This isn't officially compatible with the oldest version of Node that
Jasmine supports, but it works. If it stops working, we can always disable
linting in CI builds on older Node versions.
* Generally simplifies error handling in browsers
* Makes Jasmine's own integration tests easier to debug
* Stack traces will be provided for more global errors
* ... but less error information will be provided in some browsers if the
error comes from a file:// URL (use `npx serve` or similar instead)
* Jasmine will no longer override existing onerror handlers in browsers
* Setting window.onerror will no longer override Jasmine's global error
handling (use jasmine.spyOnGlobalErrors instead)
This was already done for everything except spec cleanup fns, since the
various skip policies need to know the difference between afterEach and
afterAll.
This should allow us to more easily support complex skipping strategies
like skipping nested cleanup fns when the corresponding befores were
skipped, or skipping specs and suites when a beforeAll fails.
* #1533
This allows assertion failures and other errors that occcur after the async
error to be routed to the correct spec/suite.
Previously, Jasmine treated global errors and unhandled promise rejections
just like exceptions thrown from a synchronous spec: it recorded the error
as a spec failure and moved on. Now, global errors and uhandled rejections
are recorded as failures but the current queueable fn will continue until
it either signals completion or times out. Global errors and unhandled
rejections are different from synchronous exceptions: it's common for the
queueable fn that caused them to continue executing. Immediately moving on
often meant that the queueable fn would produce expectation failures or
other errors when a different spec or suite was running, thus causing
those failures to be routed to the wrong place.
This reduces the risk of incorrectly passing a spec due to not correctly
detecting that an argument is an `Error` instance. Detecting Error instances
in a way that's reliable and portable across different browsers, TrustedTypes,
and frames is difficult.
[Finishes #178267587]
* De-duplication now happens in core, not in reporters. This ensures that
the console doesn't get flooded.
* Stack traces are opt-out, not opt-in.
* The current runnable is not reported or logged for certain deprecations
where it's irrelevant.
* HtmlReporter shows stack traces in expandable widgets.
* Env#deprecated and Env#deprecatedOnceWithStack are merged.
The first number is the error message in HTML5 browser, which does not include
the call stack. The error instance allows logging the complete call stack in
reporters.
Note: afterEach() functions are still run, because skipping them is
highly likely to pollute specs that run after the failure.
[Finishes #92252330]
- Fixes#577
- Fixes#807
Besides surfacing the error in the hopefully-correct place, this also
prevents the queue runners for sibling suites from interleaving, which
in turn prevents all kinds of internal state corruption.
Signed-off-by: Gregg Van Hove <gvanhove@pivotal.io>
