Rework header name scan to use optimised ByteString.indexOf

[pjfanning]

• Needs a change in 2.0.0 ByteString (new indexOf(char, start, end))
• does the illegal char check slightly later but rolls it in with the code to create the headerName String so it shouldn't reduce happy path perf - but the illegal char will be reported slightly later in the worst case
• the ByteString indexOf should perform better than the existing code in the happy path

[mdedetrich]

Are there benchmarks for this, if so can you post the difference otherwise would it be possible to create some benchmarks and post the results?

[Copilot]

Pull request overview

This PR refactors the HTTP header name scanning logic to use an optimized ByteString.indexOf method instead of character-by-character scanning. The primary goal is to improve performance in the happy path by leveraging native indexOf implementation.

• Replaces scanHeaderNameAndReturnIndexOfColon with direct indexOf call and two new helper functions
• Moves illegal character validation to occur during header name string construction rather than during colon search
• Validates the entire header name (from lineStart to colon) rather than only the portion after trie matching

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[pjfanning]

Are there benchmarks for this, if so can you post the difference otherwise would it be possible to create some benchmarks and post the results?

I'll come up with some benchmark numbers

[pjfanning]

I ran the existing benchmark and the results are similar, with and without this change.

HeaderParserBenchmark.bench_parse_headers

I'll leave this for now but may come back to it to see if it is worth continuing.