diff --git a/e2e/app/plain.html b/e2e/app/plain.html new file mode 100644 index 0000000..ed661bf --- /dev/null +++ b/e2e/app/plain.html @@ -0,0 +1,13 @@ + + + + + office-engine e2e plain import + + + + + diff --git a/e2e/plain.spec.ts b/e2e/plain.spec.ts new file mode 100644 index 0000000..c8e05ff --- /dev/null +++ b/e2e/plain.spec.ts @@ -0,0 +1,73 @@ +import { describe, expect, test } from 'bun:test' +import { parasToDocx } from '../src/core/convert' +import { BlobSource } from '../src/core/source' +import { ZipArchive } from '../src/core/zip' +import { rebuildZip } from '../src/core/zipwrite' +import { docxParas, useBrowser, viewer } from './harness' + +useBrowser() + +describe('large plain imports in Chromium', () => { + for (const kind of ['markdown', 'html', 'code'] as const) { + test(`${kind}: imports beyond V8's function-argument limit`, async () => { + const v = await viewer({}, { page: 'plain.html' }) + const result = await v.page.evaluate((kind) => { + const { htmlParas, markdownParas } = (window as unknown as { __plain: typeof import('../src/core/plain') }).__plain + const count = 150_000 + const ps = kind === 'html' + ? htmlParas('

First

' + '

paragraph

'.repeat(count) + '

Last

', false) + : markdownParas(kind === 'code' ? '```text\n' + 'line\n'.repeat(count) + '```\n\nLast\n' : '# First\n\n' + 'paragraph\n\n'.repeat(count) + '# Last\n', false) + const text = (i: number) => ps[i].runs.map((r) => r.text).join('') + return { count: ps.length, first: kind === 'code' ? text(0).split('line').length - 1 : text(0), last: text(ps.length - 1) } + }, kind) + expect(result).toEqual({ count: kind === 'code' ? 2 : 150_002, first: kind === 'code' ? 150_000 : 'First', last: 'Last' }) + expect(v.errors).toEqual([]) + }) + } +}) + +test('a document beyond the browser scroll-height limit reaches, finds and edits its last page', async () => { + // Tall pages exercise the same scroll limit as a many-thousand-page Markdown + // import, without making the regression test parse a 50 MB fixture. + const blob = await parasToDocx(Array.from({ length: 200 }, (_, i) => ({ runs: [{ text: i === 199 ? 'END_MARKER' : `Page ${i + 1}` }] }))) + const zip = await ZipArchive.open(new BlobSource(blob)) + const xml = (await zip.text('word/document.xml')) + .replace(/]*\/>/, '') + .replaceAll('', '') + const file = await rebuildZip(zip, new Map([['word/document.xml', xml]]), blob.type) + const v = await viewer({ editable: true, onSave: true }) + await v.page.evaluate((bytes) => window.__setProps({ src: new Uint8Array(bytes), name: 'long.docx' }), [...new Uint8Array(await file.arrayBuffer())]) + await v.ready('docx') + await v.page.waitForFunction(() => document.querySelector('.de-bar')?.textContent?.includes('of 200')) + const height = await v.page.locator('.de-scroll').evaluate((el) => { + el.scrollTop = el.scrollHeight + return el.scrollHeight + }) + expect(height).toBeLessThanOrEqual(12_000_000) + await v.page.locator('.de-page[data-page="200"]').waitFor() + expect(await v.page.locator('.de-page[data-page="200"]').textContent()).toContain('END_MARKER') + + await v.btn('Find & replace').click() + const input = v.page.locator('.de-find input[aria-label="Find"]') + await input.fill('END_MARKER') + await input.press('Enter') + await v.page.waitForFunction(() => { + const p = document.querySelector('.de-page[data-page="200"] [data-pa]')?.getBoundingClientRect() + const sc = document.querySelector('.de-scroll')?.getBoundingClientRect() + return p && sc && p.top >= sc.top && p.bottom <= sc.bottom + }, null, { timeout: 5000 }).catch(async (e) => { + const state = await v.page.evaluate(() => { + const s = document.querySelector('.de-scroll')! + const p = document.querySelector('.de-page[data-page="200"] [data-pa]') + return { top: s.scrollTop, height: s.scrollHeight, viewport: s.getBoundingClientRect().toJSON(), paragraph: p?.getBoundingClientRect().toJSON(), status: document.querySelector('.de-bar')?.textContent, matches: document.querySelector('.de-find-count')?.textContent, pages: [...document.querySelectorAll('.de-page')].map(p => ({ page: p.getAttribute('data-page'), top: (p as HTMLElement).style.top })) } + }) + throw new Error(`${e}\n${JSON.stringify(state)}`) + }) + await input.press('Escape') + await v.selectText('END_MARKER', 'end') + await v.page.keyboard.type('!') + await v.page.waitForFunction(() => document.querySelector('.de-page[data-page="200"]')?.textContent?.includes('END_MARKER!')) + const saved = await docxParas((await v.save()).bytes) + expect(saved.at(-1)?.runs.filter((r) => r.t === 'text').map((r) => r.text).join('')).toBe('END_MARKER!') + expect(v.errors).toEqual([]) +}) diff --git a/src/core/plain.ts b/src/core/plain.ts index 6583d30..c60f55a 100644 --- a/src/core/plain.ts +++ b/src/core/plain.ts @@ -280,13 +280,18 @@ function diagramParas(lang: DiagramLang, src: string): ImportPara[] { } } +/** Append without passing a whole document as function arguments (which can exceed the engine's argument limit). */ +function append(target: T[], values: T[]) { + for (const value of values) target.push(value) +} + /** Markdown: the text, and code blocks in a diagram language (```mermaid, ```plantuml, ```dot) as drawings */ export function markdownParas(md: string, dom = true): ImportPara[] { const lines = md.replace(/^/, '').split(/\r\n|\r|\n/) const out: ImportPara[] = [] let text: string[] = [] const flush = () => { - if (text.length) out.push(...fromHtml(markdownToHtml(text.join('\n')), dom)) + if (text.length) append(out, fromHtml(markdownToHtml(text.join('\n')), dom)) text = [] } for (let i = 0; i < lines.length; i++) { @@ -301,8 +306,8 @@ export function markdownParas(md: string, dom = true): ImportPara[] { const lang = FENCE_LANG[fence[2].toLowerCase()] if (lang) { flush() - out.push(...diagramParas(lang, lines.slice(i + 1, j).join('\n'))) - } else text.push(...lines.slice(i, j + 1)) + append(out, diagramParas(lang, lines.slice(i + 1, j).join('\n'))) + } else append(text, lines.slice(i, j + 1)) i = j } flush() @@ -319,11 +324,11 @@ export function htmlParas(html: string, dom = true): ImportPara[] { for (const m of html.matchAll(HTML_DIAGRAM)) { const lang = FENCE_LANG[(m[2] ?? m[4]).toLowerCase()] if (!lang) continue - out.push(...fromHtml(html.slice(at, m.index), dom)) - out.push(...diagramParas(lang, decodeEntities((m[3] ?? m[5]).replace(//gi, '\n').replace(/<[^>]+>/g, '')))) + append(out, fromHtml(html.slice(at, m.index), dom)) + append(out, diagramParas(lang, decodeEntities((m[3] ?? m[5]).replace(//gi, '\n').replace(/<[^>]+>/g, '')))) at = m.index! + m[0].length } - out.push(...fromHtml(html.slice(at), dom)) + append(out, fromHtml(html.slice(at), dom)) return out } diff --git a/src/react/docx/DocxView.tsx b/src/react/docx/DocxView.tsx index 358f61b..9c569ea 100644 --- a/src/react/docx/DocxView.tsx +++ b/src/react/docx/DocxView.tsx @@ -29,6 +29,8 @@ import { pageHF, paginate, type PageSlice } from './paginate' const PX = 4 / 3 const GAP = 16 const PAD = 24 +// Stay below browser element-height limits, as the spreadsheet view does. +const MAX_SCROLL_PX = 12_000_000 const started = new WeakSet() @@ -210,11 +212,37 @@ export function DocxView({ client, onStatus, initialZoom = 'fit' }: DocxViewProp } const skeleton = complete ? 0 : 1 const totalH = acc + skeleton * (pageH * z + GAP) + PAD - GAP + const spacerH = Math.min(totalH, MAX_SCROLL_PX) + const maxLogicalY = Math.max(0, totalH - size.h) + const maxScrollY = Math.max(0, spacerH - size.h) + const scrollScale = maxScrollY > 0 ? Math.max(1, maxLogicalY / maxScrollY) : 1 + const scrollY = Math.min(maxLogicalY, scroll.y * scrollScale) + const scrollOffset = scrollY - scroll.y + const scrollToY = (y: number) => { + if (scrollRef.current) scrollRef.current.scrollTop = Math.max(0, Math.min(maxLogicalY, y)) / scrollScale + } + const scrollByY = (delta: number) => { + if (scrollRef.current) scrollRef.current.scrollTop += delta / scrollScale + } + const renderedScroll = useRef(scroll.y) + useIsoLayoutEffect(() => { renderedScroll.current = scroll.y }) + // A native scroll changes the viewport before useScroll's frame-batched + // update repositions the pages. Measure a target only after that commit. + const afterScroll = (run: () => void) => { + let frames = 0 + const tick = () => { + const sc = scrollRef.current + if (!sc || ++frames > 120) return + if (Math.abs(sc.scrollTop - renderedScroll.current) > 1) requestAnimationFrame(tick) + else run() + } + requestAnimationFrame(tick) + } const totalW = Math.max(size.w, pageW * z + PAD * 2) const left = Math.max(PAD, (totalW - pageW * z) / 2) - const first = tops.length ? lowerIndex(tops, scroll.y - pageH * z) : 0 - const last = tops.length ? lowerIndex(tops, scroll.y + size.h + 200) : -1 + const first = tops.length ? lowerIndex(tops, scrollY - pageH * z) : 0 + const last = tops.length ? lowerIndex(tops, scrollY + size.h + 200) : -1 const pageCount = complete ? pages.length : 0 // Both ends of the selection stay mounted, so the DOM selection survives scrolling. A removed page moves its end of // the range to the pages container, and a page mounted again there lands after that point: outside a range that ends on it. @@ -226,7 +254,7 @@ export function DocxView({ client, onStatus, initialZoom = 'fit' }: DocxViewProp editor.contentW = contentW editor.defaultFont = meta?.font editor.defaultSz = meta?.sz - const cur = tops.length ? lowerIndex(tops, scroll.y + size.h / 3) + 1 : 0 + const cur = tops.length ? lowerIndex(tops, scrollY + size.h / 3) + 1 : 0 // word count (debounced, off the typing path) const [words, setWords] = useState(0) useEffect(() => { @@ -337,11 +365,15 @@ export function DocxView({ client, onStatus, initialZoom = 'fit' }: DocxViewProp if (isEditing) editor.input.repair() return } + const bi = Number(sel.f.pa.split('.')[0]) + const pi = pages.findIndex((p) => bi >= p.start && bi < p.end) + if (pi >= 0 && (pi < first || pi > last)) { + scrollToY(tops[pi] - 16) + return + } if (!editor.input.restore(sel)) { // caret paragraph not rendered (scrolled away): bring its page into view, retry next render - const bi = Number(sel.f.pa.split('.')[0]) - const pi = pages.findIndex((p) => bi >= p.start && bi < p.end) - if (pi >= 0 && scrollRef.current) scrollRef.current.scrollTop = Math.max(0, tops[pi] - 16) + if (pi >= 0) scrollToY(tops[pi] - 16) if (pi < 0 || (pi >= first && pi <= last)) editor.pendingSel = null return } @@ -353,8 +385,8 @@ export function DocxView({ client, onStatus, initialZoom = 'fit' }: DocxViewProp if (range && sc) { const r = range.getBoundingClientRect() const box = sc.getBoundingClientRect() - if (r.height && r.bottom > box.bottom - 24) sc.scrollTop += r.bottom - box.bottom + 48 - else if (r.height && r.top < box.top + 8) sc.scrollTop -= box.top - r.top + 48 + if (r.height && r.bottom > box.bottom - 24) scrollByY(r.bottom - box.bottom + 48) + else if (r.height && r.top < box.top + 8) scrollByY(-(box.top - r.top + 48)) } }) @@ -380,24 +412,22 @@ export function DocxView({ client, onStatus, initialZoom = 'fit' }: DocxViewProp const sc = scrollRef.current if (!m || !sc) return const pi = pages.findIndex((p) => m.bi >= p.start && m.bi < p.end) - if (pi >= 0) sc.scrollTop = Math.max(0, tops[pi] - 16) + if (pi >= 0) scrollToY(tops[pi] - 16) setFind((f) => ({ ...f, idx })) - requestAnimationFrame(() => - requestAnimationFrame(() => { - const root = pagesRef.current - if (!root) return - const a = domFromPos(root, { pa: m.pa, o: m.a }) - const b = domFromPos(root, { pa: m.pa, o: m.b }) - if (!a || !b) return - const r = document.createRange() - r.setStart(a[0], a[1]) - r.setEnd(b[0], b[1]) - const rect = r.getBoundingClientRect() - const box = sc.getBoundingClientRect() - sc.scrollTop += rect.top - box.top - box.height / 3 - if (isEditing) editor.input.restore({ a: { pa: m.pa, o: m.a }, f: { pa: m.pa, o: m.b } }, false) - }), - ) + afterScroll(() => { + const root = pagesRef.current + if (!root) return + const a = domFromPos(root, { pa: m.pa, o: m.a }) + const b = domFromPos(root, { pa: m.pa, o: m.b }) + if (!a || !b) return + const r = document.createRange() + r.setStart(a[0], a[1]) + r.setEnd(b[0], b[1]) + const rect = r.getBoundingClientRect() + const box = sc.getBoundingClientRect() + scrollByY(rect.top - box.top - box.height / 3) + if (isEditing) editor.input.restore({ a: { pa: m.pa, o: m.a }, f: { pa: m.pa, o: m.b } }, false) + }) } // highlight visible matches (CSS Custom Highlight API) @@ -703,12 +733,12 @@ export function DocxView({ client, onStatus, initialZoom = 'fit' }: DocxViewProp if (!el) return false const r = el.getBoundingClientRect() const box = sc.getBoundingClientRect() - if (r.top < box.top + 24 || r.bottom > box.bottom - 24) sc.scrollTop += r.top - box.top - box.height / 3 + if (r.top < box.top + 24 || r.bottom > box.bottom - 24) scrollByY(r.top - box.top - box.height / 3) return true } if (show()) return - sc.scrollTop = Math.max(0, tops[pi] - 16) - requestAnimationFrame(() => requestAnimationFrame(() => void show())) + scrollToY(tops[pi] - 16) + afterScroll(() => void show()) } useEffect(() => { if (!cm) return @@ -767,14 +797,12 @@ export function DocxView({ client, onStatus, initialZoom = 'fit' }: DocxViewProp const sc = scrollRef.current const pi = pages.findIndex((p) => bi >= p.start && bi < p.end) if (!sc || pi < 0) return - sc.scrollTop = Math.max(0, tops[pi] - 16) - requestAnimationFrame(() => - requestAnimationFrame(() => { - const el = pagesRef.current?.querySelector(`[data-pa="${bi}"]`) - if (el) sc.scrollTop += el.getBoundingClientRect().top - sc.getBoundingClientRect().top - 24 - if (isEditing) editor.input.restore({ a: { pa: String(bi), o: 0 }, f: { pa: String(bi), o: 0 } }) - }), - ) + scrollToY(tops[pi] - 16) + afterScroll(() => { + const el = pagesRef.current?.querySelector(`[data-pa="${bi}"]`) + if (el) scrollByY(el.getBoundingClientRect().top - sc.getBoundingClientRect().top - 24) + if (isEditing) editor.input.restore({ a: { pa: String(bi), o: 0 }, f: { pa: String(bi), o: 0 } }) + }) } return ( @@ -800,7 +828,7 @@ export function DocxView({ client, onStatus, initialZoom = 'fit' }: DocxViewProp
1 ? 'clip' : undefined }} contentEditable={isEditing || undefined} suppressContentEditableWarning spellCheck={false} @@ -842,7 +870,7 @@ export function DocxView({ client, onStatus, initialZoom = 'fit' }: DocxViewProp blocks={st.blocks} meta={meta} page={page} - top={tops[i]} + top={tops[i] - scrollOffset} left={left} height={pageHeights[i]} zoom={z} @@ -854,7 +882,7 @@ export function DocxView({ client, onStatus, initialZoom = 'fit' }: DocxViewProp /> ), )} - {skeleton > 0 &&
} + {skeleton > 0 &&
}
{imgBox && (
diff --git a/test/plain.test.ts b/test/plain.test.ts index ceb8b07..2a02f7b 100644 --- a/test/plain.test.ts +++ b/test/plain.test.ts @@ -1,7 +1,7 @@ import { describe, expect, test } from 'bun:test' import type { Block, Paragraph } from '../src/core/docx/types' import { htmlToParas, parseHtml } from '../src/core/html' -import { decodeText, plainFormat } from '../src/core/plain' +import { decodeText, htmlParas, markdownParas, plainFormat } from '../src/core/plain' import { svgImage } from '../src/core/svgfile' import { Engine } from '../src/core/worker/engine' @@ -156,6 +156,36 @@ describe('plain sources open as Office files', () => { }) }) +describe('large plain documents', () => { + const count = 150_000 + const importedText = (p: { runs: { text: string }[] }) => p.runs.map((r) => r.text).join('') + + test('Markdown keeps every paragraph beyond the browser function-argument limit', () => { + const ps = markdownParas('# First\n\n' + 'paragraph\n\n'.repeat(count) + '# Last\n', false) + expect(ps).toHaveLength(count + 2) + expect(ps[0].heading).toBe(1) + expect(importedText(ps[0])).toBe('First') + expect(importedText(ps[1])).toBe('paragraph') + expect(importedText(ps[count])).toBe('paragraph') + expect(importedText(ps[count + 1])).toBe('Last') + }) + + test('HTML keeps every paragraph beyond the browser function-argument limit', () => { + const ps = htmlParas('

First

' + '

paragraph

'.repeat(count) + '

Last

', false) + expect(ps).toHaveLength(count + 2) + expect(importedText(ps[0])).toBe('First') + expect(importedText(ps[count])).toBe('paragraph') + expect(importedText(ps[count + 1])).toBe('Last') + }) + + test('Markdown keeps a fenced code block with more lines than the function-argument limit', () => { + const ps = markdownParas('```text\n' + 'line\n'.repeat(count) + '```\n\nAfter\n', false) + expect(ps).toHaveLength(2) + expect(importedText(ps[0]).split('line')).toHaveLength(count + 1) + expect(importedText(ps[1])).toBe('After') + }) +}) + describe('plain format detection', () => { const fmt = (name: string | undefined, type: string | undefined, s: string, whole = true) => plainFormat(name, type, enc(s), whole)