Hao123.com Convert failed because the apex host only returns a meta-refresh stub; the fetch API now follows that hop to www.hao123.com.

Date: 2026-09-07 21:46 Summary: Hao123.com Convert failed because the apex host only returns a meta-refresh stub; the fetch API now follows that hop to www.hao123.com. Visibility: people

[question] 本地测试：https://hao123.com 点击转换还是失败

Tool links (English)

<convert-html-web-pages-to-word-document>

[try to solve]

Context

Local Convert on the HTML-to-Word URL tab with https://hao123.com still showed the generic convert error after the sanitize/Packer hardening. https://www.hao123.com on the same page succeeded.

Process

Called GET /api/tools/convert-html-to-pdf?url=https://hao123.com on wrangler at 127.0.0.1:8788. The Worker followed HTTP redirects to https://img.hao123.com/manual_res_2023/redict/to_www.html and returned 160 bytes: a <meta http-equiv="refresh" content="0.5;url= https://www.hao123.com?from=offline_host ">. No article text. htmlToBlocks then threw empty.

The same API for https://www.hao123.com returned ~528 KB with <title>hao123_上网从这里开始</title>. Playwright Convert of that URL enabled Download in 1.8s. Playwright Convert of the apex URL failed in 387ms with Error: empty.

Browsers follow that meta refresh; fetchHtmlFollowingRedirects only follows HTTP 3xx. The www homepage also contains a noscript refresh to /?__noscript__-=1; blindly following every refresh would fetch the noscript variant of the real site.

Root cause / analysis

hao123’s bare domain is an interstitial, not the portal. Mapping the interstitial is correctly “no blocks,” but that is the wrong document. Follow refresh only when the HTML is a small stub (≤8 KB and almost no visible text), delay ≤ 2s, target is http(s) and not a blocked host, and noscript tags are stripped first so the real homepage is left alone.

Solution

handleConvertHtmlToPdfFetchApi now loops up to three extra hops through such stubs, then absolutizes the final page. The Word page also follows one client hop if the API still returns a stub. After the Worker reload, the API for https://hao123.com returned finalUrl=https://www.hao123.com/?from=offline_host (527 KB). Playwright Convert of https://hao123.com then enabled Download (3s).

Notes / boundaries

This is the shared webpage-fetch API used by HTML-to-PDF and the batch webpage converters, so they pick up the same hop. CSS layout is still not mapped. Login walls and JS-only apps still fail. Cache version 4.58 is for the Word HTML that contains the client follow.

[actions]

• Updated src/tools/convertHtmlToPdf.ts to follow short meta-refresh stubs.
• Updated src/pages/convertHtmlWebPagesToWordDocumentPage.ts fetch to retry one refresh hop.
• Prerendered/gzipped this slug; local R2 seed; wrangler cache 4.58.