libs/JSON.bend source
libs/JSON.bend on the hub · documented module
import Base # JSON values preserve number lexemes as strings. This keeps integers and # fractional/exponent forms lossless without imposing a fixed numeric width. # Number carries grammar evidence so every constructible number is valid JSON. # Use number(text) or parse(text) to create numbers from runtime strings. def BoolType() -> Data: Bool type NumberCert<-s: String> is Data: Certified{witness: NumberProof(s)} type Value is Data: Null{} Bool{value: BoolType()} Number{lexeme: String, certificate: NumberCert<lexeme>} Str{value: String} Arr{values: List<&2, Value>} Obj{fields: List<&2, Sigma<&2, &2, String, key => Value>>} type Error is Data: Error{} type Parsed is Type: Parsed{value: Value, rest: String} type StringScan is Data: StringScan{value: String, rest: String} type StringScanState is Type: ScanText{input: String, reversed: String} ScanControl{character: Char, rest: String, reversed: String, control: BoolType()} ScanUnicode{parsed: Result<&1, &1, Error, HexScan>, reversed: String} ScanUnicodeHigh{code: U32, rest: String, reversed: String, is_high: BoolType()} ScanUnicodeLow{high: U32, original: String, parsed: Result<&1, &1, Error, HexScan>, reversed: String} ScanUnicodeLowClass{high: U32, low: U32, rest: String, original: String, reversed: String, valid: BoolType()} type NumberState is Data: NumberStart{} NumberMinus{} NumberZero{} NumberInteger{} NumberDot{} NumberFraction{} NumberExponent{} NumberExponentSign{} NumberExponentDigits{} NumberInvalid{} def whitespace(s: String) -> String: match s: case SNil{}: SNil{} case SCon{+h, t}: match h: case ' ': whitespace(t) case '\t': whitespace(t) case '\r': whitespace(t) case '\n': whitespace(t) case _: SCon{h, t} def is_control(c: Char) -> BoolType(): match c: case '\u{0}': True{} case '\u{1}': True{} case '\u{2}': True{} case '\u{3}': True{} case '\u{4}': True{} case '\u{5}': True{} case '\u{6}': True{} case '\u{7}': True{} case '\u{8}': True{} case '\u{9}': True{} case '\u{a}': True{} case '\u{b}': True{} case '\u{c}': True{} case '\u{d}': True{} case '\u{e}': True{} case '\u{f}': True{} case '\u{10}': True{} case '\u{11}': True{} case '\u{12}': True{} case '\u{13}': True{} case '\u{14}': True{} case '\u{15}': True{} case '\u{16}': True{} case '\u{17}': True{} case '\u{18}': True{} case '\u{19}': True{} case '\u{1a}': True{} case '\u{1b}': True{} case '\u{1c}': True{} case '\u{1d}': True{} case '\u{1e}': True{} case '\u{1f}': True{} case _: False{} def escape_char_special(c: Char) -> String: match c: case '\u{0}': "\\u0000" case '\u{1}': "\\u0001" case '\u{2}': "\\u0002" case '\u{3}': "\\u0003" case '\u{4}': "\\u0004" case '\u{5}': "\\u0005" case '\u{6}': "\\u0006" case '\u{7}': "\\u0007" case '\u{8}': "\\b" case '\u{9}': "\\t" case '\u{a}': "\\n" case '\u{b}': "\\u000b" case '\u{c}': "\\f" case '\u{d}': "\\r" case '\u{e}': "\\u000e" case '\u{f}': "\\u000f" case '\u{10}': "\\u0010" case '\u{11}': "\\u0011" case '\u{12}': "\\u0012" case '\u{13}': "\\u0013" case '\u{14}': "\\u0014" case '\u{15}': "\\u0015" case '\u{16}': "\\u0016" case '\u{17}': "\\u0017" case '\u{18}': "\\u0018" case '\u{19}': "\\u0019" case '\u{1a}': "\\u001a" case '\u{1b}': "\\u001b" case '\u{1c}': "\\u001c" case '\u{1d}': "\\u001d" case '\u{1e}': "\\u001e" case '\u{1f}': "\\u001f" case '\u{22}': "\\\"" case '\u{5c}': "\\\\" case _: SCon{c, SNil{}} type EscapeClass is Data: Raw{} Backslash{} Quote{} Control0{} Control1{} Control2{} Control3{} Control4{} Control5{} Control6{} Control7{} Control8{} Control9{} Control10{} Control11{} Control12{} Control13{} Control14{} Control15{} Control16{} Control17{} Control18{} Control19{} Control20{} Control21{} Control22{} Control23{} Control24{} Control25{} Control26{} Control27{} Control28{} Control29{} Control30{} Control31{} def escape_class(character: Char) -> EscapeClass: match character: case '\u{0}': Control0{} case '\u{1}': Control1{} case '\u{2}': Control2{} case '\u{3}': Control3{} case '\u{4}': Control4{} case '\u{5}': Control5{} case '\u{6}': Control6{} case '\u{7}': Control7{} case '\u{8}': Control8{} case '\u{9}': Control9{} case '\u{a}': Control10{} case '\u{b}': Control11{} case '\u{c}': Control12{} case '\u{d}': Control13{} case '\u{e}': Control14{} case '\u{f}': Control15{} case '\u{10}': Control16{} case '\u{11}': Control17{} case '\u{12}': Control18{} case '\u{13}': Control19{} case '\u{14}': Control20{} case '\u{15}': Control21{} case '\u{16}': Control22{} case '\u{17}': Control23{} case '\u{18}': Control24{} case '\u{19}': Control25{} case '\u{1a}': Control26{} case '\u{1b}': Control27{} case '\u{1c}': Control28{} case '\u{1d}': Control29{} case '\u{1e}': Control30{} case '\u{1f}': Control31{} case '"': Quote{} case '\\': Backslash{} case _: Raw{} type QuoteAction is Data: QuoteBackslash{} QuoteEnd{} QuotePlain{} def quote_action_from_class(class: EscapeClass) -> QuoteAction: match class: case Backslash{}: QuoteBackslash{} case Quote{}: QuoteEnd{} case _: QuotePlain{} def quote_action(character: Char) -> QuoteAction: quote_action_from_class(escape_class(character)) def quote_lex_action(action: QuoteAction) -> LexAction: match action: case QuoteBackslash{}: LexBackslash{} case QuoteEnd{}: LexStringEnd{} case QuotePlain{}: LexStringChar{} def escape_char_from_class(character: Char, class: EscapeClass) -> String: match class: case Raw{}: SCon{character, SNil{}} case Backslash{}: "\\\\" case Quote{}: "\\\"" case Control0{}: "\\u0000" case Control1{}: "\\u0001" case Control2{}: "\\u0002" case Control3{}: "\\u0003" case Control4{}: "\\u0004" case Control5{}: "\\u0005" case Control6{}: "\\u0006" case Control7{}: "\\u0007" case Control8{}: "\\b" case Control9{}: "\\t" case Control10{}: "\\n" case Control11{}: "\\u000b" case Control12{}: "\\f" case Control13{}: "\\r" case Control14{}: "\\u000e" case Control15{}: "\\u000f" case Control16{}: "\\u0010" case Control17{}: "\\u0011" case Control18{}: "\\u0012" case Control19{}: "\\u0013" case Control20{}: "\\u0014" case Control21{}: "\\u0015" case Control22{}: "\\u0016" case Control23{}: "\\u0017" case Control24{}: "\\u0018" case Control25{}: "\\u0019" case Control26{}: "\\u001a" case Control27{}: "\\u001b" case Control28{}: "\\u001c" case Control29{}: "\\u001d" case Control30{}: "\\u001e" case Control31{}: "\\u001f" def escape_char(+character: Char) -> String: escape_char_from_class(character, escape_class(character)) def escape_string(s: String) -> String: match s: case SNil{}: "" case SCon{h, t}: escape_char(h) ++ escape_string(t) # Structural rendering has no fuel limit. Tail mode emits the comma before # each subsequent member and the closing delimiter at the empty tail. def render_object_field(prefix: String, key: String, value_text: String) -> String: (prefix ++ ("\"" ++ escape_string(key) ++ "\":")) ++ value_text def render(value: Value, members: BoolType(), suffix: String) -> String: match value: case Null{}: "null" ++ suffix case Bool{truth}: match truth: case True{}: "true" ++ suffix case False{}: "false" ++ suffix case Number{lexeme, certificate}: lexeme ++ suffix case Str{text}: "\"" ++ escape_string(text) ++ "\"" ++ suffix case Arr{Nil{}}: match members: case False{}: "[]" ++ suffix case True{}: "]" ++ suffix case Arr{head <> tail}: match members: case False{}: "[" ++ render(head, False{}, render(Arr{tail}, True{}, suffix)) case True{}: "," ++ render(head, False{}, render(Arr{tail}, True{}, suffix)) case Obj{Nil{}}: match members: case False{}: "{}" ++ suffix case True{}: "}" ++ suffix case Obj{(key, value) <> tail}: match members: case False{}: render_object_field("{", key, render(value, False{}, render(Obj{tail}, True{}, suffix))) case True{}: render_object_field(",", key, render(value, False{}, render(Obj{tail}, True{}, suffix))) def stringify(value: Value) -> String: render(value, False{}, SNil{}) # A hexadecimal escape is exactly four digits. The code preserves standalone # UTF-16 surrogate escapes and combines adjacent high/low pairs into a scalar. # This accepts the JSON string grammar without silently replacing code units. type HexScan is Data: HexScan{code: U32, rest: String} def hex_digit(c: Char) -> Maybe<&2, U32>: match c: case '0': Some{0} case '1': Some{1} case '2': Some{2} case '3': Some{3} case '4': Some{4} case '5': Some{5} case '6': Some{6} case '7': Some{7} case '8': Some{8} case '9': Some{9} case 'a': Some{10} case 'b': Some{11} case 'c': Some{12} case 'd': Some{13} case 'e': Some{14} case 'f': Some{15} case 'A': Some{10} case 'B': Some{11} case 'C': Some{12} case 'D': Some{13} case 'E': Some{14} case 'F': Some{15} case _: None{} def scan_hex(n: Nat, s: String, acc: U32) -> Result<&1, &1, Error, HexScan>: match n: case 0n: Done{HexScan{acc, s}} case 1n+p: match s: case SNil{}: Fail{Error{}} case SCon{'0', t}: scan_hex(p, t, (acc * 16 + 0 : U32)) case SCon{'1', t}: scan_hex(p, t, (acc * 16 + 1 : U32)) case SCon{'2', t}: scan_hex(p, t, (acc * 16 + 2 : U32)) case SCon{'3', t}: scan_hex(p, t, (acc * 16 + 3 : U32)) case SCon{'4', t}: scan_hex(p, t, (acc * 16 + 4 : U32)) case SCon{'5', t}: scan_hex(p, t, (acc * 16 + 5 : U32)) case SCon{'6', t}: scan_hex(p, t, (acc * 16 + 6 : U32)) case SCon{'7', t}: scan_hex(p, t, (acc * 16 + 7 : U32)) case SCon{'8', t}: scan_hex(p, t, (acc * 16 + 8 : U32)) case SCon{'9', t}: scan_hex(p, t, (acc * 16 + 9 : U32)) case SCon{'a', t}: scan_hex(p, t, (acc * 16 + 10 : U32)) case SCon{'b', t}: scan_hex(p, t, (acc * 16 + 11 : U32)) case SCon{'c', t}: scan_hex(p, t, (acc * 16 + 12 : U32)) case SCon{'d', t}: scan_hex(p, t, (acc * 16 + 13 : U32)) case SCon{'e', t}: scan_hex(p, t, (acc * 16 + 14 : U32)) case SCon{'f', t}: scan_hex(p, t, (acc * 16 + 15 : U32)) case SCon{'A', t}: scan_hex(p, t, (acc * 16 + 10 : U32)) case SCon{'B', t}: scan_hex(p, t, (acc * 16 + 11 : U32)) case SCon{'C', t}: scan_hex(p, t, (acc * 16 + 12 : U32)) case SCon{'D', t}: scan_hex(p, t, (acc * 16 + 13 : U32)) case SCon{'E', t}: scan_hex(p, t, (acc * 16 + 14 : U32)) case SCon{'F', t}: scan_hex(p, t, (acc * 16 + 15 : U32)) case _: Fail{Error{}} def string_push(c: Char, result: Result<&1, &1, Error, StringScan>) -> Result<&1, &1, Error, StringScan>: match result: case Fail{e}: Fail{e} case Done{scan}: match scan: case StringScan{s, rest}: Done{StringScan{SCon{c, s}, rest}} def scan_string_machine(fuel: Nat, state: StringScanState) -> Result<&1, &1, Error, StringScan>: match fuel: case 0n: Fail{Error{}} case 1n+p: match state: case ScanText{input, reversed}: match input: case SNil{}: Fail{Error{}} case SCon{+h, +t}: match h: case '"': Done{StringScan{String.reverse(reversed), t}} case '\\': match t: case SNil{}: Fail{Error{}} case SCon{e, et}: match e: case '"': scan_string_machine(p, ScanText{et, SCon{'"', reversed}}) case '\\': scan_string_machine(p, ScanText{et, SCon{'\\', reversed}}) case '/': scan_string_machine(p, ScanText{et, SCon{'/', reversed}}) case 'b': scan_string_machine(p, ScanText{et, SCon{'\u{8}', reversed}}) case 'f': scan_string_machine(p, ScanText{et, SCon{'\u{c}', reversed}}) case 'n': scan_string_machine(p, ScanText{et, SCon{'\n', reversed}}) case 'r': scan_string_machine(p, ScanText{et, SCon{'\r', reversed}}) case 't': scan_string_machine(p, ScanText{et, SCon{'\t', reversed}}) case 'u': scan_string_machine(p, ScanUnicode{scan_hex(4n, et, 0), reversed}) case _: Fail{Error{}} case _: scan_string_machine(p, ScanControl{h, t, reversed, is_control(h)}) case ScanControl{character, rest, reversed, control}: match control: case True{}: Fail{Error{}} case False{}: scan_string_machine(p, ScanText{rest, SCon{character, reversed}}) case ScanUnicode{parsed, reversed}: match parsed: case Fail{e}: Fail{e} case Done{scan}: match scan: case HexScan{+code, +rest}: scan_string_machine(p, ScanUnicodeHigh{code, rest, reversed, U32.is_ge(code, 55296) && U32.is_le(code, 56319)}) case ScanUnicodeHigh{code, +rest, +reversed, is_high}: match rest: case SCon{'\\', SCon{'u', t}}: match is_high: case True{}: scan_string_machine(p, ScanUnicodeLow{code, rest, scan_hex(4n, t, 0), reversed}) case False{}: scan_string_machine(p, ScanText{rest, SCon{Char.from_u32(code), reversed}}) case _: scan_string_machine(p, ScanText{rest, SCon{Char.from_u32(code), reversed}}) case ScanUnicodeLow{high, original, parsed, reversed}: match parsed: case Fail{e}: scan_string_machine(p, ScanText{original, SCon{Char.from_u32(high), reversed}}) case Done{scan}: match scan: case HexScan{+low, +rest}: scan_string_machine(p, ScanUnicodeLowClass{high, low, rest, original, reversed, U32.is_ge(low, 56320) && U32.is_le(low, 57343)}) case ScanUnicodeLowClass{high, low, rest, original, reversed, valid}: match valid: case True{}: scan_string_machine(p, ScanText{rest, SCon{Char.from_u32((65536 + (high - 55296) * 1024 + (low - 56320) : U32)), reversed}}) case False{}: scan_string_machine(p, ScanText{original, SCon{Char.from_u32(high), reversed}}) def scan_string(+s: String) -> Result<&1, &1, Error, StringScan>: scan_string_machine((String.length(s) * 2n + 16n : Nat), ScanText{s, SNil{}}) def scan_number(s: String, acc: String) -> StringScan: match s: case SNil{}: StringScan{String.reverse(acc), SNil{}} case SCon{h, t}: match h: case '-': scan_number(t, SCon{h, acc}) case '+': scan_number(t, SCon{h, acc}) case '.': scan_number(t, SCon{h, acc}) case 'e': scan_number(t, SCon{h, acc}) case 'E': scan_number(t, SCon{h, acc}) case '0': scan_number(t, SCon{h, acc}) case '1': scan_number(t, SCon{h, acc}) case '2': scan_number(t, SCon{h, acc}) case '3': scan_number(t, SCon{h, acc}) case '4': scan_number(t, SCon{h, acc}) case '5': scan_number(t, SCon{h, acc}) case '6': scan_number(t, SCon{h, acc}) case '7': scan_number(t, SCon{h, acc}) case '8': scan_number(t, SCon{h, acc}) case '9': scan_number(t, SCon{h, acc}) case _: StringScan{String.reverse(acc), SCon{h, t}} type NumberChar is Data: NMinus{} NPlus{} NDot{} NExp{} NZero{} NDigit{} NOther{} def number_char(c: Char) -> NumberChar: match c: case '-': NMinus{} case '+': NPlus{} case '.': NDot{} case 'e': NExp{} case 'E': NExp{} case '0': NZero{} case '1': NDigit{} case '2': NDigit{} case '3': NDigit{} case '4': NDigit{} case '5': NDigit{} case '6': NDigit{} case '7': NDigit{} case '8': NDigit{} case '9': NDigit{} case _: NOther{} def number_step(character: NumberChar, state: NumberState) -> NumberState: match character state: case NOther{} _: NumberInvalid{} case NMinus{} NumberStart{}: NumberMinus{} case NMinus{} NumberExponent{}: NumberExponentSign{} case NPlus{} NumberExponent{}: NumberExponentSign{} case NDot{} NumberZero{}: NumberDot{} case NDot{} NumberInteger{}: NumberDot{} case NExp{} NumberZero{}: NumberExponent{} case NExp{} NumberInteger{}: NumberExponent{} case NExp{} NumberFraction{}: NumberExponent{} case NZero{} NumberStart{}: NumberZero{} case NZero{} NumberMinus{}: NumberZero{} case NZero{} NumberInteger{}: NumberInteger{} case NZero{} NumberDot{}: NumberFraction{} case NZero{} NumberFraction{}: NumberFraction{} case NZero{} NumberExponent{}: NumberExponentDigits{} case NZero{} NumberExponentSign{}: NumberExponentDigits{} case NZero{} NumberExponentDigits{}: NumberExponentDigits{} case NDigit{} NumberStart{}: NumberInteger{} case NDigit{} NumberMinus{}: NumberInteger{} case NDigit{} NumberInteger{}: NumberInteger{} case NDigit{} NumberDot{}: NumberFraction{} case NDigit{} NumberFraction{}: NumberFraction{} case NDigit{} NumberExponent{}: NumberExponentDigits{} case NDigit{} NumberExponentSign{}: NumberExponentDigits{} case NDigit{} NumberExponentDigits{}: NumberExponentDigits{} case _ _: NumberInvalid{} def number_next(state: NumberState, c: Char) -> NumberState: match state: case NumberInvalid{}: NumberInvalid{} case _: number_step(number_char(c), state) def number_valid_end(state: NumberState) -> BoolType(): match state: case NumberZero{}: True{} case NumberInteger{}: True{} case NumberFraction{}: True{} case NumberExponentDigits{}: True{} case _: False{} def number_valid_step(s: String, state: NumberState) -> NumberState: match s state: case _ NumberInvalid{}: NumberInvalid{} case SNil{} _: state case SCon{h, t} _: number_valid_step(t, number_next(state, h)) def number_valid(+s: String) -> BoolType(): number_valid_end(number_valid_step(s, NumberStart{})) def NumberProofFrom(valid: BoolType()) -> Data: match valid: case True{}: Unit case False{}: Empty def NumberProof(s: String) -> Data: NumberProofFrom(number_valid(s)) def null_token(s: String, matches: BoolType()) -> Result<&1, &1, Error, Parsed>: match matches: case True{}: Done{Parsed{Null{}, String.drop(s, 4n)}} case False{}: Fail{Error{}} def true_token(s: String, matches: BoolType()) -> Result<&1, &1, Error, Parsed>: match matches: case True{}: Done{Parsed{Bool{True{}}, String.drop(s, 4n)}} case False{}: Fail{Error{}} def false_token(s: String, matches: BoolType()) -> Result<&1, &1, Error, Parsed>: match matches: case True{}: Done{Parsed{Bool{False{}}, String.drop(s, 5n)}} case False{}: Fail{Error{}} def from_string_scan(scanned: Result<&1, &1, Error, StringScan>) -> Result<&1, &1, Error, Parsed>: match scanned: case Fail{e}: Fail{e} case Done{scan}: match scan: case StringScan{value, rest}: Done{Parsed{Str{value}, rest}} def number_proof(+lexeme: String, evidence: {number_valid(lexeme) == True{} : BoolType()}) -> NumberProof(lexeme): %Equal.sym(BoolType(), number_valid(lexeme), True{}, evidence) : NumberProofFrom(_) Unit{} def number_lexeme(+lexeme: String, rest: String, valid: BoolType(), evidence: {number_valid(lexeme) == valid : BoolType()}) -> Result<&1, &1, Error, Parsed>: match valid: case True{}: Done{Parsed{Number{lexeme, Certified{number_proof(lexeme, evidence)}}, rest}} case False{}: Fail{Error{}} def from_number_scan(scan: StringScan) -> Result<&1, &1, Error, Parsed>: match scan: case StringScan{+lexeme, rest}: number_lexeme(lexeme, rest, number_valid(lexeme), {==}) def parse_number(s: String) -> Result<&1, &1, Error, Parsed>: from_number_scan(scan_number(s, SNil{})) def number_from_digit(s: String, digit: BoolType()) -> Result<&1, &1, Error, Parsed>: match digit: case True{}: parse_number(s) case False{}: Fail{Error{}} # The parser is an explicit machine. Its fuel decreases on every state # transition, including transitions between container contexts. type ParserKont is Data: Top{} InArray{acc: List<&2, Value>, parent: ParserKont} InObject{key: String, acc: List<&2, Sigma<&2, &2, String, key => Value>>, parent: ParserKont} type ParserState is Type: ParseValue{clean: String, kont: ParserKont} ParsedResult{result: Result<&1, &1, Error, Parsed>, kont: ParserKont} StringResult{result: Result<&1, &1, Error, StringScan>, kont: ParserKont} ArrayStart{clean: String, acc: List<&2, Value>, empty_allowed: BoolType(), kont: ParserKont} ArrayAfter{value: Value, clean: String, acc: List<&2, Value>, kont: ParserKont} ObjectStart{clean: String, acc: List<&2, Sigma<&2, &2, String, key => Value>>, empty_allowed: BoolType(), kont: ParserKont} ObjectKey{result: Result<&1, &1, Error, StringScan>, acc: List<&2, Sigma<&2, &2, String, key => Value>>, kont: ParserKont} ObjectColon{key: String, clean: String, acc: List<&2, Sigma<&2, &2, String, key => Value>>, kont: ParserKont} ObjectAfter{key: String, value: Value, clean: String, acc: List<&2, Sigma<&2, &2, String, key => Value>>, kont: ParserKont} Complete{parsed: Parsed} def resume_value(value: Value, rest: String, kont: ParserKont) -> ParserState: match kont: case Top{}: Complete{Parsed{value, rest}} case InArray{acc, parent}: ArrayAfter{value, whitespace(rest), acc, parent} case InObject{key, acc, parent}: ObjectAfter{key, value, whitespace(rest), acc, parent} def parse_machine(fuel: Nat, state: ParserState) -> Result<&1, &1, Error, Parsed>: match fuel: case 0n: Fail{Error{}} case 1n+p: match state: case Complete{parsed}: Done{parsed} case ParseValue{+clean, +kont}: match clean: case SNil{}: Fail{Error{}} case SCon{+h, +t}: match h: case 'n': parse_machine(p, ParsedResult{null_token(clean, String.starts_with(clean, "null")), kont}) case 't': parse_machine(p, ParsedResult{true_token(clean, String.starts_with(clean, "true")), kont}) case 'f': parse_machine(p, ParsedResult{false_token(clean, String.starts_with(clean, "false")), kont}) case '"': +budget = p parse_machine(budget, StringResult{scan_string_machine(budget, ScanText{t, SNil{}}), kont}) case '[': parse_machine(p, ArrayStart{whitespace(t), Nil{}, True{}, kont}) case '{': parse_machine(p, ObjectStart{whitespace(t), Nil{}, True{}, kont}) case '-': parse_machine(p, ParsedResult{parse_number(clean), kont}) case _: parse_machine(p, ParsedResult{number_from_digit(clean, Char.is_digit(h)), kont}) case ParsedResult{result, kont}: match result: case Fail{e}: Fail{e} case Done{parsed}: match parsed: case Parsed{value, rest}: parse_machine(p, resume_value(value, rest, kont)) case StringResult{result, kont}: match result: case Fail{e}: Fail{e} case Done{scan}: match scan: case StringScan{value, rest}: parse_machine(p, resume_value(Str{value}, rest, kont)) case ArrayStart{clean, acc, empty_allowed, kont}: match clean: case SNil{}: Fail{Error{}} case SCon{h, t}: match h: case ']': match empty_allowed: case True{}: parse_machine(p, resume_value(Arr{List.reverse(&2, Value, acc)}, t, kont)) case False{}: Fail{Error{}} case _: parse_machine(p, ParseValue{clean, InArray{acc, kont}}) case ArrayAfter{value, clean, acc, kont}: match clean: case SNil{}: Fail{Error{}} case SCon{h, t}: match h: case ']': parse_machine(p, resume_value(Arr{List.reverse(&2, Value, value <> acc)}, t, kont)) case ',': parse_machine(p, ArrayStart{whitespace(t), value <> acc, False{}, kont}) case _: Fail{Error{}} case ObjectStart{clean, acc, empty_allowed, kont}: match clean: case SNil{}: Fail{Error{}} case SCon{h, t}: match h: case '}': match empty_allowed: case True{}: parse_machine(p, resume_value(Obj{List.reverse(&2, Sigma<&2, &2, String, key => Value>, acc)}, t, kont)) case False{}: Fail{Error{}} case '"': +budget = p parse_machine(budget, ObjectKey{scan_string_machine(budget, ScanText{t, SNil{}}), acc, kont}) case _: Fail{Error{}} case ObjectKey{result, acc, kont}: match result: case Fail{e}: Fail{e} case Done{scan}: match scan: case StringScan{key, rest}: parse_machine(p, ObjectColon{key, whitespace(rest), acc, kont}) case ObjectColon{key, clean, acc, kont}: match clean: case SNil{}: Fail{Error{}} case SCon{h, t}: match h: case ':': parse_machine(p, ParseValue{whitespace(t), InObject{key, acc, kont}}) case _: Fail{Error{}} case ObjectAfter{key, value, clean, acc, kont}: match clean: case SNil{}: Fail{Error{}} case SCon{h, t}: match h: case '}': parse_machine(p, resume_value(Obj{List.reverse(&2, Sigma<&2, &2, String, key => Value>, (key, value) <> acc)}, t, kont)) case ',': parse_machine(p, ObjectStart{whitespace(t), (key, value) <> acc, False{}, kont}) case _: Fail{Error{}} def parse_value(+s: String) -> Result<&1, &1, Error, Parsed>: parse_machine((String.length(s) * 8n + 16n : Nat), ParseValue{whitespace(s), Top{}}) def finish_rest_clean(value: Value, clean: String) -> Result<&1, &1, Error, Value>: match clean: case SNil{}: Done{value} case _: Fail{Error{}} def finish_rest(value: Value, rest: String) -> Result<&1, &1, Error, Value>: finish_rest_clean(value, whitespace(rest)) def finish(r: Result<&1, &1, Error, Parsed>) -> Result<&1, &1, Error, Value>: match r: case Fail{e}: Fail{e} case Done{parsed}: match parsed: case Parsed{value, rest}: finish_rest(value, rest) # Common JSON escapes have a structural decoder. Less common Unicode # escape spellings use the complete UTF-16 decoder below through the fallback. type FastString is Data: Decoded{text: String} InvalidString{} UnicodeFallback{} def fast_push(result: FastString, character: Char) -> FastString: match result: case Decoded{text}: Decoded{SCon{character, text}} case InvalidString{}: InvalidString{} case UnicodeFallback{}: UnicodeFallback{} def fast_raw(control: BoolType(), result: FastString, character: Char) -> FastString: match control: case True{}: InvalidString{} case False{}: fast_push(result, character) def fast_string(text: String) -> FastString: match text: case SCon{'"', SNil{}}: Decoded{SNil{}} case SCon{'\\', SCon{'"', tail}}: fast_push(fast_string(tail), '\u{22}') case SCon{'\\', SCon{'\\', tail}}: fast_push(fast_string(tail), '\u{5c}') case SCon{'\\', SCon{'/', tail}}: fast_push(fast_string(tail), '/') case SCon{'\\', SCon{'b', tail}}: fast_push(fast_string(tail), '\u{8}') case SCon{'\\', SCon{'f', tail}}: fast_push(fast_string(tail), '\u{c}') case SCon{'\\', SCon{'n', tail}}: fast_push(fast_string(tail), '\n') case SCon{'\\', SCon{'r', tail}}: fast_push(fast_string(tail), '\r') case SCon{'\\', SCon{'t', tail}}: fast_push(fast_string(tail), '\t') case SCon{'\\', SCon{'u', SCon{'0', SCon{'0', SCon{'0', SCon{'0', tail}}}}}}: fast_push(fast_string(tail), '\u{0}') case SCon{'\\', SCon{'u', SCon{'0', SCon{'0', SCon{'0', SCon{'1', tail}}}}}}: fast_push(fast_string(tail), '\u{1}') case SCon{'\\', SCon{'u', SCon{'0', SCon{'0', SCon{'0', SCon{'2', tail}}}}}}: fast_push(fast_string(tail), '\u{2}') case SCon{'\\', SCon{'u', SCon{'0', SCon{'0', SCon{'0', SCon{'3', tail}}}}}}: fast_push(fast_string(tail), '\u{3}') case SCon{'\\', SCon{'u', SCon{'0', SCon{'0', SCon{'0', SCon{'4', tail}}}}}}: fast_push(fast_string(tail), '\u{4}') case SCon{'\\', SCon{'u', SCon{'0', SCon{'0', SCon{'0', SCon{'5', tail}}}}}}: fast_push(fast_string(tail), '\u{5}') case SCon{'\\', SCon{'u', SCon{'0', SCon{'0', SCon{'0', SCon{'6', tail}}}}}}: fast_push(fast_string(tail), '\u{6}') case SCon{'\\', SCon{'u', SCon{'0', SCon{'0', SCon{'0', SCon{'7', tail}}}}}}: fast_push(fast_string(tail), '\u{7}') case SCon{'\\', SCon{'u', SCon{'0', SCon{'0', SCon{'0', SCon{'8', tail}}}}}}: fast_push(fast_string(tail), '\u{8}') case SCon{'\\', SCon{'u', SCon{'0', SCon{'0', SCon{'0', SCon{'9', tail}}}}}}: fast_push(fast_string(tail), '\u{9}') case SCon{'\\', SCon{'u', SCon{'0', SCon{'0', SCon{'0', SCon{'a', tail}}}}}}: fast_push(fast_string(tail), '\u{a}') case SCon{'\\', SCon{'u', SCon{'0', SCon{'0', SCon{'0', SCon{'b', tail}}}}}}: fast_push(fast_string(tail), '\u{b}') case SCon{'\\', SCon{'u', SCon{'0', SCon{'0', SCon{'0', SCon{'c', tail}}}}}}: fast_push(fast_string(tail), '\u{c}') case SCon{'\\', SCon{'u', SCon{'0', SCon{'0', SCon{'0', SCon{'d', tail}}}}}}: fast_push(fast_string(tail), '\u{d}') case SCon{'\\', SCon{'u', SCon{'0', SCon{'0', SCon{'0', SCon{'e', tail}}}}}}: fast_push(fast_string(tail), '\u{e}') case SCon{'\\', SCon{'u', SCon{'0', SCon{'0', SCon{'0', SCon{'f', tail}}}}}}: fast_push(fast_string(tail), '\u{f}') case SCon{'\\', SCon{'u', SCon{'0', SCon{'0', SCon{'1', SCon{'0', tail}}}}}}: fast_push(fast_string(tail), '\u{10}') case SCon{'\\', SCon{'u', SCon{'0', SCon{'0', SCon{'1', SCon{'1', tail}}}}}}: fast_push(fast_string(tail), '\u{11}') case SCon{'\\', SCon{'u', SCon{'0', SCon{'0', SCon{'1', SCon{'2', tail}}}}}}: fast_push(fast_string(tail), '\u{12}') case SCon{'\\', SCon{'u', SCon{'0', SCon{'0', SCon{'1', SCon{'3', tail}}}}}}: fast_push(fast_string(tail), '\u{13}') case SCon{'\\', SCon{'u', SCon{'0', SCon{'0', SCon{'1', SCon{'4', tail}}}}}}: fast_push(fast_string(tail), '\u{14}') case SCon{'\\', SCon{'u', SCon{'0', SCon{'0', SCon{'1', SCon{'5', tail}}}}}}: fast_push(fast_string(tail), '\u{15}') case SCon{'\\', SCon{'u', SCon{'0', SCon{'0', SCon{'1', SCon{'6', tail}}}}}}: fast_push(fast_string(tail), '\u{16}') case SCon{'\\', SCon{'u', SCon{'0', SCon{'0', SCon{'1', SCon{'7', tail}}}}}}: fast_push(fast_string(tail), '\u{17}') case SCon{'\\', SCon{'u', SCon{'0', SCon{'0', SCon{'1', SCon{'8', tail}}}}}}: fast_push(fast_string(tail), '\u{18}') case SCon{'\\', SCon{'u', SCon{'0', SCon{'0', SCon{'1', SCon{'9', tail}}}}}}: fast_push(fast_string(tail), '\u{19}') case SCon{'\\', SCon{'u', SCon{'0', SCon{'0', SCon{'1', SCon{'a', tail}}}}}}: fast_push(fast_string(tail), '\u{1a}') case SCon{'\\', SCon{'u', SCon{'0', SCon{'0', SCon{'1', SCon{'b', tail}}}}}}: fast_push(fast_string(tail), '\u{1b}') case SCon{'\\', SCon{'u', SCon{'0', SCon{'0', SCon{'1', SCon{'c', tail}}}}}}: fast_push(fast_string(tail), '\u{1c}') case SCon{'\\', SCon{'u', SCon{'0', SCon{'0', SCon{'1', SCon{'d', tail}}}}}}: fast_push(fast_string(tail), '\u{1d}') case SCon{'\\', SCon{'u', SCon{'0', SCon{'0', SCon{'1', SCon{'e', tail}}}}}}: fast_push(fast_string(tail), '\u{1e}') case SCon{'\\', SCon{'u', SCon{'0', SCon{'0', SCon{'1', SCon{'f', tail}}}}}}: fast_push(fast_string(tail), '\u{1f}') case SCon{'\\', SCon{'u', _}}: UnicodeFallback{} case SCon{'\\', _}: InvalidString{} case SCon{'"', _}: InvalidString{} case SNil{}: InvalidString{} case SCon{+c, tail}: fast_raw(is_control(c), fast_string(tail), c) def fast_result(result: FastString, original: String) -> Result<&1, &1, Error, Value>: match result: case Decoded{text}: Done{Str{text}} case InvalidString{}: Fail{Error{}} case UnicodeFallback{}: finish(parse_value(original)) def payload_other(text: String) -> Result<&1, &1, Error, Value>: match text: case "null": Done{Null{}} case "true": Done{Bool{True{}}} case "false": Done{Bool{False{}}} case SCon{'"', +tail}: fast_result(fast_string(tail), SCon{'"', tail}) case _: Fail{Error{}} def payload_number(+text: String, valid: BoolType(), evidence: {number_valid(text) == valid : BoolType()}) -> Result<&1, &1, Error, Value>: match valid: case True{}: Done{Number{text, Certified{number_proof(text, evidence)}}} case False{}: payload_other(text) def decode_payload(+text: String) -> Result<&1, &1, Error, Value>: payload_number(text, number_valid(text), {==}) # Lexical boundaries are found once. Token payloads are then parsed on a # balanced fork tree, so strings and numbers in one document can use different # GPU lanes. Delimiter assembly follows token order. type Lexeme is Data: Punctuation{character: Char} Payload{text: String} def flush_lexeme(reversed: String, acc: List<&2, Lexeme>) -> List<&2, Lexeme>: match reversed: case SNil{}: acc case SCon{h, t}: Payload{String.reverse(SCon{h, t})} <> acc type LexMode is Data: Outside{} Quoted{} Escaped{} type LexAction is Data: LexStop{} LexEscapedChar{} LexBackslash{} LexStringEnd{} LexStringChar{} LexQuote{} LexWhitespace{} LexPunctuation{character: Char} LexRaw{} def lex_outside_other(character: Char) -> LexAction: match character: case '"': LexQuote{} case ' ': LexWhitespace{} case '\t': LexWhitespace{} case '\r': LexWhitespace{} case '\n': LexWhitespace{} case '[': LexPunctuation{'['} case ']': LexPunctuation{']'} case '{': LexPunctuation{'{'} case '}': LexPunctuation{'}'} case ':': LexPunctuation{':' } case ',': LexPunctuation{','} case _: LexRaw{} def lex_outside_action(character: Char, number_kind: NumberChar) -> LexAction: match number_kind: case NOther{}: lex_outside_other(character) case _: LexRaw{} def lex_action(mode: LexMode, character: Char, number_kind: NumberChar) -> LexAction: match mode: case Escaped{}: LexEscapedChar{} case Quoted{}: quote_lex_action(quote_action(character)) case Outside{}: lex_outside_action(character, number_kind) def lex_seed_action(text: String, mode: LexMode) -> LexAction: match text: case SNil{}: LexStop{} case SCon{+character, _}: lex_action(mode, character, number_char(character)) def lex_end(+mode: LexMode, reversed: String, acc: List<&2, Lexeme>) -> List<&2, Lexeme>: match mode: case Outside{}: List.reverse(&2, Lexeme, flush_lexeme(reversed, acc)) case Quoted{}: List.reverse(&2, Lexeme, Payload{"\""} <> acc) case Escaped{}: List.reverse(&2, Lexeme, Payload{"\""} <> acc) def lexemes(text: String, +mode: LexMode, reversed: String, acc: List<&2, Lexeme>, action: LexAction) -> List<&2, Lexeme>: match text action: case SNil{} LexStop{}: lex_end(mode, reversed, acc) case SNil{} _: List.reverse(&2, Lexeme, flush_lexeme(reversed, acc)) case SCon{h, +t} LexStop{}: lexemes(t, mode, SCon{h, reversed}, acc, lex_seed_action(t, mode)) case SCon{h, +t} LexEscapedChar{}: lexemes(t, Quoted{}, SCon{h, reversed}, acc, lex_seed_action(t, Quoted{})) case SCon{_, +t} LexBackslash{}: lexemes(t, Escaped{}, SCon{'\\', reversed}, acc, lex_seed_action(t, Escaped{})) case SCon{_, +t} LexStringEnd{}: lexemes(t, Outside{}, SNil{}, Payload{String.reverse(SCon{'"', reversed})} <> acc, lex_seed_action(t, Outside{})) case SCon{h, +t} LexStringChar{}: lexemes(t, Quoted{}, SCon{h, reversed}, acc, lex_seed_action(t, Quoted{})) case SCon{_, +t} LexQuote{}: lexemes(t, Quoted{}, SCon{'"', SNil{}}, flush_lexeme(reversed, acc), lex_seed_action(t, Quoted{})) case SCon{_, +t} LexWhitespace{}: lexemes(t, Outside{}, SNil{}, flush_lexeme(reversed, acc), lex_seed_action(t, Outside{})) case SCon{_, +t} LexPunctuation{character}: lexemes(t, Outside{}, SNil{}, Punctuation{character} <> flush_lexeme(reversed, acc), lex_seed_action(t, Outside{})) case SCon{h, +t} LexRaw{}: lexemes(t, Outside{}, SCon{h, reversed}, acc, lex_seed_action(t, Outside{})) def lex_scan(+text: String, +mode: LexMode, reversed: String, acc: List<&2, Lexeme>) -> List<&2, Lexeme>: lexemes(text, mode, reversed, acc, lex_seed_action(text, mode)) # The reverse token stream that corresponds to render(value, mode, suffix). # It is used for the structural parser/renderer proof. def source_text(text: String, acc: List<&2, Lexeme>) -> List<&2, Lexeme>: Payload{text} <> acc def source(value: Value, members: BoolType(), acc: List<&2, Lexeme>) -> List<&2, Lexeme>: match value: case Null{}: source_text("null", acc) case Bool{truth}: match truth: case True{}: source_text("true", acc) case False{}: source_text("false", acc) case Number{lexeme, certificate}: source_text(lexeme, acc) case Str{text}: source_text("\"" ++ escape_string(text) ++ "\"", acc) case Arr{Nil{}}: match members: case False{}: Punctuation{'['} <> Punctuation{']'} <> acc case True{}: Punctuation{']'} <> acc case Arr{head <> tail}: match members: case False{}: Punctuation{'['} <> source(head, False{}, source(Arr{tail}, True{}, acc)) case True{}: Punctuation{','} <> source(head, False{}, source(Arr{tail}, True{}, acc)) case Obj{Nil{}}: match members: case False{}: Punctuation{'{'} <> Punctuation{'}'} <> acc case True{}: Punctuation{'}'} <> acc case Obj{(key, value) <> tail}: match members: case False{}: Punctuation{'{'} <> source_text("\"" ++ escape_string(key) ++ "\"", Punctuation{':'} <> source(value, False{}, source(Obj{tail}, True{}, acc))) case True{}: Punctuation{','} <> source_text("\"" ++ escape_string(key) ++ "\"", Punctuation{':'} <> source(value, False{}, source(Obj{tail}, True{}, acc))) type LexTree is Data: LexEmpty{} LexLeaf{lexeme: Lexeme} LexBranch{left: LexTree, right: LexTree} type Token is Type: Delimiter{character: Char} Atom{result: Result<&1, &1, Error, Value>} type TokenTree is Type: TokensEmpty{} TokenLeaf{token: Token} TokenBranch{left: TokenTree, right: TokenTree} def lex_leaves(items: List<&2, Lexeme>) -> List<&2, LexTree>: match items: case Nil{}: Nil{} case h <> t: LexLeaf{h} <> lex_leaves(t) def lex_pairs(trees: List<&2, LexTree>) -> List<&2, LexTree>: match trees: case Nil{}: Nil{} case h <> Nil{}: h <> Nil{} case a <> b <> t: LexBranch{a, b} <> lex_pairs(t) def lex_join(trees: List<&2, LexTree>) -> LexTree: match trees: case Nil{}: LexEmpty{} case head <> tail: LexBranch{head, lex_join(tail)} def lex_balance(fuel: Nat, trees: List<&2, LexTree>) -> LexTree: match fuel: case 0n: lex_join(trees) case 1n+p: match trees: case Nil{}: LexEmpty{} case h <> Nil{}: h case a <> b <> t: lex_balance(p, lex_pairs(a <> b <> t)) def lex_tree(+items: List<&2, Lexeme>) -> LexTree: lex_balance(1n+List.length(&2, Lexeme, items), lex_leaves(items)) def decode_lexeme(lexeme: Lexeme) -> Token: match lexeme: case Punctuation{c}: Delimiter{c} case Payload{text}: Atom{decode_payload(text)} def decode_tree(tree: LexTree) -> TokenTree: match tree: case LexEmpty{}: TokensEmpty{} case LexLeaf{lexeme}: TokenLeaf{decode_lexeme(lexeme)} case LexBranch{left, right}: l r = decode_tree(left) decode_tree(right) TokenBranch{l, r} def token_list(tree: TokenTree, tail: List<Token>) -> List<Token>: match tree: case TokensEmpty{}: tail case TokenLeaf{token}: token <> tail case TokenBranch{left, right}: token_list(left, token_list(right, tail)) type Assembly is Type: NeedValue{kont: ParserKont} ArrayFirst{kont: ParserKont} ArrayNext{acc: List<&2, Value>, kont: ParserKont} ArrayComma{acc: List<&2, Value>, kont: ParserKont} ObjectFirst{kont: ParserKont} ObjectNext{acc: List<&2, Sigma<&2, &2, String, key => Value>>, kont: ParserKont} NeedColon{key: String, acc: List<&2, Sigma<&2, &2, String, key => Value>>, kont: ParserKont} ObjectComma{acc: List<&2, Sigma<&2, &2, String, key => Value>>, kont: ParserKont} Assembled{value: Value} AssemblyError{} def assembly_resume(kont: ParserKont, value: Value) -> Assembly: match kont: case Top{}: Assembled{value} case InArray{acc, parent}: ArrayComma{value <> acc, parent} case InObject{key, acc, parent}: ObjectComma{(key, value) <> acc, parent} def assembly_atom(result: Result<&1, &1, Error, Value>, kont: ParserKont) -> Assembly: match result: case Fail{_}: AssemblyError{} case Done{value}: assembly_resume(kont, value) def assembly_value(token: Token, kont: ParserKont) -> Assembly: match token: case Atom{result}: assembly_atom(result, kont) case Delimiter{'['}: ArrayFirst{kont} case Delimiter{'{'}: ObjectFirst{kont} case _: AssemblyError{} def assembly_key_value(result: Result<&1, &1, Error, Value>, acc: List<&2, Sigma<&2, &2, String, key => Value>>, kont: ParserKont) -> Assembly: match result: case Done{Str{key}}: NeedColon{key, acc, kont} case _: AssemblyError{} def assembly_key(token: Token, acc: List<&2, Sigma<&2, &2, String, key => Value>>, kont: ParserKont) -> Assembly: match token: case Atom{result}: assembly_key_value(result, acc, kont) case _: AssemblyError{} def assembly_step(state: Assembly, token: Token) -> Assembly: match state: case NeedValue{kont}: assembly_value(token, kont) case ArrayFirst{kont}: match token: case Delimiter{']'}: assembly_resume(kont, Arr{Nil{}}) case _: assembly_value(token, InArray{Nil{}, kont}) case ArrayNext{acc, kont}: assembly_value(token, InArray{acc, kont}) case ArrayComma{acc, kont}: match token: case Delimiter{','}: ArrayNext{acc, kont} case Delimiter{']'}: assembly_resume(kont, Arr{List.reverse(&2, Value, acc)}) case _: AssemblyError{} case ObjectFirst{kont}: match token: case Delimiter{'}'}: assembly_resume(kont, Obj{Nil{}}) case _: assembly_key(token, Nil{}, kont) case ObjectNext{acc, kont}: assembly_key(token, acc, kont) case NeedColon{key, acc, kont}: match token: case Delimiter{':'}: NeedValue{InObject{key, acc, kont}} case _: AssemblyError{} case ObjectComma{acc, kont}: match token: case Delimiter{','}: ObjectNext{acc, kont} case Delimiter{'}'}: assembly_resume(kont, Obj{List.reverse(&2, Sigma<&2, &2, String, key => Value>, acc)}) case _: AssemblyError{} case _: AssemblyError{} def assembly_finish(state: Assembly) -> Result<&1, &1, Error, Value>: match state: case Assembled{value}: Done{value} case _: Fail{Error{}} def assemble(tokens: List<Token>, state: Assembly) -> Result<&1, &1, Error, Value>: match tokens: case Nil{}: assembly_finish(state) case token <> tail: assemble(tail, assembly_step(state, token)) def parse_normalized(text: String) -> Result<&1, &1, Error, Value>: assemble(token_list(decode_tree(lex_tree(lex_scan(text, Outside{}, SNil{}, Nil{}))), Nil{}), NeedValue{Top{}}) # Call as parse!(text) to place the complete parse on Bend's GPU scheduler # when one is available. The ordinary call remains useful on small inputs. def parse(+text: String) -> Result<&1, &1, Error, Value>: parse_normalized(text) # Smart constructor. Validation evidence is checked by Bend and is not an # unchecked assertion supplied by callers. def number(+text: String) -> Result<&1, &1, Error, Value>: finish(number_lexeme(text, SNil{}, number_valid(text), {==})) # Independent documents form a fork/join tree. Keep sibling subtrees similar # in total input size to distribute work across GPU lanes (or CPU workers). type Batch is Data: Document{text: String} Documents{left: Batch, right: Batch} type BatchResult is Type: DocumentResult{result: Result<&1, &1, Error, Value>} DocumentResults{left: BatchResult, right: BatchResult} def parse_batch(batch: Batch) -> BatchResult: match batch: case Document{text}: DocumentResult{parse(text)} case Documents{left, right}: l r = parse_batch(left) parse_batch(right) DocumentResults{l, r} # The ! entry points request GPU execution; Bend provides the CPU fallback. # A single document has sequential grammar dependencies. For parallel work, # use parse_batch_gpu with independent documents. def parse_gpu(text: String) -> Result<&1, &1, Error, Value>: parse!(text) def parse_batch_gpu(batch: Batch) -> BatchResult: parse_batch!(batch) # Build a balanced fork tree from documents in source order. Pairing adjacent # trees on each pass keeps the depth logarithmic without copying input text. # For best lane balance, supply documents of similar lengths/complexity. def batch_leaves(texts: List<&2, String>) -> List<&2, Batch>: match texts: case Nil{}: Nil{} case text <> tail: Document{text} <> batch_leaves(tail) def batch_pairs(trees: List<&2, Batch>) -> List<&2, Batch>: match trees: case Nil{}: Nil{} case tree <> Nil{}: tree <> Nil{} case left <> right <> tail: Documents{left, right} <> batch_pairs(tail) def batch_build(fuel: Nat, trees: List<&2, Batch>) -> Maybe<&2, Batch>: match fuel: case 0n: List.head(&2, Batch, trees) case 1n+p: match trees: case Nil{}: None{} case tree <> Nil{}: Some{tree} case left <> right <> tail: batch_build(p, batch_pairs(left <> right <> tail)) def batch_from_list(+texts: List<&2, String>) -> Maybe<&2, Batch>: batch_build(List.length(&2, String, texts), batch_leaves(texts)) def parse_optional_batch_gpu(batch: Maybe<&2, Batch>) -> Maybe<&1, BatchResult>: match batch: case None{}: None{} case Some{tree}: Some{parse_batch!(tree)} # Build the fork tree before entering the device. None represents an empty # input list; each document's parse failure remains at its own result leaf. def parse_many_gpu(texts: List<&2, String>) -> Maybe<&1, BatchResult>: parse_optional_batch_gpu(batch_from_list(texts))