[Vm-dev] pinned object behaviour during Image save/load

[Ben Coman]

On Sat, Feb 18, 2017 at 11:46 PM, GitHub <noreply at github.com> wrote:
>
>   Branch: refs/heads/Cog
>   Home:   https://github.com/OpenSmalltalk/opensmalltalk-vm
>   Commit: 0076679f0aa45970e6774f69d84f7e81d548bc40
>       https://github.com/OpenSmalltalk/opensmalltalk-vm/commit/0076679f0aa45970e6774f69d84f7e81d548bc40
>   Author: Eliot Miranda <eliot.miranda at gmail.com>
>   Date:   2017-02-18 (Sat, 18 Feb 2017)
>
>   Log Message:
>   -----------
>   CogVM source as per VMMaker.oscog-eem.2134
>
> Move the SpurPlanningCompactor to tip.  It is now ready for prime time.
> SpurPlanningCompactor is a three-finger algorithm that slides objects down to
> the first finger, which points at the first free object, from the first
> unpinned, marked object, while stepping around intervening pinned objects
> pointed at by the third finger.

Great to see this progress. Thanks Eliot.

Just curious about the strategy of handling pinned objects during
Image save/load.
Could the pinned memory possibly be considered to have been invalidated
much like when the VM exits? And thus the VM could move and compress
the pinned objects at some point, or automatically unpin all objects
in a saved Image?

Or are their cases where the pin is useful to persist across saves?
Maybe if an external library relies on one pinned object refering
 to the memory location of another pinned in some domain document structure?
Thus it is Image's responsibility to unpin all objects to minimise Image size?
I guess unpinning by the Image would be best at Image load rather when saving
since the Image may continue.

cheers -ben