That way it does not depend on the return value for the same type.
This hopefully makes the code more clear, and avoids errors if something
changes in the future (for example, we could want to pass slices as
sequence arguments).
The function do_create_interface_objects is removed in favour of 4 functions:
create_callback_interface_object, create_interface_prototype_object,
create_noncallback_interface_object and create_named_constructors.
While this increases the amount of codegen'd code, this greatly improves the
readability of the code involved in this part of DOM, instead of having one function
doing 4 different things. We can always find a more adequate abstraction later.
NativeProperties and everything related to the interface objects have been removed
from the utils module.
This makes codegen not emit anything strictly related to the interface which
are never used in the case of abstract interfaces, such as the Wrap method
or the DOMJSClass.
This is mostly stolen from Gecko. As there, we define the unforgeable members
on an object stored in the slots of the prototype object. They are then copied
onto instance objects when they are instantiated. It should be noted that
proxy objects see their unforgeable memebers defined on their expando object.
Unforgeable attributes aren't properly inherited in codegen (in a similar
fashion as getters and setters as filed in #5875) and require to be redefined
in derived interfaces. Fortunately, there are currently no such interfaces.
No unforgeable members can be included into the TestBinding interfaces for good
measure because they are not compatible with setters.
Given the unforgeable holder object has the same prototype as actual instances
of the interface, the finalize hook needs to check its slot pointer for nullity
before dropping it.
The new failing test isn't related to Unforgeable attributes, but to the fact
that all Document instances currently have a Location, even if their window
isn't in a browsing context.
Opaque DOMString
This patch makes DOMString an opaque wrapper round String (currently it's a transparent wrapper).
The changes are:
* Replacing DOMString(foo) by DOMString::from(foo).
* Replacing foo.0 by String::from(foo).
* Adding functions clear, push_str and extend for in-place mutation of DOMStrings.
* Replacing DOMString by String in other threads (devtools, storage and filereader).
* Making DOMString implement !Send.
* Removing the pub attribute from the contents of DOMString.
This enables experimenting with other string representations in the DOM.
<!-- Reviewable:start -->
[<img src="https://reviewable.io/review_button.png" height=40 alt="Review on Reviewable"/>](https://reviewable.io/reviews/servo/servo/8477)
<!-- Reviewable:end -->
Functions returning `Root<T>` are prefixed by "root_" and the ones returning
`*const T` by "native_".
Functions taking `*mut JSObject` are now suffixed by "_from_object" and the ones
taking `&T` by "_from_reflector".
Remove unnecessary uses of #[no_move]
The patch makes RootCollection a bit safer by making the StackRootTLS hold
it in place.
RootedVec was doing an extremely delicate dance and just hoping nobody
messed it up; switch to a Box to be safe.
CodeGenRust seemed to be using no_move for no particularly good reason.
<!-- Reviewable:start -->
[<img src="https://reviewable.io/review_button.png" height=40 alt="Review on Reviewable"/>](https://reviewable.io/reviews/servo/servo/8286)
<!-- Reviewable:end -->
The patch makes RootCollection a bit safer by making the StackRootTLS hold
it in place.
The use of no_move in CodeGenRust was leftover from when roots couldn't
be moved.
This should make it somewhat easier to experiment with alternative
representations in the future. To reduce churn, this commit leaves the String
field public, though.
Also, this will allow us to use the default String type to represent the IDL
USVString type, which explicitly forbids unpaired surrogates, ans as such is
a better match to the Rust String type.
unblock using JS_GC_ZEAL
I think these patches move the `JS_SetReservedSlot` call to the right place for #6057. I'm not sure that the interface to `create_dom_global` is the best; passing a `JSVal` or a `*libc::c_void` seemed about equal, so I'd welcome feedback there.
<!-- Reviewable:start -->
[<img src="https://reviewable.io/review_button.png" height=40 alt="Review on Reviewable"/>](https://reviewable.io/reviews/servo/servo/8226)
<!-- Reviewable:end -->
Moving this slot setting prior to initializing standard classes and
other initialization tasks means that we can effectively use JS_GC_ZEAL.
Fixes#6057.
We're going to wind up tacking on different pieces of code for the
global vs. non-global case once we move the JS_SetReservedSlot call into
create_dom_global. We might as well separate the indentation changes
into a separate commit.
generate JIT argument type information for methods
This enhances `CodegenRust.py` to output `JSTypedMethodJitInfo` structures where appropriate. This brings a notable speedup to tests like Dromaeo's `dom-attr/getAttribute`, which improves by several orders of magnitude with these patches applied.
If there are tricks for addressing the XXX comments, I would appreciate hearing them.
I think this addresses all of #6904.
<!-- Reviewable:start -->
[<img src="https://reviewable.io/review_button.png" height=40 alt="Review on Reviewable"/>](https://reviewable.io/reviews/servo/servo/8040)
<!-- Reviewable:end -->
