add some comments and send out as a follow-up to this.

/Erik

> On 28 Apr 2020, at 16:17, coleen.phillimore at oracle.com wrote:
> 
>> On 4/27/20 11:28 AM, Erik Österlund wrote:
>> Hi Frederic,
>> 
>> I hear your concern about hardcoded offsets making the JVM more fragile.
>> 
>> First of all, let me assure you that you would not be able to run hello world if the offsets ever were to be
>> wrong. They are verified during bootstrapping in JavaClasses::check_offsets(), so you have to be quite skilled
>> to manage to break anything without immediately noticing it. You literally can't have run a single test or java -version
>> without immediately crashing. This makes this significantly less fragile. So I don't think I believe that this is fragile.
>> 
>> Having said that, I do agree it could be annoying and a pain to maintain and update the layout code if such changes
>> are occasionally incompatible with the random selection of hardcoded offsets, forcing new awkward updates of said offsets,
>> and that hardcoding things in general is undesirable, compared to a more solid solution.
>> I had a look at an approach for removing hardcoded offsets completely. It looks something like this:
>> 
>> http://cr.openjdk.java.net/~eosterlund/8241825/webrev.00..not_01/
>> 
>> The change is a bit involved and complicated though. Because of that, I would prefer to push it as a follow-up RFE
>> focused on removing hardcoded offsets, if you are okay with that. This change is already quite tricky, and I think
>> these things ought to be separate.
> 
> This is nice and should be a follow-up RFE.  I sent you some comments to add to it.  It'd be great to not have hc offsets anymore!
> 
> Coleen