The mixed `number | bigint` representation was useful optimization for
pointers. Now, pointers are represented as V8 externals. As part of the
FFI stabilization effort we want to make `bigint` the only
representation for `u64` and `i64`.
BigInt representation performance is almost on par with mixed
representation with the added benefit that its less confusing and users
don't need manual checks and conversions for doing operations on the
value.
```
cpu: AMD Ryzen 5 7530U with Radeon Graphics
runtime: deno 1.43.6+92a8d09 (x86_64-unknown-linux-gnu)
file:///home/divy/gh/ffi/main.ts
benchmark time (avg) iter/s (min … max) p75 p99 p995
-------------------------------------------------------------------------- -----------------------------
nop 4.01 ns/iter 249,533,690.5 (3.97 ns … 10.8 ns) 3.97 ns 4.36 ns 9.03 ns
ret bigint 7.74 ns/iter 129,127,186.8 (7.72 ns … 10.46 ns) 7.72 ns 8.11 ns 8.82 ns
ret i32 7.81 ns/iter 128,087,100.5 (7.77 ns … 12.72 ns) 7.78 ns 8.57 ns 9.75 ns
ret bigint (add op) 15.02 ns/iter 66,588,253.2 (14.64 ns … 24.99 ns) 14.76 ns 19.13 ns 19.44 ns
ret i32 (add op) 12.02 ns/iter 83,209,131.8 (11.95 ns … 18.18 ns) 11.98 ns 13.11 ns 14.5 ns
```
This is a primordialization effort to improve resistance against users
tampering with the global `Object` prototype.
---------
Co-authored-by: Bartek Iwańczuk <biwanczuk@gmail.com>
Follow up to #22157.
This leaves us with 4 usages of `ensureFastOps()` in `deno` itself.
There's also about 150 usages of `Deno.core.ops.<op_name>` left as well.
`opAsync` requires a lookup by name on each async call. This is a
mechanical translation of all opAsync calls to ensureFastOps.
The `opAsync` API on Deno.core will be removed at a later time.
This commit refactors how we access "core", "internals" and
"primordials" objects coming from `deno_core`, in our internal JavaScript code.
Instead of capturing them from "globalThis.__bootstrap" namespace, we
import them from recently added "ext:core/mod.js" file.
We can move all promise ID knowledge to deno_core, allowing us to better
experiment with promise implementation in deno_core.
`{un,}refOpPromise(promise)` is equivalent to
`{un,}refOp(promise[promiseIdSymbol])`
Migrate to op2. Making a few decisions to get this across the line:
- Empty slices, no matter where the come from, are null pointers. The v8
bugs (https://bugs.chromium.org/p/v8/issues/detail?id=13489) and
(https://bugs.chromium.org/p/v8/issues/detail?id=13488) make passing
around zero-length slice pointers too dangerous as they might be
uninitialized or null data.
- Offsets and lengths are `#[number] isize` and `#[number] usize`
respectively -- 53 bits should be enough for anyone
- Pointers are bigints. This is a u64 in the fastcall world, and can
accept Integer/Int32/Number/BigInt v8 types in the slow world.
This commit renames "deno_core::InternalModuleLoader" to
"ExtModuleLoader" and changes the specifiers used by the
modules loaded from this loader to "ext:".
"internal:" scheme was really ambiguous and it's more characters than
"ext:", which should result in slightly smaller snapshot size.
Closes https://github.com/denoland/deno/issues/18020
This API is required by several extensions like "ext/node", "ext/ffi"
and also FS APIs that we want to move to a separate crate. Because
of that "pathFromURL" API was moved to "deno_web" extension so
other extension crates can rely on it.
This PR refactors all internal js files (except core) to be written as
ES modules.
`__bootstrap`has been mostly replaced with static imports in form in
`internal:[path to file from repo root]`.
To specify if files are ESM, an `esm` method has been added to
`Extension`, similar to the `js` method.
A new ModuleLoader called `InternalModuleLoader` has been added to
enable the loading of internal specifiers, which is used in all
situations except when a snapshot is only loaded, and not a new one is
created from it.
---------
Co-authored-by: Bartek Iwańczuk <biwanczuk@gmail.com>
Updated third_party dlint to v0.37.0 for GitHub Actions. This PR
includes following changes:
* fix(prefer-primordials): Stop using array pattern assignments
* fix(prefer-primordials): Stop using global intrinsics except for
`SharedArrayBuffer`
* feat(guard-for-in): Apply new guard-for-in rule
Adds support for passing and returning structs as buffers to FFI. This does not implement fastapi support for structs. Needed for certain system APIs such as AppKit on macOS.
Makes `op_ffi_ptr_of` fast. One of the tests changed from printing
`false` to `true` as the fast `&[u8]` slice path creates the slice with
a null pointer. Thus the `op_ffi_ptr_of` will now return a null pointer
value whereas previously it returned a dangling pointer value.
This PR makes pointer read methods of `Deno.UnsafePointerView` Fast API
compliant, with the exception of `getCString` which cannot be made fast
with current V8 Fast API.
Fixes #15136
Currently `UnsafeCallback` class' `ref()` and `unref()` methods rely on
the `event_loop_middleware` implementation in core. If even a single
`UnsafeCallback` is ref'ed, then the FFI event loop middleware will
always return `true` to signify that there may still be more work for
the event loop to do.
The middleware handling in core does not wait a moment to check again,
but will instead synchronously directly re-poll the event loop and
middlewares for more work. This becomes a live-loop.
This PR introduces a `Future` implementation for the `CallbackInfo`
struct that acts as the intermediary data storage between an
`UnsafeCallback` and the `libffi` C callback. Ref'ing a callback now
means calling an async op that binds to the `CallbackInfo` Future and
only resolves once the callback is unref'ed. The `libffi` C callback
will call the waker of this Future when it fires to make sure that the
main thread wakes up to receive the callback.
The return type checking for `"function"` type FFI values was incorrect
and presumed that functions were still being registered as objects
containing a "function" key.
While here, I also removed the whole return type checking logic as it
was needed for optionally creating BigInts on return when needed, but
serde_v8 does this automatically now (I think).
Welcome to better optimised op calls! Currently opSync is called with parameters of every type and count. This most definitely makes the call megamorphic. Additionally, it seems that spread params leads to V8 not being able to optimise the calls quite as well (apparently Fast Calls cannot be used with spread params).
Monomorphising op calls should lead to some improved performance. Now that unwrapping of sync ops results is done on Rust side, this is pretty simple:
```
opSync("op_foo", param1, param2);
// -> turns to
ops.op_foo(param1, param2);
```
This means sync op calls are now just directly calling the native binding function. When V8 Fast API Calls are enabled, this will enable those to be called on the optimised path.
Monomorphising async ops likely requires using callbacks and is left as an exercise to the reader.
This commit adds support for unstable FFI
callbacks. A callback is registered using
the `Deno.UnsafeCallback` API.
The backing memory for the callback can
be disposed of using `Deno.UnsafeCallback#close`.
It is not safe to pass the callback after calling
close.
Callbacks from other than the isolate thread
are not supported.
Co-authored-by: Divy Srivastava <dj.srivastava23@gmail.com>
Co-authored-by: Bert Belder <bertbelder@gmail.com>
