Skip to content

ISSUE-818-HANDOFF.md

Latest commit

 

History

History
76 lines (51 loc) · 5.29 KB

ISSUE-818-HANDOFF.md

File metadata and controls

76 lines (51 loc) · 5.29 KB
Raw

Issue #818 — Handoff for a dedicated session

Goal: achieve ≥1 clean, non-guarded daily by modernizing test selectors/steps that drifted against a nightly node-UI overhaul. This is the entry gate for #773 and the wave's agent/MCP test promotions.

Branch: fix/issue-818-testid-drift-baseline (off main; no spec touched yet — investigation only).

Verdict already established (do not re-investigate from scratch)

The daily's ~29 failures are UI testid DRIFT, NOT a Langflow product regression. This corrects #816/#830 ("product regression, flag upstream"). The nightly restructured the node interaction surface (menu / expand / rename) between dev40 and dev46; our specs still use the old testids. Features work; the DOM changed.

Evidence: of 11 distinct testids that timed out across the failing specs, 10 are gone from the dev46 frontend (only add-mcp-server-button survives → that one is behavior/flow, investigate separately).

Therefore:

  • ❌ Do NOT flag upstream. ❌ Do NOT remove @stable. ❌ Do NOT pin the daily image.
  • ✅ DO modernize the selectors/steps to the new DOM (keeps coverage). ✅ Verify each fix against a local nightly.

Environment (the nightly the daily runs — NOT what the start script pulls)

scripts/start-langflow-docker.sh pulls stable (langflow:latest), not nightly. Start the nightly explicitly:

docker run -d --rm --name lf-nightly -p 7860:7860 \
  -e LANGFLOW_AUTO_LOGIN=true -e LANGFLOW_SUPERUSER=langflow -e LANGFLOW_SUPERUSER_PASSWORD=langflow123 \
  -e LANGFLOW_ALLOW_CUSTOM_COMPONENTS=true \
  langflowai/langflow-nightly:latest
# wait for: curl -sf http://localhost:7860/api/v1/version

Run a spec against it: PLAYWRIGHT_BASE_URL=http://localhost:7860/ npx playwright test <file> --workers=1 --retries=0.

Confirmed mapping — node name/description rename (biggest cluster)

Old flow (our specs) → new flow (dev46), verified live:

Old New
hover panel-description wrapper removed — no hover step
edit-name-description-button node-edit-name-description-button
inspection-panel-name input-title-<currentName> (DYNAMIC — scope to the focused node; duplicate node names → strict-mode hazard)
save-name-description-button node-save-name-description-button
description display generic-node-desc (a DIV; the description-EDIT input testid still needs a live check for edit-name-description-node.spec)

New rename sequence: click generic-node-title-arrangement → click node-edit-name-description-button → fill input-title-<name> → click node-save-name-description-button.

Also drifted in the node-menu/expand flow: expand-button-modal (used by the local expandFocusedNode helper in if-else-component-regression.spec.ts) is gone; more-options-modal still opens the menu (icon-Expand seen). Map the new expand item live before fixing.

Remaining clusters to map live (each: reproduce → read live DOM → find new testid)

Timed-out testids still needing their new-name map: edit-button-modal, expand-button-modal, edit-fields-button, toggle_bool_use_double_brackets, tweaks-button, icon-lock, div-table_headers, canvas_controls_dropdown_toggle_inspector-toggle. Plus add-mcp-server-button (EXISTS in bundle → not a rename; investigate its flow/behavior separately).

Mapping technique:

  • Static testids: grep the container bundle — docker exec lf-nightly sh -c "grep -rhoE '\"TESTID\"' /app/.venv/lib/python3.14/site-packages/langflow/frontend/assets/". Reliable only for literal testids.
  • Dynamic testids (runtime-composed): grep gives false-MISSING → drive the UI with playwright-cli and page.evaluate(() => [...document.querySelectorAll('[data-testid]')].map(e=>e.getAttribute('data-testid'))).

Footprint (files using the drifted node-rename testids)

if-else-component-regression.spec.ts (8 fails), edit-name-description-node.spec.ts (2), general-bugs-reset-flow-run.spec.ts, group.spec.ts, langflowShortcuts.spec.ts. Other clusters live in the agent/*, prompt-template, webhook, api-request, chat-input, auth/logout, mcp-client specs (see the #816 failing-set list).

Approach (recommended)

  1. Helper-first for leverage: fix the shared node-UI interaction (rename flow, expandFocusedNode, and any node-menu POM/helper) → cascades to many specs. Check tests/helpers/ + tests/pages/ for shared node-menu/rename helpers before editing specs one-by-one.
  2. Fix cluster → verify green against local nightly (--workers=1 --retries=0) before moving on.
  3. Repeat per cluster until the failing set is empty.
  4. Force-fail each touched test() (repo rule) + id-scoped flow cleanup on any spec touched (repo rule — audit afterEach/deleteFlow).
  5. Clean up any flow created during live scouting (DELETE via API; the browser-created "Basic Prompting" starter from the last session may still exist on the local instance).
  6. PR only after explicit user authorization.

Reference material from the investigation session

  • Memory: daily-failures-are-testid-drift.md (+ daily-slowdown-verdict.md, issue-817-runner-sizing.md).
  • #818 issue comment (2026-07-19) has the verdict + mapping table.
  • Non-sharded reference failing set: run 29658914382; sharded (dev48) reference: run 29674263823.