🍕 Stop component client.js from running in Kiln edit mode

[jjpaulino]

TL;DR — for reviewers short on time

• Risk: medium — this changes which bootstrap every edit-mode page loads. View mode is byte-for-byte unchanged, and a public/js built by an older claycli degrades to today's behavior rather than failing.
• If you only review one thing: lib/cmd/vite/index.js → getViteEditBootstrapScripts L83–L101 — the entry-point swap and the stale-manifest fallback. Everything else follows from it.
• Changes, in logical review order (each link jumps to where to start reading):
  1. lib/cmd/vite/generate-bootstrap.js → buildInitPrelude L208–L265 — foundational. Separates the initializer prelude, which edit mode genuinely needs, from the mount runtime, which it must not run. No behavior change on its own.
  2. lib/cmd/vite/generate-bootstrap.js → generateViteBootstrap L267–L355 — writes two entries from that prelude: vite-bootstrap.js (view, same output as before) and the new vite-bootstrap-no-mount.js (edit).
  3. lib/cmd/vite/scripts.js → withNoMountBootstrap L666–L684 — adds the new entry to the view Rollup graph and the watcher, so it gets a content hash and shares the initializer chunks instead of duplicating them.
  4. lib/cmd/vite/index.js → resolveModuleScripts L133–L165 — serves the no-mount entry in edit mode, and preloads the entry it actually loads rather than the view bootstrap.
  5. lib/cmd/vite/index.test.js L229 and generators.test.js L139 — new coverage for the swap, the fallback, and the generated no-mount content.
  6. CLAY-VITE.md L450 — documents the parity guarantee. Docs only.

---

Feature Info

Jira Ticket Url

• None yet. This was reported in Slack while validating 6.0.0-rc.4 on dev — happy to file a ticket and add the ID to the title if we want one on the board.

Description

Component client.js files were executing in Kiln edit mode under clay vite. They never did under clay compile.

getViteViewScripts() returns the hashed view bootstrap, and resolveModuleScripts() put it first in edit mode ([...viewScripts, ...editScripts]). That bootstrap calls mountComponentModules() at module scope, so the moment the module evaluates it scans the DOM and imports a client.js chunk for every component on the page.

The legacy pipeline never did this: the edit branch of getDependencies() resolves model.js, kiln.js, templates and kiln plugins, but no client dependency graph and no _client-init.js — the runtime that mounts controllers. Editors only ever ran component controllers on rendered pages.

The bootstrap was included in edit mode because its _env-init.js import has to run before the kiln edit bundle. That is a side effect of bundling env init together with component mounting; env init is the only part edit mode needs.

So the fix is a second entry rather than dropping the bootstrap. Both are generated from the same prelude — the window.modules stub Kiln's preloader reads, _env-init.js, _globals-init.js, and the sticky-events shim — and only the view entry gets _clayClientModules plus the mount runtime.

No opt-in flag. An earlier revision of this branch had a mountInEditMode option. It's gone: mounting in edit mode has no legitimate use, it was never possible under clay compile, and a component that wants client-rendered output visible while editing should provide that through kiln.js rather than by running every ad call, analytics beacon and comment embed on the page inside the editing surface. So this needs no resolve-media.js change — sites get the fix on upgrade.

QA Testing Notes

• Load any article with ?edit=true and confirm no component client.js side effects: no GPT ad iframes injected into slots, no Coral comment embed, no analytics beacons.
• Confirm edit mode still works end to end — Kiln boots, components are selectable, model.js and kiln plugins still read env (the prelude is what makes that work, so a regression here would show up as Kiln failing outright rather than subtly).
• Load the same page without ?edit=true and confirm view mode is unchanged: ads, comments and analytics all mount.
• clay vite --only js then check public/js/_manifest.json has both .clay/vite-bootstrap and .clay/vite-bootstrap-no-mount.

Validation Points

• Stale public/js: a manifest built before this change has no no-mount entry. getViteEditBootstrapScripts() falls back to the mounting bootstrap, which reintroduces the bug but keeps Kiln working — serving no initializers at all would break editing outright. Re-run clay vite to pick up the fix. This is the one case where edit mode can still mount, and it's a build-artifact mismatch rather than a configuration choice.
• Chunk duplication: the no-mount entry rides in the same Rollup pass as the view bootstrap, so the shared initializer modules are emitted once and both entries import them.
• Watch mode: the new entry is in the watcher's input map, so it rebuilds with the view bootstrap on client.js and global/js add/unlink.
• Preloads: edit mode now preloads the bootstrap it actually loads. Previously it preloaded the view bootstrap while loading it too, so this removes a wasted hint rather than adding one.
• npm test (lint + jest): 447 tests pass, including 6 new.

---

Related: clay/clay-kiln#1575 hardens addIframeOverlays so Kiln can't produce a page-covering overlay regardless of where an iframe came from. This PR removes the trigger; that one removes the failure mode. They're independent and can land in either order.

[coveralls]

coverage: 87.018% (-0.07%) from 87.084% — jordan/edit-no-mount into master