It has more context than the env about when this should be called. Specifically
it can be called once per session, whereas, in the env, we can only call it
once per context - which could be too often.
A loop interval will no longer stop the loop from returning from `run`, and
no longer requires mutating event_nb on each iteration.
Re-enable microtask loop, which I accidentally stopped in a previous commit.
NamedFunction is important for displaying good error messages when there's
something wrong with the Zig structs we're trying to bind to JS. By making it
a normal struct, it's easier and cheaper to pass wherever an @compileError
might be needed.
When the browser microtask was added, zig-specific timeout functions were
added to the loop. This was necessary for two reasons:
1 - The existing functions were JS specific
2 - We wanted a different reset counter for JS and Zig
Like we did in https://github.com/lightpanda-io/browser/pull/577, the loop is
now JS-agnostic. It gets a Zig callback, and the Zig callback can execute JS
(or do whatever). An intrusive node, like with events, is used to minimize
allocations.
Also, because the microtask was recently moved to the page, there is no longer
a need for separate event counters. All timeouts are scoped to the page.
The new timeout callback can now be used to efficiently reschedule a task. This
reuses the IO.completion and Context, avoiding 2 allocations. More importantly
it makes the internal timer_id static for the lifetime of an "interval". This
is important for window.setInterval, where the callback can itself clear the
interval, which we would need to detect in the callback handler to avoid
re-scheduling. With the stable timer_id, the existing cancel mechanism works
as expected.
The loop no longer has a cbk_error. Callback code is expected to try/catch
callbacks (or use callback.tryCall) and handle errors accordingly.
Previously, we were passing our WebAPIs directly as an anonymous tuple. This
resulted in Env(T) having an _awful_ name - a name composed of hundreds of
classes.
By wrapping the anonymous tuple into a normal struct, the Env now gets a sane
name which helps improve stack traces (and profiling, and debugging, ...)
There's ambiguity in mapping due to the flexible nature of JavaScript. Hopefully
most types are unambiguous, like a string or am *parser.Node.
We need to "probe" each field to see if it's a possible candidate for the JS
value. On a perfect match, we stop probing and set the appropriate union field.
There are 2 levels of possible matches: candidate and coerce. A "candidate"
match has higher precedence. This is necessary because, in JavaScript, a lot
of things can be coerced to a lot of other, seemingly wrong, things.
For example, say we have this union:
a: i32,
b: bool,
Field `a` is a perfect match for the value 123. And field b is a coerce match
(because, yes, 123 can be coerced to a boolean). So we map it to `a`.
Field `a` is a candidate match for the value 34.2, because float -> int are both
"Numbers" in JavaScript. And field b is a coerce match. So we map it to `a`.
Both field `a` and field `b` are coerce matches for "hello". So we map it to `a`
because it's declared first (this relies on how Zig currently works, but I don't
think the ordering of type declarations is guaranteed, so that's an issue).
cancel on linux was a "real" cancel, but the implementation was unsafe. It took
whatever `id` it was given and @ptrFromInt'd it. This is problematic since the
`id` is user-supplied with virtually no validation.
Using the existing MacOS canceled lookup seems both easier and safer than trying
to validate the cancellation id.
- Pages within the same session have proper isolation
- they have their own window
- they have their own SessionState
- they have their own v8.Context
- Move inspector to CDP browser context
- Browser now knows nothing about the inspector
- Use notification to emit a context-created message
- This is still a bit hacky, but again, it decouples browser from CDP
The approach borrows heavily from Zig's new LinkedList API.
The main benefit is that it unifies how event callbacks are done. When the
Page.windowClick event was added, the Event structure was changed to a union,
supporting a distinct Zig and JS event.
This new approach more or less treats everything like a Zig event. A JS event
is just a Zig struct that has a Env.Callback which it can invoke in its handle
method.
The intrusive nature of the EventNode means that what used to be 1 or 2
allocations is now 0 or 1.
It also has the benefit of making netsurf completely unaware of Env.Callbacks.
- Fix get_removedNodes (it was returning addedNodes)
- get_removedNodes and get addedNodes now return references
- used enum for dispatching and clean up dispatching in general
- Remove MutationRecords and simply return an array
- this allows the returned records to be iterable (as they should be)
- jsruntime ZigToJs will now map a Zig array to a JS array
-Rely on default initialize of NodeList
-Batch observed records
- Callback only executed when call_depth == 0
- Fixes crashes when a MutationObserver callback mutated the nodes being
observes.
- Fixes some WPT issues, but Netsurf's mutationEventRelatedNode does not
appear to be to spec, so most tests fail.
- Allow zig methods to execute arbitrary code when call_depth == 0
- This is a preview of how I hope to make XHR not crash if the CDP session
ends while there's still network activity
The call_arena is still re-added, and the call_depth is still used, but
the call_arena and scope_arenas are no longer part of the Env - they remain on
the Executor.
This is to accommodate upcoming changes where multiple executors will exist at
once. While the shared allocators _might_ have been safe in some cases, the
performance gains don't justify the risk of having 2 executors interacting in a
way where sharing the allocators would cause issues.
Because of callbacks, calls into Zig can be nested. Previously, the call_arena
was reset after _every_ call. When calls are nested, this doesn't work - the
nested call resets the arena, which the caller might still need. A `call_depth`
integer was added to the Executor. Each call starts by incrementing the
call_depth and, on deinit, decrements the call_depth. The call_arena is only
reset when the call_depth == 0. This could result in lower memory use.
Also promoted the call_arena and scope_arena to the Env. Practically speaking,
nothing has changed, since they're still reset under the exact same conditions.
However, when an executor ends and a new one is started, it can now reuse the
retained_capacity of the previous arenas. This should result in fewer
allocations.
Some APIs need a null-terminated string. Currently, they have to ask for a
`[]const u8` and then convert it to a `[:0]const u8`. This is 2 allocations: 1
for jsruntime to get the `[]const u8` from v8, and then one to get the [:0]. By
supporting `[:0]const u8` directly, this is now a single allocation.
JsObject can now be used as a normal parameter. It'll receive the opaque value.
This is largely useful when a Zig function takes an argument which it needs
to pass back into a callback.
JsThis is now a thin wrapper around JsObject for functions that was the JsObject
of the receiver. This is for advanced usage where the Zig function wants to
manipulate the v8.Object that represents the zig value. postAttach is an example
of such usage.
Add value setter, keys(), values(), entries() and forEach().
Like nodelist, forEach still doesn't support `this` arg (gotta think about
how to do this).
I think these iterable methods are missing in a few places, so I added a
generic Entries iterator and a generic Iterable.
jsruntime will now map a Zig tuple to a JS array.
TypeMeta constains the index and the subtype. This allows retrieving the subtype
based on the actual value being bound, as opposed to the struct type. (I.e. it
returns the correct subtype when a Zig class is a proxy for another using the
Self declaration).
Internally store the subtype as an enum. Reduces @sizeOf(TaggedAnyOpaque) from
32 to 16.
Finally, sub_type renamed to subtype for consistency with v8.
