[6.2.5] - 2026-06-26
Added
ClassGraph: preRegisterClassesWithAnnotation() for build-time discovered classes (#200)
Adds a pre-registration hook to ClassGraphCache so frameworks that know all annotated classes at build time (e.g. the quarkus-morphium extension via Jandex) can inject them and skip the runtime ClassGraph scan — essential for Quarkus native images, where a live scan finds nothing. Pre-registrations live in a separate map that always wins over the scan cache and survives invalidate(); clearPreRegistrations() drops them explicitly. Empty lists are valid pins (skip the scan, return empty). Covers the name-based getClassesWithAnnotation() path.
DNS: resolve TXT seedlist options for mongodb+srv:// (#169)
mongodb+srv:// URLs previously resolved the SRV host list but ignored the companion TXT record, forcing Atlas users to set authSource/replicaSet by hand. DnsSrvResolver now also resolves and parses the TXT record (RFC 1035 length-prefixed character-strings, k=v&k=v options). Morphium.resolveAtlasUrlIfNeeded() applies authSource → mongoAuthDb and replicaSet → requiredReplicaSetName, but only when not already configured, so explicit user configuration always wins (per the DNS Seedlist spec). TXT resolution failures yield empty options and never block a connection.
Changed
Messaging/ChangeStream: configurable change stream batch size
The change stream getMore batch size is no longer hardcoded to 1. It is now configurable via DriverSettings.changeStreamBatchSize (default 100) and can be overridden per monitor through ChangeStreamMonitor.setBatchSize().
A batch size of 1 delivers exactly one event per getMore round-trip, which caps stream throughput at roughly one event per network round-trip. On localhost this is unnoticeable, but over a high-latency link (e.g. an SSH/SOCKS tunnel with tens of milliseconds RTT) a busy stream cannot keep up: it drains a backlog at only ~1/RTT events per second and falls behind, delivering events — including messaging answers awaited by sendAndAwaitAnswers() — up to tens of seconds late, until traffic drops and the cursor catches up.
Because awaitData returns as soon as the first event is available, a larger batch size adds no latency at low traffic but lets a single round-trip drain many backlogged events. The original reason for batchSize=1 (a multi-document-batch hang in the previous watch() implementation) no longer reproduces after the change stream rewrite. The effective batch is still bounded by MongoDB's ~16MB per-reply limit regardless of the configured count.
Fixed
InMemoryDriver: seed upserted document from equality predicates nested in $and (#201)
On upsert the InMemoryDriver seeded the new document only from top-level non-$ filter keys. With a filter like {$and:[{_id:"lock"},{expires_at:{$lte:now}}]} the _id equality was never seeded, so the upserted document got a generated ObjectId and a later delete({_id:"lock"}) never matched (lock leak in the quarkus-morphium migration runner). collectUpsertEqualityFields() now seeds the document the way MongoDB does: scalar and $eq predicates are seeded, $and is recursed, dotted names become nested documents, and operator predicates / $or / $nor are not seeded. Verified against MongoDB 8.0.13.
DNS: only use public DNS as a last-resort fallback (#170)
DnsSrvResolver.systemDnsServers() appended 8.8.8.8/1.1.1.1 unconditionally, even when system name-servers were present. In split-DNS / private-Atlas setups this could resolve SRV records against public DNS (wrong results) and caused a per-server timeout when outbound UDP/53 is firewalled. Public DNS is now only added when no system name-server is configured (e.g. a minimal container without /etc/resolv.conf); an existing system resolver is treated as authoritative and fails fast.
Aggregation: field name translation in unset(Enum...) and lookup foreignField (#198)
Follow-up to #198: two remaining field-name translation gaps. unset(Enum...) in AggregatorImpl and InMemAggregator passed Enum.name() raw to the pipeline instead of translating via tf(), and AggregatorImpl.lookup(Class, Enum, Enum, ...) did not translate the foreignField with the lookup type. Both Aggregator implementations are now covered by explicit tests.
InMemoryDriver: $expr queries with aggregation operators no longer rejected
QueryHelper.validateQuery now only recurses into operators whose payload is a query document ($and, $or, $nor, $not, $elemMatch), so aggregation expression operators inside $expr (e.g. $dateFromString) are no longer misclassified as unknown query operators. Unknown top-level and field-level operators are still rejected.
ObjectMapping: BigDecimalMapper.unmarshall tolerates Integer/Long
unmarshall did new BigDecimal((double) d) and threw ClassCastException when MongoDB returned an integer-literal field as int32/int64. It now goes through Number#doubleValue() for any Number type, with a passthrough for already-decoded BigDecimal.