TN002: Persistent Object Data Format


For the latest documentation on Visual Studio 2017 RC, see Visual Studio 2017 RC Documentation.

This note describes the MFC routines that support persistent C++ objects and the format of the object data when it is stored in a file. This applies only to classes with the DECLARE_SERIAL and IMPLEMENT_SERIAL macros.

The MFC implementation for persistent data stores data for many objects in a single contiguous part of a file. The object's Serialize method translates the object's data into a compact binary format.

The implementation guarantees that all data is saved in the same format by using the CArchive Class. It uses a CArchive object as a translator. This object persists from the time it is created until you call CArchive::Close. This method can be called either explicitly by the programmer or implicitly by the destructor when the program exits the scope that contains the CArchive.

This note describes the implementation of the CArchive members CArchive::ReadObject and CArchive::WriteObject. You will find the code for these functions in Arcobj.cpp, and the main implementation for CArchive in Arccore.cpp. User code does not call ReadObject and WriteObject directly. Instead, these objects are used by class-specific type-safe insertion and extraction operators that are generated automatically by the DECLARE_SERIAL and IMPLEMENT_SERIAL macros. The following code shows how WriteObject and ReadObject are implicitly called:

class CMyObject : public CObject  
// example usage (ar is a CArchive&)  
CMyObject* pObj;  
CArchive& ar;  
ar <<pObj;        // calls ar.WriteObject(pObj)  
ar>> pObj;        // calls ar.ReadObject(RUNTIME_CLASS(CObj))  

The method CArchive::WriteObject writes header data that is used to reconstruct the object. This data consists of two parts: the type of the object and the state of the object. This method is also responsible for maintaining the identity of the object being written out, so that only a single copy is saved, regardless of the number of pointers to that object (including circular pointers).

Saving (inserting) and restoring (extracting) objects relies on several "manifest constants." These are values that are stored in binary and provide important information to the archive (note the "w" prefix indicates 16-bit quantities):

wNullTagUsed for NULL object pointers (0).
wNewClassTagIndicates class description that follows is new to this archive context (-1).
wOldClassTagIndicates class of the object being read has been seen in this context (0x8000).

When storing objects, the archive maintains a CMapPtrToPtr (the m_pStoreMap) which is a mapping from a stored object to a 32-bit persistent identifier (PID). A PID is assigned to every unique object and every unique class name that is saved in the context of the archive. These PIDs are handed out sequentially starting at 1. These PIDs have no significance outside the scope of the archive and, in particular, are not to be confused with record numbers or other identity items.

In the CArchive class, PIDs are 32-bit, but they are written out as 16-bit unless they are larger than 0x7FFE. Large PIDs are written as 0x7FFF followed by the 32-bit PID. This maintains compatibility with projects that were created in earlier versions.

When a request is made to save an object to an archive (usually by using the global insertion operator), a check is made for a NULL CObject pointer. If the pointer is NULL, the wNullTag is inserted into the archive stream.

If the pointer is not NULL and can be serialized (the class is a DECLARE_SERIAL class), the code checks the m_pStoreMap to see whether the object has been saved already. If it has, the code inserts the 32-bit PID associated with that object into the archive stream.

If the object has not been saved before, there are two possibilities to consider: either both the object and the exact type (that is, class) of the object are new to this archive context, or the object is of an exact type already seen. To determine whether the type has been seen, the code queries the m_pStoreMap for a CRuntimeClass object that matches the CRuntimeClass object associated with the object being saved. If there is a match, WriteObject inserts a tag that is the bit-wise OR of wOldClassTag and this index. If the CRuntimeClass is new to this archive context, WriteObject assigns a new PID to that class and inserts it into the archive, preceded by the wNewClassTag value.

The descriptor for this class is then inserted into the archive using the CRuntimeClass::Store method. CRuntimeClass::Store inserts the schema number of the class (see below) and the ASCII text name of the class. Note that the use of the ASCII text name does not guarantee uniqueness of the archive across applications. Therefore, you should tag your data files to prevent corruption. Following the insertion of the class information, the archive puts the object into the m_pStoreMap and then calls the Serialize method to insert class-specific data. Placing the object into the m_pStoreMap before calling Serialize prevents multiple copies of the object from being saved to the store.

When returning to the initial caller (usually the root of the network of objects), you must call CArchive::Close. If you plan to perform other CFileoperations, you must call the CArchive method Flush to prevent corruption of the archive.

System_CAPS_ICON_note.jpg Note

This implementation imposes a hard limit of 0x3FFFFFFE indices per archive context. This number represents the maximum number of unique objects and classes that can be saved in a single archive, but a single disk file can have an unlimited number of archive contexts.

Loading (extracting) objects uses the CArchive::ReadObject method and is the converse of WriteObject. As with WriteObject, ReadObject is not called directly by user code; user code should call the type-safe extraction operator that calls ReadObject with the expected CRuntimeClass. This insures the type integrity of the extract operation.

Since the WriteObject implementation assigned increasing PIDs, starting with 1 (0 is predefined as the NULL object), the ReadObject implementation can use an array to maintain the state of the archive context. When a PID is read from the store, if the PID is larger than the current upper bound of the m_pLoadArray, ReadObject knows that a new object (or class description) follows.

The schema number, which is assigned to the class when the IMPLEMENT_SERIAL method of the class is encountered, is the "version" of the class implementation. The schema refers to the implementation of the class, not to the number of times a given object has been made persistent (usually referred to as the object version).

If you intend to maintain several different implementations of the same class over time, incrementing the schema as you revise your object's Serialize method implementation will enable you to write code that can load objects stored by using older versions of the implementation.

The CArchive::ReadObject method will throw a CArchiveException when it encounters a schema number in the persistent store that differs from the schema number of the class description in memory. It is not easy to recover from this exception.

You can use VERSIONABLE_SCHEMA combined with (bitwise OR) your schema version to keep this exception from being thrown. By using VERSIONABLE_SCHEMA, your code can take the appropriate action in its Serialize function by checking the return value from CArchive::GetObjectSchema.

In many cases the overhead of the general object archive scheme of WriteObject and ReadObject is not necessary. This is the common case of serializing the data into a CDocument. In this case, the Serialize method of the CDocument is called directly, not with the extract or insert operators. The contents of the document may in turn use the more general object archive scheme.

Calling Serialize directly has the following advantages and disadvantages:

  • No extra bytes are added to the archive before or after the object is serialized. This not only makes the saved data smaller, but allows you to implement Serialize routines that can handle any file formats.

  • The MFC is tuned so the WriteObject and ReadObject implementations and related collections will not be linked into your application unless you need the more general object archive scheme for some other purpose.

  • Your code does not have to recover from old schema numbers. This makes your document serialization code responsible for encoding schema numbers, file format version numbers, or whatever identifying numbers you use at the start of your data files.

  • Any object that is serialized with a direct call to Serialize must not use CArchive::GetObjectSchema or must handle a return value of (UINT)-1 indicating that the version was unknown.

Because Serialize is called directly on your document, it is not usually possible for the sub-objects of the document to archive references to their parent document. These objects must be given a pointer to their container document explicitly or you must use CArchive::MapObject function to map the CDocument pointer to a PID before these back pointers are archived.

As noted earlier, you should encode the version and class information yourself when you call Serialize directly, enabling you to change the format later while still maintaining backward compatibility with older files. The CArchive::SerializeClass function can be called explicitly before directly serializing an object or before calling a base class.

Technical Notes by Number
Technical Notes by Category