A checked-dynamic method call looked its name up in the method table on
every call: a strlen, a hash and a byte comparison of the name.
The compiler now gives each method-name constant an i32 slot that starts
at -1 and passes it with the call (scr_dyn_invoke_prepared_id, and the
Buffer method entries, which forward it). The first call stores the
name's id in the slot, and later calls read it. Ids stay the runtime's
own, so the slot holds whatever this runtime's table says. Names are
still matched by content on the first call, and the slot is written with
a relaxed atomic store, since every thread writes the same id.
The runtime ABI is now 28.
A 'message' listener with a typed parameter received each message as a
checked-dynamic value: the port decoded the message into dyn objects, the
listener's thunk matched the value against each union arm, read every
field again to build the record, and the dyn objects were then released.
For bench/channel-messages that was about a third of the delivery time.
- A closure registered with on, once, addListener, prependListener or
prependOnceListener for 'message' gets a typed message entry when its
parameter is a number, boolean, string, a record of those without union
fields, or a boxed union of those arms with null or undefined. The
entry builds the parameter from the message's nodes and calls the
closure.
- The entry handles a message only when it has the parameter's exact
layout: the union arm the checked conversion picks, and a record's
declared keys in declaration order with no others. Any other message
runs through the thunk on the decoded value, as before, so undeclared
keys, other orders and mismatches behave as they did.
- Ports, workers and parentPort deliver a message through the entries
when no value in it is reached twice and it holds no ports or native
handles. Listeners without an entry share one decoded value, made when
the first of them runs.
- The message readers are runtime entry points, so the runtime ABI is now
27.
Corpus 7130 and 7131 check typed listeners on MessageChannel ports, a
worker and parentPort against Node: exact layouts, extra keys, shared
values, nested records, untyped listeners beside typed ones, once
listeners and a transferred port.
Readable, Writable, Duplex, Transform and PassThrough take the
objectMode option, and a Duplex, Transform or PassThrough the
readableObjectMode and writableObjectMode options, as true or false
literals in an options object literal or in an options value. An
object-mode side holds any value that converts to a checked-dynamic
value: push, unshift, write, end, pipe, pipeline, for await, 'data'
listeners, Readable.from and the transform callback carry values in
TypeScript and JavaScript, and constructor-function streams and options
held in variables do the same. The object-mode high-water mark counts
values and defaults to 16. A readable buffer that mixes strings, Buffers
and other values holds them all, and push(undefined) and other values on
a byte-mode stream report ERR_INVALID_ARG_TYPE through 'error', as in
Node.
In typed code a write or transform callback, or a _write or _transform
method, may declare its chunk with any such type, which converts when
the callback runs. Writes pass the encoding argument to _write and
_transform as Node does: undefined when absent, null when a callback
took its place, otherwise the value unchanged. A parameter declared
string | null | undefined receives that value and one declared string
receives "" for undefined and null. write(chunk, encoding) on a
byte-mode stream encodes a string chunk with any known encoding and
throws ERR_UNKNOWN_ENCODING for an unknown one. Readable.from takes
typed Maps and Sets of any such value, and arrays that mix strings and
Buffers.
for await over a readable-only stream that will emit 'close' now ends
after 'close', as Node's eos does. Buffered writes that complete inside
clearBuffer run their callbacks in order, and 'finish' waits for a
deferred afterWrite that carries a callback, as Node's pendingcb does,
unless a _final callback schedules it.
The 'data' listener adapters take three slots and write and transform
callbacks read their chunk and encoding through runtime accessors, so
RUNTIME_ABI_VERSION is now 26.
Cipheriv and Decipheriv values are Transforms, as in Node: they can be
written and ended, piped into, used as pipeline() and stream/promises
pipeline stages, composed, read, iterated with for await, read by
stream/consumers, passed to finished(), and listened to, and their stream
properties read as in Node. Each chunk goes through the cipher's
update(), and the end pushes final() or reports its error through
'error'; writing after final() throws out of write(). The stream unit
builds a cipher's Transform on its first stream use from operations the
cipher's handle provides (ScrDynHandleOps.transform), so programs that
use ciphers without streams link as before. The cipher is a
collector-headered object behind a traced handle: its Transform's steps
hold it, and the pair is reclaimed as a cycle. Through untyped values, in
a program that uses streams, a cipher's stream properties and its on,
once, off, pipe, unpipe, read, pause, resume, setEncoding, write, end,
destroy, cork and uncork members run on its Transform; other stream
members throw a coded SC2020 error.
A Transform holds a write's callback while its readable side is at the
high-water mark and runs it at the next read, and runs it on a tick once
the readable side has ended, as Transform._write does. write() on a
PassThrough, a Transform, a Hash or a cipher used to return true however
much the readable side buffered. zlib streams applied the rule already
and keep their timing.
pipeline() accepts function stages anywhere among its streams, so runs of
async generator stages can sit on both sides of a stream. A stage can be
named by a function declaration, or a const bound to a function
expression, that nothing reassigns; its body lowers again at the
pipeline with its source typed by the pipeline. The source can also be
an array, a string, a synchronous generator object, or a synchronous
generator function without parameters: an async generator steps the
source's iterator and yields each element, and the next stage receives
each element on the same tick as with Node's for await over the source.
The runtime's function-stage entry points take a bitmask of the function
positions and one (next, read, return) helper triple for each generator
pumped into a stream, so RUNTIME_ABI_VERSION is now 25. In JavaScript,
the chunks a stage receives from the previous stage's generator read as
the strings or Buffers it yields, so their string and Buffer methods
compile.
Corpus 7055-7059 cover cipher streams in TypeScript and JavaScript, their
errors, close order and backpressure, the pipeline stages and sources
with their tick order, and Transform backpressure.
crypto.createDiffieHellman, crypto.getDiffieHellman and
crypto.createDiffieHellmanGroup return native DiffieHellman and
DiffieHellmanGroup handles backed by mbedTLS bignums. generateKeys,
computeSecret, getPrime, getGenerator, getPublicKey, getPrivateKey,
setPublicKey, setPrivateKey and verifyError follow Node 24.22:
- the modp1, modp2, modp5 and modp14 through modp18 groups, with the
private key widths of Node's OpenSSL 3 path (200 to 400 bits for
modp5 and up, [2, p - 2] for modp1 and modp2);
- explicit primes and generators in every argument form and encoding,
with keys of bits(p) - 2 bits, or below min(2^N, q) when the
parameters equal an OpenSSL named group (RFC 3526, RFC 7919,
RFC 5114);
- generated safe primes with OpenSSL's congruences for the generator;
- verifyError's DH_check bits, computed when it is first read;
- Node's argument, generator, modulus, key range, state and encoding
errors, including the OpenSSL errors' library and reason properties.
Safe-prime generation sieves candidates with the odd primes below a
bound that grows with the prime's length and tests them with
Miller-Rabin rounds over a variable-time Montgomery exponentiation that
only reads public values; key and secret exponentiations stay on
mbedTLS's constant-time path. Primes longer than 8192 bits, property
reads other than verifyError, property writes, Object.prototype methods,
and calls whose output encoding is typed any or matches no @types/node
overload are refused.
The runtime unit scr_dh.c links, with the mbedTLS archive, only into
programs that use DiffieHellman. wasm32-wasi builds refuse it. The ECDH
unit's argument and encoding helpers move to scr_crypto_args.h, which
both units include.
crypto.createECDH returns a native ECDH handle backed by mbedTLS for
prime192v1, secp224r1, prime256v1, secp384r1, secp521r1, secp192k1,
secp224k1, secp256k1, brainpoolP256r1, brainpoolP384r1 and
brainpoolP512r1. generateKeys, computeSecret, getPublicKey,
getPrivateKey, setPrivateKey, setPublicKey and ECDH.convertKey follow
Node 24: compressed, uncompressed and hybrid points, input and output
encodings, the private key range, key pair validation, and Node's error
codes and messages.
Curve names follow OpenSSL's short-name lookup: names of other curves
from crypto.getCurves() are refused at compile time when literal and
throw a coded SC2020 error when computed, other OpenSSL short names
throw ERR_CRYPTO_OPERATION_FAILED, and unknown names throw
ERR_CRYPTO_INVALID_CURVE. Calls whose arguments no @types/node overload
accepts, and computeSecret(view, encoding), are refused at compile time.
The runtime unit scr_ecdh.c links, with the mbedTLS archive, only into
programs that use ECDH. wasm32-wasi builds refuse it.
ScrArr stores len and cap as 32-bit fields and no longer stores the
state pointer. The dense states share the values' allocation and run
backward from its end: the state of slot i is at
data + cap * (slot width + 1) - 1 - i. Removing slots from the front
(shift, splice(0, n)) advances data and lowers cap, which leaves every
remaining state at its address; the head room size, needed only to free
or reallocate the block, is kept in the vacated slot before data. On
64-bit targets the header goes from 64 to 48 bytes, and from 36 to 32
bytes on wasm32. Array lengths were already limited to 2^32 - 1.
The emitted code loads len and cap as i32 and zero-extends them, and
computes state addresses from data and cap. Runtime loops compute the
end of the states once per call. Local array reads load data once for
both the state and value paths. A local or strict element read whose
index is a loop's induction variable reads slots above the hole-free
prefix through the runtime: an inline state check there makes LLVM keep
-1 - i as a second induction variable, which costs a register in the
loop.
Bulk operations do not read back states they just wrote: an append to
an array whose hole-free prefix reaches its length moves the prefix bound
to the new length, a copy from below the source's prefix writes value
states without reading the source's, and releasing a string array skips
the state checks below the prefix. Copies of up to eight slots and of
short state runs use overlapping word moves instead of a call. Emitted
element access computes the state address only on the state-checked
path. Without these, programs that build many small arrays with split or
slice ran 2-4% slower than with the separate state pointer.
Integer loops over an array's length compare the counter with the i32
length field in 32 bits; the counter never exceeds 2^32 - 1 there. The
array-length stability proof accepts Map and Set reads (get, has, size)
and checked union extractions (`x!`), which run no user code, so loops
that look up maps keep the dense fast copy, which tests the hole-free
prefix once before the loop.
Small constructions with the small allocator request 6 slots, so their
blocks stay 112 bytes, now with 7 inline slots (6 after a collector
header). With 96-byte blocks of 5 slots, programs that free and reuse
many small arrays ran 15-20% slower.
RUNTIME_ABI_VERSION goes from 22 to 23.
The process.emitWarning replacement and the stdin encoding, string data
listener and stdio 'resize' support changed runtime functions every
program with the dyn-async or events unit links (the shared warning
paths, stdin delivery, the process stream object) and grew their
thread-local state. That code now compiles only into new runtime-pack
variants, SCR_WARNING_REPLACE of scr_async_dyn.c and SCR_STDIO_EXTENDED
of scr_events.c, which the link selects when the program assigns
process.emitWarning or reaches setEncoding, a string 'data' listener or
'resize'. The default objects are byte-identical to before, so other
executables do not change. The development and sanitizer builds pass the
same defines.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
SCRIPTC_YOUNG_NURSERY=1 at compile time makes an executable allocate
traced, program-owned class instances and records young and link the
collector unit built with SCR_YIN (a runtime pack variant of
scr_cycle.c; libraries never get it). The emitted constructor lists the
object in a per-thread nursery and marks its header: generation
SCR_CYC_YOUNG and its nursery index under the buffered bit, so every
existing release path treats it as a buffered candidate (no call while it
stays alive; scr_cyc_on_dead takes it out of the nursery at its death).
A nursery collection (whenever 4096 young objects are live, at collection
points, before full sweeps and limited passes) counts each young object's
references from other young objects, marks from the objects whose count is
below their reference count (referenced from outside the nursery: stacks,
globals, older objects, runtime structures, FFI, fibers, coroutine
frames), frees what stays unmarked, and gives up the dead set's outgoing
edges after the teardowns. The counts are recomputed at every collection
instead of maintained by a store barrier, so soundness rests only on the
invariants trial deletion already needs. Survivors join trial deletion's
nursery and outside roots become candidates; trial deletion walks young
objects it reaches as nursery-level objects. Per-type admission statistics
send types that do not die in young cycles back to trial deletion for a
while. Publication takes young objects out of the nursery. The RC audit
build checks nursery membership and counts at every collection.
With the flag off, the compiler emits and links exactly what it did
before (249 programs and the self-hosted compiler compared
byte-identical). Measured with the flag on: cg-rings -38%, cg-pairs -27%,
the runtime suite geomean -0.3% (published-graphs +3.4%), the self-hosted
compiler neutral; shapes whose cycles go through arrays or closures
(phase 2) pay without gaining.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Release executables now import small runtime functions from runtime
bitcode, as speed executables do, so LLVM can inline them into program
code. Release units keep the runtime compiler's own objects, unchanged,
and additionally ship the bitcode of the same compilation reduced by the
helper's prune-import. Their unit-local symbols stay local, so only
bodies that need none of them can be imported, which keeps the import
small: about 1.3 MB per program.
Only unsanitized programs small enough for one codegen partition import.
Every partition reads the bitcode again; on the Effect fixtures (64
partitions) import cost +2..18% compile time and +4.3..4.9% size, so
partitioned release builds stay exactly as before. Speed is unchanged.
Measured against main in Linux Vercel Sandboxes, same-VM interleaved
medians (bench-runtime --layouts=4 --runs=16), on two VMs:
- bench-runtime (31 workloads): -1.3% (8 faster, 0 slower) and -1.0%
(1 slower, typed-array-scan-workers +3.2%); size +0.0%
- js-compiled suite (22 programs): -2.2% and -1.8%; size -0.1%
- cold builds of all 53 programs, interleaved: +3.1%; helper time per
build +40..235 ms
- Effect fixtures: byte-identical to main
Release builds now emit the string retain and the dyn retain and release
fast paths inline. The dyn release decides the cases scr_dyn_release
decides by kind and calls scr_dyn_release_survivor for capsules and
handles. String releases and the other families stay runtime calls, as
string-heavy programs have many release sites. The speed posture keeps
every fast path and gains the dyn release. The runtime ABI version moves
to 21 because program objects call the new runtime entry point.
Every SCR_ELEM_REF array stored its element type's retain, release and trace entry points in its own header (24 bytes), so ScrArr was 80 bytes: an 80-byte block for arrays of acyclic elements and a 96-byte block for cycle-capable ones. The three pointers now live in one constant ScrArrOps descriptor per element type, which the compiler emits once per module (@sc_arrops.*) and the runtime keeps static, and the header holds a pointer to it plus a 4-byte `cyc` flag beside `elem` that the inline RC fast paths test. ScrArr is 64 bytes: arrays take 64- and 80-byte blocks.
scr_arr_new_ref(retain, release, trace, cap) becomes scr_arr_new_ops(ops, cap); scr_set_to_arr_ref takes the result's descriptor from its caller instead of building an array from the set's key adapters. The backend names the %ScrArr fields in SCR_ARR_FIELD. The layout change bumps the runtime ABI marker to scr_runtime_abi_v13.
Nearly every long-lived object becomes a cycle candidate once, soon after
it is built: the constructing frame's release leaves it held only by the
structure it was stored into. A nursery pass then walks it (about 75 ns per
object), it survives, and it is re-buffered for the mature level, whose
next pass walks it again. In tsc-ts single-threaded, nursery passes walked
6.0M objects on Excalidraw and 36.5M on Sentry and freed 7%; 99% of what
they walked was promoted. Most of that walking goes to types whose objects
never die in a young cycle: SymbolLinks, NodeLinks and Symbol were 2.0M of
Excalidraw's candidates and the collector freed none of them, while young
garbage comes from a few types (inference contexts and mappers, closures,
some arrays). Types with a collector yield under 0.5% are 42-55% of what
nursery passes walk.
So the collector now keeps per-thread statistics per type and level. Each
ScrCycType gains a `kind` index: emitted descriptors number themselves from
32 (FIRST_EMITTED_CYC_KIND / SCR_CYC_KIND_EMITTED), the runtime's own
families have fixed kinds below that, 0 keeps no statistics. For the
nursery and mature levels, a pass counts the candidates of each kind it
walks and the objects of each kind it frees. After every 8192 walked
candidates a kind bypasses that level for the next window if fewer than 1
in 128 of them were freed. A release that buffers an object at a level its
type bypasses promotes it past the level first, and a nursery pass that
re-buffers a survivor into a bypassed mature level moves it to the old
level. One candidate in 16 stays put, so the yield keeps being measured
and a type whose objects start dying young again is walked again after
one window.
Only the generation an object is judged in changes. Any assignment is
sound (a restricted pass counts edges from older objects as external), a
bypassing candidate stays buffered, and full passes walk everything they
did before, so the old level's bounds hold unchanged. Cyclic garbage made
only of objects of a bypassing type waits for the next pass at an older
level. Programs that never complete a window (8192 walked candidates of one
type) are collected exactly as before.
The descriptor layout is part of the emitted-code/runtime ABI, so the
program-object marker moves to scr_runtime_abi_v12.
Pass log (instrumented build, single-threaded, one run each, with the
preceding null-slot change): all collector passes Excalidraw 606 -> 313 ms
(nursery 435 -> 153), TypeORM 629 -> 383 ms, tsc-ts self-check 377 -> 215
ms, Sentry 7.61 -> 5.75 s (nursery 2.92 -> 1.04 s, mature 1.17 -> 1.02 s;
the full pass is unchanged at 3.6 s).
tsc-ts (Linux sandbox, interleaved, medians; 10 rounds, Sentry 5 rounds
pinned to 8 CPUs; both changes against int/best-v6): single-threaded
Excalidraw -8.0%, TypeORM -4.2%, self-check -4.3%, Sentry -5.3%;
--checkers 8 -12.2%, -0.8%, -2.3%, -1.6%. Peak RSS -0.8% to +0.6%.
Diagnostics identical.
Tests: test_cycle gains check_nursery_bypass, check_mature_bypass and
check_rebuffer_past_mature (a test-only entry point, scr_cyc_pass_at under
SCR_CYC_TESTING, runs passes at exact levels); each fails under the
corresponding mutation (no bypass, a sticky decision, no sampling, no freed
or walked counts, a bypassing candidate left unbuffered, survivors
re-buffered into a bypassed level). Corpus 4690 builds a long-lived
declaration graph large enough for both levels to bypass and lets objects
of the same classes die in cycles; under --sanitize the exit RC audit
proves they are reclaimed.
Every ScrArr carried sixteen words, seven of them (sparse slots, their
length and capacity, noncanonical numeric properties with length and
capacity, and match-result metadata) used only by sparse arrays, arrays
with numeric non-index properties and regex match results. They now live in
a ScrArrExt the array points at, allocated on first use: a dense array
shrinks from 128 to 80 bytes, and a cycle-capable one (with its 16-byte
collector header) from 144 to 96.
Runtime readers go through a shared empty ext; writers allocate it. The
backend mirror loses the seven fields, tagged-template arrays are emitted
with a NULL ext, live stream commits swap the ext pointer (which swapped all
seven fields before), and the cached typed-view refresh swaps sparse slots
and properties through scr_arr_swap_indexed_ext, keeping each array's
metadata as before. The program-object ABI marker moves to
scr_runtime_abi_v11.
A corpus program (4610-array-ext-dyn-refresh.ts, with its order baseline) pins numeric non-index properties of a dynamically typed array through its refreshed typed view, sparse slots written directly and through a live reference, truncation, and regex match metadata read directly, through `any` and after a copy.
Every cycle-capable object carried a 32-byte header: trace and teardown function pointers, a 32-bit color, the buffered flag, the generation and a 64-bit candidate index. The two function pointers are the same for every object of a type, so the header now points at one constant ScrCycType { trace, free_fn } descriptor per type, and the remaining fields pack into the second word: u8 color, u8 gen, u16 buffered, u32 buf_index (candidate buffers stop at 2^31 entries). wasm32 pads before the descriptor, so every object-relative offset is the same on all targets.
Emitted shapes define their descriptor next to the trace and teardown functions; inline allocation stores one header word; the inline release reads buffered at obj-6 and the inline free reads the generation as an i8 at obj-7. Runtime families allocate with static descriptors. The program-object ABI marker moves to scr_runtime_abi_v10.
tsc-ts (interleaved, 5 rounds): peak RSS -8% to -11% on every app and mode; single-threaded wall -0.3% to -5.5% on top of the typed-rest-spread change.
- Synchronize published packages and the compiler surface manifest at v0.2.8, and bump the runtime ABI marker to v9.
- Add release notes for worker graph sharing, catchable deep recursion, parallel code generation, runtime performance, and compatibility fixes.
Worker runtime variants link only into executables, whose own TLS block
holds every runtime thread-local, but they were compiled with the
default general-dynamic model. The linker relaxes each __tls_get_addr
call to a %fs-relative address, yet the compiler had already treated it
as a call: hot runtime leaves saved and restored extra registers around
every thread-local access (scr_mg_visit pushed five registers instead
of one). With -ftls-model=initial-exec each access is one %fs-relative
operand. Darwin, Windows and wasm variants are unchanged.
tsc-ts self-check, Linux sandbox, Worker build vs non-worker build:
1.04x -> 1.03x (9 interleaved runs).
The default build's largest size contributor on the stack was H23's
-falign-functions=64 on every x86-64 runtime unit (2-11 KB per program;
results/h45-size-budget). Release and library flavors now keep the
compiler's default function alignment; the opt-in speed flavor keeps the
blanket 64-byte alignment. Vendored archives stay 64-byte aligned
(lre_exec keeps its pinned placement) except libunicode.c, whose small
table helpers were most of the regex programs' padding. The runtime-first
link order is unchanged.
Docs and usage text describe speed without the reverted runtime profile.
Chris Tate