The London Perl and Raku Workshop takes place on 26th Oct 2024. If your company depends on Perl, please consider sponsoring and/or attending.


Text::CSV_XS - comma-separated values manipulation routines


 # Functional interface
 use Text::CSV_XS qw( csv );

 # Read whole file in memory
 my $aoa = csv (in => "data.csv");    # as array of array
 my $aoh = csv (in => "data.csv",
                headers => "auto");   # as array of hash

 # Write array of arrays as csv file
 csv (in => $aoa, out => "file.csv", sep_char => ";");

 # Only show lines where "code" is odd
 csv (in => "data.csv", filter => { code => sub { $_ % 2 }});

 # Object interface
 use Text::CSV_XS;

 my @rows;
 # Read/parse CSV
 my $csv = Text::CSV_XS->new ({ binary => 1, auto_diag => 1 });
 open my $fh, "<:encoding(utf8)", "test.csv" or die "test.csv: $!";
 while (my $row = $csv->getline ($fh)) {
     $row->[2] =~ m/pattern/ or next; # 3rd field should match
     push @rows, $row;
 close $fh;

 # and write as CSV
 open $fh, ">:encoding(utf8)", "new.csv" or die "new.csv: $!";
 $csv->say ($fh, $_) for @rows;
 close $fh or die "new.csv: $!";


Text::CSV_XS provides facilities for the composition and decomposition of comma-separated values. An instance of the Text::CSV_XS class will combine fields into a CSV string and parse a CSV string into fields.

The module accepts either strings or files as input and support the use of user-specified characters for delimiters, separators, and escapes.

Embedded newlines

Important Note: The default behavior is to accept only ASCII characters in the range from 0x20 (space) to 0x7E (tilde). This means that the fields can not contain newlines. If your data contains newlines embedded in fields, or characters above 0x7E (tilde), or binary data, you must set binary => 1 in the call to "new". To cover the widest range of parsing options, you will always want to set binary.

But you still have the problem that you have to pass a correct line to the "parse" method, which is more complicated from the usual point of usage:

 my $csv = Text::CSV_XS->new ({ binary => 1, eol => $/ });
 while (<>) {           #  WRONG!
     $csv->parse ($_);
     my @fields = $csv->fields ();

this will break, as the while might read broken lines: it does not care about the quoting. If you need to support embedded newlines, the way to go is to not pass eol in the parser (it accepts \n, \r, and \r\n by default) and then

 my $csv = Text::CSV_XS->new ({ binary => 1 });
 open my $fh, "<", $file or die "$file: $!";
 while (my $row = $csv->getline ($fh)) {
     my @fields = @$row;

The old(er) way of using global file handles is still supported

 while (my $row = $csv->getline (*ARGV)) { ... }


Unicode is only tested to work with perl-5.8.2 and up.

See also "BOM".

The simplest way to ensure the correct encoding is used for in- and output is by either setting layers on the filehandles, or setting the "encoding" argument for "csv".

 open my $fh, "<:encoding(UTF-8)", "in.csv"  or die "in.csv: $!";
 my $aoa = csv (in => "in.csv",     encoding => "UTF-8");

 open my $fh, ">:encoding(UTF-8)", "out.csv" or die "out.csv: $!";
 csv (in => $aoa, out => "out.csv", encoding => "UTF-8");

On parsing (both for "getline" and "parse"), if the source is marked being UTF8, then all fields that are marked binary will also be marked UTF8.

On combining ("print" and "combine"): if any of the combining fields was marked UTF8, the resulting string will be marked as UTF8. Note however that all fields before the first field marked UTF8 and contained 8-bit characters that were not upgraded to UTF8, these will be bytes in the resulting string too, possibly causing unexpected errors. If you pass data of different encoding, or you don't know if there is different encoding, force it to be upgraded before you pass them on:

 $csv->print ($fh, [ map { utf8::upgrade (my $x = $_); $x } @data ]);

For complete control over encoding, please use Text::CSV::Encoded:

 use Text::CSV::Encoded;
 my $csv = Text::CSV::Encoded->new ({
     encoding_in  => "iso-8859-1", # the encoding comes into   Perl
     encoding_out => "cp1252",     # the encoding comes out of Perl

 $csv = Text::CSV::Encoded->new ({ encoding  => "utf8" });
 # combine () and print () accept *literally* utf8 encoded data
 # parse () and getline () return *literally* utf8 encoded data

 $csv = Text::CSV::Encoded->new ({ encoding  => undef }); # default
 # combine () and print () accept UTF8 marked data
 # parse () and getline () return UTF8 marked data


BOM (or Byte Order Mark) handling is available only inside the "header" method. This method supports the following encodings: utf-8, utf-1, utf-32be, utf-32le, utf-16be, utf-16le, utf-ebcdic, scsu, bocu-1, and gb-18030. See Wikipedia.

If a file has a BOM, the easiest way to deal with that is

 my $aoh = csv (in => $file, detect_bom => 1);

All records will be encoded based on the detected BOM.

This implies a call to the "header" method, which defaults to also set the "column_names". So this is not the same as

 my $aoh = csv (in => $file, headers => "auto");

which only reads the first record to set "column_names" but ignores any meaning of possible present BOM.


While no formal specification for CSV exists, RFC 4180 (1) describes the common format and establishes text/csv as the MIME type registered with the IANA. RFC 7111 (2) adds fragments to CSV.

Many informal documents exist that describe the CSV format. "How To: The Comma Separated Value (CSV) File Format" (3) provides an overview of the CSV format in the most widely used applications and explains how it can best be used and supported.


The basic rules are as follows:

CSV is a delimited data format that has fields/columns separated by the comma character and records/rows separated by newlines. Fields that contain a special character (comma, newline, or double quote), must be enclosed in double quotes. However, if a line contains a single entry that is the empty string, it may be enclosed in double quotes. If a field's value contains a double quote character it is escaped by placing another double quote character next to it. The CSV file format does not require a specific character encoding, byte order, or line terminator format.

  • Each record is a single line ended by a line feed (ASCII/LF=0x0A) or a carriage return and line feed pair (ASCII/CRLF=0x0D 0x0A), however, line-breaks may be embedded.

  • Fields are separated by commas.

  • Allowable characters within a CSV field include 0x09 (TAB) and the inclusive range of 0x20 (space) through 0x7E (tilde). In binary mode all characters are accepted, at least in quoted fields.

  • A field within CSV must be surrounded by double-quotes to contain a separator character (comma).

Though this is the most clear and restrictive definition, Text::CSV_XS is way more liberal than this, and allows extension:

  • Line termination by a single carriage return is accepted by default

  • The separation-, quote-, and escape character(s) can be any ASCII character in the range from 0x20 (space) to 0x7E (tilde). Characters outside this range may or may not work as expected. Multibyte characters, like UTF U+060C (ARABIC COMMA), U+FF0C (FULLWIDTH COMMA), U+241B (SYMBOL FOR ESCAPE), U+2424 (SYMBOL FOR NEWLINE), U+FF02 (FULLWIDTH QUOTATION MARK), and U+201C (LEFT DOUBLE QUOTATION MARK) (to give some examples of what might look promising) work for newer versions of perl for sep_char, and quote_char but not for escape_char.

    If you use perl-5.8.2 or higher these three attributes are utf8-decoded, to increase the likelihood of success. This way U+00FE will be allowed as a quote character.

  • A field in CSV must be surrounded by double-quotes to make an embedded double-quote, represented by a pair of consecutive double-quotes, valid. In binary mode you may additionally use the sequence "0 for representation of a NULL byte. Using 0x00 in binary mode is just as valid.

  • Several violations of the above specification may be lifted by passing some options as attributes to the object constructor.



(Class method) Returns the current module version.


(Class method) Returns a new instance of class Text::CSV_XS. The attributes are described by the (optional) hash ref \%attr.

 my $csv = Text::CSV_XS->new ({ attributes ... });

The following attributes are available:


 my $csv = Text::CSV_XS->new ({ eol => $/ });
           $csv->eol (undef);
 my $eol = $csv->eol;

The end-of-line string to add to rows for "print" or the record separator for "getline".

When not passed in a parser instance, the default behavior is to accept \n, \r, and \r\n, so it is probably safer to not specify eol at all. Passing undef or the empty string behave the same.

When not passed in a generating instance, records are not terminated at all, so it is probably wise to pass something you expect. A safe choice for eol on output is either $/ or \r\n.

Common values for eol are "\012" (\n or Line Feed), "\015\012" (\r\n or Carriage Return, Line Feed), and "\015" (\r or Carriage Return). The eol attribute cannot exceed 7 (ASCII) characters.

If both $/ and eol equal "\015", parsing lines that end on only a Carriage Return without Line Feed, will be "parse"d correct.


 my $csv = Text::CSV_XS->new ({ sep_char => ";" });
         $csv->sep_char (";");
 my $c = $csv->sep_char;

The char used to separate fields, by default a comma. (,). Limited to a single-byte character, usually in the range from 0x20 (space) to 0x7E (tilde). When longer sequences are required, use sep.

The separation character can not be equal to the quote character or to the escape character.

See also "CAVEATS"


 my $csv = Text::CSV_XS->new ({ sep => "\N{FULLWIDTH COMMA}" });
           $csv->sep (";");
 my $sep = $csv->sep;

The chars used to separate fields, by default undefined. Limited to 8 bytes.

When set, overrules sep_char. If its length is one byte it acts as an alias to sep_char.

See also "CAVEATS"


 my $csv = Text::CSV_XS->new ({ quote_char => "'" });
         $csv->quote_char (undef);
 my $c = $csv->quote_char;

The character to quote fields containing blanks or binary data, by default the double quote character ("). A value of undef suppresses quote chars (for simple cases only). Limited to a single-byte character, usually in the range from 0x20 (space) to 0x7E (tilde). When longer sequences are required, use quote.

quote_char can not be equal to sep_char.


 my $csv = Text::CSV_XS->new ({ quote => "\N{FULLWIDTH QUOTATION MARK}" });
             $csv->quote ("'");
 my $quote = $csv->quote;

The chars used to quote fields, by default undefined. Limited to 8 bytes.

When set, overrules quote_char. If its length is one byte it acts as an alias to quote_char.

This method does not support undef. Use quote_char to disable quotation.

See also "CAVEATS"


 my $csv = Text::CSV_XS->new ({ escape_char => "\\" });
         $csv->escape_char (":");
 my $c = $csv->escape_char;

The character to escape certain characters inside quoted fields. This is limited to a single-byte character, usually in the range from 0x20 (space) to 0x7E (tilde).

The escape_char defaults to being the double-quote mark ("). In other words the same as the default quote_char. This means that doubling the quote mark in a field escapes it:

 "foo","bar","Escape ""quote mark"" with two ""quote marks""","baz"

If you change the quote_char without changing the escape_char, the escape_char will still be the double-quote ("). If instead you want to escape the quote_char by doubling it you will need to also change the escape_char to be the same as what you have changed the quote_char to.

Setting escape_char to undef or "" will completely disable escapes and is greatly discouraged. This will also disable escape_null.

The escape character can not be equal to the separation character.


 my $csv = Text::CSV_XS->new ({ binary => 1 });
         $csv->binary (0);
 my $f = $csv->binary;

If this attribute is 1, you may use binary characters in quoted fields, including line feeds, carriage returns and NULL bytes. (The latter could be escaped as "0.) By default this feature is off.

If a string is marked UTF8, binary will be turned on automatically when binary characters other than CR and NL are encountered. Note that a simple string like "\x{00a0}" might still be binary, but not marked UTF8, so setting { binary => 1 } is still a wise option.


 my $csv = Text::CSV_XS->new ({ strict => 1 });
         $csv->strict (0);
 my $f = $csv->strict;

If this attribute is set to 1, any row that parses to a different number of fields than the previous row will cause the parser to throw error 2014.

Empty rows or rows that result in no fields (like comment lines) are exempt from these checks.


 my $csv = Text::CSV_XS->new ({ skip_empty_rows => 1 });
         $csv->skip_empty_rows ("eof");
 my $f = $csv->skip_empty_rows;

This attribute defines the behavior for empty rows: an "eol" immediately following the start of line. Default behavior is to return one single empty field.

This attribute is only used in parsing. This attribute is ineffective when using "parse" and "fields".

Possible values for this attribute are

0 | undef
 my $csv = Text::CSV_XS->new ({ skip_empty_rows => 0 });
 $csv->skip_empty_rows (undef);

No special action is taken. The result will be one single empty field.

1 | "skip"
 my $csv = Text::CSV_XS->new ({ skip_empty_rows => 1 });
 $csv->skip_empty_rows ("skip");

The row will be skipped.

2 | "eof" | "stop"
 my $csv = Text::CSV_XS->new ({ skip_empty_rows => 2 });
 $csv->skip_empty_rows ("eof");

The parsing will stop as if an "eof" was detected.

3 | "die"
 my $csv = Text::CSV_XS->new ({ skip_empty_rows => 3 });
 $csv->skip_empty_rows ("die");

The parsing will stop. The internal error code will be set to 2015 and the parser will die.

4 | "croak"
 my $csv = Text::CSV_XS->new ({ skip_empty_rows => 4 });
 $csv->skip_empty_rows ("croak");

The parsing will stop. The internal error code will be set to 2015 and the parser will croak.

5 | "error"
 my $csv = Text::CSV_XS->new ({ skip_empty_rows => 5 });
 $csv->skip_empty_rows ("error");

The parsing will fail. The internal error code will be set to 2015.

 my $csv = Text::CSV_XS->new ({ skip_empty_rows => sub { [] } });
 $csv->skip_empty_rows (sub { [ 42, $., undef, "empty" ] });

The callback is invoked and its result used instead. If you want the parse to stop after the callback, make sure to return a false value.

The returned value from the callback should be an array-ref. Any other type will cause the parse to stop, so these are equivalent in behavior:

 csv (in => $fh, skip_empty_rows => "stop");
 csv (in => $fh. skip_empty_rows => sub { 0; });

Without arguments, the current value is returned: 0, 1, eof, die, croak or the callback.


Alias for "formula"


 my $csv = Text::CSV_XS->new ({ formula => "none" });
         $csv->formula ("none");
 my $f = $csv->formula;

This defines the behavior of fields containing formulas. As formulas are considered dangerous in spreadsheets, this attribute can define an optional action to be taken if a field starts with an equal sign (=).

For purpose of code-readability, this can also be written as

 my $csv = Text::CSV_XS->new ({ formula_handling => "none" });
         $csv->formula_handling ("none");
 my $f = $csv->formula_handling;

Possible values for this attribute are


Take no specific action. This is the default.

 $csv->formula ("none");

Cause the process to die whenever a leading = is encountered.

 $csv->formula ("die");

Cause the process to croak whenever a leading = is encountered. (See Carp)

 $csv->formula ("croak");

Report position and content of the field whenever a leading = is found. The value of the field is unchanged.

 $csv->formula ("diag");

Replace the content of fields that start with a = with the empty string.

 $csv->formula ("empty");
 $csv->formula ("");

Replace the content of fields that start with a = with undef.

 $csv->formula ("undef");
 $csv->formula (undef);
a callback

Modify the content of fields that start with a = with the return-value of the callback. The original content of the field is available inside the callback as $_;

 # Replace all formula's with 42
 $csv->formula (sub { 42; });

 # same as $csv->formula ("empty") but slower
 $csv->formula (sub { "" });

 # Allow =4+12
 $csv->formula (sub { s/^=(\d+\+\d+)$/$1/eer });

 # Allow more complex calculations
 $csv->formula (sub { eval { s{^=([-+*/0-9()]+)$}{$1}ee }; $_ });

All other values will give a warning and then fallback to diag.


 my $csv = Text::CSV_XS->new ({ decode_utf8 => 1 });
         $csv->decode_utf8 (0);
 my $f = $csv->decode_utf8;

This attributes defaults to TRUE.

While parsing, fields that are valid UTF-8, are automatically set to be UTF-8, so that

  $csv->parse ("\xC4\xA8\n");

results in

  PV("\304\250"\0) [UTF8 "\x{128}"]

Sometimes it might not be a desired action. To prevent those upgrades, set this attribute to false, and the result will be



 my $csv = Text::CSV_XS->new ({ auto_diag => 1 });
         $csv->auto_diag (2);
 my $l = $csv->auto_diag;

Set this attribute to a number between 1 and 9 causes "error_diag" to be automatically called in void context upon errors.

In case of error 2012 - EOF, this call will be void.

If auto_diag is set to a numeric value greater than 1, it will die on errors instead of warn. If set to anything unrecognized, it will be silently ignored.

Future extensions to this feature will include more reliable auto-detection of autodie being active in the scope of which the error occurred which will increment the value of auto_diag with 1 the moment the error is detected.


 my $csv = Text::CSV_XS->new ({ diag_verbose => 1 });
         $csv->diag_verbose (2);
 my $l = $csv->diag_verbose;

Set the verbosity of the output triggered by auto_diag. Currently only adds the current input-record-number (if known) to the diagnostic output with an indication of the position of the error.


 my $csv = Text::CSV_XS->new ({ blank_is_undef => 1 });
         $csv->blank_is_undef (0);
 my $f = $csv->blank_is_undef;

Under normal circumstances, CSV data makes no distinction between quoted- and unquoted empty fields. These both end up in an empty string field once read, thus

 1,"",," ",2

is read as

 ("1", "", "", " ", "2")

When writing CSV files with either always_quote or quote_empty set, the unquoted empty field is the result of an undefined value. To enable this distinction when reading CSV data, the blank_is_undef attribute will cause unquoted empty fields to be set to undef, causing the above to be parsed as

 ("1", "", undef, " ", "2")

Note that this is specifically important when loading CSV fields into a database that allows NULL values, as the perl equivalent for NULL is undef in DBI land.


 my $csv = Text::CSV_XS->new ({ empty_is_undef => 1 });
         $csv->empty_is_undef (0);
 my $f = $csv->empty_is_undef;

Going one step further than blank_is_undef, this attribute converts all empty fields to undef, so

 1,"",," ",2

is read as

 (1, undef, undef, " ", 2)

Note that this affects only fields that are originally empty, not fields that are empty after stripping allowed whitespace. YMMV.


 my $csv = Text::CSV_XS->new ({ allow_whitespace => 1 });
         $csv->allow_whitespace (0);
 my $f = $csv->allow_whitespace;

When this option is set to true, the whitespace (TAB's and SPACE's) surrounding the separation character is removed when parsing. If either TAB or SPACE is one of the three characters sep_char, quote_char, or escape_char it will not be considered whitespace.

Now lines like:

 1 , "foo" , bar , 3 , zapp

are parsed as valid CSV, even though it violates the CSV specs.

Note that all whitespace is stripped from both start and end of each field. That would make it more than a feature to enable parsing bad CSV lines, as

 1,   2.0,  3,   ape  , monkey

will now be parsed as

 ("1", "2.0", "3", "ape", "monkey")

even if the original line was perfectly acceptable CSV.


 my $csv = Text::CSV_XS->new ({ allow_loose_quotes => 1 });
         $csv->allow_loose_quotes (0);
 my $f = $csv->allow_loose_quotes;

By default, parsing unquoted fields containing quote_char characters like

 1,foo "bar" baz,42

would result in parse error 2034. Though it is still bad practice to allow this format, we cannot help the fact that some vendors make their applications spit out lines styled this way.

If there is really bad CSV data, like

 1,"foo "bar" baz",42


 1,""foo bar baz"",42

there is a way to get this data-line parsed and leave the quotes inside the quoted field as-is. This can be achieved by setting allow_loose_quotes AND making sure that the escape_char is not equal to quote_char.


 my $csv = Text::CSV_XS->new ({ allow_loose_escapes => 1 });
         $csv->allow_loose_escapes (0);
 my $f = $csv->allow_loose_escapes;

Parsing fields that have escape_char characters that escape characters that do not need to be escaped, like:

 my $csv = Text::CSV_XS->new ({ escape_char => "\\" });
 $csv->parse (qq{1,"my bar\'s",baz,42});

would result in parse error 2025. Though it is bad practice to allow this format, this attribute enables you to treat all escape character sequences equal.


 my $csv = Text::CSV_XS->new ({ allow_unquoted_escape => 1 });
         $csv->allow_unquoted_escape (0);
 my $f = $csv->allow_unquoted_escape;

A backward compatibility issue where escape_char differs from quote_char prevents escape_char to be in the first position of a field. If quote_char is equal to the default " and escape_char is set to \, this would be illegal:


Setting this attribute to 1 might help to overcome issues with backward compatibility and allow this style.


 my $csv = Text::CSV_XS->new ({ always_quote => 1 });
         $csv->always_quote (0);
 my $f = $csv->always_quote;

By default the generated fields are quoted only if they need to be. For example, if they contain the separator character. If you set this attribute to 1 then all defined fields will be quoted. (undef fields are not quoted, see "blank_is_undef"). This makes it quite often easier to handle exported data in external applications. (Poor creatures who are better to use Text::CSV_XS. :)


 my $csv = Text::CSV_XS->new ({ quote_space => 1 });
         $csv->quote_space (0);
 my $f = $csv->quote_space;

By default, a space in a field would trigger quotation. As no rule exists this to be forced in CSV, nor any for the opposite, the default is true for safety. You can exclude the space from this trigger by setting this attribute to 0.


 my $csv = Text::CSV_XS->new ({ quote_empty => 1 });
         $csv->quote_empty (0);
 my $f = $csv->quote_empty;

By default the generated fields are quoted only if they need to be. An empty (defined) field does not need quotation. If you set this attribute to 1 then empty defined fields will be quoted. (undef fields are not quoted, see "blank_is_undef"). See also always_quote.


 my $csv = Text::CSV_XS->new ({ quote_binary => 1 });
         $csv->quote_binary (0);
 my $f = $csv->quote_binary;

By default, all "unsafe" bytes inside a string cause the combined field to be quoted. By setting this attribute to 0, you can disable that trigger for bytes >= 0x7F.


 my $csv = Text::CSV_XS->new ({ escape_null => 1 });
         $csv->escape_null (0);
 my $f = $csv->escape_null;

By default, a NULL byte in a field would be escaped. This option enables you to treat the NULL byte as a simple binary character in binary mode (the { binary => 1 } is set). The default is true. You can prevent NULL escapes by setting this attribute to 0.

When the escape_char attribute is set to undefined, this attribute will be set to false.

The default setting will encode "=\x00=" as


With escape_null set, this will result in


The default when using the csv function is false.

For backward compatibility reasons, the deprecated old name quote_null is still recognized.


 my $csv = Text::CSV_XS->new ({ keep_meta_info => 1 });
         $csv->keep_meta_info (0);
 my $f = $csv->keep_meta_info;

By default, the parsing of input records is as simple and fast as possible. However, some parsing information - like quotation of the original field - is lost in that process. Setting this flag to true enables retrieving that information after parsing with the methods "meta_info", "is_quoted", and "is_binary" described below. Default is false for performance.

If you set this attribute to a value greater than 9, then you can control output quotation style like it was used in the input of the the last parsed record (unless quotation was added because of other reasons).

 my $csv = Text::CSV_XS->new ({
    binary         => 1,
    keep_meta_info => 1,
    quote_space    => 0,

 my $row = $csv->parse (q{1,,"", ," ",f,"g","h""h",help,"help"});

 $csv->print (*STDOUT, \@row);
 # 1,,, , ,f,g,"h""h",help,help
 $csv->keep_meta_info (11);
 $csv->print (*STDOUT, \@row);
 # 1,,"", ," ",f,"g","h""h",help,"help"


 my $csv = Text::CSV_XS->new ({ undef_str => "\\N" });
         $csv->undef_str (undef);
 my $s = $csv->undef_str;

This attribute optionally defines the output of undefined fields. The value passed is not changed at all, so if it needs quotation, the quotation needs to be included in the value of the attribute. Use with caution, as passing a value like ",",,,,""" will for sure mess up your output. The default for this attribute is undef, meaning no special treatment.

This attribute is useful when exporting CSV data to be imported in custom loaders, like for MySQL, that recognize special sequences for NULL data.

This attribute has no meaning when parsing CSV data.


 my $csv = Text::CSV_XS->new ({ comment_str => "#" });
         $csv->comment_str (undef);
 my $s = $csv->comment_str;

This attribute optionally defines a string to be recognized as comment. If this attribute is defined, all lines starting with this sequence will not be parsed as CSV but skipped as comment.

This attribute has no meaning when generating CSV.

Comment strings that start with any of the special characters/sequences are not supported (so it cannot start with any of "sep_char", "quote_char", "escape_char", "sep", "quote", or "eol").

For convenience, comment is an alias for comment_str.


 my $csv = Text::CSV_XS->new ({ verbatim => 1 });
         $csv->verbatim (0);
 my $f = $csv->verbatim;

This is a quite controversial attribute to set, but makes some hard things possible.

The rationale behind this attribute is to tell the parser that the normally special characters newline (NL) and Carriage Return (CR) will not be special when this flag is set, and be dealt with as being ordinary binary characters. This will ease working with data with embedded newlines.

When verbatim is used with "getline", "getline" auto-chomp's every line.

Imagine a file format like

 M^^Hans^Janssen^Klas 2\n2A^Ja^11-06-2007#\r\n

where, the line ending is a very specific "#\r\n", and the sep_char is a ^ (caret). None of the fields is quoted, but embedded binary data is likely to be present. With the specific line ending, this should not be too hard to detect.

By default, Text::CSV_XS' parse function is instructed to only know about "\n" and "\r" to be legal line endings, and so has to deal with the embedded newline as a real end-of-line, so it can scan the next line if binary is true, and the newline is inside a quoted field. With this option, we tell "parse" to parse the line as if "\n" is just nothing more than a binary character.

For "parse" this means that the parser has no more idea about line ending and "getline" chomps line endings on reading.


A set of column types; the attribute is immediately passed to the "types" method.


See the "Callbacks" section below.


To sum it up,

 $csv = Text::CSV_XS->new ();

is equivalent to

 $csv = Text::CSV_XS->new ({
     eol                   => undef, # \r, \n, or \r\n
     sep_char              => ',',
     sep                   => undef,
     quote_char            => '"',
     quote                 => undef,
     escape_char           => '"',
     binary                => 0,
     decode_utf8           => 1,
     auto_diag             => 0,
     diag_verbose          => 0,
     blank_is_undef        => 0,
     empty_is_undef        => 0,
     allow_whitespace      => 0,
     allow_loose_quotes    => 0,
     allow_loose_escapes   => 0,
     allow_unquoted_escape => 0,
     always_quote          => 0,
     quote_empty           => 0,
     quote_space           => 1,
     escape_null           => 1,
     quote_binary          => 1,
     keep_meta_info        => 0,
     strict                => 0,
     skip_empty_rows       => 0,
     formula               => 0,
     verbatim              => 0,
     undef_str             => undef,
     comment_str           => undef,
     types                 => undef,
     callbacks             => undef,

For all of the above mentioned flags, an accessor method is available where you can inquire the current value, or change the value

 my $quote = $csv->quote_char;
 $csv->binary (1);

It is not wise to change these settings halfway through writing CSV data to a stream. If however you want to create a new stream using the available CSV object, there is no harm in changing them.

If the "new" constructor call fails, it returns undef, and makes the fail reason available through the "error_diag" method.

 $csv = Text::CSV_XS->new ({ ecs_char => 1 }) or
     die "".Text::CSV_XS->error_diag ();

"error_diag" will return a string like

 "INI - Unknown attribute 'ecs_char'"


 @attr = Text::CSV_XS->known_attributes;
 @attr = Text::CSV_XS::known_attributes;
 @attr = $csv->known_attributes;

This method will return an ordered list of all the supported attributes as described above. This can be useful for knowing what attributes are valid in classes that use or extend Text::CSV_XS.


 $status = $csv->print ($fh, $colref);

Similar to "combine" + "string" + "print", but much more efficient. It expects an array ref as input (not an array!) and the resulting string is not really created, but immediately written to the $fh object, typically an IO handle or any other object that offers a "print" method.

For performance reasons print does not create a result string, so all "string", "status", "fields", and "error_input" methods will return undefined information after executing this method.

If $colref is undef (explicit, not through a variable argument) and "bind_columns" was used to specify fields to be printed, it is possible to make performance improvements, as otherwise data would have to be copied as arguments to the method call:

 $csv->bind_columns (\($foo, $bar));
 $status = $csv->print ($fh, undef);

A short benchmark

 my @data = ("aa" .. "zz");
 $csv->bind_columns (\(@data));

 $csv->print ($fh, [ @data ]);   # 11800 recs/sec
 $csv->print ($fh,  \@data  );   # 57600 recs/sec
 $csv->print ($fh,   undef  );   # 48500 recs/sec


 $status = $csv->say ($fh, $colref);

Like print, but eol defaults to $\.

 $csv->print_hr ($fh, $ref);

Provides an easy way to print a $ref (as fetched with "getline_hr") provided the column names are set with "column_names".

It is just a wrapper method with basic parameter checks over

 $csv->print ($fh, [ map { $ref->{$_} } $csv->column_names ]);


 $status = $csv->combine (@fields);

This method constructs a CSV record from @fields, returning success or failure. Failure can result from lack of arguments or an argument that contains an invalid character. Upon success, "string" can be called to retrieve the resultant CSV string. Upon failure, the value returned by "string" is undefined and "error_input" could be called to retrieve the invalid argument.


 $line = $csv->string ();

This method returns the input to "parse" or the resultant CSV string of "combine", whichever was called more recently.


 $colref = $csv->getline ($fh);

This is the counterpart to "print", as "parse" is the counterpart to "combine": it parses a row from the $fh handle using the "getline" method associated with $fh and parses this row into an array ref. This array ref is returned by the function or undef for failure. When $fh does not support getline, you are likely to hit errors.

When fields are bound with "bind_columns" the return value is a reference to an empty list.

The "string", "fields", and "status" methods are meaningless again.


 $arrayref = $csv->getline_all ($fh);
 $arrayref = $csv->getline_all ($fh, $offset);
 $arrayref = $csv->getline_all ($fh, $offset, $length);

This will return a reference to a list of getline ($fh) results. In this call, keep_meta_info is disabled. If $offset is negative, as with splice, only the last abs ($offset) records of $fh are taken into consideration. Parameters $offset and $length are expected to be integers. Non-integer values are interpreted as integer without check.

Given a CSV file with 10 lines:

 lines call
 ----- ---------------------------------------------------------
 0..9  $csv->getline_all ($fh)         # all
 0..9  $csv->getline_all ($fh,  0)     # all
 8..9  $csv->getline_all ($fh,  8)     # start at 8
 -     $csv->getline_all ($fh,  0,  0) # start at 0 first 0 rows
 0..4  $csv->getline_all ($fh,  0,  5) # start at 0 first 5 rows
 4..5  $csv->getline_all ($fh,  4,  2) # start at 4 first 2 rows
 8..9  $csv->getline_all ($fh, -2)     # last 2 rows
 6..7  $csv->getline_all ($fh, -4,  2) # first 2 of last  4 rows


The "getline_hr" and "column_names" methods work together to allow you to have rows returned as hashrefs. You must call "column_names" first to declare your column names.

 $csv->column_names (qw( code name price description ));
 $hr = $csv->getline_hr ($fh);
 print "Price for $hr->{name} is $hr->{price} EUR\n";

"getline_hr" will croak if called before "column_names".

Note that "getline_hr" creates a hashref for every row and will be much slower than the combined use of "bind_columns" and "getline" but still offering the same easy to use hashref inside the loop:

 my @cols = @{$csv->getline ($fh)};
 $csv->column_names (@cols);
 while (my $row = $csv->getline_hr ($fh)) {
     print $row->{price};

Could easily be rewritten to the much faster:

 my @cols = @{$csv->getline ($fh)};
 my $row = {};
 $csv->bind_columns (\@{$row}{@cols});
 while ($csv->getline ($fh)) {
     print $row->{price};

Your mileage may vary for the size of the data and the number of rows. With perl-5.14.2 the comparison for a 100_000 line file with 14 columns:

            Rate hashrefs getlines
 hashrefs 1.00/s       --     -76%
 getlines 4.15/s     313%       --


 $arrayref = $csv->getline_hr_all ($fh);
 $arrayref = $csv->getline_hr_all ($fh, $offset);
 $arrayref = $csv->getline_hr_all ($fh, $offset, $length);

This will return a reference to a list of getline_hr ($fh) results. In this call, keep_meta_info is disabled.


 $status = $csv->parse ($line);

This method decomposes a CSV string into fields, returning success or failure. Failure can result from a lack of argument or the given CSV string is improperly formatted. Upon success, "fields" can be called to retrieve the decomposed fields. Upon failure calling "fields" will return undefined data and "error_input" can be called to retrieve the invalid argument.

You may use the "types" method for setting column types. See "types"' description below.

The $line argument is supposed to be a simple scalar. Everything else is supposed to croak and set error 1500.


This function tries to implement RFC7111 (URI Fragment Identifiers for the text/csv Media Type) -

 my $AoA = $csv->fragment ($fh, $spec);

In specifications, * is used to specify the last item, a dash (-) to indicate a range. All indices are 1-based: the first row or column has index 1. Selections can be combined with the semi-colon (;).

When using this method in combination with "column_names", the returned reference will point to a list of hashes instead of a list of lists. A disjointed cell-based combined selection might return rows with different number of columns making the use of hashes unpredictable.

 $csv->column_names ("Name", "Age");
 my $AoH = $csv->fragment ($fh, "col=3;8");

If the "after_parse" callback is active, it is also called on every line parsed and skipped before the fragment.


In cell-based selection, the comma (,) is used to pair row and column


The range operator (-) using cells can be used to define top-left and bottom-right cell location


The * is only allowed in the second part of a pair

 cell=3,2-*,2    # row 3 till end, only column 2
 cell=3,2-3,*    # column 2 till end, only row 3
 cell=3,2-*,*    # strip row 1 and 2, and column 1

Cells and cell ranges may be combined with ;, possibly resulting in rows with different numbers of columns


Disjointed selections will only return selected cells. The cells that are not specified will not be included in the returned set, not even as undef. As an example given a CSV like

 :            :

with cell=1,1-2,2;3,3-4,4;1,4;4,1 will return:


Overlapping cell-specs will return those cells only once, So cell=1,1-3,3;2,2-4,4;2,3;4,2 will return:


RFC7111 does not allow different types of specs to be combined (either row or col or cell). Passing an invalid fragment specification will croak and set error 2013.


Set the "keys" that will be used in the "getline_hr" calls. If no keys (column names) are passed, it will return the current setting as a list.

"column_names" accepts a list of scalars (the column names) or a single array_ref, so you can pass the return value from "getline" too:

 $csv->column_names ($csv->getline ($fh));

"column_names" does no checking on duplicates at all, which might lead to unexpected results. Undefined entries will be replaced with the string "\cAUNDEF\cA", so

 $csv->column_names (undef, "", "name", "name");
 $hr = $csv->getline_hr ($fh);

will set $hr->{"\cAUNDEF\cA"} to the 1st field, $hr->{""} to the 2nd field, and $hr->{name} to the 4th field, discarding the 3rd field.

"column_names" croaks on invalid arguments.

This method does NOT work in perl-5.6.x

Parse the CSV header and set sep, column_names and encoding.

 my @hdr = $csv->header ($fh);
 $csv->header ($fh, { sep_set => [ ";", ",", "|", "\t" ] });
 $csv->header ($fh, { detect_bom => 1, munge_column_names => "lc" });

The first argument should be a file handle.

This method resets some object properties, as it is supposed to be invoked only once per file or stream. It will leave attributes column_names and bound_columns alone if setting column names is disabled. Reading headers on previously process objects might fail on perl-5.8.0 and older.

Assuming that the file opened for parsing has a header, and the header does not contain problematic characters like embedded newlines, read the first line from the open handle then auto-detect whether the header separates the column names with a character from the allowed separator list.

If any of the allowed separators matches, and none of the other allowed separators match, set sep to that separator for the current CSV_XS instance and use it to parse the first line, map those to lowercase, and use that to set the instance "column_names":

 my $csv = Text::CSV_XS->new ({ binary => 1, auto_diag => 1 });
 open my $fh, "<", "file.csv";
 binmode $fh; # for Windows
 $csv->header ($fh);
 while (my $row = $csv->getline_hr ($fh)) {

If the header is empty, contains more than one unique separator out of the allowed set, contains empty fields, or contains identical fields (after folding), it will croak with error 1010, 1011, 1012, or 1013 respectively.

If the header contains embedded newlines or is not valid CSV in any other way, this method will croak and leave the parse error untouched.

A successful call to header will always set the sep of the $csv object. This behavior can not be disabled.

return value

On error this method will croak.

In list context, the headers will be returned whether they are used to set "column_names" or not.

In scalar context, the instance itself is returned. Note: the values as found in the header will effectively be lost if set_column_names is false.


 $csv->header ($fh, { sep_set => [ ";", ",", "|", "\t" ] });

The list of legal separators defaults to [ ";", "," ] and can be changed by this option. As this is probably the most often used option, it can be passed on its own as an unnamed argument:

 $csv->header ($fh, [ ";", ",", "|", "\t", "::", "\x{2063}" ]);

Multi-byte sequences are allowed, both multi-character and Unicode. See sep.

 $csv->header ($fh, { detect_bom => 1 });

The default behavior is to detect if the header line starts with a BOM. If the header has a BOM, use that to set the encoding of $fh. This default behavior can be disabled by passing a false value to detect_bom.

Supported encodings from BOM are: UTF-8, UTF-16BE, UTF-16LE, UTF-32BE, and UTF-32LE. BOM also supports UTF-1, UTF-EBCDIC, SCSU, BOCU-1, and GB-18030 but Encode does not (yet). UTF-7 is not supported.

If a supported BOM was detected as start of the stream, it is stored in the object attribute ENCODING.

 my $enc = $csv->{ENCODING};

The encoding is used with binmode on $fh.

If the handle was opened in a (correct) encoding, this method will not alter the encoding, as it checks the leading bytes of the first line. In case the stream starts with a decoded BOM (U+FEFF), {ENCODING} will be "" (empty) instead of the default undef.


This option offers the means to modify the column names into something that is most useful to the application. The default is to map all column names to lower case.

 $csv->header ($fh, { munge_column_names => "lc" });

The following values are available:

  lc     - lower case
  uc     - upper case
  db     - valid DB field names
  none   - do not change
  \%hash - supply a mapping
  \&cb   - supply a callback
Lower case
 $csv->header ($fh, { munge_column_names => "lc" });

The header is changed to all lower-case

 $_ = lc;
Upper case
 $csv->header ($fh, { munge_column_names => "uc" });

The header is changed to all upper-case

 $_ = uc;
 $csv->header ($fh, { munge_column_names => "none" });
 $csv->header ($fh, { munge_column_names => { foo => "sombrero" });

if a value does not exist, the original value is used unchanged

 $csv->header ($fh, { munge_column_names => "db" });



all sequences of non-word characters are replaced with an underscore


all leading underscores are removed

 $_ = lc (s/\W+/_/gr =~ s/^_+//r);
 $csv->header ($fh, { munge_column_names => sub { fc } });
 $csv->header ($fh, { munge_column_names => sub { "column_".$col++ } });
 $csv->header ($fh, { munge_column_names => sub { lc (s/\W+/_/gr) } });

As this callback is called in a map, you can use $_ directly.

 $csv->header ($fh, { set_column_names => 1 });

The default is to set the instances column names using "column_names" if the method is successful, so subsequent calls to "getline_hr" can return a hash. Disable setting the header can be forced by using a false value for this option.

As described in "return value" above, content is lost in scalar context.


When receiving CSV files from external sources, this method can be used to protect against changes in the layout by restricting to known headers (and typos in the header fields).

 my %known = (
     "record key" => "c_rec",
     "rec id"     => "c_rec",
     "id_rec"     => "c_rec",
     "kode"       => "code",
     "code"       => "code",
     "vaule"      => "value",
     "value"      => "value",
 my $csv = Text::CSV_XS->new ({ binary => 1, auto_diag => 1 });
 open my $fh, "<", $source or die "$source: $!";
 $csv->header ($fh, { munge_column_names => sub {
     $known{lc $_} or die "Unknown column '$_' in $source";
 while (my $row = $csv->getline_hr ($fh)) {
     say join "\t", $row->{c_rec}, $row->{code}, $row->{value};


Takes a list of scalar references to be used for output with "print" or to store in the fields fetched by "getline". When you do not pass enough references to store the fetched fields in, "getline" will fail with error 3006. If you pass more than there are fields to return, the content of the remaining references is left untouched. Under strict the two should match, otherwise "getline" will fail with error 2014.

 $csv->bind_columns (\$code, \$name, \$price, \$description);
 while ($csv->getline ($fh)) {
     print "The price of a $name is \x{20ac} $price\n";

To reset or clear all column binding, call "bind_columns" with the single argument undef. This will also clear column names.

 $csv->bind_columns (undef);

If no arguments are passed at all, "bind_columns" will return the list of current bindings or undef if no binds are active.

Note that in parsing with bind_columns, the fields are set on the fly. That implies that if the third field of a row causes an error (or this row has just two fields where the previous row had more), the first two fields already have been assigned the values of the current row, while the rest of the fields will still hold the values of the previous row. If you want the parser to fail in these cases, use the strict attribute.


 $eof = $csv->eof ();

If "parse" or "getline" was used with an IO stream, this method will return true (1) if the last call hit end of file, otherwise it will return false (''). This is useful to see the difference between a failure and end of file.

Note that if the parsing of the last line caused an error, eof is still true. That means that if you are not using "auto_diag", an idiom like

 while (my $row = $csv->getline ($fh)) {
     # ...
 $csv->eof or $csv->error_diag;

will not report the error. You would have to change that to

 while (my $row = $csv->getline ($fh)) {
     # ...
 +$csv->error_diag and $csv->error_diag;


 $csv->types (\@tref);

This method is used to force that (all) columns are of a given type. For example, if you have an integer column, two columns with doubles and a string column, then you might do a

 $csv->types ([Text::CSV_XS::IV (),
               Text::CSV_XS::NV (),
               Text::CSV_XS::NV (),
               Text::CSV_XS::PV ()]);

Column types are used only for decoding columns while parsing, in other words by the "parse" and "getline" methods.

You can unset column types by doing a

 $csv->types (undef);

or fetch the current type settings with

 $types = $csv->types ();

Set field type to integer.


Set field type to numeric/float.


Set field type to string.


 @columns = $csv->fields ();

This method returns the input to "combine" or the resultant decomposed fields of a successful "parse", whichever was called more recently.

Note that the return value is undefined after using "getline", which does not fill the data structures returned by "parse".


 @flags = $csv->meta_info ();

This method returns the "flags" of the input to "combine" or the flags of the resultant decomposed fields of "parse", whichever was called more recently.

For each field, a meta_info field will hold flags that inform something about the field returned by the "fields" method or passed to the "combine" method. The flags are bit-wise-or'd like:


The field was quoted.


The field was binary.


The field was invalid.

Currently only used when allow_loose_quotes is active.


The field was missing.

See the is_*** methods below.


 my $quoted = $csv->is_quoted ($column_idx);

where $column_idx is the (zero-based) index of the column in the last result of "parse".

This returns a true value if the data in the indicated column was enclosed in quote_char quotes. This might be important for fields where content ,20070108, is to be treated as a numeric value, and where ,"20070108", is explicitly marked as character string data.

This method is only valid when "keep_meta_info" is set to a true value.


 my $binary = $csv->is_binary ($column_idx);

where $column_idx is the (zero-based) index of the column in the last result of "parse".

This returns a true value if the data in the indicated column contained any byte in the range [\x00-\x08,\x10-\x1F,\x7F-\xFF].

This method is only valid when "keep_meta_info" is set to a true value.


 my $missing = $csv->is_missing ($column_idx);

where $column_idx is the (zero-based) index of the column in the last result of "getline_hr".

 $csv->keep_meta_info (1);
 while (my $hr = $csv->getline_hr ($fh)) {
     $csv->is_missing (0) and next; # This was an empty line

When using "getline_hr", it is impossible to tell if the parsed fields are undef because they where not filled in the CSV stream or because they were not read at all, as all the fields defined by "column_names" are set in the hash-ref. If you still need to know if all fields in each row are provided, you should enable keep_meta_info so you can check the flags.

If keep_meta_info is false, is_missing will always return undef, regardless of $column_idx being valid or not. If this attribute is true it will return either 0 (the field is present) or 1 (the field is missing).

A special case is the empty line. If the line is completely empty - after dealing with the flags - this is still a valid CSV line: it is a record of just one single empty field. However, if keep_meta_info is set, invoking is_missing with index 0 will now return true.


 $status = $csv->status ();

This method returns the status of the last invoked "combine" or "parse" call. Status is success (true: 1) or failure (false: undef or 0).

Note that as this only keeps track of the status of above mentioned methods, you are probably looking for error_diag instead.


 $bad_argument = $csv->error_input ();

This method returns the erroneous argument (if it exists) of "combine" or "parse", whichever was called more recently. If the last invocation was successful, error_input will return undef.

Depending on the type of error, it might also hold the data for the last error-input of "getline".


 Text::CSV_XS->error_diag ();
 $csv->error_diag ();
 $error_code               = 0  + $csv->error_diag ();
 $error_str                = "" . $csv->error_diag ();
 ($cde, $str, $pos, $rec, $fld) = $csv->error_diag ();

If (and only if) an error occurred, this function returns the diagnostics of that error.

If called in void context, this will print the internal error code and the associated error message to STDERR.

If called in list context, this will return the error code and the error message in that order. If the last error was from parsing, the rest of the values returned are a best guess at the location within the line that was being parsed. Their values are 1-based. The position currently is index of the byte at which the parsing failed in the current record. It might change to be the index of the current character in a later release. The records is the index of the record parsed by the csv instance. The field number is the index of the field the parser thinks it is currently trying to parse. See examples/csv-check for how this can be used.

If called in scalar context, it will return the diagnostics in a single scalar, a-la $!. It will contain the error code in numeric context, and the diagnostics message in string context.

When called as a class method or a direct function call, the diagnostics are that of the last "new" call.


Note: This is an internal function only, and output cannot be relied upon. Use at own risk.

If debugging beyond what "error_diag" is able to show, the internal cache can be shown with this function.

 # Something failed ..
 $csv->_cache_diag ();


 $recno = $csv->record_number ();

Returns the records parsed by this csv instance. This value should be more accurate than $. when embedded newlines come in play. Records written by this instance are not counted.


 $csv->SetDiag (0);

Use to reset the diagnostics if you are dealing with errors.


By default none of these are exported.

 use Text::CSV_XS qw( csv );

Import the function "csv" function. See below.

 use Text::CSV_XS qw( :CONSTANTS );

Import module constants "CSV_FLAGS_IS_QUOTED", "CSV_FLAGS_IS_BINARY", "CSV_FLAGS_ERROR_IN_FIELD", "CSV_FLAGS_IS_MISSING", "CSV_TYPE_PV", "CSV_TYPE_IV", and "CSV_TYPE_NV". Each can be imported alone




This function is not exported by default and should be explicitly requested:

 use Text::CSV_XS qw( csv );

This is a high-level function that aims at simple (user) interfaces. This can be used to read/parse a CSV file or stream (the default behavior) or to produce a file or write to a stream (define the out attribute). It returns an array- or hash-reference on parsing (or undef on fail) or the numeric value of "error_diag" on writing. When this function fails you can get to the error using the class call to "error_diag"

 my $aoa = csv (in => "test.csv") or
     die Text::CSV_XS->error_diag;

This function takes the arguments as key-value pairs. This can be passed as a list or as an anonymous hash:

 my $aoa = csv (  in => "test.csv", sep_char => ";");
 my $aoh = csv ({ in => $fh, headers => "auto" });

The arguments passed consist of two parts: the arguments to "csv" itself and the optional attributes to the CSV object used inside the function as enumerated and explained in "new".

If not overridden, the default option used for CSV is

 auto_diag   => 1
 escape_null => 0

The option that is always set and cannot be altered is

 binary      => 1

As this function will likely be used in one-liners, it allows quote to be abbreviated as quo, and escape_char to be abbreviated as esc or escape.

Alternative invocations:

 my $aoa = Text::CSV_XS::csv (in => "file.csv");

 my $csv = Text::CSV_XS->new ();
 my $aoa = $csv->csv (in => "file.csv");

In the latter case, the object attributes are used from the existing object and the attribute arguments in the function call are ignored:

 my $csv = Text::CSV_XS->new ({ sep_char => ";" });
 my $aoh = $csv->csv (in => "file.csv", bom => 1);

will parse using ; as sep_char, not ,.


Used to specify the source. in can be a file name (e.g. "file.csv"), which will be opened for reading and closed when finished, a file handle (e.g. $fh or FH), a reference to a glob (e.g. \*ARGV), the glob itself (e.g. *STDIN), or a reference to a scalar (e.g. \q{1,2,"csv"}).

When used with "out", in should be a reference to a CSV structure (AoA or AoH) or a CODE-ref that returns an array-reference or a hash-reference. The code-ref will be invoked with no arguments.

 my $aoa = csv (in => "file.csv");

 open my $fh, "<", "file.csv";
 my $aoa = csv (in => $fh);

 my $csv = [ [qw( Foo Bar )], [ 1, 2 ], [ 2, 3 ]];
 my $err = csv (in => $csv, out => "file.csv");

If called in void context without the "out" attribute, the resulting ref will be used as input to a subsequent call to csv:

 csv (in => "file.csv", filter => { 2 => sub { length > 2 }})

will be a shortcut to

 csv (in => csv (in => "file.csv", filter => { 2 => sub { length > 2 }}))

where, in the absence of the out attribute, this is a shortcut to

 csv (in  => csv (in => "file.csv", filter => { 2 => sub { length > 2 }}),
      out => *STDOUT)


 csv (in => $aoa, out => "file.csv");
 csv (in => $aoa, out => $fh);
 csv (in => $aoa, out =>   STDOUT);
 csv (in => $aoa, out =>  *STDOUT);
 csv (in => $aoa, out => \*STDOUT);
 csv (in => $aoa, out => \my $data);
 csv (in => $aoa, out =>  undef);
 csv (in => $aoa, out => \"skip");

 csv (in => $fh,  out => \@aoa);
 csv (in => $fh,  out => \@aoh, bom => 1);
 csv (in => $fh,  out => \%hsh, key => "key");

In output mode, the default CSV options when producing CSV are

 eol       => "\r\n"

The "fragment" attribute is ignored in output mode.

out can be a file name (e.g. "file.csv"), which will be opened for writing and closed when finished, a file handle (e.g. $fh or FH), a reference to a glob (e.g. \*STDOUT), the glob itself (e.g. *STDOUT), or a reference to a scalar (e.g. \my $data).

 csv (in => sub { $sth->fetch },            out => "dump.csv");
 csv (in => sub { $sth->fetchrow_hashref }, out => "dump.csv",
      headers => $sth->{NAME_lc});

When a code-ref is used for in, the output is generated per invocation, so no buffering is involved. This implies that there is no size restriction on the number of records. The csv function ends when the coderef returns a false value.

If out is set to a reference of the literal string "skip", the output will be suppressed completely, which might be useful in combination with a filter for side effects only.

 my %cache;
 csv (in    => "dump.csv",
      out   => \"skip",
      on_in => sub { $cache{$_[1][1]}++ });

Currently, setting out to any false value (undef, "", 0) will be equivalent to \"skip".

If the in argument point to something to parse, and the out is set to a reference to an ARRAY or a HASH, the output is appended to the data in the existing reference. The result of the parse should match what exists in the reference passed. This might come handy when you have to parse a set of files with similar content (like data stored per period) and you want to collect that into a single data structure:

 my %hash;
 csv (in => $_, out => \%hash, key => "id") for sort glob "foo-[0-9]*.csv";

 my @list; # List of arrays
 csv (in => $_, out => \@list)              for sort glob "foo-[0-9]*.csv";

 my @list; # List of hashes
 csv (in => $_, out => \@list, bom => 1)    for sort glob "foo-[0-9]*.csv";


If passed, it should be an encoding accepted by the :encoding() option to open. There is no default value. This attribute does not work in perl 5.6.x. encoding can be abbreviated to enc for ease of use in command line invocations.

If encoding is set to the literal value "auto", the method "header" will be invoked on the opened stream to check if there is a BOM and set the encoding accordingly. This is equal to passing a true value in the option detect_bom.

Encodings can be stacked, as supported by binmode:

 # Using PerlIO::via::gzip
 csv (in       => \@csv,
      out      => "test.csv:via.gz",
      encoding => ":via(gzip):encoding(utf-8)",
 $aoa = csv (in => "test.csv:via.gz",  encoding => ":via(gzip)");

 # Using PerlIO::gzip
 csv (in       => \@csv,
      out      => "test.csv:via.gz",
      encoding => ":gzip:encoding(utf-8)",
 $aoa = csv (in => "test.csv:gzip.gz", encoding => ":gzip");


If detect_bom is given, the method "header" will be invoked on the opened stream to check if there is a BOM and set the encoding accordingly.

detect_bom can be abbreviated to bom.

This is the same as setting encoding to "auto".

Note that as the method "header" is invoked, its default is to also set the headers.


If this attribute is not given, the default behavior is to produce an array of arrays.

If headers is supplied, it should be an anonymous list of column names, an anonymous hashref, a coderef, or a literal flag: auto, lc, uc, or skip.


When skip is used, the header will not be included in the output.

 my $aoa = csv (in => $fh, headers => "skip");

skip is invalid/ignored in combinations with detect_bom.


If auto is used, the first line of the CSV source will be read as the list of field headers and used to produce an array of hashes.

 my $aoh = csv (in => $fh, headers => "auto");

If lc is used, the first line of the CSV source will be read as the list of field headers mapped to lower case and used to produce an array of hashes. This is a variation of auto.

 my $aoh = csv (in => $fh, headers => "lc");

If uc is used, the first line of the CSV source will be read as the list of field headers mapped to upper case and used to produce an array of hashes. This is a variation of auto.

 my $aoh = csv (in => $fh, headers => "uc");

If a coderef is used, the first line of the CSV source will be read as the list of mangled field headers in which each field is passed as the only argument to the coderef. This list is used to produce an array of hashes.

 my $aoh = csv (in      => $fh,
                headers => sub { lc ($_[0]) =~ s/kode/code/gr });

this example is a variation of using lc where all occurrences of kode are replaced with code.


If headers is an anonymous list, the entries in the list will be used as field names. The first line is considered data instead of headers.

 my $aoh = csv (in => $fh, headers => [qw( Foo Bar )]);
 csv (in => $aoa, out => $fh, headers => [qw( code description price )]);

If headers is a hash reference, this implies auto, but header fields that exist as key in the hashref will be replaced by the value for that key. Given a CSV file like

 post-kode,city,name,id number,fubble


 csv (headers => { "post-kode" => "pc", "id number" => "ID" }, ...

will return an entry like

 { pc     => "1234AA",
   city   => "Duckstad",
   name   => "Donald",
   ID     => "13",
   fubble => "X313DF",

See also munge_column_names and set_column_names.


If munge_column_names is set, the method "header" is invoked on the opened stream with all matching arguments to detect and set the headers.

munge_column_names can be abbreviated to munge.


If passed, will default headers to "auto" and return a hashref instead of an array of hashes. Allowed values are simple scalars or array-references where the first element is the joiner and the rest are the fields to join to combine the key.

 my $ref = csv (in => "test.csv", key => "code");
 my $ref = csv (in => "test.csv", key => [ ":" => "code", "color" ]);

with test.csv like


the first example will return

  { 1   => {
        code    => 1,
        color   => 'gray',
        price   => 850,
        product => 'pc'
    2   => {
        code    => 2,
        color   => 'white',
        price   => 12,
        product => 'keyboard'
    3   => {
        code    => 3,
        color   => 'black',
        price   => 5,
        product => 'mouse'

the second example will return

  { "1:gray"    => {
        code    => 1,
        color   => 'gray',
        price   => 850,
        product => 'pc'
    "2:white"   => {
        code    => 2,
        color   => 'white',
        price   => 12,
        product => 'keyboard'
    "3:black"   => {
        code    => 3,
        color   => 'black',
        price   => 5,
        product => 'mouse'

The key attribute can be combined with headers for CSV date that has no header line, like

 my $ref = csv (
     in      => "foo.csv",
     headers => [qw( c_foo foo bar description stock )],
     key     =>     "c_foo",


Used to create key-value hashes.

Only allowed when key is valid. A value can be either a single column label or an anonymous list of column labels. In the first case, the value will be a simple scalar value, in the latter case, it will be a hashref.

 my $ref = csv (in => "test.csv", key   => "code",
                                  value => "price");
 my $ref = csv (in => "test.csv", key   => "code",
                                  value => [ "product", "price" ]);
 my $ref = csv (in => "test.csv", key   => [ ":" => "code", "color" ],
                                  value => "price");
 my $ref = csv (in => "test.csv", key   => [ ":" => "code", "color" ],
                                  value => [ "product", "price" ]);

with test.csv like


the first example will return

  { 1 => 850,
    2 =>  12,
    3 =>   5,

the second example will return

  { 1   => {
        price   => 850,
        product => 'pc'
    2   => {
        price   => 12,
        product => 'keyboard'
    3   => {
        price   => 5,
        product => 'mouse'

the third example will return

  { "1:gray"    => 850,
    "2:white"   =>  12,
    "3:black"   =>   5,

the fourth example will return

  { "1:gray"    => {
        price   => 850,
        product => 'pc'
    "2:white"   => {
        price   => 12,
        product => 'keyboard'
    "3:black"   => {
        price   => 5,
        product => 'mouse'


When using hashes, keep the column names into the arrayref passed, so all headers are available after the call in the original order.

 my $aoh = csv (in => "file.csv", keep_headers => \my @hdr);

This attribute can be abbreviated to kh or passed as keep_column_names.

This attribute implies a default of auto for the headers attribute.

The headers can also be kept internally to keep stable header order:

 csv (in      => csv (in => "file.csv", kh => "internal"),
      out     => "new.csv",
      kh      => "internal");

where internal can also be 1, yes, or true. This is similar to

 my @h;
 csv (in      => csv (in => "file.csv", kh => \@h),
      out     => "new.csv",
      headers => \@h);


Only output the fragment as defined in the "fragment" method. This option is ignored when generating CSV. See "out".

Combining all of them could give something like

 use Text::CSV_XS qw( csv );
 my $aoh = csv (
     in       => "test.txt",
     encoding => "utf-8",
     headers  => "auto",
     sep_char => "|",
     fragment => "row=3;6-9;15-*",
 say $aoh->[15]{Foo};


If sep_set is set, the method "header" is invoked on the opened stream to detect and set sep_char with the given set.

sep_set can be abbreviated to seps. If neither sep_set not seps is given, but sep is defined, sep_set defaults to [ sep ]. This is only supported for perl version 5.10 and up.

Note that as the "header" method is invoked, its default is to also set the headers.


If set_column_names is passed, the method "header" is invoked on the opened stream with all arguments meant for "header".

If set_column_names is passed as a false value, the content of the first row is only preserved if the output is AoA:

With an input-file like


This call

 my $aoa = csv (in => $file, set_column_names => 0);

will result in

 [[ "bar", "foo"     ],
  [ "1",   "2"       ],
  [ "3",   "4",  "5" ]]


 my $aoa = csv (in => $file, set_column_names => 0, munge => "none");

will result in

 [[ "bAr", "foo"     ],
  [ "1",   "2"       ],
  [ "3",   "4",  "5" ]]


The function "csv" can also be called as a method or with an existing Text::CSV_XS object. This could help if the function is to be invoked a lot of times and the overhead of creating the object internally over and over again would be prevented by passing an existing instance.

 my $csv = Text::CSV_XS->new ({ binary => 1, auto_diag => 1 });

 my $aoa = $csv->csv (in => $fh);
 my $aoa = csv (in => $fh, csv => $csv);

both act the same. Running this 20000 times on a 20 lines CSV file, showed a 53% speedup.


Callbacks enable actions triggered from the inside of Text::CSV_XS.

While most of what this enables can easily be done in an unrolled loop as described in the "SYNOPSIS" callbacks can be used to meet special demands or enhance the "csv" function.

 $csv->callbacks (error => sub { $csv->SetDiag (0) });

the error callback is invoked when an error occurs, but only when "auto_diag" is set to a true value. A callback is invoked with the values returned by "error_diag":

 my ($c, $s);

 sub ignore3006 {
     my ($err, $msg, $pos, $recno, $fldno) = @_;
     if ($err == 3006) {
         # ignore this error
         ($c, $s) = (undef, undef);
         Text::CSV_XS->SetDiag (0);
     # Any other error
     } # ignore3006

 $csv->callbacks (error => \&ignore3006);
 $csv->bind_columns (\$c, \$s);
 while ($csv->getline ($fh)) {
     # Error 3006 will not stop the loop
 $csv->callbacks (after_parse => sub { push @{$_[1]}, "NEW" });
 while (my $row = $csv->getline ($fh)) {
     $row->[-1] eq "NEW";

This callback is invoked after parsing with "getline" only if no error occurred. The callback is invoked with two arguments: the current CSV parser object and an array reference to the fields parsed.

The return code of the callback is ignored unless it is a reference to the string "skip", in which case the record will be skipped in "getline_all".

 sub add_from_db {
     my ($csv, $row) = @_;
     $sth->execute ($row->[4]);
     push @$row, $sth->fetchrow_array;
     } # add_from_db

 my $aoa = csv (in => "file.csv", callbacks => {
     after_parse => \&add_from_db });

This hook can be used for validation:


Die if any of the records does not validate a rule:

 after_parse => sub {
     $_[1][4] =~ m/^[0-9]{4}\s?[A-Z]{2}$/ or
         die "5th field does not have a valid Dutch zipcode";

Replace invalid fields with a default value:

 after_parse => sub { $_[1][2] =~ m/^\d+$/ or $_[1][2] = 0 }

Skip records that have invalid fields (only applies to "getline_all"):

 after_parse => sub { $_[1][0] =~ m/^\d+$/ or return \"skip"; }
 my $idx = 1;
 $csv->callbacks (before_print => sub { $_[1][0] = $idx++ });
 $csv->print (*STDOUT, [ 0, $_ ]) for @members;

This callback is invoked before printing with "print" only if no error occurred. The callback is invoked with two arguments: the current CSV parser object and an array reference to the fields passed.

The return code of the callback is ignored.

 sub max_4_fields {
     my ($csv, $row) = @_;
     @$row > 4 and splice @$row, 4;
     } # max_4_fields

 csv (in => csv (in => "file.csv"), out => *STDOUT,
     callbacks => { before_print => \&max_4_fields });

This callback is not active for "combine".

Callbacks for csv ()

The "csv" allows for some callbacks that do not integrate in XS internals but only feature the "csv" function.

  csv (in        => "file.csv",
       callbacks => {
           filter       => { 6 => sub { $_ > 15 } },    # first
           after_parse  => sub { say "AFTER PARSE";  }, # first
           after_in     => sub { say "AFTER IN";     }, # second
           on_in        => sub { say "ON IN";        }, # third

  csv (in        => $aoh,
       out       => "file.csv",
       callbacks => {
           on_in        => sub { say "ON IN";        }, # first
           before_out   => sub { say "BEFORE OUT";   }, # second
           before_print => sub { say "BEFORE PRINT"; }, # third

This callback can be used to filter records. It is called just after a new record has been scanned. The callback accepts a:


The keys are the index to the row (the field name or field number, 1-based) and the values are subs to return a true or false value.

 csv (in => "file.csv", filter => {
            3 => sub { m/a/ },       # third field should contain an "a"
            5 => sub { length > 4 }, # length of the 5th field minimal 5

 csv (in => "file.csv", filter => { foo => sub { $_ > 4 }});

If the keys to the filter hash contain any character that is not a digit it will also implicitly set "headers" to "auto" unless "headers" was already passed as argument. When headers are active, returning an array of hashes, the filter is not applicable to the header itself.

All sub results should match, as in AND.

The context of the callback sets $_ localized to the field indicated by the filter. The two arguments are as with all other callbacks, so the other fields in the current row can be seen:

 filter => { 3 => sub { $_ > 100 ? $_[1][1] =~ m/A/ : $_[1][6] =~ m/B/ }}

If the context is set to return a list of hashes ("headers" is defined), the current record will also be available in the localized %_:

 filter => { 3 => sub { $_ > 100 && $_{foo} =~ m/A/ && $_{bar} < 1000  }}

If the filter is used to alter the content by changing $_, make sure that the sub returns true in order not to have that record skipped:

 filter => { 2 => sub { $_ = uc }}

will upper-case the second field, and then skip it if the resulting content evaluates to false. To always accept, end with truth:

 filter => { 2 => sub { $_ = uc; 1 }}
 csv (in => "file.csv", filter => sub { $n++; 0; });

If the argument to filter is a coderef, it is an alias or shortcut to a filter on column 0:

 csv (filter => sub { $n++; 0 });

is equal to

 csv (filter => { 0 => sub { $n++; 0 });
 csv (in => "file.csv", filter => "not_blank");
 csv (in => "file.csv", filter => "not_empty");
 csv (in => "file.csv", filter => "filled");

These are predefined filters

Given a file like (line numbers prefixed for doc purpose only):

 6:, ,
 8:" "

Filter out the blank lines

This filter is a shortcut for

 filter => { 0 => sub { @{$_[1]} > 1 or
             defined $_[1][0] && $_[1][0] ne "" } }

Due to the implementation, it is currently impossible to also filter lines that consists only of a quoted empty field. These lines are also considered blank lines.

With the given example, lines 2 and 4 will be skipped.


Filter out lines where all the fields are empty.

This filter is a shortcut for

 filter => { 0 => sub { grep { defined && $_ ne "" } @{$_[1]} } }

A space is not regarded being empty, so given the example data, lines 2, 3, 4, 5, and 7 are skipped.


Filter out lines that have no visible data

This filter is a shortcut for

 filter => { 0 => sub { grep { defined && m/\S/ } @{$_[1]} } }

This filter rejects all lines that not have at least one field that does not evaluate to the empty string.

With the given example data, this filter would skip lines 2 through 8.

One could also use modules like Types::Standard:

 use Types::Standard -types;

 my $type   = Tuple[Str, Str, Int, Bool, Optional[Num]];
 my $check  = $type->compiled_check;

 # filter with compiled check and warnings
 my $aoa = csv (
    in     => \$data,
    filter => {
        0 => sub {
            my $ok = $check->($_[1]) or
                warn $type->get_message ($_[1]), "\n";
            return $ok;

This callback is invoked for each record after all records have been parsed but before returning the reference to the caller. The hook is invoked with two arguments: the current CSV parser object and a reference to the record. The reference can be a reference to a HASH or a reference to an ARRAY as determined by the arguments.

This callback can also be passed as an attribute without the callbacks wrapper.


This callback is invoked for each record before the record is printed. The hook is invoked with two arguments: the current CSV parser object and a reference to the record. The reference can be a reference to a HASH or a reference to an ARRAY as determined by the arguments.

This callback can also be passed as an attribute without the callbacks wrapper.

This callback makes the row available in %_ if the row is a hashref. In this case %_ is writable and will change the original row.


This callback acts exactly as the "after_in" or the "before_out" hooks.

This callback can also be passed as an attribute without the callbacks wrapper.

This callback makes the row available in %_ if the row is a hashref. In this case %_ is writable and will change the original row. So e.g. with

  my $aoh = csv (
      in      => \"foo\n1\n2\n",
      headers => "auto",
      on_in   => sub { $_{bar} = 2; },

$aoh will be:

  [ { foo => 1,
      bar => 2,
    { foo => 2,
      bar => 2,


Combine (...)
Parse (...)

The arguments to these internal functions are deliberately not described or documented in order to enable the module authors make changes it when they feel the need for it. Using them is highly discouraged as the API may change in future releases.


Reading a CSV file line by line:

 my $csv = Text::CSV_XS->new ({ binary => 1, auto_diag => 1 });
 open my $fh, "<", "file.csv" or die "file.csv: $!";
 while (my $row = $csv->getline ($fh)) {
     # do something with @$row
 close $fh or die "file.csv: $!";


 my $aoa = csv (in => "file.csv", on_in => sub {
     # do something with %_

Reading only a single column

 my $csv = Text::CSV_XS->new ({ binary => 1, auto_diag => 1 });
 open my $fh, "<", "file.csv" or die "file.csv: $!";
 # get only the 4th column
 my @column = map { $_->[3] } @{$csv->getline_all ($fh)};
 close $fh or die "file.csv: $!";

with "csv", you could do

 my @column = map { $_->[0] }
     @{csv (in => "file.csv", fragment => "col=4")};

Parsing CSV strings:

 my $csv = Text::CSV_XS->new ({ keep_meta_info => 1, binary => 1 });

 my $sample_input_string =
     qq{"I said, ""Hi!""",Yes,"",2.34,,"1.09","\x{20ac}",};
 if ($csv->parse ($sample_input_string)) {
     my @field = $csv->fields;
     foreach my $col (0 .. $#field) {
         my $quo = $csv->is_quoted ($col) ? $csv->{quote_char} : "";
         printf "%2d: %s%s%s\n", $col, $quo, $field[$col], $quo;
 else {
     print STDERR "parse () failed on argument: ",
         $csv->error_input, "\n";
     $csv->error_diag ();

Parsing CSV from memory

Given a complete CSV data-set in scalar $data, generate a list of lists to represent the rows and fields

 # The data
 my $data = join "\r\n" => map { join "," => 0 .. 5 } 0 .. 5;

 # in a loop
 my $csv = Text::CSV_XS->new ({ binary => 1, auto_diag => 1 });
 open my $fh, "<", \$data;
 my @foo;
 while (my $row = $csv->getline ($fh)) {
     push @foo, $row;
 close $fh;

 # a single call
 my $foo = csv (in => \$data);

Printing CSV data

The fast way: using "print"

An example for creating CSV files using the "print" method:

 my $csv = Text::CSV_XS->new ({ binary => 1, eol => $/ });
 open my $fh, ">", "foo.csv" or die "foo.csv: $!";
 for (1 .. 10) {
     $csv->print ($fh, [ $_, "$_" ]) or $csv->error_diag;
 close $fh or die "$tbl.csv: $!";

The slow way: using "combine" and "string"

or using the slower "combine" and "string" methods:

 my $csv = Text::CSV_XS->new;

 open my $csv_fh, ">", "hello.csv" or die "hello.csv: $!";

 my @sample_input_fields = (
     'You said, "Hello!"',   5.67,
     '"Surely"',   '',   '3.14159');
 if ($csv->combine (@sample_input_fields)) {
     print $csv_fh $csv->string, "\n";
 else {
     print "combine () failed on argument: ",
         $csv->error_input, "\n";
 close $csv_fh or die "hello.csv: $!";

Generating CSV into memory

Format a data-set (@foo) into a scalar value in memory ($data):

 # The data
 my @foo = map { [ 0 .. 5 ] } 0 .. 3;

 # in a loop
 my $csv = Text::CSV_XS->new ({ binary => 1, auto_diag => 1, eol => "\r\n" });
 open my $fh, ">", \my $data;
 $csv->print ($fh, $_) for @foo;
 close $fh;

 # a single call
 csv (in => \@foo, out => \my $data);

Rewriting CSV

Rewrite CSV files with ; as separator character to well-formed CSV:

 use Text::CSV_XS qw( csv );
 csv (in => csv (in => "bad.csv", sep_char => ";"), out => *STDOUT);

As STDOUT is now default in "csv", a one-liner converting a UTF-16 CSV file with BOM and TAB-separation to valid UTF-8 CSV could be:

 $ perl -C3 -MText::CSV_XS=csv -we\
    'csv(in=>"utf16tab.csv",encoding=>"utf16",sep=>"\t")' >utf8.csv

Dumping database tables to CSV

Dumping a database table can be simple as this (TIMTOWTDI):

 my $dbh = DBI->connect (...);
 my $sql = "select * from foo";

 # using your own loop
 open my $fh, ">", "foo.csv" or die "foo.csv: $!\n";
 my $csv = Text::CSV_XS->new ({ binary => 1, eol => "\r\n" });
 my $sth = $dbh->prepare ($sql); $sth->execute;
 $csv->print ($fh, $sth->{NAME_lc});
 while (my $row = $sth->fetch) {
     $csv->print ($fh, $row);

 # using the csv function, all in memory
 csv (out => "foo.csv", in => $dbh->selectall_arrayref ($sql));

 # using the csv function, streaming with callbacks
 my $sth = $dbh->prepare ($sql); $sth->execute;
 csv (out => "foo.csv", in => sub { $sth->fetch            });
 csv (out => "foo.csv", in => sub { $sth->fetchrow_hashref });

Note that this does not discriminate between "empty" values and NULL-values from the database, as both will be the same empty field in CSV. To enable distinction between the two, use quote_empty.

 csv (out => "foo.csv", in => sub { $sth->fetch }, quote_empty => 1);

If the database import utility supports special sequences to insert NULL values into the database, like MySQL/MariaDB supports \N, use a filter or a map

 csv (out => "foo.csv", in => sub { $sth->fetch },
                     on_in => sub { $_ //= "\\N" for @{$_[1]} });

 while (my $row = $sth->fetch) {
     $csv->print ($fh, [ map { $_ // "\\N" } @$row ]);

Note that this will not work as expected when choosing the backslash (\) as escape_char, as that will cause the \ to need to be escaped by yet another \, which will cause the field to need quotation and thus ending up as "\\N" instead of \N. See also undef_str.

 csv (out => "foo.csv", in => sub { $sth->fetch }, undef_str => "\\N");

These special sequences are not recognized by Text::CSV_XS on parsing the CSV generated like this, but map and filter are your friends again

 while (my $row = $csv->getline ($fh)) {
     $sth->execute (map { $_ eq "\\N" ? undef : $_ } @$row);

 csv (in => "foo.csv", filter => { 1 => sub {
     $sth->execute (map { $_ eq "\\N" ? undef : $_ } @{$_[1]}); 0; }});

Converting CSV to JSON

 use Text::CSV_XS qw( csv );
 use JSON; # or Cpanel::JSON::XS for better performance

 # AoA (no header interpretation)
 say encode_json (csv (in => "file.csv"));

 # AoH (convert to structures)
 say encode_json (csv (in => "file.csv", bom => 1));

Yes, it is that simple.

The examples folder

For more extended examples, see the examples/ 1. sub-directory in the original distribution or the git repository 2.


The following files can be found there:

This can be used as a boilerplate to parse invalid CSV and parse beyond (expected) errors alternative to using the "error" callback.

 $ perl examples/ bad.csv >good.csv

This is a command-line tool that uses techniques to check the CSV file and report on its content.

 $ csv-check files/utf8.csv
 Checked files/utf8.csv  with csv-check 1.9
 using Text::CSV_XS 1.32 with perl 5.26.0 and Unicode 9.0.0
 OK: rows: 1, columns: 2
     sep = <,>, quo = <">, bin = <1>, eol = <"\n">

This command splits CSV files into smaller files, keeping (part of) the header. Options include maximum number of (data) rows per file and maximum number of columns per file or a combination of the two.


A script to convert CSV to Microsoft Excel (XLS). This requires extra modules Date::Calc and Spreadsheet::WriteExcel. The converter accepts various options and can produce UTF-8 compliant Excel files.


A script to convert CSV to Microsoft Excel (XLSX). This requires the modules Date::Calc and Spreadsheet::Writer::XLSX. The converter does accept various options including merging several CSV files into a single Excel file.


A script that provides colorized diff on sorted CSV files, assuming first line is header and first field is the key. Output options include colorized ANSI escape codes or HTML.

 $ csvdiff --html --output=diff.html file1.csv file2.csv

A script to rewrite (in)valid CSV into valid CSV files. Script has options to generate confusing CSV files or CSV files that conform to Dutch MS-Excel exports (using ; as separation).

Script - by default - honors BOM and auto-detects separation converting it to default standard CSV with , as separator.


Text::CSV_XS is not designed to detect the characters used to quote and separate fields. The parsing is done using predefined (default) settings. In the examples sub-directory, you can find scripts that demonstrate how you could try to detect these characters yourself.

Microsoft Excel

The import/export from Microsoft Excel is a risky task, according to the documentation in Text::CSV::Separator. Microsoft uses the system's list separator defined in the regional settings, which happens to be a semicolon for Dutch, German and Spanish (and probably some others as well). For the English locale, the default is a comma. In Windows however, the user is free to choose a predefined locale, and then change every individual setting in it, so checking the locale is no solution.

As of version 1.17, a lone first line with just


will be recognized and honored when parsing with "getline".


More Errors & Warnings

New extensions ought to be clear and concise in reporting what error has occurred where and why, and maybe also offer a remedy to the problem.

"error_diag" is a (very) good start, but there is more work to be done in this area.

Basic calls should croak or warn on illegal parameters. Errors should be documented.

setting meta info

Future extensions might include extending the "meta_info", "is_quoted", and "is_binary" to accept setting these flags for fields, so you can specify which fields are quoted in the "combine"/"string" combination.

 $csv->meta_info (0, 1, 1, 3, 0, 0);
 $csv->is_quoted (3, 1);

Metadata Vocabulary for Tabular Data (a W3C editor's draft) could be an example for supporting more metadata.

Parse the whole file at once

Implement new methods or functions that enable parsing of a complete file at once, returning a list of hashes. Possible extension to this could be to enable a column selection on the call:

 my @AoH = $csv->parse_file ($filename, { cols => [ 1, 4..8, 12 ]});

returning something like

 [ { fields => [ 1, 2, "foo", 4.5, undef, "", 8 ],
     flags  => [ ... ],
   { fields => [ ... ],

Note that the "csv" function already supports most of this, but does not return flags. "getline_all" returns all rows for an open stream, but this will not return flags either. "fragment" can reduce the required rows or columns, but cannot combine them.

 csv (in => $fh) vs csv (provider => sub { get_line });

Whatever the attribute name might end up to be, this should make it easier to add input providers for parsing. Currently most special variations for the in attribute are aimed at CSV generation: e.g. a callback is defined to return a reference to a record. This new attribute should enable passing data to parse, like getline.

Suggested by Johan Vromans.


Write a document that has recipes for most known non-standard (and maybe some standard) CSV formats, including formats that use TAB, ;, |, or other non-comma separators.

Examples could be taken from W3C's CSV on the Web: Use Cases and Requirements


Steal good new ideas and features from PapaParse or csvkit.

Raku support

Raku support can be found here. The interface is richer in support than the Perl5 API, as Raku supports more types.

The Raku version does not (yet) support pure binary CSV datasets.


combined methods

Requests for adding means (methods) that combine "combine" and "string" in a single call will not be honored (use "print" instead). Likewise for "parse" and "fields" (use "getline" instead), given the problems with embedded newlines.

Release plan

No guarantees, but this is what I had in mind some time ago:

  • DIAGNOSTICS section in pod to *describe* the errors (see below)


Everything should now work on native EBCDIC systems. As the test does not cover all possible codepoints and Encode does not support utf-ebcdic, there is no guarantee that all handling of Unicode is done correct.

Opening EBCDIC encoded files on ASCII+ systems is likely to succeed using Encode's cp37, cp1047, or posix-bc:

 open my $fh, "<:encoding(cp1047)", "ebcdic_file.csv" or die "...";


Still under construction ...

If an error occurs, $csv->error_diag can be used to get information on the cause of the failure. Note that for speed reasons the internal value is never cleared on success, so using the value returned by "error_diag" in normal cases - when no error occurred - may cause unexpected results.

If the constructor failed, the cause can be found using "error_diag" as a class method, like Text::CSV_XS->error_diag.

The $csv->error_diag method is automatically invoked upon error when the contractor was called with auto_diag set to 1 or 2, or when autodie is in effect. When set to 1, this will cause a warn with the error message, when set to 2, it will die. 2012 - EOF is excluded from auto_diag reports.

Errors can be (individually) caught using the "error" callback.

The errors as described below are available. I have tried to make the error itself explanatory enough, but more descriptions will be added. For most of these errors, the first three capitals describe the error category:

  • INI

    Initialization error or option conflict.

  • ECR

    Carriage-Return related parse error.

  • EOF

    End-Of-File related parse error.

  • EIQ

    Parse error inside quotation.

  • EIF

    Parse error inside field.

  • ECB

    Combine error.

  • EHR

    HashRef parse related error.

And below should be the complete list of error codes that can be returned:

  • 1001 "INI - sep_char is equal to quote_char or escape_char"

    The separation character cannot be equal to the quotation character or to the escape character, as this would invalidate all parsing rules.

  • 1002 "INI - allow_whitespace with escape_char or quote_char SP or TAB"

    Using the allow_whitespace attribute when either quote_char or escape_char is equal to SPACE or TAB is too ambiguous to allow.

  • 1003 "INI - \r or \n in main attr not allowed"

    Using default eol characters in either sep_char, quote_char, or escape_char is not allowed.

  • 1004 "INI - callbacks should be undef or a hashref"

    The callbacks attribute only allows one to be undef or a hash reference.

  • 1005 "INI - EOL too long"

    The value passed for EOL is exceeding its maximum length (16).

  • 1006 "INI - SEP too long"

    The value passed for SEP is exceeding its maximum length (16).

  • 1007 "INI - QUOTE too long"

    The value passed for QUOTE is exceeding its maximum length (16).

  • 1008 "INI - SEP undefined"

    The value passed for SEP should be defined and not empty.

  • 1010 "INI - the header is empty"

    The header line parsed in the "header" is empty.

  • 1011 "INI - the header contains more than one valid separator"

    The header line parsed in the "header" contains more than one (unique) separator character out of the allowed set of separators.

  • 1012 "INI - the header contains an empty field"

    The header line parsed in the "header" contains an empty field.

  • 1013 "INI - the header contains nun-unique fields"

    The header line parsed in the "header" contains at least two identical fields.

  • 1014 "INI - header called on undefined stream"

    The header line cannot be parsed from an undefined source.

  • 1500 "PRM - Invalid/unsupported argument(s)"

    Function or method called with invalid argument(s) or parameter(s).

  • 1501 "PRM - The key attribute is passed as an unsupported type"

    The key attribute is of an unsupported type.

  • 1502 "PRM - The value attribute is passed without the key attribute"

    The value attribute is only allowed when a valid key is given.

  • 1503 "PRM - The value attribute is passed as an unsupported type"

    The value attribute is of an unsupported type.

  • 2010 "ECR - QUO char inside quotes followed by CR not part of EOL"

    When eol has been set to anything but the default, like "\r\t\n", and the "\r" is following the second (closing) quote_char, where the characters following the "\r" do not make up the eol sequence, this is an error.

  • 2011 "ECR - Characters after end of quoted field"

    Sequences like 1,foo,"bar"baz,22,1 are not allowed. "bar" is a quoted field and after the closing double-quote, there should be either a new-line sequence or a separation character.

  • 2012 "EOF - End of data in parsing input stream"

    Self-explaining. End-of-file while inside parsing a stream. Can happen only when reading from streams with "getline", as using "parse" is done on strings that are not required to have a trailing eol.

  • 2013 "INI - Specification error for fragments RFC7111"

    Invalid specification for URI "fragment" specification.

  • 2014 "ENF - Inconsistent number of fields"

    Inconsistent number of fields under strict parsing.

  • 2015 "ERW - Empty row"

    An empty row was not allowed.

  • 2021 "EIQ - NL char inside quotes, binary off"

    Sequences like 1,"foo\nbar",22,1 are allowed only when the binary option has been selected with the constructor.

  • 2022 "EIQ - CR char inside quotes, binary off"

    Sequences like 1,"foo\rbar",22,1 are allowed only when the binary option has been selected with the constructor.

  • 2023 "EIQ - QUO character not allowed"

    Sequences like "foo "bar" baz",qu and 2023,",2008-04-05,"Foo, Bar",\n will cause this error.

  • 2024 "EIQ - EOF cannot be escaped, not even inside quotes"

    The escape character is not allowed as last character in an input stream.

  • 2025 "EIQ - Loose unescaped escape"

    An escape character should escape only characters that need escaping.

    Allowing the escape for other characters is possible with the attribute "allow_loose_escapes".

  • 2026 "EIQ - Binary character inside quoted field, binary off"

    Binary characters are not allowed by default. Exceptions are fields that contain valid UTF-8, that will automatically be upgraded if the content is valid UTF-8. Set binary to 1 to accept binary data.

  • 2027 "EIQ - Quoted field not terminated"

    When parsing a field that started with a quotation character, the field is expected to be closed with a quotation character. When the parsed line is exhausted before the quote is found, that field is not terminated.

  • 2030 "EIF - NL char inside unquoted verbatim, binary off"

  • 2031 "EIF - CR char is first char of field, not part of EOL"

  • 2032 "EIF - CR char inside unquoted, not part of EOL"

  • 2034 "EIF - Loose unescaped quote"

  • 2035 "EIF - Escaped EOF in unquoted field"

  • 2036 "EIF - ESC error"

  • 2037 "EIF - Binary character in unquoted field, binary off"

  • 2110 "ECB - Binary character in Combine, binary off"

  • 2200 "EIO - print to IO failed. See errno"

  • 3001 "EHR - Unsupported syntax for column_names ()"

  • 3002 "EHR - getline_hr () called before column_names ()"

  • 3003 "EHR - bind_columns () and column_names () fields count mismatch"

  • 3004 "EHR - bind_columns () only accepts refs to scalars"

  • 3006 "EHR - bind_columns () did not pass enough refs for parsed fields"

  • 3007 "EHR - bind_columns needs refs to writable scalars"

  • 3008 "EHR - unexpected error in bound fields"

  • 3009 "EHR - print_hr () called before column_names ()"

  • 3010 "EHR - print_hr () called with invalid arguments"


IO::File, IO::Handle, IO::Wrap, Text::CSV, Text::CSV_PP, Text::CSV::Encoded, Text::CSV::Separator, Text::CSV::Slurp, Spreadsheet::CSV and Spreadsheet::Read, and of course perl.

If you are using Raku, have a look at Text::CSV in the Raku ecosystem, offering the same features.


A CSV parser in JavaScript, also used by W3C, is the multi-threaded in-browser PapaParse.

csvkit is a python CSV parsing toolkit.


Alan Citterman <> wrote the original Perl module. Please don't send mail concerning Text::CSV_XS to Alan, who is not involved in the C/XS part that is now the main part of the module.

Jochen Wiedmann <> rewrote the en- and decoding in C by implementing a simple finite-state machine. He added variable quote, escape and separator characters, the binary mode and the print and getline methods. See ChangeLog releases 0.10 through 0.23.

H.Merijn Brand <> cleaned up the code, added the field flags methods, wrote the major part of the test suite, completed the documentation, fixed most RT bugs, added all the allow flags and the "csv" function. See ChangeLog releases 0.25 and on.


 Copyright (C) 2007-2024 H.Merijn Brand.  All rights reserved.
 Copyright (C) 1998-2001 Jochen Wiedmann. All rights reserved.
 Copyright (C) 1997      Alan Citterman.  All rights reserved.

This library is free software; you can redistribute and/or modify it under the same terms as Perl itself.