How to write KDE/Plasma e2e tests with selenium-webdriver-at-spi and raw CDP for Steam Gamescope. Use when authoring KDE tests, setting up the KDE runner image, or deciding where inputsynth must live.
Standardmäßig ist der Prompt ausgewählt, der zuerst die Quelle prüft. Sie können zu einem direkten Befehl wechseln oder eine lokale Kopie herunterladen.
Quelldateien prüfen
Lesen Sie SKILL.md und alle von SkillsMP angezeigten Begleitdateien, bevor Sie sich für eine Installation entscheiden.
Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fügen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prüfen und installieren.
Ein direkter Befehl überspringt den Prüf-Prompt. Prüfen Sie die Quelle, bevor Sie ihn ausführen.
npx skills add https://github.com/projectbluefin/testsuite --skill kde
Der Befehl bleibt in einer Zeile. Scrollen Sie horizontal, um ihn vor dem Kopieren vollständig zu prüfen.
Sie bevorzugen eine lokale Kopie? Laden Sie die Dateien herunter, die SkillsMP derzeit vorliegen.
Write KDE Plasma e2e tests with selenium-webdriver-at-spi and CDP.
entry_point
docs/skills/test-authoring/kde/SKILL.md
category
test-authoring
mcp_compliance_level
partial
status
active
dependencies
[]
tags
["kde","plasma","webdriver"]
description
How to write KDE/Plasma e2e tests with selenium-webdriver-at-spi and raw CDP for Steam Gamescope. Use when authoring KDE tests, setting up the KDE runner image, or deciding where inputsynth must live.
The ghcr.io/projectbluefin/testsuite-kde-runner:kde-runner image carries host-side orchestration dependencies only:
Layer
Ships in runner image
Ships on DUT
BDD runner
behave
—
WebDriver client
selenium
—
Raw CDP / WebSocket
websocket-client
—
Parsing helpers
lxml, PyYAML
—
AT-SPI input synthesizer
NOT here
selenium-webdriver-at-spi-inputsynth
WebDriver server
NOT here
selenium-webdriver-at-spi
selenium-webdriver-at-spi-inputsynth is a Qt6 Wayland client that links PlasmaWaylandProtocols' fake-input.xml. Plasma protocol versions and the Qt/KWin ABI differ across Fedora (Aurora/Kinoite/Bazzite), Ubuntu-based KDE Neon, and Arch-based KDE Linux. Building it once on Fedora and copying it into every DUT is not portable — it will silently fail or segfault. Upstream KDE runs it inside the SUT for exactly this reason, so we do too.
2. Required Plasma session environment
Apply these as a systemd drop-in or environment wrapper before the session starts:
KWIN_WAYLAND_NO_PERMISSION_CHECKS=1 is the most commonly forgotten; without it KWin silently ignores injected keyboard/mouse events.
3. W3C WebDriver client for AT-SPI
selenium-webdriver-at-spi is a standalone W3C WebDriver server, not an Appium driver. Do not use Appium-Python-Client; use plain Selenium with a custom options class:
Use options objects, not the removed desired_capabilities dict (deleted in Selenium 4.10).
Set driver.implicitly_wait(0) globally. Mixing implicit and explicit waits multiplies timeout durations.
Prefer accessibility id > name > class name. Avoid xpath outside quarantine.
driver.save_screenshot() uses the mandatory W3C endpoint. Do not use get_full_page_screenshot_as_file(), which is Chromium-only.
4. Gamescope / Steam CEF over raw CDP
Steam Big Picture is a CEF browser inside nested Gamescope, which has no AT-SPI bus. Drive it over raw Chrome DevTools Protocol on WebSocket, not chromedriver:
Do not ship chromedriver in the runner image. chromedriver strictly version-validates against the target Chromium, and Steam's embedded CEF is routinely older than the host chrome, so any skew raises SessionNotCreatedException.
5. org.kde.PlasmaShell.evaluateScript boundaries
org.kde.PlasmaShell.evaluateScript is the structural analogue of GNOME Shell.Eval, but it is a diagnostics and session-reset hook, not an interaction API:
It is gated by immutability() != SystemImmutableandKAuthorized::authorize("plasma-desktop/scripting_console"), so hardened/kiosk images may refuse it.
Real users do not drive the desktop through evaluateScript. Tests that open Kickoff or change a KCM through it validate a debugging path, not the user path.
Permitted uses: state inspection, layout dumps (dumpCurrentLayoutJS()), KWin support information, and between-scenario session reset. All user-facing interaction must go through AT-SPI/WebDriver, app D-Bus/CLI entry points (kcmshell6, KRunner D-Bus, desktop-file activation), or qdbus calls that mirror what a user actually triggers.
Common Rationalizations
"I'll just build inputsynth once in the runner image."
Rebuttal: It will link against the Fedora Qt/KWin ABI and break on Neon or KDE Linux. Build/install it on the DUT, keyed to the DUT's distro and Plasma version.
"Appium-Python-Client is more convenient."
Rebuttal: The KDE server is W3C WebDriver, not Appium. Appium methods like install_app and hide_keyboard return 404/501 and add mobile-only dependencies.
"I'll use chromedriver for Steam because I know Selenium."
Rebuttal: chromedriver version-checks Steam's embedded CEF and fails. Raw CDP over WebSocket removes that class of failure entirely.
"evaluateScript is the KDE equivalent of Shell.Eval, so I can use it everywhere."
Rebuttal: Only for diagnostics and reset. Primary interaction must mirror real user input.
Red Flags
selenium-webdriver-at-spi-inputsynth present in testsuite-kde-runner.
Appium-Python-Client or chromedriver in KDE runner dependencies.
Runtime pip install, curl, or urllib fetches of executable test code.
Missing KWIN_WAYLAND_NO_PERMISSION_CHECKS=1 before sending input events.
Using org.kde.PlasmaShell.evaluateScript to click UI elements or change settings.
Mixing implicit waits with explicit WebDriverWait.
xpath locators in AT-SPI tests without a documented quarantine reason.
Shared Helper API (real function names)
The kde-smoke suite imports shared helpers at module scope with no try/except guard. A missing or renamed helper must cause an immediate ImportError — never a silent skip that makes CI green while zero scenarios run.
tests/shared/kde_preconditions.py
Function
Signature
Purpose
is_kde_session
(context) -> bool
Probe whether the DUT is running a KDE/Plasma session
apply_kde_session_preconditions
(context, username="bluefin-test") -> KDEResult
Orchestrate the full KDE session precondition pipeline
has_sddm
(context) -> bool
Probe whether SDDM is the active display manager
has_kwriteconfig6
(context) -> bool
Probe whether kwriteconfig6 is available
has_plasma_wayland_session
(context) -> bool
Probe whether the Plasma Wayland session desktop file exists
configure_sddm_autologin
(context, username, session) -> KDEResult
Configure SDDM autologin
suppress_welcome_wizard
(context, username) -> KDEResult
Suppress Plasma Welcome Center and distro wizards
emit_determinism_dropin
(context) -> KDEResult
Write the systemd user-environment drop-in
seed_home
(context, username, force) -> KDEResult
Reset the test user's home directory
wait_for_plasma_session
(context, timeout) -> KDEResult
Poll until kwin_wayland, plasmashell, and AT-SPI are reachable
is_kde_image
(image_ref) -> bool
Pure string check for a KDE/Plasma image family (no SSH)
ensure_kde_session
(context, username="bluefin-test") -> KDEResult
Preferred runtime entry point: wait for Plasma, then suppress the wizard
apply_disk_prep
(context, username="bluefin-test") -> KDEResult
Full pre-boot (disk-prep) pipeline
configure_autologin
(context, username, session) -> KDEResult
DM-aware autologin drop-in (SDDM or PLM)
detect_display_manager
(context) -> str
Returns "sddm", "plm", or "unknown"
has_plm
(context) -> bool
Probe whether Plasma Login Manager is the active display manager
is_kde_image, ensure_kde_session, apply_disk_prep, configure_autologin,
detect_display_manager, and has_plm were added by the KDE session-lifecycle
work (merged in #654); configure_sddm_autologin survives as a deprecated
compat wrapper around configure_autologin. See
references/session-preconditions.md for
the phase model (disk-prep vs runtime).
is_kde_image is the canonical image-family predicate. It is a pure string
check — lowercase, take the last / segment, strip :tag and @digest, then
match the family tokens aurora, kinoite, bazzite, kde, plasma. Empty
or missing references return False. Suites must import it from
tests.shared.kde_preconditions rather than rolling a private copy: two
implementations of the same predicate drift, and the suite then gates
differently depending on which path evaluates it.
Do not invent helper names. This table plus the module source is the
complete public surface. Verify a name against tests/shared/kde_preconditions.py
on the branch you are actually targeting before importing it — the module grows,
so "it did not exist last week" is not evidence either way. The rule that never
changes is below: import at module scope, never guard the import.
tests/shared/kde_webdriver.py
Function
Signature
Purpose
new_session
(command_executor, options, app) -> WebDriver
Create a Remote WebDriver session against the KDE AT-SPI server
quit_session
(driver) -> None
Tear down a WebDriver session
launch_app
(app_id, command_executor) -> WebDriver
Start a session scoped to a single application
find
(driver, locator, timeout) -> WebElement
Wait for and return a single element
find_all
(driver, locator, timeout) -> list[WebElement]
Wait for and return all matching elements
wait_for
(driver, condition, timeout)
Poll condition until truthy
retry_atspi_action
(fn, attempts) -> T
Run fn with surgical retry for AT-SPI churn
save_screenshot
(driver, path) -> bool
Save a W3C screenshot
Do not invent helper names. This table is the complete public surface of
tests/shared/kde_webdriver.py; there is no start_driver or press_key.
Check the module before importing, and never guard the import with hasattr
or try/except — a wrong name must raise, not silently disable a feature.
No-silent-skip rule
This is the real D1 defect: tests/kde-smoke/features/environment.py imported
helper names that did not exist on that tree, a try/except swallowed the
ImportError, and all 13 scenarios were skipped while CI reported green. The
fix is not "memorise which names exist" — it is never guard a helper import,
so a wrong or renamed name fails loudly on the first run.
# WRONG — swallows ImportError, sets _KDE_HELPERS_AVAILABLE = False, skips all scenarios silentlytry:
from tests.shared.kde_preconditions import some_helper, another_helper
_KDE_HELPERS_AVAILABLE = Trueexcept Exception:
_KDE_HELPERS_AVAILABLE = False# RIGHT — bare import at module scope; ImportError propagates and fails the runfrom tests.shared.kde_preconditions import (
apply_kde_session_preconditions,
is_kde_image,
is_kde_session,
)
from tests.shared import kde_webdriver
A unit test at tests/unit/test_kde_smoke_environment.py asserts the correct names exist and are callable. If a helper is renamed, the test fails before CI can report a phantom green.
Verification
testsuite-kde-runner image contains only behave, selenium, websocket-client, lxml, and PyYAML.
No perceptual image differ is installed. The PyPI package odiff is an unrelated JSON/YAML utility, not the image differ of the same name — never add it.
No inputsynth, Appium-Python-Client, or chromedriver in the runner image.
behave --dry-run tests/<suite>/features passes for the touched suite.
Read-only KDE boundary
invent.kde.org and all KDE properties are strictly read-only in this repository. Clone, fetch, read files, and call read-only APIs are allowed. Never create issues, MRs, comments, branches, tags, forks, or authenticated writes to a KDE host.
