V3 Beta - errors not reported from preload context

[timfish]

Versions + Platform

• SDK version - @sentry/[email protected]
• Electron version - all
• Platform - all

Description

With v3 beta, errors are not reported from the preload context.

[timfish]

I'm nowhere near a machine to test this but you can probably work around this by initialising the render code in the preload too:

import '@sentry/electron/preload';
Import { init } '@sentry/electron/renderer';

init();

If you're using context isolation, you will need to init the renderer code in the preload AND the isolated renderer context.

[mehrdadmms]

Thank you. It's finally working.

For anyone who's looking at this issue in the future this is the code that worked for me.
I'm using Electron + Vue2.

Preload

require('@sentry/electron/preload');
const { init } = require('@sentry/electron/renderer');

init({
  dsn: process.env.SENTRY_DSN,
  environment: process.env.NODE_ENV,
  release: pJson.version
});

Main

import * as Sentry from '@sentry/electron/main';

Sentry.init({
  dsn: process.env.SENTRY_DSN,
  environment: process.env.NODE_ENV,
  release: pJson.version
});

Renderer

import Vue from 'vue'
import * as Sentry from '@sentry/electron'
import * as SentryIntegrations from '@sentry/integrations'

Sentry.init({
  dsn: process.env.SENTRY_DSN,
  environment: process.env.NODE_ENV,
  release: pJson.version,
  integrations: [
    new SentryIntegrations.Vue({
      Vue,
      logErrors: true,
      tracing: false,
    }),
  ],
})

[timfish]

It's worth noting that as all events are sent from the main process, you do not need to supply the dsn, release and environment in the preload and renderer.

[timfish]

I think reporting errors from the preload context should work out of the box by default but this adds some annoying complications that vary by the users Electron configuration.

For non-bundled apps, we automatically inject the preload script which is currently well under 1kB in size. If we want to support reporting errors from this context with zero config, we'll need to include all the renderer code which is currently ~78kB. If we include this code by default, we'll also need to call init and users will not be able to change the settings passed to init. In the preload context, passing options to init might not be a huge issue except...

If users have contextIsolation disabled (not the default and not recommended, but very common), the preload and renderer context are not distinct so once the SDK has been init'ed in the preload, it's effectively setup in the renderer and there is no way to pass option to init.

Why not let users just create their own preload script and choose if they want to support error reporting from this context and pass custom init options? Well, if sandbox mode is enabled (planned to become default), the preload has no access to require, so preload scripts will need to be bundled!

To summarise

• If we init the renderer SDK in the preload, users with contextIsolation: false cannot pass options to init
• If we don't init the renderer SDK in the preload, users that want errors from preload context will need to use a bundler if sandbox: true

[timfish]

This is simpler in 3.0.0-beta.2

A preload is no longer required unless you're using Electron < v5 and bundling the main process.

If you would like errors reported from the preload then you'll still need to initialise the renderer portion of the sdk there. No options are required unless you need to configure specific renderer integrations.

preload.js

const { init } = require('@sentry/electron/renderer');
init();

If you have context isolation enabled, the renderer is a different context to the preload so you'll need to init the sdk again there.

[its-tim-lee]

Hey, @timfish
So did you mean that, if I want to use Sentry in three independent contexts (ie., renderer, preload, main), and if i have enabled both contextIsolation and sandbox, as long as my Sentry configurations are the same in three contexts, i don't need to configure anything in init at the contexts of renderer and preload, since you metioned:

It's worth noting that as all events are sent from the main process, you do not need to supply the dsn, release and environment in the preload and renderer.

But I guess that I still need to configure the one in renderer, since that's the instruction from the relevant Sentry doc, that I
need to integrate @sentry/vue by doing something like:

  // this is in the renderer context
  Sentry = require("@sentry/electron/renderer")
  Sentry.init(
    /** my setting */,
    require("@sentry/vue").init
  )

Am i right?
Currently, I found that if I provide setting to the Sentry.init in the preload, I'll get an error, but the error will disappear if I don't provide any setting.

BTW,
I'm using @sentry/electron v4.20.0, @sentry/vue v7.105.0, and electron v29.1.1