Question about versioning of serialization
Hello, we are currently implementing serialization versioning using boost and I am curious to know if their is a way to support a changing class hierarchy. Let's say C derives from A for a while and then a change requires C to derive from B instead, which derives from A. Is it possible to support such a change provided we know what default values are needed for the new class? I tried and even if don't serialize anything in class C I get stream errors because I believe boost saves some base info for each class. When it tries to reload those value from a old stream, obviously it desynchronizes it because the values are not in the stream. So is their a way around this? Also is it possible to support the removal of a class. If, for exemple C derives from B which in turn derives from A and suddenly B is removed. Is their a way to deal with this without invalidating all previously saved data? Thank you for your help! -- Frédérick Martel-Lupien Étudiant en Génie Informatique Université de Sherbrooke fred.lupien@gmail.com
I've never thought about this, but my first attempt would be something like the following:
template<class Archive>
void load(Archiv &ar , C &c, const unsigned int version){
if(version < 4>)
ar >> boost::serialization::base_object<A>(c);
else
ar >> boost::serialization::base_object<B>(c);
}
"Fred Lupien"
Yea I had thought about this but haven't tried it yet. I am working on this
tonight. I have another question that could help me. Is their a
documentation available on exactly what is serialized in the archive, the
class format and related stuff seem to be serialized as a list of numbers
(at least in text archive it appears so, forgive me if I am wrong). What is
serialized exactly for each class (beside what the user specifies of
course!)?
If found that on a forum but I am not sure how relevant it is (from
basic_archive.cpp) :
//////////////////////////////////////////////////////////////////////
//
// class_information is stored as
//
// class_id* // -1 for a null pointer
// if a new class id
// [
// exported key - class name*
// tracking level - always/never
// file version
// ]
//
// if tracking
// [
// object_id
// ]
//
// [ // if a new object id
// data...
// ]
//
// * required only for pointers - optional for objects
Thank you!
On 10/19/07, Robert Ramey
I've never thought about this, but my first attempt would be something like the following:
template<class Archive> void load(Archiv &ar , C &c, const unsigned int version){ if(version < 4>) ar >> boost::serialization::base_object<A>(c); else ar >> boost::serialization::base_object<B>(c); }
"Fred Lupien"
wrote in message news:32e6d6bf0710190708x45191057ld2a23a92399f59e4@mail.gmail.com... Hello, we are currently implementing serialization versioning using boost and I am curious to know if their is a way to support a changing class hierarchy. Let's say C derives from A for a while and then a change requires C to derive from B instead, which derives from A. Is it possible to support such a change provided we know what default values are needed for the new class? I tried and even if don't serialize anything in class C I get stream errors because I believe boost saves some base info for each class. When it tries to reload those value from a old stream, obviously it desynchronizes it because the values are not in the stream. So is their a way around this?
Also is it possible to support the removal of a class. If, for exemple C derives from B which in turn derives from A and suddenly B is removed. Is their a way to deal with this without invalidating all previously saved data?
*** I doubt it. If you want to load old archives, you have know how much to "skip" over. The only thing that MIGHT be possible is something like:
template<class Archive> void load(Archiv &ar , C &c, const unsigned int version){ if(version < 4>){ class a; //no longer used ar >> a; // throw away old data on leaving scope } else ar >> boost::serialization::base_object<B>(c); }
Thank you for your help!
-- Frédérick Martel-Lupien Étudiant en Génie Informatique Université de Sherbrooke fred.lupien@gmail.com
------------------------------
_______________________________________________ Boost-users mailing list Boost-users@lists.boost.org http://lists.boost.org/mailman/listinfo.cgi/boost-users
_______________________________________________ Boost-users mailing list Boost-users@lists.boost.org http://lists.boost.org/mailman/listinfo.cgi/boost-users
-- Frédérick Martel-Lupien Étudiant en Génie Informatique Université de Sherbrooke fred.lupien@gmail.com
look in
basic_archive.cpp
Robert Ramey
"Fred Lupien"
participants (2)
-
Fred Lupien
-
Robert Ramey