From f932a1491bbcbed5ede5ae6b8d04d496553c6363 Mon Sep 17 00:00:00 2001
From: Glenn Gore <glenn.g@affinidi.com>
Date: Tue, 15 Sep 2026 14:15:51 +0200
Subject: [PATCH] fix(extension): emit no modulepreload links, which Chrome
 refuses to use
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Every generated wallet page carried `<link rel="modulepreload" crossorigin>`
for each shared chunk its entry imports, and Chrome declined all of them:

    A preload for 'chrome-extension://…/assets/bridge-protocol-….js' is
    found, but is not used because it is a cross-world extension resource
    mismatch.

The hint is fetched under a different resource world than the module graph
that `<script type="module">` then loads, so the preloaded copy never matches
the lookup and is discarded — each chunk fetched twice, and 18 warnings
across popup (4), options (6), confirm (4) and offscreen (4) on the
chrome://extensions Errors page.

Nothing was broken by them; the script tag is what actually loads the graph.
The cost is the page they filled. The Errors page is where a real fault has
to be noticed, and a screenful of benign warnings is how one gets scrolled
past — the same reasoning R7.2 applies to a lost prompt.

Preloading buys nothing here anyway: every asset is read from local disk,
with no network latency to hide. `manager.html` never had any, because
`codeSplitting: false` leaves the console no shared chunks to preload, and
`background.js` is a single bundle with no HTML at all — so this is the one
config that needed it.

Verified by building both ways: 18 links before, 0 after, with the entry's
own `<script type="module" crossorigin>` unchanged. Full three-config build,
726 extension tests and the dist guards (single background bundle, admin and
persona URIs confined to manager.js, no seeds family, one manager chunk) all
pass.

Signed-off-by: Glenn Gore <glenn.g@affinidi.com>
---
 packages/extension/vite.config.ts | 17 +++++++++++++++++
 1 file changed, 17 insertions(+)

diff --git a/packages/extension/vite.config.ts b/packages/extension/vite.config.ts
index ec98efa..e1817e9 100644
--- a/packages/extension/vite.config.ts
+++ b/packages/extension/vite.config.ts
@@ -43,6 +43,23 @@ export default defineConfig({
     // to vite's default low targets (chrome87/es2020). Safe for an MV3
     // extension, whose runtime is a current Chromium.
     target: "es2022",
+    // No `<link rel="modulepreload">` in the emitted HTML. Chrome refuses to
+    // use them here — "A preload for '…' is found, but is not used because it
+    // is a cross-world extension resource mismatch" — so each preloaded chunk
+    // was fetched once as a discarded hint and again for real, and the
+    // chrome://extensions Errors page filled with one warning per chunk per
+    // page (18 across popup, options, confirm and offscreen).
+    //
+    // Nothing was broken by them: the `<script type="module">` is what
+    // actually loads the graph. The cost is that the Errors page is where a
+    // *real* fault has to be noticed, and a page of benign warnings is how one
+    // gets scrolled past — the same reasoning as R7.2 for lost prompts.
+    //
+    // Preloading buys nothing in an extension anyway: every asset is read from
+    // local disk, not a network with latency to hide. `manager.html` never had
+    // these, because `codeSplitting: false` leaves it no shared chunks to
+    // preload.
+    modulePreload: false,
     rollupOptions: {
       input: {
         popup: resolve(__dirname, "popup.html"),