[juliasadler]

What this changes

• detectPages in scripts/publish.mjs: when two --pages entries resolve to the same key (same data-theme, or the same filename slug), the later entry now falls back to its filename slug — with a numeric suffix if that is taken too — instead of silently overwriting the earlier one. Both pages end up recorded under distinct keys.
• The inline slug expression is now a slugOf helper, plus a freeSlug helper that finds an unclaimed key. No code comments, per repo rule.
• Regression test in scripts/publish.test.mjs with a two-same-theme fixture, as suggested in the issue.
• CHANGELOG entry under [Unreleased] → Fixed.

Why

Fixes #89. Before, --pages main.html,alt.html with both files at data-theme="dark" produced figure.pages = { "dark": "src/alt.html" }: main.html vanished as a page and shipped as a dead file under dest/src. After, it produces { "dark": "src/main.html", "alt": "src/alt.html" }.

I took the issue's first suggested direction (slug fallback) rather than throwing: it records both pages, as the issue's "Expected" section asks, and it doesn't turn a previously silent success into a hard error.

Checks

• Every example rebuilds, and git status --short shows only the changes I meant. — git status --short shows only CHANGELOG.md, scripts/publish.mjs, scripts/publish.test.mjs. No example touched, so no figure rebuild applies; node --test scripts/publish.test.mjs passes 8/8 including the new regression test.
• No comments in code.
• Kit changes are opt-in, and every existing example is byte-identical — N/A, no kit change.
• I built and looked at both themes. — N/A, no visual change (publish script only).
• Figures with pipes, cables or round parts … — N/A.
• node scripts/lines.mjs … — N/A.
• Shaders … — N/A.
• 3D … — N/A.
• Every number in the figure … is true. — N/A, no figure numbers changed.
• A line under Unreleased in CHANGELOG.md.

Screenshots

N/A — no visual change.

[rahul05ranjan]

Clean fix for a real footgun — the old pages[...] = page silently ate the first page whenever two shared a theme, and the recording would just point at the later file.

I traced the fallback chain and it holds together: first same-theme page keeps the theme key, the colliding one drops to its filename slug, and a slug that's also taken gets -2, -3, … via freeSlug, so a third dark page or an unlucky slug collision can't silently overwrite either. Object.hasOwn over in is the right call too, since it ignores prototype keys.

One pre-existing edge this inherits rather than introduces: a page literally named dark.html with no theme slugs to the key dark, which consumers likely treat as the dark-theme entry. Before your change it collided head-on; now it survives as dark-2 if a themed page got there first, which is better but still slightly ambiguous. Probably not worth solving unless someone hits it — the numeric suffix at least makes the two pages distinct now.

The test covers the real case well, including the assertion that the overwritten page no longer ships as a stray source file — that's the user-visible half of the bug, easy to forget.