> The result of <XML validate> is R.
That was an oversight on my behalf, I had a hard time understanding the standard, but now the validation of DOCUMENT and CONTENT being accepted makes more sense.
The current patch has some issues.
> xmloption is document_or_content. But xmlvalidate_text_schema() always validates as a document.
As Andrey noticed, we should indeed support both a document and content. Which entails into an iterative validation (for each node provided) on content mode, so I should likely add the xmloption back. The fact it worked with the example I created was actually luck.
Also, I am not sure if some variables used inside of the PG_TRY are memory safe -- notice that none right now is set to volatile, despite being accessed in different parts of the block; other functions in xml.c do handle such correctly it seems (like xml_parse).
About the syntax proposal by Jim, I have no problems with complying to it. It does increase considerably the scope from what I originally intended, but that's the price to have something actually nice.
I can think of several useful extensions we could consider in a further implementation:
Schema Dependencies/Imports
CREATE XMLSCHEMA base AS '...';
CREATE XMLSCHEMA extended
IMPORTS base
AS '...';
Schema Versioning
CREATE XMLSCHEMA patient VERSION '1.0' AS '...';
CREATE XMLSCHEMA patient VERSION '2.0' AS '...';
XMLVALIDATE(doc ACCORDING TO XMLSCHEMA patient VERSION '2.0')
Custom Error Messages
CREATE XMLSCHEMA patient
AS '...'
ERROR MESSAGE 'Patient record does not match schema v2.0';
Schema inference from samples (if the lib supports it, that is)
CREATE XMLSCHEMA patient
INFER FROM (SELECT data FROM patient_samples);
And much more, but perhaps that's already too ambitious for a first version.
I'll wait for the others to ring their bells.
Regards, Magueta.
