[azahnen]

Problem

A feature type can be built from several main tables ("concat"). Until now, every one of those tables had to have a primary instant (or primary interval), otherwise the schema failed validation. In practice some tables simply have no date/time, so such types could not be configured.

Separately, filters on a concat type looked up columns by the plain property name, but the SQL mapping keys them with a table-index prefix (e.g. 0_name), so the right column was not always found.

Solution

Tables without a primary instant or interval are now allowed. When a filter uses that property, it is treated as null in those tables:

• IS NULL matches.
• Every other predicate on it (e.g. T_INTERSECTS, >) matches nothing.
• Negations are pushed down to the predicates first, so NOT (datetime > ...) also matches nothing, following SQL's unknown semantics.

Details:

• Removed the check in FeatureSchema that required the number of primary instants/intervals to equal the number of concatenated objects. The check that they share the same property path stays.
• New CqlAbsentPropertiesMapper and Cql.mapAbsentProperties() rewrite a filter so given properties behave as null, resolving predicates to boolean literals.
• FilterEncoderSql detects, per main-table mapping, which primary instant/interval properties are missing and applies the mapper before encoding.
• SqlQueryMapping now resolves concat-prefixed property keys (e.g. 0_name) when looking up tables, columns and schemas for a plain property name.

Testing

• Added CqlAbsentPropertiesMapperSpec covering unchanged filters, IS NULL/IS NOT NULL, negation push-down, and mixed AND/OR branches.
• Added FilterEncoderSqlConcatPrimaryInstantSpec covering SQL output for a concat type with one table with and one without a primary instant (1=0, 1=1, (1=0 OR A.id = 1)).
• Added FilterEncoderSqlColumnNameSpec covering column lookup for single and concat main tables with differing column names.