Re: [boost] Boost XML Serialization Stability across library versions

Subject: Re: [boost] Boost XML Serialization Stability across library versions
From: Adlai Shawareb (adlai_at_[hidden])
Date: 2017-01-25 17:07:10

>>On 1/24/17 10:01 PM, Robert Ramey wrote:
>>>On 1/24/17 6:29 PM, Adlai Shawareb wrote:
>>> Thank you. We are really looking at Boost Serialization as a way to manage changes in the data stored in the memory chip. In future versions we may add a float or remove an int, and need to be backward compatible with those changes.
>>>
>>> Is there a feature in Boost that would support managing data in that way?
>>>
>>> The only alternative to serialization we've come up with is to use unions and structs to selectively include or exclude data as the versions change.
>>>
>>> Thanks,
>>> Adlai
>>
>> I think you're confusing two different questions
>>
>> a) The library maintains backward compatibility to your previous data. 
>> So if your data changes, you can still retrieve data from older 
>> archives. This is referred to as "versioning".
>>
>> b) The same system is applied to maintaining backward compatibility to 
>> structs from the standard library, - vector, optional, etc....
>>
>> c) some structures from the standard library don't have a version number 
>> encoded in the data. versioning is maintained by a library version 
>> number encoded in the header record.
>>
>> d) This also maintains backward compatibility when you change your 
>> compiler to another version.
>>
>> e) This is all described in the documenation along with examples.
>>
>> f) This has worked well. In a couple of very rare instances it has 
>> occurred this this versioning between libraries has been messed up for 
>> one or another data type. I regret this. I would like to see an 
>> expanded test suite to address this and hence be able to guarantee that 
>> this problem would never occur. Unfortunately, no one has expressed the 
>> view that this is sufficiently necessary in order to justify the 
>> expenditure required.
>>
>> g) In your case, I would recommend creating some tests of your own that 
>> you could run on every new version of the library so that you would 
>> absolutly that backward compatibility has not been broken.
>>
>> I thought you were referring to abi compatibility - which I don't 
>> believe can be guaranteed. But now I see this is not your question.
>>
>> Robert Ramey
>
> Thanks Robert,
>
> Regarding your suggestion to use text serialization. Since, over time, we will be adding and deleting data from the data set that we serialize, would the following be the right approach to managing the changes in the data that is de/serialized. I'm > > thinking that we would set the BOOST_CLASS_VERSION before de/serializing.
>
> {
>                ar & mInt;
>               If(version >= 2)
>                              ar & mFloat;
>                If(version >= 3)
>                               ar & mInt2;
>                ar & mFloat2;
> }
>
> Regards,
> Adlai