The Sentry `Dedupe` integration has been filtering out our events, even
when they were never sent due to our `beforeSend` handler. It was
wrongly identifying them as duplicates because it has no knowledge of
`beforeSend` or whether they were actually sent or not.
To resolve this, the filtering we were doing in `beforeSend` has been
moved to a Sentry integration. This integration is installed ahead of
the `Dedupe` integration, so `Dedupe` should never find out about any
events that we filter out, and thus will never consider them as sent
when they were not.
* Don't send errors to sentry if users have not opted-in to participate in metametrics
* Don't capture opt-out metrics
* Move the metrics-opt in screen to immediately after the welcome screen
* Ensure that global.getSentryState is set in the background
* Fix e2e tests after rearranging onboardin flow
* Fix unit tests
* More e2e test fixes
* Remove unnecessary wrappers around capture exception
Certain build steps accidentally omitted the `version` variable. It has
now been restored to all steps, ensuring that all environment variables
are correctly injected into all bundles.
A check has been added to the Sentry setup module to ensure the release
is not omitted in the future.
The `environment` field we use for Sentry includes the build type for
all build types except `main`, but the log message indicating that
Sentry did not include this. This log message is useful for ensuring
that Sentry is setup correctly, so it should display the same
environment that Sentry is using. It has been updated to do just that.
The build type (i.e. the distribution) is now included in the Sentry
environment during setup, for all builds except the "main" build. This
will allow us to track Flask and beta errors separately from other
errors.
A constant was created for the build types. The equivalent constant in
our build scripts was updated to match it more closely, for
consistency. We can't use the same constant in both places because our
shared constants are in modules that use ES6 exports, and our build
script does not yet support ES6 exports.
The singular `BuildType` was used rather than `BuildTypes` to match our
naming conventions elsewhere for enums. We name them like classes or
types, rather than like a collection.
Relates to #11896
Sentry is now configured with environment variables, rather than with
hard-coded values. This makes it easier to test Sentry functionality
using a different Sentry account, as we did recently during QA of
v9.5.1.
The only change for the normal build process is the introduction of the
`SENTRY_DSN_DEV` variable, which can be set via `.metamaskrc` or via an
environment variable. This determines where error reports are sent. It
still defaults to our team Sentry account's `metamask-testing` project.
The `sentry:publish` script now requires SENTRY_ORG and SENTRY_PROJECT
to be set in order to publish release artifacts. The CircleCI
configuration has been updated with these values, so it should act the
same as it did before. Previously we had used a CLI flag to specify the
organization and project, but Sentry already natively supports these
environment variables [1].
[1]: https://docs.sentry.io/product/cli/configuration/#configuration-values
* Remove network config store
* Remove inline networks variable in network controller
* Re-key network controller 'rpcTarget' to 'rpcUrl'
* Require chainId in lookupNetwork, implement eth_chainId
* Require chain ID in network form
* Add alert, migrations, and tests
* Add chainId validation to addToFrequentRpcList
* Update public config state selector to match new network controller
state
* Use network enums in networks-tab.constants
* Ensure chainId in provider config is current
* Update tests
The Sentry DSN is now expected to be provided via environment variable
for production builds. The build script will fail if it is missing, and
an error will be thrown at runtime if it is missing.
The `SENTRY_DSN` environment variable has been set in CI to the old
value for `SENTRY_PROD_DSN`. We can migrate to a new DSN at some point
in the future.
In a non-production environment, Sentry was configured to send error
reports to a "test" MetaMask project. It will still do this during e2e
tests, but in development Sentry is now disabled completely.
In practice this was never useful in development.
The `extra` property of errors sent to Sentry is sometimes not
initialized when we add the application state. A check has been added
to initialize it if it's missing.
I suspect that this changed with v5 of `@sentry/browser`, though I
can't find any explicit confirmation of this in their changelog.
The state snapshot that was attached to Sentry errors was removed
recently in #8794 because it had become too large. The snapshot has
now been restored and reduced in size.
A utility function has been written to reduce the state object to just
the requested properties. This seemed safer than filtering out state
that is known to be large or to contain identifiable information.
This is not a great solution, as now knowledge about the state shape
resides in this large constant, but it will suffice for now. I am
hopeful that we can decorate our controllers with this metadata in the
future instead, as part of the upcoming background controller refactor.
A separate `getSentryState` global function has been added to get the
reduced state, so that the old `getCleanAppState` function that we used
to use could remain unchanged. It's still useful to get that full state
copy while debugging, and in e2e tests.
The state snapshot we were attaching to Sentry errors was too large.
As a temporary solution, it has been removed completely. We can re-add
it later after reducing its size.
Previously all browser globals were allowed to be used anywhere by
ESLint because we had set the `env` property to `browser` in the ESLint
config. This has made it easy to accidentally use browser globals
(e.g. #8338), so it has been removed. Instead we now have a short list
of allowed globals.
All browser globals are now accessed as properties on `window`.
Unfortunately this change resulted in a few different confusing unit
test errors, as some of our unit tests setup assumed that a particular
global would be used via `window` or `global`. In particular,
`window.fetch` didn't work correctly because it wasn't patched by the
AbortController polyfill (only `global.fetch` was being patched).
The `jsdom-global` package we were using complicated matters by setting
all of the JSDOM `window` properties directly on `global`, overwriting
the `AbortController` for example.
The `helpers.js` test setup module has been simplified somewhat by
removing `jsdom-global` and constructing the JSDOM instance manually.
The JSDOM window is set on `window`, and a few properties are set on
`global` as well as needed by various dependencies. `node-fetch` and
the AbortController polyfill/patch now work as expected as well,
though `fetch` is only available on `window` now.
Errors without stack traces would break the Sentry error processing,
which assumes the presence of a stack trace. Many errors don't have any
stack trace though, such as uncaught promises.
This breakage resulting in the app state being missing from the error
report, and a console warning.
Any error sent to Sentry will now be marked with the environment they
were sent from. The environment is set at build time, and is set
dependant upon the build flags and CI-related environment variables.
Setting the environment will let us filter error reports in Sentry to
focus specifically upon reports sent from production, release
candidates, PR testing, or whatever else.
Previously, all errors encountered during testing or production were
sent to the primary `metamask` Sentry project, whereas development
errors were sent to `test-metamask` instead. This change ensures that
errors encountered during tests are sent to `test-metamask` as well.
