[Vm-dev] becomeForward: behavior

Florin Mateoc florin.mateoc at gmail.com
Tue Jan 25 20:21:22 UTC 2022


I agree, your wording is better. I think both #becomeForward: and the bulk
become methods should include the comment (slightly differently phrased),
as people looking at the #becomeForward: method won't necessarily read the
comment of the underlying bulk become method
Both should mention that the indexes where the target(s) are now stored are
incorrect.

I also think that both primitive 72 and primitive 248 should have a comment
to that effect. In both cases we start with two different objects and we
come out with only one. Regardless of the choice of which identityHash to
use, one of the original objects loses (if it was previously included in a
hashed collection), so there is potentially extra work to be done

On Tue, Jan 25, 2022 at 3:07 PM Eliot Miranda <eliot.miranda at gmail.com>
wrote:

>
>
>
> On Jan 25, 2022, at 11:57 AM, Florin Mateoc <florin.mateoc at gmail.com>
> wrote:
>
> 
> I think what the comment should say is "Note that the identityHash of the
> receiver also changes to match its new identity, so if the receiver was
> indexed in any hashed structures, those indexes are now incorrect, so the
> structures should be rehashed"
>
>
> I think that wording is misleading.  What happens is that all references
> to the receiver(s) are replaced by references to the target(s) (this is a
> bulk become).  So the receiver, is fact, obliterated.  So hashed
> collections that used to contain the receiver(s) now contain the target(s)
> and must be rehashed.
>
>
> The comment for primitive 72 should say "Note that the identityHash of the
> argument is changed to match the identityHash that previously belonged to
> the receiver, so if the argument was indexed in any hashed structures,
> those indexes are now incorrect, so the structures should be rehashed"
>
>
> On Tue, Jan 25, 2022 at 2:15 PM Florin Mateoc <florin.mateoc at gmail.com>
> wrote:
>
>> Hi Vanessa and Marcel,
>>
>> Thank you for clarifying.
>> Indeed, my notion of a recent image was a "current release" 5.3 image
>> from a year and a half ago, which does not have the change, so I was indeed
>> referring to primitive 72.
>>
>> Sorry for the confusion,
>>
>> Florin
>>
>> On Tue, Jan 25, 2022 at 1:29 PM Vanessa Freudenberg <vanessa at codefrau.net>
>> wrote:
>>
>>>
>>> @Florin,
>>>
>>> I think nobody has mentioned yet that the behavior of
>>> Object>>becomeForward: in Squeak changed "recently".
>>>
>>> What you were describing is the behavior of primitive 72, which indeed
>>> modified the argument's identity hash.
>>>
>>> More recent Squeak releases use primitive 248 instead, which does not
>>> modify the target's hash. We also introduced primitive 249 which takes a
>>> boolean argument to indicate hash copying.
>>>
>>> @Marcel: I think the comments might be more helpful if they explicitly
>>> mentioned which collections might have to be rehashed, and which (if any)
>>> object is being modified.
>>>
>>> Vanessa
>>>
>>> On Tue, Jan 25, 2022 at 03:44 Marcel Taeumel <marcel.taeumel at hpi.de>
>>> wrote:
>>>
>>>>
>>>> Hi Florin --
>>>>
>>>> > it changes the identityHash of the argument, the non (obvious)
>>>> target.
>>>>
>>>> I updated the comment in #becomeForward: ... hope it helps:
>>>>
>>>> "Primitive. All variables in the entire system that used to point to
>>>> the receiver now point to the argument. Fails if either argument is an
>>>> immediate such as a SmallInteger, or if the receiver is read-only.
>>>> NOTE THAT the identityHash of the receiver IS NOT copied to the
>>>> argument so that the argument should still be properly indexed in any
>>>> existing hashed structures after the mutation. See
>>>> #becomeForward:copyHash:."
>>>>
>>>> Best,
>>>> Marcel
>>>>
>>>> Am 24.01.2022 15:11:09 schrieb Florin Mateoc <florin.mateoc at gmail.com>:
>>>> Hi Clément and Marcel,
>>>>
>>>> You are right, other use cases are supported, and there is probably no
>>>> need
>>>> for a separate setIdentityHash: primitive.
>>>>
>>>> I guess I would not have minded if the method currently called
>>>> #becomForward: would have been called #becomeForwardCopyHash: (plus a
>>>> big,
>>>> fat comment). As it is, it is not strictly a one way become method, it
>>>> is a
>>>> combination method instead, and its comment does not reflect that.
>>>>
>>>> Best,
>>>> Florin
>>>>
>>>>
>>>> On Mon, Jan 24, 2022 at 3:00 AM Marcel Taeumel
>>>> wrote:
>>>>
>>>> >
>>>> > Hi Florin, hi Clément --
>>>> >
>>>> > yes, that's what Squeak's #becomeForward:copyHash: is for.
>>>> >
>>>> > See this discussion, where the interference between #becomeForward:,
>>>> copy
>>>> > hash, and ModificationForbidden is explained:
>>>> >
>>>> >
>>>> http://lists.squeakfoundation.org/pipermail/squeak-dev/2020-April/208596.html
>>>> >
>>>> > Best,
>>>> > Marcel
>>>> >
>>>> > Am 24.01.2022 08:04:38 schrieb Clément Béra :
>>>> > Hi Florin,
>>>> >
>>>> > I believe there are 2 primitives for 2 different use-cases:
>>>> > - primitiveArrayBecomeOneWayNoCopyHash 248
>>>> > - primitiveArrayBecomeOneWayCopyHash 249
>>>> >
>>>> > The difference between both is which hash is preserved. I think for
>>>> your
>>>> > use-case you should use the other primitive.
>>>> >
>>>> > On Mon, Jan 24, 2022 at 5:04 AM Florin Mateoc
>>>> > wrote:
>>>> >
>>>> > >
>>>> > > Hi,
>>>> > >
>>>> > > I am a bit surprised by the #becomeForward: behavior in Squeak.
>>>> This is
>>>> > a
>>>> > > one way become, where the target of the operation is the receiver,
>>>> which
>>>> > > sheds its identity/existence. Nobody points to it after the
>>>> primitive
>>>> > > execution, so it is discarded. This understanding also conforms to
>>>> the
>>>> > > method comment.
>>>> > > As such, I remember a pattern of usage in VisualAge Smalltalk,
>>>> where one
>>>> > > way become was used as a cheap cleanup/avoidance of memory leaks,
>>>> by
>>>> > doing
>>>> > > oneWayBecome: nil. It's not that I advocate for it, but this works
>>>> in
>>>> > > Squeak too, except in Squeak #becomeForward: does an additional
>>>> thing to
>>>> > > the pointers redirection, it changes the identityHash of the
>>>> argument,
>>>> > the
>>>> > > non (obvious) target. While I understand this may be useful in
>>>> certain
>>>> > > situations, I think it is a dangerous conflation of activities. A
>>>> new
>>>> > > primitive that sets the identity hash could be used (VA has it)
>>>> > explicitly
>>>> > > instead when such behavior is desired.
>>>> > > As it is, if I do "Object new becomeForward: nil", it succeeds and
>>>> it
>>>> > > changes nil's identityHash.
>>>> > >
>>>> > > Sorry if this has been debated before,
>>>> > >
>>>> > > Cheers,
>>>> > > Florin
>>>> > >
>>>> >
>>>> >
>>>> > --
>>>> > Clément Béra
>>>> > https://clementbera.github.io/
>>>> > https://clementbera.wordpress.com/
>>>> > Hi Florin,
>>>> >
>>>> > I believe there are 2 primitives for 2 different use-cases:
>>>> > - primitiveArrayBecomeOneWayNoCopyHash 248
>>>> > - primitiveArrayBecomeOneWayCopyHash 249
>>>> >
>>>> > The difference between both is which hash is preserved. I think for
>>>> your
>>>> > use-case you should use the other primitive.
>>>> >
>>>> > On Mon, Jan 24, 2022 at 5:04 AM Florin Mateoc
>>>> > wrote:
>>>> >
>>>> >>
>>>> >> Hi,
>>>> >>
>>>> >> I am a bit surprised by the #becomeForward: behavior in Squeak. This
>>>> is a
>>>> >> one way become, where the target of the operation is the receiver,
>>>> which
>>>> >> sheds its identity/existence. Nobody points to it after the
>>>> primitive
>>>> >> execution, so it is discarded. This understanding also conforms to
>>>> the
>>>> >> method comment.
>>>> >> As such, I remember a pattern of usage in VisualAge Smalltalk, where
>>>> one
>>>> >> way become was used as a cheap cleanup/avoidance of memory leaks, by
>>>> doing
>>>> >> oneWayBecome: nil. It's not that I advocate for it, but this works
>>>> in
>>>> >> Squeak too, except in Squeak #becomeForward: does an additional
>>>> thing to
>>>> >> the pointers redirection, it changes the identityHash of the
>>>> argument, the
>>>> >> non (obvious) target. While I understand this may be useful in
>>>> certain
>>>> >> situations, I think it is a dangerous conflation of activities. A
>>>> new
>>>> >> primitive that sets the identity hash could be used (VA has it)
>>>> explicitly
>>>> >> instead
>>>> >>
>>>> >> when such behavior is desired.
>>>> >> As it is, if I do "Object new becomeForward: nil", it succeeds and
>>>> it
>>>> >> changes nil's identityHash.
>>>> >>
>>>> >> Sorry if this has been debated before,
>>>> >>
>>>> >> Cheers,
>>>> >> Florin
>>>> >>
>>>> >>
>>>> >
>>>> > --
>>>> > Clément Béra
>>>> > https://clementbera.github.io/
>>>> > https://clementbera.wordpress.com/
>>>> >
>>>> >
>>>> Hi
>>>> Clément and Marcel,
>>>>
>>>> You are right, other use cases are supported, and there is probably no
>>>> need for a separate setIdentityHash: primitive.
>>>>
>>>> I guess I would not have minded if the method currently called
>>>> #becomForward: would have been called #becomeForwardCopyHash: (plus a big,
>>>> fat comment). As it is, it is not strictly a one way become method, it is a
>>>> combination method instead, and its comment does not reflect that.
>>>>
>>>> Best,
>>>> Florin
>>>>
>>>>
>>>> On Mon, Jan 24, 2022 at 3:00 AM Marcel Taeumel <marcel.taeumel at hpi.de>
>>>> wrote:
>>>>
>>>>>
>>>>>
>>>>> Hi Florin, hi Clément --
>>>>>
>>>>> yes, that's what Squeak's #becomeForward:copyHash: is for.
>>>>>
>>>>> See this discussion, where the interference between #becomeForward:,
>>>>> copy hash, and ModificationForbidden is explained:
>>>>>
>>>>> http://lists.squeakfoundation.org/pipermail/squeak-dev/2020-April/208596.html
>>>>>
>>>>> Best,
>>>>> Marcel
>>>>>
>>>>>
>>>>> Am 24.01.2022 08:04:38 schrieb Clément Béra <bera.clement at gmail.com>:
>>>>> Hi Florin,
>>>>>
>>>>>
>>>>>
>>>>> I believe there are 2 primitives for 2 different use-cases:
>>>>>
>>>>> - primitiveArrayBecomeOneWayNoCopyHash 248
>>>>>
>>>>> - primitiveArrayBecomeOneWayCopyHash 249
>>>>>
>>>>>
>>>>>
>>>>> The difference between both is which hash is preserved. I think for
>>>>> your
>>>>>
>>>>> use-case you should use the other primitive.
>>>>>
>>>>>
>>>>>
>>>>> On Mon, Jan 24, 2022 at 5:04 AM Florin Mateoc
>>>>>
>>>>> wrote:
>>>>>
>>>>>
>>>>>
>>>>> >
>>>>>
>>>>> > Hi,
>>>>>
>>>>> >
>>>>>
>>>>> > I am a bit surprised by the #becomeForward: behavior in Squeak. This
>>>>> is a
>>>>>
>>>>> > one way become, where the target of the operation is the receiver,
>>>>> which
>>>>>
>>>>> > sheds its identity/existence. Nobody points to it after the
>>>>> primitive
>>>>>
>>>>> > execution, so it is discarded. This understanding also conforms to
>>>>> the
>>>>>
>>>>> > method comment.
>>>>>
>>>>> > As such, I remember a pattern of usage in VisualAge Smalltalk, where
>>>>> one
>>>>>
>>>>> > way become was used as a cheap cleanup/avoidance of memory leaks, by
>>>>> doing
>>>>>
>>>>> > oneWayBecome: nil. It's not that I advocate for it, but this works
>>>>> in
>>>>>
>>>>> > Squeak too, except in Squeak #becomeForward: does an additional
>>>>> thing to
>>>>>
>>>>> > the pointers redirection, it changes the identityHash of the
>>>>> argument, the
>>>>>
>>>>> > non (obvious) target. While I understand this may be useful in
>>>>> certain
>>>>>
>>>>> > situations, I think it is a dangerous conflation of activities. A
>>>>> new
>>>>>
>>>>> > primitive that sets the identity hash could be used (VA has it)
>>>>> explicitly
>>>>>
>>>>> > instead when such behavior is desired.
>>>>>
>>>>> > As it is, if I do "Object new becomeForward: nil", it succeeds and
>>>>> it
>>>>>
>>>>> > changes nil's identityHash.
>>>>>
>>>>> >
>>>>>
>>>>> > Sorry if this has been debated before,
>>>>>
>>>>> >
>>>>>
>>>>> > Cheers,
>>>>>
>>>>> > Florin
>>>>>
>>>>> >
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> --
>>>>>
>>>>> Clément Béra
>>>>>
>>>>> https://clementbera.github.io/
>>>>>
>>>>> https://clementbera.wordpress.com/
>>>>>
>>>>> Hi Florin,
>>>>>
>>>>> I believe there are 2 primitives for 2 different use-cases:
>>>>> - primitiveArrayBecomeOneWayNoCopyHash 248
>>>>> - primitiveArrayBecomeOneWayCopyHash 249
>>>>>
>>>>> The difference between both is which hash is preserved. I think for
>>>>> your use-case you should use the other primitive.
>>>>>
>>>>> On Mon, Jan 24, 2022 at 5:04 AM Florin Mateoc <florin.mateoc at gmail.com>
>>>>> wrote:
>>>>>
>>>>>>
>>>>>> Hi,
>>>>>>
>>>>>> I am a bit surprised by the #becomeForward: behavior in Squeak. This
>>>>>> is a one way become, where the target of the operation is the receiver,
>>>>>> which sheds its identity/existence. Nobody points to it after the primitive
>>>>>> execution, so it is discarded. This understanding also conforms to the
>>>>>> method comment.
>>>>>> As such, I remember a pattern of usage in VisualAge Smalltalk, where
>>>>>> one way become was used as a cheap cleanup/avoidance of memory leaks, by
>>>>>> doing oneWayBecome: nil. It's not that I advocate for it, but this works in
>>>>>> Squeak too, except in Squeak #becomeForward: does an additional thing to
>>>>>> the pointers redirection, it changes the identityHash of the argument, the
>>>>>> non (obvious) target. While I understand this may be useful in certain
>>>>>> situations, I think it is a dangerous conflation of activities. A new
>>>>>> primitive that sets the identity hash could be used (VA has it) explicitly
>>>>>>
>>>>>> instead
>>>>>>
>>>>>>
>>>>>>
>>>>>> when such behavior is desired.
>>>>>> As it is, if I do "Object new becomeForward: nil", it succeeds and it
>>>>>> changes nil's identityHash.
>>>>>>
>>>>>> Sorry if this has been debated before,
>>>>>>
>>>>>> Cheers,
>>>>>> Florin
>>>>>>
>>>>>>
>>>>>>
>>>>>
>>>>> --
>>>>> Clément Béra
>>>>> https://clementbera.github.io/
>>>>> https://clementbera.wordpress.com/
>>>>>
>>>>>
>>>>>
>>>>>
>>>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.squeakfoundation.org/pipermail/vm-dev/attachments/20220125/a3a58154/attachment-0001.html>


More information about the Vm-dev mailing list