Skip to content

e2e: the live standing check returns to the conversation that receives the firing - #1377

Merged
AbirAbbas merged 3 commits into
devfrom
fix/1344
Sep 22, 2026
Merged

AbirAbbas merged 3 commits into
devfrom
fix/1344

Conversation

@AbirAbbas

Copy link
Copy Markdown
Collaborator

Closes #1344 — which turned out not to be a product defect: the firing row is drawn.

What was true

TestTUIE2E/the_firing_reaches_the_person went red on 17ae56d with DEFECT: the firing reached the conversation but was never DRAWN in it. #1071 had changed the scenario's way back into the conversation from Escape to ctrl+t, but at that moment Home's keyboard is on the settled reminder exchange row, and ctrl+t acts only on a conversation row, so the chord was a silent no-op and the test polled Home for a row that was drawn in the conversation. The journal, the model's answer and the row were all in place.

What is true now

The scenario walks Up onto the already-open conversation, presses Enter, and waits for that conversation's own footer before the standing pass. Two focused tests pin the boundaries the report implicated so this cannot hide again: a real session.Agent over the loopback wire delivers exactly one standing update to the client's task lane (internal/remote/tasklane_test.go), and the drawn · said: row survives the reply its wake produces, once and once only (internal/tui3/standing_test.go). No product line changed.

Verified

  • The subtest red on the unfixed tree (403 s), green after: the firing is drawn in the conversation with remind me in 1 minute to · said: Time to drink water on the open conversation; the closed-window drain and home's since you left stay green.
  • Observation for the owner: ctrl+t on Home does nothing, silently, when the selection is an exchange row.

🤖 Generated with Claude Code

AbirAbbas and others added 3 commits September 21, 2026 21:44
…s the firing

The live firing scenario pressed ctrl+t while Home still selected the settled
reminder exchange. That chord acts only on a conversation row, so the test
stayed on Home and reported a delivered `said:` row as missing (#1344).

The scenario now walks onto the existing conversation, opens it, and proves the
transition before the standing pass. Two focused tests pin the boundaries the
report implicated: a real session agent over the loopback wire delivers one
standing update to the client lane, and the drawn `said:` row survives the
reply its wake produces.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
@AbirAbbas
AbirAbbas merged commit 419741f into dev Sep 22, 2026
4 checks passed
@AbirAbbas
AbirAbbas deleted the fix/1344 branch September 22, 2026 13:28
santoshkumarradha added a commit that referenced this pull request Sep 23, 2026
TestAStandingFiringReachesTheHostedConversation (from dev's #1377) opened
the lane through WatchTaskUpdates, whose ask is made off the loop, and then
repeated the ask synchronously as its receipt. When the door's own ask
reached the engine after the firing, it replaced the far subscription the
firing had gone down and the row was lost. Under parallel load this failed
on dev and on this branch alike. The test now installs the client lane and
makes the one ask itself, answered before the firing.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

An in-window standing firing is delivered but never drawn

1 participant