…into store-event-state-in-main-process

The test was supposed to check total length == matching length. Along
the way fixed some incorrect assertions around API key and hashing.

Some tests may not fire event during launch window

The actual net package on windows does not like `0.0.0.0` as localhost

The Breadcrumb.toJSON method renames the internal properties of a
breadcrumb object. This is fine, but with some changes the toJSON
method stops being called during serialisation. When this happens,
the breadcrumbs are silently dropped because the 'name' is undefined
and the core Client drops breadcrumbs without a name/message

By avoiding toJSON in all cases, this should be a bit more robust

Avoid Breadcrumb.toJSON by copying breadcrumbs to an object before sending them over IPC

Co-authored-by: Ben Gourley <[email protected]>

Add browser onerror & unhanded rejection plugins to renderers

Add console, network and interaction breadcrumb plugins

On windows the build fails intermittently in the publish step with:

    lerna ERR! ZlibError: zlib: invalid block type
    lerna ERR!     at Unzip.write (D:\a\bugsnag-electron\bugsnag-electron\node_modules\minizlib\index.js:147:22)
    lerna ERR!     at Object.write (D:\a\bugsnag-electron\bugsnag-electron\node_modules\tar\lib\parse.js:305:58)
    lerna ERR!     at Yallist.<anonymous> (D:\a\bugsnag-electron\bugsnag-electron\node_modules\minipass\index.js:374:18)
    lerna ERR!     at Yallist.forEach (D:\a\bugsnag-electron\bugsnag-electron\node_modules\minipass\node_modules\yallist\yallist.js:156:8)
    lerna ERR!     at ReadStream.emit (D:\a\bugsnag-electron\bugsnag-electron\node_modules\minipass\index.js:373:20)
    lerna ERR!     at ReadStream.emit (D:\a\bugsnag-electron\bugsnag-electron\node_modules\fs-minipass\index.js:154:22)
    lerna ERR!     at ReadStream.write (D:\a\bugsnag-electron\bugsnag-electron\node_modules\minipass\index.js:163:17)
    lerna ERR!     at ReadStream.[_handleChunk] (D:\a\bugsnag-electron\bugsnag-electron\node_modules\fs-minipass\index.js:131:19)
    lerna ERR!     at ReadStream.[_onread] (D:\a\bugsnag-electron\bugsnag-electron\node_modules\fs-minipass\index.js:109:32)
    lerna ERR!     at D:\a\bugsnag-electron\bugsnag-electron\node_modules\fs-minipass\index.js:101:22
    lerna ERR! lerna zlib: invalid block type

macOS expands to find all available space and time

We aren't even using this, removing for now to unblock the build

This commit introduces an new plugin whose job it is to manage the lifecycle
of an event created in the renderer. This ensures internal onError callbacks
in main populate the event's data fields so that all data is present for
onError callbacks in a renderer.

Co-authored-by: Delisa <[email protected]>

Allow Act to run the tests with Ubuntu

Co-authored-by: Joe Haines <[email protected]>

Co-authored-by: Joe Haines <[email protected]>

Co-authored-by: Joe Haines <[email protected]>

The more tests we have, the more likely it is that we'll see at least
one unhandled rejection timeout error from deep in playwright resulting
from calling ElectronApplication.close() and looking something like:

  TimeoutError: Timeout 30000ms exceeded.
    at ProgressController.run (server/progress.js:75:30)
    at ElectronApplication._waitForEvent (server/electron/electron.js:99:35)
    at ElectronApplication.close (server/electron/electron.js:90:29)
    at ElectronApplicationDispatcher.close (dispatchers/electronDispatcher.js:59:28)
    at DispatcherConnection.dispatch (dispatchers/dispatcher.js:178:52)

This change bypasses the close function altogether, instead killing the
app and closing the node process connection manually.

Manually it works, but for whatever reason, the request numbers differ
in the test?

Add plugin-electron-preload-error

Add plugin-electron-net-breadcrumbs

e.g. https://github.com/bugsnag/bugsnag-electron/actions/runs/744197591

A payload can be delivered multiple times if it is sent once, then
stopped, then started again before it is delivered

This will result in two scheduled deliveries — the original one and
a new one kicked off by 'start'

The second can then peek the same payload the first is sending because
payloads are only removed from the queue when they have been sent
successfully

Upload cucumber failure logs when tests fail

Fix possible double delivery

Don't start delivery until the app is ready

Electron's default exception handler doesn't exit like Node's. Instead,
it displays an error dialog:
https://github.com/electron/electron/blob/24e21467b91b4facb554cbbbc32a13ed4426f25b/lib/browser/init.ts#L44-L61

Our default onUncaughtException should therefore not exit and instead
just log the error

Use app path over CWD for the default project root

Don't exit in the default onUncaughtException

Each use of this was removed in separate PRs, so it only started
failing in master

Remove unused process import

This ensures we don't store or return payloads that will never be
possible to send. These payloads can create a loop of failures as
the same payload will be retried every delivery attempt, but will
fail if missing required attributes

Co-authored-by: Joe Haines <[email protected]>

Validate payloads before queuing/returning them from queue

It's only in the Node notifier we can't use this and have to use
the prototype, since client.leaveBreadcrumb() is monkeypatched.

refactor: Rename electron plugins and add unit tests

…ag-electron into bengourley/plugin-refactor

…ent-data

Node's stream.write can fail synchronously if something is very wrong
with the passed data, e.g. if it's null or undefined

In this case we'll never be able to deliver the event/session because
the body is broken, so now we won't attempt to retry these errors

In theory this should never happen, but bugs elsewhere in delivery
could trigger this

Don't retry delivery if writing the body fails

The metadata replacement event arguments changed recently, so updated the
status checker and the tests to handle this case in the future.

Our onBreadcrumb callbacks are added to the front of the list of
callbacks. This would usually mean they were also called first, which
would be problematic as then cancelled breadcrumbs would still be
synced. However, we run breadcrumb callbacks in reverse order (last
to first) so they are actually called _last_. This allows breadcrumbs
to be cancelled normally while also cancelling the syncing

Verify onBreadcrumb syncing can be cancelled

Some mac builds now require 6+ minutes to do an installation

https://github.com/bugsnag/bugsnag-electron/runs/2378942400

For whatever reason, only on mac, installing electron pulls in the other
deps, and downloading the test browsers takes several minutes.

Add integration test for cancelling breadcrumbs