Blame | Last modification | View Log | RSS feed
#!/usr/bin/env perl
# csv2xls: Convert csv to xls
# (m)'11 [06 Oct 2011] Copyright H.M.Brand 2007-2013
use strict;
use warnings;
our $VERSION = "1.71";
sub usage
{
my $err = shift and select STDERR;
print <<EOU;
usage: csv2xls [-s <sep>] [-q <quot>] [-w <width>] [-d <dtfmt>]
[-o <xls>] [file.csv]
-s <sep> use <sep> as seperator char. Auto-detect, default = ','
The string "tab" is allowed.
-e <esc> use <sep> as seperator char. Auto-detect, default = ','
The string "undef" is allowed.
-q <quot> use <quot> as quotation char. Default = '"'
The string "undef" will disable quotation.
-w <width> use <width> as default minimum column width (4)
-o <xls> write output to file named <xls>, defaults
to input file name with .csv replaced with .xls
if from standard input, defaults to csv2xls.xls
-F allow formula's. Otherwise fields starting with
an equal sign are forced to string
-f force usage of <xls> if already exists (unlink before use)
-d <dtfmt> use <dtfmt> as date formats. Default = 'dd-mm-yyyy'
-D cols only convert dates in columns <cols>. Default is everywhere.
-u CSV is UTF8
-v [<lvl>] verbosity (default = 1)
EOU
exit $err;
} # usage
use Getopt::Long qw(:config bundling nopermute passthrough);
my $sep; # Set after reading first line in a flurry attempt to auto-detect
my $quo = '"';
my $esc = '"';
my $wdt = 4; # Default minimal column width
my $xls; # Excel out file name
my $frc = 0; # Force use of file
my $utf = 0; # Data is encoded in Unicode
my $frm = 0; # Allow formula's
my $dtf = "dd-mm-yyyy"; # Date format
my $opt_v = 1;
my $dtc;
GetOptions (
"help|?" => sub { usage (0); },
"c|s=s" => \$sep,
"q=s" => \$quo,
"e=s" => \$esc,
"w=i" => \$wdt,
"o|x=s" => \$xls,
"d=s" => \$dtf,
"D=s" => \$dtc,
"f" => \$frc,
"F" => \$frm,
"u" => \$utf,
"v:1" => \$opt_v,
) or usage (1);
my $title = @ARGV && -f $ARGV[0] ? $ARGV[0] : "csv2xls";
($xls ||= $title) =~ s/(?:\.csv)?$/.xls/i;
-s $xls && $frc and unlink $xls;
if (-s $xls) {
print STDERR "File '$xls' already exists. Overwrite? [y/N] > N\b";
scalar <STDIN> =~ m/^[yj](es|a)?$/i or exit;
}
# Don't split ourselves when modules do it _much_ better, and follow the standards
use Text::CSV_XS;
use Date::Calc qw( Delta_Days Days_in_Month );
use Spreadsheet::WriteExcel;
use Encode qw( from_to );
my $csv;
my $wbk = Spreadsheet::WriteExcel->new ($xls);
my $wks = $wbk->add_worksheet ();
$dtf =~ s/j/y/g;
my %fmt = (
date => $wbk->add_format (
num_format => $dtf,
align => "center",
),
rest => $wbk->add_format (
align => "left",
),
);
my ($h, $w, @w) = (0, 1); # data height, -width, and default column widths
my $row;
my $firstline;
unless ($sep) { # No sep char passed, try to auto-detect;
while (<>) {
m/\S/ or next; # Skip empty leading blank lines
$sep = # start auto-detect with quoted strings
m/["\d];["\d;]/ ? ";" :
m/["\d],["\d,]/ ? "," :
m/["\d]\t["\d,]/ ? "\t" :
# If neither, then for unquoted strings
m/\w;[\w;]/ ? ";" :
m/\w,[\w,]/ ? "," :
m/\w\t[\w,]/ ? "\t" :
";" ;
# Yeah I know it should be a ',' (hence Csv), but the majority
# of the csv files to be shown comes from fucking Micky$hit,
# that uses semiColon ';' instead.
$firstline = $_;
last;
}
}
$csv = Text::CSV_XS-> new ({
sep_char => $sep eq "tab" ? "\t" : $sep,
quote_char => $quo eq "undef" ? undef : $quo,
escape_char => $esc eq "undef" ? undef : $esc,
binary => 1,
keep_meta_info => 1,
auto_diag => 1,
});
if ($firstline) {
$csv->parse ($firstline) or die $csv->error_diag ();
$row = [ $csv->fields ];
}
if ($opt_v > 3) {
foreach my $k (qw( sep_char quote_char escape_char )) {
my $c = $csv->$k () || "undef";
$c =~ s/\t/\\t/g;
$c =~ s/\r/\\r/g;
$c =~ s/\n/\\n/g;
$c =~ s/\0/\\0/g;
$c =~ s/([\x00-\x1f\x7f-\xff])/sprintf"\\x{%02x}",ord$1/ge;
printf STDERR "%-11s = %s\n", $k, $c;
}
}
if (my $rows = $dtc) {
$rows =~ s/-$/-999/; # 3,6-
$rows =~ s/-/../g;
eval "\$dtc = { map { \$_ => 1 } $rows }";
}
while ($row && @$row or $row = $csv->getline (*ARGV)) {
my @row = @$row;
@row > $w and push @w, ($wdt) x (($w = @row) - @w);
foreach my $c (0 .. $#row) {
my $val = $row[$c] || "";
my $l = length $val;
$l > $w[$c] and $w[$c] = $l;
if ($utf and $csv->is_binary ($c)) {
from_to ($val, "utf-8", "ucs2");
$wks->write_unicode ($h, $c, $val);
next;
}
if ($csv->is_quoted ($c)) {
if ($utf) {
from_to ($val, "utf-8", "ucs2");
$wks->write_unicode ($h, $c, $val);
}
else {
$wks->write_string ($h, $c, $val);
}
next;
}
if (!$dtc or $dtc->{$c + 1}) {
my @d = (0, 0, 0); # Y, M, D
$val =~ m/^(\d{4})(\d{2})(\d{2})$/ and @d = ($1, $2, $3);
$val =~ m/^(\d{2})-(\d{2})-(\d{4})$/ and @d = ($3, $2, $1);
if ( $d[2] >= 1 && $d[2] <= 31 &&
$d[1] >= 1 && $d[1] <= 12 &&
$d[0] >= 1900 && $d[0] <= 2199) {
my $dm = Days_in_Month (@d[0,1]);
$d[2] < 1 and $d[2] = 1;
$d[2] > $dm and $d[2] = $dm;
my $dt = 2 + Delta_Days (1900, 1, 1, @d);
$wks->write ($h, $c, $dt, $fmt{date});
next;
}
}
if (!$frm && $val =~ m/^=/) {
$wks->write_string ($h, $c, $val);
}
else {
$wks->write ($h, $c, $val);
}
}
++$h % 100 or $opt_v && printf STDERR "%6d x %6d\r", $w, $h;
} continue { $row = undef }
$opt_v && printf STDERR "%6d x %6d\n", $w, $h;
$wks->set_column ($_, $_, $w[$_]) for 0 .. $#w;
$wbk->close ();