Repository navigation
Conversation
editeur.org answers automated requests with a CAPTCHA, so `make schema` cannot fetch anything and code generation is stuck. For 2.1 that turns out to be avoidable: commit 9352123 deleted 2_1_rev03_schema/ when the build moved to downloading at build time, and its parent still has all six XSDs. `git show` reaches them with no network at all. `make schema-from-history` restores them into schema/v2. Verified here: all six files land (1.8 MB), including the XHTML subset, and `make -n schema` afterwards only tries to download v3 — the pattern rule treats schema/v2 as satisfied once the directory exists. This deliberately does not commit any EDItEUR file. schema/ is gitignored, so what this repository distributes is unchanged, and the licensing question ADR-0003 raises is untouched: the target is our own code, and the files it recovers have been in this repository's public history since 2021 either way. It does not help 3.0 or 3.1 — neither is in the history — so the schema acquisition decision in ADR-0003 still has to be made for those. Not reviewed by a subagent: the account is at its monthly spend limit and review agents fail to start. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01P8rZakwAiz1jpViUX34x1Z
総評動きます。狙いも機構の選び方も妥当で、ブロッカーはありません。 一時クローンで 指摘必須なし。 推奨 1: shallow clone で失敗し、空の
|
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01P8rZakwAiz1jpViUX34x1Z
Review found the failure mode that mattered. In a shallow clone — which is what actions/checkout produces by default — `git show 9352123^:...` fails, but the shell had already created the redirect target, leaving a 0-byte XSD behind. `schema/%` then treats schema/v2 as satisfied and `make schema` stops trying to fetch v2, so the next build reads a truncated schema. Restore into schema/v2.tmp and mv it into place only once every file is written, and guard the whole thing with `git cat-file -e` so a shallow clone gets a message naming the fix instead of a git error. Verified in a `--depth 1` clone: the target fails with that message, no schema/ directory is created at all, and `make -n schema` still wants both v2 and v3. The pin is now the full SHA of the parent (0134b80), not `9352123^`. Also correct a claim this PR made. `make generated/go/v2` does not work after restoring: it goes through `build` -> `schema`, which wants v3 too and still hits the CAPTCHA — confirmed with `make -n generated/go/v2`. Generating v2 means calling the executable directly, and the Makefile comment, AGENTS.md and ADR-0003 now all say so. Restore all eight files rather than six. readme.txt and readme2.txt are in the deleted tree and in BUILD.bazel's onix2p1 filegroup, and they are the files that carry EDItEUR's copyright notice. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01P8rZakwAiz1jpViUX34x1Z
editeur.org が自動リクエストに CAPTCHA を返すため
make schemaが機能せず、コード生成が止まっています(詳細は #58)。2.1 についてはこれを回避できます。足元にありました
コミット 9352123 は、ビルド時ダウンロード方式に移行する際に
2_1_rev03_schema/を削除しました。その親コミットには XSD 6 本がそのまま残っています。git showで、ネットワークを一切使わずに取り出せます。make schema-from-historyを追加しました。手元での実行結果です。計 1.8 MB、
src/Lib.hsが読むschema/v2/ONIX_BookProduct_Release2.1_reference.xsdを含みます。schema/%は prerequisite を持たないパターンルールなので、ディレクトリが存在すれば make はダウンロードを試みません。EDItEUR のファイルはコミットしていません
これは意図的です。
schema/は.gitignore済みなので、このリポジトリが配布するものは何も変わりません。追加しているのは自前の make ターゲットだけで、第三者の著作物ではありません。したがって #58 が提起しているライセンスの論点には触れません。復元されるファイルは、この変更の有無にかかわらず 2021 年からこのリポジトリの公開履歴に存在しています。
3.0 / 3.1 には効きません
履歴にあるのは 2.1 だけです。最新スキーマへの追従には、依然として #58 の判断(vendoring / ミラー /
--distdir)が必要です。確認したこと・していないこと
make -n schemaが v2 を要求しなくなることを実行して確認🤖 Generated with Claude Code
https://claude.ai/code/session_01P8rZakwAiz1jpViUX34x1Z
Generated by Claude Code