Summary
28 files under _site/ are committed to main and are not gitignored, while docs/ --- the directory _quarto.yml actually renders into --- is gitignored. So the repo tracks stale build output and ignores the real build output, exactly backwards.
Verified on a clean worktree off origin/main (63648a2):
$ git ls-files _site | wc -l
28
$ git log -1 --format='%h %ci %s' -- _site
a8eb6c8 2024-12-11 20:34:18 -0800 more
$ git check-ignore -v _site
(no output -- not ignored)
$ grep -n 'docs' .gitignore
6:docs/
The tracked copy is from 2024-12-11, roughly twenty months out of date. It contains index.html, listings.json, papers/listing.html, and a set of people/*/ images, none of which reflect the current site.
Why this is worth fixing rather than ignoring
On its own this is untidy: dead files, and a diff that occasionally shows build output. That alone would not justify an issue.
What makes it worth filing is that it converts a loud failure into a silent wrong one in the preview migration tracked by #102. Morrison-Lab/gha's preview.yml renders according to _quarto.yml (so, into docs/) but stages its artifact from a hardcoded _site/. With no _site/ present that mismatch fails immediately at ls _site/. With this stale _site/ present it passes, and the PR preview publishes the 2024 site.
That gap is filed upstream as Morrison-Lab/gha#608 and is the primary blocker on #102. Deleting the tracked _site/ does not fix that gap --- it fixes the part where the gap is invisible. Both are worth doing, and this one is entirely within this repo.
What to do
git rm -r --cached _site and commit the deletion.- Add
_site/ to .gitignore, next to the existing docs/ entry, so a local render with the default output directory does not re-add it.
- Confirm nothing references
_site/ --- notably publish.yml's folder: docs/ and preview.yml's source-dir: ./docs/, both of which already point at docs/ and so are unaffected.
Scope
Small and self-contained; no dependency on #102 or #103, and worth landing whether or not either migration proceeds.
Summary
28 files under
_site/are committed tomainand are not gitignored, whiledocs/--- the directory_quarto.ymlactually renders into --- is gitignored. So the repo tracks stale build output and ignores the real build output, exactly backwards.Verified on a clean worktree off
origin/main(63648a2):The tracked copy is from 2024-12-11, roughly twenty months out of date. It contains
index.html,listings.json,papers/listing.html, and a set ofpeople/*/images, none of which reflect the current site.Why this is worth fixing rather than ignoring
On its own this is untidy: dead files, and a diff that occasionally shows build output. That alone would not justify an issue.
What makes it worth filing is that it converts a loud failure into a silent wrong one in the preview migration tracked by #102.
Morrison-Lab/gha'spreview.ymlrenders according to_quarto.yml(so, intodocs/) but stages its artifact from a hardcoded_site/. With no_site/present that mismatch fails immediately atls _site/. With this stale_site/present it passes, and the PR preview publishes the 2024 site.That gap is filed upstream as Morrison-Lab/gha#608 and is the primary blocker on #102. Deleting the tracked
_site/does not fix that gap --- it fixes the part where the gap is invisible. Both are worth doing, and this one is entirely within this repo.What to do
git rm -r --cached _siteand commit the deletion._site/to.gitignore, next to the existingdocs/entry, so a local render with the default output directory does not re-add it._site/--- notablypublish.yml'sfolder: docs/andpreview.yml'ssource-dir: ./docs/, both of which already point atdocs/and so are unaffected.Scope
Small and self-contained; no dependency on #102 or #103, and worth landing whether or not either migration proceeds.