You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit c8ad8fb
Browse filesBrowse the repository at this point in the historyBrowse files
fix(ios): size sole-display gestures from XCTest screen metrics (callstack#4157)
* fix(ios): size the sole display from the system app, not the runner's UIScreen
The generated AgentDeviceRunnerUITests-Runner.app has no launch screen, so the
runner process can read UIScreen.main in compatibility mode (768x1024 on an
iPad, a letterbox on an iPhone). The synthesized rotation then turned landscape
touches about that box, and every press landed a fixed offset away while
reporting ok. Read the display from the AXRuntime system application frame,
the same screen the fills-screen proof already compares against.
Fixescallstack#4091
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(ios): log a failed system-app screen read and harden its test
Log AGENT_DEVICE_RUNNER_APP_SCREEN_UNRESOLVED stage=system-app when the
AXRuntime read fails, rename the reference-frame size to portraitSize (it
is points), resolve every UIScreen getter before swapping any, and skip the
display test unless the host app fills the screen.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* test(ios): read the expected screen from the host app, within the fill tolerance
The gesture-viewport test now expects the host app's frame instead of the
runner's UIScreen, and both display tests compare within the 0.5 pt the
fills-screen proof allows. The sole-screen frame doc names its unit.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(ios): refuse a synthesized gesture when the sole display cannot be measured
A sole display whose size AXRuntime would not report fell back to the
resolved window's box, which on a placed iPad window is the callstack#3899 basis.
synthesizedRotationFrame now answers nil only for that case, and tap, long
press, drag and planned gestures refuse with UNSUPPORTED_OPERATION before
they log or dispatch. A multi-screen host keeps the window route.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(ios): log every failed system-app screen read and keep the hint key-free
The system-app size read now logs an unavailable AXRuntime class too, the
refusal hint points at the runner log without quoting its key, the rotation
frame doc says an unknown orientation keeps the window box (no turn reads
it), and the display test clears its cached target.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(ios): say what the sole-screen route pays for after the display read
Admission on an iPhone still skips the fills-screen probe, but every device
now pays one AXRuntime read (~0.6-0.9 ms) when the caller sizes the display.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* perf(ios): read the sole display size once per runner
A device's display size does not change while the runner runs, so the
first successful AXRuntime read (~0.6-0.9 ms) is kept on the main-owned
state and later gestures reuse it; a failed read is not kept, so the
refusal stays reachable and the next gesture reads again.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(ios): keep the sole display size for one command, not the runner's life
Display scaling can change the system application frame while the runner
runs, so a size kept for the whole runner could turn later gestures about a
stale box. Each command clears it and reads again; a command's gestures
still share one successful read.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(ios): turn about a full-screen app's window when the display size is unreadable
A failed AXRuntime display read refused every synthesized gesture on a
single-screen device, including iPhone taps that worked through the window
before. When the sole-screen rule (RunnerResolveSoleScreen) proves the app
fills the screen, its window is the display and stays the rotation basis;
otherwise the gesture still refuses with UNSUPPORTED_OPERATION. Both routes
log AGENT_DEVICE_RUNNER_APP_SCREEN_UNRESOLVED stage=rotation-basis.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(ios): read the display box only for turns that move the axes
Portrait and an unknown orientation never read the rotation box, so an
unreadable display no longer refuses their gestures and they skip the
AXRuntime read entirely. The AX-failure tap tests now run in landscape,
where the box is read.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* ci(ios): keep the nightly iPad leg at five tests
The display-box test is stub-based and runs on the iPhone full suite, so
the iPad leg keeps the refusal test, which only an iPad can exercise.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* test(ios): share the AX-failure tap tests' launch, landscape and stub setup
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* refactor(ios): name the sole-screen admission rule and the turns that read the box
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* test(ios): pin that a portrait tap dispatches when the display is unreadable
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(ios): place the display-size read after admission and spell longpress as the CLI does
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(ios): dispatch a window-less portrait tap when the display size is unreadable
With the window query rejected and the AXRuntime display read failing there
is no box at all. A portrait tap or long press keeps its point, so on an
admitted sole screen it is now dispatched as given; any other turn still
refuses with UNSUPPORTED_OPERATION. Both log stage=rotation-basis window=none.
Tap and long press share one point-gesture resolver.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* test(ios): name exactly which turns read the display box
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(ios): read sole-display size from XCTest screen metrics
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
0 commit comments