Skip to content

Settings rows never re-freeze after the AI source changes #625

Closed

Description

@omesser

What happens

Attach a Harness, then pick Model API in AI source. The source row updates — the popup reads "Model API", the state line reads "Not attached." — and the Base URL, Model and API key rows below stay frozen. Typing into Base URL does nothing. Only a relaunch hands them back.

It fails in both directions. Start with no Harness, pick Harness · claude, wait for it to attach: the three rows stay editable while the Harness is the Completer.

Why it matters

The section's own copy promises the opposite:

Off above hands the HTTP rows back, and takes effect at once.

That sentence is the whole point of #500 — the wait is bounded by the user, not by a relaunch. On AppKit the wait is still a relaunch.

Cause

form::describe() is rebuilt on every draw, so http_row_parts reads harness::driving() live and the frozen flags in the description are always current. The AppKit renderer only ever reads them once:

  • frozen is applied in the build pass — settings_window.rs:791, :834, :869, :958, :1026, :1073.
  • draw() (settings_window.rs:505) sets values and placeholders. It never calls setEnabled or setEditable.

Closing and reopening Settings does not help: the window is hidden and reshown, not rebuilt.

Verified on

macOS 15 (Darwin 25.6), AppKit renderer, main at d152097e and on #593. Pre-existing, not from either.

scripts/verify-settings-macos.sh (once #626 lands) covers the launch-time half of this — it opens Settings after the Harness attaches for exactly this reason. The runtime switch is what it cannot cover until this is fixed.

Not just this

Two smaller things share the cause and are worth folding in:

  1. A frozen text field looks live. freeze_or_bind freezes with setEditable(false), which leaves the field at full contrast with a normal border — indistinguishable from an editable one until you type into it. The popups next to it grey out properly. The status strips (FormRow::status) exist in the data and no renderer draws them, which is where the "not in use" explanation was supposed to go.
  2. GTK and Win32 need the same check. Both build their controls in one pass too; whether they re-apply on redraw is unverified from here.

What "done" looks like

draw() re-applies frozen for every row it can reach, so switching AI source freezes or unfreezes the three HTTP rows in place — no relaunch, and no stale copy telling the user it already happened.

Activity

  1. added this to the V1 milestone on Sep 10, 2026
  2. added
    p1Next up: the default for work that should happen
    macosmacos / OSX specific
    on Sep 10, 2026
  3. omesser commented on Sep 11, 2026

    @omesser
    OwnerAuthor

    Fixed by 015cc5f7 on #593, verified live on macOS at tip eb77f282.

    draw() now re-applies frozen on the redraw path — setEditable(!frozen) for fields, setEnabled(!frozen) for popups and buttons — so the rows follow the AI source picker without a relaunch. Driven in an open window:

    • Model API → Harness · claude: attached at t=17s, the three rows froze at t=73s.
    • Harness · claude → Model API: rows live again at t=5s.

    The direction this issue quotes — "hands the HTTP rows back, and takes effect at once" — is the five-second one. The freeze direction lags about a minute behind the attach because it waits for the next redraw rather than riding the attach, which is a smaller thing than what this issue describes and not worth holding it open.

    The status strips render now as well: "Not in use: a Harness is the Completer" appears under the frozen rows, which was the other half of the complaint — a frozen field is still visually identical to a live one, but it is no longer unexplained.

    Two things this does not cover: the GTK and Win32 halves of the same commit, which cannot be compiled on the machine that verified this.

    Closing on the strength of the macOS verification. #593 is still open — if it does not land, this comes back with it.

    — Claude Code, on @omesser's behalf

  4. moved this from 👷 Todo to 🏁 Done in Fidgeton Sep 11, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    macosmacos / OSX specificp1Next up: the default for work that should happen

    Projects

    • Status
      Done

    Milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions