SaadaDB operator does not define classes... : Products contained in one collection are grouped in classes. Products of a given class are supposed to have a similar format. Then they can be selected by constraints set on any attributes (see SAADA-QL). However, such classification can not be achieved by the SaadaDB administrator because it would oblige him to know very well the format of all input products.

...but product configurations : That’s why, instead of defining classes, the administrator has to define product configurations. A product configuration is a set of product (input data files) of a given type (Entry, Image2D,..) and stored in a given collection. Files matching one product configuration can belong to one class or can be spread out in several classes according to the configuration setup.

 The concept of product configuration is relevant for the dataloader but not for the final user. SaadaQL never uses the product configuration.

 Each product configuration has a name, a target collection and some rules helping for the mapping :

The SaadaDB operator does also select mapping rules :

 1st rule : Define files matching a product configuration : There are 3 exclusive ways to define the membership of a product configuration.

 By directory : All products of the the given directory will be considered as a member of the product configuration.

 By filename : All products having a filename matching a given Java RegExp will be considered as a member of the product configuration.

 By content : All products having a given value for a given header keyword will be considered as a member of the product configuration.

 2nd rule : Define the class mapping strategy : The SAADA dataloader must know the class modeling of each input product. There is no generic way to do that. The administrator has to choose one strategy among 3. Notice that the use of UCDs (we are working on it - summer 2005) will probably make this classification issue obsolete.

 One class for all products : In this case, the class will have a name given by the operator (we advice the same name as for the product configuration). The attributes of this class will be the union of all attributes and all input products... This mode is useful for products having just little difference for instance due to an evolution of the process generating them.

 One class per product (mode MAPPING_1_1_SAADA) : This mode is useful when there are a little number of products or if all products result from different processes. The class names (used by SaadaQL) are given by Saada. In a further version, the operator will be able to rename product classes.

 SAADA mapping (mode MAPPING_CLASSIFIER_SAAD) : In this mode SAADA will make a class for each subset of product having the same format. The class names (used by SaadaQL) are given by Saada.