Repository navigation
QQQ 4.1.0-RC.4
Pre-releaseQQQ 4.1.0-RC.4 pairs the framework with Next UI 1.0.0-RC.14, published to Maven Central from release/4.1 on 2026-10-12. The owner approved the release and waived the Next acceptance-receipt gate for the exact Next RC.14 artifact (tag commit 03e7415a, Central JAR SHA-256 df7eb6706859820346adf870397163d44efe60d6e288d5b6c022d24f7d880952); the RC.1 accepted limitations still apply. QQQ 4.0.0 remains the stable release. Release tracking: #921; candidate PR #1087.
The signed annotated tag v4.1.0-RC.4 fixes source f6206e44 (the version bump on the merge of #1087). Candidate pipeline on 649dbebe: static analysis, ESB broker conformance, sample bootstrap, sample acceptance against Next RC.14, and mvn_publish all passed.
Merged since RC.3:
- #1075: core table/process permission enforcement (#1061, #1062), off by default in 4.1.x
- #1068: record search never returns a blank label and logs every search
- #1072: record search highlights, details, paging, sort and per-table provider status
- #1074: joined write locks are validated against the record as updated (#1073)
- #1076: record developer panel providers (core contract, endpoint, record security, ESB)
- #1078: permission-checked AI tool kernel in qqq-backend-core (QQQ Assistant phase 0/1)
- #1080, #1082: BOM-managed
qqq-frontend-nextmoves to1.0.0-RC.14 - Docs: #1054, #1067, #1069
This is a prerelease. Notes from the changelog:
The BOM and sample move the Next dashboard (qqq-frontend-next) from 1.0.0-RC.11 to 1.0.0-RC.14; the 4.1.0-RC.4
Next acceptance-receipt waiver is bound to that exact artifact. See the release procedure.
Security
- Permission enforcement in core actions (#1061,
#1062) —QueryAction,GetAction,CountAction,AggregateAction,
InsertAction,UpdateAction,DeleteAction,ReplaceAction,ExportAction,RunProcessAction,
RenderWidgetActionandGenerateReportActioncan now check the session's table/process/widget/report permissions
themselves (previously only the middlewares did, so MCP servers, assistants, custom routes and scripts skipped them),
andFieldSecurityLockis now enforced: denied fields are removed from read output and table meta data, and
filtering, sorting, grouping on or writing to them is a permission error. Per call,PermissionEnforcement.ENFORCE
always checks andwithPermissionBypass(reason)never does; otherwise outermost calls are checked when
QInstance.enforcePermissionsInCoreActions(or-Dqqq.enforcePermissionsInCoreActions/
QQQ_ENFORCE_PERMISSIONS_IN_CORE_ACTIONS) is true. Calls nested inside an action (process steps, customizers, widget
renderers, possible-value lookups, async children) stay trusted, andQSystemUserSessionpasses as before; the
scheduler, polling automations and SQS poller mark their runs as system bypasses. Off by default in 4.1.x (no
behavior change unless enabled); planned to default on in 4.2. See
Permission Enforcement in Core Actions.
Added
dryRunonInsertInput/UpdateInput/DeleteInput— runs pre-customizers (as a preview) and all
validations, returns would-be records (orDeleteOutput.dryRunPrimaryKeys), and stores nothing.QFieldMetaData.isSensitive— marks PII/secret fields;ExportActionleaves them out unless
ExportInput.withIncludeSensitiveFields(true), frontend field meta data carries the flag, and AI/MCP integrations must
not send their values to a model.