Persistence, storage, and retrieval

Most object-oriented applications need the ability to store object structures on persistent storage for later retrieval, and to transfer such object structures to other applications. Eiffel offers a full persistence mechanisms serving these needs.

Persistence completeness

A fundamental requirement on object persistence mechanisms is the Persistence Completeness rule, stated as follows in Eiffel: The Language:

Whenever an object is stored into an external file, the stored content contains all the dependents of that object. Conversely, retrieving a previously stored object also retrieves all its dependents.
Storing an object just by itself would usually result in wrong semantics: most objects contain references to other objects, which must also be stored and retrieved with it. The persistence completeness rule ensures that this is always the case. It also means, of course, that features used for storing and retrieving objects must do much more than simple input and output; they must perform complete traversals of object structures.

The Eiffel persistence mechanism applies Persistence Completeness.

Varieties of store operations

Different variants of the store operation are available: session, basic and independent store

  • Session store produces the most compact structure in the resulting files; but the resulting structure is dependent on the current execution of the system which executes the store operation (System is taken here, as elsewhere in this documentation, in its Eiffel sense of an executable assembly of classes, compiled together with the help of a configuration file.)
  • Basic store is like session store with the difference that the resulting structure is dependent on the system which executes the store operation (i.e. only the system creating the persistent version can retrieve it.)
  • On the other hand, independent store allows a system running on a computer with a certain architecture to retrieve, without any explicit conversion operation, object structures stored by a system running on a machine of a completely different architecture. In addition, independent store lets you retrieve an old version of the object that was saved (see more details in the recoverable section below.)

Using the storage and retrieval facilities

Historically, the persistence mechanism was offered via the helper class STORABLE which you could use as an ancestor whenever you wanted to store and retrieve objects. Via this class you would have access to basic_store and independent_store to store an object, and retrieved to retrieve one. However this was not necessary and the persistence mechanism could be used directly from any descendants of the IO_MEDIUM using routines with the same names. This manner of storing and retrieving objects is called the C persistence mechanism since it was completely written in C and is included as part of the Eiffel runtime.

Today, we recommend using the SED (SErialization Deserialization) persistence mechanism, entirely written in Eiffel. It is very flexible, since it lets you control the format of the stored structure, but in most cases it suffices to rely on the simple helper class SED_STORABLE_FACILITIES. This class offers session_store, basic_store, and store (the de-facto independent store), as well as retrieved.

In both cases, you only need to be aware of the difference between the various storing mechanism (session, basic and independent) at storage time. The stored structure will always be available through feature retrieved; this feature will figure out, from the format of the stored structure how it was stored and will decode it accordingly.

Caution: This is only true when using just the C or SED persistence format. If the structure was stored using the C format, you need to use the C retrieved feature. Conversely, if it was stored using SED, you need to use the SED retrieval mechanism.

Regardless of the mechanism used, the feature retrieved returns a result of type detachable ANY and is typically used through an object test.

With the C persistence format

The example below will show you how to store an object using the C persistence format. It uses independent_store but you could also use in-place basic_store instead:

store_object (o: ANY; p: PATH) -- Store object `o` into file located in `p`. local f: RAW_FILE do create f.make_with_path (p) f.open_write f.independent_store (o) f.close end

Then to retrieve the object you would write the following:

retrieve (p: PATH) -- Retrieve object stored in file located in `p`. local f: RAW_FILE do create f.make_with_path (p) f.open_read if attached {MY_TYPE} io_medium.retrieved as data then -- Retrieved result is of expected type. -- Proceed with processing of result, -- typically with calls of the form `data.some_feature'. else -- Result was not of expected type MY_TYPE. end f.close end

If the structure in the file has been corrupted and retrieved is unable to do its job, it will trigger an exception (the code for that exception in class EXCEPTIONS (which inherits it from EXCEP_CONST and is discussed in the next section, together with the notion of exception code) is Retrieve_exception.)

With the SED persistence format

The example below will show you how to store an object using the SED mechanism assuming the current class is a descendant of SED_STORABLE_FACILITIES. It uses the store feature but you could also use session_store or basic_store too.

store_object (o: ANY; p: PATH) -- Store object `o` into file located in `p`. local f: RAW_FILE writer: SED_MEDIUM_READER_WRITER do create f.make_with_path (p) f.open_write create writer.make_for_writing (f) store (o, writer) f.close end

Then to retrieve the object you would write the following:

retrieve (p: PATH) -- Retrieve object stored in file located in `p`. local f: RAW_FILE reader: SED_MEDIUM_READER_WRITER do create f.make_with_path (p) f.open_read create reader.make_for_reading (f) if attached {MY_TYPE} retrieved (reader, False) as data then -- Retrieved result is of expected type. -- Proceed with processing of result, -- typically with calls of the form `data.some_feature'. else -- Result was not of expected type MY_TYPE. end f.close end

In case of an error during retrieval, no objects will be returned and instead the query retrieved_errors will list all the errors it encountered.

Recoverable format

Sometimes you will be in a position where the definition of a class (its "schema") will have changed between the time you stored your object and the time you are trying to retrieve it. Such schema changes include:

  • Renaming the class.
  • Adding attributes.
  • Removing attributes.
  • Renaming attributes.
  • Changing the types of attributes.

The persistence mechanism allows you to retrieve the old version of the object only if it was saved using the independent_store facility.

The mechanism lets you define precisely what will happen in the case of a scheme change, or "mismatch". Each time you retrieve an object of a certain base class whose schema has changed, the feature correct_mismatch will be called. This feature is defined in ANY and has the following effect:

  • Its default version causes an exception. This is the proper behavior since the old object might not make sense with the new schema (for example, it might violate the invariant), and you do not want to continue the computation, without warning, with wrong objects.
  • To specify otherwise, and avoid the exception, just redefine correct_mismatch in the class whose schema has been changed.

For example, the important EiffelBase library class HASH_TABLE changed between version 5.1 and version 5.2 to use SPECIAL rather than ARRAY for its internal data storage. To retrieve a 5.1 version of HASH_TABLE, you can redefine correct_mismatch as follows: correct_mismatch -- Attempt to correct object mismatch during retrieve using `mismatch_information'. do -- In version 5.1 and earlier, `content', `keys' and `deleted_marks' -- were of base class ARRAY. In 5.2 we changed it to be a SPECIAL for -- efficiency reasons. In order to retrieve an old HASH_TABLE we -- need to convert those ARRAY instances into SPECIAL instances. -- Convert `content' from ARRAY to SPECIAL. if attached {ARRAY [G]} mismatch_information.item ("content") as l_temp then content := l_temp.area end -- Convert `keys' from ARRAY to SPECIAL. if attached {ARRAY [H]} mismatch_information.item ("keys") as l_temp then keys := l_temp.area end -- Convert `deleted_marks' from ARRAY to SPECIAL. if attached {ARRAY [BOOLEAN]} mismatch_information.item ("deleted_marks") as l_temp then deleted_marks := l_temp.area end if content = Void or keys = Void or deleted_marks = Void then -- Could not retrieve old version of HASH_TABLE. We throw an exception. Precursor {TABLE} end end

Note the use of mismatch_information. This is a once query of ANY of type MISMATCH_INFORMATION which behaves like a HASH_TABLE. The keys of the table are the names of the attributes on which a mismatch occurred, and the values are the corresponding object fields as they were originally stored. In this particular case of HASH_TABLE we know that the previous version was an ARRAY, so we do an object test and if it succeeds we assign its area to the corresponding attribute of HASH_TABLE.

If a class name changed, you should create an instance of CLASS_NAME_TRANSLATIONS, it behaves like a HASH_TABLE where the keys represent the old name, and the value the new name. This instance needs to be created before the call to retrieved.

cached: 10/07/2024 1:40:49.000 AM