Hi Bert,
[please see the "however" later on]
On Tue, Jul 18, 2017 at 7:30 AM, Bert Freudenberg bert@freudenbergs.de wrote:
On Tue, Jul 18, 2017 at 4:09 PM, Levente Uzonyi leves@caesar.elte.hu wrote:
On Tue, 18 Jul 2017, marcel.taeumel wrote:
Hi, there.
Why are there guards to prevent forward-becoming to immediates (chars, ints) but not to symbols?
The former is technically impossible due to different object representations, the latter is possible and not restricted at all. For example, true and false are not immediate objects, you can use #become*: on them to blow your image up. So, there's no restriction at all if it's technically possible to use #become*:. The responsibility model is the simplest here: use at your own risk. Since this comes up every once in a while, I suggest a comment be added to those methods stating the responsibility model explicitly.
+1
*Especially* a warning that becomeForward: does modify the target's hash.
I don't know how the VM handles immutability in this case, but it's possible that it wouldn't let #become*: affect immutable objects.
I think that would be fine, you should be able to become an immutable object and vice versa.
On the other hand, I'm sure it would let you change fields of immutable objects via #become*:, but that's not an issue in your case.
This is debatable ... I would rather have the VM raise an error when trying to become a field of an immutable object. Immutable should mean immutable, no?
Agreed. I'm modifying the Spur VM to fail becomeForward: when copyHash is true and the target object is immutable. So when we have immutable literals, #normal would be immutable and the become would fail. The operative code was
containsOnlyValidBecomeObjects: array1 and: array2 twoWay: isTwoWay copyHash: copyHash ... [fieldOffset >= self baseHeaderSize] whileTrue: [... isTwoWay ifTrue: [...] ifFalse:
[(copyHash and: [self isImmediate: oop2]) ifTrue: [^PrimErrInappropriate]].
fieldOffset := fieldOffset - self bytesPerOop]. ... ^0
and is now
containsOnlyValidBecomeObjects: array1 and: array2 twoWay: isTwoWay copyHash: copyHash ... [fieldOffset >= self baseHeaderSize] whileTrue: [... isTwoWay ifTrue: [...] ifFalse:
[(copyHash and: [(self isImmediate: oop2) or: [self isImmutable: oop2]])
ifTrue:
[^PrimErrInappropriate]].
fieldOffset := fieldOffset - self bytesPerOop]. ... ^0
This was a slip on mine and Clément's part when we added read-only object support.
However, it seems to me that becomerForward: doing a copyHash is itself a bug. What's the rationale for copying the receiver's hash to the target object? I've seen various frameworks use become to clean-up objects by doing aNoLongerWantedObject becomeFor4ward: nil, and with our current default implementation for becomeForward: that will cause havoc.
- Bert -
Levente
This will break your image: "Object new becomeForward: #normal"
... because Cursor class MNU #normal ... :-) Either this is a bug or forward-becoming to symbols is not intended at all.
Best, Marcel
_,,,^..^,,,_ best, Eliot
The former is technically impossible due to different object representations, the latter is possible and not restricted at all. For example, true and false are not immediate objects, you can use #become*: on them to blow your image up. So, there's no restriction at all if it's technically possible to use #become*:. The responsibility model is the simplest here: use at your own risk. Since this comes up every once in a while, I suggest a comment be added to those methods stating the responsibility model explicitly.
+1
Why express +1 for the responsibility model for use of become: (which I agree with), but not for mutating objects?
*Especially* a warning that becomeForward: does modify the target's hash.
I assume this is not referring to a Warning signal here, but just something in the comment. But even that wouldn't be necessary as long as the indirection remains, since the "copyHash: false" code makes that very clear.
I don't know how the VM handles immutability in this case, but it's possible that it wouldn't let #become*: affect immutable objects.
I think that would be fine, you should be able to become an immutable object and vice versa.
On the other hand, I'm sure it would let you change fields of immutable objects via #become*:, but that's not an issue in your case.
This is debatable ... I would rather have the VM raise an error when trying to become a field of an immutable object. Immutable should mean immutable, no?
So for bulkBecome:, will the error mean it becomed some of them (the ones that that could), but not all? It's getting complex!
However, it seems to me that becomerForward: doing a copyHash is itself a bug. What's the rationale for copying the receiver's hash to the target object?
For when you become a Proxy object to a Symbol selector to perform, which are keys in MethodDictionary's. It's absolutely essential.
Best, Chris
Hi Chris,
On Thu, Jul 20, 2017 at 2:26 PM, Chris Muller asqueaker@gmail.com wrote:
The former is technically impossible due to different object representations, the latter is possible and not restricted at all. For example, true and false are not immediate objects, you can use
#become*: on
them to blow your image up. So, there's no restriction at all if it's technically possible to use #become*:. The responsibility model is the simplest here: use at your own risk. Since this comes up every once in a while, I suggest a comment be added to those methods stating the responsibility model explicitly.
+1
Why express +1 for the responsibility model for use of become: (which I agree with), but not for mutating objects?
*Especially* a warning that becomeForward: does modify the target's
hash.
I assume this is not referring to a Warning signal here, but just something in the comment. But even that wouldn't be necessary as long as the indirection remains, since the "copyHash: false" code makes that very clear.
I don't know how the VM handles immutability in this case, but it's possible that it wouldn't let #become*: affect immutable objects.
I think that would be fine, you should be able to become an immutable object and vice versa.
On the other hand, I'm sure it would let you change fields of immutable objects via #become*:, but that's not an issue in your case.
This is debatable ... I would rather have the VM raise an error when trying to become a field of an immutable object. Immutable should mean immutable, no?
So for bulkBecome:, will the error mean it becomed some of them (the ones that that could), but not all? It's getting complex!
None of them. The semantics of primitives should be they either succeed and answer their result, or they fail without side-effects. Alas the ImageSegment loading primitive is an exception, but I believe it's the only one. So in the become primitives, validation is done before any becomes, including computing an estimate of any memory needed, and the entire bulk become will fail if any single object fails validation and/or if there is not enough memory to complete the operation (two-way become may involve creating copies of objects).
However, it seems to me that becomerForward: doing a copyHash is itself a bug. What's the rationale for copying the receiver's hash to the target object?
For when you become a Proxy object to a Symbol selector to perform, which are keys in MethodDictionary's. It's absolutely essential.
This doesn't make sense to me. Are you saying one has a proxy for a Symbol that is not present, this gets entered as a key into MethodDictionaries, and later is becalmed into the Symbol it represents? If not, can you take me through the use-case?
It seems to me that what one wants is the ability to determine an Object's identityHash. I've long thought that Symbols should set their identityHash on creation so that Symbols identityHashes are constant across all systems. This would men for example, that binary code loaders would not need to rehash method dictionaries because they would already be correctly hashed. In these cases one definitely doesn't want to overwrite the target's identityHash; instead, taking your example, one would want to create the proxy with the identityHash of its target.
Best, Chris
This is debatable ... I would rather have the VM raise an error when trying to become a field of an immutable object. Immutable should mean immutable, no?
So for bulkBecome:, will the error mean it becomed some of them (the ones that that could), but not all? It's getting complex!
None of them. The semantics of primitives should be they either succeed and answer their result, or they fail without side-effects. Alas the ImageSegment loading primitive is an exception, but I believe it's the only one. So in the become primitives, validation is done before any becomes, including computing an estimate of any memory needed, and the entire bulk become will fail if any single object fails validation and/or if there is not enough memory to complete the operation (two-way become may involve creating copies of objects).
Okay, that sounds like a good principle.
However, it seems to me that becomerForward: doing a copyHash is itself a bug. What's the rationale for copying the receiver's hash to the target object?
For when you become a Proxy object to a Symbol selector to perform, which are keys in MethodDictionary's. It's absolutely essential.
This doesn't make sense to me. Are you saying one has a proxy for a Symbol that is not present, this gets entered as a key into MethodDictionaries, and later is becalmed into the Symbol it represents? If not, can you take me through the use-case?
It seems to me that what one wants is the ability to determine an Object's identityHash. I've long thought that Symbols should set their identityHash on creation so that Symbols identityHashes are constant across all systems. This would men for example, that binary code loaders would not need to rehash method dictionaries because they would already be correctly hashed. In these cases one definitely doesn't want to overwrite the target's identityHash; instead, taking your example, one would want to create the proxy with the identityHash of its target.
Woops, I misread this sentence:
However, it seems to me that becomerForward: doing a copyHash is itself a
I thought you meant NOT copying the hash was a bug. In fact, we agree that copying (changing) the identityHash of the target object is not very safe or useful. Sorry for the confusion.
So, yeah, the scenario is the Magma app materialized an object which references a Proxy. Then that proxy was agitated (sent a message --> DNU), whereupon the Symbol string is read from the DB and the Proxy becomeFoward'ed to it. But it was a Symbol that already existed in MethodDictionary's, so its identityHash shouldn't change just because of that.
Best.
PS -- Here is the first message of the first thread which was the birth of the #becomeForward:copyHash: addition to the VM. :)
http://lists.squeakfoundation.org/pipermail/squeak-dev/2002-May/039735.html
vm-dev@lists.squeakfoundation.org
-
Chris Muller -
Eliot Miranda