Additionally, use the second bit (instead of the first) to differentiate
between strings and symbols there. This will allow transparent
conversion of DFS to StringBase in the future.
When compiling code like this:
x = { foo: x }
We don't want to put a new JS::Object in `x` until *after* we've
evaluated `x` for the `foo` field.
This fixes an issue when loading https://puter.com/ :^)
This works very similarly to MarkedVector<T>, but instead of expecting
T to be Value or a GC-allocated pointer type, T can be anything.
Every pointer-sized value in the vector's storage will be checked during
conservative root scanning.
In other words, this allows you to put something like this in a
ConservativeVector<Foo> and it will be protected from GC:
struct Foo {
i64 number;
Value some_value;
GCPtr<Object> some_object;
};
Instead of having Call refer to a range of VM registers, it now has
a trailing list of argument operands as part of the instruction.
This means we no longer have to shuffle every argument value into
a register before making a call, making bytecode smaller & faster. :^)
By handling common cases like Int32 arithmetic directly in the
instruction handler, we can avoid the cost of calling the generic helper
functions in Value.cpp.
Instead of splitting the postfix variants into ToNumeric + Inc/Dec,
we now have dedicated PostfixIncrement and PostfixDecrement instructions
that handle both outputs in one go.
This adapts our implementation to the editorial change in the temporal
proposal: 737baf2d
The changes to CalendarMethodsRecordLookup had already been implemented,
but we had followed the typo in the spec for CalendarMethodsRecordCall.
The issue in CalendarMethodsRecordCall hasn't surfaced yet, as the AOs
using Calendar Methods Record are currently not passing through a String
to represent a Calendar builtin.
No change to test-262.
This number is pure guesswork but it appears to fix GCC builds with
both ASAN and UBSAN hitting a native stack overflow before we have
a chance to catch it on our Azure CI.
This patch moves us away from the accumulator-based bytecode format to
one with explicit source and destination registers.
The new format has multiple benefits:
- ~25% faster on the Kraken and Octane benchmarks :^)
- Fewer instructions to accomplish the same thing
- Much easier for humans to read(!)
Because this change requires a fundamental shift in how bytecode is
generated, it is quite comprehensive.
Main implementation mechanism: generate_bytecode() virtual function now
takes an optional "preferred dst" operand, which allows callers to
communicate when they have an operand that would be optimal for the
result to go into. It also returns an optional "actual dst" operand,
which is where the completion value (if any) of the AST node is stored
after the node has "executed".
One thing of note that's new: because instructions can now take locals
as operands, this means we got rid of the GetLocal instruction.
A side-effect of that is we have to think about the temporal deadzone
(TDZ) a bit differently for locals (GetLocal would previously check
for empty values and interpret that as a TDZ access and throw).
We now insert special ThrowIfTDZ instructions in places where a local
access may be in the TDZ, to maintain the correct behavior.
There are a number of progressions and regressions from this test:
A number of async generator tests have been accidentally fixed while
converting the implementation to the new bytecode format. It didn't
seem useful to preserve bugs in the original code when converting it.
Some "does eval() return the correct completion value" tests have
regressed, in particular ones related to propagating the appropriate
completion after control flow statements like continue and break.
These are all fairly obscure issues, and I believe we can continue
working on them separately.
The net test262 result is a progression though. :^)
This is pure prep work for refactoring the bytecode to use more operands
instead of only registers.
generate_bytecode() virtuals now return an Optional<Operand>, and the
idea is to return an Operand referring to the value produced by this
AST node.
They also take an Optional<Operand> "preferred_dst" input. This is
intended to communicate the caller's preference for an output operand,
if any. This will be used to elide temporaries when we can store the
result directly in a local, for example.
The JIT compiler was an interesting experiment, but ultimately the
security & complexity cost of doing arbitrary code generation at runtime
is far too high.
In subsequent commits, the bytecode format will change drastically, and
instead of rewriting the JIT to fit the new bytecode, this patch simply
removes the JIT instead.
Other engines, JavaScriptCore in particular, have already proven that
it's possible to handle the vast majority of contemporary web content
with an interpreter. They are currently ~5x faster than us on benchmarks
when running without a JIT. We need to catch up to them before
considering performance techniques with a heavy security cost.
This begins the process of aligning our implementation with the spec
with regard to using CalendarMethodsRecord. The main intent here is to
make it much easier to make normative changes to AOs which have been
updated to CalendarMethodsRecord.
While this does resolve various FIXMEs, many others above need to be
added in order to be able to pass through a CalendarMethodsRecord. The
use here aligns with what I can gather from the spec of what the
arguments to CreateCalendarMethodsRecord should be, but various AOs have
been updated so much with other changes it's not completely obvious.
Other AOs do not even exist in the latest version of the spec, but we
still rely on them.
As part of these updates, this commit coincidentally also fixes two
PlainDate roundingmode issues seen in test262 - a test of which is also
added in test-js. This issue boiled down to what appears to be an
observable optimization in the spec, where it can avoid calling
dateUntil in certain situations (roundingGranularityIsNoop).
However, the main goal here is to make it much easier to fix many more
issues in the future :^)
since/calendar-dateuntil-called-with-singular-largestunit.js ❌ -> ✅
until/calendar-dateuntil-called-with-singular-largestunit.js ❌ -> ✅
This is part of a large refactor made as part of the temporal spec.
Most AOs using the calendar now pass through this record. There will
need to be a long process of going through updating AOs to use this
record.
This makes it possible to use MakeIndexSequqnce in functions like:
template<typename T, size_t N>
constexpr auto foo(T (&a)[N])
This means AK/StdLibExtraDetails.h must now include AK/Types.h
for size_t, which means AK/Types.h can no longer include
AK/StdLibExtras.h (which arguably it shouldn't do anyways),
which requires rejiggering some things.
(IMHO Types.h shouldn't use AK::Details metaprogramming at all.
FlatPtr doesn't necessarily have to use Conditional<> and ssize_t could
maybe be in its own header or something. But since it's tangential to
this PR, going with the tried and true "lift things that cause the
cycle up to the top" approach.)
Previously, attempting to load a value from an invalid reference would
cause a crash. We now return a CodeGenerationError rather than hitting
an assertion. This is not a complete solution, as ideally we would want
to return a ReferenceError, but this now matches the behavior we see
when we attempt to store something to an invalid reference.
This is a bit tangled in that updating these functions involves a slew
of other spec changes.
However those spec updates fix a bunch of rounding issues, fixing 32
test cases.
Diff Tests:
+32 ✅ -32 ❌
This has the guts of the old temporal AO BalanceDuration with some
differences such as an extra precision of one unit. This appears to be
important for different rounding modes to act as a tiebreaker.
It also does not have any logic regarding a zoned date time 'relative
to' - the spec seems to have this factored in a way where callers are
expected to perform this logic if neccessary.
This is an editorial change in the ECMA-262 spec. See:
12d3687
This AO is meant to replace usages of IteratorNext followed by
IteratorValue with a single operation.
This allows, for example:
ThrowCompletionOr<Optional<Value>> foo()
{
return OptionalNone {};
}
The constructors and constraints here are lifted verbatim from
AK::Optional.
This helps us avoid from needing to construct a Function<T> when
invoking `create_heap_function` with a lambda.
Co-Authored-By: Ali Mohammad Pur <mpfard@serenityos.org>
perform_call() wants a ReadonlySpan<Value>, so just grab a slice of the
current register window instead of making a MarkedVector.
10% speed-up on this function call microbenchmark:
function callee(a, b, c) { }
function caller(callee) {
for (let i = 0; i < 10_000_000; ++i)
callee(1, 2, 3)
}
caller(callee)
Value::Value(double) already converts double to int when it is safe, no
need to check for this here explicitly. While this technically removes
an optimization, I doubt that it will regress performance in any
measurable way.
For parameters that exist strictly as "locals", we can save time and
space by not adding them to the function environment.
This is a speed-up across the board on basically every test.
For example, ~11% on Octane/typescript.js :^)
