fix(stream): node v26 read() chunk-at-a-time, no spurious EOF readable, pipe dedup

[proggeramlug]

Summary

Closes the largest real-gap cluster in test-parity/node-suite/stream by matching node v26's internal/streams/readable byte-for-byte. Baseline (pristine origin/main, node v26.3.0): 758 pass / 10 real fail (after excluding NODE-FAIL tests where node itself throws). This fixes 6 of the 10 via three distinct root causes; the remaining 4 are a single separate codegen gap (see below).

Root causes & fixes

1. read() with no size returns the head chunk, not the whole buffer.
node v26 howMuchToRead(NaN) has a fast path: without a string decoder it always returns state.buffer[bufferIndex].length (one buffer at a time), even when paused. Only a setEncoding (decoded) paused stream concatenates the whole buffer into one string. Perry drained the entire buffer whenever paused.
→ fixes readable/readable-mode-only, readable/read-in-readable-listener, readable/unshift-after-read, readable/iter-yields-buffer-no-encoding

# readable-mode-only, before → after
- drained: aabb
+ drained: aa|bb

2. objectMode read() of the last item no longer emits a final readable.
node never emits readable just to deliver a null at EOF — after the last item of an ended stream it goes straight to end (endReadable). The extra emit made a fixed-count consumer observe a spurious read()===null pair.
→ fixes readable/object-mode-read-returns-object

  first: {"a":1}
  second: {"b":2}
- first: null
- second: null

3. pipe() destinations now consume-on-emit.
A flowing destination (piped-into PassThrough/Duplex) must drop each chunk from its own readable buffer when it emits data live; otherwise the chunk lingers and the destination's drain microtask re-emits it, duplicating every piped chunk. pipeline() already marks both ends via mark_live_pipe_consume_on_emit; pipe() now marks the destination too.
→ fixes pipe/repipe-after-unpipe

- joined: after-repipeafter-repipe
+ joined: after-repipe

Also updates two perry-runtime unit tests (node_stream_tests_extra.rs) that asserted the pre-v26 drain-whole behavior; verified against node v26, read() after unshift returns the unshifted head chunk and leaves the remainder.

Verification (node v26.3.0, macOS)

• The 6 target tests pass.
• Focused 232-test regression over the read/pipe/objectMode/flowing/transform/duplex/consumers surface: 226 pass, 6 node-fail, 0 compile-fail, 0 diffs — no new regressions.
• Cross-module: pipe fan-out (one→two sinks) and chained pipes match node exactly.
• cargo test -p perry-runtime --lib: green (the Date.prototype global-state test passes in isolation — a pre-existing flake unrelated to this change).

A full single-machine harness sweep was not feasible during this run because the shared host was at ~6× CPU load from concurrent builds; the full stream/http/etc. sweep runs in CI.

Not addressed — separate codegen gap

subclass/extends-{readable,writable,duplex,transform} still throw TypeError: Class extends value is not a constructor. User-defined class X extends Readable requires native-stream subclassing support in codegen (synthesized super() initialising native stream state on this, routing push()/_read to the user object) — a much larger, independent change, not a stream-runtime bug.

Summary by CodeRabbit

• Bug Fixes

  • Improved stream reading behavior for better Node.js v26 compatibility
  • Fixed duplicate chunk re-emission when piping to stream destinations
  • Refined EOF signaling to prevent extra read observations

• Tests

  • Updated tests to verify Node v26 stream reading behavior

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: e61d6d35-ced3-423b-9ccb-a7d1f0fdcbe2

📥 Commits

Reviewing files that changed from the base of the PR and between 12ae086 and 1b4dd55.

📒 Files selected for processing (3)

• crates/perry-runtime/src/node_stream_readable_read.rs
• crates/perry-runtime/src/node_stream_readwrite.rs
• crates/perry-runtime/src/node_stream_tests_extra.rs

---

📝 Walkthrough

Walkthrough

This PR refines Perry's readable stream implementation to align with Node v26 semantics. Buffered chunk draining in paused mode now requires both non-flowing state and active encoding. End-of-stream handling removes duplicate readable events. Pipe destination streams are explicitly marked for live consumption. Tests validate chunk-at-a-time read behavior.

Changes

Stream Read and Pipe Behavior

Layer / File(s)	Summary
Readable stream buffer drain and EOF behavior crates/perry-runtime/src/node_stream_readable_read.rs, crates/perry-runtime/src/node_stream_tests_extra.rs	Buffer drain condition now requires both non-flowing state and readable encoding tag; EOF handling removes duplicate queue_readable_event call, leaving only schedule_readable_end. Two tests updated to verify chunk-at-a-time read() returns head chunk first after unshift, consistent with Node v26.
Pipe destination consumption marking crates/perry-runtime/src/node_stream_readwrite.rs	pipe_stream_to_destination adds mark_live_pipe_consume_on_emit(dest) before pipe listener installation to ensure flowing piped destinations consume readable buffers on data emission, preventing duplicate re-emission.

Estimated Code Review Effort

🎯 3 (Moderate) | ⏱️ ~22 minutes

Poem

🐇 A stream flows true, chunk by chunk we read,
No drain unless decoder plants the seed,
At EOF's edge, no double-event spree,
Piped paths consume what they should see! ✨

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly summarizes the main changes: chunk-at-a-time read() behavior, EOF readable event fix, and pipe deduplication.
Description check	✅ Passed	The description is comprehensive and follows the template structure with Summary, Changes, Related issue, Test plan, and Checklist sections clearly addressed.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/stream-read-pipe-dup-diffs

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}