Stop a mid-turn rename offering to switch to the conversation on screen

[springfall2008]

Summary

The "Replying in 'X' — switch to it" banner exists to say a reply is running in a conversation the user is not looking at. setBusy() got that right. handleTitle() restated the same rule inverted:

if (state.busy && state.busy.conversation_id === state.conversation) {
    showBanner(state.busy.conversation_id, data.title);
}

title events are scoped server-side to the conversation being viewed, so that branch is true exactly when the busy conversation is the one already open — the case setBusy() deliberately hides the banner for. So the moment the model named a new conversation with set_chat_title, a banner appeared offering to switch the user to the transcript already in front of them.

The same handler also never updated state.titles, which updateChatTitle() reads, so the header went on saying "New chat" while the row in the list already showed the real name — both halves of the reported screenshot, one handler.

Fix

The decision gets a single owner, refreshBanner(), which setBusy() and handleTitle() both route through. Restating the rule at each call site is precisely what let the two drift apart. handleTitle() now refreshes the header and keeps state.busy.title current, so switching away afterwards shows the banner with the name the conversation actually has rather than the one it was created with.

Testing

Written test-first; the assertions were watched failing before the fix existed.

Beyond the string assertions this repo uses for its JS, I ran the real generated script in a DOM shim, before and after:

Step	Before	After
setBusy on the conversation being viewed	hidden	hidden
title event arrives mid-turn	visible	hidden
header title after the rename	undefined	"Octopus saving sessions today"
reply is in a different conversation	visible	visible

That reproduces the reported screenshot and shows both defects fixed with no regression on the case the banner is actually for.

./run_pre_commit — exit 0, all hooks passed, full --quick suite green.

Note on existing coverage

test_busy_banner_only_points_at_another_conversation asserted the rule inside setBusy(). Rather than dropping it, it is retargeted at refreshBanner() and additionally checks that setBusy() still delegates — so the coverage moves with the rule, and the rule cannot quietly migrate back into one caller and drift again.

[Copilot]

🟢 Approval recommended

The behavioral fix is localized, aligns setBusy() and handleTitle() behind a shared decision point, and is covered by targeted regression tests (only a minor comment wording nit noted).

Pull request overview

This PR fixes a UI logic mismatch in the Chat tab where a mid-turn conversation rename could incorrectly display the “Replying in ‘X’ — switch to it” banner for the currently open conversation, and could also leave the header title stale. The change centralizes the banner visibility decision and ensures title events update both the conversation list row and the header state.

Changes:

• Introduce refreshBanner() as the single owner of the banner show/hide decision and route setBusy() / handleTitle() through it.
• Update handleTitle() to keep state.titles (and thus the header via updateChatTitle()) in sync with server title events, and to keep state.busy.title current when applicable.
• Add/retarget tests to assert the banner rule is centralized and that title events don’t trigger a “switch to it” banner for the open conversation.

File summaries

File	Description
apps/predbat/web_chat.py	Centralizes busy-banner visibility logic and updates title-event handling to keep header/list/banner state consistent.
apps/predbat/tests/test_web_chat.py	Updates existing banner rule test to target refreshBanner() and adds a regression test for mid-turn title events.

Review details

• Files reviewed: 2/2 changed files
• Comments generated: 1
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[springfall2008]

The red pre-commit check was a test-harness bug, not a code regression: test_title_event_does_not_offer_to_switch_to_the_open_conversation (added in 11f54ba) ended without return failed, so it returned None and the runner's failed |= test(...) raised TypeError right after the test printed its own 'passed' line — aborting the whole web_chat module. Fixed in 921811f by adding the missing return; web_chat now passes end to end locally (exit 0) and the full run_pre_commit gate is green. No code change — the earlier merge-into-main and the Copilot thread's setIdle() resolution were already in.