Design question: why not reuse the CMS slots in mark word?

[Christine Flood]

I don't know that it breaks serialization, I was just speculating.

The object header format has been baked into Hotspot since approximately 1998, that 
was the first thing I thought of that might break.

I think there are much better ways to improve Shenandoah rather than making a change which
requires a different object format for Shenandoah than for any of the other collectors.  

At the very least it would require more Shenandoah specific changes to C2, which we are trying to minimize.



Christine



----- Original Message -----
> From: "Aleksey Shipilev" <shade at redhat.com>
> To: "Christine Flood" <chf at redhat.com>, "Roman Kennke" <rkennke at redhat.com>
> Cc: shenandoah-dev at openjdk.java.net
> Sent: Tuesday, September 20, 2016 1:23:57 PM
> Subject: Re: Design question: why not reuse the CMS slots in mark word?
> 
> On 09/20/2016 02:42 PM, Christine Flood wrote:
> > Make sure you don't break object serialization.
> 
> I must be missing something. Why do we care about (Java?) object
> serialization here?
> 
> If forwarding pointer is the part of object metadata (i.e. header), it
> does not participate in the serialization at all.
> 
> If forwarding pointer is the part of "fields" block, then it is puzzling
> how runtime would ever touch it, if it enumerates the fields by offsets,
> not reads the entire "bitmap" of fields. (It has to, because field block
> is not contiguous due to per-field alignment requirements that are
> specific to the current running mode).
> 
> Thanks,
> -Aleksey
> 
>