Repository navigation
Simfil adds shared schema storage, typed bitmasks, and APIs for cancellable queries and retaining complete typed results. Recursive search-scope detection, interrupted dictionary transfers and C++ compiler compatibility are corrected.
Added
- Introduce shared feature-model Schema 2.0 (mapget#199)
- Support bitmask values in feature inspection and queries (MAPV-498)
- Cancel or limit expensive queries while keeping partial results (simfil#164)
- Extract complete query results without losing value types (simfil#165)
Fixed
- Support temporary ArrayArena ranges in C++20 view pipelines (simfil#162)
- Prevent interrupted transfers from corrupting the string dictionary (simfil#166)
- Fix attribute-scope detection for recursive searches (simfil#167)
Implementation commits
Compatibility Notes
-
Upgrade mapget clients, remote/process datasources, Python wheels and erdblick WASM together. Tile streams require protocol 6.3, MapSchema codec 7 and simfil SchemaPool codec 4; older wire/schema artifacts are incompatible. Canonical /sources and /info metadata now identifies mapFormat and an exact schemaId instead of embedding feature definitions and JSON Schema. Resolve the matching binary schema before decoding partitions. Legacy inline featureTypes/featureModelSchema configuration remains accepted as input.
Native producers must retain the shared MapSchema owner and declare produced layer types separately from map-wide feature identities. Python canonical metadata dictionaries also need their matching map_schema artifact; the dictionary alone cannot reconstruct a datasource. Rebuild native consumers against the updated APIs.
Simfil native callers must migrate the previous Schema/SchemaBinding/SchemaReference/SchemaModel interfaces to SchemaPool, SchemaType and SchemaTypeModelNode. Type IDs belong to the exact schema owner, and dynamic string IDs now start at 256. Serve a matching frontend build including its workers and reload existing browser tabs. mapget clears incompatible server-side caches at startup, but external stored payloads are not converted.
-
Bitmask values now retain typed flags instead of formatted strings. Queries and custom styles that parse those strings should use named flag operations such as hasFlag(...). JSON consumers must handle the tagged _bitmask, bits, schema and typeId representation.
(MAPV-498)
-
Query limits are opt-in and cooperative, not hard preemption. The existing vector-returning API remains unbounded. A zero limit means no allowance, not unlimited; custom functions, regex and result processing need their own limits.
-
Filter projection consumers must read values[expression][result] and hostValues[expression][result], rather than one scalar per expression. Empty sequences, one null and one array-valued result remain distinct; consult valueErrors/hostValueErrors for failed slots. Upgrade binary readers with the matching stream protocol.
Native result copying requires destination field names to exist and uses limits separate from query evaluation. Failed copies return no partial root, but an append-only destination can retain bounded temporary allocations; copied containers are intended for read-only results.
-
Applications using schema field-path analysis must honor hasUnresolvedAccess. A nonempty list of known matching paths is not necessarily exhaustive.