PureYAML/Sources/Parsing/Scanner.Step.swift

extension PureYAML.Parsing.Scanner {
    /// Result of one scanning iteration over the source. Both the batch ``scan``
    /// loop and the incremental ``PureYAML/Parsing/TokenCursor`` driver share the
    /// same step so the two paths cannot diverge (a divergence is what once let a
    /// tab-indentation error slip past the streaming parser while the batch scanner
    /// caught it).
    enum Step {
        /// Nothing remains; the caller appends stream end and stops.
        case endOfInput
        /// A line break, blank line, or comment was consumed; restart the loop.
        case advanced
        /// A directive or document marker was scanned; restart the loop.
        case scannedMarker
        /// An ordinary token was scanned.
        case scannedToken
    }

    /// Performs one scanning iteration: consumes a leading line break or blank line,
    /// scans indentation at a line start, then scans either a directive / document
    /// marker or one ordinary token. Resets `simpleKeyAllowed` at every block-context
    /// line break so a tab at the next line start is recognized as (illegal)
    /// indentation rather than (legal) in-line separation.
    func scanStep(_ state: inout State) throws -> Step {
        if state.reader.isAtEnd {
            return .endOfInput
        }
        if isLineBreak(state.reader.peek()) {
            state.reader.advance()
            state.isAtLineStart = true
            if state.flowDepth == 0 {
                state.simpleKeyAllowed = true
            }
            return .advanced
        }

        if state.isAtLineStart {
            // A line whose entire content is blanks (spaces and/or tabs) before the
            // break is a blank line: it carries no indentation and no token, matching
            // scanner.c's scan_to_next_token, which eats the blanks and the break
            // before any indent processing. Consuming it here keeps a whitespace-only
            // line (e.g. `· \t`) from emitting a spurious indent token or being
            // mis-scanned as content.
            if consumeBlankLineRemainder(&state) {
                return .advanced
            }
            try scanIndentation(&state)
            if state.reader.isAtEnd || isLineBreak(state.reader.peek()) {
                return .advanced
            }
            if try scanDirectiveOrDocumentMarker(&state) {
                return .scannedMarker
            }
            state.isAtLineStart = false
        }

        try scanToken(&state)
        return .scannedToken
    }

    /// When the reader sits on a run of blanks (spaces and/or tabs) that ends the
    /// line (a line break or end of input follows), consumes that run plus the line
    /// break and reports true: the line was blank. Otherwise leaves the reader where
    /// it was and reports false. Used so a whitespace-only line (e.g. `· \t`) is
    /// skipped rather than mis-scanned as plain-scalar content.
    func consumeBlankLineRemainder(_ state: inout State) -> Bool {
        var offset = 0
        while let character = state.reader.peek(offset: offset), character == " " || character == "\t" {
            offset += 1
        }
        let next = state.reader.peek(offset: offset)
        guard next == nil || isLineBreak(next) else {
            return false
        }
        for _ in 0 ..< offset {
            state.reader.advance()
        }
        if isLineBreak(state.reader.peek()) {
            state.reader.advance()
            if state.flowDepth == 0 {
                state.simpleKeyAllowed = true
            }
        }
        return true
    }
}