mirror of
https://github.com/kanaka/mal.git
synced 2024-11-10 02:45:44 +03:00
1c76df7dc6
* Change "Array" to "ArraySlice" * Change "as" keyword to "as!" * Use count(str.utf16) rather than str.utf16count) * Add "combine:" parameter label to call to slice.reduce
566 lines
18 KiB
Swift
566 lines
18 KiB
Swift
//******************************************************************************
|
|
// MAL - step 7 - quote
|
|
//******************************************************************************
|
|
// This file is automatically generated from templates/step.swift. Rather than
|
|
// editing it directly, it's probably better to edit templates/step.swift and
|
|
// regenerate this file. Otherwise, your change might be lost if/when someone
|
|
// else performs that process.
|
|
//******************************************************************************
|
|
|
|
import Foundation
|
|
|
|
// The number of times EVAL has been entered recursively. We keep track of this
|
|
// so that we can protect against overrunning the stack.
|
|
//
|
|
var EVAL_level = 0
|
|
|
|
// The maximum number of times we let EVAL recurse before throwing an exception.
|
|
// Testing puts this at some place between 1800 and 1900. Let's keep it at 500
|
|
// for safety's sake.
|
|
//
|
|
let EVAL_leval_max = 500
|
|
|
|
// Control whether or not tail-call optimization (TCO) is enabled. We want it
|
|
// `true` most of the time, but may disable it for debugging purposes (it's
|
|
// easier to get a meaningful backtrace that way).
|
|
//
|
|
let TCO = true
|
|
|
|
// Control whether or not we emit debugging statements in EVAL.
|
|
//
|
|
let DEBUG_EVAL = false
|
|
|
|
// String used to prefix information logged in EVAL. Increasing lengths of the
|
|
// string are used the more EVAL is recursed.
|
|
//
|
|
let INDENT_TEMPLATE = "|----|----|----|----|----|----|----|----|" +
|
|
"----|----|----|----|----|----|----|----|----|----|----|" +
|
|
"----|----|----|----|----|----|----|----|----|----|----|" +
|
|
"----|----|----|----|----|----|----|----|----|----|----|" +
|
|
"----|----|----|----|----|----|----|----|----|----|----|" +
|
|
"----|----|----|----|----|----|----|----|----|----|----|" +
|
|
"----|----|----|----|----|----|----|----|----|----|----|" +
|
|
"----|----|----|----|----|----|----|----|----|----|----|" +
|
|
"----|----|----|----|----|----|----|----|----|----|----|" +
|
|
"----|----|----|----|----|----|----|----|----|----|----|" +
|
|
"----|----|----|----|----|----|----|----|----|----|----|"
|
|
|
|
// Holds the prefix of INDENT_TEMPLATE used for actual logging.
|
|
//
|
|
var indent = String()
|
|
|
|
// Symbols used in this module.
|
|
//
|
|
let kSymbolArgv = MalSymbol(symbol: "*ARGV*")
|
|
let kSymbolConcat = MalSymbol(symbol: "concat")
|
|
let kSymbolCons = MalSymbol(symbol: "cons")
|
|
let kSymbolDef = MalSymbol(symbol: "def!")
|
|
let kSymbolDo = MalSymbol(symbol: "do")
|
|
let kSymbolEval = MalSymbol(symbol: "eval")
|
|
let kSymbolFn = MalSymbol(symbol: "fn*")
|
|
let kSymbolIf = MalSymbol(symbol: "if")
|
|
let kSymbolLet = MalSymbol(symbol: "let*")
|
|
let kSymbolQuasiQuote = MalSymbol(symbol: "quasiquote")
|
|
let kSymbolQuote = MalSymbol(symbol: "quote")
|
|
let kSymbolSpliceUnquote = MalSymbol(symbol: "splice-unquote")
|
|
let kSymbolUnquote = MalSymbol(symbol: "unquote")
|
|
|
|
// Class to help control the incrementing and decrementing of EVAL_level. We
|
|
// create one of these on entry to EVAL, incrementing the level. When the
|
|
// variable goes out of scope, the object is destroyed, decrementing the level.
|
|
//
|
|
class EVAL_Counter {
|
|
init() {
|
|
++EVAL_level
|
|
}
|
|
deinit {
|
|
--EVAL_level
|
|
}
|
|
}
|
|
|
|
// Parse the string into an AST.
|
|
//
|
|
func READ(str: String) -> MalVal {
|
|
return read_str(str)
|
|
}
|
|
|
|
// Return whether or not `val` is a non-empty list.
|
|
//
|
|
func is_pair(val:MalVal) -> Bool {
|
|
if !is_sequence(val) { return false }
|
|
let list = val as! MalSequence
|
|
return !list.isEmpty
|
|
}
|
|
|
|
// Evaluate `quasiquote`, possibly recursing in the process.
|
|
//
|
|
// As with quote, unquote, and splice-unquote, quasiquote takes a single
|
|
// parameter, typically a list. In the general case, this list is processed
|
|
// recursively as:
|
|
//
|
|
// (quasiquote (first rest...)) -> (cons (quasiquote first) (quasiquote rest))
|
|
//
|
|
// In the processing of the parameter passed to it, quasiquote handles three
|
|
// special cases:
|
|
//
|
|
// * If the parameter is an atom or an empty list, the following expression
|
|
// is formed and returned for evaluation:
|
|
//
|
|
// (quasiquote atom-or-empty-list) -> (quote atom-or-empty-list)
|
|
//
|
|
// * If the first element of the non-empty list is the symbol "unquote"
|
|
// followed by a second item, the second item is returned as-is:
|
|
//
|
|
// (quasiquote (unquote fred)) -> fred
|
|
//
|
|
// * If the first element of the non-empty list is another list containing
|
|
// the symbol "splice-unquote" followed by a list, that list is catenated
|
|
// with the quasiquoted result of the remaining items in the non-empty
|
|
// parent list:
|
|
//
|
|
// (quasiquote (splice-unquote list) rest...) -> (items-from-list items-from-quasiquote(rest...))
|
|
//
|
|
// Note the inconsistent handling between "quote" and "splice-quote". The former
|
|
// is handled when this function is handed a list that starts with "quote",
|
|
// whereas the latter is handled when this function is handled a list whose
|
|
// first element is a list that starts with "splice-quote". The handling of the
|
|
// latter is forced by the need to incorporate the results of (splice-quote
|
|
// list) with the remaining items of the list containing that splice-quote
|
|
// expression. However, it's not clear to me why the handling of "unquote" is
|
|
// not handled similarly, for consistency's sake.
|
|
|
|
func quasiquote(qq_arg:MalVal) -> MalVal {
|
|
|
|
// If the argument is an atom or empty list:
|
|
//
|
|
// Return: (quote <argument>)
|
|
|
|
if !is_pair(qq_arg) {
|
|
return MalList(objects: kSymbolQuote, qq_arg)
|
|
}
|
|
|
|
// The argument is a non-empty list -- that is (item rest...)
|
|
|
|
// If the first item from the list is a symbol and it's "unquote" -- that
|
|
// is, (unquote item ignored...):
|
|
//
|
|
// Return: item
|
|
|
|
let qq_list = qq_arg as! MalSequence
|
|
if is_symbol(qq_list.first()) {
|
|
let sym = qq_list.first() as! MalSymbol
|
|
if sym == kSymbolUnquote {
|
|
return qq_list.count >= 2 ? qq_list[1] : MalNil()
|
|
}
|
|
}
|
|
|
|
// If the first item from the list is itself a non-empty list starting with
|
|
// "splice-unquote"-- that is, ((splice-unquote item ignored...) rest...):
|
|
//
|
|
// Return: (concat item quasiquote(rest...))
|
|
|
|
if is_pair(qq_list.first()) {
|
|
let qq_list_item0 = qq_list.first() as! MalSequence
|
|
if is_symbol(qq_list_item0.first()) {
|
|
let sym = qq_list_item0.first() as! MalSymbol
|
|
if sym == kSymbolSpliceUnquote {
|
|
let result = quasiquote(qq_list.rest())
|
|
if is_error(result) { return result }
|
|
return MalList(array: [kSymbolConcat, qq_list_item0[1], result])
|
|
}
|
|
}
|
|
}
|
|
|
|
// General case: (item rest...):
|
|
//
|
|
// Return: (cons (quasiquote item) (quasiquote (rest...))
|
|
|
|
let first = quasiquote(qq_list.first())
|
|
if is_error(first) { return first }
|
|
|
|
let rest = quasiquote(qq_list.rest())
|
|
if is_error(rest) { return rest }
|
|
|
|
return MalList(objects: kSymbolCons, first, rest)
|
|
}
|
|
|
|
// Perform a simple evaluation of the `ast` object. If it's a symbol,
|
|
// dereference it and return its value. If it's a collection, call EVAL on all
|
|
// elements (or just the values, in the case of the hashmap). Otherwise, return
|
|
// the object unchanged.
|
|
//
|
|
func eval_ast(ast: MalVal, env: Environment) -> MalVal {
|
|
switch ast.type {
|
|
case .TypeSymbol:
|
|
let symbol = ast as! MalSymbol
|
|
if let val = env.get(symbol) {
|
|
return val
|
|
}
|
|
return MalError(message: "'\(symbol)' not found") // Specific text needed to match MAL unit tests
|
|
case .TypeList:
|
|
let list = ast as! MalList
|
|
var result = [MalVal]()
|
|
result.reserveCapacity(list.count)
|
|
for item in list {
|
|
let eval = EVAL(item, env)
|
|
if is_error(eval) { return eval }
|
|
result.append(eval)
|
|
}
|
|
return MalList(array: result)
|
|
case .TypeVector:
|
|
let vec = ast as! MalVector
|
|
var result = [MalVal]()
|
|
result.reserveCapacity(vec.count)
|
|
for item in vec {
|
|
let eval = EVAL(item, env)
|
|
if is_error(eval) { return eval }
|
|
result.append(eval)
|
|
}
|
|
return MalVector(array: result)
|
|
case .TypeHashMap:
|
|
let hash = ast as! MalHashMap
|
|
var result = [MalVal]()
|
|
result.reserveCapacity(hash.count * 2)
|
|
for (k, v) in hash {
|
|
let new_v = EVAL(v, env)
|
|
if is_error(new_v) { return new_v }
|
|
result.append(k)
|
|
result.append(new_v)
|
|
}
|
|
return MalHashMap(array: result)
|
|
default:
|
|
return ast
|
|
}
|
|
}
|
|
|
|
enum TCOVal {
|
|
case NoResult
|
|
case Return(MalVal)
|
|
case Continue(MalVal, Environment)
|
|
|
|
init() { self = .NoResult }
|
|
init(_ result: MalVal) { self = .Return(result) }
|
|
init(_ ast: MalVal, _ env: Environment) { self = .Continue(ast, env) }
|
|
init(_ e: String) { self = .Return(MalError(message: e)) }
|
|
}
|
|
|
|
// EVALuate "def!".
|
|
//
|
|
func eval_def(list: MalSequence, env: Environment) -> TCOVal {
|
|
if list.count != 3 {
|
|
return TCOVal("expected 2 arguments to def!, got \(list.count - 1)")
|
|
}
|
|
let arg1 = list[1]
|
|
let arg2 = list[2]
|
|
if !is_symbol(arg1) {
|
|
return TCOVal("expected symbol for first argument to def!")
|
|
}
|
|
let sym = arg1 as! MalSymbol
|
|
let value = EVAL(arg2, env)
|
|
if is_error(value) { return TCOVal(value) }
|
|
return TCOVal(env.set(sym, value))
|
|
}
|
|
|
|
// EVALuate "let*".
|
|
//
|
|
func eval_let(list: MalSequence, env: Environment) -> TCOVal {
|
|
if list.count != 3 {
|
|
return TCOVal("expected 2 arguments to let*, got \(list.count - 1)")
|
|
}
|
|
let arg1 = list[1]
|
|
let arg2 = list[2]
|
|
if !is_sequence(arg1) {
|
|
return TCOVal("expected list for first argument to let*")
|
|
}
|
|
let bindings = arg1 as! MalSequence
|
|
if bindings.count % 2 == 1 {
|
|
return TCOVal("expected even number of elements in bindings to let*, got \(bindings.count)")
|
|
}
|
|
var new_env = Environment(outer: env)
|
|
for var index = 0; index < bindings.count; index += 2 {
|
|
let binding_name = bindings[index]
|
|
let binding_value = bindings[index + 1]
|
|
|
|
if !is_symbol(binding_name) {
|
|
return TCOVal("expected symbol for first element in binding pair")
|
|
}
|
|
let binding_symbol = binding_name as! MalSymbol
|
|
let evaluated_value = EVAL(binding_value, new_env)
|
|
if is_error(evaluated_value) { return TCOVal(evaluated_value) }
|
|
new_env.set(binding_symbol, evaluated_value)
|
|
}
|
|
if TCO {
|
|
return TCOVal(arg2, new_env)
|
|
}
|
|
return TCOVal(EVAL(arg2, new_env))
|
|
}
|
|
|
|
// EVALuate "do".
|
|
//
|
|
func eval_do(list: MalSequence, env: Environment) -> TCOVal {
|
|
if TCO {
|
|
let eval = eval_ast(MalList(slice: list[1..<list.count-1]), env)
|
|
if is_error(eval) { return TCOVal(eval) }
|
|
return TCOVal(list.last(), env)
|
|
}
|
|
|
|
let evaluated_ast = eval_ast(list.rest(), env)
|
|
if is_error(evaluated_ast) { return TCOVal(evaluated_ast) }
|
|
let evaluated_seq = evaluated_ast as! MalSequence
|
|
return TCOVal(evaluated_seq.last())
|
|
}
|
|
|
|
// EVALuate "if".
|
|
//
|
|
func eval_if(list: MalSequence, env: Environment) -> TCOVal {
|
|
if list.count < 3 {
|
|
return TCOVal("expected at least 2 arguments to if, got \(list.count - 1)")
|
|
}
|
|
let cond_result = EVAL(list[1], env)
|
|
var new_ast = MalVal()
|
|
if is_truthy(cond_result) {
|
|
new_ast = list[2]
|
|
} else if list.count == 4 {
|
|
new_ast = list[3]
|
|
} else {
|
|
return TCOVal(MalNil())
|
|
}
|
|
if TCO {
|
|
return TCOVal(new_ast, env)
|
|
}
|
|
return TCOVal(EVAL(new_ast, env))
|
|
}
|
|
|
|
// EVALuate "fn*".
|
|
//
|
|
func eval_fn(list: MalSequence, env: Environment) -> TCOVal {
|
|
if list.count != 3 {
|
|
return TCOVal("expected 2 arguments to fn*, got \(list.count - 1)")
|
|
}
|
|
if !is_sequence(list[1]) {
|
|
return TCOVal("expected list or vector for first argument to fn*")
|
|
}
|
|
return TCOVal(MalClosure(eval: EVAL, args:list[1] as! MalSequence, body:list[2], env:env))
|
|
}
|
|
|
|
// EVALuate "quote".
|
|
//
|
|
func eval_quote(list: MalSequence, env: Environment) -> TCOVal {
|
|
if list.count >= 2 {
|
|
return TCOVal(list[1])
|
|
}
|
|
return TCOVal(MalNil())
|
|
}
|
|
|
|
// EVALuate "quasiquote".
|
|
//
|
|
func eval_quasiquote(list: MalSequence, env: Environment) -> TCOVal {
|
|
if list.count >= 2 {
|
|
if TCO {
|
|
return TCOVal(quasiquote(list[1]), env)
|
|
}
|
|
return TCOVal(EVAL(quasiquote(list[1]), env))
|
|
}
|
|
return TCOVal("Expected non-nil parameter to 'quasiquote'")
|
|
}
|
|
|
|
// Walk the AST and completely evaluate it, handling macro expansions, special
|
|
// forms and function calls.
|
|
//
|
|
func EVAL(var ast: MalVal, var env: Environment) -> MalVal {
|
|
let x = EVAL_Counter()
|
|
if EVAL_level > EVAL_leval_max {
|
|
return MalError(message: "Recursing too many levels (> \(EVAL_leval_max))")
|
|
}
|
|
|
|
if DEBUG_EVAL {
|
|
indent = prefix(INDENT_TEMPLATE, EVAL_level)
|
|
}
|
|
|
|
while true {
|
|
if is_error(ast) { return ast }
|
|
if DEBUG_EVAL { println("\(indent)> \(ast)") }
|
|
|
|
if !is_list(ast) {
|
|
|
|
// Not a list -- just evaluate and return.
|
|
|
|
let answer = eval_ast(ast, env)
|
|
if DEBUG_EVAL { println("\(indent)>>> \(answer)") }
|
|
return answer
|
|
}
|
|
|
|
// Special handling if it's a list.
|
|
|
|
var list = ast as! MalList
|
|
if DEBUG_EVAL { println("\(indent)>. \(list)") }
|
|
|
|
if list.isEmpty {
|
|
return list
|
|
}
|
|
|
|
// Check for special forms, where we want to check the operation
|
|
// before evaluating all of the parameters.
|
|
|
|
let arg0 = list.first()
|
|
if is_symbol(arg0) {
|
|
var res: TCOVal
|
|
let fn_symbol = arg0 as! MalSymbol
|
|
|
|
switch fn_symbol {
|
|
case kSymbolDef: res = eval_def(list, env)
|
|
case kSymbolLet: res = eval_let(list, env)
|
|
case kSymbolDo: res = eval_do(list, env)
|
|
case kSymbolIf: res = eval_if(list, env)
|
|
case kSymbolFn: res = eval_fn(list, env)
|
|
case kSymbolQuote: res = eval_quote(list, env)
|
|
case kSymbolQuasiQuote: res = eval_quasiquote(list, env)
|
|
default: res = TCOVal()
|
|
}
|
|
switch res {
|
|
case let .Return(result): return result
|
|
case let .Continue(new_ast, new_env): ast = new_ast; env = new_env; continue
|
|
case .NoResult: break
|
|
}
|
|
}
|
|
|
|
// Standard list to be applied. Evaluate all the elements first.
|
|
|
|
let eval = eval_ast(ast, env)
|
|
if is_error(eval) { return eval }
|
|
|
|
// The result had better be a list and better be non-empty.
|
|
|
|
let eval_list = eval as! MalList
|
|
if eval_list.isEmpty {
|
|
return eval_list
|
|
}
|
|
|
|
if DEBUG_EVAL { println("\(indent)>> \(eval)") }
|
|
|
|
// Get the first element of the list and execute it.
|
|
|
|
let first = eval_list.first()
|
|
let rest = eval_list.rest()
|
|
|
|
if is_builtin(first) {
|
|
let fn = first as! MalBuiltin
|
|
let answer = fn.apply(rest)
|
|
if DEBUG_EVAL { println("\(indent)>>> \(answer)") }
|
|
return answer
|
|
} else if is_closure(first) {
|
|
let fn = first as! MalClosure
|
|
var new_env = Environment(outer: fn.env)
|
|
let result = new_env.set_bindings(fn.args, with_exprs:rest)
|
|
if is_error(result) { return result }
|
|
if TCO {
|
|
env = new_env
|
|
ast = fn.body
|
|
continue
|
|
}
|
|
let answer = EVAL(fn.body, new_env)
|
|
if DEBUG_EVAL { println("\(indent)>>> \(answer)") }
|
|
return answer
|
|
}
|
|
|
|
// The first element wasn't a function to be executed. Return an
|
|
// error saying so.
|
|
|
|
return MalError(message: "first list item does not evaluate to a function: \(first)")
|
|
}
|
|
}
|
|
|
|
// Convert the value into a human-readable string for printing.
|
|
//
|
|
func PRINT(exp: MalVal) -> String? {
|
|
if is_error(exp) { return nil }
|
|
return pr_str(exp, true)
|
|
}
|
|
|
|
// Perform the READ and EVAL steps. Useful for when you don't care about the
|
|
// printable result.
|
|
//
|
|
func RE(text: String, env: Environment) -> MalVal? {
|
|
if text.isEmpty { return nil }
|
|
let ast = READ(text)
|
|
if is_error(ast) {
|
|
println("Error parsing input: \(ast)")
|
|
return nil
|
|
}
|
|
let exp = EVAL(ast, env)
|
|
if is_error(exp) {
|
|
println("Error evaluating input: \(exp)")
|
|
return nil
|
|
}
|
|
return exp
|
|
}
|
|
|
|
// Perform the full READ/EVAL/PRINT, returning a printable string.
|
|
//
|
|
func REP(text: String, env: Environment) -> String? {
|
|
let exp = RE(text, env)
|
|
if exp == nil { return nil }
|
|
return PRINT(exp!)
|
|
}
|
|
|
|
// Perform the full REPL.
|
|
//
|
|
func REPL(env: Environment) {
|
|
while true {
|
|
if let text = _readline("user> ") {
|
|
if let output = REP(text, env) {
|
|
println("\(output)")
|
|
}
|
|
} else {
|
|
println()
|
|
break
|
|
}
|
|
}
|
|
}
|
|
|
|
// Process any command line arguments. Any trailing arguments are incorporated
|
|
// into the environment. Any argument immediately after the process name is
|
|
// taken as a script to execute. If one exists, it is executed in lieu of
|
|
// running the REPL.
|
|
//
|
|
func process_command_line(args:[String], env:Environment) -> Bool {
|
|
var argv = MalList()
|
|
if args.count > 2 {
|
|
let args1 = args[2..<args.count]
|
|
let args2 = args1.map { MalString(unescaped: $0) as MalVal }
|
|
let args3 = [MalVal](args2)
|
|
argv = MalList(array: args3)
|
|
}
|
|
env.set(kSymbolArgv, argv)
|
|
|
|
if args.count > 1 {
|
|
RE("(load-file \"\(args[1])\")", env)
|
|
return false
|
|
}
|
|
|
|
return true
|
|
}
|
|
|
|
func main() {
|
|
var env = Environment(outer: nil)
|
|
|
|
load_history_file()
|
|
load_builtins(env)
|
|
|
|
RE("(def! not (fn* (a) (if a false true)))", env)
|
|
RE("(def! load-file (fn* (f) (eval (read-string (str \"(do \" (slurp f) \")\")))))", env)
|
|
|
|
env.set(kSymbolEval, MalBuiltin(function: {
|
|
unwrap($0) {
|
|
(ast:MalVal) -> MalVal in
|
|
EVAL(ast, env)
|
|
}
|
|
}))
|
|
|
|
if process_command_line(Process.arguments, env) {
|
|
REPL(env)
|
|
}
|
|
|
|
save_history_file()
|
|
}
|