fix(core): ignore extra blank lines in event stream decoder

[dinwwwh]

Event streams that contain extra blank lines now decode cleanly. Previously a stream ending with an extra newline made the decoder throw "Event Stream ended before complete" after delivering every event, so toAsyncIteratorObject consumers saw an EventStreamDecoderError instead of normal completion. The spec treats extra blank lines as no-ops, so any producer or proxy that sends one triggered this.

Fixes

• Streams ending in extra blank lines (\n, \r or \r\n) no longer throw on end
• Consecutive blank lines between messages no longer emit a spurious empty {} event
• Leading blank lines before the first message are ignored
• Truncated streams still throw as before

Performance

• Chunks containing a message boundary are scanned once instead of twice
• Decode benchmarks are flat or faster; a 100KB message in one chunk decodes about 1.8x faster

Testing

• The existing every-split-position and chunk-size tests now include leading, repeated and trailing blank lines for every line-ending style; both fail on main
• Added an end-to-end toAsyncIteratorObject test covering the reported failure

[pkg-pr-new]

• @standard-server/hono-node-playground
• @standard-server/node-http-playground

@standard-server/aws-lambda

npm i https://pkg.pr.new/@standard-server/aws-lambda@111

@standard-server/core

npm i https://pkg.pr.new/@standard-server/core@111

@standard-server/fastify

npm i https://pkg.pr.new/@standard-server/fastify@111

@standard-server/fetch

npm i https://pkg.pr.new/@standard-server/fetch@111

@standard-server/node

npm i https://pkg.pr.new/@standard-server/node@111

@standard-server/peer

npm i https://pkg.pr.new/@standard-server/peer@111

@standard-server/shared

npm i https://pkg.pr.new/@standard-server/shared@111

commit: 1439bb1

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[codspeed]

Merging this PR will improve performance by 27.58%

⚠️ Different runtime environments detected

Some benchmarks with significant performance changes were compared across different runtime environments,
which may affect the accuracy of the results.

Open the report in CodSpeed to investigate

⚡ 1 improved benchmark
✅ 25 untouched benchmarks
⏩ 108 skipped benchmarks¹

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
⚡	large message 100KB in one chunk	848.4 µs	665 µs	+27.58%

Tip

Curious why performance improved? Comment @codspeedbot explain why performance improved on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing claude/event-stream-decoder-blank-lines-ad4cf7 (1439bb1) with main (90876d0)}

Footnotes

1. 108 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[pullfrog]

ℹ️ No critical issues — one small test-naming nit inline.

Reviewed changes

• Greedy message delimiter — MESSAGE_DELIMITER_REGEX now requires 2+ line endings, so runs of blank lines collapse into a single delimiter and no longer emit a spurious empty {} event.
• Leading blank-line stripping — when pending is empty, feed() drops a leading [\r\n]+ run, replacing the removed discardLeadingLF state.
• Overlap window reduced — MAX_DELIMITER_OVERLAP 3 → 2, justified because pending never contains a blank line.
• Single-pass scan — the global regex is driven via exec/lastIndex instead of test + matchAll.
• Tests — split-position and chunk-size tests extended with leading/repeated/trailing blank lines; a new end-to-end toAsyncIteratorObject case reproduces the reported failure.

I re-ran the changed tests both ways: they fail on the base decoder (90876d0) and pass on this head (71665f7); the core decoder and fetch event-stream suites are green (50 tests).

  ^{｜ Fix all ➔ ｜ Fix 👍s ➔ ｜ View workflow run ｜ Using DeepSeek Flash (default — pick a model for stronger reviews) ｜ 𝕏}

[pullfrog]

✅ No new issues found.

Reviewed changes

Since the prior pullfrog review (71665f7), the branch added one test-only commit (1439bb1) that broadens coverage of the blank-line handling fixed earlier:

• Blank-line-only streams — new case asserts a stream of only \n/\r/\r\n blank lines emits nothing and does not throw on end().
• Comment-only messages preserved — new case pins that comment-only messages (keep-alives) surrounded by extra blank lines are still emitted, so blank-line skipping cannot swallow them.
• Incomplete message after blanks — new case asserts end() still throws when extra blank lines are followed by an incomplete message.
• End-to-end fetch coverage — new toAsyncIteratorObject case feeds leading, repeated and trailing blank lines plus comment pings, and expects clean message delivery followed by normal completion.
• Chunk-boundary tests extended — the split-position matrix now includes multi-ending delimiters and a leading delimiter; the chunk-size parity stream carries leading/extra/trailing blank lines; the empty-chunk test was renamed to describe what it actually covers.

I re-ran the changed tests both ways: reverting decoder.ts to the base (90876d0) makes the four new behavioral cases fail (spurious {} events and EventStream ended before complete), and they pass on the head (1439bb1). Both suites are green (53 tests).

  ^{｜ View workflow run ｜ Using DeepSeek Flash (default — pick a model for stronger reviews) ｜ 𝕏}