Slice 5+6 of the display block, one coherent unit (they share the state
machine the goal's dual review covers).
GlfwDisplayModeSwitcher (#376) ports retail's fullscreen semantics -
Device::ForceDisplayResolution @gmClient::Init 0x004047af is a REAL
video-mode change - through native glfwSetWindowMonitor on the same
IWindow.Native.Glfw handle path #348's cursor cache proved. Primary
monitor (retail's primary display device); refresh = the monitor's
highest for the picked WxH; the windowed placement is remembered for the
exit path; every failure is a no-throw (bool, reason) result.
SilkRuntimeDisplayWindowTarget.Apply (#388) becomes the state-aware
machine: fullscreen target = validated native mode switch (mode must be
in #391's DisplayModeCatalog - an offered mode is supported by
construction, making the "Graphics mode not supported" crash class
unreachable from the dropdown); windowed target while fullscreen = the
native exit (which sets the client size itself); plain windowed pick =
the proven #387 size write. A raw Size write NEVER happens against a
fullscreen window - on GLFW that is a video-mode request, and an
unsupported one was the exact unhandled-GlfwException that killed the
user's 2026-08-13 session. The old Silk borderless WindowState path is
deleted from the apply. New IWindowedSizeSurface narrows the window
dependency so the machine is unit-testable (FakePacingSurface idiom).
Live-verified on this machine (goal-sanctioned automated run):
display: fullscreen mode switch 1920x1080@300 -> framebuffer resize
event 1920x1080 -> vulkan: swapchain recreated 1920x1080 ok=True ->
graceful close, desktop mode restored.
Tests: 5 state-machine facts (validated switch/never-size-write,
unoffered refusal, failed-switch usability, native exit, plain windowed
write). App suite 4,972/3 skips. Gate script sections D4-D6 written
(black-screen-risk steps flagged). Dual Opus review of the pair follows
as its own round.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Decomp-first per the block's rule: the research doc
(docs/research/2026-08-13-retail-ui-display-change.md, committed here)
pulled retail's actual mechanism before any code. A display change runs
UIElementManager::RefreshEvent @0x0045C530 ->
UIElement::UpdateForParentSizeChange @0x00462640, which unconditionally
re-applies every floating window's own clamping MoveTo override
(x = max(0, min(x, parentW - selfW)) - top-left priority, oversized
windows pin to 0), then broadcasts global message 0xE whose sole
listener reloads the per-resolution auto layout. No proportional moves,
no resets; retail saves layouts only via @saveui.
Port: RetailWindowLayoutPersistence.ClampAllToScreen() is the cascade
clamp (no store I/O; _restoring suppresses the per-move save so a live
drag-resize cannot write settings.json per frame), and
RetailUiRuntime.Draw carries a two-step screen-size edge detector:
change frame -> clamp; first stable frame -> one
RestoreAll(saveBack:false) per-resolution reload (the 0xE analog; no
lazy save-back, matching retail's save-only-on-command). The login
restore path already used retail's exact clamp math (Apply) - the live
trigger was the missing half, which is precisely the stranding the user
reported.
Deliberate deviation, register AD-91: retail's gmFloatyChatUI windows
have NO clamp and can strand; the block's requirement ("UI windows must
stay reachable") clamps every registered window uniformly.
Tests: 5 new persistence facts (clamp/top-left-pin/no-move/no-save-on-
clamp/no-save-on-live-reload). App suite 4,967/3 skips. Gate script
section D3 filled in.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
