if (typeof module !== 'undefined' && module.exports) { var chai = require('chai'); var Papa = require('../papaparse.js'); } var assert = chai.assert; var RECORD_SEP = String.fromCharCode(30); var UNIT_SEP = String.fromCharCode(31); var FILES_ENABLED = false; try { new File([""], ""); FILES_ENABLED = true; } catch (e) {} // safari, ie var XHR_ENABLED = false; try { new XMLHttpRequest(); XHR_ENABLED = true; } catch (e) {} // safari, ie // Tests for the core parser using new Papa.Parser().parse() (CSV to JSON) var CORE_PARSER_TESTS = [ { description: "One row", input: 'A,b,c', expected: { data: [['A', 'b', 'c']], errors: [] } }, { description: "Two rows", input: 'A,b,c\nd,E,f', expected: { data: [['A', 'b', 'c'], ['d', 'E', 'f']], errors: [] } }, { description: "Three rows", input: 'A,b,c\nd,E,f\nG,h,i', expected: { data: [['A', 'b', 'c'], ['d', 'E', 'f'], ['G', 'h', 'i']], errors: [] } }, { description: "Whitespace at edges of unquoted field", input: 'a, b ,c', notes: "Extra whitespace should graciously be preserved", expected: { data: [['a', ' b ', 'c']], errors: [] } }, { description: "Quoted field", input: 'A,"B",C', expected: { data: [['A', 'B', 'C']], errors: [] } }, { description: "Quoted field with extra whitespace on edges", input: 'A," B ",C', expected: { data: [['A', ' B ', 'C']], errors: [] } }, { description: "Quoted field with delimiter", input: 'A,"B,B",C', expected: { data: [['A', 'B,B', 'C']], errors: [] } }, { description: "Quoted field with line break", input: 'A,"B\nB",C', expected: { data: [['A', 'B\nB', 'C']], errors: [] } }, { description: "Quoted fields with line breaks", input: 'A,"B\nB","C\nC\nC"', expected: { data: [['A', 'B\nB', 'C\nC\nC']], errors: [] } }, { description: "Quoted fields at end of row with delimiter and line break", input: 'a,b,"c,c\nc"\nd,e,f', expected: { data: [['a', 'b', 'c,c\nc'], ['d', 'e', 'f']], errors: [] } }, { description: "Quoted field with escaped quotes", input: 'A,"B""B""B",C', expected: { data: [['A', 'B"B"B', 'C']], errors: [] } }, { description: "Quoted field with escaped quotes at boundaries", input: 'A,"""B""",C', expected: { data: [['A', '"B"', 'C']], errors: [] } }, { description: "Unquoted field with quotes at end of field", notes: "The quotes character is misplaced, but shouldn't generate an error or break the parser", input: 'A,B",C', expected: { data: [['A', 'B"', 'C']], errors: [] } }, { description: "Quoted field with quotes around delimiter", input: 'A,""",""",C', notes: "For a boundary to exist immediately before the quotes, we must not already be in quotes", expected: { data: [['A', '","', 'C']], errors: [] } }, { description: "Quoted field with quotes on right side of delimiter", input: 'A,",""",C', notes: "Similar to the test above but with quotes only after the comma", expected: { data: [['A', ',"', 'C']], errors: [] } }, { description: "Quoted field with quotes on left side of delimiter", input: 'A,""",",C', notes: "Similar to the test above but with quotes only before the comma", expected: { data: [['A', '",', 'C']], errors: [] } }, { description: "Quoted field with 5 quotes in a row and a delimiter in there, too", input: '"1","cnonce="""",nc=""""","2"', notes: "Actual input reported in issue #121", expected: { data: [['1', 'cnonce="",nc=""', '2']], errors: [] } }, { description: "Quoted field with whitespace around quotes", input: 'A, "B" ,C', notes: "The quotes must be immediately adjacent to the delimiter to indicate a quoted field", expected: { data: [['A', ' "B" ', 'C']], errors: [] } }, { description: "Misplaced quotes in data, not as opening quotes", input: 'A,B "B",C', notes: "The input is technically malformed, but this syntax should not cause an error", expected: { data: [['A', 'B "B"', 'C']], errors: [] } }, { description: "Quoted field has no closing quote", input: 'a,"b,c\nd,e,f', expected: { data: [['a', 'b,c\nd,e,f']], errors: [{ "type": "Quotes", "code": "MissingQuotes", "message": "Quoted field unterminated", "row": 0, "index": 3 }] } }, { description: "Line starts with quoted field", input: 'a,b,c\n"d",e,f', expected: { data: [['a', 'b', 'c'], ['d', 'e', 'f']], errors: [] } }, { description: "Line ends with quoted field", input: 'a,b,c\nd,e,f\n"g","h","i"\n"j","k","l"', expected: { data: [['a', 'b', 'c'], ['d', 'e', 'f'], ['g', 'h', 'i'], ['j', 'k', 'l']], errors: [] } }, { description: "Quoted field at end of row (but not at EOF) has quotes", input: 'a,b,"c""c"""\nd,e,f', expected: { data: [['a', 'b', 'c"c"'], ['d', 'e', 'f']], errors: [] } }, { description: "Empty quoted field at EOF is empty", input: 'a,b,""\na,b,""', expected: { data: [['a', 'b', ''], ['a', 'b', '']], errors: [] } }, { description: "Multiple consecutive empty fields", input: 'a,b,,,c,d\n,,e,,,f', expected: { data: [['a', 'b', '', '', 'c', 'd'], ['', '', 'e', '', '', 'f']], errors: [] } }, { description: "Empty input string", input: '', expected: { data: [], errors: [] } }, { description: "Input is just the delimiter (2 empty fields)", input: ',', expected: { data: [['', '']], errors: [] } }, { description: "Input is just empty fields", input: ',,\n,,,', expected: { data: [['', '', ''], ['', '', '', '']], errors: [] } }, { description: "Input is just a string (a single field)", input: 'Abc def', expected: { data: [['Abc def']], errors: [] } }, { description: "Commented line at beginning", input: '# Comment!\na,b,c', config: { comments: true }, expected: { data: [['a', 'b', 'c']], errors: [] } }, { description: "Commented line in middle", input: 'a,b,c\n# Comment\nd,e,f', config: { comments: true }, expected: { data: [['a', 'b', 'c'], ['d', 'e', 'f']], errors: [] } }, { description: "Commented line at end", input: 'a,true,false\n# Comment', config: { comments: true }, expected: { data: [['a', 'true', 'false']], errors: [] } }, { description: "Two comment lines consecutively", input: 'a,b,c\n#comment1\n#comment2\nd,e,f', config: { comments: true }, expected: { data: [['a', 'b', 'c'], ['d', 'e', 'f']], errors: [] } }, { description: "Two comment lines consecutively at end of file", input: 'a,b,c\n#comment1\n#comment2', config: { comments: true }, expected: { data: [['a', 'b', 'c']], errors: [] } }, { description: "Three comment lines consecutively at beginning of file", input: '#comment1\n#comment2\n#comment3\na,b,c', config: { comments: true }, expected: { data: [['a', 'b', 'c']], errors: [] } }, { description: "Entire file is comment lines", input: '#comment1\n#comment2\n#comment3', config: { comments: true }, expected: { data: [], errors: [] } }, { description: "Comment with non-default character", input: 'a,b,c\n!Comment goes here\nd,e,f', config: { comments: '!' }, expected: { data: [['a', 'b', 'c'], ['d', 'e', 'f']], errors: [] } }, { description: "Bad comments value specified", notes: "Should silently disable comment parsing", input: 'a,b,c\n5comment\nd,e,f', config: { comments: 5 }, expected: { data: [['a', 'b', 'c'], ['5comment'], ['d', 'e', 'f']], errors: [] } }, { description: "Multi-character comment string", input: 'a,b,c\n=N(Comment)\nd,e,f', config: { comments: "=N(" }, expected: { data: [['a', 'b', 'c'], ['d', 'e', 'f']], errors: [] } }, { description: "Input with only a commented line", input: '#commented line', config: { comments: true, delimiter: ',' }, expected: { data: [], errors: [] } }, { description: "Input with only a commented line and blank line after", input: '#commented line\n', config: { comments: true, delimiter: ',' }, expected: { data: [['']], errors: [] } }, { description: "Input with only a commented line, without comments enabled", input: '#commented line', config: { delimiter: ',' }, expected: { data: [['#commented line']], errors: [] } }, { description: "Input without comments with line starting with whitespace", input: 'a\n b\nc', config: { delimiter: ',' }, notes: "\" \" == false, but \" \" !== false, so === comparison is required", expected: { data: [['a'], [' b'], ['c']], errors: [] } }, { description: "Multiple rows, one column (no delimiter found)", input: 'a\nb\nc\nd\ne', expected: { data: [['a'], ['b'], ['c'], ['d'], ['e']], errors: [] } }, { description: "One column input with empty fields", input: 'a\nb\n\n\nc\nd\ne\n', expected: { data: [['a'], ['b'], [''], [''], ['c'], ['d'], ['e'], ['']], errors: [] } }, { description: "Fast mode, basic", input: 'a,b,c\nd,e,f', config: { fastMode: true }, expected: { data: [['a', 'b', 'c'], ['d', 'e', 'f']], errors: [] } }, { description: "Fast mode with comments", input: '// Commented line\na,b,c', config: { fastMode: true, comments: "//" }, expected: { data: [['a', 'b', 'c']], errors: [] } }, { description: "Fast mode with preview", input: 'a,b,c\nd,e,f\nh,j,i\n', config: { fastMode: true, preview: 2 }, expected: { data: [['a', 'b', 'c'], ['d', 'e', 'f']], errors: [] } }, { description: "Fast mode with blank line at end", input: 'a,b,c\n', config: { fastMode: true }, expected: { data: [['a', 'b', 'c'], ['']], errors: [] } } ]; describe('Core Parser Tests', function() { function generateTest(test) { (test.disabled ? it.skip : it)(test.description, function() { var actual = new Papa.Parser(test.config).parse(test.input); assert.deepEqual(JSON.stringify(actual.errors), JSON.stringify(test.expected.errors)); assert.deepEqual(actual.data, test.expected.data); }); } for (var i = 0; i < CORE_PARSER_TESTS.length; i++) { generateTest(CORE_PARSER_TESTS[i]); } }); // Tests for Papa.parse() function -- high-level wrapped parser (CSV to JSON) var PARSE_TESTS = [ { description: "Two rows, just \\r", input: 'A,b,c\rd,E,f', expected: { data: [['A', 'b', 'c'], ['d', 'E', 'f']], errors: [] } }, { description: "Two rows, \\r\\n", input: 'A,b,c\r\nd,E,f', expected: { data: [['A', 'b', 'c'], ['d', 'E', 'f']], errors: [] } }, { description: "Quoted field with \\r\\n", input: 'A,"B\r\nB",C', expected: { data: [['A', 'B\r\nB', 'C']], errors: [] } }, { description: "Quoted field with \\r", input: 'A,"B\rB",C', expected: { data: [['A', 'B\rB', 'C']], errors: [] } }, { description: "Quoted field with \\n", input: 'A,"B\nB",C', expected: { data: [['A', 'B\nB', 'C']], errors: [] } }, { description: "Mixed slash n and slash r should choose first as precident", input: 'a,b,c\nd,e,f\rg,h,i\n', expected: { data: [['a', 'b', 'c'], ['d', 'e', 'f\rg', 'h', 'i'], ['']], errors: [] } }, { description: "Header row with one row of data", input: 'A,B,C\r\na,b,c', config: { header: true }, expected: { data: [{"A": "a", "B": "b", "C": "c"}], errors: [] } }, { description: "Header row only", input: 'A,B,C', config: { header: true }, expected: { data: [], errors: [] } }, { description: "Row with too few fields", input: 'A,B,C\r\na,b', config: { header: true }, expected: { data: [{"A": "a", "B": "b"}], errors: [{ "type": "FieldMismatch", "code": "TooFewFields", "message": "Too few fields: expected 3 fields but parsed 2", "row": 0 }] } }, { description: "Row with too many fields", input: 'A,B,C\r\na,b,c,d,e\r\nf,g,h', config: { header: true }, expected: { data: [{"A": "a", "B": "b", "C": "c", "__parsed_extra": ["d", "e"]}, {"A": "f", "B": "g", "C": "h"}], errors: [{ "type": "FieldMismatch", "code": "TooManyFields", "message": "Too many fields: expected 3 fields but parsed 5", "row": 0 }] } }, { description: "Row with enough fields but blank field at end", input: 'A,B,C\r\na,b,', config: { header: true }, expected: { data: [{"A": "a", "B": "b", "C": ""}], errors: [] } }, { description: "Tab delimiter", input: 'a\tb\tc\r\nd\te\tf', config: { delimiter: "\t" }, expected: { data: [['a', 'b', 'c'], ['d', 'e', 'f']], errors: [] } }, { description: "Pipe delimiter", input: 'a|b|c\r\nd|e|f', config: { delimiter: "|" }, expected: { data: [['a', 'b', 'c'], ['d', 'e', 'f']], errors: [] } }, { description: "ASCII 30 delimiter", input: 'a'+RECORD_SEP+'b'+RECORD_SEP+'c\r\nd'+RECORD_SEP+'e'+RECORD_SEP+'f', config: { delimiter: RECORD_SEP }, expected: { data: [['a', 'b', 'c'], ['d', 'e', 'f']], errors: [] } }, { description: "ASCII 31 delimiter", input: 'a'+UNIT_SEP+'b'+UNIT_SEP+'c\r\nd'+UNIT_SEP+'e'+UNIT_SEP+'f', config: { delimiter: UNIT_SEP }, expected: { data: [['a', 'b', 'c'], ['d', 'e', 'f']], errors: [] } }, { description: "Bad delimiter (\\n)", input: 'a,b,c', config: { delimiter: "\n" }, notes: "Should silently default to comma", expected: { data: [['a', 'b', 'c']], errors: [] } }, { description: "Multi-character delimiter", input: 'a, b, c', config: { delimiter: ", " }, expected: { data: [['a', 'b', 'c']], errors: [] } }, { description: "Callback delimiter", input: 'a$ b$ c', config: { delimiter: function(input) { return input[1] + ' '; } }, expected: { data: [['a', 'b', 'c']], errors: [] } }, { description: "Dynamic typing converts numeric literals", input: '1,2.2,1e3\r\n-4,-4.5,-4e-5\r\n-,5a,5-2', config: { dynamicTyping: true }, expected: { data: [[1, 2.2, 1000], [-4, -4.5, -0.00004], ["-", "5a", "5-2"]], errors: [] } }, { description: "Dynamic typing converts boolean literals", input: 'true,false,T,F,TRUE,FALSE,True,False', config: { dynamicTyping: true }, expected: { data: [[true, false, "T", "F", true, false, "True", "False"]], errors: [] } }, { description: "Dynamic typing doesn't convert other types", input: 'A,B,C\r\nundefined,null,[\r\nvar,float,if', config: { dynamicTyping: true }, expected: { data: [["A", "B", "C"], ["undefined", "null", "["], ["var", "float", "if"]], errors: [] } }, { description: "Dynamic typing applies to specific columns", input: 'A,B,C\r\n1,2.2,1e3\r\n-4,-4.5,-4e-5', config: { header: true, dynamicTyping: { A: true, C: true } }, expected: { data: [{"A": 1, "B": "2.2", "C": 1000}, {"A": -4, "B": "-4.5", "C": -0.00004}], errors: [] } }, { description: "Dynamic typing applies to specific columns by index", input: '1,2.2,1e3\r\n-4,-4.5,-4e-5\r\n-,5a,5-2', config: { dynamicTyping: { 1: true } }, expected: { data: [["1", 2.2, "1e3"], ["-4", -4.5, "-4e-5"], ["-", "5a", "5-2"]], errors: [] } }, { description: "Dynamic typing can be applied to `__parsed_extra`", input: 'A,B,C\r\n1,2.2,1e3,5.5\r\n-4,-4.5,-4e-5', config: { header: true, dynamicTyping: { A: true, C: true, __parsed_extra: true } }, expected: { data: [{"A": 1, "B": "2.2", "C": 1000, "__parsed_extra": [ 5.5 ]}, {"A": -4, "B": "-4.5", "C": -0.00004}], errors: [{ "type": "FieldMismatch", "code": "TooManyFields", "message": "Too many fields: expected 3 fields but parsed 4", "row": 0 }] } }, { description: "Dynamic typing by indices can be determined by function", input: '001,002,003', config: { dynamicTyping: function(field) { return (field % 2) === 0; } }, expected: { data: [[1, "002", 3]], errors: [] } }, { description: "Dynamic typing by headers can be determined by function", input: 'A_as_int,B,C_as_int\r\n001,002,003', config: { header: true, dynamicTyping: function(field) { return /_as_int$/.test(field); } }, expected: { data: [{"A_as_int": 1, "B": "002", "C_as_int": 3}], errors: [] } }, { description: "Blank line at beginning", input: '\r\na,b,c\r\nd,e,f', config: { newline: '\r\n' }, expected: { data: [[''], ['a', 'b', 'c'], ['d', 'e', 'f']], errors: [] } }, { description: "Blank line in middle", input: 'a,b,c\r\n\r\nd,e,f', config: { newline: '\r\n' }, expected: { data: [['a', 'b', 'c'], [''], ['d', 'e', 'f']], errors: [] } }, { description: "Blank lines at end", input: 'a,b,c\nd,e,f\n\n', expected: { data: [['a', 'b', 'c'], ['d', 'e', 'f'], [''], ['']], errors: [] } }, { description: "Blank line in middle with whitespace", input: 'a,b,c\r\n \r\nd,e,f', expected: { data: [['a', 'b', 'c'], [" "], ['d', 'e', 'f']], errors: [] } }, { description: "First field of a line is empty", input: 'a,b,c\r\n,e,f', expected: { data: [['a', 'b', 'c'], ['', 'e', 'f']], errors: [] } }, { description: "Last field of a line is empty", input: 'a,b,\r\nd,e,f', expected: { data: [['a', 'b', ''], ['d', 'e', 'f']], errors: [] } }, { description: "Other fields are empty", input: 'a,,c\r\n,,', expected: { data: [['a', '', 'c'], ['', '', '']], errors: [] } }, { description: "Empty input string", input: '', expected: { data: [], errors: [{ "type": "Delimiter", "code": "UndetectableDelimiter", "message": "Unable to auto-detect delimiting character; defaulted to ','" }] } }, { description: "Input is just the delimiter (2 empty fields)", input: ',', expected: { data: [['', '']], errors: [] } }, { description: "Input is just a string (a single field)", input: 'Abc def', expected: { data: [['Abc def']], errors: [ { "type": "Delimiter", "code": "UndetectableDelimiter", "message": "Unable to auto-detect delimiting character; defaulted to ','" } ] } }, { description: "Preview 0 rows should default to parsing all", input: 'a,b,c\r\nd,e,f\r\ng,h,i', config: { preview: 0 }, expected: { data: [['a', 'b', 'c'], ['d', 'e', 'f'], ['g', 'h', 'i']], errors: [] } }, { description: "Preview 1 row", input: 'a,b,c\r\nd,e,f\r\ng,h,i', config: { preview: 1 }, expected: { data: [['a', 'b', 'c']], errors: [] } }, { description: "Preview 2 rows", input: 'a,b,c\r\nd,e,f\r\ng,h,i', config: { preview: 2 }, expected: { data: [['a', 'b', 'c'], ['d', 'e', 'f']], errors: [] } }, { description: "Preview all (3) rows", input: 'a,b,c\r\nd,e,f\r\ng,h,i', config: { preview: 3 }, expected: { data: [['a', 'b', 'c'], ['d', 'e', 'f'], ['g', 'h', 'i']], errors: [] } }, { description: "Preview more rows than input has", input: 'a,b,c\r\nd,e,f\r\ng,h,i', config: { preview: 4 }, expected: { data: [['a', 'b', 'c'], ['d', 'e', 'f'], ['g', 'h', 'i']], errors: [] } }, { description: "Preview should count rows, not lines", input: 'a,b,c\r\nd,e,"f\r\nf",g,h,i', config: { preview: 2 }, expected: { data: [['a', 'b', 'c'], ['d', 'e', 'f\r\nf', 'g', 'h', 'i']], errors: [] } }, { description: "Preview with header row", notes: "Preview is defined to be number of rows of input not including header row", input: 'a,b,c\r\nd,e,f\r\ng,h,i\r\nj,k,l', config: { header: true, preview: 2 }, expected: { data: [{"a": "d", "b": "e", "c": "f"}, {"a": "g", "b": "h", "c": "i"}], errors: [] } }, { description: "Empty lines", input: '\na,b,c\n\nd,e,f\n\n', config: { delimiter: ',' }, expected: { data: [[''], ['a', 'b', 'c'], [''], ['d', 'e', 'f'], [''], ['']], errors: [] } }, { description: "Skip empty lines", input: 'a,b,c\n\nd,e,f', config: { skipEmptyLines: true }, expected: { data: [['a', 'b', 'c'], ['d', 'e', 'f']], errors: [] } }, { description: "Skip empty lines, with newline at end of input", input: 'a,b,c\r\n\r\nd,e,f\r\n', config: { skipEmptyLines: true }, expected: { data: [['a', 'b', 'c'], ['d', 'e', 'f']], errors: [] } }, { description: "Skip empty lines, with empty input", input: '', config: { skipEmptyLines: true }, expected: { data: [], errors: [ { "type": "Delimiter", "code": "UndetectableDelimiter", "message": "Unable to auto-detect delimiting character; defaulted to ','" } ] } }, { description: "Skip empty lines, with first line only whitespace", notes: "A line must be absolutely empty to be considered empty", input: ' \na,b,c', config: { skipEmptyLines: true, delimiter: ',' }, expected: { data: [[" "], ['a', 'b', 'c']], errors: [] } }, { description: "Single quote as quote character", notes: "Must parse correctly when single quote is specified as a quote character", input: "a,b,'c,d'", config: { quoteChar: "'"}, expected: { data: [['a', 'b', 'c,d']], errors: [] } }, { description: "Header row with preceding comment", notes: "Must parse correctly headers if they are preceded by comments", input: '#Comment\na,b\nc,d\n', config: { header: true, comments: '#', skipEmptyLines: true, delimiter: ','}, expected: { data: [{'a': 'c', 'b': 'd'}], errors: [] } } ]; describe('Parse Tests', function() { function generateTest(test) { (test.disabled ? it.skip : it)(test.description, function() { var actual = Papa.parse(test.input, test.config); assert.deepEqual(JSON.stringify(actual.errors), JSON.stringify(test.expected.errors)); assert.deepEqual(actual.data, test.expected.data); }); } for (var i = 0; i < PARSE_TESTS.length; i++) { generateTest(PARSE_TESTS[i]); } }); // Tests for Papa.parse() that involve asynchronous operation var PARSE_ASYNC_TESTS = [ { description: "Simple worker", input: "A,B,C\nX,Y,Z", config: { worker: true, }, expected: { data: [['A','B','C'],['X','Y','Z']], errors: [] } }, { description: "Simple download", input: "sample.csv", config: { download: true }, disabled: !XHR_ENABLED, expected: { data: [['A','B','C'],['X','Y','Z']], errors: [] } }, { description: "Simple download + worker", input: "tests/sample.csv", config: { worker: true, download: true }, disabled: !XHR_ENABLED, expected: { data: [['A','B','C'],['X','Y','Z']], errors: [] } }, { description: "Simple file", disabled: !FILES_ENABLED, input: FILES_ENABLED ? new File(["A,B,C\nX,Y,Z"], "sample.csv") : false, config: { }, expected: { data: [['A','B','C'],['X','Y','Z']], errors: [] } }, { description: "Simple file + worker", disabled: !FILES_ENABLED, input: FILES_ENABLED ? new File(["A,B,C\nX,Y,Z"], "sample.csv") : false, config: { worker: true, }, expected: { data: [['A','B','C'],['X','Y','Z']], errors: [] } } ]; describe('Parse Async Tests', function() { function generateTest(test) { (test.disabled ? it.skip : it)(test.description, function(done) { var config = test.config; config.complete = function(actual) { assert.deepEqual(JSON.stringify(actual.errors), JSON.stringify(test.expected.errors)); assert.deepEqual(actual.data, test.expected.data); done(); }; config.error = function(err) { throw err; }; Papa.parse(test.input, config); }); } for (var i = 0; i < PARSE_ASYNC_TESTS.length; i++) { generateTest(PARSE_ASYNC_TESTS[i]); } }); // Tests for Papa.unparse() function (JSON to CSV) var UNPARSE_TESTS = [ { description: "A simple row", notes: "Comma should be default delimiter", input: [['A', 'b', 'c']], expected: 'A,b,c' }, { description: "Two rows", input: [['A', 'b', 'c'], ['d', 'E', 'f']], expected: 'A,b,c\r\nd,E,f' }, { description: "Data with quotes", input: [['a', '"b"', 'c'], ['"d"', 'e', 'f']], expected: 'a,"""b""",c\r\n"""d""",e,f' }, { description: "Data with newlines", input: [['a', 'b\nb', 'c'], ['d', 'e', 'f\r\nf']], expected: 'a,"b\nb",c\r\nd,e,"f\r\nf"' }, { description: "Array of objects (header row)", input: [{ "Col1": "a", "Col2": "b", "Col3": "c" }, { "Col1": "d", "Col2": "e", "Col3": "f" }], expected: 'Col1,Col2,Col3\r\na,b,c\r\nd,e,f' }, { description: "With header row, missing a field in a row", input: [{ "Col1": "a", "Col2": "b", "Col3": "c" }, { "Col1": "d", "Col3": "f" }], expected: 'Col1,Col2,Col3\r\na,b,c\r\nd,,f' }, { description: "With header row, with extra field in a row", notes: "Extra field should be ignored; first object in array dictates header row", input: [{ "Col1": "a", "Col2": "b", "Col3": "c" }, { "Col1": "d", "Col2": "e", "Extra": "g", "Col3": "f" }], expected: 'Col1,Col2,Col3\r\na,b,c\r\nd,e,f' }, { description: "Specifying column names and data separately", input: { fields: ["Col1", "Col2", "Col3"], data: [["a", "b", "c"], ["d", "e", "f"]] }, expected: 'Col1,Col2,Col3\r\na,b,c\r\nd,e,f' }, { description: "Specifying column names only (no data)", notes: "Papa should add a data property that is an empty array to prevent errors (no copy is made)", input: { fields: ["Col1", "Col2", "Col3"] }, expected: 'Col1,Col2,Col3' }, { description: "Specifying data only (no field names), improperly", notes: "A single array for a single row is wrong, but it can be compensated.
Papa should add empty fields property to prevent errors.", input: { data: ["abc", "d", "ef"] }, expected: 'abc,d,ef' }, { description: "Specifying data only (no field names), properly", notes: "An array of arrays, even if just a single row.
Papa should add empty fields property to prevent errors.", input: { data: [["a", "b", "c"]] }, expected: 'a,b,c' }, { description: "Custom delimiter (semicolon)", input: [['A', 'b', 'c'], ['d', 'e', 'f']], config: { delimiter: ';' }, expected: 'A;b;c\r\nd;e;f' }, { description: "Custom delimiter (tab)", input: [['Ab', 'cd', 'ef'], ['g', 'h', 'ij']], config: { delimiter: '\t' }, expected: 'Ab\tcd\tef\r\ng\th\tij' }, { description: "Custom delimiter (ASCII 30)", input: [['a', 'b', 'c'], ['d', 'e', 'f']], config: { delimiter: RECORD_SEP }, expected: 'a'+RECORD_SEP+'b'+RECORD_SEP+'c\r\nd'+RECORD_SEP+'e'+RECORD_SEP+'f' }, { description: "Bad delimiter (\\n)", notes: "Should default to comma", input: [['a', 'b', 'c'], ['d', 'e', 'f']], config: { delimiter: '\n' }, expected: 'a,b,c\r\nd,e,f' }, { description: "Custom line ending (\\r)", input: [['a', 'b', 'c'], ['d', 'e', 'f']], config: { newline: '\r' }, expected: 'a,b,c\rd,e,f' }, { description: "Custom line ending (\\n)", input: [['a', 'b', 'c'], ['d', 'e', 'f']], config: { newline: '\n' }, expected: 'a,b,c\nd,e,f' }, { description: "Custom, but strange, line ending ($)", input: [['a', 'b', 'c'], ['d', 'e', 'f']], config: { newline: '$' }, expected: 'a,b,c$d,e,f' }, { description: "Force quotes around all fields", input: [['a', 'b', 'c'], ['d', 'e', 'f']], config: { quotes: true }, expected: '"a","b","c"\r\n"d","e","f"' }, { description: "Force quotes around all fields (with header row)", input: [{ "Col1": "a", "Col2": "b", "Col3": "c" }, { "Col1": "d", "Col2": "e", "Col3": "f" }], config: { quotes: true }, expected: '"Col1","Col2","Col3"\r\n"a","b","c"\r\n"d","e","f"' }, { description: "Force quotes around certain fields only", input: [['a', 'b', 'c'], ['d', 'e', 'f']], config: { quotes: [true, false, true] }, expected: '"a",b,"c"\r\n"d",e,"f"' }, { description: "Force quotes around certain fields only (with header row)", input: [{ "Col1": "a", "Col2": "b", "Col3": "c" }, { "Col1": "d", "Col2": "e", "Col3": "f" }], config: { quotes: [true, false, true] }, expected: '"Col1",Col2,"Col3"\r\n"a",b,"c"\r\n"d",e,"f"' }, { description: "Empty input", input: [], expected: '' }, { description: "Mismatched field counts in rows", input: [['a', 'b', 'c'], ['d', 'e'], ['f']], expected: 'a,b,c\r\nd,e\r\nf' }, { description: "JSON null is treated as empty value", input: [{ "Col1": "a", "Col2": null, "Col3": "c" }], expected: 'Col1,Col2,Col3\r\na,,c' }, { description: "Custom quote character (single quote)", input: [['a,d','b','c']], config: { quoteChar: "'"}, expected: "'a,d',b,c" }, { description: "Don't print header if header:false option specified", input: [{ "Col1": "a", "Col2": "b", "Col3": "c" }, { "Col1": "d", "Col2": "e", "Col3": "f" }], config: { header: false }, expected: 'a,b,c\r\nd,e,f' } ]; describe('Unparse Tests', function() { function generateTest(test) { (test.disabled ? it.skip : it)(test.description, function() { var actual; try { actual = Papa.unparse(test.input, test.config); } catch (e) { if (e instanceof Error) { throw e; } actual = e; } assert.strictEqual(actual, test.expected); }); } for (var i = 0; i < UNPARSE_TESTS.length; i++) { generateTest(UNPARSE_TESTS[i]); } }); var CUSTOM_TESTS = [ { description: "Complete is called with all results if neither step nor chunk is defined", expected: [['A', 'b', 'c'], ['d', 'E', 'f'], ['G', 'h', 'i']], disabled: !FILES_ENABLED, run: function(callback) { Papa.parse(new File(['A,b,c\nd,E,f\nG,h,i'], 'sample.csv'), { chunkSize: 3, complete: function(response) { callback(response.data); } }); } }, { description: "Step is called for each row", expected: 2, run: function(callback) { var callCount = 0; Papa.parse('A,b,c\nd,E,f', { step: function() { callCount++; }, complete: function() { callback(callCount); } }); } }, { description: "Step is called with the contents of the row", expected: ['A', 'b', 'c'], run: function(callback) { Papa.parse('A,b,c', { step: function(response) { callback(response.data[0]); } }); } }, { description: "Step is called with the last cursor position", expected: [6, 12, 17], run: function(callback) { var updates = []; Papa.parse('A,b,c\nd,E,f\nG,h,i', { step: function(response) { updates.push(response.meta.cursor); }, complete: function() { callback(updates); } }); } }, { description: "Step exposes cursor for downloads", expected: [129, 287, 452, 595, 727, 865, 1031, 1209], disabled: !XHR_ENABLED, run: function(callback) { var updates = []; Papa.parse("/tests/long-sample.csv", { download: true, step: function(response) { updates.push(response.meta.cursor); }, complete: function() { callback(updates); } }); } }, { description: "Step exposes cursor for chunked downloads", expected: [129, 287, 452, 595, 727, 865, 1031, 1209], disabled: !XHR_ENABLED, run: function(callback) { var updates = []; Papa.parse("/tests/long-sample.csv", { download: true, chunkSize: 500, step: function(response) { updates.push(response.meta.cursor); }, complete: function() { callback(updates); } }); } }, { description: "Step exposes cursor for workers", expected: [452, 452, 452, 865, 865, 865, 1209, 1209], disabled: !XHR_ENABLED, run: function(callback) { var updates = []; Papa.parse("/tests/long-sample.csv", { download: true, chunkSize: 500, worker: true, step: function(response) { updates.push(response.meta.cursor); }, complete: function() { callback(updates); } }); } }, { description: "Chunk is called for each chunk", expected: [3, 3, 2], disabled: !XHR_ENABLED, run: function(callback) { var updates = []; Papa.parse("/tests/long-sample.csv", { download: true, chunkSize: 500, chunk: function(response) { updates.push(response.data.length); }, complete: function() { callback(updates); } }); } }, { description: "Chunk is called with cursor position", expected: [452, 865, 1209], disabled: !XHR_ENABLED, run: function(callback) { var updates = []; Papa.parse("/tests/long-sample.csv", { download: true, chunkSize: 500, chunk: function(response) { updates.push(response.meta.cursor); }, complete: function() { callback(updates); } }); } }, { description: "Step exposes indexes for files", expected: [6, 12, 17], disabled: !FILES_ENABLED, run: function(callback) { var updates = []; Papa.parse(new File(['A,b,c\nd,E,f\nG,h,i'], 'sample.csv'), { download: true, step: function(response) { updates.push(response.meta.cursor); }, complete: function() { callback(updates); } }); } }, { description: "Step exposes indexes for chunked files", expected: [6, 12, 17], disabled: !FILES_ENABLED, run: function(callback) { var updates = []; Papa.parse(new File(['A,b,c\nd,E,f\nG,h,i'], 'sample.csv'), { chunkSize: 3, step: function(response) { updates.push(response.meta.cursor); }, complete: function() { callback(updates); } }); } }, { description: "Quoted line breaks near chunk boundaries are handled", expected: [['A', 'B', 'C'], ['X', 'Y\n1\n2\n3', 'Z']], disabled: !FILES_ENABLED, run: function(callback) { var updates = []; Papa.parse(new File(['A,B,C\nX,"Y\n1\n2\n3",Z'], 'sample.csv'), { chunkSize: 3, step: function(response) { updates.push(response.data[0]); }, complete: function() { callback(updates); } }); } }, { description: "Step functions can abort parsing", expected: [['A', 'b', 'c']], run: function(callback) { var updates = []; Papa.parse('A,b,c\nd,E,f\nG,h,i', { step: function(response, handle) { updates.push(response.data[0]); handle.abort(); callback(updates); }, chunkSize: 6 }); } }, { description: "Complete is called after aborting", expected: true, run: function(callback) { var updates = []; Papa.parse('A,b,c\nd,E,f\nG,h,i', { step: function(response, handle) { handle.abort(); }, chunkSize: 6, complete: function (response) { callback(response.meta.aborted); } }); } }, { description: "Step functions can pause parsing", expected: [['A', 'b', 'c']], run: function(callback) { var updates = []; Papa.parse('A,b,c\nd,E,f\nG,h,i', { step: function(response, handle) { updates.push(response.data[0]); handle.pause(); callback(updates); }, complete: function() { callback('incorrect complete callback'); } }); } }, { description: "Step functions can resume parsing", expected: [['A', 'b', 'c'], ['d', 'E', 'f'], ['G', 'h', 'i']], run: function(callback) { var updates = []; var handle = null; var first = true; Papa.parse('A,b,c\nd,E,f\nG,h,i', { step: function(response, h) { updates.push(response.data[0]); if (!first) return; handle = h; handle.pause(); first = false; }, complete: function() { callback(updates); } }); setTimeout(function() { handle.resume(); }, 500); } }, { description: "Step functions can abort workers", expected: 1, disabled: !XHR_ENABLED, run: function(callback) { var updates = 0; Papa.parse("/tests/long-sample.csv", { worker: true, download: true, chunkSize: 500, step: function(response, handle) { updates++; handle.abort(); }, complete: function() { callback(updates); } }); } }, { description: "beforeFirstChunk manipulates only first chunk", expected: 7, disabled: !XHR_ENABLED, run: function(callback) { var updates = 0; Papa.parse("/tests/long-sample.csv", { download: true, chunkSize: 500, beforeFirstChunk: function(chunk) { return chunk.replace(/.*?\n/, ''); }, step: function(response) { updates++; }, complete: function() { callback(updates); } }); } }, { description: "First chunk not modified if beforeFirstChunk returns nothing", expected: 8, disabled: !XHR_ENABLED, run: function(callback) { var updates = 0; Papa.parse("/tests/long-sample.csv", { download: true, chunkSize: 500, beforeFirstChunk: function(chunk) { }, step: function(response) { updates++; }, complete: function() { callback(updates); } }); } }, { description: "Should not assume we own the worker unless papaworker is in the search string", disabled: typeof Worker === 'undefined', expected: [false, true, true, true, true], run: function(callback) { var searchStrings = [ '', '?papaworker', '?x=1&papaworker', '?x=1&papaworker&y=1', '?x=1&papaworker=1' ]; var results = searchStrings.map(function () { return false; }); var workers = []; // Give it .5s to do something setTimeout(function () { workers.forEach(function (w) { w.terminate(); }); callback(results); }, 500); searchStrings.forEach(function (searchString, idx) { var w = new Worker('../papaparse.js' + searchString); workers.push(w); w.addEventListener('message', function () { results[idx] = true; }); w.postMessage({input: 'a,b,c\n1,2,3'}); }); } } ]; describe('Custom Tests', function() { function generateTest(test) { (test.disabled ? it.skip : it)(test.description, function(done) { test.run(function (actual) { assert.deepEqual(JSON.stringify(actual), JSON.stringify(test.expected)); done(); }); }); } for (var i = 0; i < CUSTOM_TESTS.length; i++) { generateTest(CUSTOM_TESTS[i]); } });