Fix package-backed Svelte/Vue scripts and delegated authority #428
No reviewers
Labels
No labels
code-review
correctness
dos
performance
security
severity/high
severity/low
severity/medium
tech-debt
Kind/Breaking
Kind/Bug
Kind/Documentation
Kind/Enhancement
Kind/Feature
Kind/Security
Kind/Testing
Priority
Critical
Priority
High
Priority
Low
Priority
Medium
Reviewed
Confirmed
Reviewed
Duplicate
Reviewed
Invalid
Reviewed
Won't Fix
Status
Abandoned
Status
Blocked
Status
Need More Info
No milestone
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Dependencies
No dependencies set
Reference
h-dv/code-index!428
Loading…
Reference in a new issue
No description provided.
Delete branch "codex/package-embedded-dispatch"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Svelte and Vue components with embedded scripts fail on the shipped generic runtime because delegation only consults native parsers. Reuse Claude's package-dispatch and base-source extraction implementation, then harden it so scripts use the project's approved TypeScript/JavaScript producer and template handlers bind to their own script declarations.
The wrapper retains contribution ownership. Each embedded contribution records its actual producer and container digest; resolver and read-only admissions intersect both producers' grants under the same active or pending generation. Delegation stays one level deep and shares the file deadline and aggregate fact/decoded-byte budget. Inner diagnostics, profiles and receiverless facts survive storage.
Installing a missing producer now reparses wrapper files on parser-free builds. Promotion and rollback also project the incoming contribution's verdict onto file metadata, so repaired facts do not retain an old refusal. Base-side diff extraction uses the same package producers, including embedded symbols.
Validation covers signed Svelte JS/TS and Vue TS packages through the four explicitly built parser-free binaries, template bindings, watching/restart, deleted base symbols, missing-producer refusal and live repair without deleting the database. Negative regressions cover nested delegation, aggregate budgets, nested/private names, foreign files, grant revocation, pending authority and container identity. An official signed-package v0.41.13 database upgrade passes. The existing generic acceptance lane carries these checks. The Linux GNU release job additionally runs the same acceptance against its four already-built binaries before packaging, so nightly-only coverage cannot allow a broken release. No additional Cargo build or CI lane was added.
Windows CI also exposed a daemon-restart disclosure defect when the operating system reuses a PID. Reconnect now compares the retained private authentication token as well as the PID, including after a timeout drops the channel. A deterministic regression covers changed-token replacements and same-token reconnects with and without a retained channel; no token is exposed in diagnostics.
Schema 89, host extraction epoch 4 and embedded dispatch semantics 2 intentionally invalidate stale caches. The first project start after upgrade rebuilds the index. Existing signed language packages remain compatible.
Adapted from Claude WIP commits
4f0bf992and58231e61; the original branches and worktrees are preserved. Related: #119, #279.Fixes #424.
Local validation: 5,060 workspace tests passed; 121 focused generation/embedded checks passed; 32 full-corpus stage/lifecycle checks passed with condition-only baseline updates; Clippy with warnings denied and workspace doc tests passed. All 37 timing checks passed across the serial run and an isolated rerun of the documented shared-page memory measurement; its limits are unchanged. The release workflow's parsed acceptance step and release-script guards also passed. The parser-free four-binary acceptance and official-package v0.41.13 upgrade proofs both passed.
CI follow-up validation: all 281 daemon library tests, the full/minimal MCP restart E2E, and 43 release-gate tests passed. Daemon all-targets Clippy passed with warnings denied. The release acceptance step is mirrored in both Linux job blocks to preserve their parity guard; its GNU-only condition avoids duplicate execution.
Linux CI exposed five large XAML fixtures losing their measurement to the fixed 35-second host deadline under parallel load. All nine XAML checks pass serially with unchanged production limits. Seven large-fixture tests now run in the existing
ci-timingprofile; two lightweight checks stay in the parallel suite. The registry verifies helper-bound test declarations and exact profile partitioning; removing a term from both profiles is mutation-tested to fail. The registry, all 50 path-gating checks, cadence checks, and Clippy pass. No retries, new CI lane, test skip, or raised budget was introduced.