[Pharo-project] How to copy a CompiledMethod?

Stéphane Ducasse stephane.ducasse at inria.fr
Fri Nov 26 22:15:47 CET 2010


veryDeepCopy should be replaced by copy and postCopy.
So could you publish a version of postCopy with a couple of tests?
Because this would be good to make progress from this side.


On Nov 26, 2010, at 6:14 PM, Gabriel Hernán Barbuto wrote:

> Eliot
> Thanks for your response.
> On Fri, Nov 26, 2010 at 5:49 PM, Eliot Miranda <eliot.miranda at gmail.com> wrote:
>> On Thu, Nov 25, 2010 at 1:14 PM, Igor Stasenko <siguctua at gmail.com> wrote:
>>> On 25 November 2010 20:44, Gabriel Hernán Barbuto <gbarbuto at gmail.com>
>>> wrote:
>>>> Hi
>>>> I am trying to make a complete duplicate of a CompiledMethod. I
>>>> started with CompiledMethod>>#copyWithTrailerBytes: but I think that
>>>> this way still leaves some things to copy. It duplicates all the
>>>> literals, but I think that it doesn't duplicate a Pragma if for
>>>> example, the method has one.
>>> AFAIR, pragmas are stored as one of literals.
>>> So, it actually should also make a copy of pragmas.
>> Yes. pragmas are effectively referenced from literals.  If a method needs ti
>> the penultimate literal (which otherwise is the method;s selector) holds an
>> AdditionalMethodState which refers back to its method.  This needs to be
>> copied if one is copying a method, since the backpointer needs to refer to
>> the owning method, not the method form which the copy was made.
>>>> Is there a proper way to completely make deep copy of a
>>>> CompiledMethod? I mean, without writing my own. If there is a way to
>>>> do it, I don't want to duplicate it.
>>> if you tell more, why #copy is not enough for you.
>> If literals are read-only (and Array/ByteArray/String/Float literals are
>> conceptually read-only but not actually) you don't need to make a deep copy.
>>  You only need to copy the AdditionalMethodState, if present, and the class
>> association, if you want to install the method somewhere else.
> Yes that was exactly what I did. I handle both Association and
> AdditionalMethodState in a special way and just send copy to all the
> other literals. I was not sure what I could encounter in the literals
> so I opted to send copy to them. Maybe they will just return
> themselves or make an unnecessary duplicate.
> I have also taken care of copying the Pragmas referenced from any
> AdditionalMethodState. Igor mentioned that there could be Pragmas
> directly on the literals but I am not sure about that. I will have to
> get all the CompiledMethods and check if anyone has a Pragma among its
> literals, but that is not enough, maybe my image doesn't have any.
> I understand that all the literals are conceptually read only, and
> that copying anything besides the Associations, AdditionalMethodStates
> and Pragmas can waste memory and CPU cycles, but I am not interested
> in speed nor memory consumption right now. But I will take a note, in
> case some day I need to improve the copy.
> After copying all the Associations, AdditionalMethodStates and Pragmas
> that I've found I fix the references of both AdditionalMethodStates
> and Pragmas to the new method in order to break all references to
> objects in the current image that I don't want to show up in the new
> image.
> Thanks again.
> Best
> Gabriel
>> best
>> Eliot
>>>> Thanks in advance.
>>>> Gabriel
>>> --
>>> Best regards,
>>> Igor Stasenko AKA sig.

More information about the Pharo-project mailing list