[linghengqian]

Two commits:

1. fix: seven plugin problems found while verifying the docs
2. docs: detailed English and Chinese documentation

Plugin fixes

Each was found by checking the docs against real runs (vfox 1.0.12 and 1.0.13, the live Google, GitHub and gitcode.com endpoints).

Problem	Fix
Installing an official release asked gitcode.com (PreInstall → Available → ohos.list()). While gitcode.com answered HTTP 429 the plugin waited 30 + 60 + 120 + 180 s and printed nothing: vfox search took 392 s and vfox install flutter@3.47.4 212 s	PreInstall asks for the official releases only. vfox search and latest still list the OpenHarmony releases, but try twice, stop at a 429 and warn on stderr. Installing an OpenHarmony build waits as before
3.13.3, 3.9.0-0.1.pre and 3.17.0-0.0.pre are in the index two or three times for one file, with different checksums, and table.sort is not stable, so an old entry could win: checksum error on Linux	The newest release_date of each archive of a version is kept, and the sort order is total
+ohos- tags (four releases since 2026-10-15) were ignored	Listed and installable
A global core.autocrlf=true wrote CRLF into bin/internal/shared.sh of Linux ARM64 and OpenHarmony-on-Linux source installs: $'\r': command not found	git.init sets core.autocrlf=false in the checkout
-arm64 source installs were offered for every release with an x64 build, but before 2.2.0 (Linux) and 3.44.0-0.1.pre (Windows) the release's own tool takes the x64 Dart SDK	Offered from those releases on
vfox install flutter@latest installed the newest beta	Stable releases sort before pre-releases
flutter does not support386architecture	Spaces

Behavior changes to look at

• vfox search lists the stable releases first, so flutter@latest is 3.47.7-x64 and not 3.50.0-0.1.pre-x64.
• ARM64 source installs: Linux no longer offers 1.20.0 to 2.0.6 as -arm64, Windows no longer offers the releases before 3.44.0-0.1.pre. On Linux those could not run. On Windows this is a judgment call: they did run, with the x64 Dart SDK through Windows' emulation, under a name that says arm64. The -x64 archives are still there, but flutter@3.24.0 (no suffix) now needs flutter@3.24.0-x64 on Windows ARM64, as it already does on Apple silicon for old releases. To keep the old Windows behavior, drop the "windows/arm64" line of OLDEST_NATIVE in lib/source.lua.

Docs

• README.md is the way in; docs/ has a page per subject (versions and names, channels, platforms, architecture selection, mirrors, OpenHarmony builds, flutter upgrade, troubleshooting, install from source, development, e2e, changes), in the style of vfox-java: exact messages, data with the day it was taken, a list of things that look wrong and are not.
• Every page has a Chinese twin (README_CN.md, docs/zh-CN/) and a language switch line.
• Behavior that is only on main is marked "Unreleased" / "未发布" and listed in docs/changes.md. After the next release, move those entries under the version and drop the marks: grep -rn -i -e unreleased -e 未发布 docs README*.
• The commands and messages were run, not copied. Checked on the docs: relative links and anchors (the Chinese headings too), all 41 external links, and every English page against its Chinese twin (headings, code blocks, inline code, table rows, numbers).
• Corrections to the existing pages: vfox info prints the directory above the SDK, so the .vfox-manifest is below it (flutter-upgrade.md); 2.10.0 is not a release without architecture metadata, 1.22.6 is, and the unverified claim that the Windows ARM64 -x64 archive cannot run is gone (arm64.md); the OpenHarmony list was three names, there are 25, and Flutter has no 3.41.10, 3.35.8, 3.27.5 or 3.22.4, so the fork is not "based on Flutter 3.41.10" (ohos.md).

Verification

• tests/hooks_test.lua: 22 groups pass, 10 of them new. On the code of main, 9 of the new ones fail; the tenth guards the wait for installing an OpenHarmony build, which must stay.
• stylua --check ., lua-language-server --check . --checklevel=Warning, the license header check (apache/skywalking-eyes:0.9.0), and gofmt, go vet and go test -short in tests/e2e pass.
• Linux container e2e, official (3.47.4) and ohos (3.41.10-ohos-1.0.0) with the latest vfox: pass.
• Real vfox, gitcode.com unreachable: an official install takes under 0.3 s (vfox 1.0.12 and 1.0.13), and with 1.0.13 a proxy that logs CONNECT saw no request to gitcode.com during it; vfox search returns after 2.1 s with the warning; latest installs 3.47.7-x64. 3.41.9+ohos-1.0.2 installs, lists, is used and uninstalled on live gitcode.com.
• Real index: PreInstall picks the newest checksum for 3.13.3, 3.9.0-0.1.pre and 3.17.0-0.0.pre.
• core.autocrlf: with the real global core.autocrlf=true, the checkout of the old code has CRLF in shared.sh, the new one has LF and flutter --version runs.
• ARM64 boundaries: the bootstrap scripts of all 300 releases with an x64 build were classified (Linux: 283 pick the Linux zip by CPU, from 2.2.0; Windows: 148 mention ARM64, from 3.20.0-1.1.pre), the Dart SDK of each was looked up in Flutter's storage, and the Linux scripts of eight releases were run as an aarch64 machine: dart-sdk-linux-x64.zip up to 2.0.6, dart-sdk-linux-arm64.zip from 2.2.0.

Only CI can tell

• Windows, macOS and real ARM64 hardware: this PR's ubuntu-26.04-arm, windows-latest and windows-11-arm jobs install 3.47.4 on them. The ARM64 ones are source installs, so they run core.autocrlf=false and the cutoff.
• That core.autocrlf=false writes the same CRLF files as true on Windows (core.eol is native there) was checked on Linux with core.eol=crlf and Flutter's .gitattributes, not on Windows.
• That an x64 Dart SDK does not run on an ARM64 Linux machine follows from the architecture and was not tried.

Squash merge is the only option here, so the PR title becomes the commit title.