Repository navigation
Description
Report
Open https://chromewebstore.google.com/ and click any extension card. The whole app crashes. Reported by @meinharrd on 2026-09-11 against the shipped build (platform not yet recorded here, presumably the macOS 0.8.5 package).
What I could not reproduce (Linux x64, xvfb, real network, real nodes)
Every run below loaded the store, dismissed Google's EU consent page and the "Switch to Chrome?" popup, then clicked the first in-viewport card (Google Translate). The app stayed up and the detail page rendered (screenshot below).
| build | click style | result |
|---|---|---|
main d69affb, from source |
real mouse click via host window | detail page, no crash |
main, from source |
in-page a.click() (SPA route) |
detail page, no crash |
main, from source |
middle-click | opened in new tab, no crash |
main, from source |
Ctrl+click | opened in new tab, no crash |
main, from source |
address-bar navigation to the detail URL | detail page, no crash |
0.8.5 39f1cd5, from source |
real mouse click | detail page, no crash |
main packaged (electron-builder --linux --x64, linux-unpacked/freedom) |
real mouse click | detail page, no crash |
Electron 44.3.0 / Chromium 152 in all runs. No render-process-gone, child-process-gone or uncaught exception in the main-process log during the click. The only main-process crash seen is at quit and is unrelated to this page: #345.
So on Linux the click path is fine. The crash is platform- or environment-specific, and the next step needs data from the machine that crashes.
Needed from the reporting machine
- OS and architecture, Freedom version (About), and whether it is the DMG/DEB/AppImage/NSIS install or a source run.
- The main-process log: macOS
~/Library/Logs/Freedom/main.log, Linux~/.config/Freedom/logs/main.log, Windows%APPDATA%\Freedom\logs\main.log. The last ~50 lines before the crash. - The OS crash report. macOS: the newest
Freedom-*.ipsin~/Library/Logs/DiagnosticReports/(Console.app → Crash Reports). Itsterminationblock and the crashed thread's stack tell us whether this is a main-process abort (like ipfs: main process aborts (SIGABRT) on quit — napi_throw fatal in freedom-ipfs-event-worker #345'snapi_throw), a ChromiumCHECK, or a GPU/renderer death that took the window with it. - Whether it also happens with the wallet locked / a fresh profile (
Profiles → New), and with adblock off. That separates profile state from the page itself. - Whether navigating straight to a detail URL (e.g.
https://chromewebstore.google.com/detail/google-translate/aapbdbdomjkkjkaonfhkkikfgjllcleb) crashes too, or only the click from the store front.
Note that Freedom never calls crashReporter.start(), so there are no minidumps of our own under crash-reports/; the OS report is the only trace.
Also observed on this page (not the crash, worth their own issues if confirmed)
webview-preloadfails to inject the Swarm and Radicle providers on Google's pages:Failed to set the 'textContent' property on 'HTMLScriptElement': This document requires 'TrustedScript' assignment. Any site with a Trusted Types CSP gets nowindow.swarm/window.radicle. Caught, not fatal.[adblock-update] feed read failed: Cannot read properties of undefined (reading 'makeReader')a few seconds after the page loads. The filter engine itself reported ready.- Google's own bundle logs
Uncaught (in promise) Error: wdafter the SPA route; harmless.
Repro driver used
Playwright _electron launcher, FREEDOM_TEST_MODE unset (live network), FREEDOM_TEST_USER_DATA pointed at a scratch dir; stdout/stderr of the Electron process captured to a file; click delivered with win.mouse.click(webviewRect.x + cardRect.x, webviewRect.y + cardRect.y) after scrollIntoView inside the guest. The ELECTRON_ENABLE_LOGGING=1 stream is what shows a crash if one happens.
Activity
meinharrd commentedon Sep 11, 2026
Reproduced on macOS, root-caused: Electron 44 exposes
chrome.webstorePrivateto the store pageEnvironment: macOS 26.6.2 (25G83), arm64,
/Applications/Freedom.app0.8.5 (DMG install), Electron 44.3.0 / Chromium 152.0.7977.78, fresh scratch profile (FREEDOM_TEST_USER_DATA), live network. Driver: Playwright_electronwithexecutablePathon the installed app./detail/google-translate/aapbdbdomjkkjkaonfhkkikfgjllclebdelete chrome.webstorePrivateexecuted in the page firstelectron .from the source tree with the Electron 43.4.1 innode_modulesFresh profile crashes too, so profile state, wallet lock state and adblock are ruled out. The three
Freedom-*.ipsreports from the original machine have the identical stack.Crash report
EXC_BAD_ACCESS (SIGSEGV),KERN_INVALID_ADDRESS at 0x0, faulting threadCrBrowserMain(browser/main process),x0 = 0. Symbolicated with Electron's published breakpad symbols for v44.3.0 (Electron FrameworkUUID4C4C44E3-5555-3144-A193-1D6578CCBF56matches):Chromium 152
extensions/browser/api/webstore_private/webstore_private_api.cc:1440:Why
Since the Chromium 152 roll the
webstorePrivateAPI lives in//extensionsinstead of//chrome, so Electron 44 compiles and registers it and ships the core_api_features.jsonentry that exposes it to ordinary web pages onhttps://chromewebstore.google.com/*. Electron provides noWebstorePrivateAPIDelegate, so the firstchrome.webstorePrivate.*call from the page is a null-pointer method call in the browser process. Probing the store page in 0.8.5 confirms it gets a realchrome.webstorePrivatewith 19 functions (getReferrerChain,install,beginInstallWithManifest3,getExtensionStatus, ...), all configurable/writable; the store callsgetReferrerChainas soon as a detail page is shown. Not Freedom code.Upstream fix: electron/electron#53752 "fix: don't expose chrome.webstorePrivate to web pages" (merged to
main2026-09-09, backported to 45-x-y). The 44-x-y backport electron/electron#53776 is still open, so no released Electron 44.x contains it yet. Electron 43.x predates the roll and is unaffected, matching the source control run above.About the Linux non-repro
The repo's
node_modules/electronis 43.4.1 whilepackage-lock.jsonpins 44.3.0. Afrom sourcerun or anelectron-builderpackage built from stalenode_modulesuses 43.4.1, which does not expose the API. Worth re-checking the Linux runs withprocess.versions.electronfrom inside the app rather than the lockfile.Options
electrononce a 44.x release with #53776 ships (or move to 45 when stable).chromewebstore.google.com/chrome.google.com/webstore, runwebFrame.executeJavaScript('delete chrome.webstorePrivate')at document start. Deleting the property is exactly what kept the app alive in the control run, andexecuteJavaScriptis not subject to Google's Trusted Types CSP that blocks the current<script>provider injection.[-]crash: chromewebstore.google.com → click any extension → hard crash (reported on 0.8.5; not reproducible on Linux)[/-][+]crash: chromewebstore.google.com detail page → SIGSEGV in browser process (Electron 44 exposes chrome.webstorePrivate without a delegate)[/+]Merge pull request #384 from solardev-xyz/alan/346-electron-44-4-1-we…
fix: bump Electron to 44.4.1 so the Chrome Web Store no longer crashe…