In order to uses less space and improve the readability.
zig fmt allows only 1 switch case per line or all in one line.
When having a lot of conditions, splitting the line is useful.
Virtually all string are page-owned (in the page.arena), but because of the
Small String Optimization we use (string.zig), a string could be stack-allocated
The correct solution is probably to change the key to be a string.String. But
I want to give more thought to memory in general, and strings specifically need
to be thought about. So this is a quick fix for crashing.
It now returns a Caught struct which contains all information. The Caught struct
can be logged directly, providing more consistent logs for caught errors.
We previously treated v8::Object and v8::Values interchangeably, and would just
ptrCast one to the other. So, if an API was defined with a js.Object but was
given a non-object value, e.g. 9001, it would still work.
This has since been tightened. If an API takes a js.Object, than the v8 value
must be an object. Passing a non-object will result in a InvalidArgument error.
CustomEvent.detail and PerformanceMark.detail can both be any value, so the
apis/fields have been updated from js.Object -> js.Value.
On CDP.BrowserContext.deinit, clear the isolated world ExecutionContext before
terminating the session. This is important as the isolated_world list is
allocated from the session.arena.
Also, semi-revert 63f1c85964. Before all this
we were running microtasks on ExecutionWorld.removeContext. That didn't seem
right (and I thought it was the original source of the bug). But, for the "real"
Page context, this is critical, since Microtasks can reference the Page object.
Since microTasks are isolation-level, it's possible for a microtasks for Page1
to execute after Page1 goes away (if we create a new page, Page2). This re-adds
the microtask "draining", but only for the Page (i.e. in Page.deinit).
This crashes linux in releasesafe without an embedded snapshot. Not sure why,
but it shouldn't be necessary. This was added back when we were executing
microtasks on a schedule, rather than manually at explicit points.
