Semantic CLI/MCP tests conflate initial readiness with query and replacement behavior under parallel CI #416
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#416
Loading…
Reference in a new issue
No description provided.
Delete branch "%!s()"
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?
Full local nextest CI profile (12 cores, candidate
908852b1) ran 5,039 tests: 5,033 passed, four CLI query tests hit one-shot warming_up and one daemon replacement-envelope test killed an owner before proving the initial reply was ready. A sixth failure was a candidate fixture naming the current version and is separately corrected.The CLI failures cover content, envelope and PATH-selection semantics; they use the production ten-second one-shot deadline as an implicit debug-startup bound. The same 22-test binary passes alone. A loaded direct probe of the one-shot answer test passed in 8.73 seconds. Actual compiled CLI/MCP/daemon host, builtin and worker fingerprints were checked equal after the candidate feature-alias fix, so no remaining identity split was found in these binaries.
The replacement test only checks that its first reply lacks daemon_restarted. That also passes for warming_up, after which it kills the still-starting daemon and expects a normal replacement reply. It must first observe its expected symbol before testing a replacement. Replacement calls themselves should remain single-call assertions.
Proposed narrow fix: content-only query commands use the existing debug-only startup-wait override, while the separately classified hook-budget/deadline tests retain the actual production deadline. Prove initial daemon readiness before the deliberate kill. No production timeout changes, automatic retries, CI gate removal or new lane machinery.
Related #283/#301 (one-shot query behavior), #415 (long-startup lifecycle).
Targeted verification now passes: all 22 CLI query tests, the replacement-envelope regression with an explicit initial-ready premise, and the 11 fingerprint/9 version-site checks. The next parallel full run exposed the same pressure in blast_radius_hook::a_public_change_reports_its_dependents: its content/session-deduplication assertion received warming_up after 52.28s including nested build/startup work. Added a narrow resource group for query_cli and blast_radius_hook in the existing ci profile: maximum two concurrent tests, each reserves four global slots. This follows the existing wasm-precompile grouping, changes no production/hook deadlines and removes no tests. Timing profile and no-retry policy remain unchanged. Full validation is still in progress; not merged or released.