[serialization] Stateful serialization
Hello, In a recent post, I asked whether it was possible to swap deserialized pointer values for already-existing ones so I could merge the incoming data with objects already in memory. I guess the answer is no... An alternative is to serialize some other identifier instead of writing pointers, and then on de-serialization I find my existing objects via those identifiers. This works fine, but it leads to another question: How do I make my internal objects available to the deserialization routines so that they have an opportunity to look up the necessary information? Or more generally, How do I do stateful serialization / deserialization? For example, maybe I want to supply options that fine-tune how my data is serialized or deserialized. The only parameter that is passed around to all the serialization functions is the Archive, so if I want to add options or other context information, it seems like I should add this to the Archive. That's simple enough to do; I just made a wrapper Archive that can sit atop any other Archive and contain any other options / context data I need. On the one hand it seems sneaky to hijack the Archive for this purpose, but on the other hand iostreams do basically the same thing via the various ios_base calls (setf, width, precision, etc.) so maybe it's not so bad. Am I perhaps overlooking some facility already present in the serialization library? Thanks, Scott
Scott Howlett wrote:
Hello,
In a recent post, I asked whether it was possible to swap deserialized pointer values for already-existing ones so I could merge the incoming data with objects already in memory. I guess the answer is no...
An alternative is to serialize some other identifier instead of writing pointers, and then on de-serialization I find my existing objects via those identifiers. This works fine, but it leads to another question:
How do I make my internal objects available to the deserialization routines so that they have an opportunity to look up the necessary information? Or more generally, How do I do stateful serialization / deserialization? For example, maybe I want to supply options that fine-tune how my data is serialized or deserialized.
I may be evil for suggesting this: http://uint32t.blogspot.com/2008/03/lame-implementation-of-dynamic-scoping.h... I am only suggesting the idea, not the implementation!!!! -- Sohail Somani http://uint32t.blogspot.com
Make your own archive derived from one of the archives along with your own helper. This technique is used in the current archive to implement the special case of shared pointers - which do not model model the serializablt concept. You can do the same thing with your own "special types" Robert Ramey Scott Howlett wrote:
Hello,
In a recent post, I asked whether it was possible to swap deserialized pointer values for already-existing ones so I could merge the incoming data with objects already in memory. I guess the answer is no...
An alternative is to serialize some other identifier instead of writing pointers, and then on de-serialization I find my existing objects via those identifiers. This works fine, but it leads to another question:
How do I make my internal objects available to the deserialization routines so that they have an opportunity to look up the necessary information? Or more generally, How do I do stateful serialization / deserialization? For example, maybe I want to supply options that fine-tune how my data is serialized or deserialized.
The only parameter that is passed around to all the serialization functions is the Archive, so if I want to add options or other context information, it seems like I should add this to the Archive. That's simple enough to do; I just made a wrapper Archive that can sit atop any other Archive and contain any other options / context data I need.
On the one hand it seems sneaky to hijack the Archive for this purpose, but on the other hand iostreams do basically the same thing via the various ios_base calls (setf, width, precision, etc.) so maybe it's not so bad.
Am I perhaps overlooking some facility already present in the serialization library?
Thanks, Scott
Thanks! That's basically what I have done. My derived Archive is a wrapper that can be put around any underlying Archive. Scott Robert Ramey wrote:
Make your own archive derived from one of the archives along with your own helper. This technique is used in the current archive to implement the special case of shared pointers - which do not model model the serializablt concept. You can do the same thing with your own "special types"
Robert Ramey
Scott Howlett wrote:
Hello,
In a recent post, I asked whether it was possible to swap deserialized pointer values for already-existing ones so I could merge the incoming data with objects already in memory. I guess the answer is no...
An alternative is to serialize some other identifier instead of writing pointers, and then on de-serialization I find my existing objects via those identifiers. This works fine, but it leads to another question:
How do I make my internal objects available to the deserialization routines so that they have an opportunity to look up the necessary information? Or more generally, How do I do stateful serialization / deserialization? For example, maybe I want to supply options that fine-tune how my data is serialized or deserialized.
The only parameter that is passed around to all the serialization functions is the Archive, so if I want to add options or other context information, it seems like I should add this to the Archive. That's simple enough to do; I just made a wrapper Archive that can sit atop any other Archive and contain any other options / context data I need.
On the one hand it seems sneaky to hijack the Archive for this purpose, but on the other hand iostreams do basically the same thing via the various ios_base calls (setf, width, precision, etc.) so maybe it's not so bad.
Am I perhaps overlooking some facility already present in the serialization library?
Thanks, Scott
_______________________________________________ Boost-users mailing list Boost-users@lists.boost.org http://lists.boost.org/mailman/listinfo.cgi/boost-users
participants (3)
-
Robert Ramey -
Scott Howlett -
Sohail Somani