#883 regressed performance by ~7%

[ahoppen]

#883 regressed swift-format’s performance by 7-8%. To reproduce run

cd /tmp
git clone https://github.com/apple/swift-syntax.git
cd swift-syntax
git checkout -f 509.0.0

cd path/to/swift-format
swift build -c release
.build/release/swift-format lint --measure-instructions --recursive /tmp/swift-syntax

And compare the instruction count between swift-format with and without #883.

[ahoppen]

Synced to Apple’s issue tracker as rdar://141233035

[macshome]

In case anyone was wondering, here is the difference.

Before #883: Instructions executed: 67167400341
With #883: Instructions executed: 71260352299

[macshome]

Changing the code to:

let lines = text.count { $0 == "\n" }
    lineNumber += lines
    guard lines > 1, let range = text.range(of: "\n", options: .backwards) else {
      // Handle case where no newline exists
      column += text.count
      return
    }
    let lastLine = text[range.upperBound...]
    column = lastLine.count

Lowers the instructions executed to: 68378386615

[macshome]

After playing around in Godbolt it seems that lastIndex(of:_) uses a few less instructions per-loop than range(of:_). Changing the code to:

 let lines = text.count { $0 == "\n" }
    lineNumber += lines
    guard lines > 1, let lastNewlineIndex = text.lastIndex(of: "\n") else {
      // Handle case where no newline exists
      column += text.count
      return
    }
    let lastLine = text[text.index(after: lastNewlineIndex)...]
    column = lastLine.count

Further lowers the instruction executed count to: 68259258073

That's only about ~ 1.5% difference from before the changes.