Posted by request from COMCIFS for the information of software
developers interested in possible future directions for CIF.
-----------------------------------------------------------------------------------------------------
Adoption of a next generation dictionary definition language: DDLm
-------------------------------------------------------------------------------------
CIF has now been in active use as a data exchange framework for about
fifteen years. The exchange, archiving and deposition of CIF data are
supported by data dictionaries that define the items most widely used
in the different crystallographic applications. Dictionaries are CIF
documents organised according to a specific set of rules referred to
as the dictionary definition language (DDL). In crystallography two
DDL's have evolved to satisfy an expanding definition need; DDL1 is
the simplest and is used for the core, powder, modulated and precision
density dictionaries; DDL2 has more relational attributes and is used
for the macromolecular, image and symmetry dictionaries. While the
DDL1 and DDL2 attributes are similar, the differences add an extra
level of complexity to the application and the maintenance of the CIF
dictionaries. Knowledge management technologies have also advanced
considerably since either DDL was introduced and it was for these
reasons COMCIFS decided at the Florence Congress to charge a working
group (Syd Hall, Nick Spadaccini and John Westbrook) with the task of
proposing a "next generation" DDL for use with future crystallographic
definitions.
The brief for this group was that a new DDL should, at a minimum:
meet the attribute capabilities of existing DDLs; increase the
semantic richness and precision in definitions; and provide
mechanisms that will enable common data definitions to be shared
across domain-specific dictionaries (i.e. the same item need not
appear in more than one dictionary). An overarching requirement was
that definitions be simple to write, to maintain and to apply in CIF
exchange processes.
A DDL proposal has been received from the working group for our
consideration. It has been called DDLm to reflect the inclusion of
methods attributes for relating defined items. The name of the new
DDL is not important at this stage, and can be discussed later prior
to adoption.
COMCIFS now needs to discuss and assess this proposal. Details of
DDLm are available on the IUCr web at
http://www.iucr.org/iucr-top/cif/ddlm/index.html.
This page provides four URLs.
* README summarises the files available in the File Distribution URL.
* Descriptive Documents contains RTF and PDF files describing the new DDL.
* File Distribution provides individual files for downloading.
* Zip Distribution provides an automatic download of all files as a .zip
file.
ddl.dic contains the formal definitions of the DDLm attributes and is
the main document for consideration by COMCIFS. The other files
describe the DDLm attributes, and show their typical application in
crystallographic definitions. The TEST domain dictionaries cif.dic,
cif_core.dic, core_*.dic and com_*.dic have been used to test the
DDLm attributes, and these files are provided here to illustrate the
typical application of the attributes. They do not constitute a
proposal to replace the current core dictionaries.
It is recommended you read the descriptive documents DDLm_spec.rtf,
ddl_attr.pdf, ddl_import.pdf and dREL_spec.rtf in that order. It
certainly helps to appreciate the role "importation" plays in the new
DDL; this is the mechanism that ensures that an item is defined in
only one file and is imported only when individual dictionaries are
used in applications. To best understand the use of the DDLm
attributes in definitions, look at the test domain dictionaries
core_diffr.dic, core_struc.dic, etc. Note that the division of the
core definitions into these particular dictionary modules is not
important at the moment, and has been done simply to illustrate how
definitions may now be organized into their natural groupings (and be
maintained independently) without any need to duplicate common
definitions. Importation attributes will expand these dictionaries
automatically as they are applied and thus ensure that the latest
unique definitions are sourced and used.
Look carefully at the use of the symbolic methods expressions in the
definition of derivative data items (i.e. items that are related to
other data items). These inbuilt methods serve a number of functions;
foremost they permit a much more precise definition of an item in
terms of its relationship to other items, but methods may be also
applied actively to specific data instantiations for the purposes of
evaluation and validation. This is possible because the symbolic
language dREL is executable using a Jython engine that will be
supplied to software developers. It is also recommended that
developers be encouraged to develop their own "methods expressions"
(written in a computer language of choice) and link these to the
dictionaries via the item tags. Such external "methods systems" would
not be overseen by COMCIFS, though it may wish to provide standard
data files for conformance testing of derivative calculations.
All of the new definition features of DDLm will not be elaborated on
here, however COMCIFS members are asked in particular to carefully
review the new TYPE attributes. These are much more comprehensive
than DDL1 and DDL2 and provide for more precise definition and
validation. Note that container types List, Array, Tuple and Table
depend on the acceptance of new multi-line string delimiters based on
matching square, round and curley brackets. It is proposed that List
and Array strings be bounded by square [ ] brackets; Tuples by
round ( ) brackets and Tables by curly { } brackets. Bracketed
strings may be nested and extend over multiple lines. This syntax
change has been advocated in COMCIFS discussions for several years now
and it is will be applied to the Star File specifications. These
string delimiters are the only change to affect CIF data files (all
existing files will be backwards compatible) and the proposed syntax
is unlikely to pose problems for future parsers.
The _type.purpose attribute identifies the function and origin of a
data item; and in particular if its value is a number deemed to be
"Measured" (either directly from the experiment or by derivation) and
that therefore has an associated standard uncertainty value. This is
important in order to cope with the current practice of allowing the
SU value to be appended to the measured value of a single item
(e.g.
_blat 0.622(6)
)
or as two items with separate tags (e.g.
_blat 0.622
_blat_su 0.006
).
It is proposed with DDLm that the existence of
the item <tag>_su be implicit if the item <tag> is defined as
Measured; and that both the concatenated and the separated value-su
constructions be accepted interchangeably. That is, it is not
necessary to define <tag>_su in a dictionary but its presence in an
instance document is understood and acceptable.
It should be stated that DDLm, as a next generation definition
language, is intended principally for use in future dictionaries.
Because of its added semantic richness, application to existing
dictionaries will be encouraged but not be insisted on, and it is
expected that the existing dictionaries will be supported as long as
they are widely used. The motivation for the adoption of DDLm must be
its inherent advantages in meeting current and future knowledge
management challenges, and its attractiveness to software developers
so that dictionaries are used routinely for handling data.
The voting members of COMCIFS will be asked to give COMCIFS
approval to DDLm later in 2007. In the meantime comments are
welcome from all members of this mailing list.
David Brown
Chair, COMCIFS
--
I.David Brown
Professor Emeritus
McMaster University
Brockhouse Institute for Materials Research
King St. W
Hamilton
Ontario
L8S 4M1
Canada
[log in to unmask]
Tel: +905 525 9140 x 24710
Fax: +905 521 2773
|
