`UIParent.lua`'s registry moved under three tests at once:
- `ItemTextFrame` is `pushable = 5`, so the reader shares the screen instead of replacing what is
there; `QuestLogFrame` is `area = "doublewide"` with `width = 690`, not a left-slot window, so a
`whileDead` open seats at the doublewide slot and leaves the left one empty; and
`UIChildWindows` has five rows, the realm having added `ReputationDetailFrame`.
- `SetUIPanel` runs `CloseChildWindows()` for **any** slot it seats (`UIParent.lua:792-796`), not
just the centre one, so a window merely pushed across also puts the open letter away. The test
that pinned the stock trigger now watches both halves go.
- `user_placed_layouts()` is not empty after a manifest load, and should not be: the realm's own
movable windows stamp themselves in their OnLoad — `PlayerFrame.xml:488-489` is
`SetMovable(1) SetUserPlaced(true)`, the same feature as the `MOVE`/`MOVE_RESET` rows in their
right-click menus — and `TargetFrame`, the Turtle shop, transmog and guild bank do likewise. The
map test now asserts about its own row rather than about the file being empty.
- The blip tooltip calls `UnitName("player")` (`WorldMapFrame.lua:660`), which reads the player
record; `load_default_ui` seats a placeholder one so the realm's files have a race and class at
load, so a test that wants a named player names the record, after the manifest.
And one real race, not a realm change: `a_binding_table_resolves_whatever_case_the_package_spelled_it_in`
moves `BENILLA_HOME` and `WOW_CAPTURE` without taking `ENV_LOCK`. It was the only env-moving test
in that module missing it, and the collision shows up as a *neighbour* reading the wrong roster —
which is how it surfaced here, two addons tests failing in the full run and passing alone.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SgstFWtn1RxEJ5kVrpgYwn
gromandClaude Opus 5
34097b6164