[kivitendo-erp.git] / SL / Helper / Csv.pm

package SL::Helper::Csv;

use strict;
use warnings;

use version 0.77;
use Carp;
use IO::File;
use Params::Validate qw(:all);
use List::MoreUtils qw(all pairwise firstidx);
use Text::CSV_XS;
use Rose::Object::MakeMethods::Generic scalar => [ qw(
  file encoding sep_char quote_char escape_char header profile
  numberformat dateformat ignore_unknown_columns strict_profile is_multiplexed
  _row_header _io _csv _objects _parsed _data _errors all_cvar_configs case_insensitive_header
  _multiplex_datatype_position
) ];

use SL::Helper::Csv::Dispatcher;
use SL::Helper::Csv::Error;

# public interface

sub new {
  my $class  = shift;
  my %params = validate(@_, {
    sep_char               => { default => ';' },
    quote_char             => { default => '"' },
    escape_char            => { default => '"' },
    header                 => { type    => ARRAYREF, optional => 1 },
    profile                => { type    => ARRAYREF, optional => 1 },
    file                   => 1,
    encoding               => 0,
    numberformat           => 0,
    dateformat             => 0,
    ignore_unknown_columns => 0,
    strict_profile         => 0,
    case_insensitive_header => 0,
  });
  my $self = bless {}, $class;

  $self->$_($params{$_}) for keys %params;

  $self->_io(IO::File->new);
  $self->_csv(Text::CSV_XS->new({
    binary => 1,
    sep_char    => $self->sep_char,
    quote_char  => $self->quote_char,
    escape_char => $self->escape_char,

  }));
  $self->_errors([]);

  return $self;
}

sub parse {
  my ($self, %params) = @_;

  $self->_open_file;
  return if ! $self->_check_multiplexed;
  return if ! $self->_check_header;
  return if ! $self->_check_multiplex_datatype_position;
  return if ! $self->dispatcher->parse_profile;
  return if ! $self->_parse_data;

  $self->_parsed(1);
  return $self;
}

sub get_data {
  $_[0]->_data;
}

sub get_objects {
  my ($self, %params) = @_;
  croak 'must parse first' unless $self->_parsed;

  $self->_make_objects unless $self->_objects;
  return $self->_objects;
}

sub errors {
  @{ $_[0]->_errors }
}

sub check_header {
  $_[0]->_check_header;
}

# private stuff

sub _open_file {
  my ($self, %params) = @_;

  $self->encoding($self->_guess_encoding) if !$self->encoding;

  $self->_io->open($self->file, '<' . $self->_encode_layer)
    or die "could not open file " . $self->file;

  return $self->_io;
}

# check, if data is multiplexed and if all nessesary infos are given
sub _check_multiplexed {
  my ($self, %params) = @_;

  $self->is_multiplexed(0);

  # If more than one profile is given, it is multiplexed.
  if ($self->profile) {
    my @profile = @{ $self->profile };
    if (scalar @profile > 1) {
      # Each profile needs a class and a row_ident
      my $info_ok = all { defined $_->{class} && defined $_->{row_ident} } @profile;
      $self->_push_error([
        0,
        "missing class or row_ident in one of the profiles for multiplexed data",
        0,
        0]) unless $info_ok;

      # If header is given, there needs to be a header for each profile
      # and no empty headers.
      if ($info_ok && $self->header) {
        my @header = @{ $self->header };
        my $t_ok = scalar @profile == scalar @header;
        $self->_push_error([
          0,
          "number of headers and number of profiles must be the same for multiplexed data",
          0,
          0]) unless $t_ok;
        $info_ok = $info_ok && $t_ok;

        $t_ok = all { scalar @$_ > 0} @header;
        $self->_push_error([
          0,
          "no empty headers are allowed for multiplexed data",
          0,
          0]) unless $t_ok;
        $info_ok = $info_ok && $t_ok;
      }
      $self->is_multiplexed($info_ok);
      return $info_ok;
    }
  }

  # ok, if not multiplexed
  return 1;
}

sub _check_header {
  my ($self, %params) = @_;
  my $header;

  $header = $self->header;
  if (!$header) {
    my $n_header = ($self->is_multiplexed)? scalar @{ $self->profile } : 1;
    foreach my $p_num (0..$n_header - 1) {
      my $h = $self->_csv->getline($self->_io);

      my ($code, $string, $position, $record, $field) = $self->_csv->error_diag;

      $self->_push_error([
        $self->_csv->error_input,
        $code, $string, $position, $record // 0,
      ]) unless $h;

      if ($self->is_multiplexed) {
        push @{ $header }, $h;
      } else {
        $header = $h;
      }
    }
  }

  # Special case: utf8 BOM.
  # certain software (namely MS Office and notepad.exe insist on prefixing
  # data with a discouraged but valid byte order mark
  # if not removed, the first header field will not be recognized
  if ($header) {
    my $h = ($self->is_multiplexed)? $header->[0] : $header;

    if ($h && $h->[0] && $self->encoding =~ /utf-?8/i) {
      $h->[0] =~ s/^\x{FEFF}//;
    }
  }

  # check, if all header fields are parsed well
  if ($self->is_multiplexed) {
    return unless $header && all { $_ } @$header;
  } else {
    return unless $header;
  }

  # Special case: human stupidity
  # people insist that case sensitivity doesn't exist and try to enter all
  # sorts of stuff. at this point we've got a profile (with keys that represent
  # valid methods), and a header full of strings. if two of them match, the user
  # most likely meant that field, so rewrite the header
  if ($self->case_insensitive_header) {
    die 'case_insensitive_header is only possible with profile' unless $self->profile;
    if ($header) {
      my $h_aref = ($self->is_multiplexed)? $header : [ $header ];
      my $p_num  = 0;
      foreach my $h (@{ $h_aref }) {
        my %names = (
          (map { $_ => $_                                     } keys %{ $self->profile->[$p_num]->{profile} || {} }),
          (map { $_ => $self->profile->[$p_num]{mapping}{$_}  } keys %{ $self->profile->[$p_num]->{mapping} || {} }),
        );
        for my $name (keys %names) {
          for my $i (0..$#$h) {
            $h->[$i] = $names{$name} if lc $h->[$i] eq lc $name;
          }
        }
        $p_num++;
      }
    }
  }

  return $self->header($header);
}

sub _check_multiplex_datatype_position {
  my ($self) = @_;

  return 1 if !$self->is_multiplexed; # ok if not multiplexed

  my @positions = map { firstidx { 'datatype' eq lc($_) } @{ $_ } } @{ $self->header };
  my $first_pos = $positions[0];
  if (all { $first_pos == $_ } @positions) {
    $self->_multiplex_datatype_position($first_pos);
    return 1;
  } else {
    $self->_push_error([0,
                        "datatype field must be at the same position for all datatypes for multiplexed data",
                        0,
                        0]);
    return 0;
  }
}

sub _is_empty_row {
  return !!all { !$_ } @{$_[0]};
}

sub _parse_data {
  my ($self, %params) = @_;
  my (@data, @errors);

  while (1) {
    my $row = $self->_csv->getline($self->_io);
    if ($row) {
      next if _is_empty_row($row);
      my $header = $self->_header_by_row($row);
      if (!$header) {
        push @errors, [
          0,
          "Cannot get header for row. Maybe row name and datatype field not matching.",
          0,
          0];
        last;
      }
      my %hr;
      @hr{@{ $header }} = @$row;
      push @data, \%hr;
    } else {
      last if $self->_csv->eof;

      # Text::CSV_XS 0.89 added record number to error_diag
      my ($code, $string, $position, $record, $field) = $self->_csv->error_diag;

      push @errors, [
        $self->_csv->error_input,
        $code, $string, $position,
        $record // $self->_io->input_line_number,
      ];
    }
    last if $self->_csv->eof;
  }

  $self->_data(\@data);
  $self->_push_error(@errors);

  return ! @errors;
}

sub _header_by_row {
  my ($self, $row) = @_;

  # initialize lookup hash if not already done
  if ($self->is_multiplexed && ! defined $self->_row_header ) {
    $self->_row_header({ pairwise { no warnings 'once'; $a->{row_ident} => $b } @{ $self->profile }, @{ $self->header } });
  }

  if ($self->is_multiplexed) {
    return $self->_row_header->{$row->[$self->_multiplex_datatype_position]}
  } else {
    return $self->header;
  }
}

sub _encode_layer {
  ':encoding(' . $_[0]->encoding . ')';
}

sub _make_objects {
  my ($self, %params) = @_;
  my @objs;

  local $::myconfig{numberformat} = $self->numberformat if $self->numberformat;
  local $::myconfig{dateformat}   = $self->dateformat   if $self->dateformat;

  for my $line (@{ $self->_data }) {
    my $tmp_obj = $self->dispatcher->dispatch($line);
    push @objs, $tmp_obj;
  }

  $self->_objects(\@objs);
}

sub dispatcher {
  my ($self, %params) = @_;

  $self->{_dispatcher} ||= $self->_make_dispatcher;
}

sub _make_dispatcher {
  my ($self, %params) = @_;

  die 'need a header to make a dispatcher' unless $self->header;

  return SL::Helper::Csv::Dispatcher->new($self);
}

sub _guess_encoding {
  # won't fix
  'utf-8';
}

sub _push_error {
  my ($self, @errors) = @_;
  my @new_errors = ($self->errors, map { SL::Helper::Csv::Error->new(@$_) } @errors);
  $self->_errors(\@new_errors);
}

sub specs {
  $_[0]->dispatcher->_specs
}

1;

__END__

=encoding utf-8

=head1 NAME

SL::Helper::Csv - take care of csv file uploads

=head1 SYNOPSIS

  use SL::Helper::Csv;

  my $csv = SL::Helper::Csv->new(
    file        => \$::form->{upload_file},
    encoding    => 'utf-8', # undef means utf8
    sep_char    => ',',     # default ';'
    quote_char  => '\'',    # default '"'
    escape_char => '"',     # default '"'
    header      => [ qw(id text sellprice word) ], # see later
    profile     => [ { profile => { sellprice => 'sellprice_as_number'},
                       class   => 'SL::DB::Part' } ],
  );

  my $status  = $csv->parse;
  my $hrefs   = $csv->get_data;
  my $objects = $csv->get_objects;

  my @errors  = $csv->errors;

=head1 DESCRIPTION

See Synopsis.

Text::CSV already offers good functions to get lines out of a csv file, but in
most cases you will want those lines to be parsed into hashes or even objects,
so this model just skips ahead and gives you objects.

Its basic assumptions are:

=over 4

=item You do know what you expect to be in that csv file.

This means first and foremost that you have knowledge about encoding, number and
date format, csv parameters such as quoting and separation characters. You also
know what content will be in that csv and what L<Rose::DB> is responsible for
it. You provide valid header columns and their mapping to the objects.

=item You do NOT know if the csv provider yields to your expectations.

Stuff that does not work with what you expect should not crash anything, but
give you a hint what went wrong. As a result, if you remember to check for
errors after each step, you should be fine.

=item Data does not make sense. It's just data.

Almost all data imports have some type of constraints. Some data needs to be
unique, other data needs to be connected to existing data sets. This will not
happen here. You will receive a plain mapping of the data into the class tree,
nothing more.

=item Multiplex data

This module can handle multiplexed data of different class types. In that case
multiple profiles with classes and row identifiers must be given. Multiple
headers may also be given or read from csv data. Data must contain the row
identifier in the column named 'datatype'.

=back

=head1 METHODS

=over 4

=item C<new> PARAMS

Standard constructor. You can use this to set most of the data.

=item C<parse>

Do the actual work. Will return true ($self actually) if success, undef if not.

=item C<get_objects>

Parse the data into objects and return those.

This method will return an arrayref of all objects.

=item C<get_data>

Returns an arrayref of the raw lines as hashrefs.

=item C<errors>

Return all errors that came up during parsing. See error handling for detailed
information.

=back

=head1 PARAMS

=over 4

=item C<file>

The file which contents are to be read. Can be a name of a physical file or a
scalar ref for memory data.

=item C<encoding>

Encoding of the CSV file. Note that this module does not do any encoding
guessing. Know what your data is. Defaults to utf-8.

=item C<sep_char>

=item C<quote_char>

=item C<escape_char>

Same as in L<Text::CSV>

=item C<header> \@HEADERS

If given, it contains an ARRAY of the header fields for not multiplexed data.
Or an ARRAYREF for each different class type for multiplexed data. These
ARRAYREFS are the header fields which are an array of columns. In this case
the first lines are not used as a header. Empty header fields will be ignored
in objects.

If not given, headers are taken from the first n lines of data, where n is the
number of different class types.

In case of multiplexed data there must be a column named 'datatype'. This
column must be given in each header and must be at the same position in each
header.

Examples:

  classic data of one type:
  [ 'name', 'street', 'zipcode', 'city' ]

  multiplexed data with two different types:
  [ [ 'datatype', 'ordernumber', 'customer', 'transdate' ],
    [ 'datatype', 'partnumber', 'qty', 'sellprice' ] ]

=item C<profile> PROFILE_DATA

The profile mapping csv to the objects.

See section L</PROFILE> for information on this topic.

=item C<ignore_unknown_columns>

If set, the import will ignore unknown header columns. Useful for lazy imports,
but deactivated by default.

=item C<case_insensitive_header>

If set, header columns will be matched against profile entries case
insensitive, and on match the profile name will be taken.

Only works if a profile is given, will die otherwise.

If both C<case_insensitive_header> and C<strict_profile> is set, matched header
columns will be accepted.

=item C<strict_profile>

If set, all columns to be parsed must be specified in C<profile>. Every header
field not listed there will be treated like an unknown column.

If both C<case_insensitive_header> and C<strict_profile> is set, matched header
columns will be accepted.

=back

=head1 PROFILE

The profile is needed for mapping csv data to the accessors in the data object.

The basic structure is:

  PROFILE       := [ CLASS_PROFILE, CLASS_PROFILE* ]
  CLASS_PROFILE := {
                      profile   => { ACCESSORS+ },
                      class     => $classname,
                      row_ident => $row_ident,
                      mapping   => { MAPPINGS* },
                   }
  ACCESSORS     := $field => $accessor
  MAPPINGS      := $alias => $field

The C<ACCESSORS> may be used to map header fields to custom
accessors. Example:

  profile => {
    listprice => 'listprice_as_number',
  }

In this case C<listprice_as_number> will be used to store the values from the
C<listprice> column.

In case of a One-To-One relationship these can also be set over
relationships by separating the steps with a dot (C<.>). This will work:

  customer => 'customer.name',

And will result in something like this:

  $obj->customer($obj->meta->relationship('customer')->class->new);
  $obj->customer->name($csv_line->{customer})

Beware, this will not try to look up anything in the database! You will
simply receive objects that represent what the profile defined. If some of
these information are unique, or should be connected to preexisting data, you
will have to do that for yourself. Since you provided the profile, it is
assumed you know what to do in this case.

If no profile is given, any header field found will be taken as is.

If the path in a profile entry is empty, the field will be subjected to
C<strict_profile> and C<case_insensitive_header> checking and will be parsed
into C<get_data>, but will not be attempted to be dispatched into objects.

C<class> must be present. A new instance will be created for each line before
dispatching into it.

C<row_ident> is used to determine the correct profile in multiplexed data and
must be given there. It's not used in non-multiplexed data.

If C<mappings> is present, it must contain a hashref that maps strings to known
fields. This can be used to add custom profiles for known sources, that don't
comply with the expected header identities.

Without strict profiles, mappings can also directly map header fields that
should end up in the same accessor.

With case insensitive headings, mappings will also modify the headers, to fit
the expected profile.

Mappings can be identical to known fields and will be prefered during lookup,
but will not replace the field, meaning that:

  profile => {
    name        => 'name',
    description => 'description',
  }
  mapping => {
    name        => 'description',
    shortname   => 'name',
  }

will work as expected, and shortname will not end up in description. This also
works with the case insensitive option. Note however that the case insensitive
option will not enable true unicode collating.


Here's a full example:

  [
    {
      class     => 'SL::DB::Order',
      row_ident => 'O'
    },
    {
      class     => 'SL::DB::OrderItem',
      row_ident => 'I',
      profile   => { sellprice => 'sellprice_as_number' },
      mapping   => { 'Verkaufspreis' => 'sellprice' }
    },
  ]

=head1 ERROR HANDLING

After parsing a file all errors will be accumulated into C<errors>.
Each entry is an object with the following attributes:

 raw_input:  offending raw input,
 code:   Text::CSV error code if Text:CSV signalled an error, 0 else,
 diag:   error diagnostics,
 line:   position in line,
 col:    estimated line in file,

Note that the last entry can be off, but will give an estimate.

Error handling is also known to break on new Perl versions and need to be
adjusted from time to time due to changes in Text::CSV_XS.

=head1 CAVEATS

=over 4

=item *

sep_char, quote_char, and escape_char are passed to Text::CSV on creation.
Changing them later has no effect currently.

=item *

Encoding errors are not dealt with properly.

=back

=head1 TODO

Dispatch to child objects, like this:

 $csv = SL::Helper::Csv->new(
   file    => ...
   profile => [ {
     profile => [
       makemodel => {
         make_1  => make,
         model_1 => model,
       },
       makemodel => {
         make_2  => make,
         model_2 => model,
       },
     ],
     class   => SL::DB::Part,
   } ]
 );

=head1 AUTHOR

Sven Schöling E<lt>s.schoeling@linet-services.deE<gt>

=cut