node:stream: sync the stream suite to Node v26.3.0 and fix the gaps it surfaces — stream/iter + zlib/iter behind --experimental-stream-iter, read()-one-buffer semantics, compose/pipeline/web-adapter/http/fs/net parity (247/251 upstream tests, +63 vendored)#31826
Open
cirospaciari wants to merge 28 commits into
Open
node:stream: sync the stream suite to Node v26.3.0 and fix the gaps it surfaces — stream/iter + zlib/iter behind --experimental-stream-iter, read()-one-buffer semantics, compose/pipeline/web-adapter/http/fs/net parity (247/251 upstream tests, +63 vendored)#31826cirospaciari wants to merge 28 commits into
cirospaciari wants to merge 28 commits into
Conversation
Collaborator
|
Updated 9:37 AM PT - Jun 10th, 2026
❌ @robobun, your commit 699fc81 has 1 failures in
🧪 To try this PR locally: bunx bun-pr 31826That installs a local version of the PR into your bun-31826 --bun |
Contributor
|
Note Reviews pausedIt looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the Use the following commands to manage reviews:
Use the checkboxes below for quick actions:
WalkthroughAdds experimental stream/iter modules, web/classic adapters, pull/push/share/broadcast primitives, compression transforms, runtime gating for node:stream/iter, async_hooks–nextTick init calls, stream/http/net adjustments, error-code/binding updates, process exception hardening, and extensive test coverage. ChangesExperimental Streams Iteration Stack
Suggested reviewers:
|
Contributor
|
This PR may be a duplicate of:
🤖 Generated with Claude Code |
![claude[bot]](https://avatars.githubusercontent.com/in/1236702?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Additional findings (outside current diff — PR may have been updated during review):
-
🟡
src/js/internal/streams/operators.ts:397-400— Nit: removingcomposefromstreamReturningOperatorsleavesfunction compose(stream, options)(line 21) and itsstaticComposerequire (line 6) unreferenced —Readable.prototype.composenow lives in readable.ts with its own lazy require. TheisWritable/isNodeStream/addAbortSignalNoValidateimports (lines 7-8) are also only used by that function, so all of these can be deleted.Extended reasoning...
What
This PR moves
Readable.prototype.composeoff the operators registry and onto the prototype directly inreadable.ts(so it returns the composed Duplex rather than wrapping it in a Readable). The diff tooperators.tsremovescomposefrom thestreamReturningOperatorsexport object — but it leaves thefunction compose(stream, options)definition at line 21 in place, along with thestaticComposerequire at line 6 that only this function uses.Step-by-step
- Before this PR,
operators.tsdefinedfunction composeand exported it viastreamReturningOperators.compose;stream.tsthen iterated that object to installReadable.prototype.composeas a registry-wrapped operator. - This PR adds
Readable.prototype.compose = function compose(stream, options) { … }directly insrc/js/internal/streams/readable.ts:1289, with its owncomposeImpl ??= require("internal/streams/compose")lazy load. - The diff hunk at
operators.ts:397-400removescompose,fromstreamReturningOperatorsso the registry no longer installs a competing prototype method. - After that removal, grepping
operators.tsshowscomposereferenced only at line 21 (the definition) and line 33 (staticCompose(this, stream)inside its own body). Nothing else in the file — and nothing outside it, since the function was never exported by name — references it. - The
staticComposeimport at line 6 is used solely at line 33 insidecompose. Likewise,addAbortSignalNoValidate(line 7) is used only at line 37, andisWritable/isNodeStream(line 8) only at line 29 — all inside the now-dead function body.
Why nothing prevents it
composeis a top-level function declaration with no other call sites, so it simply sits in the module as dead code. Bun's builtin bundler may or may not tree-shake the function and its CommonJS-stylerequire("internal/streams/compose")import; either way there is no observable runtime behavior change.Impact
None at runtime — this is purely a maintenance/cleanliness issue. The dead function is ~20 lines plus four unused import bindings across three
requirecalls. It's mildly confusing to leave acomposehere when the real implementation has moved toreadable.ts, but it cannot cause incorrect behavior.Fix
Delete
function compose(stream, options) { … }(lines 21-41) and the now-unused imports:const staticCompose = require("internal/streams/compose");(line 6),const { addAbortSignalNoValidate } = require("internal/streams/add-abort-signal");(line 7), andconst { isWritable, isNodeStream } = require("internal/streams/utils");(line 8). - Before this PR,
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Actionable comments posted: 10
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@src/js/internal/streams/iter/broadcast.ts`:
- Around line 146-147: Replace truthy checks that treat falsy cancellation
reasons as absent by explicitly checking for undefined: change occurrences of
"if (self.#error)" to "if (self.#error !== undefined)" (and similarly for the
abort-idempotency flag, e.g., replace "if (this.#aborted)" with "if
(this.#aborted !== undefined)"). Update all spots in broadcast.ts that currently
use truthy checks on private fields (notably the self.#error checks and the
abort/idempotency check) so cancel(0), cancel(""), or cancel(false) correctly
trigger rejection instead of returning kDone.
- Line 528: The end(options) method in broadcast.ts has an unused parameter
causing eslint no-unused-vars; rename the parameter to _options (e.g.,
end(_options)) or prefix it with an underscore to silence the linter, or remove
the parameter entirely if the method signature isn't required by an interface;
update any internal references to that parameter accordingly and keep the method
body unchanged.
In `@src/js/internal/streams/iter/classic.ts`:
- Around line 511-517: The sync stub methods writeSync and writevSync are
triggering eslint(no-unused-vars); rename their parameters to unused-prefixed
names (e.g., _chunk for writeSync and _chunks for writevSync) or otherwise
prefix them with an underscore so the linter recognizes them as intentionally
unused, leaving the bodies returning false unchanged; update the function
signatures for writeSync(chunk) and writevSync(chunks) to use the new names
(_chunk, _chunks) to resolve the lint errors.
In `@src/js/internal/streams/iter/duplex.ts`:
- Around line 112-123: The abort listener currently stays registered if both
endpoints close before abort; keep a reference to the listener (abortBoth) as
you already have, attach it with signal.addEventListener("abort", abortBoth, {
__proto__: null, once: true }), and ensure you call
signal.removeEventListener("abort", abortBoth) when both channels are closed
(e.g., in the completion/close handlers for the aWriter and bWriter endpoints)
so the listener and its captured channel state can be released; also call
removeEventListener inside abortBoth after failing both writers to guarantee
cleanup when abort actually fires.
In `@src/js/internal/streams/iter/from.ts`:
- Around line 289-310: The async-iterable branch can emit arbitrarily large
batches (via isUint8ArrayBatch and the accumulated `batch` from
normalizeAsyncValue), bypassing FROM_BATCH_SIZE; modify the logic in the
isAsyncIterable path so that pre-batched arrays from isUint8ArrayBatch are split
into chunks of at most FROM_BATCH_SIZE before yielding, and when collecting
normalized chunks from normalizeAsyncValue push into a temporary buffer and
yield slices of size FROM_BATCH_SIZE as they fill (yield any final partial slice
at the end). Keep the existing fast path for single Uint8Array (isUint8Array
yielding [value]) but ensure all emitted batches never exceed FROM_BATCH_SIZE.
Use the existing symbols isAsyncIterable, isUint8ArrayBatch, isUint8Array,
normalizeAsyncValue and the FROM_BATCH_SIZE constant to locate and implement the
changes.
In `@src/js/internal/streams/iter/pull.ts`:
- Around line 870-875: The code calls writer.writev(batch, opts).then(...)
assuming a Promise; change this to handle non-Promise returns by normalizing the
result (e.g., assign result = writer.writev(batch, opts) and use
Promise.resolve(result).then(...) or check for a thenable before calling .then)
so that if writev returns undefined or a non-Promise the byte-counting loop
still runs without throwing; update the same pattern if writer.write(...) is
used elsewhere in this pull flow to ensure consistent handling.
In `@src/js/internal/streams/iter/push.ts`:
- Line 609: The method end(options) declares an unused parameter causing lint
failures; either remove the parameter or rename it to _options (or /* options
*/) in the end function signature to mark it as intentionally unused. Locate the
end function declaration (symbol: end) and update its signature to drop the
options argument or underscore it, and run tests/lint to confirm the warning is
resolved.
In `@src/js/internal/streams/iter/transform.ts`:
- Around line 586-610: The sync transform loop currently only finalizes when
receiving a null batch and can finalize multiple times; add a boolean guard
(e.g., finalized) alongside the existing async pattern so finalization (calling
processSyncInput(..., finishFlag) and draining via drainBatch()) happens exactly
once when the source ends or when a null flush marker is seen; update the for-of
loop over source to check/trigger finalization on end-of-loop (before leaving
the try) and ensure the finally block also uses the finalized guard to avoid
duplicate finalize calls—modify references in this function to processSyncInput,
drainBatch, pending, pendingBytes, BATCH_HWM, finishFlag and processFlag
accordingly.
In `@src/js/internal/webstreams_adapters.ts`:
- Around line 441-445: Guard against null/undefined `options` before
dereferencing `options.type`: first check that `options` is an object (e.g. if
(options != null && options.type !== undefined) { validateOneOf(options.type,
"options.type", ["bytes", undefined]); } const isBYOB = options != null &&
options.type === "bytes";) Update the code paths that use `options.type`,
referencing the existing `validateOneOf` and `isBYOB` identifiers so property
access is always preceded by the null check.
In `@src/js/node/net.ts`:
- Around line 744-757: The adoption path for FIFO/character-device/file
descriptors can leave the stream readable forever if options.readable === true
but there is no backing reader (_handle), so update the fd-adoption branch
(where stats.isFIFO(), stats.isCharacterDevice(), stats.isFile(), and
stats.isSocket()) to detect absence of a reader (e.g. this._handle or
equivalent) and explicitly close the readable side in that case by calling
this.push(null) and this.read(0); ensure this logic runs regardless of
options.readable value when no _handle exists so the stream does not hang
(adjust the condition around kSyncWriteFd/_write/_writev assignment and the
push(null)/read(0) calls to handle the "readable but no _handle" case).
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: ASSERTIVE
Plan: Pro
Run ID: 9deac0d9-283a-4f76-a5f4-25d4408e8ad6
📒 Files selected for processing (171)
src/js/builtins/ProcessObjectInternals.tssrc/js/builtins/ReadableStreamInternals.tssrc/js/internal-for-testing.tssrc/js/internal/async_context_frame.tssrc/js/internal/async_hooks.tssrc/js/internal/async_hooks_tick.tssrc/js/internal/fs/streams.tssrc/js/internal/http.tssrc/js/internal/streams/duplex.tssrc/js/internal/streams/duplexify.tssrc/js/internal/streams/from.tssrc/js/internal/streams/iter/broadcast.tssrc/js/internal/streams/iter/classic.tssrc/js/internal/streams/iter/consumers.tssrc/js/internal/streams/iter/duplex.tssrc/js/internal/streams/iter/from.tssrc/js/internal/streams/iter/pull.tssrc/js/internal/streams/iter/push.tssrc/js/internal/streams/iter/ringbuffer.tssrc/js/internal/streams/iter/share.tssrc/js/internal/streams/iter/transform.tssrc/js/internal/streams/iter/types.tssrc/js/internal/streams/iter/utils.tssrc/js/internal/streams/operators.tssrc/js/internal/streams/pipeline.tssrc/js/internal/streams/readable.tssrc/js/internal/streams/writable.tssrc/js/internal/webstreams_adapters.tssrc/js/node/_http_client.tssrc/js/node/_http_incoming.tssrc/js/node/_http_server.tssrc/js/node/async_hooks.tssrc/js/node/net.tssrc/js/node/stream.iter.tssrc/js/node/zlib.iter.tssrc/jsc/ErrorCode.rssrc/jsc/bindings/ErrorCode.cppsrc/jsc/bindings/ErrorCode.tssrc/jsc/bindings/JSCommonJSModule.cppsrc/resolve_builtins/HardcodedModule.rssrc/resolve_builtins/lib.rssrc/runtime/jsc_hooks.rssrc/runtime/node/zlib/NativeBrotli.rstest/js/node/stream/node-stream-uint8array.test.tstest/js/node/test/common/index.jstest/js/node/test/parallel/test-crypto-cipheriv-decipheriv.jstest/js/node/test/parallel/test-stream-add-abort-signal.jstest/js/node/test/parallel/test-stream-big-push.jstest/js/node/test/parallel/test-stream-compose.jstest/js/node/test/parallel/test-stream-construct.jstest/js/node/test/parallel/test-stream-consumers.jstest/js/node/test/parallel/test-stream-destroy.jstest/js/node/test/parallel/test-stream-drop-take.jstest/js/node/test/parallel/test-stream-duplex-destroy.jstest/js/node/test/parallel/test-stream-duplex-from.jstest/js/node/test/parallel/test-stream-duplex-readable-writable.jstest/js/node/test/parallel/test-stream-duplex-writable-finished.jstest/js/node/test/parallel/test-stream-duplex.jstest/js/node/test/parallel/test-stream-duplexpair.jstest/js/node/test/parallel/test-stream-end-paused.jstest/js/node/test/parallel/test-stream-filter.jstest/js/node/test/parallel/test-stream-finished-async-local-storage.jstest/js/node/test/parallel/test-stream-finished-bindAsyncResource-path.jstest/js/node/test/parallel/test-stream-finished-default-path.jstest/js/node/test/parallel/test-stream-forEach.jstest/js/node/test/parallel/test-stream-iter-broadcast-backpressure.jstest/js/node/test/parallel/test-stream-iter-broadcast-basic.jstest/js/node/test/parallel/test-stream-iter-broadcast-coverage.jstest/js/node/test/parallel/test-stream-iter-broadcast-from.jstest/js/node/test/parallel/test-stream-iter-consumers-bytes.jstest/js/node/test/parallel/test-stream-iter-consumers-merge.jstest/js/node/test/parallel/test-stream-iter-consumers-tap.jstest/js/node/test/parallel/test-stream-iter-consumers-text.jstest/js/node/test/parallel/test-stream-iter-cross-realm.jstest/js/node/test/parallel/test-stream-iter-disabled.jstest/js/node/test/parallel/test-stream-iter-duplex.jstest/js/node/test/parallel/test-stream-iter-from-async.jstest/js/node/test/parallel/test-stream-iter-from-coverage.jstest/js/node/test/parallel/test-stream-iter-from-sync.jstest/js/node/test/parallel/test-stream-iter-from-writable-cache-options.jstest/js/node/test/parallel/test-stream-iter-namespace.jstest/js/node/test/parallel/test-stream-iter-pipeto-edge.jstest/js/node/test/parallel/test-stream-iter-pipeto-signal.jstest/js/node/test/parallel/test-stream-iter-pipeto-writev.jstest/js/node/test/parallel/test-stream-iter-pipeto.jstest/js/node/test/parallel/test-stream-iter-pull-async.jstest/js/node/test/parallel/test-stream-iter-pull-sync.jstest/js/node/test/parallel/test-stream-iter-push-backpressure.jstest/js/node/test/parallel/test-stream-iter-push-basic.jstest/js/node/test/parallel/test-stream-iter-push-writer.jstest/js/node/test/parallel/test-stream-iter-readable-interop-disabled.jstest/js/node/test/parallel/test-stream-iter-readable-interop.jstest/js/node/test/parallel/test-stream-iter-share-async.jstest/js/node/test/parallel/test-stream-iter-share-coverage.jstest/js/node/test/parallel/test-stream-iter-share-from.jstest/js/node/test/parallel/test-stream-iter-share-sync.jstest/js/node/test/parallel/test-stream-iter-sharedarraybuffer.jstest/js/node/test/parallel/test-stream-iter-to-readable.jstest/js/node/test/parallel/test-stream-iter-transform-compat.jstest/js/node/test/parallel/test-stream-iter-transform-coverage.jstest/js/node/test/parallel/test-stream-iter-transform-errors.jstest/js/node/test/parallel/test-stream-iter-transform-output.jstest/js/node/test/parallel/test-stream-iter-transform-roundtrip.jstest/js/node/test/parallel/test-stream-iter-transform-sync.jstest/js/node/test/parallel/test-stream-iter-validation.jstest/js/node/test/parallel/test-stream-iter-writable-from.jstest/js/node/test/parallel/test-stream-iter-writable-interop.jstest/js/node/test/parallel/test-stream-map.jstest/js/node/test/parallel/test-stream-pipe-await-drain.jstest/js/node/test/parallel/test-stream-pipe-flow.jstest/js/node/test/parallel/test-stream-pipe-objectmode-to-non-objectmode.jstest/js/node/test/parallel/test-stream-pipeline-http2.jstest/js/node/test/parallel/test-stream-pipeline-listeners.jstest/js/node/test/parallel/test-stream-pipeline.jstest/js/node/test/parallel/test-stream-push-strings.jstest/js/node/test/parallel/test-stream-readable-aborted.jstest/js/node/test/parallel/test-stream-readable-async-iterators.jstest/js/node/test/parallel/test-stream-readable-compose.jstest/js/node/test/parallel/test-stream-readable-didRead.jstest/js/node/test/parallel/test-stream-readable-dispose.jstest/js/node/test/parallel/test-stream-readable-emittedReadable.jstest/js/node/test/parallel/test-stream-readable-ended.jstest/js/node/test/parallel/test-stream-readable-event.jstest/js/node/test/parallel/test-stream-readable-hwm-0-no-flow-data.jstest/js/node/test/parallel/test-stream-readable-infinite-read.jstest/js/node/test/parallel/test-stream-readable-needReadable.jstest/js/node/test/parallel/test-stream-readable-readable-one.jstest/js/node/test/parallel/test-stream-readable-strategy-option.jstest/js/node/test/parallel/test-stream-readable-to-web-byob.jstest/js/node/test/parallel/test-stream-readable-to-web-termination-byob.jstest/js/node/test/parallel/test-stream-readable-to-web-termination.jstest/js/node/test/parallel/test-stream-readable-to-web.mjstest/js/node/test/parallel/test-stream-reduce.jstest/js/node/test/parallel/test-stream-toArray.jstest/js/node/test/parallel/test-stream-transform-destroy.jstest/js/node/test/parallel/test-stream-transform-final.jstest/js/node/test/parallel/test-stream-transform-flush-data.jstest/js/node/test/parallel/test-stream-typedarray.jstest/js/node/test/parallel/test-stream-uint8array.jstest/js/node/test/parallel/test-stream-unpipe-event.jstest/js/node/test/parallel/test-stream-unshift-empty-chunk.jstest/js/node/test/parallel/test-stream-unshift-read-race.jstest/js/node/test/parallel/test-stream-wrap-drain.jstest/js/node/test/parallel/test-stream-wrap-encoding.jstest/js/node/test/parallel/test-stream-wrap.jstest/js/node/test/parallel/test-stream-writable-change-default-encoding.jstest/js/node/test/parallel/test-stream-writable-decoded-encoding.jstest/js/node/test/parallel/test-stream-writable-destroy.jstest/js/node/test/parallel/test-stream-writable-ended-state.jstest/js/node/test/parallel/test-stream-writable-finished-state.jstest/js/node/test/parallel/test-stream-writable-finished.jstest/js/node/test/parallel/test-stream-writable-null.jstest/js/node/test/parallel/test-stream-writable-samecb-singletick.jstest/js/node/test/parallel/test-stream-writable-write-error.jstest/js/node/test/parallel/test-stream-writev.jstest/js/node/test/parallel/test-stream2-base64-single-char-read-end.jstest/js/node/test/parallel/test-stream2-basic.jstest/js/node/test/parallel/test-stream2-compatibility.jstest/js/node/test/parallel/test-stream2-httpclient-response-end.jstest/js/node/test/parallel/test-stream2-objects.jstest/js/node/test/parallel/test-stream2-push.jstest/js/node/test/parallel/test-stream2-read-correct-num-bytes-in-utf8.jstest/js/node/test/parallel/test-stream2-readable-wrap.jstest/js/node/test/parallel/test-stream2-transform.jstest/js/node/test/parallel/test-stream2-writable.jstest/js/node/test/parallel/test-stream3-cork-end.jstest/js/node/test/parallel/test-stream3-cork-uncork.jstest/js/node/test/parallel/test-stream3-pause-then-read.jstest/js/node/test/parallel/test-stream3-pipeline-async-iterator.jstest/js/node/test/parallel/test-zlib-flush-write-sync-interleaved.jstest/js/node/test/sequential/test-stream2-stderr-sync.js
💤 Files with no reviewable changes (1)
- src/js/internal/streams/operators.ts
Member
Author
|
Review triage (claude[bot] + coderabbit rounds): Fixed:
Already fixed before the review snapshot (stale findings):
Faithful-port rebuttals — these match node v26.3.0 source line-for-line, so changing them would diverge from upstream (and from the vendored upstream tests):
🤖 PR feedback addressed by Claude Code. |
Contributor
There was a problem hiding this comment.
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
src/js/internal/streams/readable.ts (1)
1269-1275:⚠️ Potential issue | 🟠 Major | 🏗️ Heavy liftUse the CLI-parsed gate here instead of
process.execArgv.
process.execArgvis userland-mutable, so this feature gate can be flipped after startup: pushing--experimental-stream-iterenables the API without the CLI flag, and removing it makes this method throw even though resolver gating still uses the write-once flag fromsrc/resolve_builtins/HardcodedModule.rs. Please plumb the same immutable startup bit into this builtin so both resolution andReadable[Symbol.for("Stream.toAsyncStreamable")]()stay in sync.🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@src/js/internal/streams/readable.ts` around lines 1269 - 1275, The code checks process.execArgv for "--experimental-stream-iter" which is user-mutable; replace that check in Readable.prototype[toAsyncStreamable] (the block referencing createBatchedAsyncIterator, normalizeBatch and kValidatedSource) with the immutable CLI-parsed startup bit used by module resolution; locate the existing startup flag exported into the JS runtime (the same flag the resolver/hardcoded gating logic reads) and use that boolean here to decide whether to require("internal/streams/iter/classic") and to throw $ERR_STREAM_ITER_MISSING_FLAG() when unset, so resolution and Readable[Symbol.for("Stream.toAsyncStreamable")] use the same write-once gate.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Outside diff comments:
In `@src/js/internal/streams/readable.ts`:
- Around line 1269-1275: The code checks process.execArgv for
"--experimental-stream-iter" which is user-mutable; replace that check in
Readable.prototype[toAsyncStreamable] (the block referencing
createBatchedAsyncIterator, normalizeBatch and kValidatedSource) with the
immutable CLI-parsed startup bit used by module resolution; locate the existing
startup flag exported into the JS runtime (the same flag the resolver/hardcoded
gating logic reads) and use that boolean here to decide whether to
require("internal/streams/iter/classic") and to throw
$ERR_STREAM_ITER_MISSING_FLAG() when unset, so resolution and
Readable[Symbol.for("Stream.toAsyncStreamable")] use the same write-once gate.
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: ASSERTIVE
Plan: Pro
Run ID: b5448f35-9cc3-43d3-b389-8c70e887ee7b
📒 Files selected for processing (14)
src/js/internal/streams/iter/broadcast.tssrc/js/internal/streams/iter/classic.tssrc/js/internal/streams/iter/push.tssrc/js/internal/streams/operators.tssrc/js/internal/streams/readable.tssrc/js/internal/webstreams_adapters.tssrc/jsc/ErrorCode.rssrc/jsc/bindings/ErrorCode.cppsrc/resolve_builtins/HardcodedModule.rssrc/resolve_builtins/lib.rssrc/runtime/cli/Arguments.rstest/expectations.txttest/js/bun/crypto/cipheriv-decipheriv.test.tstest/js/node/crypto/crypto.test.ts
💤 Files with no reviewable changes (1)
- src/js/internal/streams/operators.ts
Contributor
There was a problem hiding this comment.
Actionable comments posted: 1
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@src/runtime/cli/Arguments.rs`:
- Around line 1280-1283: The flag handling for
args.flag(b"--experimental-stream-iter") is only in the runtime-command branch
so bun build cannot enable resolver behavior; move the
bun_resolve_builtins::set_stream_iter_enabled(true) and
Bun__streamIterEnabled.store(...) out of the runtime-only block and into the
common/BUILD_PARAMS path (expose the flag in the build table as well) so that
when BUILD_PARAMS is constructed the resolver gate is reachable for both build
and runtime flows; update any parameter parsing that currently writes only to
RUNTIME_PARAMS_ to also populate the build params so the resolver sees the flag.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: ASSERTIVE
Plan: Pro
Run ID: 6e2c2bf0-ba6d-4442-80fa-49f5b67a715d
📒 Files selected for processing (5)
src/js/internal/streams/readable.tssrc/js/node/_http_incoming.tssrc/jsc/bindings/BunProcess.cppsrc/jsc/bindings/BunProcess.hsrc/runtime/cli/Arguments.rs
cirospaciari
force-pushed
the
claude/streams-node26-sync
branch
from
e66c8ab to
660f2b4
Compare
Member
Author
|
Finding from the macOS CI failures ( The With the exception fix in this PR the throw finally surfaces, and the child reports ~70 MB RSS growth over 5000 watch cycles (~14 KB per Marked expected-fail on darwin in 🤖 Generated with Claude Code |
Contributor
There was a problem hiding this comment.
Additional findings (outside current diff — PR may have been updated during review):
-
🔴
src/io/posix_event_loop.rs:1024-1031— Commit 5f0b651 ("test: expect fs.watch leak test failure on macOS") silently reverts merged PR #31821 insrc/io/posix_event_loop.rs— deletingderegistration_already_gone()and its three POSIX call sites, restoring the file byte-for-byte to its pre-#31821 state. This is unrelated to the node:stream sync, isn't mentioned in the commit message or PR description, and re-introduces the regression #31821 fixed (ENOENT/EBADF on deregistration early-returns before clearing registration flags, leaving the poll re-issuing doomed deletes). Please drop theposix_event_loop.rshunk from this PR.Extended reasoning...
What the bug is
Commit 5f0b651 in this PR is titled "test: expect fs.watch leak test failure on macOS" and the PR description discusses only the node:stream sync. But
git show 5f0b651a --statshows it touches two files:test/expectations.txt(the described change) andsrc/io/posix_event_loop.rs(+23/−39, never mentioned anywhere).The
posix_event_loop.rshunk is a byte-for-byte revert of merged PR #31821 ("Treat unregistering an already-removed poll registration as success", commit 29f2c7e, merged by Dylan Conway 2026-06-04). Verified viagit diff 29f2c7e6^..29f2c7e6 -- src/io/posix_event_loop.rsbeing the exact inverse of this PR's hunk for that file: the file blob goes from6a82cf50(post-#31821) back to630461f4(pre-#31821). PR #31821 was merged ~5h before this PR was opened, consistent with a rebase-conflict mis-resolution where the branch's older copy of the file was kept.Step-by-step proof
The diff for
src/io/posix_event_loop.rsin this PR:- Deletes the
deregistration_already_gone()helper (the function whose doc comment explained "the registration flags must still be cleared, which an error return would skip, leaving the poll claiming to be registered and re-issuing doomed deletes on later teardown calls"). - Removes its call site in the Linux epoll
CTL_DELbranch (lines 1024-1029): thematch sys::get_errno(ctl)that treatedderegistration_already_gone(e)as success is replaced with the oldif let Some(errno) = errno_sys(ctl, …) { return errno; }. - Removes its call site in the macOS kevent64
EV_DELETEbranch (lines 1124-1132): the loop that skippedEV_ERRORresults matchingderegistration_already_goneis replaced with the old unconditionalreturn kevent_change_error(...). - Removes its call site in the FreeBSD kevent
EV_DELETEbranch (lines 1171-1176): same pattern. - Restores the pre-#31821 doc comment on
kevent_change_errorat lines 58-61 ("the deinit path tolerates that race by discarding the returned error") — this is not new documentation; it's the old comment that #31821 deleted, reappearing because the file went back to its prior state.
This is the entirety of #31821, undone.
Why existing code doesn't prevent it
There is no test asserting the tolerance behavior (per #31821's own commit message, "there's no JS-observable surface" — callers discard the error). The file change is buried in a 180-file PR and the commit message describes only the test-expectation half. Per CLAUDE.md: "Before deleting odd-looking code … git-blame why it was written; it is usually load-bearing."
Functional impact
Per #31821's commit message: when an fd is closed while still registered (or a pty master close marks the slave's knotes
EV_EOF|EV_ONESHOTand the kernel auto-deletes them — "happens on every terminal window/tab close while a tty is polled"), the deregistration syscall returns ENOENT/EBADF. With the tolerance removed,unregister_with_fd_implnow early-returns at lines 1027-1029 / 1127-1132 / 1174-1176 before reaching the flag-clearing block at lines 1179-1184.The
deinitpath itself is fine (deinit_possibly_deferat line 408 doeslet _ = self.unregister(...)thenself.flags = FlagsSet::empty()at line 412, clearing flags regardless). But the non-deinit callers —PipeReader::pause()at PipeReader.rs:308 anddns.rs:4972— callunregister()directly and discard the result without a subsequent flag clear. So after a failed unregister the poll'sis_registered()keeps returning true, and a laterdeinit()re-issues another doomedEV_DELETE— exactly the double-delete #31821 instrumented ("observed via instrumentation: the same poll failing the delete twice during one pty teardown"). libuv ignores these same errnos for the same reason.The runtime impact is mild (callers discard the error; no crash/hang). But silently reverting another contributor's merged fix via an unrelated commit in a 180-file PR is exactly what review must catch — it would otherwise re-land as a regression with no audit trail.
How to fix
Drop the
src/io/posix_event_loop.rshunk from commit 5f0b651 (e.g.git checkout origin/main -- src/io/posix_event_loop.rsand amend). The test-expectation half of that commit is fine and should stay. - Deletes the
Contributor
There was a problem hiding this comment.
Actionable comments posted: 13
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (2)
src/js/node/_http_server.ts (1)
1552-1559:⚠️ Potential issue | 🟠 Major | ⚡ Quick winFall back to
this.socketwhen the request already detached its alias.
ServerResponse.end()passesreq.sockethere. AfterIncomingMessage._destroy()sets that alias tonull, this guard skips clearing the live response socket’s_httpMessage/"close"listener and only nullsthis.socket, leaving a stale back-reference on the keep-alive socket.🐛 Proposed fix
ServerResponse.prototype.detachSocket = function (socket) { + socket ??= this.socket; // socket can be null when the stream destroyer detached the request's // socket (req.socket = null) before the response finished. if (socket && socket._httpMessage === this) { if (socket[kCloseCallback]) socket[kCloseCallback] = undefined; socket.removeListener("close", onServerResponseClose); socket._httpMessage = null; } this.socket = null; };🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@src/js/node/_http_server.ts` around lines 1552 - 1559, ServerResponse.prototype.detachSocket currently only operates on the passed-in socket, which can be null after IncomingMessage._destroy(), leaving a stale back-reference on the real keep-alive socket; change the logic in detachSocket to fall back to the live socket (use socket || this.socket) and operate on that alternativeSocket: check alternativeSocket._httpMessage === this, clear alternativeSocket[kCloseCallback], removeListener("close", onServerResponseClose) from alternativeSocket, set alternativeSocket._httpMessage = null, and ensure this.socket is also nulled so no stale references remain (refer to ServerResponse.prototype.detachSocket, kCloseCallback, onServerResponseClose, _httpMessage, and ServerResponse.end).src/js/internal/webstreams_adapters.ts (1)
576-597:⚠️ Potential issue | 🟡 Minor | ⚡ Quick winValidate mapped readable type before fast paths.
readableOptions.typeis only validated when the readable branch callsnewReadableStreamFromStreamReadable(). The destroyed/non-readable branches bypass that and pass the value directly intonew ReadableStream(...), which can produce inconsistent error behavior for invalid values.Suggested fix
function newReadableWritablePairFromDuplex(duplex, options = kEmptyObject) { @@ const readableOptions = { type: options.readableType }; if (options.readableType == null && options.type != null) { // 'options.type' is a deprecated alias for 'options.readableType'. emitDEP0201(); readableOptions.type = options.type; } + if (readableOptions.type !== undefined) { + const typeArgName = + options.readableType == null && options.type != null + ? "options.type" + : "options.readableType"; + validateOneOf(readableOptions.type, typeArgName, ["bytes", undefined]); + }🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@src/js/internal/webstreams_adapters.ts` around lines 576 - 597, The code constructs ReadableStream with readableOptions.type in the destroyed and non-readable fast paths without validating it, causing inconsistent errors; before any fast-path creation (inside isDestroyed and the non-readable branch where new ReadableStream is used), run the same validation/conversion logic used by newReadableStreamFromStreamReadable on readableOptions.type (extract or call the validator used there) and use the validated value when constructing new ReadableStream so invalid types throw the same error as in newReadableStreamFromStreamReadable; update references around readableOptions, isDestroyed(duplex), isReadable(duplex), and newReadableStreamFromStreamReadable accordingly.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@src/js/builtins/ProcessObjectInternals.ts`:
- Around line 380-387: Iterating tickInitHooks directly can skip callbacks if
hooks mutate the array during iteration; fix by taking a snapshot (e.g., const
hooks = tickInitHooks.slice()) after obtaining asyncId (from
asyncHooksTick.newAsyncId()) and then iterate over that snapshot, calling each
hook with (asyncId, "TickObject", 0, tock) so mutations to tickInitHooks inside
callbacks do not affect the current dispatch.
In `@src/js/builtins/ReadableStreamInternals.ts`:
- Around line 1609-1617: The BYOB read-into queue draining logic currently only
runs in cancel() causing BYOB readers to hang on EOF; move the readIntoRequests
drain into the shared readableStreamClose() path (use
$getByIdDirectPrivate/$isReadableStreamBYOBReader to detect a BYOB reader and
$getByIdDirectPrivate to read its "readIntoRequests", then $putByIdDirectPrivate
to replace it with a new FIFO and $fulfillPromise pending requests with {value:
undefined, done: true}) and update cancel() to call/reuse readableStreamClose()
so both close and cancel share the same drain behavior for readIntoRequests.
In `@src/js/internal/streams/iter/consumers.ts`:
- Around line 379-437: The ready.shift() draining is O(n^2); replace the queue
with a head-pointer ring: keep the existing ready array but add a numeric head
(e.g., readyHead = 0) and change pushes to ready.push(...) as-is, then drain by
reading items with const item = ready[readyHead++]; treat FIFO emptiness as
readyHead < ready.length (and update the outer loop condition to use readyHead
vs ready.length); when readyHead grows large (or readyHead === ready.length)
reset the buffer by doing ready = ready.slice(readyHead); readyHead = 0 to avoid
unbounded memory; update both the synchronous drain loop (where ready.shift() is
used) and any code paths that check ready.length (e.g., the outer while
condition and waitResolve logic) to use readyHead/ready.length semantics; keep
onSettled, iterator.next().then, and the error-push logic unchanged aside from
the new empty-checks.
In `@src/js/internal/streams/iter/push.ts`:
- Around line 213-237: The method `#createPendingWrite` currently pushes the entry
into `#pendingWrites` before validating/using the provided signal; change the flow
so the entry is only enqueued after any signal work succeeds: if signal is
falsy, push immediately; if signal is provided, prepare onAbort and the wrapped
resolve/reject, attempt to addEventListener inside a try/catch, and only push
the entry into `#pendingWrites` after addEventListener succeeds (on failure,
throw/reject without mutating `#pendingWrites`); ensure you still remove the
listener in the wrapped entry.resolve/entry.reject and remove the entry from the
queue in onAbort as before.
In `@src/js/node/_http_incoming.ts`:
- Around line 413-421: The connection getter must be kept in sync with the
socket getter: stop using the `??=` pattern on the same backing field
`this[fakeSocketSymbol]` (which resurrects a FakeSocket when the field is
intentionally set to null) and instead match the socket getter's logic (only
create a new FakeSocket when `this[fakeSocketSymbol] === undefined`, otherwise
return the current value including null); update the `connection` accessor (and
any code that sets `stream.socket`) to reference `fakeSocketSymbol`,
`FakeSocket`, and the same undefined-only lazy-init behavior so `socket` and
`connection` remain alias-consistent.
In `@src/js/node/async_hooks.ts`:
- Around line 382-403: The current disable() removes the shared function
reference stored in enabledInit which can remove another hook using the same
init callback; instead create and store a per-instance wrapper/token when
registering init (e.g. assign a unique wrapper value alongside enabledInit
inside the installation path where enabledInit = init and
tickInitHooks.push(init)), push that wrapper/token into
require("internal/async_hooks_tick").tickInitHooks, and in disable() remove that
specific wrapper/token (use the stored wrapper variable rather than enabledInit)
and clear the wrapper variable; adjust any references that read enabledInit to
use the instance wrapper token so other hooks with the same raw init function
aren’t affected.
In `@src/js/node/net.ts`:
- Around line 1368-1397: The synchronous fd writers fdSyncWrite and fdSyncWritev
perform writes but never update the socket's byte counters; update
this[kBytesWritten] and this._bytesDispatched with the number of bytes returned
from each fs.writeSync call so socket.bytesWritten and _bytesDispatched reflect
the writes. In both fdSyncWrite and fdSyncWritev, capture the return value from
fs.writeSync for each write chunk/iteration, add that value to
this[kBytesWritten] and this._bytesDispatched (accumulating across the inner
loop and across v-write entries), then proceed as before with offset and
callback handling.
- Around line 1157-1170: When releasing the adopted fd (this[kSyncWriteFd]) also
restore the instance's sync-write function overrides so future reconnects don't
keep using the stale path: after clearing this[kSyncWriteFd] in the Socket
shutdown/cleanup code, restore or remove the fdSyncWrite and fdSyncWritev
overrides (the properties set earlier around lines where fdSyncWrite /
fdSyncWritev were installed) back to their default/original implementations (or
delete them from the instance) so subsequent Socket.connect() calls use the
fresh handle's normal write path instead of the stale sync-fd handlers.
In `@test/js/node/stream/node-stream.test.js`:
- Around line 582-589: The test spawns a subprocess into variable proc
(Bun.spawn) and only drains proc.stdout, risking a deadlock because stderr is
piped but never read; update the Promise.all call that awaits proc.stdout.text()
and proc.exited to also include proc.stderr.text() so both stdout and stderr are
consumed concurrently (i.e., use Promise.all([proc.stdout.text(),
proc.stderr.text(), proc.exited])) and adjust the destructuring accordingly to
capture and ignore or assert stderr as needed.
In `@test/js/node/test/parallel/test-stream-iter-broadcast-basic.js`:
- Around line 221-230: The test testCancelWithFalsyReason is asserting that
bc.cancel(0) should reject (expects result 0) but upstream semantics treat falsy
cancel reasons as normal completion; update the test to expect successful
completion instead of rejection: call text(consumer) without the .catch(...) (or
change the promise handling) and assert that the consumer completes normally
(e.g., resolved value rather than 0). Adjust assertions in
testCancelWithFalsyReason to reflect completion behavior for broadcast().cancel
when given a falsy reason.
In `@test/js/node/test/parallel/test-stream-iter-from-sync.js`:
- Around line 195-207: The test claims zero-copy for DataView inputs but only
checks content; either enforce aliasing or drop the claim. In
testFromSyncDataView, if zero-copy is required, add assertions that the returned
Uint8Array (from fromSync(view), e.g. batches[0][0]) shares the same underlying
ArrayBuffer and offsets as the original DataView (compare .buffer === buf and
.byteOffset/.byteLength with view.byteOffset/view.byteLength); otherwise remove
or reword the comment about zero-copy and leave only the content assertion.
In `@test/js/node/test/parallel/test-stream-iter-share-from.js`:
- Around line 103-125: The testShareBlockBackpressure test currently starts both
consumers together so the slow consumer never lags; modify it to force the fast
consumer to run ahead before the slow one is allowed to pull. Specifically, in
testShareBlockBackpressure (the shared = share(...), fast = shared.pull(), slow
= shared.pull() block) advance the fast reader (call fast.next() repeatedly or
await text(fast) partially) to consume enough items to exceed highWaterMark
(e.g., consume 2–3 items) and only then start/await the slow reader (or gate
slow.next() with a deferred promise) so the source must stall under
backpressure:'block' and the test actually verifies blocking behavior. Ensure
final assertions still check both readers receive all items.
In `@test/js/node/test/parallel/test-stream-iter-writable-from.js`:
- Line 270: Replace the non-deterministic sleep with an event-based wait: in the
tests testDestroyDelegatesToFail, testDestroyWithoutError, testDestroyWithError,
and testDestroyWithoutFail, remove the await new Promise(resolve =>
setTimeout(resolve, 10)) and instead await the stream close event (for example
using once(writable, 'close') from the 'events' module or stream.finished) so
the test proceeds only after writable emits 'close' following destroy().
---
Outside diff comments:
In `@src/js/internal/webstreams_adapters.ts`:
- Around line 576-597: The code constructs ReadableStream with
readableOptions.type in the destroyed and non-readable fast paths without
validating it, causing inconsistent errors; before any fast-path creation
(inside isDestroyed and the non-readable branch where new ReadableStream is
used), run the same validation/conversion logic used by
newReadableStreamFromStreamReadable on readableOptions.type (extract or call the
validator used there) and use the validated value when constructing new
ReadableStream so invalid types throw the same error as in
newReadableStreamFromStreamReadable; update references around readableOptions,
isDestroyed(duplex), isReadable(duplex), and newReadableStreamFromStreamReadable
accordingly.
In `@src/js/node/_http_server.ts`:
- Around line 1552-1559: ServerResponse.prototype.detachSocket currently only
operates on the passed-in socket, which can be null after
IncomingMessage._destroy(), leaving a stale back-reference on the real
keep-alive socket; change the logic in detachSocket to fall back to the live
socket (use socket || this.socket) and operate on that alternativeSocket: check
alternativeSocket._httpMessage === this, clear
alternativeSocket[kCloseCallback], removeListener("close",
onServerResponseClose) from alternativeSocket, set
alternativeSocket._httpMessage = null, and ensure this.socket is also nulled so
no stale references remain (refer to ServerResponse.prototype.detachSocket,
kCloseCallback, onServerResponseClose, _httpMessage, and ServerResponse.end).
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: ASSERTIVE
Plan: Pro
Run ID: 12ad007f-2b22-408e-b158-d95e9ecc5945
📒 Files selected for processing (179)
src/js/builtins/ProcessObjectInternals.tssrc/js/builtins/ReadableStreamInternals.tssrc/js/internal-for-testing.tssrc/js/internal/async_context_frame.tssrc/js/internal/async_hooks.tssrc/js/internal/async_hooks_tick.tssrc/js/internal/fs/streams.tssrc/js/internal/http.tssrc/js/internal/streams/duplex.tssrc/js/internal/streams/duplexify.tssrc/js/internal/streams/from.tssrc/js/internal/streams/iter/broadcast.tssrc/js/internal/streams/iter/classic.tssrc/js/internal/streams/iter/consumers.tssrc/js/internal/streams/iter/duplex.tssrc/js/internal/streams/iter/from.tssrc/js/internal/streams/iter/pull.tssrc/js/internal/streams/iter/push.tssrc/js/internal/streams/iter/ringbuffer.tssrc/js/internal/streams/iter/share.tssrc/js/internal/streams/iter/transform.tssrc/js/internal/streams/iter/types.tssrc/js/internal/streams/iter/utils.tssrc/js/internal/streams/operators.tssrc/js/internal/streams/pipeline.tssrc/js/internal/streams/readable.tssrc/js/internal/streams/writable.tssrc/js/internal/webstreams_adapters.tssrc/js/node/_http_client.tssrc/js/node/_http_incoming.tssrc/js/node/_http_server.tssrc/js/node/async_hooks.tssrc/js/node/net.tssrc/js/node/stream.iter.tssrc/js/node/zlib.iter.tssrc/jsc/ErrorCode.rssrc/jsc/bindings/BunProcess.cppsrc/jsc/bindings/ErrorCode.cppsrc/jsc/bindings/ErrorCode.tssrc/jsc/bindings/JSCommonJSModule.cppsrc/resolve_builtins/HardcodedModule.rssrc/resolve_builtins/lib.rssrc/runtime/cli/Arguments.rssrc/runtime/jsc_hooks.rssrc/runtime/node/zlib/NativeBrotli.rstest/expectations.txttest/js/bun/crypto/cipheriv-decipheriv.test.tstest/js/node/crypto/crypto.test.tstest/js/node/net/node-net.test.tstest/js/node/stream/node-stream-uint8array.test.tstest/js/node/stream/node-stream.test.jstest/js/node/test/common/index.jstest/js/node/test/parallel/test-crypto-cipheriv-decipheriv.jstest/js/node/test/parallel/test-stream-add-abort-signal.jstest/js/node/test/parallel/test-stream-big-push.jstest/js/node/test/parallel/test-stream-compose.jstest/js/node/test/parallel/test-stream-construct.jstest/js/node/test/parallel/test-stream-consumers.jstest/js/node/test/parallel/test-stream-destroy.jstest/js/node/test/parallel/test-stream-drop-take.jstest/js/node/test/parallel/test-stream-duplex-destroy.jstest/js/node/test/parallel/test-stream-duplex-from.jstest/js/node/test/parallel/test-stream-duplex-readable-writable.jstest/js/node/test/parallel/test-stream-duplex-writable-finished.jstest/js/node/test/parallel/test-stream-duplex.jstest/js/node/test/parallel/test-stream-duplexpair.jstest/js/node/test/parallel/test-stream-end-paused.jstest/js/node/test/parallel/test-stream-filter.jstest/js/node/test/parallel/test-stream-finished-async-local-storage.jstest/js/node/test/parallel/test-stream-finished-bindAsyncResource-path.jstest/js/node/test/parallel/test-stream-finished-default-path.jstest/js/node/test/parallel/test-stream-forEach.jstest/js/node/test/parallel/test-stream-iter-broadcast-backpressure.jstest/js/node/test/parallel/test-stream-iter-broadcast-basic.jstest/js/node/test/parallel/test-stream-iter-broadcast-coverage.jstest/js/node/test/parallel/test-stream-iter-broadcast-from.jstest/js/node/test/parallel/test-stream-iter-consumers-bytes.jstest/js/node/test/parallel/test-stream-iter-consumers-merge.jstest/js/node/test/parallel/test-stream-iter-consumers-tap.jstest/js/node/test/parallel/test-stream-iter-consumers-text.jstest/js/node/test/parallel/test-stream-iter-cross-realm.jstest/js/node/test/parallel/test-stream-iter-disabled.jstest/js/node/test/parallel/test-stream-iter-duplex.jstest/js/node/test/parallel/test-stream-iter-from-async.jstest/js/node/test/parallel/test-stream-iter-from-coverage.jstest/js/node/test/parallel/test-stream-iter-from-sync.jstest/js/node/test/parallel/test-stream-iter-from-writable-cache-options.jstest/js/node/test/parallel/test-stream-iter-namespace.jstest/js/node/test/parallel/test-stream-iter-pipeto-edge.jstest/js/node/test/parallel/test-stream-iter-pipeto-signal.jstest/js/node/test/parallel/test-stream-iter-pipeto-writev.jstest/js/node/test/parallel/test-stream-iter-pipeto.jstest/js/node/test/parallel/test-stream-iter-pull-async.jstest/js/node/test/parallel/test-stream-iter-pull-sync.jstest/js/node/test/parallel/test-stream-iter-push-backpressure.jstest/js/node/test/parallel/test-stream-iter-push-basic.jstest/js/node/test/parallel/test-stream-iter-push-writer.jstest/js/node/test/parallel/test-stream-iter-readable-interop-disabled.jstest/js/node/test/parallel/test-stream-iter-readable-interop.jstest/js/node/test/parallel/test-stream-iter-share-async.jstest/js/node/test/parallel/test-stream-iter-share-coverage.jstest/js/node/test/parallel/test-stream-iter-share-from.jstest/js/node/test/parallel/test-stream-iter-share-sync.jstest/js/node/test/parallel/test-stream-iter-sharedarraybuffer.jstest/js/node/test/parallel/test-stream-iter-to-readable.jstest/js/node/test/parallel/test-stream-iter-transform-compat.jstest/js/node/test/parallel/test-stream-iter-transform-coverage.jstest/js/node/test/parallel/test-stream-iter-transform-errors.jstest/js/node/test/parallel/test-stream-iter-transform-output.jstest/js/node/test/parallel/test-stream-iter-transform-roundtrip.jstest/js/node/test/parallel/test-stream-iter-transform-sync.jstest/js/node/test/parallel/test-stream-iter-validation.jstest/js/node/test/parallel/test-stream-iter-writable-from.jstest/js/node/test/parallel/test-stream-iter-writable-interop.jstest/js/node/test/parallel/test-stream-map.jstest/js/node/test/parallel/test-stream-pipe-await-drain.jstest/js/node/test/parallel/test-stream-pipe-flow.jstest/js/node/test/parallel/test-stream-pipe-objectmode-to-non-objectmode.jstest/js/node/test/parallel/test-stream-pipeline-http2.jstest/js/node/test/parallel/test-stream-pipeline-listeners.jstest/js/node/test/parallel/test-stream-pipeline.jstest/js/node/test/parallel/test-stream-push-strings.jstest/js/node/test/parallel/test-stream-readable-aborted.jstest/js/node/test/parallel/test-stream-readable-async-iterators.jstest/js/node/test/parallel/test-stream-readable-compose.jstest/js/node/test/parallel/test-stream-readable-didRead.jstest/js/node/test/parallel/test-stream-readable-dispose.jstest/js/node/test/parallel/test-stream-readable-emittedReadable.jstest/js/node/test/parallel/test-stream-readable-ended.jstest/js/node/test/parallel/test-stream-readable-event.jstest/js/node/test/parallel/test-stream-readable-hwm-0-no-flow-data.jstest/js/node/test/parallel/test-stream-readable-infinite-read.jstest/js/node/test/parallel/test-stream-readable-needReadable.jstest/js/node/test/parallel/test-stream-readable-readable-one.jstest/js/node/test/parallel/test-stream-readable-strategy-option.jstest/js/node/test/parallel/test-stream-readable-to-web-byob.jstest/js/node/test/parallel/test-stream-readable-to-web-termination-byob.jstest/js/node/test/parallel/test-stream-readable-to-web-termination.jstest/js/node/test/parallel/test-stream-readable-to-web.mjstest/js/node/test/parallel/test-stream-reduce.jstest/js/node/test/parallel/test-stream-toArray.jstest/js/node/test/parallel/test-stream-transform-destroy.jstest/js/node/test/parallel/test-stream-transform-final.jstest/js/node/test/parallel/test-stream-transform-flush-data.jstest/js/node/test/parallel/test-stream-typedarray.jstest/js/node/test/parallel/test-stream-uint8array.jstest/js/node/test/parallel/test-stream-unpipe-event.jstest/js/node/test/parallel/test-stream-unshift-empty-chunk.jstest/js/node/test/parallel/test-stream-unshift-read-race.jstest/js/node/test/parallel/test-stream-wrap-drain.jstest/js/node/test/parallel/test-stream-wrap-encoding.jstest/js/node/test/parallel/test-stream-wrap.jstest/js/node/test/parallel/test-stream-writable-change-default-encoding.jstest/js/node/test/parallel/test-stream-writable-decoded-encoding.jstest/js/node/test/parallel/test-stream-writable-destroy.jstest/js/node/test/parallel/test-stream-writable-ended-state.jstest/js/node/test/parallel/test-stream-writable-finished-state.jstest/js/node/test/parallel/test-stream-writable-finished.jstest/js/node/test/parallel/test-stream-writable-null.jstest/js/node/test/parallel/test-stream-writable-samecb-singletick.jstest/js/node/test/parallel/test-stream-writable-write-error.jstest/js/node/test/parallel/test-stream-writev.jstest/js/node/test/parallel/test-stream2-base64-single-char-read-end.jstest/js/node/test/parallel/test-stream2-basic.jstest/js/node/test/parallel/test-stream2-compatibility.jstest/js/node/test/parallel/test-stream2-httpclient-response-end.jstest/js/node/test/parallel/test-stream2-objects.jstest/js/node/test/parallel/test-stream2-push.jstest/js/node/test/parallel/test-stream2-read-correct-num-bytes-in-utf8.jstest/js/node/test/parallel/test-stream2-readable-wrap.jstest/js/node/test/parallel/test-stream2-transform.jstest/js/node/test/parallel/test-stream2-writable.jstest/js/node/test/parallel/test-stream3-cork-end.jstest/js/node/test/parallel/test-stream3-cork-uncork.jstest/js/node/test/parallel/test-stream3-pause-then-read.jstest/js/node/test/parallel/test-stream3-pipeline-async-iterator.jstest/js/node/test/parallel/test-zlib-flush-write-sync-interleaved.jstest/js/node/test/sequential/test-stream2-stderr-sync.jstest/napi/node-napi-tests/test/node-api/test_uv_threadpool_size/do.test.ts
💤 Files with no reviewable changes (8)
- test/js/node/test/parallel/test-stream-pipe-flow.js
- test/js/node/test/parallel/test-stream-pipe-await-drain.js
- test/js/node/test/parallel/test-stream-pipe-objectmode-to-non-objectmode.js
- test/js/node/test/parallel/test-stream-map.js
- test/js/node/test/parallel/test-stream-pipeline-listeners.js
- src/js/internal/streams/operators.ts
- test/js/node/test/parallel/test-stream-pipeline-http2.js
- test/js/node/test/parallel/test-stream-iter-writable-interop.js
cirospaciari
force-pushed
the
claude/streams-node26-sync
branch
from
22307d5 to
2305b6e
Compare
cirospaciari
force-pushed
the
claude/streams-node26-sync
branch
from
f6fd167 to
22fd91f
Compare
Sync the vendored test/js/node/test stream tests (61 files, plus the zlib/crypto transform-stream consumers) to their node v26.3.0 versions and fix the behavior gaps they expose: - readable: read() returns one buffered chunk at a time instead of concatenating the whole buffer (nodejs/node#60441). Port the "fast path for buffers" branch of howMuchToRead. - writable: write(string, 'buffer') now throws ERR_UNKNOWN_ENCODING. - duplexify: an error on the readable side of a { readable, writable } pair now destroys the writable side (was destroying the readable side itself - port typo). - Duplex.toWeb: forward options; support readableType: 'bytes' and the deprecated type alias (emits DEP0201). - Readable.toWeb: support BYOB byte streams (options.type = 'bytes'); readable completion no longer waits for a half-open writable side of a duplex, so the web stream closes when the readable side ends. - ErrorCode: format "Unknown encoding:" values like node's %s (inspect objects -> "{}", not "[object Object]"), and drop the cipher name from ERR_CRYPTO_UNKNOWN_CIPHER to match node's exact-match-asserted "Unknown cipher" message. test/js/node/stream/node-stream-uint8array.test.ts asserted the old read() concatenation behavior; updated to match node v26.3.0. Verified: all 193 vendored stream tests pass; full node-compat sweep (2455 entries) shows no regressions vs the unmodified binary; bun-native stream/web-stream test files pass.
…3.0 stream tests
Brings the vendored stream suite to all 251 node v26.3.0 stream test
files (249 parallel + 2 sequential); 238 pass.
New feature: node's experimental iterable-streams API behind
--experimental-stream-iter (also BUN_EXPERIMENTAL_STREAM_ITER=1):
- node:stream/iter and node:zlib/iter entry modules plus the 13
internal/streams/iter/* modules ported from node v26.3.0 (~7.5k
lines: push/pull/from/share/broadcast/transform/consumers/classic/
duplex/ringbuffer/types/utils).
- Resolution gated in the module loader (HardcodedModule + alias gate):
without the flag, node:stream/iter reports "No such built-in module"
and bare stream/iter falls through to filesystem resolution, matching
node. Readable.prototype[Symbol.for("Stream.toAsyncStreamable")]
interop with lazy flag check (ERR_STREAM_ITER_MISSING_FLAG).
- New error codes: ERR_OPERATION_FAILED, ERR_STREAM_ITER_MISSING_FLAG,
ERR_INVALID_ARG_VALUE RangeError variant. src/jsc/ErrorCode.rs had
drifted from the generated C++ list; renumbered all ordinals from
ErrorCode+List.h (a stale mirror made native code throw adjacent
error codes, e.g. ERR_INVALID_HANDLE_TYPE for ERR_INVALID_HTTP_TOKEN).
Behavior fixes exposed by the newly vendored tests:
- readable: fromList decoder slow path compared n against the buffer
array length instead of the chunk length, over-returning on read(n)
with an encoding set.
- readable: resume()/pause() are no-ops on destroyed streams.
- readable: compose() moved off the operators registry onto
Readable.prototype.compose returning the Duplex directly (node
v26.3.0 layout); registry wrapper hid the writable side.
- pipeline: an AbortError no longer wins over the real failure
(node's error precedence clause).
- from: _destroy now chains the instance's original _destroy (e.g.
duplexify's abort) before closing the iterator, so generators parked
on a pending source settle; errors aggregate via aggregateTwoErrors.
- duplexify: fromAsyncGen destroy unblocks a generator awaiting the
next write/final signal (port of node's resolver flush).
- writable: write(string, "buffer") throws ERR_UNKNOWN_ENCODING.
- web adapters: ReadableStream cancel settles pending BYOB reads with
done:true; Readable.toWeb completion only waits for the readable
side of a half-open duplex.
- http: client response EOF is pushed synchronously with the final
chunk (single 'readable' for data+EOF, like node's parser);
ClientRequest.destroy(err) emits 'error' on the request and destroys
an incomplete response with ECONNRESET (node's socketCloseListener
semantics); IncomingMessage.socket getter no longer resurrects a
FakeSocket after the stream destroyer detached it (null signals the
connection must outlive the request); ServerResponse.detachSocket
tolerates that null.
- zlib: brotli decoder error codes match node ("ERR_" + enum suffix
with its leading underscore, e.g. ERR__ERROR_FORMAT_PADDING_2).
- crypto: ERR_CRYPTO_UNKNOWN_CIPHER message is "Unknown cipher"
(node dropped the cipher name).
13 vendored files still fail, all for runtime gaps outside the stream
JS layer: 7 need node-private internals (internal/js_stream_socket,
internal/test/binding, internal/async_context_frame, --expose-internals
userland requires), 2 need async_hooks init events / synchronous fd
adoption in net.Socket, 2 need per-segment http client body delivery
(native fetch path coalesces), and the 2 stream-iter disabled tests
need the CLI to report uncaught require failures under -e (currently
exits 0 silently; pre-existing).
…failures
- --expose-internals: vendored tests can require node-internal modules.
bun:internal-for-testing gains an exposedInternals map (static requires);
the bun-patched test common/index.js installs a require interceptor for
internal/* when a test declares the flag. New internal shims:
internal/async_context_frame (enabled=true, current() undefined - bun
tracks context natively and never materializes frames) and
internal/async_hooks (enabledHooksExist backed by real enable/disable
tracking).
- async_hooks: createHook init events now fire for TickObject resources.
process.nextTick emits one init per call when an init hook is enabled
(array-length check on the hot path, nothing when disabled). Hooks with
only init no longer warn; before/after/destroy still do.
- stdio: process.stdout/stderr writers are born constructed (instance
_construct removed on the already-open-fd fast path), so they no longer
allocate construct TickObjects - node's stdio (net.Socket/tty) never has
a _construct. Without this, lazily creating stdout inside an init-hooked
window miscounts tick objects.
- fs: ReadStream._read marks kIsPerformingIO and bails out of the read
callback when the stream was destroyed mid-read (node parity), so
close() between reads yields ERR_STREAM_PREMATURE_CLOSE instead of a
clean end; _destroy waits for in-flight IO (kIoDone) before closing the
fd.
- net: Socket({fd}) adopts pipe/character-device/file/socketpair fds with
synchronous write(2)-based _write, matching node's effective stdio
semantics (writes survive an immediate process.exit()). Network-socket
fd adoption (handle layer) remains unsupported.
- cjs eval entry: use the out-param JSC::evaluate overload and rethrow.
The 3-arg convenience overload catches the exception into a discarded
NakedPtr, so an uncaught require() failure in `bun -e` (CJS-transpiled)
exited 0 with no output; now it reports and exits nonzero.
Vendored stream suite: 247/251 node v26.3.0 files pass. Remaining 4:
test-stream-wrap{,-drain,-encoding} need a net.Socket({handle}) libuv
compat layer (internal/js_stream_socket); test-stream-pipeline needs
per-socket-read http client body delivery (native fetch path coalesces).
…ator, _-prefix unused params - clippy disallows std::env::var_os in bun_resolve_builtins (and bun_core is not a dependency there); stream/iter is now gated on the CLI flag only, exactly like node. The readable interop check drops the env clause to stay consistent with the resolver. - operators.ts: remove the dead compose() and its now-unused imports (Readable.prototype.compose replaced the registry entry). - iter ports: underscore-prefix interface-compliance params that are intentionally unused.
…DE_STR table The renumbering pass missed three things the review caught: - MODULE_NOT_FOUND (the bare code, distinct from ERR_MODULE_NOT_FOUND) was mapped to the ERR_-prefixed entry's ordinal (155) instead of its own (268). - COUNT stayed at 312; it is 315 with the three new codes. - The CODE_STR table was not re-aligned; it is now regenerated wholesale from the generated ErrorCode+Data.h ordering, so code_str() is correct for every ordinal including the three insertions.
Node validates the options object before reading options.type (lib/internal/webstreams/adapters.js); the port skipped it, so an explicit null produced a raw property-access TypeError instead of ERR_INVALID_ARG_TYPE.
Review caught that the resolver gate scanned raw argv, so the flag was honored even when it appeared after the script name, where node treats it as a script argument (and process.execArgv correctly excludes it). The flag is now a declared CLI param; parsing sets an AtomicBool in bun_resolve_builtins, and the resolver/alias gates read that instead of scanning argv. `bun script.js --experimental-stream-iter` no longer enables the module.
…xpectations for known stream gaps
The two bun-native crypto tests asserted the pre-v26 read()-concatenates
behavior (single cipher.read() returning the whole ciphertext). Node
v26.3.0 fails the same pattern (verified: 48 of 64 bytes on first read,
then ERR_OSSL_BAD_DECRYPT). Both now drain the stream; with that, the
cipher event order is byte-identical to node v26.3.0 - including the
ordering the old test marked as a TODO bug.
expectations.txt gains the four vendored stream tests blocked on
missing native subsystems (http per-segment delivery; the
net.Socket({handle}) libuv compat layer), so CI tracks them as known
rather than new failures. test-stream-pipeline is SKIP because its
failure mode is a timeout.
…rWithCode
The RangeError variant had no case in the switch, so
$ERR_INVALID_ARG_VALUE_RangeError("options.encoding", value) fell through
to the default branch and produced an error whose message was just the
property name. Route it through the same formatter as ERR_INVALID_ARG_VALUE,
parameterized on the error code so the RangeError structure is used.
A bare { fd } is the connect({ fd }) path: child_process extra stdio
creates sockets that attach a native duplex handle in
Socket.prototype.connect. Adopting in the constructor ended their readable
side with push(null) and replaced the native write path, which broke
puppeteer's CDP pipe (fds 3/4) and, on Windows, threw EBADF from
fstatSync on uv pseudo-fds. Node's own stdio-style fd sockets always pass
explicit flags (e.g. new Socket({ fd: 2, readable: false, writable: true })),
so keying adoption on writable: true keeps the stdio semantics while
restoring the native path for bare fds.
Also address review findings on the adoption path:
- close the adopted fd in _destroy (fds 0-2 exempt, matching
SyncWriteStream's autoClose gating); node closes the wrapping libuv
handle in the equivalent path
- rethrow fstat failures as ERR_INVALID_FD_TYPE('UNKNOWN'), matching
node's createHandle -> guessHandleType behavior
reifyAllStaticProperties (hit by the node:process ESM module generator)
performs no exception check between PropertyCallback invocations, so any
builder that leaves a throw-scope obligation aborts debug/ASAN builds
under BUN_JSC_validateExceptionChecks once the next builder declares a
scope. Convert the throw-capable builders (config, _preload_modules /
moduleLoadList stub, env, mainModule, nextTick, channel) to the
DECLARE_TOP_EXCEPTION_SCOPE + clear pattern already used by the stdio
builders. Exposed by the new test-stream-readable-to-web.mjs, which is
the first vendored .mjs test to import node:process.
Also account for two tests that only passed while bun -e swallowed
child-process exceptions (fixed earlier in this branch):
- test-inspector-enabled.js needs process.binding('inspector'); mark it
as a known failure in test/expectations.txt
- test_uv_threadpool_size/node-options.js runs the same addon as test.js
(unsupported uv_sleep) and its env fixture was never vendored; mark it
todo alongside test.js
Per the spec (whatwg/streams #1103), ReadableStreamClose performs each pending readIntoRequest's close steps with undefined, resolving them as { value: undefined, done: true }. We only did this on the cancel path, so a BYOB read pending at ordinary EOF (controller.close()) hung forever. Move the drain into readableStreamClose and let cancel reuse it.
Readable.prototype[toAsyncStreamable] gated on
process.execArgv.includes("--experimental-stream-iter"), which user code
can mutate. Expose the resolver's write-once AtomicBool through a tiny
$cpp accessor (createStreamIterEnabledFlag) and consult that instead.
- _http_incoming: get connection() used ??= on the shared backing field, so reading req.connection from an 'aborted' listener resurrected a FakeSocket after the stream destroyer assigned stream.socket = null and defeated the socket-null teardown gate. Make it a strict alias of the socket getter, like node. - async_hooks: register a per-instance wrapper in tickInitHooks so two hooks sharing the same init function stay independently removable (removal is by identity and could otherwise reorder the survivor). - nextTick init dispatch: snapshot the hook list (mutations from inside a hook must not affect the in-flight emit, matching node's tmp_array staging) and treat a throwing init hook as fatal like node's fatalError: print (shielded from a user-replaced console) and exit 1 instead of surfacing the throw to the process.nextTick caller.
- drop the always-true stats !== undefined guard (the fstat try/catch rethrows unconditionally) - restore prototype _write/_writev when _destroy releases an adopted fd so a later connect() on the reused socket writes through the new handle - account synchronous fd writes in kBytesWritten so bytesWritten and _bytesDispatched reflect them
…atch leak - node-stream.test.js: the spawned child pipes stderr (it emits the stream/iter ExperimentalWarning there); drain it concurrently and assert the warning, per the subprocess-pipe convention - expectations.txt: fs.watch leak test false-passed everywhere while the eval-entry bug swallowed the child's thrown error; the underlying darwin PathWatcher leak is pre-existing and tracked separately
The runner uppercases machine modifiers (DARWIN, X64, ...) and compares them case-sensitively against expectations.txt tags, so the lowercase [ darwin ] entry never matched any lane and the known PathWatcher leak failure kept surfacing on darwin testers. Matches the existing [ WINDOWS-AARCH64 ] / [ ASAN ] casing.
…spec Moving the drain into readableStreamClose was wrong: the WHATWG spec's ReadableStreamClose only settles default-reader read requests, while the drain-with-undefined step belongs to ReadableStreamCancel (step 6). With the drain in close, the standard byte-source EOF pattern - controller.close() then byobRequest.respond(0) - found readIntoRequests already empty and the caller's transferred buffer was never returned (node/browsers hand back a zero-length view of it). Restore the cancel-only drain and replace the regression test with two spec-shaped ones: close()+respond(0) returns the zero-length view over the caller's buffer, and cancel() resolves the pending read with undefined.
Node defines IncomingMessage.prototype.connection with both accessors; getter-only meant req.connection = x threw in strict mode instead of writing through to req.socket.
ModuleLoader__isBuiltin (behind require.resolve.paths and Module._resolveLookupPaths) scans the alias tables directly and never saw the --experimental-stream-iter gate that Alias::get applies, so without the flag the introspection APIs reported stream/iter as a builtin (null) while require() correctly failed to resolve it - node returns the lookup paths array there. Factor the gate into stream_iter_alias_gated() and consult it from both readers; test asserts paths-array without the flag and null with it.
ClientRequest.destroy(err) scheduled 'error' on a nextTick, but the abort chain it triggers (onAbort -> socketCloseListener) emits 'close' synchronously inside the same destroy() call, reversing node's order where 'close' is the terminal event - listeners cleaned up on 'close' missed the error entirely. Emit the error first (re-entry is a no-op via the destroyed guard); test asserts the [error, close] order.
- JSCommonJSModule: the CJS function-wrapper evaluate still used the 3-arg JSC::evaluate overload that swallows exceptions (making the following RETURN_IF_EXCEPTION dead); use the out-param overload and rethrow like the eval-entry path above it, so a wrapper parse failure surfaces the real error instead of the 'function wrapper' TypeError - BunProcess: the six lazy property builders converted to TopExceptionScope now also report the cleared exception via reportUncaughtExceptionAtEventLoop, matching the established constructStdin/constructStdioWriteStream pattern, so the original failure is diagnosable instead of silently yielding undefined
The synchronous 'error' emit added for event ordering threw straight out of destroy() when no listener was attached (standard EventEmitter behavior), skipping the abort/teardown chain and leaking the in-flight fetch - destroyed was already set, so a retry no-oped. Emit synchronously only when a listener exists (preserving the error-then-close order); otherwise defer to a nextTick so the teardown always runs, a listener attached later in the same tick still catches, and an unhandled error crashes asynchronously - matching node, where destroy() never throws. Also drop the stray un-unref'd 5s timeout from the ordering test (Bun's per-test timeout covers the hang case) and add a no-listener regression test.
…throwing error handler The listenerCount gate only covered the no-listener half of the 'destroy() never throws' invariant: a present 'error' handler that itself throws still unwound out of the synchronous emit before the abort/socket-destroy chain ran, leaking the in-flight fetch with destroyed already set. Catch the listener's throw and re-throw it on a fresh tick, matching node where a throwing socketErrorListener becomes an async uncaught exception after socket.destroy(err) already ran. Subprocess test asserts destroy() returns, the teardown ran, and the handler bug surfaces as an async uncaught exception.
… import Matches the file's existing ESM import style and the test convention of reserving require/dynamic import for tests that exercise them.
cirospaciari
force-pushed
the
claude/streams-node26-sync
branch
from
dd7d9dd to
74a5aff
Compare
cirospaciari
force-pushed
the
claude/streams-node26-sync
branch
from
2789aa3 to
699fc81
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Brings the node:stream suite to full parity with Node v26.3.0 by vendoring every upstream stream test, implementing the experimental
stream/iter/zlib/iterAPI behind--experimental-stream-iter, and fixing the runtime gaps the tests expose.All 251 upstream stream test files are now vendored (63 added, 64 resynced to v26.3.0 content); before this PR 124 were byte-identical to upstream and 188 existed at all.
New feature:
stream/iter+zlib/iter(experimental)Node v26's iterable-streams API, ported (~7.5k lines: push/pull/from/share/broadcast/transform/consumers/classic/duplex/ringbuffer + entry modules), gated like Node:
--experimental-stream-iter: without the flag,node:stream/iterreportsNo such built-in moduleand barestream/iterfalls through to filesystem resolution, byte-matching Node's errors.Readable.prototype[Symbol.for("Stream.toAsyncStreamable")]interop with Node's lazy flag check (ERR_STREAM_ITER_MISSING_FLAG).test-stream-iter-*files pass (the 2disabledtests also required the eval-entry exception fix below).Behavior changes
read()returns one buffered chunk at a time instead of concatenating the whole buffer (stream: readable read one buffer at a time nodejs/node#60441): thehowMuchToReadfast path for buffers.read(n)with a decoder no longer over-returns:fromListcomparednagainst the buffer-array length instead of the chunk length.resume()/pause()are no-ops on destroyed streams.Readable.prototype.composemoved off the operators registry onto the prototype returning the composed Duplex directly (Node v26 layout); the registry wrapper hid the writable side.Readable.fromdestroy chaining:_destroychains the instance's original_destroy(e.g. duplexify's abort) before closing the iterator, andduplexify'sfromAsyncGen.destroyunblocks a generator parked on its source — errors now propagate through nested.compose()chains instead of hanging.Duplex.from({ readable, writable }): a readable-side error destroys the writable side (was destroying the readable itself).write(string, "buffer")throwsERR_UNKNOWN_ENCODING.Duplex.toWeb/Readable.toWebsupport byte streams (readableType: 'bytes'+ the deprecatedtypealias with DEP0201);ReadableStream.cancel()settles pending BYOB reads with{done: true}; readable completion no longer waits for a half-open writable side, so the web stream closes when the readable side ends.'readable'for data+EOF, like Node's parser);ClientRequest.destroy(err)emits'error'on the request and destroys an incomplete response withECONNRESET(Node's socketCloseListener semantics).IncomingMessage.socketgetter no longer resurrects a FakeSocket after the stream destroyer detached it —req.socket = nullnow signals "connection must outlive the request" like Node, sores.end()afterdestroy(req)still delivers the body;detachSockettolerates the detached state.kIsPerformingIO/kIoDone, Node-verbatim), soclose()between reads yieldsERR_STREAM_PREMATURE_CLOSEinstead of a clean end, and_destroywaits for in-flight IO before closing the fd.net.Socket({ fd })adopts pipe/character-device/file/socketpair fds with synchronouswrite(2)-backed writes — Node's effective stdio-socket semantics (data survives an immediateprocess.exit()). Network-socket fd adoption still needs the handle layer.createHookinitevents now fire forTickObjectresources — one perprocess.nextTick()call, with a single array-length check on the disabled hot path. Init-only hooks no longer emit the "not implemented" warning. stdio writers are born constructed (no construct TickObjects; Node's stdio streams never have_construct).--expose-internals: vendored node tests can requireinternal/*modules, served frombun:internal-for-testingvia a require interceptor installed by the test harness; newinternal/async_context_frameandinternal/async_hooksshims back thefinished()ALS tests.JSC::evaluateoverload silently discards exceptions into an unused out-param, sobun -e 'require("./missing")'exited 0 with no output; it now prints the resolution error and exits nonzero (also un-hangs--printfor the same case).ERR__ERROR_FORMAT_PADDING_2style —"ERR_"+ enum suffix with its leading underscore);ERR_CRYPTO_UNKNOWN_CIPHERmessage isUnknown cipher;Unknown encoding:values format like Node's%s(objects inspect to{}); new codesERR_OPERATION_FAILED,ERR_STREAM_ITER_MISSING_FLAG,ERR_INVALID_ARG_VALUERangeError variant.src/jsc/ErrorCode.rshad drifted from the generated C++ table (191 stale ordinals — native code threw adjacent error codes, e.g.ERR_INVALID_HANDLE_TYPEforERR_INVALID_HTTP_TOKEN); it is renumbered fromErrorCode+List.h.Known limitations / follow-ups
test-stream-wrap{,-drain,-encoding}needinternal/js_stream_socket, i.e. anet.Socket({ handle })libuv-handle compatibility layer — the same subsystem TLS-over-arbitrary-duplex wants; left as the follow-up.test-stream-pipelineneeds per-socket-read http client body delivery: the native fetch path coalesces response-body segments between JS pulls (one merged buffer instead of Node's one'data'per TCP read). Pre-existing, proven on the unmodified baseline; a fix wants an opt-in per-segment mode on the native response stream to avoid regressing fetch throughput.AsyncContextFrame.current()returnsundefined(bun tracks async context natively and never materializes frame objects);enabledis honestlytrue. Tests couplingenabledto a non-nullcurrent()would not pass.initis emitted for TickObject only; promise/timer/native resources and before/after/destroy remain unimplemented (and still warn).test-stream-readable-default-encoding-error, held-back pair) run pre-v26 content because their upstream versions require internals not in the exposed map.Testing
Validated locally with the debug build:
test-streamsweep (CI-parity mode): 247/251 files pass; the 4 failures are the documented native gaps above, each verified pre-existing on an unmodified baseline build.test-httpsweep (488 files — the only deltas were the ErrorCode-ordinal regressions, fixed in this PR), vendored zlib/crypto tests, and the bun-nativeconsole/process/fs/net/child_process/async_hooks/ web-streams suites — identical or better everywhere.-ehappy path, missing-require error+exit-1, stdin scripts,--print.