Section 5.2.6.3.  CIFOBJ

The CIFOBJ class library (Schirripa & Westbrook, 1996) was developed to provide an object view of the mmCIF dictionary and to complement the relational CIFLIB class library (Westbrook et al., 1997) for handling mmCIF data. As such, it is very much tuned to crystallographic applications and it handles only the subset of STAR features used in CIF dictionaries. This does, however, include some limited handling of save frames, and is therefore a little more complex than applications interested only in CIF data input/output processing.

CIFOBJ has two components. The first builds a persistent store of objects of types item, subcategory, category and dictionary. Each such object is a container for all relevant attributes permitted for that object type. The object store is populated from the mmCIF dictionary by the CIFOBJ loader class (using methods provided by CIFLIB). This loader class assembles the dictionary objects and passes them to an object-storage manager. The second component of the CIFOBJ class library provides the methods necessary for building dictionary objects from the persistent store and for passing attribute strings to procedures concerned with establishing the integrity of data values.

The main reason for discussing this implementation here is to indicate the rapid growth in complexity needed to impose an application-specific object view on even a relatively simple STAR data structure. In generic prototype STAR projects such as OOSTAR, half a dozen or so classes and associated methods suffice to represent the highest-level abstract concepts implied in STAR constructions. In CIFOBJ, however, dozens of methods are associated with dictionary access. Dictionary-driven validation of an mmCIF numerical data value can involve:

(i) retrieval from the dictionary of the extended type declaration associated with the data item;

(ii) validation of the basic type (i.e. that it is indeed numeric) against the primitive data types supported by the dictionary;

(iii) validation of the extended type by regular-expression matching of the string representation of the value against the allowed patterns stored in the dictionary;

(iv) retrieval of any existing range constraints specified in the dictionary and comparison with the data value;

(v) location and evaluation of any associated standard uncertainty;

(vi) identification of the units in which the physical quantity is expressed;

(vii) if necessary, conversion of the units according to the conversion tables stored in the dictionary.

More complexity arises from the relationships between data items expressed through dictionary attributes such as name aliasing, parent–child dependencies and category membership.

Nevertheless, the complexity of these relationships is an indication of the richness of the metadata available through the dictionary approach, and the availability of well defined object representations simplifies the construction of well designed large-scale application frameworks, such as underpin the Protein Data Bank (Berman, Battistuz et al., 2002) and Nucleic Acid Database (Berman, Westbrook et al., 2002).