Workflow Editor Tests: Playwright Migration Status

Salvaged from the playwright_backlog branch (worktree ~/projects/worktrees/galaxy/branch/playwright_backlog, uncommitted WORKFLOW_EDITOR_CONNECT_PLAYWRIGHT_{PLAN,IMPLEMENTATION}.md, March 2026). Re-triaged against origin/dev @ 062e1411676 (2026-09-16).

Branch disposition

All code from that branch has landed in dev — verified feature-by-feature, not by commit subject:

Dev is strictly ahead except one line: the branch dropped @selenium_only from test_rendering_rules_workflow_1; dev still has it. That is the only salvageable delta, and it needs re-verification against current dev regardless of the March result.

The sibling branch playwright_backlog_1 is fully merged into dev (0 commits ahead).

Post-salvage cleanup (2026-09-16): worktree removed via ghwt rm galaxy playwright_backlog; branches playwright_backlog (tip ffd6655b125, 8 unmerged commits, all superseded), playwright_backlog_1 (2825bb09e42) and the stale test_stories (aa477d3b599) deleted. The test-stories branch/worktree (PR #21199) is untouched. Note the hyphen/underscore trap: test-stories is the live PR branch; test_stories was the stale merged one.

Remaining @selenium_only in test_workflow_editor.py (3 of an original 46)

Tiers A/B/C came from the March triage; C is now empty (dev migrated test_editor_place_comments, test_editor_snapping, test_editor_selection once mouse_drag() existed, plus test_editor_create_conditional_step and test_map_over_output_indicator).

Established patterns (still current)

Backend branching

if self._driver_impl.backend_type == "playwright":
    pw_driver = cast("HasPlaywrightDriver", self._driver_impl)
    # Playwright-specific code
else:
    # Selenium code

Accessing the Playwright page

• From NavigatesGalaxy: self.page property (raises for Selenium).
• When also needing _unwrap_element: cast to HasPlaywrightDriver.

Keyboard input

• page.keyboard.type("text") / page.keyboard.press("Enter").
• element.press("Control+a") for combos on a specific element.
• PlaywrightElement.send_keys translates Selenium Keys.* automatically.

Gotchas found the hard way

• assert_connected: SVG <g> elements report as hidden under Playwright — use wait_for_present, not wait_for_visible.
• open_in_workflow_editor must dismiss StateUpgradeModal; it blocks clicks under Playwright.
• Tooltips can intercept pointer events on adjacent terminals — fixed upstream in NodeOutput.vue with v-b-tooltip.hover.noninteractive. That was a real UI bug, not a test artifact.

Open questions

• test_aria_connections_menu: is there a Playwright equivalent of switch_to.active_element that returns a usable element wrapper (page.evaluate("document.activeElement"))? Needs a driver-layer addition.
• test_collection_input_sample_sheet_chipseq_example: why was it ever decorated?

Migration loop validated (2026-09-16)

Branch playwright_rendering_rules_workflow_1 (2 commits, pushed to the jmchilton fork, no PR opened). test_rendering_rules_workflow_1 passes under Playwright with the decorator simply removed — 3/3 green, ~20s per run. The March triage was right: it was a stale decorator, no code change needed.

Local environment gotchas worth reusing

• ~/galaxy_selenium_context.yml points local_galaxy_url at http://localhost:5173, and framework.py maps that config key onto GALAXY_TEST_EXTERNAL. So the browser goes through the Vite dev server and the framework does not start its own Galaxy — both run.sh (8080) and make client-dev-server (5173) must be up.
• A fresh worktree has an empty DB; the first run dies in setup with Timeout waiting on CSS selector [.loggedin-only]. Register the user first (/galaxy-register-user).
• A cold Vite causes a spurious first-run failure ([id^="g-card-action-workflow-edit-"] not clickable, ~265s). Warm it up and re-run before believing any timeout.
• GALAXY_TEST_SELENIUM_HEADLESS=1 cannot work for the Selenium backend on macOS — driver_factory.virtual_display_if_enabled wants Xvfb. Playwright headless is fine.

Tool conf: which conf has which tools

CI does not rely on sample_tool_conf.xml alone. run_tests.sh sets, for both backends:

-selenium   GALAXY_TEST_TOOL_CONF="lib/galaxy/config/sample/tool_conf.xml.sample,test/functional/tools/sample_tool_conf.xml"
-playwright GALAXY_TEST_TOOL_CONF="lib/galaxy/config/sample/tool_conf.xml.sample,test/functional/tools/sample_tool_conf.xml"

tool_conf.xml.sample already lists filters/randomlines.xml and filters/catWrapper.xml, so random_lines1 and cat1 are present in CI. Local run.sh with GALAXY_RUN_WITH_TEST_TOOLS=1 loads only the test conf, which is why random_lines1 was missing here and the workflow node rendered with no input terminal. -framework also uses the test conf alone.

Separately, sample_tool_conf.xml has referenced for_workflows/randomlines.xml since the samples_tool_conf rename while the symlink has never existed in any branch, so that conf logs Error reading tool configuration file from path 'for_workflows/randomlines.xml' on every startup. The added symlink mirrors dccab49baa8 (catWrapper) and fixes it.

Tier A/B re-triage (2026-09-16) — both remaining candidates are blocked on real driver gaps

Tried dropping the decorator on the other two and running under Playwright. Neither passes; the March “may just work now” guesses were wrong for both. Decorators left in place.

test_conditional_subworkflow_step — no ActionChains emulation

test_workflow_editor.py:1339
AttributeError: '_PlaywrightDriverImpl' object has no attribute 'move_to_element'

The test calls self.action_chains().move_to_element(conditional_toggle).click().perform(). HasPlaywrightDriver.action_chains() (has_playwright_driver.py:831) is a stub that returns self with the comment “Playwright doesn’t use ActionChains pattern” — so any chain method raises AttributeError. There is no emulation, only an object that exists.

This is the same underlying gap as Tier D’s test_aria_connections_menu. A real chainable shim (move_to_element, click, click_and_hold, move_by_offset, release, key_down, key_up, send_keys, perform) mapping onto page.mouse.* / page.keyboard.* would unlock several remaining tests across files rather than one — the reusable-abstraction move. mouse_drag() already proves the pattern for a single operation.

test_collection_input_sample_sheet_chipseq_example — trailing newline from send_enter

test_workflow_editor.py:306
AssertionError: assert 'The column i...le reports.\n' == 'The column i...able reports.'

Not a .text normalization issue — the assert compares the downloaded workflow’s tool_state against the CHIPSEQ_COLUMNS constant. workflow_editor_enter_column_definition (navigates_galaxy.py:1420-1424) types the description and then calls self.send_enter(elem), which carries the standing comment “seems like a Galaxy bug that these enter’s are needed?”. Under Playwright that ENTER lands in the field as a literal newline and is saved into tool_state; under Selenium it is not. Fixing it in the test (stripping, or dropping the assert) is the wrong direction — the backends should agree on what send_enter does to a committed field value.