PureXML/Tests/ContentModelOracle.swift

@testable import PureXML

/// An independent, brute-force content-model membership oracle over element names,
/// deliberately implemented separately from PureXML's matcher so the instance
/// differential can adjudicate a disagreement with libxml2: libxml2's content
/// validation has leniency bugs (it accepts some invalid nested/nullable/unbounded
/// models), so "PureXML rejects, libxml2 accepts" is not by itself a PureXML fault.
/// The generated instances are empty `xs:string` elements, so an instance is valid
/// iff its element-name sequence is in this language; this oracle is that decision,
/// verified against an external regex engine on the known cases.
indirect enum ContentModel: Sendable {
    case element(String)
    case any
    case sequence([ContentModel])
    case choice([ContentModel])
    /// An `xs:all` group; each member is its own `occurs(element, min, 1)`.
    case all([ContentModel])
    case occurs(ContentModel, Int, Int?)
}

enum ContentModelOracle {
    /// Whether the model can match the empty sequence.
    static func nullable(_ model: ContentModel) -> Bool {
        switch model {
        case .element, .any: false
        case let .sequence(parts): parts.allSatisfy(nullable)
        case let .choice(parts): parts.contains(where: nullable)
        case let .all(parts): parts.allSatisfy(nullable)
        case let .occurs(sub, low, _): low == 0 || nullable(sub)
        }
    }

    /// Every position reachable after matching `model` starting at `start`.
    static func ends(_ model: ContentModel, _ tokens: [String], _ start: Int) -> Set<Int> {
        switch model {
        case let .element(name):
            return start < tokens.count && tokens[start] == name ? [start + 1] : []
        case .any:
            return start < tokens.count ? [start + 1] : []
        case let .sequence(parts):
            var positions: Set<Int> = [start]
            for part in parts {
                var next: Set<Int> = []
                for position in positions {
                    next.formUnion(ends(part, tokens, position))
                }
                positions = next
                if positions.isEmpty {
                    break
                }
            }
            return positions
        case let .choice(parts):
            var out: Set<Int> = []
            for part in parts {
                out.formUnion(ends(part, tokens, start))
            }
            return out
        case let .all(parts):
            return allEnds(parts, Array(repeating: false, count: parts.count), tokens, start)
        case let .occurs(sub, low, high):
            return occursEnds(sub, low, high, tokens, start)
        }
    }

    /// Positions reachable by matching `sub` between `low` and `high` times. A
    /// nullable `sub` is bounded by stopping once the reachable set stops growing.
    private static func occursEnds(_ sub: ContentModel, _ low: Int, _ high: Int?, _ tokens: [String], _ start: Int) -> Set<Int> {
        var byReps: [Set<Int>] = [[start]]
        let cap = min(high ?? Int.max, tokens.count + low + 2)
        var reps = 0
        while reps < cap {
            var next: Set<Int> = []
            for position in byReps[reps] {
                next.formUnion(ends(sub, tokens, position))
            }
            byReps.append(next)
            reps += 1
            let priorPositions = byReps[0 ..< reps].reduce(into: Set<Int>()) { $0.formUnion($1) }
            if next.isSubset(of: priorPositions), reps >= low {
                break
            }
        }
        var out: Set<Int> = []
        if low == 0 {
            out.insert(start)
        }
        for count in low ... cap where count < byReps.count {
            out.formUnion(byReps[count])
        }
        return out
    }

    /// Positions reachable by matching the remaining (not-`done`) members of an
    /// `xs:all` group in any order; a member that is nullable may be skipped.
    private static func allEnds(_ members: [ContentModel], _ done: [Bool], _ tokens: [String], _ start: Int) -> Set<Int> {
        var out: Set<Int> = []
        if members.indices.allSatisfy({ done[$0] || nullable(members[$0]) }) {
            out.insert(start)
        }
        for index in members.indices where !done[index] {
            for end in ends(members[index], tokens, start) where end > start {
                var next = done
                next[index] = true
                out.formUnion(allEnds(members, next, tokens, end))
            }
        }
        return out
    }

    static func matches(_ model: ContentModel, _ tokens: [String]) -> Bool {
        ends(model, tokens, 0).contains(tokens.count)
    }
}