De-genericize processor, engine, etc. (#502)

* Avoid double execution by avoiding Array init

* De-genericize processor, engine, etc.

Provides only modest performance improvements (it was already getting
specialized), but makes it possible to add String-specific specializations.

Author: milseman

Sources/_StringProcessing/Algorithms/Matching/Matches.swift

@@ -349,6 +349,12 @@ extension BidirectionalCollection where SubSequence == Substring {
   public func matches<Output>(
     of r: some RegexComponent<Output>
   ) -> [Regex<Output>.Match] {
-    Array(_matches(of: r))
+    // FIXME: Array init calls count, which double-executes the regex :-(
+    // FIXME: just return some Collection<Regex<Output>.Match>
+    var result = Array<Regex<Output>.Match>()
+    for match in _matches(of: r) {
+      result.append(match)
+    }
+    return result
   }
 }

Sources/_StringProcessing/ByteCodeGen.swift

@@ -3,7 +3,7 @@
 extension Compiler {
   struct ByteCodeGen {
     var options: MatchingOptions
-    var builder = Program.Builder()
+    var builder = MEProgram.Builder()
     /// A Boolean indicating whether the first matchable atom has been emitted.
     /// This is used to determine whether to apply initial options.
     var hasEmittedFirstMatchableAtom = false
@@ -16,7 +16,7 @@ extension Compiler {
 }
 
 extension Compiler.ByteCodeGen {
-  mutating func emitRoot(_ root: DSLTree.Node) throws -> Program {
+  mutating func emitRoot(_ root: DSLTree.Node) throws -> MEProgram {
     // The whole match (`.0` element of output) is equivalent to an implicit
     // capture over the entire regex.
     try emitNode(.capture(name: nil, reference: nil, root))

Sources/_StringProcessing/Compiler.swift

@@ -25,7 +25,7 @@ class Compiler {
     self.tree = tree
   }
 
-  __consuming func emit() throws -> Program {
+  __consuming func emit() throws -> MEProgram {
     // TODO: Handle global options
     var codegen = ByteCodeGen(
       options: options, captureList: tree.captureList

Sources/_StringProcessing/ConsumerInterface.swift

@@ -18,7 +18,7 @@ extension DSLTree.Node {
   /// the front of an input range
   func generateConsumer(
     _ opts: MatchingOptions
-  ) throws -> MEProgram<String>.ConsumeFunction? {
+  ) throws -> MEProgram.ConsumeFunction? {
     switch self {
     case .atom(let a):
       return try a.generateConsumer(opts)
@@ -56,7 +56,7 @@ extension DSLTree.Atom {
   // top-level nodes, but it's also invoked for `.atom` members of a custom CC
   func generateConsumer(
     _ opts: MatchingOptions
-  ) throws -> MEProgram<String>.ConsumeFunction? {
+  ) throws -> MEProgram.ConsumeFunction? {
     let isCaseInsensitive = opts.isCaseInsensitive
 
     switch self {
@@ -142,7 +142,7 @@ extension String {
   }
 }
 
-func consumeName(_ name: String, opts: MatchingOptions) -> MEProgram<String>.ConsumeFunction {
+func consumeName(_ name: String, opts: MatchingOptions) -> MEProgram.ConsumeFunction {
   let consume = consumeFunction(for: opts)
   return consume(propertyScalarPredicate {
     // FIXME: name aliases not covered by $0.nameAlias are missed
@@ -180,7 +180,7 @@ extension AST.Atom {
 
   func generateConsumer(
     _ opts: MatchingOptions
-  ) throws -> MEProgram<String>.ConsumeFunction? {
+  ) throws -> MEProgram.ConsumeFunction? {
     // TODO: Wean ourselves off of this type...
     if let cc = self.characterClass?.withMatchLevel(
       opts.matchLevel
@@ -237,7 +237,7 @@ extension AST.Atom {
 extension DSLTree.CustomCharacterClass.Member {
   func generateConsumer(
     _ opts: MatchingOptions
-  ) throws -> MEProgram<String>.ConsumeFunction {
+  ) throws -> MEProgram.ConsumeFunction {
     switch self {
     case let .atom(a):
       guard let c = try a.generateConsumer(opts) else {
@@ -344,7 +344,7 @@ extension DSLTree.CustomCharacterClass.Member {
 extension DSLTree.CustomCharacterClass {
   func generateConsumer(
     _ opts: MatchingOptions
-  ) throws -> MEProgram<String>.ConsumeFunction {
+  ) throws -> MEProgram.ConsumeFunction {
     // NOTE: Easy way to implement, obviously not performant
     let consumers = try members.map {
       try $0.generateConsumer(opts)
@@ -386,7 +386,7 @@ private func propertyScalarPredicate(_ p: @escaping (Unicode.Scalar.Properties)
 
 func consumeScalar(
   _ p: @escaping ScalarPredicate
-) -> MEProgram<String>.ConsumeFunction {
+) -> MEProgram.ConsumeFunction {
   { input, bounds in
     // TODO: bounds check?
     let curIdx = bounds.lowerBound
@@ -399,7 +399,7 @@ func consumeScalar(
 }
 func consumeCharacterWithLeadingScalar(
   _ p: @escaping ScalarPredicate
-) -> MEProgram<String>.ConsumeFunction {
+) -> MEProgram.ConsumeFunction {
   { input, bounds in
     let curIdx = bounds.lowerBound
     if p(input[curIdx].unicodeScalars.first!) {
@@ -410,7 +410,7 @@ func consumeCharacterWithLeadingScalar(
 }
 func consumeCharacterWithSingleScalar(
   _ p: @escaping ScalarPredicate
-) -> MEProgram<String>.ConsumeFunction {
+) -> MEProgram.ConsumeFunction {
   { input, bounds in
     let curIdx = bounds.lowerBound
 
@@ -423,7 +423,7 @@ func consumeCharacterWithSingleScalar(
 
 func consumeFunction(
   for opts: MatchingOptions
-) -> (@escaping ScalarPredicate) -> MEProgram<String>.ConsumeFunction {
+) -> (@escaping ScalarPredicate) -> MEProgram.ConsumeFunction {
   opts.semanticLevel == .graphemeCluster
     ? consumeCharacterWithLeadingScalar
     : consumeScalar
@@ -432,11 +432,11 @@ func consumeFunction(
 extension AST.Atom.CharacterProperty {
   func generateConsumer(
     _ opts: MatchingOptions
-  ) throws -> MEProgram<String>.ConsumeFunction {
+  ) throws -> MEProgram.ConsumeFunction {
     // Handle inversion for us, albeit not efficiently
     func invert(
-      _ p: @escaping MEProgram<String>.ConsumeFunction
-    ) -> MEProgram<String>.ConsumeFunction {
+      _ p: @escaping MEProgram.ConsumeFunction
+    ) -> MEProgram.ConsumeFunction {
       return { input, bounds in
         if p(input, bounds) != nil { return nil }
 
@@ -448,7 +448,7 @@ extension AST.Atom.CharacterProperty {
     }
 
     let consume = consumeFunction(for: opts)
-    let preInversion: MEProgram<String>.ConsumeFunction =
+    let preInversion: MEProgram.ConsumeFunction =
     try {
       switch kind {
         // TODO: is this modeled differently?
@@ -533,7 +533,7 @@ extension Unicode.BinaryProperty {
   // FIXME: Semantic level, vet for precise defs
   func generateConsumer(
     _ opts: MatchingOptions
-  ) throws -> MEProgram<String>.ConsumeFunction {
+  ) throws -> MEProgram.ConsumeFunction {
     let consume = consumeFunction(for: opts)
 
     // Note if you implement support for any of the below, you need to adjust
@@ -701,7 +701,7 @@ extension Unicode.POSIXProperty {
   // FIXME: Semantic level, vet for precise defs
   func generateConsumer(
     _ opts: MatchingOptions
-  ) -> MEProgram<String>.ConsumeFunction {
+  ) -> MEProgram.ConsumeFunction {
     let consume = consumeFunction(for: opts)
 
     // FIXME: modes, etc
@@ -749,7 +749,7 @@ extension Unicode.ExtendedGeneralCategory {
   // FIXME: Semantic level
   func generateConsumer(
     _ opts: MatchingOptions
-  ) throws -> MEProgram<String>.ConsumeFunction {
+  ) throws -> MEProgram.ConsumeFunction {
     let consume = consumeFunction(for: opts)
 
     switch self {

Sources/_StringProcessing/Engine/Consume.swift

@@ -13,8 +13,8 @@ var checkComments = true
 
 extension Engine {
   func makeProcessor(
-    input: Input, bounds: Range<Input.Index>, matchMode: MatchMode
-  ) -> Processor<Input> {
+    input: String, bounds: Range<String.Index>, matchMode: MatchMode
+  ) -> Processor {
     Processor(
       program: program,
       input: input,
@@ -24,7 +24,7 @@ extension Engine {
   }
 }
 
-extension Processor where Input == String {
+extension Processor {
   // TODO: Should we throw here?
   mutating func consume() -> Input.Index? {
     while true {

Sources/_StringProcessing/Engine/Engine.swift

@@ -11,9 +11,9 @@
 
 // Currently, engine binds the type and consume binds an instance.
 // But, we can play around with this.
-struct Engine<Input: BidirectionalCollection> where Input.Element: Hashable {
+struct Engine {
 
-  var program: MEProgram<Input>
+  var program: MEProgram
 
   // TODO: Pre-allocated register banks
 
@@ -25,7 +25,7 @@ struct Engine<Input: BidirectionalCollection> where Input.Element: Hashable {
   }
 
   init(
-    _ program: MEProgram<Input>,
+    _ program: MEProgram,
     enableTracing: Bool? = nil
   ) {
     var program = program

Sources/_StringProcessing/Engine/MEBuilder.swift

@@ -11,7 +11,7 @@
 
 @_implementationOnly import _RegexParser // For errors
 
-extension MEProgram where Input.Element: Hashable {
+extension MEProgram {
   struct Builder {
     var instructions: [Instruction] = []
 
@@ -71,7 +71,7 @@ extension MEProgram.Builder {
   // TODO: We want a better strategy for fixups, leaving
   // the operand in a different form isn't great...
 
-  init<S: Sequence>(staticElements: S) where S.Element == Input.Element {
+  init<S: Sequence>(staticElements: S) where S.Element == Character {
     staticElements.forEach { elements.store($0) }
   }
 
@@ -183,14 +183,14 @@ extension MEProgram.Builder {
     instructions.append(.init(.advance, .init(distance: n)))
   }
 
-  mutating func buildMatch(_ e: Input.Element) {
+  mutating func buildMatch(_ e: Character) {
     instructions.append(.init(
       .match, .init(element: elements.store(e))))
   }
 
   mutating func buildMatchSequence<S: Sequence>(
     _ s: S
-  ) where S.Element == Input.Element {
+  ) where S.Element == Character {
     instructions.append(.init(
       .matchSequence,
       .init(sequence: sequences.store(.init(s)))))
@@ -219,7 +219,7 @@ extension MEProgram.Builder {
   }
 
   mutating func buildAssert(
-    _ e: Input.Element, into cond: BoolRegister
+    _ e: Character, into cond: BoolRegister
   ) {
     instructions.append(.init(.assertion, .init(
       element: elements.store(e), bool: cond)))

Sources/_StringProcessing/Engine/MECapture.swift

@@ -95,7 +95,7 @@ extension Processor._StoredCapture: CustomStringConvertible {
 }
 
 struct MECaptureList {
-  var values: Array<Processor<String>._StoredCapture>
+  var values: Array<Processor._StoredCapture>
   var referencedCaptureOffsets: [ReferenceID: Int]
 
   func latestUntyped(from input: String) -> Array<Substring?> {

Sources/_StringProcessing/Engine/MEProgram.swift

@@ -11,12 +11,14 @@
 
 @_implementationOnly import _RegexParser
 
-struct MEProgram<Input: BidirectionalCollection> where Input.Element: Equatable {
+struct MEProgram {
+  typealias Input = String
+
   typealias ConsumeFunction = (Input, Range<Input.Index>) -> Input.Index?
   typealias AssertionFunction =
     (Input, Input.Index, Range<Input.Index>) throws -> Bool
   typealias TransformFunction =
-    (Input, Processor<Input>._StoredCapture) throws -> Any?
+    (Input, Processor._StoredCapture) throws -> Any?
   typealias MatcherFunction =
     (Input, Input.Index, Range<Input.Index>) throws -> (Input.Index, Any)?
 

Sources/_StringProcessing/Engine/Processor.swift

@@ -14,8 +14,6 @@ enum MatchMode {
   case partialFromFront
 }
 
-typealias Program = MEProgram<String>
-
 /// A concrete CU. Somehow will run the concrete logic and
 /// feed stuff back to generic code
 struct Controller {
@@ -26,9 +24,8 @@ struct Controller {
   }
 }
 
-struct Processor<
-  Input: BidirectionalCollection
-> where Input.Element: Equatable { // maybe Hashable?
+struct Processor {
+  typealias Input = String
   typealias Element = Input.Element
 
   let input: Input
@@ -75,7 +72,7 @@ extension Processor {
 
 extension Processor {
   init(
-    program: MEProgram<Input>,
+    program: MEProgram,
     input: Input,
     bounds: Range<Position>,
     matchMode: MatchMode,

Sources/_StringProcessing/Engine/Registers.swift

@@ -29,15 +29,15 @@ extension Processor {
     // TODO: Degenericize Processor and store Strings
     var sequences: [[Element]] = []
 
-    var consumeFunctions: [MEProgram<Input>.ConsumeFunction]
+    var consumeFunctions: [MEProgram.ConsumeFunction]
 
-    var assertionFunctions: [MEProgram<Input>.AssertionFunction]
+    var assertionFunctions: [MEProgram.AssertionFunction]
 
     // Captured-value constructors
-    var transformFunctions: [MEProgram<Input>.TransformFunction]
+    var transformFunctions: [MEProgram.TransformFunction]
 
     // Value-constructing matchers
-    var matcherFunctions: [MEProgram<Input>.MatcherFunction]
+    var matcherFunctions: [MEProgram.MatcherFunction]
 
     // currently, these are for comments and abort messages
     var strings: [String]
@@ -58,6 +58,8 @@ extension Processor {
 }
 
 extension Processor.Registers {
+  typealias Input = String
+
   subscript(_ i: StringRegister) -> String {
     strings[i.rawValue]
   }
@@ -85,24 +87,24 @@ extension Processor.Registers {
   subscript(_ i: ElementRegister) -> Input.Element {
     elements[i.rawValue]
   }
-  subscript(_ i: ConsumeFunctionRegister) -> MEProgram<Input>.ConsumeFunction {
+  subscript(_ i: ConsumeFunctionRegister) -> MEProgram.ConsumeFunction {
     consumeFunctions[i.rawValue]
   }
-  subscript(_ i: AssertionFunctionRegister) -> MEProgram<Input>.AssertionFunction {
+  subscript(_ i: AssertionFunctionRegister) -> MEProgram.AssertionFunction {
     assertionFunctions[i.rawValue]
   }
-  subscript(_ i: TransformRegister) -> MEProgram<Input>.TransformFunction {
+  subscript(_ i: TransformRegister) -> MEProgram.TransformFunction {
     transformFunctions[i.rawValue]
   }
-  subscript(_ i: MatcherRegister) -> MEProgram<Input>.MatcherFunction {
+  subscript(_ i: MatcherRegister) -> MEProgram.MatcherFunction {
     matcherFunctions[i.rawValue]
   }
 }
 
 extension Processor.Registers {
   init(
-    _ program: MEProgram<Input>,
-    _ sentinel: Input.Index
+    _ program: MEProgram,
+    _ sentinel: String.Index
   ) {
     let info = program.registerInfo
 

Sources/_StringProcessing/Engine/Tracing.swift

@@ -54,7 +54,7 @@ extension Instruction.Payload: CustomStringConvertible {
 }
 
 extension Processor.SavePoint {
-  func describe(in input: Input) -> String {
+  func describe(in input: String) -> String {
     let posStr: String
     if let p = self.pos {
       posStr = "\(input.distance(from: input.startIndex, to: p))"