基于缩进解析类似 ML 的语法，以及所有被认为是 instruction/expression 的东西
Parsing an ML-like syntax based on indentation, and everything considered to be an instruction/expression
注意：不久前，我已经问过 。这不是重复，但要求的澄清不属于主题本身的范围。因此，我允许自己打开另一个职位，处理基于缩进的类似 ML 语法的分析，并将所有内容视为指令/表达式。
例如：
"Hello" 是一个表达式，
let foo = 2 + 1是使用表达式（2 + 1）的指令，
print foo 是一条指令，...
简而言之，一种非常模块化和动态的语法和语义。像 F# 或 OCaml。
为此，我使用 F# 和 API（在 nuget 上可用）FParsec。 FParsec wiki 提供了an example of a syntax based on indentation，所以我重新整理了一下。下面代码中使用的模块是IndentationParserWithoutBacktracking.
要解析的示例代码使用基本缩进，不混合 "literal" 和 "instructions/expressions":
loop i 1 10
loop k 1 10
print k
print i
print j
一个简单的代码，没有上下文（但目前这并不重要）。
我的实现允许这样的代码：
let foo = a + b
let foo =
let a = 9
let b = 1
a + b
let foo = 7
let foo =
loop i 1 10
print i
例如。 （loop 和 print 只是为了测试...）
这个问题困扰了我一个星期，但我无法解决，每次解析器中需要一条新行指令时，缩进模块都会询问我...这是屏幕截图：
这适用于上述所有示例。我不太明白这个问题，因此不知道如何解决。
这是针对此问题测试的代码，它满足最小和功能代码标准，但是，必须使用 FParsec：
open FParsec
// This module come from 'https://github.com/stephan-tolksdorf/fparsec/wiki/Parsing-indentation-based-syntax-with-FParsec'
// I used the second module: 'IndentationParserWithoutBacktracking'
module IndentationParserWithoutBacktracking =
let tabStopDistance = 8
type LastParsedIndentation() =
[<DefaultValue>]
val mutable Value: int32
[<DefaultValue>]
val mutable EndIndex: int64
type UserState =
{Indentation: int
// We put LastParsedIndentation into the UserState so that we
// can conveniently use a separate instance for each stream.
// The members of the LastParsedIndentation instance will be mutated
// directly and hence won't be affected by any stream backtracking.
LastParsedIndentation: LastParsedIndentation}
with
static member Create() = {Indentation = -1
LastParsedIndentation = LastParsedIndentation(EndIndex = -1L)}
type CharStream = CharStream<UserState>
type Parser<'t> = Parser<'t, UserState>
// If this function is called at the same index in the stream
// where the function previously stopped, then the previously
// returned indentation will be returned again.
// This way we can avoid backtracking at the end of indented blocks.
let skipIndentation (stream: CharStream) =
let lastParsedIndentation = stream.UserState.LastParsedIndentation
if lastParsedIndentation.EndIndex = stream.Index then
lastParsedIndentation.Value
else
let mutable indentation = stream.SkipNewlineThenWhitespace(tabStopDistance, false)
lastParsedIndentation.EndIndex <- stream.Index
lastParsedIndentation.Value <- indentation
indentation
let indentedMany1 (p: Parser<'t>) label : Parser<'t list> =
fun stream ->
let oldIndentation = stream.UserState.Indentation
let indentation = skipIndentation stream
if indentation <= oldIndentation then
Reply(Error, expected (if indentation < 0 then "newline" else "indented " + label))
else
stream.UserState <- {stream.UserState with Indentation = indentation}
let results = ResizeArray()
let mutable stateTag = stream.StateTag
let mutable reply = p stream // parse the first element
let mutable newIndentation = 0
while reply.Status = Ok
&& (results.Add(reply.Result)
newIndentation <- skipIndentation stream
newIndentation = indentation)
do
stateTag <- stream.StateTag
reply <- p stream
if reply.Status = Ok
|| (stream.IsEndOfStream && results.Count > 0 && stream.StateTag = stateTag)
then
if newIndentation < indentation || stream.IsEndOfStream then
stream.UserState <- {stream.UserState with Indentation = oldIndentation}
Reply(List.ofSeq results)
else
Reply(Error, messageError "wrong indentation")
else // p failed
Reply(reply.Status, reply.Error)
open IndentationParserWithoutBacktracking
let isBlank = fun c -> c = ' ' || c = '\t'
let ws = spaces
let ws1 = skipMany1SatisfyL isBlank "whitespace"
let str s = pstring s .>> ws
let keyword str = pstring str >>? nextCharSatisfiesNot (fun c -> isLetter c || isDigit c) <?> str
// AST
type Identifier = Identifier of string
// A value is just a literal or a data name, called here "Variable"
type Value =
| Int of int | Float of float
| Bool of bool | String of string
| Char of char | Variable of Identifier
// All is an instruction, but there are some differences:
type Instr =
// Arithmetic
| Literal of Value | Infix of Instr * InfixOp * Instr
// Statements (instructions needing another instructions)
| Let of Identifier * Instr list
| Loop of Identifier * Instr * Instr * Instr list
// Other - the "print" function, from the link seen above
| Print of Identifier
and InfixOp =
// Arithmetic
| Sum | Sub | Mul | Div
// Logic
| And | Or | Equal | NotEqual | Greater | Smaller | GreaterEqual | SmallerEqual
// Literals
let numberFormat = NumberLiteralOptions.AllowMinusSign ||| NumberLiteralOptions.AllowFraction |||
NumberLiteralOptions.AllowHexadecimal ||| NumberLiteralOptions.AllowOctal |||
NumberLiteralOptions.AllowBinary
let literal_numeric =
numberLiteral numberFormat "number" |>> fun nl ->
if nl.IsInteger then Literal (Int(int nl.String))
else Literal (Float(float nl.String))
let literal_bool =
(choice [
(stringReturn "true" (Literal (Bool true)))
(stringReturn "false" (Literal (Bool false)))
]
.>> ws) <?> "boolean"
let literal_string =
(between (pstring "\"") (pstring "\"") (manyChars (satisfy (fun c -> c <> '"')))
|>> fun s -> Literal (String s)) <?> "string"
let literal_char =
(between (pstring "'") (pstring "'") (satisfy (fun c -> c <> '''))
|>> fun c -> Literal (Char c)) <?> "character"
let identifier =
(many1Satisfy2L isLetter (fun c -> isLetter c || isDigit c) "identifier"
|>> Identifier) <?> "identifier"
let betweenParentheses p =
(between (str "(") (str ")") p) <?> ""
let variable = identifier |>> fun id -> Literal (Variable id)
let literal = (attempt literal_numeric <|>
attempt literal_bool <|>
attempt literal_char <|>
attempt literal_string <|>
attempt variable)
// Instressions and statements
let pInstrs, pInstrimpl = createParserForwardedToRef()
// `ploop` is located here to force `pInstrs` to be of the type `Instr list`, `ploop` requesting an instression list.
let ploop =
pipe4
(keyword "loop" >>. ws1 >>. identifier)
(ws1 >>. literal)
(ws1 >>. literal)
(pInstrs)
(fun id min max stmts -> Loop(id, min, max, stmts))
// `singlepInstr` allows to use only one Instression, used just after.
let singlepInstr =
pInstrs |>> fun ex -> ex.Head
let term =
(ws >>. singlepInstr .>> ws) <|>
(betweenParentheses (ws >>. singlepInstr)) <|>
(ws >>. literal .>> ws) <|>
(betweenParentheses (ws >>. literal))
let infixOperator (p: OperatorPrecedenceParser<_, _, _>) op prec map =
p.AddOperator(InfixOperator(op, ws, prec, Associativity.Left, map))
let ops =
// Arithmetic
[ "+"; "-"; "*"; "/"; "%" ] @
// Logical
[ "&&"; "||"; "=="; "!="; ">"; "<"; ">="; "<=" ]
let opCorrespondance op =
match op with
// Arithmetic operators
| "+" -> Sum | "-" -> Sub
| "*" -> Mul | "/" -> Div
// Logical operators
| "&&" -> And | "||" -> Or
| "==" -> Equal | "!=" -> NotEqual
| ">" -> Greater | "<" -> Smaller
| ">=" -> GreaterEqual | "<=" -> SmallerEqual
| _ -> failwith ("Unknown operator: " + op)
let opParser = new OperatorPrecedenceParser<Instr, unit, UserState>()
for op in ops do
infixOperator opParser op 1 (fun x y -> Infix(x, opCorrespondance op, y))
opParser.TermParser <- term
// Statements
(*
- let:
let <identifier> = <instruction(s) / value>
- print:
print <identifier>
- loop:
loop <identifier> <literal> <literal> <indented statements>
*)
let plet =
pipe2
(keyword "let" >>. ws1 >>. identifier)
(ws >>. str "=" >>. ws >>. pInstrs)
(fun id exp -> Let(id, exp))
let print =
keyword "print" >>. ws1 >>. identifier
|>> Print
let instruction =
print <|> ploop <|> plet <|>
opParser.ExpressionParser <|>
literal
pInstrimpl := indentedMany1 instruction "instruction"
let document = pInstrs .>> spaces .>> eof
let test str =
match runParserOnString document (UserState.Create()) "" str with
| Success(result, _, _) -> printfn "%A" result
| Failure(errorMsg, _, _) -> printfn "%s" errorMsg
System.Console.Clear()
let code = test @"
let foo = a + b
"
我想首先了解为什么它不起作用，而且能够找到解决我的问题的方法，并且该解决方案可以扩展到解析器的潜在语法添加。
等待有益的回答，谢谢。
为了理解为什么您的解析器不起作用，您需要隔离问题。
如果我没理解错的话，您希望 let 解析器支持同一行上的单个指令或后续行上的缩进指令，例如：
let x = instruction
let b =
instruction
instruction
如果您无法让现有的实现正常工作，我建议您回到 Wiki 上的实现并尝试只添加对 let 语句的支持。
例如，我通过以下修改使 Wiki 解析器接受简单的 let 语句：
type Statement = Loop of Identifier * int * int * Statement list
| Print of Identifier
| Let of Identifier * Statement list
let ws = skipManySatisfy isBlank
let str s = pstring s .>> ws
let statement, statementRef = createParserForwardedToRef()
let indentedStatements = indentedMany1 statement "statement"
let plet = keyword "let" >>. pipe2 (ws1 >>. identifier)
(ws >>. str "=" >>. ws
>>. (indentedStatements
<|> (statement |>> fun s -> [s])))
(fun id exp -> Let(id, exp))
statementRef := print <|> loop <|> plet
请注意，在修改后的版本中 statement 现在是解析器转发到引用单元格，而不是 indentedStatements。
另请注意，ws 未使用 spaces 实现，就像在您的解析器中一样。这很重要，因为 spaces 也会消耗换行符，这会阻止 indentedMany1 看到换行符并正确计算缩进。
您的解析器产生 "Expecting: newline" 错误的原因是 indentedMany1 需要在缩进序列的开头换行，以便能够计算缩进。如果你想支持例如，你将不得不修改 indentedMany1 的实现。以下缩进模式：
let x = instruction
instruction
instruction
注意：不久前，我已经问过
例如：
"Hello" 是一个表达式，
let foo = 2 + 1是使用表达式（2 + 1）的指令，
print foo 是一条指令，...
简而言之，一种非常模块化和动态的语法和语义。像 F# 或 OCaml。
为此，我使用 F# 和 API（在 nuget 上可用）FParsec。 FParsec wiki 提供了an example of a syntax based on indentation，所以我重新整理了一下。下面代码中使用的模块是IndentationParserWithoutBacktracking.
要解析的示例代码使用基本缩进，不混合 "literal" 和 "instructions/expressions":
loop i 1 10
loop k 1 10
print k
print i
print j
一个简单的代码，没有上下文（但目前这并不重要）。
我的实现允许这样的代码：
let foo = a + b
let foo =
let a = 9
let b = 1
a + b
let foo = 7
let foo =
loop i 1 10
print i
例如。 （loop 和 print 只是为了测试...）
这个问题困扰了我一个星期，但我无法解决，每次解析器中需要一条新行指令时，缩进模块都会询问我...这是屏幕截图：
这适用于上述所有示例。我不太明白这个问题，因此不知道如何解决。
这是针对此问题测试的代码，它满足最小和功能代码标准，但是，必须使用 FParsec：
open FParsec
// This module come from 'https://github.com/stephan-tolksdorf/fparsec/wiki/Parsing-indentation-based-syntax-with-FParsec'
// I used the second module: 'IndentationParserWithoutBacktracking'
module IndentationParserWithoutBacktracking =
let tabStopDistance = 8
type LastParsedIndentation() =
[<DefaultValue>]
val mutable Value: int32
[<DefaultValue>]
val mutable EndIndex: int64
type UserState =
{Indentation: int
// We put LastParsedIndentation into the UserState so that we
// can conveniently use a separate instance for each stream.
// The members of the LastParsedIndentation instance will be mutated
// directly and hence won't be affected by any stream backtracking.
LastParsedIndentation: LastParsedIndentation}
with
static member Create() = {Indentation = -1
LastParsedIndentation = LastParsedIndentation(EndIndex = -1L)}
type CharStream = CharStream<UserState>
type Parser<'t> = Parser<'t, UserState>
// If this function is called at the same index in the stream
// where the function previously stopped, then the previously
// returned indentation will be returned again.
// This way we can avoid backtracking at the end of indented blocks.
let skipIndentation (stream: CharStream) =
let lastParsedIndentation = stream.UserState.LastParsedIndentation
if lastParsedIndentation.EndIndex = stream.Index then
lastParsedIndentation.Value
else
let mutable indentation = stream.SkipNewlineThenWhitespace(tabStopDistance, false)
lastParsedIndentation.EndIndex <- stream.Index
lastParsedIndentation.Value <- indentation
indentation
let indentedMany1 (p: Parser<'t>) label : Parser<'t list> =
fun stream ->
let oldIndentation = stream.UserState.Indentation
let indentation = skipIndentation stream
if indentation <= oldIndentation then
Reply(Error, expected (if indentation < 0 then "newline" else "indented " + label))
else
stream.UserState <- {stream.UserState with Indentation = indentation}
let results = ResizeArray()
let mutable stateTag = stream.StateTag
let mutable reply = p stream // parse the first element
let mutable newIndentation = 0
while reply.Status = Ok
&& (results.Add(reply.Result)
newIndentation <- skipIndentation stream
newIndentation = indentation)
do
stateTag <- stream.StateTag
reply <- p stream
if reply.Status = Ok
|| (stream.IsEndOfStream && results.Count > 0 && stream.StateTag = stateTag)
then
if newIndentation < indentation || stream.IsEndOfStream then
stream.UserState <- {stream.UserState with Indentation = oldIndentation}
Reply(List.ofSeq results)
else
Reply(Error, messageError "wrong indentation")
else // p failed
Reply(reply.Status, reply.Error)
open IndentationParserWithoutBacktracking
let isBlank = fun c -> c = ' ' || c = '\t'
let ws = spaces
let ws1 = skipMany1SatisfyL isBlank "whitespace"
let str s = pstring s .>> ws
let keyword str = pstring str >>? nextCharSatisfiesNot (fun c -> isLetter c || isDigit c) <?> str
// AST
type Identifier = Identifier of string
// A value is just a literal or a data name, called here "Variable"
type Value =
| Int of int | Float of float
| Bool of bool | String of string
| Char of char | Variable of Identifier
// All is an instruction, but there are some differences:
type Instr =
// Arithmetic
| Literal of Value | Infix of Instr * InfixOp * Instr
// Statements (instructions needing another instructions)
| Let of Identifier * Instr list
| Loop of Identifier * Instr * Instr * Instr list
// Other - the "print" function, from the link seen above
| Print of Identifier
and InfixOp =
// Arithmetic
| Sum | Sub | Mul | Div
// Logic
| And | Or | Equal | NotEqual | Greater | Smaller | GreaterEqual | SmallerEqual
// Literals
let numberFormat = NumberLiteralOptions.AllowMinusSign ||| NumberLiteralOptions.AllowFraction |||
NumberLiteralOptions.AllowHexadecimal ||| NumberLiteralOptions.AllowOctal |||
NumberLiteralOptions.AllowBinary
let literal_numeric =
numberLiteral numberFormat "number" |>> fun nl ->
if nl.IsInteger then Literal (Int(int nl.String))
else Literal (Float(float nl.String))
let literal_bool =
(choice [
(stringReturn "true" (Literal (Bool true)))
(stringReturn "false" (Literal (Bool false)))
]
.>> ws) <?> "boolean"
let literal_string =
(between (pstring "\"") (pstring "\"") (manyChars (satisfy (fun c -> c <> '"')))
|>> fun s -> Literal (String s)) <?> "string"
let literal_char =
(between (pstring "'") (pstring "'") (satisfy (fun c -> c <> '''))
|>> fun c -> Literal (Char c)) <?> "character"
let identifier =
(many1Satisfy2L isLetter (fun c -> isLetter c || isDigit c) "identifier"
|>> Identifier) <?> "identifier"
let betweenParentheses p =
(between (str "(") (str ")") p) <?> ""
let variable = identifier |>> fun id -> Literal (Variable id)
let literal = (attempt literal_numeric <|>
attempt literal_bool <|>
attempt literal_char <|>
attempt literal_string <|>
attempt variable)
// Instressions and statements
let pInstrs, pInstrimpl = createParserForwardedToRef()
// `ploop` is located here to force `pInstrs` to be of the type `Instr list`, `ploop` requesting an instression list.
let ploop =
pipe4
(keyword "loop" >>. ws1 >>. identifier)
(ws1 >>. literal)
(ws1 >>. literal)
(pInstrs)
(fun id min max stmts -> Loop(id, min, max, stmts))
// `singlepInstr` allows to use only one Instression, used just after.
let singlepInstr =
pInstrs |>> fun ex -> ex.Head
let term =
(ws >>. singlepInstr .>> ws) <|>
(betweenParentheses (ws >>. singlepInstr)) <|>
(ws >>. literal .>> ws) <|>
(betweenParentheses (ws >>. literal))
let infixOperator (p: OperatorPrecedenceParser<_, _, _>) op prec map =
p.AddOperator(InfixOperator(op, ws, prec, Associativity.Left, map))
let ops =
// Arithmetic
[ "+"; "-"; "*"; "/"; "%" ] @
// Logical
[ "&&"; "||"; "=="; "!="; ">"; "<"; ">="; "<=" ]
let opCorrespondance op =
match op with
// Arithmetic operators
| "+" -> Sum | "-" -> Sub
| "*" -> Mul | "/" -> Div
// Logical operators
| "&&" -> And | "||" -> Or
| "==" -> Equal | "!=" -> NotEqual
| ">" -> Greater | "<" -> Smaller
| ">=" -> GreaterEqual | "<=" -> SmallerEqual
| _ -> failwith ("Unknown operator: " + op)
let opParser = new OperatorPrecedenceParser<Instr, unit, UserState>()
for op in ops do
infixOperator opParser op 1 (fun x y -> Infix(x, opCorrespondance op, y))
opParser.TermParser <- term
// Statements
(*
- let:
let <identifier> = <instruction(s) / value>
- print:
print <identifier>
- loop:
loop <identifier> <literal> <literal> <indented statements>
*)
let plet =
pipe2
(keyword "let" >>. ws1 >>. identifier)
(ws >>. str "=" >>. ws >>. pInstrs)
(fun id exp -> Let(id, exp))
let print =
keyword "print" >>. ws1 >>. identifier
|>> Print
let instruction =
print <|> ploop <|> plet <|>
opParser.ExpressionParser <|>
literal
pInstrimpl := indentedMany1 instruction "instruction"
let document = pInstrs .>> spaces .>> eof
let test str =
match runParserOnString document (UserState.Create()) "" str with
| Success(result, _, _) -> printfn "%A" result
| Failure(errorMsg, _, _) -> printfn "%s" errorMsg
System.Console.Clear()
let code = test @"
let foo = a + b
"
我想首先了解为什么它不起作用，而且能够找到解决我的问题的方法，并且该解决方案可以扩展到解析器的潜在语法添加。
等待有益的回答，谢谢。
为了理解为什么您的解析器不起作用，您需要隔离问题。
如果我没理解错的话，您希望 let 解析器支持同一行上的单个指令或后续行上的缩进指令，例如：
let x = instruction
let b =
instruction
instruction
如果您无法让现有的实现正常工作，我建议您回到 Wiki 上的实现并尝试只添加对 let 语句的支持。
例如，我通过以下修改使 Wiki 解析器接受简单的 let 语句：
type Statement = Loop of Identifier * int * int * Statement list
| Print of Identifier
| Let of Identifier * Statement list
let ws = skipManySatisfy isBlank
let str s = pstring s .>> ws
let statement, statementRef = createParserForwardedToRef()
let indentedStatements = indentedMany1 statement "statement"
let plet = keyword "let" >>. pipe2 (ws1 >>. identifier)
(ws >>. str "=" >>. ws
>>. (indentedStatements
<|> (statement |>> fun s -> [s])))
(fun id exp -> Let(id, exp))
statementRef := print <|> loop <|> plet
请注意，在修改后的版本中 statement 现在是解析器转发到引用单元格，而不是 indentedStatements。
另请注意，ws 未使用 spaces 实现，就像在您的解析器中一样。这很重要，因为 spaces 也会消耗换行符，这会阻止 indentedMany1 看到换行符并正确计算缩进。
您的解析器产生 "Expecting: newline" 错误的原因是 indentedMany1 需要在缩进序列的开头换行，以便能够计算缩进。如果你想支持例如，你将不得不修改 indentedMany1 的实现。以下缩进模式：
let x = instruction
instruction
instruction