reader.go

Documentation: encoding/csv

     1  // Copyright 2011 The Go Authors. All rights reserved.
     2  // Use of this source code is governed by a BSD-style
     3  // license that can be found in the LICENSE file.
     4  
     5  // Package csv reads and writes comma-separated values (CSV) files.
     6  // There are many kinds of CSV files; this package supports the format
     7  // described in RFC 4180.
     8  //
     9  // A csv file contains zero or more records of one or more fields per record.
    10  // Each record is separated by the newline character. The final record may
    11  // optionally be followed by a newline character.
    12  //
    13  //	field1,field2,field3
    14  //
    15  // White space is considered part of a field.
    16  //
    17  // Carriage returns before newline characters are silently removed.
    18  //
    19  // Blank lines are ignored. A line with only whitespace characters (excluding
    20  // the ending newline character) is not considered a blank line.
    21  //
    22  // Fields which start and stop with the quote character " are called
    23  // quoted-fields. The beginning and ending quote are not part of the
    24  // field.
    25  //
    26  // The source:
    27  //
    28  //	normal string,"quoted-field"
    29  //
    30  // results in the fields
    31  //
    32  //	{`normal string`, `quoted-field`}
    33  //
    34  // Within a quoted-field a quote character followed by a second quote
    35  // character is considered a single quote.
    36  //
    37  //	"the ""word"" is true","a ""quoted-field"""
    38  //
    39  // results in
    40  //
    41  //	{`the "word" is true`, `a "quoted-field"`}
    42  //
    43  // Newlines and commas may be included in a quoted-field
    44  //
    45  //	"Multi-line
    46  //	field","comma is ,"
    47  //
    48  // results in
    49  //
    50  //	{`Multi-line
    51  //	field`, `comma is ,`}
    52  package csv
    53  
    54  import (
    55  	"bufio"
    56  	"bytes"
    57  	"errors"
    58  	"fmt"
    59  	"io"
    60  	"unicode"
    61  	"unicode/utf8"
    62  )
    63  
    64  // A ParseError is returned for parsing errors.
    65  // Line numbers are 1-indexed and columns are 0-indexed.
    66  type ParseError struct {
    67  	StartLine int   // Line where the record starts
    68  	Line      int   // Line where the error occurred
    69  	Column    int   // Column (1-based byte index) where the error occurred
    70  	Err       error // The actual error
    71  }
    72  
    73  func (e *ParseError) Error() string {
    74  	if e.Err == ErrFieldCount {
    75  		return fmt.Sprintf("record on line %d: %v", e.Line, e.Err)
    76  	}
    77  	if e.StartLine != e.Line {
    78  		return fmt.Sprintf("record on line %d; parse error on line %d, column %d: %v", e.StartLine, e.Line, e.Column, e.Err)
    79  	}
    80  	return fmt.Sprintf("parse error on line %d, column %d: %v", e.Line, e.Column, e.Err)
    81  }
    82  
    83  func (e *ParseError) Unwrap() error { return e.Err }
    84  
    85  // These are the errors that can be returned in ParseError.Err.
    86  var (
    87  	ErrTrailingComma = errors.New("extra delimiter at end of line") // Deprecated: No longer used.
    88  	ErrBareQuote     = errors.New("bare \" in non-quoted-field")
    89  	ErrQuote         = errors.New("extraneous or missing \" in quoted-field")
    90  	ErrFieldCount    = errors.New("wrong number of fields")
    91  )
    92  
    93  var errInvalidDelim = errors.New("csv: invalid field or comment delimiter")
    94  
    95  func validDelim(r rune) bool {
    96  	return r != 0 && r != '"' && r != '\r' && r != '\n' && utf8.ValidRune(r) && r != utf8.RuneError
    97  }
    98  
    99  // A Reader reads records from a CSV-encoded file.
   100  //
   101  // As returned by NewReader, a Reader expects input conforming to RFC 4180.
   102  // The exported fields can be changed to customize the details before the
   103  // first call to Read or ReadAll.
   104  //
   105  // The Reader converts all \r\n sequences in its input to plain \n,
   106  // including in multiline field values, so that the returned data does
   107  // not depend on which line-ending convention an input file uses.
   108  type Reader struct {
   109  	// Comma is the field delimiter.
   110  	// It is set to comma (',') by NewReader.
   111  	// Comma must be a valid rune and must not be \r, \n,
   112  	// or the Unicode replacement character (0xFFFD).
   113  	Comma rune
   114  
   115  	// Comment, if not 0, is the comment character. Lines beginning with the
   116  	// Comment character without preceding whitespace are ignored.
   117  	// With leading whitespace the Comment character becomes part of the
   118  	// field, even if TrimLeadingSpace is true.
   119  	// Comment must be a valid rune and must not be \r, \n,
   120  	// or the Unicode replacement character (0xFFFD).
   121  	// It must also not be equal to Comma.
   122  	Comment rune
   123  
   124  	// FieldsPerRecord is the number of expected fields per record.
   125  	// If FieldsPerRecord is positive, Read requires each record to
   126  	// have the given number of fields. If FieldsPerRecord is 0, Read sets it to
   127  	// the number of fields in the first record, so that future records must
   128  	// have the same field count. If FieldsPerRecord is negative, no check is
   129  	// made and records may have a variable number of fields.
   130  	FieldsPerRecord int
   131  
   132  	// If LazyQuotes is true, a quote may appear in an unquoted field and a
   133  	// non-doubled quote may appear in a quoted field.
   134  	LazyQuotes bool
   135  
   136  	// If TrimLeadingSpace is true, leading white space in a field is ignored.
   137  	// This is done even if the field delimiter, Comma, is white space.
   138  	TrimLeadingSpace bool
   139  
   140  	// ReuseRecord controls whether calls to Read may return a slice sharing
   141  	// the backing array of the previous call's returned slice for performance.
   142  	// By default, each call to Read returns newly allocated memory owned by the caller.
   143  	ReuseRecord bool
   144  
   145  	TrailingComma bool // Deprecated: No longer used.
   146  
   147  	r *bufio.Reader
   148  
   149  	// numLine is the current line being read in the CSV file.
   150  	numLine int
   151  
   152  	// offset is the input stream byte offset of the current reader position.
   153  	offset int64
   154  
   155  	// rawBuffer is a line buffer only used by the readLine method.
   156  	rawBuffer []byte
   157  
   158  	// recordBuffer holds the unescaped fields, one after another.
   159  	// The fields can be accessed by using the indexes in fieldIndexes.
   160  	// E.g., For the row `a,"b","c""d",e`, recordBuffer will contain `abc"de`
   161  	// and fieldIndexes will contain the indexes [1, 2, 5, 6].
   162  	recordBuffer []byte
   163  
   164  	// fieldIndexes is an index of fields inside recordBuffer.
   165  	// The i'th field ends at offset fieldIndexes[i] in recordBuffer.
   166  	fieldIndexes []int
   167  
   168  	// fieldPositions is an index of field positions for the
   169  	// last record returned by Read.
   170  	fieldPositions []position
   171  
   172  	// lastRecord is a record cache and only used when ReuseRecord == true.
   173  	lastRecord []string
   174  }
   175  
   176  // NewReader returns a new Reader that reads from r.
   177  func NewReader(r io.Reader) *Reader {
   178  	return &Reader{
   179  		Comma: ',',
   180  		r:     bufio.NewReader(r),
   181  	}
   182  }
   183  
   184  // Read reads one record (a slice of fields) from r.
   185  // If the record has an unexpected number of fields,
   186  // Read returns the record along with the error ErrFieldCount.
   187  // Except for that case, Read always returns either a non-nil
   188  // record or a non-nil error, but not both.
   189  // If there is no data left to be read, Read returns nil, io.EOF.
   190  // If ReuseRecord is true, the returned slice may be shared
   191  // between multiple calls to Read.
   192  func (r *Reader) Read() (record []string, err error) {
   193  	if r.ReuseRecord {
   194  		record, err = r.readRecord(r.lastRecord)
   195  		r.lastRecord = record
   196  	} else {
   197  		record, err = r.readRecord(nil)
   198  	}
   199  	return record, err
   200  }
   201  
   202  // FieldPos returns the line and column corresponding to
   203  // the start of the field with the given index in the slice most recently
   204  // returned by Read. Numbering of lines and columns starts at 1;
   205  // columns are counted in bytes, not runes.
   206  //
   207  // If this is called with an out-of-bounds index, it panics.
   208  func (r *Reader) FieldPos(field int) (line, column int) {
   209  	if field < 0 || field >= len(r.fieldPositions) {
   210  		panic("out of range index passed to FieldPos")
   211  	}
   212  	p := &r.fieldPositions[field]
   213  	return p.line, p.col
   214  }
   215  
   216  // InputOffset returns the input stream byte offset of the current reader
   217  // position. The offset gives the location of the end of the most recently
   218  // read row and the beginning of the next row.
   219  func (r *Reader) InputOffset() int64 {
   220  	return r.offset
   221  }
   222  
   223  // pos holds the position of a field in the current line.
   224  type position struct {
   225  	line, col int
   226  }
   227  
   228  // ReadAll reads all the remaining records from r.
   229  // Each record is a slice of fields.
   230  // A successful call returns err == nil, not err == io.EOF. Because ReadAll is
   231  // defined to read until EOF, it does not treat end of file as an error to be
   232  // reported.
   233  func (r *Reader) ReadAll() (records [][]string, err error) {
   234  	for {
   235  		record, err := r.readRecord(nil)
   236  		if err == io.EOF {
   237  			return records, nil
   238  		}
   239  		if err != nil {
   240  			return nil, err
   241  		}
   242  		records = append(records, record)
   243  	}
   244  }
   245  
   246  // readLine reads the next line (with the trailing endline).
   247  // If EOF is hit without a trailing endline, it will be omitted.
   248  // If some bytes were read, then the error is never io.EOF.
   249  // The result is only valid until the next call to readLine.
   250  func (r *Reader) readLine() ([]byte, error) {
   251  	line, err := r.r.ReadSlice('\n')
   252  	if err == bufio.ErrBufferFull {
   253  		r.rawBuffer = append(r.rawBuffer[:0], line...)
   254  		for err == bufio.ErrBufferFull {
   255  			line, err = r.r.ReadSlice('\n')
   256  			r.rawBuffer = append(r.rawBuffer, line...)
   257  		}
   258  		line = r.rawBuffer
   259  	}
   260  	readSize := len(line)
   261  	if readSize > 0 && err == io.EOF {
   262  		err = nil
   263  		// For backwards compatibility, drop trailing \r before EOF.
   264  		if line[readSize-1] == '\r' {
   265  			line = line[:readSize-1]
   266  		}
   267  	}
   268  	r.numLine++
   269  	r.offset += int64(readSize)
   270  	// Normalize \r\n to \n on all input lines.
   271  	if n := len(line); n >= 2 && line[n-2] == '\r' && line[n-1] == '\n' {
   272  		line[n-2] = '\n'
   273  		line = line[:n-1]
   274  	}
   275  	return line, err
   276  }
   277  
   278  // lengthNL reports the number of bytes for the trailing \n.
   279  func lengthNL(b []byte) int {
   280  	if len(b) > 0 && b[len(b)-1] == '\n' {
   281  		return 1
   282  	}
   283  	return 0
   284  }
   285  
   286  // nextRune returns the next rune in b or utf8.RuneError.
   287  func nextRune(b []byte) rune {
   288  	r, _ := utf8.DecodeRune(b)
   289  	return r
   290  }
   291  
   292  func (r *Reader) readRecord(dst []string) ([]string, error) {
   293  	if r.Comma == r.Comment || !validDelim(r.Comma) || (r.Comment != 0 && !validDelim(r.Comment)) {
   294  		return nil, errInvalidDelim
   295  	}
   296  
   297  	// Read line (automatically skipping past empty lines and any comments).
   298  	var line []byte
   299  	var errRead error
   300  	for errRead == nil {
   301  		line, errRead = r.readLine()
   302  		if r.Comment != 0 && nextRune(line) == r.Comment {
   303  			line = nil
   304  			continue // Skip comment lines
   305  		}
   306  		if errRead == nil && len(line) == lengthNL(line) {
   307  			line = nil
   308  			continue // Skip empty lines
   309  		}
   310  		break
   311  	}
   312  	if errRead == io.EOF {
   313  		return nil, errRead
   314  	}
   315  
   316  	// Parse each field in the record.
   317  	var err error
   318  	const quoteLen = len(`"`)
   319  	commaLen := utf8.RuneLen(r.Comma)
   320  	recLine := r.numLine // Starting line for record
   321  	r.recordBuffer = r.recordBuffer[:0]
   322  	r.fieldIndexes = r.fieldIndexes[:0]
   323  	r.fieldPositions = r.fieldPositions[:0]
   324  	pos := position{line: r.numLine, col: 1}
   325  parseField:
   326  	for {
   327  		if r.TrimLeadingSpace {
   328  			i := bytes.IndexFunc(line, func(r rune) bool {
   329  				return !unicode.IsSpace(r)
   330  			})
   331  			if i < 0 {
   332  				i = len(line)
   333  				pos.col -= lengthNL(line)
   334  			}
   335  			line = line[i:]
   336  			pos.col += i
   337  		}
   338  		if len(line) == 0 || line[0] != '"' {
   339  			// Non-quoted string field
   340  			i := bytes.IndexRune(line, r.Comma)
   341  			field := line
   342  			if i >= 0 {
   343  				field = field[:i]
   344  			} else {
   345  				field = field[:len(field)-lengthNL(field)]
   346  			}
   347  			// Check to make sure a quote does not appear in field.
   348  			if !r.LazyQuotes {
   349  				if j := bytes.IndexByte(field, '"'); j >= 0 {
   350  					col := pos.col + j
   351  					err = &ParseError{StartLine: recLine, Line: r.numLine, Column: col, Err: ErrBareQuote}
   352  					break parseField
   353  				}
   354  			}
   355  			r.recordBuffer = append(r.recordBuffer, field...)
   356  			r.fieldIndexes = append(r.fieldIndexes, len(r.recordBuffer))
   357  			r.fieldPositions = append(r.fieldPositions, pos)
   358  			if i >= 0 {
   359  				line = line[i+commaLen:]
   360  				pos.col += i + commaLen
   361  				continue parseField
   362  			}
   363  			break parseField
   364  		} else {
   365  			// Quoted string field
   366  			fieldPos := pos
   367  			line = line[quoteLen:]
   368  			pos.col += quoteLen
   369  			for {
   370  				i := bytes.IndexByte(line, '"')
   371  				if i >= 0 {
   372  					// Hit next quote.
   373  					r.recordBuffer = append(r.recordBuffer, line[:i]...)
   374  					line = line[i+quoteLen:]
   375  					pos.col += i + quoteLen
   376  					switch rn := nextRune(line); {
   377  					case rn == '"':
   378  						// `""` sequence (append quote).
   379  						r.recordBuffer = append(r.recordBuffer, '"')
   380  						line = line[quoteLen:]
   381  						pos.col += quoteLen
   382  					case rn == r.Comma:
   383  						// `",` sequence (end of field).
   384  						line = line[commaLen:]
   385  						pos.col += commaLen
   386  						r.fieldIndexes = append(r.fieldIndexes, len(r.recordBuffer))
   387  						r.fieldPositions = append(r.fieldPositions, fieldPos)
   388  						continue parseField
   389  					case lengthNL(line) == len(line):
   390  						// `"\n` sequence (end of line).
   391  						r.fieldIndexes = append(r.fieldIndexes, len(r.recordBuffer))
   392  						r.fieldPositions = append(r.fieldPositions, fieldPos)
   393  						break parseField
   394  					case r.LazyQuotes:
   395  						// `"` sequence (bare quote).
   396  						r.recordBuffer = append(r.recordBuffer, '"')
   397  					default:
   398  						// `"*` sequence (invalid non-escaped quote).
   399  						err = &ParseError{StartLine: recLine, Line: r.numLine, Column: pos.col - quoteLen, Err: ErrQuote}
   400  						break parseField
   401  					}
   402  				} else if len(line) > 0 {
   403  					// Hit end of line (copy all data so far).
   404  					r.recordBuffer = append(r.recordBuffer, line...)
   405  					if errRead != nil {
   406  						break parseField
   407  					}
   408  					pos.col += len(line)
   409  					line, errRead = r.readLine()
   410  					if len(line) > 0 {
   411  						pos.line++
   412  						pos.col = 1
   413  					}
   414  					if errRead == io.EOF {
   415  						errRead = nil
   416  					}
   417  				} else {
   418  					// Abrupt end of file (EOF or error).
   419  					if !r.LazyQuotes && errRead == nil {
   420  						err = &ParseError{StartLine: recLine, Line: pos.line, Column: pos.col, Err: ErrQuote}
   421  						break parseField
   422  					}
   423  					r.fieldIndexes = append(r.fieldIndexes, len(r.recordBuffer))
   424  					r.fieldPositions = append(r.fieldPositions, fieldPos)
   425  					break parseField
   426  				}
   427  			}
   428  		}
   429  	}
   430  	if err == nil {
   431  		err = errRead
   432  	}
   433  
   434  	// Create a single string and create slices out of it.
   435  	// This pins the memory of the fields together, but allocates once.
   436  	str := string(r.recordBuffer) // Convert to string once to batch allocations
   437  	dst = dst[:0]
   438  	if cap(dst) < len(r.fieldIndexes) {
   439  		dst = make([]string, len(r.fieldIndexes))
   440  	}
   441  	dst = dst[:len(r.fieldIndexes)]
   442  	var preIdx int
   443  	for i, idx := range r.fieldIndexes {
   444  		dst[i] = str[preIdx:idx]
   445  		preIdx = idx
   446  	}
   447  
   448  	// Check or update the expected fields per record.
   449  	if r.FieldsPerRecord > 0 {
   450  		if len(dst) != r.FieldsPerRecord && err == nil {
   451  			err = &ParseError{
   452  				StartLine: recLine,
   453  				Line:      recLine,
   454  				Column:    1,
   455  				Err:       ErrFieldCount,
   456  			}
   457  		}
   458  	} else if r.FieldsPerRecord == 0 {
   459  		r.FieldsPerRecord = len(dst)
   460  	}
   461  	return dst, err
   462  }
   463
View as plain text