// Ported from Go: // https://github.com/golang/go/blob/go1.12.5/src/encoding/csv/ // Copyright 2018-2019 the Deno authors. All rights reserved. MIT license. import { BufReader } from "../io/bufio.ts"; import { TextProtoReader } from "../textproto/mod.ts"; import { StringReader } from "../io/readers.ts"; const INVALID_RUNE = ["\r", "\n", '"']; export class ParseError extends Error { StartLine: number; Line: number; constructor(start: number, line: number, message: string) { super(message); this.StartLine = start; this.Line = line; } } /** * @property comma - Character which separates values. Default: ',' * @property comment - Character to start a comment. Default: '#' * @property trimLeadingSpace - Flag to trim the leading space of the value. * Default: 'false' * @property lazyQuotes - Allow unquoted quote in a quoted field or non double * quoted quotes in quoted field Default: 'false' * @property fieldsPerRecord - Enabling the check of fields for each row. * If == 0, first row is used as referal for the number of fields. */ export interface ParseOptions { comma?: string; comment?: string; trimLeadingSpace?: boolean; lazyQuotes?: boolean; fieldsPerRecord?: number; } function chkOptions(opt: ParseOptions): void { if (!opt.comma) opt.comma = ","; if (!opt.trimLeadingSpace) opt.trimLeadingSpace = false; if ( INVALID_RUNE.includes(opt.comma!) || INVALID_RUNE.includes(opt.comment!) || opt.comma === opt.comment ) { throw new Error("Invalid Delimiter"); } } async function read( Startline: number, reader: BufReader, opt: ParseOptions = { comma: ",", trimLeadingSpace: false } ): Promise { const tp = new TextProtoReader(reader); let line: string; let result: string[] = []; let lineIndex = Startline; const r = await tp.readLine(); if (r === Deno.EOF) return Deno.EOF; line = r; // Normalize \r\n to \n on all input lines. if ( line.length >= 2 && line[line.length - 2] === "\r" && line[line.length - 1] === "\n" ) { line = line.substring(0, line.length - 2); line = line + "\n"; } const trimmedLine = line.trimLeft(); if (trimmedLine.length === 0) { return []; } // line starting with comment character is ignored if (opt.comment && trimmedLine[0] === opt.comment) { return []; } result = line.split(opt.comma!); let quoteError = false; result = result.map((r): string => { if (opt.trimLeadingSpace) { r = r.trimLeft(); } if (r[0] === '"' && r[r.length - 1] === '"') { r = r.substring(1, r.length - 1); } else if (r[0] === '"') { r = r.substring(1, r.length); } if (!opt.lazyQuotes) { if (r[0] !== '"' && r.indexOf('"') !== -1) { quoteError = true; } } return r; }); if (quoteError) { throw new ParseError(Startline, lineIndex, 'bare " in non-quoted-field'); } return result; } export async function readAll( reader: BufReader, opt: ParseOptions = { comma: ",", trimLeadingSpace: false, lazyQuotes: false } ): Promise { const result: string[][] = []; let _nbFields: number; let lineResult: string[]; let first = true; let lineIndex = 0; chkOptions(opt); for (;;) { const r = await read(lineIndex, reader, opt); if (r === Deno.EOF) break; lineResult = r; lineIndex++; // If fieldsPerRecord is 0, Read sets it to // the number of fields in the first record if (first) { first = false; if (opt.fieldsPerRecord !== undefined) { if (opt.fieldsPerRecord === 0) { _nbFields = lineResult.length; } else { _nbFields = opt.fieldsPerRecord; } } } if (lineResult.length > 0) { if (_nbFields! && _nbFields! !== lineResult.length) { throw new ParseError(lineIndex, lineIndex, "wrong number of fields"); } result.push(lineResult); } } return result; } /** * HeaderOption provides the column definition * and the parse function for each entry of the * column. */ export interface HeaderOption { name: string; parse?: (input: string) => unknown; } export interface ExtendedParseOptions extends ParseOptions { header: boolean | string[] | HeaderOption[]; parse?: (input: unknown) => unknown; } /** * Csv parse helper to manipulate data. * Provides an auto/custom mapper for columns and parse function * for columns and rows. * @param input Input to parse. Can be a string or BufReader. * @param opt options of the parser. * @param [opt.header=false] HeaderOptions * @param [opt.parse=null] Parse function for rows. * Example: * const r = await parseFile('a,b,c\ne,f,g\n', { * header: ["this", "is", "sparta"], * parse: (e: Record) => { * return { super: e.this, street: e.is, fighter: e.sparta }; * } * }); * // output * [ * { super: "a", street: "b", fighter: "c" }, * { super: "e", street: "f", fighter: "g" } * ] */ export async function parse( input: string | BufReader, opt: ExtendedParseOptions = { header: false } ): Promise { let r: string[][]; if (input instanceof BufReader) { r = await readAll(input, opt); } else { r = await readAll(new BufReader(new StringReader(input)), opt); } if (opt.header) { let headers: HeaderOption[] = []; let i = 0; if (Array.isArray(opt.header)) { if (typeof opt.header[0] !== "string") { headers = opt.header as HeaderOption[]; } else { const h = opt.header as string[]; headers = h.map( (e): HeaderOption => { return { name: e }; } ); } } else { headers = r.shift()!.map( (e): HeaderOption => { return { name: e }; } ); i++; } return r.map((e): unknown => { if (e.length !== headers.length) { throw `Error number of fields line:${i}`; } i++; let out: Record = {}; for (let j = 0; j < e.length; j++) { const h = headers[j]; if (h.parse) { out[h.name] = h.parse(e[j]); } else { out[h.name] = e[j]; } } if (opt.parse) { return opt.parse(out); } return out; }); } if (opt.parse) { return r.map((e: string[]): unknown => opt.parse!(e)); } return r; }