A 'task check' run on clean main at b724c22 (pre-release, 2026-09-12) failed with 'task: Failed to run task "test": exit status 1'. The failing package was not captured: output was tailed, and the visible tail (internal/logging through ui, alphabetically) showed no FAIL, so the failure is in a package sorting before internal/logging (cmd/, internal/a-l*, or a domain/infrastructure package).
Five subsequent full runs (2x 'task check', 3x 'go test ./... -count=1') all passed with exit 0, on the same clean tree. Release archied/v1.23.0 + archie/v1.23.0 was cut on a green gate.
Next step: run 'go test ./... -count=1' in a loop capturing full output (not tailed), or 'go test ./... -count=10' on the suspect packages, to name the test. Then write the regression test rather than silently re-running until green.
Acceptance criteria
Flaky test identified by name and covered by a deterministic regression test, or the race/ordering cause fixed at its producer.
Close reason
Closed
Imported from beads archie-core-u4xu · GitHub #808 · type bug · priority P2 · created 2026-09-11 · closed 2026-09-18
A 'task check' run on clean main at b724c22 (pre-release, 2026-09-12) failed with 'task: Failed to run task "test": exit status 1'. The failing package was not captured: output was tailed, and the visible tail (internal/logging through ui, alphabetically) showed no FAIL, so the failure is in a package sorting before internal/logging (cmd/*, internal/a*-l*, or a domain/infrastructure package).
Five subsequent full runs (2x 'task check', 3x 'go test ./... -count=1') all passed with exit 0, on the same clean tree. Release archied/v1.23.0 + archie/v1.23.0 was cut on a green gate.
Next step: run 'go test ./... -count=1' in a loop capturing full output (not tailed), or 'go test ./... -count=10' on the suspect packages, to name the test. Then write the regression test rather than silently re-running until green.
### Acceptance criteria
Flaky test identified by name and covered by a deterministic regression test, or the race/ordering cause fixed at its producer.
### Close reason
Closed
---
_Imported from beads `archie-core-u4xu` · GitHub #808 · type bug · priority P2 · created 2026-09-11 · closed 2026-09-18_
sam
added the mediumbug labels 2026-10-27 13:41:03 +00:00
Candidate diagnosis, not confirmed, found as a side-discovery while adding a
real-process test for the configurable listen addresses (archie-core-4rk0).
TestUIProcessServesTheDashboardAgainstLiveDependencies failed 2/2 full gates
with that change and 3/3 in isolation, while HEAD was green 3/3. The cause was
not the production change: holding the new test back left the production change
passing. The added real-process test increased suite load, and that test's
shutdown deadline is load-sensitive.
This matches u4xu's signature: internal/app/archieui sorts alphabetically before
internal/logging, so it sits inside the window this bead identified as the
suspect range, and load-sensitivity explains "not reproducible in 5 runs" on an
otherwise idle tree.
Not proven to be the same failure. The way to settle it is the next step this
bead already names, with the shutdown deadline as the first place to look rather
than a blind bisect of the package range.
Resolved in the originating change by moving the wiring proof into the existing
smoke test (same coverage, no extra process), so nothing was suppressed and no
timeout was widened to paper over it.
_sam wrote (originally 2026-09-18):_
Candidate diagnosis, not confirmed, found as a side-discovery while adding a
real-process test for the configurable listen addresses (archie-core-4rk0).
TestUIProcessServesTheDashboardAgainstLiveDependencies failed 2/2 full gates
with that change and 3/3 in isolation, while HEAD was green 3/3. The cause was
not the production change: holding the new test back left the production change
passing. The added real-process test increased suite load, and that test's
shutdown deadline is load-sensitive.
This matches u4xu's signature: internal/app/archieui sorts alphabetically before
internal/logging, so it sits inside the window this bead identified as the
suspect range, and load-sensitivity explains "not reproducible in 5 runs" on an
otherwise idle tree.
Not proven to be the same failure. The way to settle it is the next step this
bead already names, with the shutdown deadline as the first place to look rather
than a blind bisect of the package range.
Resolved in the originating change by moving the wiring proof into the existing
smoke test (same coverage, no extra process), so nothing was suppressed and no
timeout was widened to paper over it.
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
A 'task check' run on clean main at b724c22 (pre-release, 2026-09-12) failed with 'task: Failed to run task "test": exit status 1'. The failing package was not captured: output was tailed, and the visible tail (internal/logging through ui, alphabetically) showed no FAIL, so the failure is in a package sorting before internal/logging (cmd/, internal/a-l*, or a domain/infrastructure package).
Five subsequent full runs (2x 'task check', 3x 'go test ./... -count=1') all passed with exit 0, on the same clean tree. Release archied/v1.23.0 + archie/v1.23.0 was cut on a green gate.
Next step: run 'go test ./... -count=1' in a loop capturing full output (not tailed), or 'go test ./... -count=10' on the suspect packages, to name the test. Then write the regression test rather than silently re-running until green.
Acceptance criteria
Flaky test identified by name and covered by a deterministic regression test, or the race/ordering cause fixed at its producer.
Close reason
Closed
Imported from beads
archie-core-u4xu· GitHub #808 · type bug · priority P2 · created 2026-09-11 · closed 2026-09-18sam wrote (originally 2026-09-18):
Candidate diagnosis, not confirmed, found as a side-discovery while adding a
real-process test for the configurable listen addresses (archie-core-4rk0).
TestUIProcessServesTheDashboardAgainstLiveDependencies failed 2/2 full gates
with that change and 3/3 in isolation, while HEAD was green 3/3. The cause was
not the production change: holding the new test back left the production change
passing. The added real-process test increased suite load, and that test's
shutdown deadline is load-sensitive.
This matches u4xu's signature: internal/app/archieui sorts alphabetically before
internal/logging, so it sits inside the window this bead identified as the
suspect range, and load-sensitivity explains "not reproducible in 5 runs" on an
otherwise idle tree.
Not proven to be the same failure. The way to settle it is the next step this
bead already names, with the shutdown deadline as the first place to look rather
than a blind bisect of the package range.
Resolved in the originating change by moving the wiring proof into the existing
smoke test (same coverage, no extra process), so nothing was suppressed and no
timeout was widened to paper over it.