jakubklimek
commented
10 years ago Ok, see the SPARQL TRANSFORMER TEST pipeline. It takes the source data where there are s:geo properties from s:PostalAddress to s:GeoCoordinates. In the target graph, the properties are missing (deleted and not inserted) which should not be possible because SPARQL DELETE/INSERT query is supposed to be atomic according to the specification.
Source: http://internal.opendata.cz:8890/sparql?default-graph-uri=http%3A%2F%2Flinked.opendata.cz%2Fresource%2Fdataset%2Fseznam.gov.cz%2Fovm%2Flist%2Fnotransform&query=prefix+s%3A+%3Chttp%3A%2F%2Fschema.org%2F%3E%0D%0Aselect+*+where+%7B%3Fs+s%3Ageo+%3Fo%7D+limit+100&format=text%2Fhtml&timeout=0&debug=on Target: http://internal.opendata.cz:8891/sparql?default-graph-uri=http%3A%2F%2Flinked.opendata.cz%2Fresource%2Fdataset%2Fseznam.gov.cz%2Flist%2Fsparql-transformer-test&query=prefix+s%3A+%3Chttp%3A%2F%2Fschema.org%2F%3E%0D%0Aselect+*+where+%7B%3Fs+s%3Ageo+%3Fo%7D+limit+100&format=text%2Fhtml&timeout=0&debug=on
tomas-knap
tomesj
OK, this is a bit tricky to explain. Look at the gov.cz pipeline on ODCS. The extractor creates 2 dataunits DU1 - from list of organizations, DU2 - from details of organizations. They are XMLs. Each DU is transformed by a different XSLT. Then, the output goes to Geocoders, which for each s:PostalAddress create s:GeoCoordinates, which is connected to s:PostalAddress using s:geo, which is not correct because s:geo goes from s:Place, not s:PostalAddress. On the other hand, for the Geocoder to be universal, it needs to process only s:PostalAddress and connect the coordinates somehow. OK, so this is what then the SPARQL transformer should correct. On the input, it has the geocoder output and non-geocoded data, which should be merged (it points to the same input). Then, the query (the query is OK) in the SPARQL transformer should reconnect the s:geo to whatever is connected via s:address to the PostalAddress (and I assume it is s:Place, but don't use it in the query).
The issue is, that this does not work. What is even weirder is that it actually deletes the original s:geo properties, but does not insert the new ones. In addition, when I orginally suspected this issue, the "list" branch transformer did nothing - did not change the data, but this didn't happen the second time.