[Vm-dev] Weak finalization enh. are still not integrated into VM(s)

[David T. Lewis]

On Tue, Sep 21, 2010 at 03:17:15AM +0300, Igor Stasenko wrote:
>  
> On 21 September 2010 02:08, Eliot Miranda <eliot.miranda at gmail.com> wrote:
> >
> >
> >
> > On Mon, Sep 20, 2010 at 4:02 PM, Igor Stasenko <siguctua at gmail.com> wrote:
> >>
> >> Hello guys,
> >>
> >> i'd like to remind you that this stuff is not yet made own way to VM and images.
> >>
> >> Magma suffers badly from old finalization implementation, and as a
> >> consequence it losing in speed competition
> >> comparing to other OODBs.
> >> I'd really like to see this enhancement to be integrated to both
> >> Squeak VM and Cog VM.
> >>
> >> My quick analyzis shows that Squeak 4.2 image contains some things,
> >> which needed:
> >>
> >> ?? ?? ?? ??"Weak reference finalization"
> >> ?? ?? ?? ??newArray at: 56 put: (self at: #WeakFinalizer ifAbsent: []).
> >>
> >> while Pharo still not.
> >>
> >> I'd like to remind you that VM-side changes are really simple and
> >> backwards compatible (given the right size of special objects array).
> >> Once we will have new VMs, supporting this feature , we could start
> >> integrating the image-side part of it.
> >>
> >> Please, inform me, what i can do to help closing this issue.
> >
> > you can either send me a change set that is w.r.t. to a Cog version or save a Monticello version that is a delta from Cog. ??I also need image level test code to pound on this.
> 
> 
> The issue #7473 are dedicated for both vm-side changes and
> image-side changes.

The VM side changes should be in place as of VMMaker-dtl.175, version
4.2.2 (sorry I forgot to update the Mantis issue when I did this). That
means that the changes are in the VMMaker package, but are not yet in
published VM releases (e.g. the Unix VM Squeak-4.0.3.2202 was built
from VMM version 4.0.3, prior to 4.2.2).

Eliot - the ObjectMemory>>initializeSpecialObjectIndices in VMMaker-dtl.175
ends as follows, which is intended to be consistent with Cog and Alien
requirements, and should match the #recreateSpecialObjectsArray that
Igor provided.

	SelectorAboutToReturn := 48.
	SelectorRunWithIn := 49.
	SelectorAttemptToAssign := 50.

	"PrimErrTableIndex := 51. in Interpreter class>>initializePrimitiveErrorCodes"

	ClassAlien := 52.
	InvokeCallbackSelector := 53.
	ClassUnsafeAlien := 54.

	ClassWeakFinalizer := 55

Dave

> 
> The image-side scripts are:
> 
> - Weak-finalizers-classes.2.cs
> - new-finalization-auto.2.cs
> 
> install them in the above order, then run
> 
> WeakArray restartFinalizationProcess
> 
> I added the test case for new kind of registry, and all tests which
> are green for old registry
> also green for a new one.
> The new registry are automatically detects if VM supports new
> finalization or not,
> and switching to suitable finalization logic  (it takes a single GC
> in order to detect it).
> 
> 
> The last thing is #recreateSpecialObjectsArray  [2].
> A new entry is already added in Squeak, thanks Levente,
> except that naming is wrong, and should refer to #WeakFinalizationList,
> instead of  #WeakFinalizer , i.e.
> 
> 	"Weak reference finalization"
> 	newArray at: 56 put: (self at: #WeakFinalizationList ifAbsent: []).
> 
> 
> The VM-side changes is in file
>  weak-finalizers-VM.1.cs ,
>  but they are for Squeak VM, and a bit outdated, since index of
> special object is wrong
> (obviously, should be in sync with one in spl objs array).
> 
> There are two methods, which don't require changes:
> 
> ObjectMemory>>finalizeReference:
> and
> ObjectMemory>>weakFinalizerCheck:
> 
> And on a class side, the #intitializeSpecialObjectIndices, should use
> a correct index for
> WeakFinalizer class var, i.e.
> 
> WeakFinalizer := 55.
> 
> 
> For Cog VM, i made everything fixed (see changesets in attachments).
> I was able to build new Cog VM and it works fine, reporting:
> 
> WeakFinalizationRegistry hasNewFinalization  ==> true
> 
> after starting an image under this VM.
> 
> 1. http://bugs.squeak.org/view.php?id=7473
> 2. http://bugs.squeak.org/view.php?id=7525
> 
> > best,
> > Eliot
> 
> -- 
> Best regards,
> Igor Stasenko AKA sig.