Generating Error objects got a lot slower after the introduction of
SourceCode in b0b022507b.
This was noticeable with `test-js` which generates a lot of errors,
so walking the source code over and over to compute (line, column)
was eating a ton of time.
This patch makes repeated lookups a lot faster by building a cache
of line break offsets in the source code. The cache is built on first
offset lookup, so we only pay for this in code that actually throws.
On my machine, this takes `test-js` runtime from 6.7 sec to 4.3 sec.
This patch does two things:
- We now use u32 instead of size_t for the hops and index fields
in EnvironmentCoordinate. This means we're limited to an environment
nesting level and variable count of 4Gs respectively.
- Instead of wrapping it in an Optional, EnvironmentCoordinate now has
a custom valid/invalid state using a magic marker value.
These two changes reduce the size of Identifier by 16 bytes. :^)
Before this change, each AST node had a 64-byte SourceRange member.
This SourceRange had the following layout:
filename: StringView (16 bytes)
start: Position (24 bytes)
end: Position (24 bytes)
The Position structs have { line, column, offset }, all members size_t.
To reduce memory consumption, AST nodes now only store the following:
source_code: NonnullRefPtr<SourceCode> (8 bytes)
start_offset: u32 (4 bytes)
end_offset: u32 (4 bytes)
SourceCode is a new ref-counted data structure that keeps the filename
and original parsed source code in a single location, and all AST nodes
have a pointer to it.
The start_offset and end_offset can be turned into (line, column) when
necessary by calling SourceCode::range_from_offsets(). This will walk
the source code string and compute line/column numbers on the fly, so
it's not necessarily fast, but it should be rare since this information
is primarily used for diagnostics and exception stack traces.
With this, ASTNode shrinks from 80 bytes to 32 bytes. This gives us a
~23% reduction in memory usage when loading twitter.com/awesomekling
(330 MiB before, 253 MiB after!) :^)
We can't be nuking the ESO while its owned execution context is still on
the VM's execution context stack, as that may lead to a use-after-free.
This patch solves this by adding a `context_owner` field to each context
and treating it as a GC root.
We were mistakenly trying to append UTF-16 code units to a StringBuilder
via the append(char) API. This patch fixes that by accumulating the
result in a Vector<u16> instead.
This'll be a bit worse for performance, since we're now doing additional
UTF-16 string conversions, but we're going for correctness at this stage
and can worry about performance later.
These lambdas were marked mutable as they captured a Ptr wrapper
class by value, which then only returned const-qualified references
to the value they point from the previous const pointer operators.
Nothing is actually mutating in the lambdas state here, and now
that the Ptr operators don't add extra const qualifiers these
can be removed.
Even if the pointer value is const, the value they point to is not
necessarily const, so these functions should not add the qualifier.
This also removes the redundant non-const implementations of these
operators.
Before this could give the `must be followed by in` error before the
undeclared private identifier error. Fixing the `in` error would not
have resolved the other error so this order makes the errors more
actionable.
If we don't check that a private identifier is valid this can break the
assumption that we have a private environment when evaluation the
private identifier. Also an unknown private identifier this should
be a SyntaxError.
