feat(cli): add migrate-images for KBs ingested before note-relative links#182
feat(cli): add migrate-images for KBs ingested before note-relative links#182Aldominguez12 wants to merge 2 commits into
Conversation
…inks Follow-up to VectifyAI#181: already-ingested KBs keep wiki-root-relative sources/images/... links in their wiki/sources/*.md pages, which render broken in Obsidian, GitHub, and VS Code. `openkb migrate-images` rewrites them to the note-relative images/... form now emitted at ingest (`--dry-run` to preview). Scope matches the writers VectifyAI#181 changed: only .md files directly under wiki/sources/ — long-doc JSON metadata intentionally keeps wiki-root-relative paths, and pages outside sources/ never carried the old prefix. Idempotent, takes the ingest lock, and logs to wiki/log.md. Verified against two real pre-VectifyAI#181 KBs (32 files, 2,061 links): output is byte-identical to a hand-verified migration, and a second run is a no-op. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 7c11d107dd
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| verb = "Would rewrite" if dry_run else "Rewrote" | ||
| click.echo(f"{verb} {total} image link(s) across {len(changed)} file(s).") | ||
| if not dry_run: | ||
| append_log(wiki, "migrate-images", f"{total} link(s) across {len(changed)} file(s)") |
There was a problem hiding this comment.
When openkb migrate-images runs concurrently with another mutating command, this append_log executes after the exclusive kb_ingest_lock block has already been released. Since append_log writes wiki/log.md, this leaves a wiki write outside the repo's locking invariant from AGENTS.md and can interleave with another process's locked log update; keep the log append inside the same exclusive lock as the file rewrites.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Good catch — fixed in 29da160. The append_log call now runs inside the same kb_ingest_lock block as the file rewrites, matching the pattern used by the lint command. tests/test_migrate.py still passes (11/11).
Codex review flagged that append_log ran after kb_ingest_lock was released, leaving a wiki/log.md write outside the locking invariant. Move it inside the exclusive block, matching the lint command. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
1 participant
Follow-up to #181, as offered there and welcomed in the review.
Problem
KBs ingested before #181 keep wiki-root-relative
sources/images/...links in theirwiki/sources/*.mdpages. Theread_wiki_imagefallback keeps agent flows working, but renderers (Obsidian, GitHub, VS Code) still resolve those links against the containing file and show every figure broken.Fix
New one-time migration command:
](sources/images/prefix to](images/— the formmd_image_ref()emits at ingest since fix(images): write note-relative image links in sources pages #181. Only the prefix is swapped, so path tails never need parsing and image embeds / plain links are handled alike..mdfiles directly underwiki/sources/. Long-doc per-page JSON intentionally keeps wiki-root-relative paths (internal metadata), and pages outsidesources/never carried the old prefix.wiki/log.md— same conventions aslint --fix.openkb/migrate.pyso the command doesn't drag inpymupdfviaimages.py.Verification
tests/test_migrate.py: 11 tests covering rewrite/alt-preservation, idempotency, dry-run, JSON and non-sources pages left untouched, sorted multi-file output, and the CLI surface (no KB, nothing to migrate, apply + log, dry-run).ruff check/formatandmypyclean on the touched files.🤖 Generated with Claude Code