test(e2e): add editor smoke tests for typing, formulas and slides

[juliusknorr]

Summary

Adds Playwright smoke tests exercising basic editing in each editor, run against the example app:

• Document — type text, apply bold/italic, verify the text round-trips via the automation API
• Spreadsheet — enter numbers and a =A1+A2 formula, verify the formula is accepted in the cell
• Presentation — build a three-slide deck and verify the slide count

A shared helpers.ts opens a new document from the example page, waits for the editor iframe to finish loading, and exposes editorApi() to call window.Asc.editor inside the iframe.

Notes

• Element ids are renamed in this build, so tests drive the editors via keyboard shortcuts and the asc_* automation API rather than clicking toolbar buttons.
• Spreadsheet test asserts the formula is accepted, not that it computes 15: asc_getCellInfo().asc_getText() returns the formula string and GetSelectedText() is empty for cell selections. A TODO documents the clipboard-based computed-value check (needs clipboard permissions in the project config).
• Presentation slides are added through AddSlide() because Ctrl+M only fires when the thumbnail panel is focused.

Testing

Verified locally against a running example server:

3 passed (11.7s)

AI assistance

Tests authored with AI assistance (Claude Code, claude-opus-4-8); reviewed and verified by the contributor.

[Alex-Arsys]

LGTM

[j-base64]

Looks good,

Just one tiny comment inline about bold/italic test.

also, not sure of importance or root cause, noticed that running with --workers 1 all tests pass clean.
with parallel workers, tests intermittently open the wrong editor/extension

[j-base64]

commented these lines and the test still passed.
maybe we can test these two bold/italic actions with

const isBold = await editorApi(editorPage, (api) => api.get_TextProps().get_TextPr().get_Bold());
const isItalic = await editorApi(editorPage, (api) => api.get_TextProps().get_TextPr().get_Italic());
expect(isBold).toBe(true);
expect(isItalic).toBe(true);

Tested it locally, fails commenting line 10-11 and passes with them uncommented.

[moodyjmz]

TL;DR: The spreadsheet cell reads (spreadsheet-formula.spec.ts:32-40) race the selection change and can flake. nameBox.press('Enter') resolves when the key is dispatched, not when the editor finishes moving the selection and asc_getCellInfo() reflects the new cell — and expect(await cellText(...)).toBe(...) is a one-shot with no retry. Wrap the reads in expect.poll (the presentation test already does this for the analogous async-count case):

await select('A1');
await expect.poll(() => cellText(editorPage)).toBe('5');
await select('A2');
await expect.poll(() => cellText(editorPage)).toBe('10');
await select('A3');
await expect.poll(() => cellText(editorPage)).toBe('=A1+A2');

retries: 1 will mostly hide the flake rather than fix it, which is worse — hence the poll.

Detail

The spreadsheet runs calc off the UI thread, so there's a genuine async gap between the name-box navigation and asc_getCellInfo() returning the new cell's info. The tell is that presentation-slides.spec.ts already uses await expect.poll(() => slideCount(...)).toBe(3) for exactly this class of problem (a value that updates asynchronously after a mutation) — the same treatment just wasn't applied to the cell reads.

Reviewed with AI assistance (Claude Code, claude-opus-4-8); grounded against the PR diff and the existing example-page.spec.ts / playwright.config.ts on main.

[chrip]

Verification

Claim / Item	Reality	Status
Document test verifies bold/italic (PR desc: "apply bold/italic")	Only assertions are #slot-btn-undo button not disabled and asc_GetSelectedText() contains 'Hello world'. Plain typing enables undo and produces the text, so the Ctrl+b/Ctrl+i keystrokes (lines 10-11) are effectively unverified — matches @j-base64's local finding.	❌
Spreadsheet reads are deterministic	select() does nameBox.press('Enter') (resolves on key dispatch), then a one-shot expect(await cellText(...)).toBe(...). asc_getCellInfo() can lag the selection move → flake. Presentation test uses expect.poll for the same async-value class; spreadsheet does not.	❌
Presentation slide count uses proper retry wait	Uses await expect.poll(() => slideCount(editorPage)).toBe(3) — correct, deterministic.	✓
Spreadsheet asserts formula accepted, not computed (=15)	Correct and honestly documented in the TODO; asc_getText() returns the formula string, computed-value path needs clipboard perms.	✓
No arbitrary sleep/waitForTimeout	Confirmed — waits are expect-with-timeout, waitForURL, toBeHidden on #loading-mask, expect.poll. Good.	✓
e2e CI job runs these specs and is green	build.yml has an e2e job (needs: [build, manifest], self-hosted) that runs npx playwright test in e2e/ — wiring exists. But on this PR's run the build jobs FAILED, so e2e was SKIPPED. The tests were never exercised in CI.	❌
Build failure is caused by this PR	No. Root cause is a 3rd-party core build step: gclient_paths.patch failed … patch does not apply / Common/3dParty/build_3rdparty.py failed! — unrelated to adding .ts test files.	✓ (unrelated)
Commit hygiene (conventional, DCO, AI disclosure)	test(e2e): …, Signed-off-by: present (DCO check green), Assisted-by: ClaudeCode:claude-opus-4-8 trailer, AI assistance disclosed in body.	✓
Diff scope	Exactly 4 new files under e2e/tests/; package.json/playwright.config.ts/constants.ts/example-page.spec.ts are pre-existing scaffold, not touched here.	✓

Issues & Suggestions

🔴 Blocking

• Document test does not verify the formatting it advertises — e2e/tests/document-editing.spec.ts:10-11. The Ctrl+b/Ctrl+i presses have no corresponding assertion; @j-base64 verified locally that commenting them out still passes. A "bold/italic" smoke test that stays green when formatting is entirely removed gives false confidence. Adopt @j-base64's concrete fix:

  const isBold = await editorApi(editorPage, (api) => api.get_TextProps().get_TextPr().get_Bold());
  const isItalic = await editorApi(editorPage, (api) => api.get_TextProps().get_TextPr().get_Italic());
  expect(isBold).toBe(true);
  expect(isItalic).toBe(true);

  (Or rename the test/PR to "type text" and drop the formatting claim — but verifying is clearly the intent.)

⚠️ Major

• Spreadsheet cell reads can flake — e2e/tests/spreadsheet-formula.spec.ts (the select→expect(await cellText(...)) block, ~lines 28-37). The name-box Enter resolves before asc_getCellInfo() reflects the new cell. Wrap each read in expect.poll(() => cellText(editorPage)).toBe(...), mirroring the presentation test. Note retries: 1 in playwright.config.ts will mask rather than fix this — @moodyjmz's point is correct.
• Parallel-worker cross-talk — @j-base64 reports tests intermittently open the wrong editor with multiple workers (clean with --workers 1). playwright.config.ts sets fullyParallel: false but does not pin workers: 1, so spec files still run concurrently and appear to collide on the shared example app / new-tab targeting. Worth pinning workers: 1 (the workflow comment already notes the e2e job assumes a single runner) or root-causing the tab mismatch before this lands, otherwise CI will be flaky.
• No CI validation yet — the e2e job was SKIPPED because build failed on an unrelated 3rd-party patch error. There is currently no green run proving these specs pass in CI (only the author's local "3 passed"). Recommend rebasing onto current main to get a clean build + an actual e2e run before merging.

ℹ️ Minor / 💡 Suggestions

• Presentation test drives the API, not the UI — presentation-slides.spec.ts adds slides via api.AddSlide() because Ctrl+M needs thumbnail focus (documented). Reasonable for a smoke test, but it exercises the automation API more than user interaction; a follow-up focusing the thumbnail panel and using the shortcut would be closer to a real user path. 💡
• Selector robustness — tests rely on internal ONLYOFFICE ids (#slot-btn-undo, #ce-cell-name, #editor_sdk, #loading-mask). These can shift on upstream bumps, but this matches the existing example-page.spec.ts convention, so it's consistent, not new debt. ℹ️
• editorApi serialization — helpers.ts rebuilds the passed fn via new Function inside frame.evaluate. The "must be self-contained (no closures)" caveat is documented; fine for tests. ℹ️
• SPDX headers — the new .ts files carry no SPDX header, but the pre-existing example-page.spec.ts also has none, so the e2e dir convention is header-less. Non-blocking; flag only if the repo later standardizes. ℹ️

Verdict

Request changes — Solid, well-scoped test scaffolding, but the headline document test provably does not verify bold/italic (already flagged inline by @j-base64), the spreadsheet reads have a genuine race (@moodyjmz), and no CI run has actually executed these specs (build failed on an unrelated 3rd-party patch, skipping the e2e job). Fix the bold/italic assertion, poll the spreadsheet reads, and get a green e2e run — after that this is an easy approve.

Assisted-by: ClaudeCode:claude-opus-4-8