WEBDEV-9066: Guard custom element registration against duplicate defines

[jbuckner]

Importing two element subpaths on the same page throws and one of them never registers. Repros on the published 0.4.0 in headless Chromium:

Failed to execute 'define' on 'CustomElementRegistry': the name "ia-status-indicator" has already been used with this registry

Each subpath esm.archive.org serves is its own self-contained bundle, so a shared component is built into every entry point that imports it. Two subpaths means two copies and two customElements.define() calls for the same tag. The second throws mid-evaluation, which aborts the rest of that bundle, so it's partway through: some tags from the failed bundle register and the rest are lost. Bundler consumers like offshoot never see it because Vite dedupes to one instance.

12 tags are defined by more than one entry point. ia-status-indicator is in ia-button, ia-dropdown-search-bar, ia-otp-form, ia-radio-player and its own subpath, so any two of those collide.

So this adds src/util/custom-element.ts, a @customElement that skips the define when the tag is already claimed, and points every call site at it. First copy wins. That's safe here: nothing does instanceof on a shared element class and the only class-level imports of one are import type, so the copies are interchangeable. There's an eslint rule to keep Lit's version from coming back.

The demo and story files are swapped too. They aren't published so they can't hit this, but leaving them on the Lit decorator just invites someone to copy the wrong import into a new element.

What I didn't do

There's a second, structural fix: a single-URL entry point would give non-bundler consumers one module instance and no duplicates at all. That's a bigger change and it argues with #81, which drops the package root. Skipping it for now since petabox usage is getting deprecated anyway. The guard is worth having on its own for any non-bundler consumer.

Verifying

Unit tests cover the guard, including that registration keeps working after a duplicate is skipped, which is the part that actually breaks.

The real failure mode needs two separately-built bundles, which no unit test can produce, so I checked that by hand: esbuild dist/.../ia-dropdown-search-bar and dist/.../ia-radio-player into two self-contained bundles and load both on one page.

• Without the guard: radio-player throws, doesn't register, doesn't upgrade, and ia-transcript-view and ia-waveform-progress are lost with it.
• With it: both import, all six tags register, both elements attach a shadow root.

694 tests pass, lint 0 errors, prettier clean.

https://webarchive.jira.com/browse/WEBDEV-9066

🤖 Generated with Claude Code

https://claude.ai/code/session_01PoknqYwGpzkxFE3kfcYW81

[github-actions]

PR Preview Action v1.8.1
Preview removed because the pull request was closed.
2026-09-16 17:14 UTC

[codecov-commenter]

Codecov Report

❌ Patch coverage is 87.50000% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 85.80%. Comparing base (631605d) to head (e1df5b2).
⚠️ Report is 1 commits behind head on main.

Files with missing lines	Patch %	Lines
src/util/custom-element.ts	87.50%	0 Missing and 1 partial ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #106   +/-   ##
=======================================
  Coverage   85.79%   85.80%           
=======================================
  Files          74       75    +1     
  Lines        2619     2627    +8     
  Branches      577      579    +2     
=======================================
+ Hits         2247     2254    +7     
  Misses        195      195           
- Partials      177      178    +1     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[latonv]

LGTM, just had one thought about possible logging when we skip a define.