Word→PDF looked blank because html2pdf snapped the small gray HTML preview box onto A4; capture now uses an isolated A4-ratio iframe document.

Date: 2026-09-06 14:28 Summary: Word→PDF looked blank because html2pdf snapped the small gray HTML preview box onto A4; capture now uses an isolated A4-ratio iframe document. Visibility: people

[question] http://127.0.0.1:8788/zh/tools/convert-word-document-to-pdf 转换后的pdf是空白的

Tool links (English)

https://onlinefreetools.org/tools/convert-word-document-to-pdf Chinese: https://onlinefreetools.org/zh/tools/convert-word-document-to-pdf

[try to solve]

Context

On the local preview (http://127.0.0.1:8788/zh/tools/convert-word-document-to-pdf) the sample .docx converted without an error (status「完成」, download enabled), but the PDF preview looked empty: a full A4 page with at most a thin gray strip at the top. The HTML preview still showed the sample sentence「用于转换的 Word 示例段落。」, so mammoth was not the failure.

Process

Confirmed html2pdf.bundle was already loaded (the earlier jsPDF undefined crash from 2026-09-06 09:39 was already fixed). Conversion reached OftPdfWork.showPreview.
Playwright against Chrome measured the pdf.js canvas: 744×1052 (A4 at scale 1.25) with only ~126 dark pixels in an 180×12 box at the top-left, plus ~84k light-gray pixels. That matches a screenshot of #wordPdfHtmlPreview (gray 8rem min-height box) scaled to page width, not a missing PDF.
Compared with Convert HTML to PDF: that tool snapshots a sandboxed iframe document. Word→PDF called html2pdf().from(previewEl) on the in-page preview div.
Replaced the preview <div> with an iframe, wrote mammoth HTML as a full document (white background, #111 text, CJK-capable fonts, Word-like padding), and captured documentElement with html2canvas windowWidth/height padded to whole A4-ratio pages so a short sample stays one page.

Root cause / analysis

html2pdf.js rasterizes the source node, then fits that bitmap to A4 width. The Word preview node was a short gray Bootstrap box (~960×128). Fitting it to 595pt made the paragraph ~9pt of ink on an otherwise white page—easy to read as “blank,” especially below the fold. A second blank page appeared when capture height was a few pixels taller than one A4 after width-fit (html2pdf sliced a sliver page).

This is not a CJK font embedding issue: the pipeline already stores a JPEG of the canvas, so glyphs are pixels. Isolation from the tool page also avoids html2canvas failing on Bootstrap color functions.

Solution

Preview and PDF share one iframe document (sandbox="allow-same-origin"), max-width 794px.
Capture options: backgroundColor: #ffffff, scrollX/Y: 0, pagebreak: css+legacy, height rounded to N × width × (841.89/595.28).
Local verification (Playwright, cache PAGES_CACHE_VERSION 4.37): HTML iframe text matches the sample; PDF preview「第 1 / 1 页」; canvas shows the same sentence at the top-left with white margins; no pageerror.

Notes / boundaries

mammoth still flattens headers, text boxes, and complex Word layout; a file whose HTML preview is empty will still produce an empty PDF.
Long documents paginate by screenshot height, not native Word page breaks.
Production needs the usual upload:r2 plus Worker PAGES_CACHE_VERSION 4.37 (or later) so cached HTML is not the old div-capture page.

[actions]