I noticed today that TSVOutput may produce output that contains prefixed names which is invalid per my reading of the relevant specification - http://www.w3.org/TR/sparql11-results-csv-tsv/
This is due to the fact that TSVOutput called FmtUtils.stringForNode() with only a Node resulting in it using the ARQ default prefix mapping for output.
Attached is a simple patch which fixes the issue, it should also speed up TSVOutput marginally as the existing code requires a SerializationContext to be created for every term serialized and incurs the cost of trying to turn URIs into prefixed names. Essentially the patch creates a null SerializationContext variable and just passes that to every call to FmtUtils.stringForNode() so that the ARQ default prefix mapping never gets used.
The second part of the issue is that this malformed TSV input may be accepted because TSVInputIterator uses NodeFactory.parseNode() to parse terms which calls SSE.parseNode() without any prefix mapping and thus internally ends up using the default SSE prefix mapping which means some prefixed names get permitted as valid when they should be rejected.
The second patch attached fixes this part of the issue by keeping an empty static prefix map and calling SSE.parseNode() directly and passing in this map.
|Status||Resolved [ 5 ]||Closed [ 6 ]|
|Status||Open [ 1 ]||Resolved [ 5 ]|
|Assignee||Andy Seaborne [ andy.seaborne ]|
|Resolution||Fixed [ 1 ]|