diff options
Diffstat (limited to 'public/prism/prism-scheme.js')
| -rw-r--r-- | public/prism/prism-scheme.js | 132 |
1 files changed, 132 insertions, 0 deletions
diff --git a/public/prism/prism-scheme.js b/public/prism/prism-scheme.js new file mode 100644 index 0000000..4af5d04 --- /dev/null +++ b/public/prism/prism-scheme.js @@ -0,0 +1,132 @@ +(function (Prism) { + Prism.languages.scheme = { + // this supports "normal" single-line comments: + // ; comment + // and (potentially nested) multiline comments: + // #| comment #| nested |# still comment |# + // (only 1 level of nesting is supported) + comment: + /;.*|#;\s*(?:\((?:[^()]|\([^()]*\))*\)|\[(?:[^\[\]]|\[[^\[\]]*\])*\])|#\|(?:[^#|]|#(?!\|)|\|(?!#)|#\|(?:[^#|]|#(?!\|)|\|(?!#))*\|#)*\|#/, + string: { + pattern: /"(?:[^"\\]|\\.)*"/, + greedy: true, + }, + symbol: { + pattern: /'[^()\[\]#'\s]+/, + greedy: true, + }, + char: { + pattern: + /#\\(?:[ux][a-fA-F\d]+\b|[-a-zA-Z]+\b|[\uD800-\uDBFF][\uDC00-\uDFFF]|\S)/, + greedy: true, + }, + 'lambda-parameter': [ + // https://www.cs.cmu.edu/Groups/AI/html/r4rs/r4rs_6.html#SEC30 + { + pattern: + /((?:^|[^'`#])[(\[]lambda\s+)(?:[^|()\[\]'\s]+|\|(?:[^\\|]|\\.)*\|)/, + lookbehind: true, + }, + { + pattern: /((?:^|[^'`#])[(\[]lambda\s+[(\[])[^()\[\]']+/, + lookbehind: true, + }, + ], + keyword: { + pattern: + /((?:^|[^'`#])[(\[])(?:begin|case(?:-lambda)?|cond(?:-expand)?|define(?:-library|-macro|-record-type|-syntax|-values)?|defmacro|delay(?:-force)?|do|else|except|export|guard|if|import|include(?:-ci|-library-declarations)?|lambda|let(?:rec)?(?:-syntax|-values|\*)?|let\*-values|only|parameterize|prefix|(?:quasi-?)?quote|rename|set!|syntax-(?:case|rules)|unless|unquote(?:-splicing)?|when)(?=[()\[\]\s]|$)/, + lookbehind: true, + }, + builtin: { + // all functions of the base library of R7RS plus some of built-ins of R5Rs + pattern: + /((?:^|[^'`#])[(\[])(?:abs|and|append|apply|assoc|ass[qv]|binary-port\?|boolean=?\?|bytevector(?:-append|-copy|-copy!|-length|-u8-ref|-u8-set!|\?)?|caar|cadr|call-with-(?:current-continuation|port|values)|call\/cc|car|cdar|cddr|cdr|ceiling|char(?:->integer|-ready\?|\?|<\?|<=\?|=\?|>\?|>=\?)|close-(?:input-port|output-port|port)|complex\?|cons|current-(?:error|input|output)-port|denominator|dynamic-wind|eof-object\??|eq\?|equal\?|eqv\?|error|error-object(?:-irritants|-message|\?)|eval|even\?|exact(?:-integer-sqrt|-integer\?|\?)?|expt|features|file-error\?|floor(?:-quotient|-remainder|\/)?|flush-output-port|for-each|gcd|get-output-(?:bytevector|string)|inexact\??|input-port(?:-open\?|\?)|integer(?:->char|\?)|lcm|length|list(?:->string|->vector|-copy|-ref|-set!|-tail|\?)?|make-(?:bytevector|list|parameter|string|vector)|map|max|member|memq|memv|min|modulo|negative\?|newline|not|null\?|number(?:->string|\?)|numerator|odd\?|open-(?:input|output)-(?:bytevector|string)|or|output-port(?:-open\?|\?)|pair\?|peek-char|peek-u8|port\?|positive\?|procedure\?|quotient|raise|raise-continuable|rational\?|rationalize|read-(?:bytevector|bytevector!|char|error\?|line|string|u8)|real\?|remainder|reverse|round|set-c[ad]r!|square|string(?:->list|->number|->symbol|->utf8|->vector|-append|-copy|-copy!|-fill!|-for-each|-length|-map|-ref|-set!|\?|<\?|<=\?|=\?|>\?|>=\?)?|substring|symbol(?:->string|\?|=\?)|syntax-error|textual-port\?|truncate(?:-quotient|-remainder|\/)?|u8-ready\?|utf8->string|values|vector(?:->list|->string|-append|-copy|-copy!|-fill!|-for-each|-length|-map|-ref|-set!|\?)?|with-exception-handler|write-(?:bytevector|char|string|u8)|zero\?)(?=[()\[\]\s]|$)/, + lookbehind: true, + }, + operator: { + pattern: /((?:^|[^'`#])[(\[])(?:[-+*%/]|[<>]=?|=>?)(?=[()\[\]\s]|$)/, + lookbehind: true, + }, + number: { + // The number pattern from [the R7RS spec](https://small.r7rs.org/attachment/r7rs.pdf). + // + // <number> := <num 2>|<num 8>|<num 10>|<num 16> + // <num R> := <prefix R><complex R> + // <complex R> := <real R>(?:@<real R>|<imaginary R>)?|<imaginary R> + // <imaginary R> := [+-](?:<ureal R>|(?:inf|nan)\.0)?i + // <real R> := [+-]?<ureal R>|[+-](?:inf|nan)\.0 + // <ureal R> := <uint R>(?:\/<uint R>)? + // | <decimal R> + // + // <decimal 10> := (?:\d+(?:\.\d*)?|\.\d+)(?:e[+-]?\d+)? + // <uint R> := <digit R>+ + // <prefix R> := <radix R>(?:#[ei])?|(?:#[ei])?<radix R> + // <radix 2> := #b + // <radix 8> := #o + // <radix 10> := (?:#d)? + // <radix 16> := #x + // <digit 2> := [01] + // <digit 8> := [0-7] + // <digit 10> := \d + // <digit 16> := [0-9a-f] + // + // The problem with this grammar is that the resulting regex is way to complex, so we simplify by grouping all + // non-decimal bases together. This results in a decimal (dec) and combined binary, octal, and hexadecimal (box) + // pattern: + pattern: RegExp( + SortedBNF({ + '<ureal dec>': + /\d+(?:\/\d+)|(?:\d+(?:\.\d*)?|\.\d+)(?:[esfdl][+-]?\d+)?/.source, + '<real dec>': /[+-]?<ureal dec>|[+-](?:inf|nan)\.0/.source, + '<imaginary dec>': /[+-](?:<ureal dec>|(?:inf|nan)\.0)?i/.source, + '<complex dec>': + /<real dec>(?:@<real dec>|<imaginary dec>)?|<imaginary dec>/.source, + '<num dec>': /(?:#d(?:#[ei])?|#[ei](?:#d)?)?<complex dec>/.source, + + '<ureal box>': /[0-9a-f]+(?:\/[0-9a-f]+)?/.source, + '<real box>': /[+-]?<ureal box>|[+-](?:inf|nan)\.0/.source, + '<imaginary box>': /[+-](?:<ureal box>|(?:inf|nan)\.0)?i/.source, + '<complex box>': + /<real box>(?:@<real box>|<imaginary box>)?|<imaginary box>/.source, + '<num box>': /#[box](?:#[ei])?|(?:#[ei])?#[box]<complex box>/.source, + + '<number>': /(^|[()\[\]\s])(?:<num dec>|<num box>)(?=[()\[\]\s]|$)/ + .source, + }), + 'i' + ), + lookbehind: true, + }, + boolean: { + pattern: /(^|[()\[\]\s])#(?:[ft]|false|true)(?=[()\[\]\s]|$)/, + lookbehind: true, + }, + function: { + pattern: + /((?:^|[^'`#])[(\[])(?:[^|()\[\]'\s]+|\|(?:[^\\|]|\\.)*\|)(?=[()\[\]\s]|$)/, + lookbehind: true, + }, + identifier: { + pattern: /(^|[()\[\]\s])\|(?:[^\\|]|\\.)*\|(?=[()\[\]\s]|$)/, + lookbehind: true, + greedy: true, + }, + punctuation: /[()\[\]']/, + }; + + /** + * Given a topologically sorted BNF grammar, this will return the RegExp source of last rule of the grammar. + * + * @param {Record<string, string>} grammar + * @returns {string} + */ + function SortedBNF(grammar) { + for (var key in grammar) { + grammar[key] = grammar[key].replace(/<[\w\s]+>/g, function (key) { + return '(?:' + grammar[key].trim() + ')'; + }); + } + // return the last item + return grammar[key]; + } +})(Prism); |
