Hi Carlos,

First off: thanks for the feedback, it is very useful to discuss various real 
world scenarios.

> On 28 Apr 2016, at 02:33, Carlos Lombardi <[email protected]> wrote:
> 
> Hi,
> 
> I want to use STON to enable the backup a graph of objects pertaining to an 
> application.

Would it be possible to provide more details (maybe off list), I would like to 
have a better idea of the issue you are trying to solve.

What does that mean 'not supported by STON', why is that so ?
Why do you need to write extra information ? what is in it ? how do you deal 
with it afterwards ?

> Some of the objects in this graph include, in their state, a Map that is not 
> supported by STON.
> To write these objects, I would like to write via the standard procedure all 
> the instance variables unless that pointing to such a map, and then add a 
> modified version of that map by hand. 
> 
> To exclude the conflicting variable is easy, I just redefine 
> #stonAllInstVarNames at the class side as follows:
>      ^self instVarNames reject: [ :theName | (theName = #references) ]
> in this way, I do not need to modify the method if the state of these objects 
> changes.

Yes, that is one extension point.

> On the other hand, I found not easy to use the standard mechanism for the 
> included instance variables. This mechanism is embedded in 
> STONWriter>>writeObject: . I found no option but repeat the logic codified 
> there, in the method #stonOn: of my class.
> 
> The easiest way to let STON integrate the standard mechanism to store the 
> state of an object, with by-hand additions, is to slightly modify 
> STONWriter>>writeObject:, adding just one line. Objects including instance 
> variables would be written as follows:
> 
>       self writeObject: anObject streamMap: [ :dictionary | 
>               instanceVariableNames do: [ :each | 
>                       (anObject instVarNamed: each)
>                               ifNotNil: [ :value | 
>                                       dictionary at: each asSymbol put: value 
> ]
>                               ifNil: [ 
>                                       anObject stonShouldWriteNilInstVars 
>                                               ifTrue: [ dictionary at: each 
> asSymbol put: nil ] ] ] .
>               anObject stonAdditionalInfoOn: dictionary . ]
> 
> the added line is the last one, the invocation to #stonAdditionalInfoOn:. Of 
> course, the method #stonAdditionalInfoOn: should be added to the Object class.
> By redefining this method in my class, I was able to easily add the modified, 
> STON-compatlble value, to the values handled by the STON code.

Overwriting #stonOn: is meant to be an extension point as well.

> I did not need to modify the STON code to read the STON String to generate a 
> new object.

How can that work ? Via additional accessors ?

Like I said, a concrete example, maybe of the STON output would be helpful.

> Do you think that doing this modification to STON could be a good idea?

Customising STON is OK, it should indeed be possible without changing the core 
code. But I would like to better understand your use case before really 
commenting.

Regards,

Sven

> Cordially - Carlos
> 
> 
> PS: I envisaged as an alternative, to allow the following method in my class
> 
>     stonOn:stonWriter
>         stonWriter writeObjectAsMap: self
>               do: [:mapWriter | 
>                    // added code goes here
>                    stonWriter writeInstVarsOf: self on: mapWriter
>                    // added code goes here
>               ]
> 
> To this end, the methods #writeObjectAsMap:do: and #writeInstVarsOf:on should 
> be added to STONWriter. The method 
>      writeInstVar: aSymbol  of: anObject  on: mapWriter
> could be offered as well, to allow more fine-grained redefinitions of stonOn: 
> , while keeping the ability to use the standard mechanism to store an 
> instance variable.
> 
> I think that the option I described earlier is simpler, I include this one 
> because maybe other people would prefere it.
> 
> 


Reply via email to