An array constructed with a known element count of up to 16 (array
literals, copies, slices and the arrays the runtime builds) keeps its
slots in its own block, as empty arrays built for appends already do, so
it costs one allocation and one free.
scr_arr_make_ops is inlined into its two constructors, so the size class
of scr_arr_new_ops_small folds to a table index. It writes the element
ops id with the other header fields and records ops past the registry
in the array's ext after the storage is set up: the ext pointer precedes
the dense allocation, so inline slots must move out before the ext is
created.
A class carries the collector header when any class in its hierarchy can
be on a cycle, so instances of classes whose own fields hold nothing
cycle-capable were buffered as candidates and walked by every level. Such
an object has no traced children and is never on a cycle.
Their descriptors now name scr_cyc_no_children, and they are allocated
with buf set to SCR_CYC_ACYCLIC. That value has the buffered bit set, so
a release that leaves the object alive makes no call; scr_cyc_is_buffered
is false for it, and an emitted death or publication mark takes the object
out of a candidate buffer only when buf is below -1. A pass that walks
such an object as a child restores the value when it survives.
The runtime call reads the cert and key options before the SNICallback,
so a cert or key initializer after the callback that reassigned the
callback's binding changed the callback the server used, and the server
could serve the wrong certificate. When SNICallback comes before a cert
or key entry, the callback is now saved just before the next of them is
evaluated. With the callback read in order, a callback expression with
side effects lowers instead of being refused.
A counted for loop needs a stable limit to run its counter in integer
storage. Array lengths qualified when the loop provably keeps them, but
a string's length did not, so `for (let i = 0; i < s.length; i++)`
kept a double counter and evaluated the length, through the UTF-16
mapping of a non-ASCII string, on every iteration. A string's length
cannot change, so it is now stable whenever the string's binding is
not rebound in the loop or boxed.
dyn.nameClassPrototype sets the class name on a class's runtime prototype
object and stores no property. The int32 field analysis did not list it as
read-only, so every program that gave a class a runtime prototype counted a
dynamic store that can name any property, and the number fields of every
class reaching dynamic code were stored as doubles.
The type mapper maps the type inferred from an empty object literal to the
empty record as an arm of an unreduced union and to a checked-dynamic
value elsewhere, but both answers shared one memo entry. Once the type had
been mapped on its own, a union such as `typeof empty | DataView` reused
the dynamic answer for its arm and refused (SC2009). The memo key now
includes whether the mapping is for such an arm.
The compiler's own source has 24 likely-hot functions that use untyped
values, so its self-builds now print SC6101. The native bootstrap
contracts compare stderr exactly; the rebuilds of the compiler entry pass
--no-dynamic-tier-warning, which changes no IR or LLVM output.
On the error path, a callback's result reads from a checked-dynamic
slot. Passing it on into a Buffer parameter still handed over the
adapter's original value, so after `if (!out) out = Buffer.from(...)`
a typed callee received the absent Buffer instead of the fallback and
crashed. The forward now tests the slot: while it holds undefined the
callee gets the absent Buffer, and otherwise the Buffer the slot holds.
An array helper specialized for its callback calls the callback's
function directly once the closure's entry matches it, and falls back to
the entry otherwise. Both paths were followed by the pending-exception
test. When the direct target is outside the may-throw set, only the
fallback can leave an exception pending, so the test now runs behind a
guard that is true on that path alone; a known closure called directly
skips it. Worker programs keep the test on every path, where it also
observes termination.
The prerelease libnode-dev-24.22.1 mirrors the dockfries v24.22.1 libnode,
Node's v24.22.1 headers and Node's license byte for byte, as CI-only inputs
(not a scriptc release). dev-pack.json lists the mirror first and the
original locations after it; fetch-dev-pack.mjs takes the first source
whose bytes match the pinned sha256 and logs which one it used.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
An installed @scriptc/libnode-<platform> pack versions in lockstep with
the compiler, whose version every executable cache key already holds;
keying on it as well made the full compiler's key differ from the
bootstrap's routed key wherever a pack directory was present, so routed
cache hits stopped. Only directories with a manifest.json count as
installed packs, so an unbuilt workspace package is not one.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Stack the libnode packs on the in-process loader (#1685) and use its
pack assembler and manifest format instead of a second one:
build-pack.mjs keeps its CLI and fields (schema, nodeVersion,
moduleVersion, bridgeAbi, platform, arch, files), adds the libnode and
bridge names, strips and (on macOS) re-signs the bridge and links it with
-z defs, and exports buildPack(), which the platform packs call with their
package, target and build inputs as extra manifest fields.
libnodeRecord falls back to the installed @scriptc/libnode-<platform>
pack, found from the entry's directory and then next to the compiler,
when neither --libnode nor SCRIPTC_LIBNODE names one.
The Linux test lanes restore the libnode build that the libnode-packs
workflow cached, assemble the linux-x64 pack from it in seconds
(build.mjs --from-cache never builds Node) and export SCRIPTC_LIBNODE, so
the Node island corpus and harness tests run instead of skipping.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Malte Ubl