Develop from/to wrappers for format adapters with ISO 19139

[eblondel]

The purpose is to develop 2 wrapper functions that will wrap custom functions dedicated to map ISO 19139 from/to other metadata language formats (with candidate adapters #82 #83).

A generic metadata mapper will be drafted to plug adapters between geometa ISOMetadata classes and
other metadata classes (in particular EML and NetCDF-CF). If enough generic this pivot metadata mapper may be externalised in a separate project and used by geometa.

This mapper will be developed first in a separate branch. The mapping tests done with EML package will be handled with the most recent version of EML (from Github) where handling of EML metadata elements has been switched from S4 to list objects. In term of package lifecycle and release to CRAN, this work will be merged to the master branch only when EML package will be up-to-date in CRAN.

[eblondel]

The generic metadata converter implemented consists in set of classes and functions available here.

The design includes the definition of the metadata formats subject to the conversion. A format is defined with the class pivot_format (R6 class). The properties describing a pivot_format are made of a unique id the name of package(s) used (for information only), followed by string expressions (using sprintf formatting):

• a reader to read metadata properties from the source object
• a checker for checking existence of a given property in the source metadata object and,
• a constructor, for creating the target metadata object for the given property. In case the constructor is NULL, then no conversion to this metadata format will be possible (case of ncdf4 in Adapter for NetCDF-CF metadata #82).

Formats for geometa, ncdf4 and EML/emld are then registered by default in geometa. Please note that are not Import packages, as such strong dependency with packages like EML/emld/ncdf4 is not expected in geometa since it is only used for this mapping/conversion feature). Registration of default formats can be done with setMappingFormats() in R user session, otherwise this function will be called when attempting a first conversion. If registered, the mapping formats can be listed with getMappingFormats(). This function will list the formats and indicate if reading/writing is enabled (in case of ncdf4, the column "to" will be FALSE, because there is no writing capacity). A function is already available to register a new mapping format with function registerMappingFormat.

The rules required to convert from one object to another are listed (property by property) into a tabular data file. By default the mapping file used is the one embedded in geometa at: https://github.com/eblondel/geometa/blob/master/inst/extdata/mappings/geometa_mapping.csv An interactive file is available as read-only (and used for developments) at: https://docs.google.com/spreadsheets/d/1N-Xpp8-YjVzQ4CORfdas_9CaRV4hdrRKPbO4xXNs8-Q/edit#gid=0 Each column of this mapping file corresponds to a format. Each row of the mapping file will be modelled as mapping with class pivot_format_mapping designed to map the property from a source object, to a target object. The content of a cell is a string representation of the way to access the property and is modelled with a class named pivot_format_rule. A mapping is then modelled with 2 formats (source,target) to which we associate 2 rules (source,target).

The methodology behind writing a pivot format rule is out of scope of this ticket, but will be documented in exhaustive way in the future to foster collaborative work around the mappings made available in geometa (starting with EML/emld and ncdf4). To summarize, the string representation allows to perform the following tasks:

• accessing (read) a property from a metadata object, nested or not, single or list items.
• filtering a property based on another property (e.g. filter a contact which role is "owner")
• splitting string properties (with "sep" parameter), array item selection (with "index" parameter")
• perform formatting of target values
  In a summary, the methodology put in place pretends to provide a generic syntax for maping R objects: a source object (in format 1) to an target object (in format 2), in analogy to syntaxes like Xpath/XSLT for XML objects. In that way the description of mapping rules can be done in a data table, instead of hardcoding rules in a R script.

A set of unit functions are in charge of reading source object (get_pivot_source_object), building the target object (create_pivot_target_object, feed_pivot_target_data) in order to apply a mapping (apply_format_mapping). The function convert_metadata constitutes the generic converter, that is then used in different setters for the default formats considered, for use with the generic as function. See #82 and #83 for use cases.