Adding `#[inline]` will typically improve codegen for optimized builds
without LTO (so far the majority in practice) by allowing functions that
otherwise couldn't be inlined across codegen units to get inlined
across codegen units.
Right now `wasm-bindgen` has a lot of functions that are very small and
delegate to other functions, but aren't otherwise candidates for
inlining because they're concrete.
I was poking around in release-mode wasm recently and noticed an
alarming number of functions for tiny pieces of functionality, which
motivates this patch!
Previously the `link_mem_intrinsics` hack actually had a runtime
overhead by storing a value into a global location, but it turns out we
can actually use a non-inlined function call as part of the *descriptor*
which requires this to be in the final binary, but we'll end up snip'ing
the value at the end.
All in all this should mean that it's not a zero-overhead solution for
linking these intrinsics! The `#[wasm_bindgen]` attribute already has
other problems if the descriptors don't show up, so that's the least of
our issues!
This commit removes the need for an injected `ConstructorToken` type and
also cleans up the story we have for generating constructors a bit.
After this commit a `constructor()` is omitted entirely if we're in
non-debug mode and there's no actual listed constructor. Additionally we
don't deal with splat arguments and rerouting constructors, Nick was
kind enough to enlighten me about `Object.create` which is creating an
instance without running the constructor!
Instances of an exported type are now created through one of two
methods:
* If `#[wasm_bindgen(constructor)]` is present, then a `constructor` is
generated with the appropriate signature. If a constructor is not
present and we're in debug mode, a throwing constructor is generated.
If we're in release mode and there's no constructor, no constructor is
generated.
* Otherwise if a binding returns an instance of a type (or otherwise
needs to manfuacture an instance, then it will cause an internal
`__wrap` function to be generated. This function will use
`Object.create` to create an instance without running the constructor.
This should ideally clean up our generated JS for classes quite a bit,
making it much more lean-and-mean!
If the setter doesn't start with `set_*` then we currently panic, but
panicking is bad! Instead let's thread through structured errors to make
sure they make their way to the top
This commit adds support for exporting a function defined in Rust that returns a
`Result`, translating the `Ok` variant to the actual return value and the `Err`
variant to an exception that's thrown in JS.
The support for return types and descriptors was rejiggered a bit to be a bit
more abstract and more well suited for this purpose. We no longer distinguish
between functions with a return value and those without a return value.
Additionally a new trait, `ReturnWasmAbi`, is used for converting return values.
This trait is an internal implementation detail, however, and shouldn't surface
itself to users much (if at all).
Closes#841
This is intended to address #834 where we don't actually want methods scoped
like this! Instead we'll provide one unique accessor for the `window` object
itself.
This is a roundabout way to say that this addresses the last comment on #23,
namely if you only use the `console` submodule from `web_sys` it doesn't
actually link correctly!
The problem here has to do with codegen units and the compiler. The compiler
will create a codegen unit for each `mod` in the source code. If a codegen unit
isn't actually used, then the codegen unit is removed from the final link step.
This causes problems for web-sys where the JSON description of our program was
part of the main CGU but not in each submodule, so when submodules were only
used the descriptor program in the main CGU was not included.
The fix in this commit is to instead generate a descriptor program in the
submodule itself instead of leaving it in the main CGU. By removing the `Module`
node in the AST this naturally happens as the descriptor is only generated in
the same module as all other associated items.
Any LongLong still present after flattening now gets translated to a `f64` type
so we can bind these types. While not a true integral value or truely 64-bits of
integer precision, it's all JS has anyway!
Instead of actually modifying the `FirstPassRecord` let's instead just skip
relevant entries when we come across them. This should help us retain knowledge
that `optional SomeImportedType arg` can be bound even though `SomeImportedType`
may not exist.
One small tweak was needed to modify the AST afterwards to remove `extends`
annotations which aren't actually defined, but other than that this should...
Closes#802
* Gate `web-sys` APIs on activated features
Currently the compile times of `web-sys` are unfortunately prohibitive,
increasing the barrier to using it. This commit updates the crate to instead
have all APIs gated by a set of Cargo features which affect what bindings are
generated at compile time (and which are then compiled by rustc). It's
significantly faster to activate only a handful of features vs all thousand of
them!
A magical env var is added to print the list of all features that should be
generated, and then necessary logic is added to ferry features from the build
script to the webidl crate which then uses that as a filter to remove items
after parsing. Currently parsing is pretty speedy so we'll unconditionally parse
all WebIDL files, but this may change in the future!
For now this will make the `web-sys` crate a bit less ergonomic to use as lots
of features will need to be specified, but it should make it much more
approachable in terms of first-user experience with compile times.
* Fix AppVeyor testing web-sys
* FIx a typo
* Udpate feature listings from rebase conflicts
* Add some crate docs and such
This commit adds further support for the `Global` attribute to not only emit
structural accessors but also emit functions that don't take `&self`. All
methods on a `[Global]` interface will not require `&self` and will call
functions and/or access properties on the global scope.
This should enable things like:
Window::location() // returns `Location`
Window::fetch(...) // invokes the `fetch` function
Closes#659
I discussed this with @fitzgen awhile back and this sort of casting seems
especially problematic when you have code along the lines of:
let mut x: HtmlElement = ...;
{
let y: &mut JsValue = x.as_ref();
*y = 3.into();
}
x.some_html_element_method();
as that will immediately throw! We didn't have a use case for mutable casting
other than consistency, so this commit removes it for now. We can possibly add
it back in later if motivated, but for now it seems reasonable to try to avoid
these sorts of pitfalls!
This commit adds support for generating bindings for dictionaries defined in
WebIDL. Dictionaries are associative arrays which are simply objects in JS with
named keys and some values. In Rust given a dictionary like:
dictionary Foo {
long field;
};
we'll generate a struct like:
pub struct Foo {
obj: js_sys::Object,
}
impl Foo {
pub fn new() -> Foo { /* make a blank object */ }
pub fn field(&mut self, val: i32) -> &mut Self {
// set the field using `js_sys::Reflect`
}
}
// plus a bunch of AsRef, From, and wasm abi impls
At the same time this adds support for partial dictionaries and dictionary
inheritance. All dictionary fields are optional by default and hence only have
builder-style setters, but dictionaries can also have required fields. Required
fields are exposed as arguments to the `new` constructor.
Closes#241
