Publish opted-in marimo notebooks at /notebooks
Notebooks from the marimo server can now appear in the garden. A notebook
opts in with an HTML comment in one of its markdown cells — invisible when
rendered, greppable in the .py source:
<!-- garden:publish
title: Dream of Spotification
order: 30
-->
Default-deny on purpose: garden.c0smere.net is public and the export bakes
each notebook's executed output into the page, not just its code.
export-notebooks.py also carries a NEVER_PUBLISH list (genome_*, coursework)
so a marker pasted into one of those refuses loudly instead of publishing.
Pipeline: export-notebooks.py runs each marked notebook in a one-shot
marimo container (same image/env/GPU as the live server, so it hits the real
databases) into notebooks-export/ + index.json; copy-notebooks.mjs stages
those to public/nb/; Astro reads index.json to build the pages and the
sidebar section.
Isolation choices worth keeping:
- Own timer and own lock, separate from the 5-min garden build — executing a
notebook takes minutes and must never hold up a build tick.
- Cached on notebook content; both index.json and .cache.json go through
write_if_changed, since auto-build.sh hashes mtimes under notebooks-export/
and an unconditional rewrite would force a full rebuild every 30 minutes.
- Per-notebook timeout; a failure keeps the previous export and continues.
- Runs against a throwaway copy of the notebook dir, so notebooks that write
scratch files don't dirty the notebooks repo.
- Notebooks are NOT injected into the garden collection — they aren't vault
notes, and doing so would move noteCount and the sitemap.
build.format:'file' makes the listing a file (notebooks.html) beside a
directory of detail pages. Verified against the running nginx: /notebooks and
/notebooks/<slug> resolve through the generic try_files but /notebooks/ does
not, so nginx.conf gets an explicit location for the trailing-slash form.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
This commit is contained in:
@@ -21,12 +21,15 @@ flock 9
|
||||
# a dirty tree can make pull fail; still rebuild whatever is checked out
|
||||
git pull -q || echo "git pull failed (dirty tree?) — building local state"
|
||||
|
||||
# signature covers committed HEAD, any uncommitted repo edits, and vault mtimes
|
||||
# signature covers committed HEAD, any uncommitted repo edits, vault mtimes,
|
||||
# and the notebook exports (gitignored, so git status/diff can't see them —
|
||||
# they land here on their own timer via deploy/export-notebooks.py)
|
||||
sig=$({
|
||||
git rev-parse HEAD
|
||||
git status --porcelain=v1
|
||||
git diff
|
||||
find "$VAULT" -name .obsidian -prune -o -type f -printf '%T@ %p\n' | sort
|
||||
find "$REPO/notebooks-export" -type f -printf '%T@ %p\n' 2>/dev/null | sort
|
||||
} | sha256sum | cut -d' ' -f1)
|
||||
|
||||
if [ "$FORCE" -eq 0 ] && [ -f "$STATE" ] && [ "$(cat "$STATE")" = "$sig" ]; then
|
||||
|
||||
Reference in New Issue
Block a user